POPULATION ECOLOGY OF ALEWIFE AND CISCO IN LAKE ECOSYSTEMS   

 

 

 

 

 

 

 

 

A Dissertation 

Presented to the Faculty of the Graduate School 

of Cornell University 

In Partial Fulfillment of the Requirements for the Degree of 

Doctor of Philosophy 

 

 

 

 

 

 

by 

Alexander L. Koeberle 

December 2025



 

 

 

 

 

 

 

 

 

 

 

 

 

© 2025 Alexander L. Koeberle



 

 

POPULATION ECOLOGY OF ALEWIFE AND CISCO IN LAKE ECOSYSTEMS   

 

Alexander L. Koeberle, Ph. D. 

Cornell University 2025 

 

Fisheries restoration is increasingly used to recover populations and improve food web 

function. Despite widespread application, the ability to evaluate restoration success is 

often limited without system-specific information, including the original drivers of 

fish population decline. In this dissertation, I present the trajectory of two 

planktivorous fishes: introduced alewife (Alosa pseudoharengus) and extirpated cisco 

(Coregonus artedi), the focal species of a reintroduction effort in Keuka Lake, New 

York. I first investigated the collapse of alewife by the mid-2010s across two lakes, 

Keuka Lake and Otsego Lake, New York, testing both top-down and bottom-up 

hypotheses for population decline. A data-driven approach across trophic levels 

revealed multiple pathways to alewife collapse, with common ecosystem 

consequences including a pelagic prey fish void left vacant in both lakes over a decade 

later. Next, I evaluated subsequent cisco reintroductions to Keuka Lake beginning in 

2018 to restore a native prey fish base, using acoustic telemetry and environmental 

DNA (eDNA) technologies to monitor outcomes. Results from a whole-lake acoustic 

telemetry experiment revealed high post-release mortality, information useful for 

adapting management practices to improve survival of stocked juvenile fish. In 

addition, a field assessment validated that eDNA-based methods successfully detected 

cold-water, pelagic schooling fish like cisco at coarse spatial scales. Yet, fine-scale 

comparisons revealed that deep lake currents, empirically measured with drifter 

devices, could transport cisco eDNA up to 3.3 km over 24 hours, complicating spatial 



 

interpretation. Finally, I believe insights from the past are important to inform the 

future: results from population modeling were used to assess the viability of future 

restoration scenarios. By incorporating in situ cisco life history parameters from 

Keuka Lake, outcomes from a population viability analysis indicated that cisco re-

establishment is unlikely with current stocking practices and lake conditions. Although 

Keuka Lake is oligotrophic, has abundant zooplankton and mysid (Mysis diluviana) 

prey, and lacks alewife, an ecosystem state more representative of historical cisco 

habitat, the still abundant top predator lake trout (Salvelinus namaycush) population 

that caused alewife collapse may similarly hinder cisco recovery. Combined, this 

integrated approach provided opportunity to evaluate fisheries restoration in Keuka 

Lake with application to cold-water lake ecosystems. 

 



 

iii 

BIOGRAPHICAL SKETCH 

Alex grew up in Cobleskill, New York and attended Cornell University where he 

earned a Bachelor’s in Natural Resources in 2013. A summer as an undergraduate 

field technician tracking brown trout in the Catskill Mountains led him to pursue a 

career in fisheries ecology and conservation since. After his undergraduate, Alex held 

various technician positions across the United States, including roles that monitored 

brook trout in the Adirondacks, steelhead in northern Idaho, and coastal cutthroat trout 

and coastal giant salamanders in the Cascades. In 2019, he earned a Master’s in 

Fisheries Science from Oregon State University, where he studied chinook salmon 

biology with Dr. Ivan Arismendi. He then worked at University of Arizona as a 

research specialist for aquatic species conservation across the desert southwest. Alex 

returned to Ithaca in 2021 for his Ph.D. in Natural Resources at Cornell University, 

with dissertation research focusing on the population ecology and ecosystem science 

of fish reintroductions under Drs. Suresh Sethi and Lars Rudstam.  

 

 

 

 

 

 

 

 



 

iv 

ACKNOWLEDGMENTS 

I especially thank my PhD committee over the past several years. First, I would like to 

thank my advisor Suresh Sethi, who saw the potential in me joining the Keuka Lake 

project and challenged me to be scientifically rigorous not only quantitatively, but also 

as a collaborator and writer. His guidance helped me appreciate that a thorough 

analysis is only as meaningful as the research questions it addresses and the broader 

context for fisheries conservation. Second, I thank Lars Rudstam, my co-advisor who 

shared with me the perspective of lakes as ecosystems. I started this program studying 

fish and now have a deeper appreciation for what they eat. Thank you to Evan Cooch 

who shared a wealth of knowledge and insights on population modeling that were 

crucial for this study. The lecture notes will be an invaluable resource throughout my 

career. Lastly, I thank Brad Hammers, who always encouraged me to think critically 

of the management implications for this work. I thoroughly enjoyed spending many 

days with Brad and crew from NYS Department of Environmental Conservation on 

Keuka Lake. I cannot think of a better setting or co-workers for graduate research.  

 

None of this research would be possible without the contributions of many other 

important project collaborators. Thank you to NYS Department of Environmental 

Conservation, US Geological Survey Tunison Laboratory of Aquatic Science, US Fish 

& Wildlife Service Northeast Fishery Center, and Keuka Lake Association for their 

long-term support with the Keuka Lake project. Specifically, I thank Web Pearsall, 

Dan Mulhall, Lew McCaffrey, Marc Chalupnicki, Jim McKenna, Jr., Meredith 

Bartron, Aaron Maloy, Chris Rees, Lauren Atkins, Rob Dintruff, Bob Lambert, Darryl 

Heckle, and many other agency personnel who have shared conversations and their 

knowledge of the Finger Lakes in the field. Whether processing acoustic telemetry 

files, configuring drifters to measure lake currents, or explaining genetic methods in 



 

v 

the lab, I could not have completed this research without their assistance. Thank you 

to Jim Watkins, Kayden Nasworthy, Sarah Lawhun, Joe Connolly, and several Shack 

student interns who helped me sample mysids and run acoustics at night in the Finger 

Lakes, and to Chris Marshall, Lucy Weisbeck, and Taylor Herne for encouraging me 

to join the Lake Guardian spring survey on Lake Superior. I thank the Otsego Lake 

Biological Field Station, especially Dan Stich, Holly Waterfield, and Matt Albright 

who shared data for our alewife comparisons between lakes and hosted me in 

Cooperstown. I have had many interesting discussions about food web modeling, and 

especially appreciate the guidance of Tom Stewart, as well as Kevin Cazelles, Kayla 

Hale, and the LIMCAT group. Thank you to the NY Chapter of the American 

Fisheries Society and International Association for Great Lakes Research. Finally, 

thank you to the funding sources who supported my graduate work including NYS 

Department of Environmental Conservation, NY Cooperative Fish and Wildlife 

Research Unit, and NY Sea Grant.  

 

At Cornell, I would like to thank the Department of Natural Resources and the 

Environment graduate students, members of the Shack-Sethi Lab, and Limnology Lab 

group for their always insightful discussions. Thank you to my lab mates Taylor 

Brown, Kimberly Fitzpatrick, and Kelsey Alvarez del Castillo for their feedback and 

input. I also thank staff from the Cornell Biological Field Station at Shackelton Point. 

Thank you to the Cornell Statistical Consulting Unit, specifically Andrew Siefert and 

Lynn Johnson. A highlight of graduate school was TAing NTRES 2100 Field Biology, 

thank you to Marc Goebel and students, who I have learned much from. In addition, I 

would like to thank many administrative staff for their help including Melanie Moss, 

Mandy Economos, Anne Marie Sheridan, Kelly Perkins, Tiffany Stauderman, 

Michelle Holeck, and Crystal Brown. Thank you to the many faculty and staff in the 



 

vi 

Department of Natural Resources and the Environment for their conversations and 

inspiration over the years. I especially thank Dan Decker, TJ Ross, Cliff Kraft, Rich 

Stedman, and Tim Martinson. 

 

Finally, I would like to thank my parents Rick and Tammy Koeberle, brothers and 

sister-in-law Brad, Nick and Melanie Koeberle, grandparents Frank and Beverly 

Koeberle and Emil and Marlene Vyskocil, my partner Zena Casteel, and 

Ithaca/Newfield-area friends for their unwavering support along the way.    



 

vii 

TABLE OF CONTENTS 

 

Biographical Sketch………………………………………………..……………….iii 

Acknowledgements………………………………………………………… ……... iv 

Table of Contents…………………………………………………………………...vii 

Introduction…………………….…………………….…………...………………. 1 

Chapter 1: Dissecting causes and consequences of alewife (Alosa  

pseudoharengus) collapse in lake ecosystems………………………………………13 

Chapter 2: Whole-lake acoustic telemetry to evaluate survival of stocked  

juvenile fish………………………………………………………………………… 84 

Chapter 3: How accurately does eDNA reflect the spatial distribution of  

cold‐water fish? Field validation from a temperate lake…………………………....120 

Chapter 4: Integrating acoustic telemetry and demographic modeling to inform  

cisco (Coregonus artedi) restoration in Keuka Lake, New York…………………...158 

A Chapter 1 Appendix………………………………………….………………… 204 

B Chapter 2 Appendix………………………………………….………………… 226 

C Chapter 3 Appendix………………………………………….……………….... 238 

D Chapter 4 Appendix………………………………………….………………… 262 

 



 

1 

INTRODUCTION 

 

Restoring native species is believed to improve food web function and increase 

ecosystem stability to disturbance (Holling 1973; Ives and Carpenter 2007; 

MacDougall et al. 2013). This is because coevolution of native species assemblages 

may foster efficient, diverse, and stable trophic structures (May 1972; Levin 1998; 

McCann 2000; Gellner et al. 2023). By contrast, introduced species are often 

generalists and may disrupt energetic pathways and trophic interaction strengths 

across a food web (Valiente-Banuet et al. 2014; Aslan 2019). While removal of 

introduced species is a common restoration strategy, this is particularly challenging in 

aquatic ecosystems where species often go undetected until they reach high levels of 

abundance (Ewel and Putz 2004; Havel et al. 2015). Restoring aquatic ecosystems 

containing suites of introduced species across multiple trophic levels to their previous 

state is therefore likely impossible (Odum 1969; Olden et al. 2004; Hobbs et al. 2009). 

Yet, population fluctuations of some introduced species may provide an opportunity to 

test the viability of restoring native species even within highly modified food webs. 

 

Successful species restoration is often limited by a failure to identify and address the 

original causes of species decline and a lack of system-specific monitoring data once 

restoration efforts have been implemented (Cochran-Biederman et al. 2015; Jachowski 

et al. 2016). My study provides a rare opportunity to evaluate lake ecosystem 

trajectories before and after the natural collapse of introduced alewife (Alosa 



 

2 

pseudoharengus) by the mid-2010s in two comparable lakes, Keuka Lake and Otsego 

Lake, New York, and during food web recovery via cisco (Coregonus artedi) 

reintroductions to Keuka Lake. Lake managers considered cisco extirpated from 

Keuka Lake since the mid-1990s, likely due to predation on cisco larvae by introduced 

prey fish. By the mid-2010s, managers observed a combination of improved water 

quality, increased zooplankton prey availability from predation release, and the mid-

trophic prey fish void from alewife collapse, ecosystem conditions more suitable for 

cisco (Bunnell et al. 2023). To stabilize the mid-trophic level and increase resilience of 

the lake trout (Salvelinus namaycush) fishery to prey fish population fluctuations, 

juvenile cisco were reintroduced from 2018 to present. This dissertation will address 

the research questions: (1) What are the ecosystem causes and consequences of 

alewife collapse, (2) can a native prey fish base be reintroduced through cisco 

stocking, and (3) what are plausible future ecosystem scenarios for Keuka Lake? 

 

Dissertation overview 

In Chapter 1, I present a comparative analysis of introduced alewife population 

collapse using long-term datasets from Keuka Lake (45 years, 1979-2024) and Otsego 

Lake (38 years, 1988-2025). Time series analysis shed light on the population 

dynamics of alewife in each lake, revealing an order of magnitude decline in Keuka 

Lake. In Otsego Lake, alewife declined from a peak of approximately 11,000 alewife 

per hectare to their extirpation. Evidence suggests that alewife collapse was driven by 

the combined effect of low zooplankton prey, high predation by top piscivores, and a 



 

3 

series of polar vortex winters. Although similar ecosystem consequences occurred in 

both lakes, the relative influence of contributing factors to alewife collapse was lake 

specific: Keuka Lake exhibited stronger top-down control by an abundant piscivore, 

while Otsego Lake showed more pronounced bottom-up limitations, coinciding with 

dreissenid mussel (Dreissena spp.) establishment (Kao et al. 2016). Reduced alewife 

zooplanktivory resulted in rapid biomass increases of the zooplankton Daphnia spp., 

in both lakes, and opossum shrimp mysids (Mysis diluviana), present only in Keuka 

Lake. Despite this, piscivory remained high, particularly from lake trout in Keuka 

Lake, which likely sustained itself on alternative prey like abundant mysids while 

continuing to suppress alewife recovery. The alewife collapse left a mid-trophic prey 

fish void in both lakes for over a decade, creating a rare opportunity to reintroduce a 

native prey fish to Keuka Lake. 

 

In Chapter 2, I demonstrate that advances in animal tracking technologies improve our 

ability to estimate important demographic parameters to evaluate fish reintroduction 

efforts. In Keuka Lake, a whole-lake acoustic telemetry array was deployed with small 

acoustic transmitter technology (including tags < 1.0 g) to track the fate of hatchery-

reared juvenile cisco across multiple cohorts of stocked fish (McMichael et al. 2010; 

McKenna et al. 2021). Using time-to-event survival modeling, I found that juvenile 

fish mortality was highest immediately after release (< 4 hours after stocking) with up 

to 75% of smaller, younger age-0 juvenile cisco perishing in a straight-to-death period, 

compared to 30% initial mortality of older, larger age-1+ juvenile cisco. High post-



 

4 

stocking mortality of juvenile cisco likely resulted from a combination of avian 

predation, lake trout predation, and physiological stress from handling, transport, and 

stocking. I found that the few older age-1+ juvenile cisco that survived the initial 

release and through an acclimation period entered a long-term period of higher 

survival, consistent with natural mortality and estimated as a 2% annual survival rate. 

These results provide fishery managers with insight into how hatchery fish size and 

age at release influence the success of native cisco restoration efforts.  

 

In Chapter 3, I assess the spatial accuracy of environmental DNA (eDNA)-based 

species detections, an emerging tool for monitoring species at low abundance. Results 

from this study validated eDNA detections at deep depths (12 m and 18 m) and at 

coarse spatial scales, consistent with tagged fish distributions inferred from the 

acoustic telemetry array in Keuka Lake. Yet, mismatches were observed at finer scales 

between eDNA and acoustic telemetry detections. I hypothesized that this was due to 

rapid transport of genetic material via lake currents. Empirical measurements of lake 

currents using drifters indicated cisco eDNA could drift up to 3.3 km at 12 m depths or 

1.5 km at 18 m depths over a 24 hr transport period. These findings highlight the need 

to account for eDNA transport and persistence in lake environments for accurate 

species occupancy assessments at fine scales. While eDNA-based detection 

technology is a promising tool for aquatic conservation, field-based validations such 

as this study are critical for accurate assessments.  

 



 

5 

Finally, in Chapter 4, I conducted a population viability analysis to assess the 

likelihood of successful cisco restoration in Keuka Lake. Post-stocking evaluation has 

long been recognized as a crucial component of fisheries management for identifying 

causes of success or failure of stocking programs (Cowx 1994). Adaptive management 

can improve long-term restoration success rates, but system-specific information on 

reintroduced populations is often rare and difficult to obtain (Armstrong and Seddon 

2008). In this study, I leveraged acoustic telemetry and a novel multi-stage survival 

model to estimate both hatchery and wild-equivalent juvenile cisco survival rates. I 

also used a long-term dataset to estimate in situ adult cisco natural mortality rates 

before their extirpation from Keuka Lake. I applied a population projection model 

parameterized with in situ cisco vital rates and calculated extinction probabilities for 

future population trajectories (Leslie 1948; Caswell 2001; Ellner and Fieberg 2003). I 

found that current fish stocking practices are unlikely to achieve management targets. 

Even under optimistic survival scenarios, population recovery remained unlikely. This 

was reflected in both low juvenile hatchery and wild-equivalent survival rates and 

adult mortality estimates. Importantly, this data-driven approach highlights that 

relying on life history parameters extrapolated from other systems would have led to 

false optimism for restoration success.  

 

Implications for fisheries conservation and management 

Long-term datasets helped address Question 1: What are the ecosystem causes and 

consequences of alewife collapse? In Keuka Lake, I found evidence for strong top-



 

6 

down control by the abundant lake trout population as the leading driver of alewife 

collapse, consistent with a trophic cascade (Carpenter et al. 1985). My results strongly 

suggest that heavy lake trout predation on adult alewife occurred first, followed by a 

compensatory increase of juvenile alewife before their population collapsed across age 

classes (Rudstam et al. 2011). This collapse coincided with two consecutive polar 

vortex winters (2014-2015), suggesting that the suppressed alewife population was 

susceptible to external perturbations (Dunlop and Riley 2013). Although declining 

primary productivity after introduced dreissenid mussel establishment in the mid-

1990s likely contributed to initial alewife declines, I lacked evidence for bottom-up 

control as a primary cause of collapse in Keuka Lake. This was apparent from a rapid 

increase in zooplankton and mysid biomass immediately after alewife collapse. While 

alewife declined by an order of magnitude, lake trout abundance remained high 

relative to other populations across North America (Hansen et al. 2021). I infer that 

continued strong predation pressure by lake trout has prevented alewife populations 

from recovering in Keuka Lake.  

 

Insights from whole-lake acoustic telemetry, implemented from the onset of cisco 

reintroductions, address Question 2: Can a native prey fish base be reintroduced 

through cisco stocking? Results from this study revealed prohibitively low survival of 

stocked juvenile fish. This is likely due to the combined effects of stocking-related 

stress and heavy predation immediately post-stocking. While contemporary lake 

conditions, e.g., oligotrophic waters, abundant zooplankton and mysid prey, and the 



 

7 

absence of alewife appear favorable for cisco, high predation by the lake trout 

population is likely a key mechanistic barrier to cisco recovery in Keuka Lake.  

 

Results from the survival analysis reveal multiple options for reducing juvenile cisco 

post-stocking mortality. To improve outcomes, an adaptive management 

recommendation is to stock fewer, but larger age-1+ juvenile cisco rather than age-0 

fish (Fonken et al. 2022). Higher post-stocking survival rates of age-1+ fish compared 

to age-0 fish are potentially due to predator gape limitation by lake trout. While 

current hatchery practices support up to 2,000 age-1+ fish per year, modeled 

optimistic stocking scenarios could achieve adult spawner targets by releasing 

approximately 4,200 age-1+ fish. Alternative stocking practices such as hatchery 

modifications to produce more wild-like age-1+ fish or at release such as night 

stocking, stocking with cold air temperatures, or use of net pen acclimation, could be 

explored to achieve these optimistic post-stocking survival rates (Brown and Day 

2002; Roberts et al. 2009; Cogliati et al. 2023). 

 

Nonetheless, population viability analysis revealed a low probability of successful 

cisco restoration in Keuka Lake, even under the most optimistic scenarios. I found that 

both hatchery-stocked and wild-equivalent juvenile survival rates would need to 

increase substantially to accrue minimum adult spawner management targets. 

Contemporary lake conditions, including a high like trout population, remain 

unfavorable for establishing a self-sustained cisco population. This is reflected in an 



 

8 

estimated adult survival rate of 51.4% in Keuka Lake, well below the approximately 

70% survival rate observed in self-sustaining coregonine populations elsewhere in 

North America and Europe.  

 

Combined, my dissertation offers insight into Question 3: What are plausible future 

ecosystem scenarios for Keuka Lake post-alewife collapse? Identifying future 

ecosystem scenarios is complex and depends on lake conditions, emerging introduced 

species, and predator-prey dynamics described throughout this dissertation. Future 

ecosystem states of Keuka Lake could range from an alewife-recovered, intermediate 

trophic-level dominated food web, to a scenario shaped by recently introduced species 

such as the top piscivore walleye (Sander vitreus, detected in 2016) and benthivore 

round goby (Neogobius melanostomus, detected by local anglers in fall 2025). Round 

goby may drastically alter energy pathways across lake food webs (Brooking et al. 

2022). Severe winters have been linked to alewife declines but also to strong cisco 

recruitment events (Fielder and McDonnell 2024; Brown et al. 2025). Potentially 

fewer cold winters could thus shift species composition from cold-water species like 

cisco to cool- and warm-water species including smallmouth bass (Micropterus 

dolomieu), yellow perch (Perca flavescens), walleye, and alewife. A cold-water, cisco-

restored lake ecosystem will likely require alternative stocking practices to improve 

juvenile cisco survival, continued suppression of alewife, and a reduction in top 

predator abundance.    



 

9 

REFERENCES 

Armstrong, D.P. and P.J. Seddon. 2008. Directions in reintroduction biology. Trends 

in Ecology & Evolution 23(1): pp.20-25. 

 

Aslan, C.E. 2019. Implications of non-native species for mutualistic network 

resistance and resilience. PloS One 14(6): p.e0217498. 

 

Brooking, T.E., Rudstam, L.G., Jackson, J.R., VanDeValk, A.J., Holeck, K.T., 

Hotaling, C.W. and J.E. Cooper. 2022. Effects of round goby on the benthic 

invertebrate community and on diets and growth of yellow perch and white 

perch in Oneida Lake, New York. Transactions of the American Fisheries 

Society 151(5): pp.641-654. 

 

Brown, C. and R.L. Day. 2002. The future of stock enhancements: lessons for 

hatchery practice from conservation biology. Fish and Fisheries 3(2):  

pp.79-94. 

 

Brown, T.A., Rudstam, L.G., Sethi, S.A., Ripple, P., Smith, J.B., Treska, T.J., Hessell, 

C., Olsen, E., He, J.X., Jonas, J.L., Rook, B.J, Blankenheim, J.E., Beech, 

S.J.H., Brown, E., Berglund, E.K., Cook, H.A., Dunlop, E.S., James, S.,  

Pothoven, S.A., Amidon, Z.J., Sweka, J.A., Carl, D.D., Hansen, S.P., Bunnell,  

D.B., Weidel, B.C., and A.E. Honsey. 2025. Reconstructing half a century of  

coregonine recruitment reveals species-specific dynamics and synchrony  

across the Laurentian Great Lakes. ICES Journal of Marine Science 82(2):  

fsae160. 

 

Bunnell, D.B., Ackiss, A.S., Alofs, K.M., Brant, C.O., Bronte, C.R., Claramunt, R.M., 

Dettmers, J.M., Honsey, A.E., Mandrak, N.E., Muir, A.M., and V.J. Santucci 

Jr. 2023. A science and management partnership to restore coregonine 

diversity to the Laurentian Great Lakes. Environmental Reviews 31(4):  

pp.716-738. 

 

Carpenter, S.R., Kitchell, J.F., and J.R. Hodgson. 1985. Cascading trophic interactions  

and lake productivity. BioScience, 35(10): pp.634-639. 

 

Caswell, H. 2001. Matrix Population models. 2nd ed. Sinauer Associates, Inc. 

Publishers, Sunderland, Massachusetts. 

 

Cochran‐Biederman, J.L., Wyman, K.E., French, W.E., and G.J. Loppnow. 2015. 

Identifying correlates of success and failure of native freshwater fish 

reintroductions. Conservation Biology 29(1): pp.175-186. 

 

Cogliati, K.M., Noakes, D.L., Khan, F., Sharpe, C.S., and C.B. Schreck, C.B. 2023. 

Producing wild fish phenotypes in hatchery-reared fish. Environmental Biology 

of Fishes 106(5): pp.1113-1134. 



 

10 

 

Cowx, I.G. 1994. Stocking strategies. Fisheries Management and Ecology 1(1):  

pp.15-30. 

 

Dunlop, E.S. and S.C. Riley. 2013. The contribution of cold winter temperatures to the 

2003 alewife population collapse in Lake Huron. Journal of Great Lakes  

Research 39(4): pp.682-689. 

 

Ellner, S.P. and J. Fieberg. 2003. Using PVA for management despite uncertainty: 

effects of habitat, hatcheries, and harvest on salmon. Ecology 84(6):  

pp.1359-1369. 

 

Ewel, J.J. and F.E. Putz. 2004. A place for alien species in ecosystem restoration. 

Frontiers in Ecology and the Environment 2(7): 354-360. 

 

Hansen, Michael J, Guy, C.S., Bronte, C.R., and N.A. Nate. 2021. Life history and  

population dynamics, in: The Lake Charr Salvelinus namaycush: Biology, 

Ecology, Distribution, and Management. Springer, pp.253-286. 

 

Havel, J.E., Kovalenko, K.E., Thomaz, S.M., Amalfitano, S., and L.B. Kats. 2015. 

Aquatic invasive species: challenges for the future. Hydrobiologia 750(1): 

pp.147-170. 

 

Hobbs, R.J., Higgs, E., and J.A. Harris. 2009. Novel ecosystems: implications for 

conservation and restoration. Trends in Ecology & Evolution 24(11):  

pp.599-605. 

 

Holling, C.S. 1973. Resilience and stability of ecological systems. Annual Review of 

Ecology and Systematics 4: pp.1-23. 

 

Ives, A.R. and S.R. Carpenter. 2007. Stability and diversity of ecosystems. Science, 

317(5834): pp.58-62. 

 

Fielder, D.G. and K.N McDonnell. 2024. Evaluation of potential factors affecting the 

success of Cisco re-introduction and re-establishment in Lake Huron. 

Ecological Modelling 496: p.110817. 

 

Fonken, D.R., Conner, M.M., Walsworth, T.E., and Thompson, P.D., 2022. Benefits  

of stocking fewer but larger individuals with implications for native fish  

recovery. Canadian Journal of Fisheries and Aquatic Sciences 80(3):  

pp.439-450. 

 

Gellner, G., McCann, K. and A. Hastings. 2023. Stable diverse food webs become 

more common when interactions are more biologically constrained. Proceedings of the 

National Academy of Sciences 120(31): p.e2212061120. 



 

11 

 

Jachowski D.S., Millspaugh J.J., Angermeier P.L., and R. Slotow (Editors). 2016. 

Reintroduction of fish and wildlife populations. University of California Press. 

 

Kao, Y.-C., Adlerstein, S.A., and E.S. Rutherford. 2016. Assessment of top-down and 

bottom-up controls on the collapse of alewives (Alosa pseudoharengus) in 

Lake Huron. Ecosystems 19: pp.803-831. 

 

Leslie, P.H. 1948. Some further notes on the use of matrices in population 

mathematics. Biometrika 35(3/4): pp.213-245. 

 

Levin, S.A. 1998. Ecosystems and the biosphere as complex adaptive systems. 

Ecosystems 1(5): 431-436. 

 

MacDougall, A., McCann, K., Gellner, G., and R. Turkington. 2013. Diversity loss 

with persistent human disturbance increases vulnerability to ecosystem 

collapse. Nature 494: pp.86-89. 

 

May, R.M. 1972. Will a large complex system be stable? Nature 238(5364): 413-414. 

 

McCann, K.S. 2000. The diversity–stability debate. Nature 405(6783): 228-233. 

 

McKenna Jr, J.E., Sethi, S.A., Scholten, G.M., Kraus, J. and M. Chalupnicki. 2021. 

Acoustic tag retention and tagging mortality of juvenile cisco Coregonus 

artedi. Journal of Great Lakes Research 47(3): pp.937-942. 

 

McMichael, G.A., Eppard, M.B., Carlson, T.J., Carter, J.A., Ebberts, B.D., Brown 

R.S., Weiland, M., Ploskey, G.R., Harnish, R.A., and Z.D. Deng. 2010. The 

juvenile salmon acoustic telemetry system: a new tool. Fisheries 35(1):  

pp.9-22. 

 

Odum, E.P. 1969. The strategy of ecosystem development: An understanding of 

ecological succession provides a basis for resolving man's conflict with nature. 

Science 164(3877): pp.262-270. 

 

Olden, J.D., Poff, N.L., Douglas, M.R., Douglas, M.E., and K.D. Fausch. 2004. 

Ecological and evolutionary consequences of biotic homogenization. Trends in 

Ecology & Evolution 19(1): pp.18-24. 

 

Roberts, L.J., Taylor, J., Gough, P.J., Forman, D.W. and C.G. de Leaniz. 2009. Night 

stocking facilitates nocturnal migration of hatchery‐reared Atlantic salmon, 

Salmo salar, smolts. Fisheries Management and Ecology 16(1): pp.10-13. 

 

Rudstam, L.G., Brooking, T.E., Krueger, S.D., Jackson, J.R., and L. Wetherbee. 2011.  

Analysis of compensatory responses in land‐locked alewives to walleye 



 

12 

predation: a tale of two lakes. Transactions of the American Fisheries Society 

140(6): pp.1587-1603. 

 

Valiente‐Banuet, A., Aizen, M.A., Alcántara, J.M., Arroyo, J., Cocucci, A., Galetti, 

M., García, M.B., García, D., Gómez, J.M., Jordano, P. and R. Medel. 2015. Beyond 

species loss: the extinction of ecological interactions in a changing world. Functional 

Ecology 29(3): 299-307. 

  



 

13 

CHAPTER 1 

DISSECTING CAUSES AND CONSEQENCES OF ALEWIFE (ALOSA 

PSEUDOHARENGUS) COLLAPSE IN LAKE ECOSYSTEMS 

 

Abstract:  

Planktivorous fish can significantly impact trophic dynamics in aquatic ecosystems. 

One example is alewife (Alosa pseudoharengus), which has modified food webs in 

lakes throughout its introduced range in North America. We conducted a comparative 

analysis of water quality, benthos, zooplankton, and fishes from Keuka Lake, New 

York (45 years, 1979-2024), a lake in the Lake Ontario basin, and from Otsego Lake, 

New York (38 years, 1988-2025), an Atlantic drainage lake. Our data driven approach 

presents evidence for the causes of alewife population collapse, which was likely the 

combined effect of low zooplankton prey, high predation by an apex piscivore, and a 

series of cold winters. While multiple pathways contributed to alewife collapse, the 

relative influence of contributing factors was system specific. Multiple datasets 

indicate stronger top-down control by abundant piscivores in Keuka Lake versus a 

more pronounced bottom-up limitation in Otsego Lake, coinciding with dreissenid 

mussel establishment. In Keuka Lake we could not detect dramatic changes in 

crustacean zooplankton and mysid shrimp Mysis diluviana (present only in Keuka 

Lake) associated with declining nutrients and the dreissenid mussel invasion, but such 

declines were apparent in Otsego Lake. Predation release from alewife collapse led to 

a rapid response in zooplankton communities, including increased Daphnia spp. 

biomass in both lakes, and in Keuka Lake, an increase in the mysid population to 

densities approaching 200 individuals/m2, higher than in most Laurentian Great Lakes. 

The number of top predator lake trout Salvelinus namaycush declined in both lakes but 

remained relatively high feeding on mysids in Keuka Lake and other fish species in 



 

14 

both lakes. Persistent piscivory likely continued to suppress alewife recovery in both 

lakes, overwhelming their well-known compensatory response. We found that 

predators switching to mysids allowed the predator population to remain high longer, 

as mysids increased to predation release from dominant alewife. By contrast, shifts to 

prey like small yellow perch Perca flavescens appear to take a longer time, perhaps 

because yellow perch do not overlap spatially with lake trout to the same extent as 

mysids. In addition, yellow perch has yet to fill the pelagic prey fish void, which 

remains vacant in both lakes over a decade later. Our study demonstrates that 

identifying the causes of collapse of a dominant planktivore in two lakes is important 

for understanding food web consequences.  

 

Introduction  

Understanding predator-prey dynamics and their cascading effects across trophic 

levels is fundamental to inform key drivers of structure and function of ecosystems. 

Two ecological factors that shape food web structure are predator driven top-down and 

resource limited bottom-up processes (Hairston et al., 1960; McQueen et al., 1986, 

1989; Maciej Gliwicz, 2002). Trophic cascades occur when top-down control by 

predators causes reciprocal changes in abundance or biomass across multiple lower 

trophic levels. In aquatic food webs this is exemplified by increased piscivore 

abundance → decreased planktivores → increased zooplankton → decreased 

phytoplankton (Paine, 1980; Carpenter et al., 1985; Carpenter and Kitchell, 1996). 

This concept has been extensively tested in freshwater systems with whole-lake 

biomanipulations and natural experiments (Shapiro et al., 1975; Jeppesen et al., 1990; 

Van Donk et al., 1990; Rudstam et al., 1993; Pace et al., 1999; Carpenter et al., 2001, 



 

15 

2011). For example, stocking piscivores as a biomanipulation management tool has 

been shown to successfully improve lake water clarity (Lathrop et al., 2002). Yet, 

identifying primary drivers of lake ecosystem change remains a challenge, as top-

down and bottom-up effects often co-occur and have complex interactions (Maciej 

Gliwicz, 2002; Kao et al., 2016; Rogers et al., 2024). By testing the trophic cascade 

hypothesis across long-term, multi-trophic level datasets, it may therefore be possible 

to identify the relative magnitude of specific drivers of ecosystem change.   

 

Mid-trophic level planktivorous fish are an important energy pathway between 

zooplankton and top predator fish (Cury et al., 2000). In aquatic systems dominated by 

an abundant, single-species forage fish base, planktivores can strongly influence food 

web structure through both top-down control of plankton and bottom-up limitation of 

piscivores (Rudstam et al., 1993; Lepak et al., 2025). Their population dynamics, 

however, are often characterized by episodic recruitment and boom-and-bust cycles, 

making them susceptible to rapid fluctuations and collapse, the causes of which are 

not always apparent (Stein et al., 1995; Fauchild et al., 2011; Buren et al., 2019; Peck 

et al., 2021; Brown et al., 2025). Such instability can have cascading ecological effects 

and significant economic consequences for commercial and recreational fisheries of 

top predators (Cury et al., 2000; Bunnell et al., 2014; Essington et al., 2015). 

 

Alewife (Alosa pseudoharengus) are an excellent study species for understanding the 

role of pelagic prey fish in lake ecosystems due to their life history plasticity and 

ecological impacts. These shads are native to the North American Atlantic coast with 



 

16 

both anadromous populations and landlocked populations in freshwater lakes. High 

trait plasticity and varied foraging ability have enabled their widespread colonization 

(Palkovacs et al., 2007; Post et al., 2008). This includes the Laurentian Great Lakes 

basin (hereafter, Great Lakes), where alewife spread in the 19th century via intentional 

introductions and bait bucket releases (Smith, 1985). In the early 20th century, alewife 

was considered a better prey species than native coregonines (Coregonus spp.) for 

salmonid fisheries and stocking was encouraged by fisheries managers (Pritchard, 

1929; Odell, 1934). Now widely established, alewife is a dominant planktivore and 

prey fish for top predators, often outcompeting native coregonid (Coregonus spp.) 

populations (Smith 1970; Madenjian et al. 2008). Despite supporting highly 

productive fisheries, including introduced Chinook salmon (Oncorhynchus 

tshawytscha) in the Great Lakes (Jones et al., 1993; Warner et al., 2008; Fitzpatrick et 

al., 2023), alewife populations fluctuate widely in both their anadromous and 

landlocked ranges (O’Gorman and Stewart, 1999; Davis and Schultz, 2009), posing 

risks to long-term fisheries sustainability.  

 

While introduced alewife form a trophic link between zooplankton and top predator 

fish, their establishment often restructures aquatic food webs (Smith, 1970; O’Gorman 

and Stewart, 1999; Madenjian et al., 2008). The arrival of alewife is associated with 

declines in walleye (Sander vitreus) and yellow perch (Perca flavescens) recruitment 

through predation on percid larvae (Brandt et al., 1987; Brooking et al., 1998). 

Alewife high in the enzyme thiaminase are a main cause for thiamine deficiencies in 



 

17 

salmonids (Fisher et al., 1996; Fitzsimons et al., 2005), including their likely 

contribution to Atlantic salmon (Salmo salar) extirpation from Lake Ontario 

(Madenjian et al., 2008). Their competitive presence is also linked to coregonine prey 

fish declines throughout the Great Lakes, prompting contemporary basin-wide 

restoration efforts (Smith, 1970; Bunnell et al., 2023). At lower trophic levels, alewife 

exhibit strong top-down control on zooplankton in their native range (Brooks and 

Dodson, 1965; Post and Walters, 2009) and in the lakes they invaded (Stewart et al., 

2009; Wang et al., 2010; Lesser et al., 2024). Alewife are visual planktivores and can 

significantly modify zooplankton species composition. High planktivory rates by 

alewife eliminate large zooplankton allowing smaller zooplankton to become 

dominant, the size-efficiency hypothesis proposed by Brooks and Dodson (1965). This 

is well-supported by alewife studies across freshwater lakes (Wells, 1970; Stewart and 

Binkowski, 1986; Post et al., 2008; Dougherty et al., 2025). Because of their ability to 

depress Daphnia zooplankton species, the arrival of alewife is often accompanied by 

increases in phytoplankton blooms and decreases in water clarity (Harman et al., 2002, 

Wang et al., 2010).  

 

Consequently, managers have sought to reduce landlocked alewife via piscivore 

stocking but with mixed success (Rudstam et al., 2011). This is due to a strong 

compensatory density dependence response (Myers et al., 1995; Rose et al., 2002), 

where declining adult abundance reduces cannibalism of young-of-year alewife. This 

release allows increased alewife growth rates, more larvae per adult, and higher larval 



 

18 

survival due to decreased competition with and predation by adults (Rudstam et al., 

2011). Combined with their high fecundity rates (Norden, 1967), these life history 

characteristics make biomanipulation strategies to reduce or eliminate introduced 

alewife populations a challenge.  

 

Natural experiments provide opportunity to understand fish population drivers of 

ecosystem change. For example, alewife collapsed in Lake Huron between 2003-2004, 

coinciding with lower primary productivity, decreased zooplankton and mysid shrimp 

(Mysis diluviana) abundance (Reavie et al., 2014; Holda et al., 2023), and abundant 

top predator populations (Johnson et al., 2010). Cold winters, poor condition due to 

low food availability, and increased salmon predation have all been proposed and 

debated to have contributed to alewife collapse and prevented their recovery (Nalepa 

et al., 2007; Barbiero et al., 2011; Dunlop and Riley, 2013; Bunnell et al., 2014; He et 

al., 2015). Alewife biomass in Lake Huron declined by over 90%, however, attributing 

a cause from top-down control by stocked Chinook salmon (Oncorhynchus 

tshawytscha) and other piscivores or bottom-up control from dreissenid mussels and 

increased water clarity remained controversial (Madenjian et al., 2013; Kao et al., 

2016; also see Riley and Dunlop (2016) and Bence et al. (2016) in response). 

Nonetheless, the causes, which were concurrent in Lake Huron, and resulting 

consequences of alewife collapse are generally not well understood. Here we 

investigate whether simplified food webs and extensive monitoring across trophic 



 

19 

levels in inland lakes allows us to identify the relative strengths of top-down and 

bottom-up control where alewife-dominated prey fish populations have declined.  

 

In this study, we leveraged datasets from two lakes where landlocked alewife 

populations independently collapsed: Keuka Lake, New York, USA (4,688 ha surface 

area, 57 m maximum depth, Lake Ontario basin drainage) and Otsego Lake, New 

York, USA (1,637 ha surface area, 50.5 m maximum depth, Atlantic coast drainage). 

Alewife likely arrived in Keuka Lake by the early 1900s (Odell, 1934; Smith, 1985), 

while alewife were illegally introduced into Otsego Lake in 1986 and became 

abundant by the 1990s (Harman et al., 2002). The lakes function as natural replicates 

for this study given their similar morphology, oligotrophic status, and species 

assemblages (Bloomfield, 1978a, 1978b). We first quantified the timing and 

magnitude of alewife declines in both lakes. Then, we used long-term datasets to 

evaluate three hypothesized drivers of population collapse at a whole-ecosystem level: 

Prey fish collapse was caused by (H1) bottom-up control via decreased lower trophic 

level productivity, (H2) top-down predation by abundant piscivores, or (H3) abiotic 

stressors such as severe winters. Our selection of hypotheses was informed by studies 

of alewife collapse in Lake Huron. While these three hypotheses are not mutually 

exclusive, we believe our comparative analysis allows us to test their relative 

importance.  

 



 

20 

Each hypothesis generates different predictions for the consequences of alewife 

collapse. If alewife collapse was driven by declining productivity (H1), as in Lake 

Huron (Nalepa et al., 2007; Barbiero et al., 2011; Bunnell et al. 2014), we predicted 

oligotrophication due to reduced nutrient inputs and no response in zooplankton 

populations despite a compensatory release from alewife predation. If driven by top-

down control (H2), as in Lake Huron (He et al., 2015), we predicted increased 

predator abundance would indirectly release zooplankton and reduce phytoplankton 

(measured via chlorophyll-a). This would follow a trophic cascade consistent with 

food web theory (Carpenter et al., 1985; Gliwicz, 2002). Abiotic effects (H3), such as 

severe winters (Dunlop and Riley, 2013), were expected to produce similar top-down 

effects on lower trophic levels but on shorter timescales without prior increases in 

piscivores. Across hypotheses, based on observations of top predators in Lake Huron 

(Kao et al., 2016), we expected declines in lake trout growth and abundance following 

alewife collapse and potential increase in the predatory invertebrate Mysis diluviana 

(present only in Keuka Lake) compensating for reduced zooplanktivory by alewife. 

We anticipate our comparative, data-driven approach offers insights into how the 

causes of prey fish collapse affect the food web consequences in lake ecosystems.   

 

Methods 

We compiled long-term datasets of water quality, zooplankton, benthic organisms, 

fish, and ice cover to investigate the relative influence of top-down, bottom-up, and 

external drivers of alewife collapse. Here, we briefly describe survey methodology and 



 

21 

data analysis conducted across trophic levels in Keuka Lake and Otsego Lake. 

Detailed field sampling procedures are available in Appendices S1 and S2.  

 

Primary productivity 

To evaluate hypothesized bottom-up drivers and predicted top-down effects of alewife 

collapse, we analyzed long-term water quality datasets collected monthly or biweekly 

during summer by Keuka Lake Association and SUNY Oneonta Otsego Lake 

Biological Field Station (OBFS). Water quality metrics included Secchi disk depth 

(m), an indicator of water quality, as well as concentrations of chlorophyll a (chl a, 

µg/L), total phosphorus (TP, µg P/L), and nitrate (Keuka Lake NOx, Otsego Lake NO3, 

mg N/L) available for both lakes. Since 1992, a volunteer network through Keuka 

Lake Association conducted monthly seasonal water quality sampling at ten 

standardized locations equally spaced across Keuka Lake. At each location, physical 

characteristics were measured, and water samples were collected with a Kemmerer 

bottle for nutrient analysis at a shallow depth (1 m, epilimnion) and a deep depth (30 

m, hypolimnion). All water samples collections followed standard state-wide protocol 

(see Appendix S1: Section S1.1 for more information) and laboratory analyses for 

nutrient concentrations were conducted by the Finger Lakes Institute (Hobart and 

William Smith Colleges, Geneva, New York, USA). In Otsego Lake, physical 

measurements and water quality samples were collected seasonally since 1988 

following similar protocols and samples were analyzed by trained personnel at OBFS. 

By contrast to a lake-wide sampling approach, all water quality data from Otsego Lake 



 

22 

were collected at a single location at the deepest point of the lake (50.5 m).  Nutrient 

samples were collected at 4 m intervals from 0-48 m depths and chlorophyll a values 

were collected from a continuous 0-20 m integrated sample. We therefore took 

average nutrient measurements for depth categories consistent with Keuka Lake, 

epilimnion (≤ 12 m depth) and hypolimnion (> 12 m depth). Lastly, we conducted 

sensitivity analysis to identify potential outliers from each dataset (Appendix S1: 

Section S1.1). 

 

Zooplankton   

We analyzed long-term zooplankton datasets to identify changes in species 

composition and abundance. Introduced zooplankton species include predatory 

fishhook water flea (Cercopagis pengoi) detected in Keuka Lake in 1999. In Keuka 

Lake, summer epilimnetic zooplankton samples were collected monthly since 1992 at 

four standard sites (east, west, and south arms and confluence region of Y-shaped 

Keuka Lake). All tows used a 153 µm mesh 0.5 m in diameter plankton net towed 

vertically during the day. All Keuka Lake zooplankton counts were completed at the 

Cornell Biological Field Station (CBFS). At least 100 individuals in each sample were 

identified to species or higher taxonomic groups and measured. Zooplankton biomass 

(μg/L) was calculated using standard length-weight regressions from the Great Lakes 

(U.S. Environmental Protection Agency Great Lakes Standard Operating Procedure 

2017; hereafter EPA-GLNPO). We applied mean species-specific weights to Keuka 

Lake zooplankton data from 1998-2023 to 1992-1997 due to known issues with 



 

23 

measuring tablets in early years. In Otsego Lake, summer epilimnetic zooplankton 

samples were collected from 2002 to 2024 approximately every three weeks at the 

same station for water quality described above. All tows used a 63 µm mesh 0.5 m 

diameter (2002-2004) or 0.2 m diameter (2005-2024) plankton net towed vertically 

from 12 m to 0 m depths during the day with a flow meter (General Oceanics, Inc.) to 

quantify sampled water volume. Zooplankton counts were conducted by trained staff 

at OBFS, following similar procedures as CBFS. This included counting at least 100 

individuals per sample or counting multiple cells until at least 100 specimens were 

achieved with 1 ml of water incrementally viewed. The number of crustaceans counted 

on average was < 100 individuals (range 25-305 individuals), including rotifers. In 

2024, paired epilimnetic and whole-water column samples were collected by New 

York State Department of Environmental Conservation (NYSDEC) to compare 

species composition in the epilimnion and hypolimnion in Keuka Lake. For 

consistency, we restricted the time series of water quality and zooplankton data from 

both lakes to epilimnetic tows and the summer sampling period June-September as an 

index of temporal change.   

 

Benthos 

Introduced populations of dreissenid mussels were established in both lakes during 

each time series. Zebra mussels (Dreissena polymorpha) were first detected in Keuka 

Lake in 1994 and Otsego Lake in 2007, and quagga mussels (Dreissena rostriformis 

bugensis) were first detected in Keuka Lake in 2008 and Otsego Lake in 2020. Benthic 



 

24 

data were more robust for Keuka Lake, where we compared three surveys to assess 

changes in invertebrate communities and sediment composition. Species of interest 

include both the introduced dreissenid mussels and Diporeia, a native burrowing 

amphipod that has declined throughout the Great Lakes but serves as an important 

trophic link between phytoplankton and fish (Watkins et al., 2012). Benthic surveys 

were conducted in summer 2010 (Watkins et al., 2012) at six sites in the northeastern 

arm (mean depth = 22.5 m; range: 10-35 m) with three replicates per sampling station; 

in March-April 2019 at 20 deep-water sites (mean depth = 43 m; range: 16-57 m) with 

one replicate per station; and August 2025 replicating 2010 survey protocol at six 

comparable sites and one additional deep site (mean depth = 30.6 m, range: 9-57 m). 

Across Keuka Lake surveys, all field samples were collected with petite Ponar grabs 

(area sampled per grab = 0.023 m2) and were sorted, identified, and enumerated by 

trained personnel from CBFS, SUNY Buffalo, or NYSDEC to estimate species density 

(converted to number of individuals/m2). In Otsego Lake, benthic survey data were 

collected to monitor dreissenid populations starting in 2021 after quagga mussels were 

detected. OBFS staff collected samples from two transects across the lake from 2 m to 

47 m depths and 2 m to 20 m depths using an Ekman dredge. No Diporeia were 

collected but other amphipods (Gammarus or Hyalella) were observed. The lack of 

glacial relict species Diporeia and Mysis in Otsego Lake is due to watershed 

differences from Keuka Lake in the Finger Lakes region (Fairchild, 1899, 1926; 

Balcer et al., 1984; Larson and Schaetzl, 2001; Dermott, et al., 2006).  

 



 

25 

Mysis  

Mysids, present in Keuka Lake but not Otsego Lake, are an important trophic link 

between the benthos and mid to upper trophic level fish through diel vertical 

migration. We estimated mysid abundance in Keuka Lake with a lake-wide, whole-

water column survey. In October 2023, we sampled 10 sites (mean depth 43 m, range 

26-52 m) equally spaced across the lake with a 500 µm mesh 1.0 m in diameter mysid 

net towed vertically at night. The survey commenced 1 hour after sunset and all 

sampling was conducted with red light on deck to avoid repelling light-sensitive 

mysids (Holda et al., 2019). Mysid samples were processed (counted, standard length 

measured, and sexed) and biomass calculated using standard length-weight regressions 

(Johannsson et al. 2011). All mysids were identified and all individuals up to 50 were 

measured from each net tow. Additional samples from June 2024 were collected at 

four deep sites with the same protocol (mean depth = 47 m, range 30-57 m). This 

survey was used to estimate mysid growth rates and evaluate age structure. All 

individuals from one tow at each site were processed from the June 2024 survey.  

 

Fish  

We examined fish communities in both lakes to evaluate hypotheses for alewife 

collapse. In Keuka Lake, cisco (Coregonus artedi) were the historical dominant 

planktivore, but were considered by NYSDEC to be extirpated by the mid-1990s. 

Lake whitefish (Coregonus clupeaformis), caught in low abundances, were last 

detected in the 1980s. Lake trout (Salvelinus namaycush) are the top piscivores, 



 

26 

reproduce naturally, and appear not to be affected by thiaminase deficiency 

(Fitzsimons et al., 2005). Hatchery stocking of brown trout (Salmo trutta) and 

landlocked Atlantic salmon ended in 2016, and NYSDEC implemented cisco 

reintroductions in 2018 towards restoring a native prey fish base (Koeberle et al., 

2023). Piscivorous walleye (Sander vitreus) also appeared in 2016. In Otsego Lake, 

abundant lake whitefish (Coregonus clupeaformis) once supported a local commercial 

fishery, and efforts in the early 1900s to increase their population with propagation 

from Lake Ontario inadvertently introduced cisco (Harman et al., 2002). Lake 

whitefish and cisco are still present. In contrast to Keuka Lake, NYSDEC managers 

augment the lake trout population with annual hatchery stocking of 2,500 to 10,000 

fish (Appendix S1: Table S1). From 2000-2014, an average of over 40,000 age-0 

walleye were stocked per year to reduce the alewife population (Appendix S1: Table 

S1). Walleye stocking ceased following the alewife collapse and establishment of 

natural walleye reproduction.  

 

A suite of sampling gears was deployed to evaluate fish community dynamics over 

time. In Keuka Lake, the primary long-term survey method for fish is a lake-wide 

standard gang gill net survey (hereafter, cold-water survey) by NYSDEC since the 

1970s. Generally, this survey is conducted every three years during the summer with 

nets set along the lake bottom at 32 deep sites. NYSDEC conducted a similar gill net 

survey on Otsego Lake since 1994, but every two years at six sites lake wide. Since 

this survey targets lake trout, forage net surveys were paired with hydroacoustic 



 

27 

surveys (see section below) in both lakes (1996-2013 in Otsego Lake, 2000-2023 in 

Keuka Lake) to assess the prey fish community. In Keuka Lake, additional sampling 

gear included a lake-wide forage gill net assessment, implemented in 2019 after cisco 

reintroductions, and a lake-wide percid gill net assessment implemented in 2022. In 

Otsego Lake, additional sampling gear included a similar percid gill net assessment, 

implemented in 2003 after walleye stocking, and a long-term trap net survey since 

1988 to monitor spawning adult alewife shortly after their introduction. Full 

specifications for each netting survey are available in Appendix S1: Section S1.3.  

 

To analyze changes in fish population size, structure, and species composition, we 

calculated catch rates (number of fish/net) as an index of fish abundance. Nets are set 

overnight for standardized surveys (e.g., cold-water, forage, percid, and trap net 

surveys) and we assume catch rates of respective gear types are comparable between 

study lakes. Fish data collected during netting surveys included counts and sizes, 

which we used to evaluate alewife and lake trout growth rates and mean relative 

weight across the time series. Age data were available across the Keuka Lake time 

series for lake trout and a subset of years for alewife. All fish were aged from scales 

by NYSDEC. When available, we also used age information to calculate lake trout and 

alewife mortality rates across the time series using the Chapman-Robson maximum 

likelihood estimator (Robson and Chapman, 1961). Finally, we analyzed lake trout 

stomach contents collected from the cold-water survey in Keuka Lake. Fish age data 

and lake trout stomach content data were not available for Otsego Lake. 



 

28 

 

Hydroacoustic analysis   

While net surveys offer valuable information on fish population dynamics, we 

complemented long-term net assessments with hydroacoustic methods to generate 

quantitative estimates of predator and prey fish abundance (density; fish/ha). Keuka 

Lake was surveyed with hydroacoustics in years 2000, 2007, 2011, 2016 and 2023. 

Otsego Lake was surveyed more frequently from 1996-2013. The equipment and 

frequency used differed across years in Keuka Lake, with year 2000 surveyed with a 

Simrad EY-500 70kHz unit, and 2007-2023 with a 120 kHz Biosonics unit. In Otsego 

Lake, surveys were conducted with similar hydroacoustic equipment; prior to 2007 

surveys were conducted with equipment from CBFS and OBFS (Simrad EY500 

70kHz unit), while after 2007 surveys were conducted by OBFS (123 kHz Biosonics 

DTX). Alewife surveys with 70 and 120 kHz produce similar fish densities (Rudstam 

et al., 2009; Brooking and Rudstam, 2009; Rudstam et al., 2011). Fall surveys were 

conducted in September and October and field methodologies were similar between 

lakes. This included conducting multiple transects across each lake at night. Surveys 

typically took 3-5 h and started 1 h after sunset with the transducer located about 0.5 

m below the surface either mounted on a pole next to a small boat or on a tow body 

located to the side of the boat. Field collections followed the Great Lakes Standard 

Operating Procedure for Fisheries Acoustic Surveys (hereafter, GLSOP-FAS; Parker-

Stetter et al. 2009) with pulse lengths of 0.4 to 0.6 ms. In addition to fall surveys, 

spring surveys were also conducted in Otsego Lake. Data analysis for both lakes 



 

29 

followed the GLSOP-FAS for single echo detection, with targets larger than -62 dB to 

-60 dB (depending on season) considered fish. In Keuka Lake, target strength 

divisions were -60 dB, -50 dB, and -40 dB in survey years 2007-2023. In Otsego Lake, 

targets between -61 dB to -55 dB and -41dB to -35 dB were assumed to be alewife and 

targets -41 dB to -35 dB and larger were assumed to be predatory fish (Warner et al., 

2002). In both lakes, a surface exclusion zone was established from the surface to 2 m 

to exclude data from surface noise by the transducer. Hydroacoustic equipment 

specifications, survey protocol, and fish data analysis procedures are detailed in 

Appendix S1: Section S1.4. Acoustic settings used during surveys are provided in 

Appendix S1: Table S2, and representative echograms are shown in Appendix S1: 

Figure S1. Mysid densities in Keuka Lake were also estimated with hydroacoustics, 

following procedures by Rudstam et al. (2008a, 2008b; see Appendix S1: Section 1.4). 

 

Climate analysis  

Alewife evolved in the Atlantic Ocean with the ability to migrate southward to warmer 

waters in winter; however, landlocked lakes often present harsh winter conditions 

associated with increased mortality (Ridgway et al., 1990; O’Gorman and Stewart, 

1999; Dunlop and Riley, 2013). While the physiological mechanisms underlying this 

response remain contested, proposed factors include immunosuppression, loss of 

homeoviscous adaptation, and poor condition coupled with small juvenile body size 

due to limited food availability leading to recruitment failure (Snyder and Hennessey, 

2003; Lepak and Kraft, 2008; O’Gorman et al., 2004; Dunlop and Riley, 2013). 



 

30 

Conversely, increased spring-summer degree days are associated with increased year-

class strength (Warren et al., 2024).  

 

A water treatment plant on Keuka Lake provided intake temperature data, but only 

from shallow depths (approximately 1.5 m depth and 6 m offshore) not reflective of 

alewife winter habitat. Therefore, we assessed winter growing degree days for Keuka 

Lake and long-term ice cover as proxies for winter severity, following climate analysis 

for alewife collapse by Dunlop and Riley (2013). Daily maximum and minimum 

temperature data from 1998-2025 were obtained from Penn Yan airport (Lat 42.64306, 

Lon -77.04944), < 2 km from Keuka Lake 

(<https://www.weather.gov/wrh/timeseries?site=kpeo> accessed 17 November 2025). 

We then calculated winter growing degree days (1-December to 31-March) as: 

𝐺𝐷𝐷𝑤 =  ∑ [0.5(𝑇𝑚𝑎𝑥,𝑑 + 𝑇𝑚𝑖𝑛,𝑑) − 𝑇𝑏𝑎𝑠𝑒]𝑤 , where 𝑇𝑚𝑎𝑥,𝑑 is maximum daily air 

temperature, 𝑇𝑚𝑖𝑛,𝑑 is minimum daily air temperature, and 𝑇𝑏𝑎𝑠𝑒 is a 10˚C base 

temperature (Dunlop and Riley, 2013). We also assumed a positive relationship 

between ice cover and winter severity, e.g., increased ice cover corresponds to harsher 

winter conditions. OBFS measures annual ice cover duration (number of days) in 

Otsego Lake, with records of ice break-up dating back to the winter of 1842-1843. Ice 

cover measurements were not available for Keuka Lake. Consequently, we evaluated 

comparable regional lake ice cover datasets: maximum ice cover extent (percentage) 

for Lake Ontario available from NOAA Great Lakes Environmental Research 

Laboratory (<https://www.glerl.noaa.gov/data/ice/#historical> accessed 17 November 

https://www.weather.gov/wrh/timeseries?site=kpeo%20
https://www.glerl.noaa.gov/data/ice/#historical


 

31 

2025) and ice cover duration for inland Lake Ontario basin lakes Cazenovia Lake 

(maximum depth = 14 m) and Oneida Lake (maximum depth = 17 m) described by 

Sharma et al. (2021; 2022). All ice cover datasets were restricted to 1970-2025.  

 

Statistical analysis 

All statistical analyses were conducted in program R version 4.5.0 (R Core Team, 

2025). Alewife abundance indices were derived from time-series netting (alewife 

catch) and hydroacoustic (alewife density) datasets to quantify both the timing and 

magnitude of alewife collapse within each study system. To assess ecosystem 

responses, we overlaid time series datasets with identified alewife decline and collapse 

periods to evaluate temporal synchrony between alewife population dynamics and 

concurrent changes in other trophic levels.   

 

We evaluated the relative strength of our hypotheses of bottom-up and top-down 

drivers of alewife collapse using visual inspection of deviations from long-term means 

and statistical tests. First, changes in associated biota were assessed by aggregating 

data into pre- and post-alewife collapse periods and applying Welch’s two-sample t-

tests, or for multiple groups, analysis of variance (ANOVA) with post-hoc Tukey’s 

HSD tests. Next, we fit univariate generalized additive models (GAMs) to evaluate 

trends before and after alewife collapse in water quality and predator fish metrics (e.g., 

mean relative weight, survival). We implemented GAMs via the package ‘mgcv’ in 

program R (Wood, 2017) with restricted maximum likelihood (REML) and restricted 



 

32 

basis dimensions (typically k = 4) to reduce overfitting. For all statistical tests, we 

considered p < 0.05 as significant and p < 0.10 as marginally significant. Finally, we 

evaluated winter severity as a driver of alewife collapse by identifying deviations (≥ 1 

SD) in winter growing degree days and ice cover duration or extent from long-term 

means during alewife collapse periods. 

 

Results 

Timing and magnitude of alewife collapse  

A period of alewife decline preceded population collapse, which we identified from 

net surveys as 2011-2015 in Keuka Lake and 2007-2010 in Otsego Lake (Figure 1). In 

Keuka Lake, the period of alewife decline began in 1997 and continued up to 2011 

(Figures 1a, c). In Otsego Lake, we inferred that alewife decline occurred over a 

shorter period from 2003-2007 (Figure 1b).  

 

Hydroacoustic surveys revealed a one to two order of magnitude decline in abundance 

as alewife populations collapsed. In Keuka Lake, alewife densities estimated by 

hydroacoustics declined by 95% from 1,302 fish/ha (2000-2011) to 70 fish/ha (2016-

2023; Figure 1a, Appendix S1: Table S3). In Otsego Lake, alewife densities declined > 

99% from ≈ 11,000 fish/ha in fall 2002 to 25 fish/ha in spring 2013 (Figure 1b; 

Appendix S1: Figure S2). These declines in hydroacoustic-estimated densities were 

also reflected across netting surveys. In Keuka Lake, alewife catch in forage nets 



 

33 

decreased by 98% between 2000-2007 and 2016-2022 (Figure 1e). In Otsego Lake, 

alewife catch in the trap net survey declined by 76% from 1997 to the early 2000s, 

with no alewife caught after 2011 (Figure 1h). While alewife hydroacoustic densities 

and catch rates in Keuka Lake declined steadily prior to collapse, alewife in Otsego 

Lake exhibited boom-and-bust recruitment during their decline, until collapse resulted 

in extirpation after 2013 (Figure 1b).  

 

In both lakes, we hypothesized that adult decline preceded the alewife population 

collapse, followed by a transient increase in sub-adult stages as recruitment briefly 

intensified before their collapse. Alewife length frequency distributions shifted from 

unimodal to bimodal during their decline, indicating increased recruitment of young-

of-year fish and increased growth rates (Appendix S1: Figures S3, S4). In Keuka 

Lake, post-alewife collapse distributions returned to unimodal, dominated by fast-

growing juveniles with few surviving adults. Age data, available only for Keuka Lake, 

showed absence of age-3+ alewife during the decline and high adult annual natural 

mortality across the time series (65-70% mortality rate; Appendix S1: Figure S5).  

 

Indicators of primary productivity 

Water clarity improved with time in both lakes. Secchi depth nearly doubled in Keuka 

Lake (Figure 2a), where the trend was significant, increasing from 4.6 m before 

alewife decline (1992-1994) to 7.9 m post-alewife collapse (2022-2023). Otsego Lake 

(Figure 2b) showed an increase following zebra mussel arrival in 2007, with mean 



 

34 

Secchi depth increasing from 3.9 m before alewife decline (1988, 1992-1993) to 6.5 m 

post-alewife collapse (2022-2024).  

 

Long-term changes in mean summer nutrient concentrations varied between lakes with 

stronger evidence of bottom-up effects occurring in Otsego Lake (Figure 2). In Otsego 

Lake, epilimnetic total phosphorus and chlorophyll a concentrations decreased 

significantly from before alewife decline to after alewife collapse (Figure 2f, h; TP by 

44% and chl a by 79%), while nitrate increased after 2007, coinciding with zebra 

mussel arrival and alewife decline (Figure 2d). In Keuka Lake, nitrate remained 

relatively stable (Figure 2c) while total phosphorus declined by 52% from pre-alewife 

decline (1992-1994) to alewife decline (2001-2003) and chlorophyll a declined by 

74% from a peak in 1994-1996 to a minimum during alewife collapse (2011-2013). In 

contrast to Otsego Lake, total phosphorus and chlorophyll a concentrations increased 

after alewife collapse in Keuka Lake (2016-2023; Figure 2e, g). Despite these 

dynamics, both lakes exhibited similar long-term mean concentrations of total 

phosphorus and chlorophyll a, while Otsego Lake maintained higher mean nitrate 

levels (Figure 2). In both lakes, hypolimnetic nutrient concentrations mirror 

epilimnetic patterns (Appendix S1: Figure S7). 

 

Zooplankton  

Long-term zooplankton surveys in Keuka Lake (30+ years) and Otsego Lake (20+ 

years) revealed a clear rapid response to predation release after alewife collapse 



 

35 

(Figure 3). A sharp increase in Keuka Lake zooplankton biomass during 2014 aligned 

with our inferred timing of alewife collapse and was inconsistent with expectations for 

a bottom-up effect (Figure 3a). In Keuka Lake, total epilimnetic zooplankton biomass 

was highest in the early 1990s at the start of the time series, then declined during 

alewife dominance, and increased (marginally significantly) by 69% from pre- to post-

alewife collapse (1998-2013; 2014-2024). Zooplankton species composition changed 

dramatically. Daphnia biomass increased (significantly) by 235%, while Bosmina and 

cyclopoid copepod biomass declined after the 1990s and remained low. By contrast, in 

Otsego Lake, we found an overall decline in zooplankton biomass through the period 

of alewife decline and that continued after their collapse (Figure 3b), consistent with 

declining lower trophic primary productivity, and with more limited changes in 

zooplankton community structure. 

 

Size distribution analyses in both lakes showed a shift toward larger-bodied 

zooplankton post-alewife collapse (Figure 4a, b), consistent with alewife-controlled 

size structure described by Brooks and Dodson (1965). Daphnia biomass increases 

were driven by size, not density in Keuka Lake (Figure 4c, e). Species composition 

also shifted, e.g., in Keuka Lake, smaller-bodied Daphnia retrocurva relative density 

declined from 68% to 17%, while larger-bodied Daphnia pulicaria increased from 4% 

to 65%. By contrast to Keuka Lake, Daphnia spp. size increased while density 

decreased significantly despite alewife collapse (Figure 4d, f). Rotifers, counted in 

Otsego Lake, also declined across the time series including after the alewife collapse 



 

36 

(Appendix S1: Figure S8). These findings suggest some consistency in lower trophic-

level response to alewife predation release but with lake-specific outcomes as Keuka 

Lake showed increased biomass and a shift to larger-bodied taxa, while Otsego Lake 

exhibited only a shift in body size. 

    

Benthos 

The primary benthic change in both lakes was dreissenid mussel colonization, 

expanding the grazer population and likely affecting bottom-up ecosystem controls by 

reducing the phytoplankton available to zooplankton. Benthic surveys showed that 

mussel populations were well-established in both lakes, with a similar shift in relative 

abundance from zebra mussels to quagga mussels. In Keuka Lake, between 2010 (pre-

alewife collapse) and 2025 (post-alewife collapse), zebra mussel density declined 

significantly, while quagga mussel density remained stable (not significant) and 

appeared to shift to deeper sites (Table 1). In addition, Diporeia density generally 

declined from pre-alewife collapse in 2010 to post-alewife collapse, though increased 

between the 2019 and 2025 post-alewife collapse surveys (Table 1). Nonetheless, at 

replicate sites between 2010-2025, excluding the deep sampling site in 2025, Diporeia 

declined significantly by 61% from pre- to post-alewife collapse. Benthic species 

groups Chironomidae and Oligochaeta, however, appeared relatively stable 

(marginally or not significant) despite different survey depths in 2010 and 2019 (Table 

1). In Otsego Lake, dreissenid invasion occurred more recently than in Keuka Lake, 

but a similar species shift appeared underway: quagga mussels have outcompeted 



 

37 

zebra mussels across depths, increasing from < 500 individuals/m2 in 2021 to ≈ 7,000 

individuals/m2 in 2023, while zebra mussels remained below 200 individuals/m2 

across survey years. Quagga mussel densities were higher than those observed in 

Keuka Lake (Table 1), increasing the potential for a strong bottom-up ecosystem 

effect.  

 

Mysids in Keuka Lake 

Increased mysid biomass after alewife collapse was also inconsistent with a bottom-up 

effect in Keuka Lake. Hydroacoustic surveys revealed mysid density increased by 

26% after alewife collapse (Appendix S1: Table S4; Figure 5a). Netting surveys also 

revealed high mysid densities after alewife collapse: 186 individuals/m2 in October 

2023 and 297 individuals/m2 in June 2024, compared to Lake Ontario (121-316 

individuals/m2; Figure 5b, d) despite Keuka Lake being much shallower. Mysid 

biomass estimated from net surveys was also high after alewife collapse, averaging 1.1 

g/m2 (dry weight) in fall of 2023 and 0.48 g/m2 in summer of 2024, with fall biomass 

exceeding Lake Ontario values by up to 2.6 × (Figure 5b). The number of mysid prey 

available per alewife increased ≈ 20 × (Table 2).  

 

Top predator fish  

Top-down control of planktivores by piscivores appeared to occur in both lakes, with 

stronger evidence for Keuka Lake given higher estimated lake trout populations 



 

38 

(Figure 6). Hydroacoustic estimates of predator fish density in Keuka Lake declined 

from 52 fish/ha in 2000 to < 20 fish/ha post-collapse (2016–2023; Figure 6c, 

Appendix S1: Table S3). Cold-water netting surveys showed peak lake trout catch of 

20 fish/net in 2000, declining to 6 fish/net after alewife collapse. In Otsego Lake, 

predator density estimated from hydroacoustics peaked at 35 fish/ha in 2001, then 

declined to < 3 fish/ha in 2012 after alewife collapse (Figure 6d). Lake trout catch in 

Otsego Lake peaked in 2006-2008 (21 fish/net), with catch rates at or below Keuka 

Lake post-alewife collapse. Walleye stocking in Otsego Lake successfully increased 

their population during the period of alewife decline, with evidence of natural walleye 

reproduction after 2014, post-alewife collapse (Figures 6f, h). Thus, with a one to two 

orders of magnitude decline in alewife abundance, lake trout abundance in Keuka 

Lake only decreased by a factor of 2.5, whereas Otsego Lake experienced an order of 

magnitude decline in predator abundance, suggesting weaker top-down control of 

alewife in Otsego Lake. Only Otsego Lake showed a marginally significant decline in 

lake trout catch over time, despite a population augmented with stocking, suggesting a 

bottom-up limitation driven by alewife collapse. 

 

The consequences of alewife collapse to top predators were apparent in both study 

systems. Piscivory rates on alewife declined sharply after alewife collapse in both 

lakes (Table 2). Despite reduced abundance, adult lake trout annual survival rates 

remained stable (non-significant trends) in Keuka Lake, indicating the reduction was 

due to decreased recruitment, perhaps from cannibalism when alewife declined 



 

39 

(Figure 6e). In Keuka Lake, mean lake trout size and growth rates of medium and 

large-sized fish decreased, with no large lake trout caught after alewife collapse 

(Figure 6g). Significant decreases of growth rates also occurred across medium and 

large lake trout size classes in Otsego Lake; however, recent data suggest a partial 

recovery in growth rates of small- to medium-sized fish (Figure 6h). In Keuka Lake, 

increased growth rate trends of small lake trout were significant, and thus smaller-

sized lake trout appeared unaffected by alewife collapse (Figure 6g).  

 

Winter severity 

Our ice cover analysis revealed unusually severe winters in 2014 and 2015, correlating 

with the timing of alewife collapse in Keuka Lake, but similar patterns were not 

evident in Otsego Lake (Figure 7). Long-term trends of ice cover duration across three 

inland New York lakes and maximum ice cover extent in Lake Ontario generally 

declined from 1970-2025. Inland lake ice coverage averaged 73 days (SD 31 days) for 

Otsego Lake, 84 days (SD 27 days) for Oneida Lake, and 98 days (SD 22 days) for 

Cazenovia Lake while maximum ice extent in Lake Ontario averaged 29% (SD 19%) 

from 1970-2025. These inland lakes showed increases in ice cover duration during 

alewife collapse in Keuka Lake from 2014-2015 (Figure 7c), including 117 days in 

Oneida Lake (> 1 SD from the long-term mean) and 120 days in Cazenovia Lake (≈ 1 

SD from the long-term mean). This is even more apparent in the ice cover extent data 

from Lake Ontario, which increased from 2% in 2012 at the start of the alewife 

collapse period to 61% in 2014 and 82% in 2015 at the end of alewife collapse, nearly 



 

40 

3 SD above the long-term mean (Figure 7a). These results are consistent with air 

temperature data collected near Keuka Lake, where winter growing degree days for 

consecutive winters 2013-2014 and 2014-2015 approached 2 SD below the mean 

(Figure 7b). In contrast, although ice cover duration peaked locally in Otsego Lake 

during alewife collapse in 2009 (Figure 7d), the increase was < 1 SD from the long-

term average, a notably weaker signal than the severe winters inferred to occur in 

Keuka Lake during alewife collapse. 

 

DISCUSSION  

Simplified food webs and extensive long-term monitoring of inland study lakes 

enabled us to disentangle the causes and consequences of fish population decline. 

Empirical evidence revealed multiple pathways to landlocked alewife collapse in 

inland lakes, with some common ecosystem consequences such as zooplankton 

predation release and increases in specific taxa biomass. In lakes with an alewife-

dominated forage base, we found that their population collapse left a vacant 

planktivore niche, prompting prey switching by top predators. These dynamics were 

strongly mediated by lake-specific factors, including alternative prey availability for 

top predators and productivity of lower trophic levels. Our data-driven analysis 

provides ecosystem-level insights into how predator-prey interactions shape food web 

structures in lakes.    

 

Competing hypotheses for causes of alewife collapse 



 

41 

What are the drivers of introduced alewife population collapse, and can they be 

generalized across lakes? Our findings indicated a similar magnitude of alewife 

decline and timing of population collapse in Keuka Lake and Otsego Lake; however, 

the underlying ecosystem drivers were different. Each of our hypotheses was 

supported in at least one study system, yet ecosystem dynamics appeared lake-specific 

and generalizations for a primary cause of alewife collapse were not supported across 

both lakes. In Otsego Lake, the evidence for causes of alewife collapse most strongly 

supports bottom-up declines in primary production and zooplankton prey availability 

(H1), with moderate support for top-down lake trout and introduced walleye predation 

(H2) but limited support for winter severity effects (H3). By contrast, evidence for 

alewife collapse in Keuka Lake most strongly supported top-down control by 

abundant lake trout (H2), with a series of severe winters leading to an abrupt 

population collapse (H3). We lacked evidence supporting the hypothesis that bottom-

up control via decreased primary production from dreissenid mussel invasion and 

zooplankton prey limitation (H1) contributed to alewife collapse in Keuka Lake.  

 

Only Otsego Lake showed trends consistent with bottom-up regulation (H1) of 

alewife, including decreased chlorophyll-a concentrations and zooplankton abundance 

before and after alewife collapse. A bottom-up effect predicts that primary production 

limitations should amplify upwards through the food web, resulting in reduced 

zooplankton and mid to upper trophic level fish (McQueen et al., 1989; Gliwicz, 

2002). In Otsego Lake, declines in lower trophic production occurred shortly after 



 

42 

zebra mussels arrived during the alewife decline, and continued to decline after 

alewife collapse and quagga mussels arrived. This trend reflects bottom-up effects 

commonly associated with dreissenid mussels (Karatayev et al., 2022) and parallels 

the reduced food availability hypothesized to have contributed to alewife collapse in 

Lake Huron (Nalepa et al., 2007; Barbiero et al., 2011; Bunnell et al., 2014; Kao et al., 

2016).  

 

In Keuka Lake, we found evidence for nutrient declines likely contributing to the 

initial alewife decline; however, bottom-up limitations were not the primary driver of 

subsequent population collapse. Nutrient declines were modest in Keuka Lake, mainly 

occurring in the 1990s along with zooplankton biomass declines after zebra mussels 

arrived and during early alewife decline, but contrary to our predictions for a bottom-

up effect, both zooplankton and chlorophyll-a (a proxy for phytoplankton) increased 

after alewife collapse. One possibility is that alewife can decline without a bottom-up 

concomitant decline in lower trophic level zooplankton. Additionally, the strongest 

effects of dreissenid mussels typically occur within 5-10 years after their arrival, and 

we hypothesize that their earlier invasion of both species in Keuka Lake provided 

more time for ecosystem recovery as mussel grazing pressure declined and 

phytoplankton resistance improved (Karatayev et al., 2022, 2023). Primary production 

in Keuka Lake appeared sufficient to support both phytoplankton and zooplankton 

growth across the time series. Using total phosphorus as an indicator of nutrient 

availability and chlorophyll a as a proxy for phytoplankton abundance, Keuka Lake 



 

43 

ecosystem responses did not fully align with our expectations for a trophic cascade, 

because primary productivity increased post-alewife collapse. A compensatory 

increase in zooplankton biomass supports the size-efficiency hypothesis (Brooks and 

Dodson, 1965; Hall et al., 1976), but not a classic trophic cascade leading to reduced 

primary production (Carpenter et al., 1985). This pattern in Keuka Lake aligns with 

Mehner (2009), who found limited top-down control on lower trophic level 

productivity in lakes.  

 

High lake trout abundance in Keuka Lake likely exerted strong top-down control (H2) 

on alewife populations, consistent with predator-driven planktivore decline hypotheses 

(Carpenter et al., 1985; Mills and Forney, 1988; O’Gorman and Stewart, 1999). In 

Keuka Lake, hydroacoustic density estimates ranged from 52 lake trout/ha to 43 lake 

trout/ha before alewife collapse and remained high immediately after collapse at 16 

lake trout/ha (in 2016; before population establishment of piscivorous walleye). Such 

densities in Keuka Lake exceeded regional averages and were consistent with high 

piscivory rates that could suppress prey fish populations (see Table 3). For 

comparison, a review of 45 North American populations reported a mean density of 

7.1 lake trout/ha, with 95% of populations ranging between 0.3 and 25 lake trout/ha 

(Hansen et al., 2021). In Otsego Lake, predator densities, comprised of both lake trout 

and walleye, ranged from a maximum 35.2 predators/ha before alewife collapse to 3.2 

predators/ha after alewife collapse. Thus, even the combined predator densities in 

Otsego Lake remained lower than those observed in Keuka Lake. Gill net catch rates 



 

44 

in Keuka Lake also declined modestly as alewife declined but were higher than catch 

rates observed in Otsego Lake, where piscivorous lake trout and walleye densities 

were consistently lower. While increases in lake trout and stocked walleye may have 

contributed to alewife extirpation in Otsego Lake, we infer they were likely not the 

primary driver of collapse, given lower abundances and an order of magnitude decline 

in piscivores after alewife collapse consistent with a bottom-up food web effect. These 

contrasting predator densities between lakes support the hypothesis that strong top-

down pressure in Keuka Lake contributed to a trophic cascade of increased predator 

lake trout and decreased planktivore alewife. 

 

The timing of severe winters coincided with declining alewife populations, suggesting 

that abiotic stressors (H3) may have compounded trophic drivers of alewife 

particularly in Keuka Lake. Regional ice cover analyses revealed unusually harsh 

winters in 2014 and 2015, associated with polar vortex events, which led to extended 

ice cover in nearby inland lakes and a maximum ice cover extent in Lake Ontario. A 

series of cold winters in Keuka Lake aligns with a third hypothesis proposed for the 

Great Lakes and other landlocked lakes, where severe winters were implicated in 

alewife collapse (O’Gorman and Stewart, 1999; Dunlop and Riley, 2013; Warren et 

al., 2024). In Keuka Lake, our analysis suggests that the declining alewife population, 

already under high predation pressure, was vulnerable to reduced overwinter condition 

or increased immunosuppression triggered by cold winter conditions (Lepak and 

Kraft, 2008). As top-down control of adult alewife by lake trout triggered a 



 

45 

compensatory response of juveniles during the decline period, we infer that extrinsic 

forces like severe winters likely amplified demographic stochasticity at low population 

sizes, increasing susceptibility to juvenile mortality and accelerating their collapse.  

 

Food web consequences of alewife collapse 

The consequences of alewife collapse were similar and apparent in both ecosystems 

across multiple trophic levels. Our analysis reveals a combination of bottom-up and 

top-down drivers contributed to prey fish decline, with food web responses that 

resemble a mid-trophic level dominated food web where alewife regulate both top 

predator and zooplankton prey structure (Cury et al., 2000; Lynam et al., 2017; 

Moyano et al., 2023). This was evident with a shift in zooplankton from Bosmina to 

Daphnia, with compensatory release in Daphnia biomass following alewife collapse. 

A clear relationship between decreased Daphnia size and alewife presence is well-

described from land-locked populations (Brooks and Dodson, 1965; Hall et al., 1976; 

Post et al., 2008). An alewife effect on Daphnia size was particularly apparent in both 

Keuka Lake and Otsego Lake. Indeed, the sharp increase in Daphnia biomass 

observed in Keuka Lake in summer 2014 strongly suggests that alewife collapse 

occurred during the first polar vortex in winter 2014. Mysid biomass in Keuka Lake 

also increased after alewife collapse. Mysids, as omnivores, consume zooplankton, 

phytoplankton, and detritus; and detritus may have increased from dreissenid mussels. 

Following alewife collapse, mysids may benefit from reduced competition for 

zooplankton, increased access to benthic detritus production, and release from alewife 



 

46 

predation pressure (Ellis et al., 2011). Combined, the compensatory increase in 

zooplankton and mysid biomass following alewife collapse suggests that alewife 

exerted strong top-down control over lower trophic levels in Keuka Lake. This 

contrasts with a bottom-up limitation of zooplankton production that occurred during 

periods of high alewife abundance and after their collapse, as observed in Otsego 

Lake. 

 

Consequences of alewife collapse generally followed our expectations for top 

predators. Lake trout condition and abundance declined in both lakes, and diets shifted 

from alewife to yellow perch, slimy sculpin (Cottus cognatus), and mysids in Keuka 

Lake after alewife collapse (Appendix S1: Figure S9). Mysid-dominated diets 

appeared sufficient to maintain condition of smaller, younger lake trout but were 

insufficient for larger individuals. Alewife collapsed earlier in Otsego Lake, where 

lake trout condition of small to medium sized fish appeared to improve in recent years. 

In both lakes, netting data indicated a prey fish shift from alewife to yellow perch in 

catches, with yellow perch catch rates increasing substantially in Otsego Lake in 

recent years (Appendix S1: Figures S10-11). Walleye natural reproduction also 

established after alewife collapse in both lakes, which could be from successful 

stocking efforts in Otsego Lake, an intentional introduction to Keuka Lake, combined 

with a release of larval walleye from alewife depredation (Brooking et al., 1998; 

Porath et al., 2003; Rudstam et al., 2011). Demersal introduced prey fish rainbow 

smelt (Osmerus mordax) have remained absent from Keuka Lake since the mid-2000s 



 

47 

(Appendix S1: Figure S12), and alewife have not recovered, likely due to continued 

predation by lake trout that can maintain a high population feeding on alternative but 

less preferred prey, such as abundant yellow perch and mysids (Ellis et al., 2020).  

 

Data limitations 

Inference from our study was limited by a lack of long-term data for several 

ecosystem components. We lacked environmental measurements that could help 

distinguish internal biotic and external abiotic drivers of nutrient dynamics, which 

could be particularly useful for understanding drivers of recent increases in nitrate in 

Otsego Lake and total phosphorus in Keuka Lake. While a decline in chlorophyll a 

during alewife decline is consistent with increased zooplankton and mussel grazing 

pressure, observed trends in total phosphorus and nitrogen likely reflect variation of 

inorganic factors like run-off intensity, watershed management practices, and land use 

changes whereas variability among years often reflects wet and dry years (Burford and 

Lu, 2024; Dillon and Molot, 2024). Differences in lake volume and a residence time of 

6.3 years in Keuka Lake and 3.8 years in Otsego Lake may impact nutrient cycling 

and food web dynamics; however, inference to watershed-level dynamics was outside 

the scope of the present study (Bloomfield, 1978a, 1978b; Andersen et al., 2024). In 

addition, zooplankton sampling was restricted to the epilimnion in both lakes, missing 

up to 75% of deeper-dwelling species-groups in Keuka Lake and therefore 

underestimating species biomass particularly for calanoid copepods (for summer 2024 

paired epilimnetic and whole-water column tow surveys see Appendix S1: Table S5 



 

48 

and Figure S13). As such, sampling efforts may have failed to capture important shifts 

in zooplankton abundance and distribution at depth after alewife collapse. In Otsego 

Lake, hydroacoustic and trap net surveys captured alewife population dynamics from 

their initial establishment through collapse. By contrast, Keuka Lake lacked pre-

collapse hydroacoustic data and long-term gill net surveys were designed for lake trout 

with alewife caught as bycatch, limiting our ability to estimate alewife abundance 

prior to the start of their decline. As a result, we lack insights into potential peak 

population size and whether alewife experienced earlier boom-and-bust cycles 

following their introduction to Keuka Lake, likely by the early 1900s.  

 

Benthic survey datasets were also intermittent, especially before and during early 

dreissenid mussel invasion, potentially limiting our capabilities to identify important 

benthic-pelagic energy pathways described for north-temperate lakes (Vander Zanden 

and Vadeboncoeur, 2002; Stewart and Sprules, 2011). In Keuka Lake, Diporeia, an 

important benthic prey item for fish, declined between 2010 (shortly after quagga 

mussels arrived) and 2025 after alewife collapsed. Diporeia declines are linked with 

dreissenid mussel invasion and declines have co-occurred with declining mysids in 

some of the Great Lakes (Watkins et al., 2012; Holda et al., 2023). Yet, mysid 

populations remain abundant in Keuka Lake and appeared to increase after alewife 

collapse, in contrast to patterns in Lake Huron and Lake Michigan. We expect 

additional benthic information could be useful to evaluate the role of benthic changes 

and visual predation in food web dynamics.  



 

49 

 

Implications for ecosystem-based management 

Our results indicate that even in ecologically similar lakes the primary mechanisms for 

alewife collapse differed. This provides two important insights for managing prey fish 

populations to improve ecosystem resilience. First, in Otsego Lake piscivore stocking 

as a method of biocontrol likely contributed to prey fish collapse; however, we lacked 

evidence that this was a primary driver of alewife decline. Nonetheless, maintaining 

the predator population (e.g., stocked lake trout, established walleye) may have 

prevented collapsed alewife populations from recovering. This may have resulted from 

an established walleye population with increased habitat overlap with alewife, 

compared to cold-water lake trout, thereby contributing to continuous predation 

pressure throughout the alewife’s life cycle. Second, for species restoration efforts like 

present cisco rehabilitation in Keuka Lake, identifying the original causes of 

population decline are therefore important to improve future restoration success 

(Cochran-Biederman et al., 2015; Mrnak et al., 2022). Although adult alewife annual 

survival rates (30-35%; Appendix S1: Figure S5) estimated in this study are lower than 

historical adult cisco annual survival rates (51.4%) from Keuka Lake, high lake trout 

predation and poor juvenile cisco survival limit population establishment (Koeberle et 

al., 2025). Time series analyses presented here demonstrate that Keuka Lake, although 

oligotrophic, has abundant zooplankton and mysid prey, and lacks alewife 

competitors, conditions more suitable to cisco than for alewife (Ridgway et al., 2020; 

Bunnell et al., 2023). Yet, top-down control by still abundant lake trout may be 



 

50 

preventing restoration of the native cisco population as cisco is a preferred prey of 

lake trout in lakes where they co-occur. Both lake trout and cisco prefer cold-water 

habitats compared to cool-water alewife and yellow perch, resulting in more thermal 

overlap between cisco and lake trout.  

 

Conclusions 

In summary, this study demonstrates that multiple pathways can lead to prey fish 

collapse, yet common food web responses to collapse emerge across lake ecosystems. 

Although the primary mechanism for alewife collapse differed between lakes, both 

systems exhibited a persistent void in the pelagic prey fish niche, consistent with an 

intermediate trophic level dominated food web structure (Stein et al., 1995; Fauchild 

et al., 2011; Buren et al., 2019). As a result, zooplankton responded rapidly to release 

from planktivore predation, but the magnitude of this response to alewife collapse 

varied by lake productivity. This highlights how food web dynamics mediate 

ecosystem stability across gradients of size, structure, and biotic and abiotic diversity 

(Tunney et al. 2012). Top predators also adapted their prey reliance on dominant 

alewife, with lake-specific prey subsidies such as mysids in Keuka Lake influencing 

post-alewife collapse predator growth rates and abundance (Ellis et al., 2010; Ridgway 

et al., 2020). Combined, while food web structural response to alewife collapse 

appeared consistent, lake-specific lower trophic level productivity and prey 

availability to top predators influenced each lake’s ecosystem trajectory.  

 



 

51 

Our study design provides a rare opportunity to assess food web structural changes 

before and after the natural collapse of an introduced dominant planktivorous fish. 

Analysis of long-term datasets reveals a partial return to a pre-alewife introduction 

ecosystem state, including increased water clarity and alewife suppression or 

extirpation. While restored food web function is unlikely to return to a pre-modified 

state (Carpenter et al., 2011; Bunnell et al., 2021; Lesser et al., 2024; Walsworth et al., 

2025), future work could explore whether functional near-equivalent species can 

restore the mid-trophic structural void left by alewife (McCann and Rooney, 2009; 

McMeans et al., 2016). We anticipate that whole-lake ecosystem models will be useful 

in conjunction with long-term datasets to evaluate ecosystem changes (Stewart and 

Sprules, 2011; Lesser et al., 2024). Such models could quantify food web structural 

changes, shifts in energy flow, or future sustainability of the fish community structure 

to improve capacity for adaptive management strategies. Our work demonstrates that a 

better understanding of past drivers of species decline can have important insights into 

future ecosystem dynamics. 

 

Acknowledgements 

The authors thank multiple local, state, and federal partners for their long-term 

collaboration and monitoring of Keuka Lake. Specifically, we thank Keuka Lake 

Association: Robert Lampert, Robert Dintruff, Darryl Heckle, Maria Hudson, Steven 

Brigham, Susan Oliver, and many local volunteers for their long-term data collection. 

At Cornell University, we thank Rory Paltridge, Ondine Morgan-Knapp, and Joe 



 

52 

Connolly for their help with mysid surveys, Chris Marshall for zooplankton analysis 

code, and members of the Shack Lab for their valuable input. We also thank Jeremy 

Kraus, Gregg Mackey, and staff (USGS-Tunison); Steve Robb, Bree Minges, Ariel 

Thoms, Matthew Sanderson, Ben Carson, Mike Disarno, Kate Riordan, and many staff 

(NYSDEC Region 7; NYSDEC Region 4); Florian Reyda (SUNY Oneonta); John 

Foster, Mark Cornwell, and many students (SUNY Cobleskill). Lastly, the authors 

would like to thank Lew McCaffrey, Evan Cooch, and Zena Casteel for their 

comments which greatly improved this manuscript. Any use of trade, firm, or product 

names is for descriptive purposes only and does not imply endorsement by the U.S. 

Government.  

 

Funding  

Keuka Lake research was funded by New York State Department of Environmental 

Conservation using Federal Aid Sport Fish Restoration Funds from “Grant F64-R” and 

New York Sea Grant “R/FBF-28”. Financial support for Otsego Lake research was 

provided in part by the Clark Foundation and a National Science Foundation Field 

Stations and Marine Laboratories (FSML) grant to W. Harman (NSF DBI 1034744). 

Survey of Dreissena in 2025 was funded by SUNY Buffalo State University’ Ruth 

Huppuch Great Lakes Center Endowed Fund. Burlakova was supported by the U.S. 

EPA through the Great Lakes Restoration Initiative via a Cooperative Agreement with 

Cornell University, Department of Natural Resources under Award GL00E03089 

“Great Lakes Biology Monitoring Program: Zooplankton, Mysis, Benthos 2022-2027” 



 

53 

(PI J. Watkins) and Subaward # 141698-21585 to SUNY Buffalo State University (Co-

PIs LEB and AYK). 



 

54 

References 

 

Andersen, A.M., Blanchfield, P.J., Heuvel, C.E., McCann, K.S., McMeans, B.C., 

Ridgway, M.S. and A.T. Fisk. 2024. “Influence of lake volume on food web 

metrics in a freshwater fish assemblage across a small range of ecosystem sizes.” 

Canadian Journal of Fisheries and Aquatic Sciences 82: 1-14. 

https://doi.org/10.1139/cjfas-2024-0226  

 

Balcer, M.D., Korda, N.L. and S.I. Dodson. 1984. Zooplankton of the Great Lakes: a 

guide to the identification and ecology of the common crustacean species. University 

of Wisconsin Press. 

 

Barbiero, R.P., Lesht, B.M., and G.J. Warren. 2011. “Evidence for bottom–up 

control of recent shifts in the pelagic food web of Lake Huron.” Journal of  

Great Lakes Research 37(1): 78–85. https://doi.org/10.1016/j.jglr.2010.11.013  

 

Bence, J.R., Madenjian, C.P., He, J.X., Fielder, D.G., Pothoven, S.A., Dobiesz, N.E., 

Johnson, J.E., Ebener, M.P., Cottrill, R.A., Mohr, L.C., and S.R. Koproski. 2016. 

“Reply to comments by Riley and Dunlop on He et al. (2016).” Canadian Journal of 

Fisheries and Aquatic Sciences 73(5), 865–868. https://doi.org/10.1139/cjfas-2015-

0522 

 

Bloomfield, J.A. 1978a. Lakes of New York State: Ecology of the Finger Lakes 

New York. Academic Press. 

 

Bloomfield, J.A. 1978b. Lakes of New York State: Ecology of the Lakes of East 

Central New York. Academic Press. 

 

Brandt, S.B., Mason, D.M., Macneill, D.B., Coates, T., and J.E. Gannon. 1987. 

“Predation by alewives on larvae of yellow perch in Lake Ontario.” 

Transactions of the American Fisheries Society 116(4): 641–645. 

https://doi.org/10.1577/1548-8659(1987)116%3C641:PBAOLO%3E2.0.CO;2  

 

Brooking, T.E. and L.G. Rudstam. 2009. “Hydroacoustic target strength distributions 

of alewives in a net‐cage compared with field surveys: deciphering target strength 

distributions and effect on density estimates.” Transactions of the American Fisheries 

Society 138(3): 471-486. https://doi.org/10.1577/T07-215.1 

 

Brooking, T.E., Rudstam, L.G., Olson, M.H., and A.J. Vandevalk. 1998. “Size 



 

55 

dependent alewife predation on larval walleyes in laboratory experiments.” 

North American Journal of Fisheries Management 18(4): 960–965. 

https://doi.org/10.1577/1548-8675(1998)018%3C0960:SDAPOL%3E2.0.CO;2  

 

Brooks, J.L. and S.I. Dodson. 1965. “Predation, Body Size, and Composition of 

Plankton: The effect of a marine planktivore on lake plankton illustrates theory 

of size, competition, and predation.” Science 150(3692): 28–35. 

https://doi.org/10.1126/science.150.3692.28  

 

Brown, T.A., Rudstam, L.G., Sethi, S.A., Ripple, P., Smith, J.B., Treska, T.J., 

Hessell, C., Olsen, E., He, J.X., Jonas, J.L., Rook, B.J, Blankenheim, J.E., 

Beech, S.J.H., Brown, E., Berglund, E.K., Cook, H.A., Dunlop, E.S., James, 

S., Pothoven, S.A., Amidon, Z.J., Sweka, J.A., Carl, D.D., Hansen, S.P., 

Bunnell, D.B., Weidel, B.C., and A.E. Honsey. 2025. “Reconstructing half a 

century of coregonine recruitment reveals species-specific dynamics and 

synchrony across the Laurentian Great Lakes.” ICES Journal of Marine 

Science 82(2): fsae160. https://doi.org/10.1093/icesjms/fsae160  

 

Bunnell, D.B., Ackiss, A.S., Alofs, K.M., Brant, C.O., Bronte, C.R., Claramunt, 

R.M., Dettmers, J.M., Honsey, A.E., Mandrak, N.E., Muir, A.M., Santucci, V.J. 

Jr., Smith, D.R., Strach, R.M., Sweka, J.A., Weidel, B.C., Mattes, W.P., and 

Newman, K.R. 2023. “A science and management partnership to restore 

coregonine diversity to the Laurentian Great Lakes.” Environmental Reviews 

31(4): 716–738. https://doi.org/10.1139/er-2022-0109  

 

Bunnell, D.B., Barbiero, R.P., Ludsin, S.A., Madenjian, C.P., Warren, G.J., Dolan, 

D.M., Brenden, T.O., Briland, R., Gorman, O.T., He, J.X., Johengen, T.H., 

Lantry, B.F., Lesht, B.M., Nalepa, T.F., Riley, S.C., Riseng, C.M., Treska, T.J., 

Tsehaye, I., Walsh, M.G., Warner, D.M., and B.C. Weidel. 2024. “Changing 

Ecosystem Dynamics in the Laurentian Great Lakes: Bottom-Up and Top 

Down Regulation.” BioScience 64: 26–39. 

https://doi.org/10.1093/biosci/bit001 

 

Bunnell, D.B., Ludsin, S.A., Knight, R.L., Rudstam, L.G., Williamson, C.E., 

Höök, T.O., Collingsworth, P.D., Lesht, B.M., Barbiero, R.P., Scofield, A.E., 

Rutherford, E.S., Gaynor, L., Vanderploeg, H.A., and M.A. Koops. 2021. 

“Consequences of changing water clarity on the fish and fisheries of the 

Laurentian Great Lakes.” Canadian Journal of Fisheries and Aquatic Sciences 

78(10): 1524-1542. https://doi.org/10.1139/cjfas-2020-0376  

https://doi.org/10.1093/icesjms/fsae160
https://doi.org/10.1139/er-2022-0109
https://doi.org/10.1139/cjfas-2020-0376


 

56 

 

Buren, A.D., Murphy, H.M., Adamack, A.T., Davoren, G.K., Koen-Alonso, M., 

Montevecchi, W.A., Mowbray, F.K., Pepin, P., Regular, P.M., Robert, D., Rose, 

G.A., Stenson, G.B., and D. Varkey. 2019. “The collapse and continued low 

productivity of a keystone forage fish species.” Marine Ecology Progress 

Series 616: 155–170. https://doi.org/10.3354/meps12924  

 

Burford, M.A. and J. Lu. 2024. The Nitrogen Cycle. In: Jones, I.D., Smol, J.P., 

eds. Wetzel’s Limnology: Lake and river ecosystems. 4th ed. London: Elsevier, 

Chapter 28. 

 

Carpenter, S.R., Cole, J.J., Hodgson, J.R., Kitchell, J.F., Pace, M.L., Bade, D., 

Cottingham, K.L., Essington, T.E., Houser, J.N., and D.E. Schindler. 2001. 

“Trophic cascades, nutrients, and lake productivity: whole‐lake experiments.” 

Ecological Monographs 71(2): 163–186. 

https://doi.org/10.1890/00129615(2001)071[0163:TCNALP]2.0.CO;2  

 

Carpenter, S.R., Cole, J.J., Pace, M.L., Batt, R., Brock, W.A., Cline, T., Coloso, J., 

Hodgson, J.R., Kitchell, J.F., Seekell, D.A., Smith, L., and B. Weidel. 2011. 

“Early warnings of regime shifts: a whole-ecosystem experiment.” Science 

332: 1079–1082. https://doi.org/10.1126/science.1203672  

 

Carpenter, S.R. and J.F. Kitchell. 1996. The Trophic Cascade in Lakes. Cambridge 

University Press. 

 

Carpenter, S.R., Kitchell, J.F., and J.R. Hodgson. 1985. “Cascading trophic 

interactions and lake productivity.” BioScience, 35(10): 634-639. 

https://doi.org/10.2307/1309989  

 

Cury, P., 2000. Small pelagics in upwelling systems: patterns of interaction and 

structural changes in “wasp-waist” ecosystems. ICES Journal of Marine 

Science 57(3): 603–618. https://doi.org/10.1006/jmsc.2000.0712 

 

Davis, J.P. and E.T. Schultz. 2009. “Temporal shifts in demography and life history 

of an anadromous alewife population in Connecticut.” Marine and Coastal 

Fisheries 1(1): 90–106. https://doi.org/10.1577/C08-003.1  

 

 

https://doi.org/10.3354/meps12924
https://doi.org/10.1126/science.1203672
https://doi.org/10.1006/jmsc.2000.0712
https://doi.org/10.1577/C08-003.1


 

57 

Dermott, R., Bonnell, R., and P. Jarvis. 2006. Changes in abundance of deep water 

amphipod Diporeia (Pontoporeiidae) in eastern North American lakes with or without 

Dreissena mussels. Canadian Technical Report of Fisheries and Aquatic Sciences 

2636.  

 

Dettmers, J.M., Goddard, C.I., and K.D. Smith. 2012. “Management of alewife 

using Pacific salmon in the Great Lakes: whether to manage for economics or 

the ecosystem?” Fisheries 37(37), 495–501. 

https://doi.org/10.1080/03632415.2012.731875  

 

Dillon, P.J. and L.A. Molot. 2024. The Phosphorus Cycle. In: Jones, I.D., Smol, 

J.P., eds. Wetzel’s Limnology: Lake and river ecosystems. 4th ed. London: 

Elsevier, Chapter 28. 

 

Dougherty, M.M., MacDonald, A., York, G., and D.M. Post. 2025. “Monitoring a 

keystone species (Alosa pseudoharengus) with environmental effects: A 

comparison with direct capture and environmental DNA.” PLoS One 20: 

e0324385. https://doi.org/10.1371/journal.pone.0324385  

 

Dunlop, E.S. and S.C. Riley. 2013. “The contribution of cold winter temperatures 

to the 2003 alewife population collapse in Lake Huron.” Journal of Great 

Lakes Research 39(4): 682–689. https://doi.org/10.1016/j.jglr.2013.08.001  

 

Ellis, B.K., Stanford, J.A., Goodman, D., Stafford, C.P., Gustafson, D.L., 

Beauchamp, D.A., Chess, D.W., Craft, J.A., Deleray, M.A., and B.S. Hansen. 

2011. “Long-term effects of a trophic cascade in a large lake ecosystem.” 

Proceedings of the National Academy of Sciences 108(3): 1070–1075. 

https://doi.org/10.1073/pnas.1013006108  

 

Essington, T.E., Moriarty, P.E., Froehlich, H.E., Hodgson, E.E., Koehn, L.E., 

Oken, K.L., Siple, M.C., and C.C. Stawitz. 2015. “Fishing amplifies forage 

fish population collapses.” Proceedings of the National Academy of Sciences 

112(21): 6648–6652. https://doi.org/10.1073/pnas.1422020112  

 

Fairchild, H.L. 1899. “Glacial waters in the Finger Lakes region of New York.” 

Bulletin of the Geological Society of America 10(1): 27-68.  

 

Fairchild, H.L. 1926. “Geologic romance of the Finger Lakes.” The Scientific 

Monthly 23(2), 161-173.  

https://doi.org/10.1371/journal.pone.0324385
https://doi.org/10.1016/j.jglr.2013.08.001
https://doi.org/10.1073/pnas.1422020112


 

58 

 

Fauchald, P., Skov, H., Skern-Mauritzen, M., Johns, D., and T. Tveraa. 2011. 

“Wasp-waist interactions in the North Sea ecosystem.” PloS one 6: e22729. 

https://doi.org/10.1371/journal.pone.0022729  

 

Fisher, J.P., Fitzsimons, J.D., Combs, G.F., and J.M. Spitsbergen. 1996. “Naturally 

occurring thiamine deficiency causing reproductive failure in Finger Lakes 

Atlantic salmon and Great Lakes lake trout.” Transactions of the American 

Fisheries Society 125(2): 167–178. 

https://doi.org/10.1577/15488659(1996)125%3C0167:NOTDCR%3E2.3.CO;2  

 

Fitzpatrick, K.B., Weidel, B.C., Connerton, M.J., Lantry, J.R., Holden, J.P., Yuille, 

M.J., Lantry, B., LaPan, S.R., Rudstam, L.G., Sullivan, P.J., Brenden, T.O., and 

S.A. Sethi. 2022. “Balancing prey availability and predator consumption: a 

multispecies stock assessment for Lake Ontario.” Canadian Journal of 

Fisheries and Aquatic Sciences 79(9): 1529–1545. 

https://doi.org/10.1139/cjfas-2021-0126  

 

Fitzsimons, J.D., Williston, B., Zajicek, J.L., Tillitt, D.E., Brown, S.B., Brown, 

L.R., Honeyfield, D.C., Warner, D.M., Rudstam, L.G., and W. Pearsall. 2005. 

Thiamine content and thiaminase activity of ten freshwater stocks and one 

marine stock of alewives. Journal of Aquatic Animal Health 17(1): 26–35. 

https://doi.org/10.1577/H04-002.1 

 

Froese, R. and D. Pauly. Editors. 2025. FishBase. World Wide Web electronic 

publication. www.fishbase.org, (04/2025). 

 

Hairston, N.G., Smith, F.E., and L.B. Slobodkin. 1960. Community structure, 

population control, and competition. The American Naturalist 94(879): 421 

425. https://doi.org/10.1086/282146  

 

Hall, D.J., Threlkeld, S.T., Burns, C.W., and P.H. Crowley. 1976. “The size 

efficiency hypothesis and the size structure of zooplankton communities.” 

Annual Review of Ecology and Systematics 7(1976): 177–208. 

https://doi.org/10.1146/annurev.es.07.110176.001141  

 

Hansen, Michael J, Guy, C.S., Bronte, C.R., and N.A. Nate. 2021. Life history and 

population dynamics, in: The Lake Charr Salvelinus Namaycush: Biology, 

Ecology, Distribution, and Management. Springer, pp. 253–286. 

https://doi.org/10.1086/282146


 

59 

 

Harman, W.N., Albright, M.F., and D.M. Warner. 2002. “Trophic changes in 

Otsego Lake, NY following the introduction of the alewife (Alosa 

pseudoharengus).” Lake and Reservoir Management 18(3): 215–226. 

https://doi.org/10.1080/07438140209354150  

 

He, J.X., Bence, J.R., Madenjian, C.P., Pothoven, S.A., Dobiesz, N.E., Fielder, 

D.G., Johnson, J.E., Ebener, M.P., Cottrill, R.A., Mohr, L.C., S.R. Koproski. 

2015. “Coupling age-structured stock assessment and fish bioenergetics 

models: a system of time-varying models for quantifying piscivory patterns 

during the rapid trophic shift in the main basin of Lake Huron.” Canadian 

Journal of Fisheries and Aquatic Sciences 72(1): 7–23. 

https://doi.org/10.1139/cjfas-2014-0161  

 

Holda, T.J., Watkins, J.M., Scofield, A.E., Pothoven, S.A., Warner, D.M., O’Brien, 

T.P., Bowen, K.L., Currie, W.J., Jude, D.J., Boynton, P.V., and L.G. Rudstam. 

2023. “Intra-lake trends and inter-lake comparisons of Mysis diluviana life 

history variables and their relationships to food limitation.” Journal of Great 

Lakes Research 49(5): 1179–1189. https://doi.org/10.1016/j.jglr.2023.08.001  

 

Jeppesen, E. J., Volta, P., and Z. Mao. 2024. Fish. In: Jones, I.D., Smol, J.P., eds. 

Wetzel’s Limnology: Lake and river ecosystems. 4th ed. London: Elsevier, 

Chapter 22. 

 

Jeppesen, E., Sondergaard, M., Mortensen, E., Kristensen, P., Riemann, B., Jensen, 

H.J., Muller, J.P., Sortkjaer, O., Jensen, J.P., Christoffersen, K., Bosselmann, S., and E. 

Dall. 1990. “Fish manipulation as a lake restoration tool in shallow, eutrophic 

temperate lakes l: cross-analysis of three Danish case-studies.” Hydrobiologia 200, 

205-218. https://doi.org/10.1007/BF02530340  

 

Johannsson, O.E., Bowen, K.L., Holeck, K.T. and M.G. Walsh. 2011. “Mysis diluviana 

population and cohort dynamics in Lake Ontario before and after the establishment of 

Dreissena spp., Cercopagis pengoi, and Bythotrephes longimanus.” Canadian Journal 

of Fisheries and Aquatic Sciences 68(5): 795-811. https://doi.org/10.1139/f2011-028 

 

Johnson, J.E., DeWitt, S.P., and D.J. Gonder. 2010. “Mass-marking reveals 

emerging self regulation of the Chinook Salmon population in Lake Huron.” 

North American Journal of Fisheries Management 30(2): 518–529. 

https://doi.org/10.1577/M09-094.1  

https://doi.org/10.1016/j.jglr.2023.08.001


 

60 

 

Jones, M.L., Koonce, J.F., and R. OˈGorman. 1993. “Sustainability of hatchery 

dependent salmonine fisheries in Lake Ontario: the conflict between predator 

demand and prey supply.” Transactions of the American Fisheries Society 

122(5): 1002–1018. 

https://doi.org/10.1577/15488659(1993)122%3C1002:SOHDSF%3E2.3.CO;2  

 

Kao, Y.-C., Adlerstein, S.A., and E.S. Rutherford. 2016. “Assessment of top-down 

and bottom-up controls on the collapse of alewives (Alosa pseudoharengus) in 

Lake Huron.” Ecosystems 19: 803–831.  

https://doi.org/10.1007/s10021-016-9969-y  

 

Karatayev, A.Y., Burlakova, L.E., Mehler, K., Elgin, A.K., Rudstam, L.G., 

Watkins, J.M., and M. Wick. 2022. “Dreissena in Lake Ontario 30 years post 

invasion.” Journal of Great Lakes Research 48(2): 264–273. 

https://doi.org/10.1016/j.jglr.2020.11.010  

 

Karatayev, V.A., Rudstam, L.G., Karatayev, A.Y., Burlakova, L.E., Adamovich, 

B.V., Zhukava, H.A., Holeck, K.T., Hetherington, A.L., Jackson, J.R., Balogh,  

C., Serfőző, Z., Hotaling, C.W., Zhukova, T.V., Mikheyeva, T.M., 

Kovalevskaya, R.Z., Makarevich, O.A., and D.V. Kruk. 2023. “Time scales of 

ecosystem impacts and recovery under individual and serial invasions.” 

Ecosystems 26: 1224–1237. https://doi.org/10.1007/s10021-023-00828-2  

 

Koeberle, A.L., Pearsall, W., Hammers, B.E., Mulhall, D., McKenna Jr, J.E., 

Chalupnicki, M., and S.A. Sethi. 2023. “Whole-lake acoustic telemetry to 

evaluate survival of stocked juvenile fish.” Scientific Reports 13(18956): 

18956. https://doi.org/10.1038/s41598-023-46330-6  

 

Koeberle, A.L., Hammers, B.E., Pearsall, W., Mulhall, D., Haley, A., Grausgruber, 

S.J., Chalupnicki, M., McKenna Jr, J.E., Cooch, E.G., Rudstam, L.G. and S.A. 

Sethi. 2025. “Integrating acoustic telemetry and demographic modeling to 

inform cisco (Coregonus artedi) restoration in Keuka Lake, New York.” 

Canadian Journal of Fisheries and Aquatic Sciences, 82: 1-15. 

https://doi.org/10.1139/cjfas-2025-0005  

 

Larson, G. and R. Schaetzl. 2001. “Origin and evolution of the Great Lakes.” Journal 

of Great Lakes Research 27(4): 518-546. https://doi.org/10.1016/S0380-

1330(01)70665-X 

https://doi.org/10.1007/s10021-023-00828-2
https://doi.org/10.1038/s41598-023-46330-6


 

61 

 

Lathrop, R.C., Johnson, B.M., Johnson, T., Vogelsang, M., Carpenter, S.R., 

Hrabik, T., Kitchell, J.F., Magnuson, J., Rudstam, L.G., and R.S. Stewart. 

2002. “Stocking piscivores to improve fishing and water clarity: a synthesis of 

the Lake Mendota biomanipulation project.” Freshwater Biology 47(12): 

2410–2424. https://doi.org/10.1046/j.1365-2427.2002.01011.x  

 

Lepak, J.M., Hansen, A.G., Johnson, B.M., Battige, K., Cristan, E.T., Farrell, C.J., 

Pate, W.M., Rogers, K.B., Treble, A.J., and T.E. Walsworth. 2025. “Cyclical, 

multi-trophic-level responses to a volatile, introduced forage fish: Learning 

from four decades of food web observation to inform management.” Fisheries 

50(2): 52–65. https://doi.org/10.1093/fshmag/vuae013  

 

Lepak, J.M. and C.E. Kraft. 2008. “Alewife mortality, condition, and immune 

response to prolonged cold temperatures.” Journal of Great Lakes Research 

34(1): 134–142. 

https://doi.org/10.3394/03801330(2008)34%255B134:AMCAIR%255D2.0.CO;2  

 

Lesser, J.S., Bruel, R., Marcy-Quay, B., McReynolds, A.T., Stockwell, J.D., and  

J.E. Marsden. 2024. “Whole-lake food web model indicates alewife invasion 

fueled lake trout restoration and altered patterns of trophic flow Lake 

Champlain.” Journal of Great Lakes Research 50(1): 102249. 

https://doi.org/10.1016/j.jglr.2023.102249  

 

Lynam, C.P., Llope, M., Möllmann, C., Helaouët, P., Bayliss-Brown, G.A., and 

N.C. Stenseth. 2017. “Interaction between top-down and bottom-up control in 

marine food webs.” Proceedings of the National Academy of Sciences 114(8): 

1952–1957. https://doi.org/10.1073/pnas.1621037114  

 

Maciej Gliwicz, Z. 2002. “On the different nature of top-down and bottom-up 

effects in pelagic food webs.” Freshwater Biology 47(12): 2296–2312. 

https://doi.org/10.1046/j.1365-2427.2002.00990.x  

 

Madenjian, C.P., O’Gorman, R., Bunnell, D.B., Argyle, R.L., Roseman, E.F., 

Warner, D.M., Stockwell, J.D., and M.A. Stapanian. 2008. “Adverse effects of 

alewives on Laurentian Great Lakes fish communities.” North American 

Journal of Fisheries Management 28(1): 263–282.  

https://doi.org/10.1577/M07-012.1  

 

https://doi.org/10.1046/j.1365-2427.2002.01011.x
https://doi.org/10.1093/fshmag/vuae013
https://doi.org/10.1073/pnas.1621037114


 

62 

Madenjian, C.P., Rutherford, E.S., Stow, C.A., Roseman, E.F. and J.X. He.2013. 

“Trophic shift, not collapse.” Environmental Science & Technology 47(21), 11915-

11916. https://doi.org/10.1021/es404089y 

 

McCann, K.S. and N. Rooney. 2009. “The more food webs change, the more they 

stay the same.” Philosophical Transactions of the Royal Society B: Biological 

Sciences 364(1524): 1789-1801. https://doi.org/10.1098/rstb.2008.0273  

 

McMeans, B.C., McCann, K.S., Tunney, T.D., Fisk, A.T., Muir, A.M., Lester, N., 

Shuter, B. and N. Rooney. 2016. “The adaptive capacity of lake food webs: 

from individuals to ecosystems.” Ecological Monographs 86(1): 4-19. 

https://doi.org/10.1890/15-0288.1  

 

McQueen, D.J., Johannes, M.R., Post, J.R., Stewart, T.J., and D.R. Lean. 1989. 

“Bottom-up and top-down impacts on freshwater pelagic community 

structure.” Ecological Monographs 59(3): 289–309. 

https://doi.org/10.2307/1942603  

 

McQueen, D.J., Post, J.R., and E.L. Mills. 1986. “Trophic relationships in 

freshwater pelagic ecosystems.” Canadian Journal of Fisheries and Aquatic 

Sciences 43(8): 1571–1581. https://doi.org/10.1139/f86-195  

 

Mehner, T. 2010. “No empirical evidence for community‐wide top‐down control 

of prey fish density and size by fish predators in lakes.” Limnology and 

Oceanography 55(1): 203-213. https://doi.org/10.4319/lo.2010.55.1.0203  

 

Mills, E.L. and J.L. Forney. 1988. Trophic dynamics and development of 

freshwater pelagic food webs, in: Complex Interactions in Lake Communities. 

Springer, pp. 11–30. 

 

Moyano, M., Illing, B., Akimova, A., Alter, K., Bartolino, V., Börner, G., 

Clemmesen, C., Finke, A., Gröhsler, T., Kotterba, P., Livdane, L., Mittermayer, 

F., Moll, D., Von Nordheim, L., Peck, M.A., Schaber, M., and P. Polte. 2022. 

“Caught in the middle: bottom-up and top-down processes impacting 

recruitment in a small pelagic fish.” Reviews in Fish Biology and Fisheries 33: 

55-84. https://doi.org/10.1007/s11160-022-09739-2  

 

Mrnak, J.T., Sikora, L.W., Zanden, M.J.V., and G.G. Sass. 2023. “Applying 

panarchy theory to aquatic invasive species management: a case study on 

https://doi.org/10.1098/rstb.2008.0273
https://doi.org/10.1139/f86-195
https://doi.org/10.4319/lo.2010.55.1.0203
https://doi.org/10.1007/s11160-022-09739-2


 

63 

invasive rainbow smelt Osmerus mordax.” Reviews in Fisheries Science & 

Aquaculture 31(1): 66–85. https://doi.org/10.1080/23308249.2022.2078951  

 

Myers, R.A., Barrowman, N.J., Hutchings, J.A., and A.A. Rosenberg. 1995. 

Population dynamics of exploited fish stocks at low population levels. Science 

269(5227): 1106–1108. https://doi.org/10.1126/science.269.5227.1106  

 

Nalepa, T.F., Fanslow, D.L., Pothoven, S.A., Foley III, A.J., and G.A. Lang. 2007. 

“Long-term trends in benthic macroinvertebrate populations in Lake Huron 

over the past four decades.” Journal of Great Lakes Research 33(2): 421–436. 

https://doi.org/10.3394/0380-1330(2007)33[421:LTIBMP]2.0.CO;2  

 

Norden, C.R. 1967. “Age, growth and fecundity of the alewife, Alosa 

pseudoharengus (Wilson), in Lake Michigan.” Transactions of American 

Fisheries Society 96(4): 387-393. 

https://doi.org/10.1577/15488659(1967)96[387:AGAFOT]2.0.CO;2  

 

Odell, T.T. 1934. “The life history and ecological relationships of the alewife 

(Pomolobus pseudoharengus-Wilson) in Seneca Lake, New York.” 

Transactions of the American Fisheries Society 64(1): 118-126.  

 

O’Gorman, R. And T.J. Stewart. 1999. “Ascent, dominance, and decline of the 

alewife in the Great Lakes: food web interactions and management strategies.” 

Great Lakes policy and management: a binational perspective 489–514. 

 

O'Gorman R., Lantry B.F., and C.P. Schneider. 2004. “Effect of stock size, climate, 

predation, and trophic status on recruitment of alewives in Lake Ontario, 

1978–2000.” Transactions of the American Fisheries Society 133(4): 855–867. 

https://doi.org/10.1577/T03-016.1  

 

Pace, M.L., Cole, J.J., Carpenter, S.R., and J.F. Kitchell. 1999. “Trophic cascades 

revealed in diverse ecosystems.” Trends in ecology & evolution 14(12): 483 

488. https://doi.org/10.1016/S0169-5347(99)01723-1  

 

Paine, R.T. 1980. “Food webs: linkage, interaction strength and community 

infrastructure.” Journal of Animal Ecology 49(3): 667–685. 

https://doi.org/10.2307/4220  

 

Palkovacs, E.P., Dion, K.B., Post, D.M., and A. Caccone. 2008. “Independent 

https://doi.org/10.1080/23308249.2022.2078951
https://doi.org/10.1126/science.269.5227.1106
https://doi.org/10.1016/S0169-5347(99)01723-1


 

64 

evolutionary origins of landlocked alewife populations and rapid parallel 

evolution of phenotypic traits.” Molecular Ecology 17(2): 582–597. 

https://doi.org/10.1111/j.1365-294X.2007.03593.x  

 

Parker-Stetter, S.L., Rudstam, L.G., Sullivan, P.J., and D.M. Warner. 2009. Standard 

operating procedures for fisheries acoustic surveys in the Great Lakes. Great Lakes 

Fisheries Commission Special Publication 2009-01.  

 

Peck, M.A., Alheit, J., Bertrand, A., Catalán, I.A., Garrido, S., Moyano, M., 

Rykaczewski, R.R., Takasuka, A., and C.D. van Der Lingen. 2021. “Small 

pelagic fish in the new millennium: a bottom-up view of global research 

effort.” Progress in Oceanography 191: 102494. 

https://doi.org/10.1016/j.pocean.2020.102494  

 

Porath, M.T., Peters, E.J. and D.L. Eichner. 2003. “Impact of alewife introduction 

on walleye and white bass condition in Lake McConaughy, Nebraska, 1980 

1995.” North American Journal of Fisheries Management 23(3): 1050-1055. 

https://doi.org/10.1577/M02-058  

 

Post, D.M., Palkovacs, E.P., Schielke, E.G., and S.I. Dodson. 2008. “Intraspecific 

variation in a predator affects community structure and cascading trophic 

interactions.” Ecology 89(7) 2019–2032. https://doi.org/10.1890/07-1216.1  

 

Post, D.M., and A.W. Walters. 2009. “Nutrient excretion rates of anadromous 

alewives during their spawning migration.” Transactions of the American 

Fisheries Society 138(2): 264–268. https://doi.org/10.1577/T08-111.1  

 

Pritchard, A.L. 1929. “The alewife (Pomolobus pseudoharengus) in Lake 

Ontario.” University of Toronto Studies Biological Series Publication 33, 

Publication of the Ontario Fisheries Research Laboratory 38. 

 

R Core Team. 2025. R: a language and environment for statistical computing. R 

foundation for statistical computing, Vienna. Available from 

https://www.rproject.org. 

 

Reavie, E.D., Barbiero, R.P., Allinger, L.E., and G.J. Warren. 2014. 

“Phytoplankton trends in the Great Lakes, 2001–2011.” Journal of Great Lakes 

Research 40(3): 618–639. https://doi.org/10.1016/j.jglr.2014.04.013  

 

https://doi.org/10.1890/07-1216.1
https://doi.org/10.1577/T08-111.1
https://doi.org/10.1016/j.jglr.2014.04.013


 

65 

Ridgway, M.S., Hurley, D., and K. Scott. 1990. “Effects of winter temperature and 

predation on the abundance of alewife (Alosa pseudoharengus) in the Bay of 

Quinte, Lake Ontario.” Journal of Great Lakes Research 16(1): 11–20. 

https://doi.org/10.1016/S0380-1330(90)71394-9  

 

Ridgway, M.S., Piette‐Lauzière, G., Bell, A.H. and J. Turgeon. 2020. “Isotopic 

niche size of Coregonus artedi (sensu lato) increases in the presence of Mysis 

diluviana, expanded habitat use and phenotypic diversity.” Ecology and Evolution 

10(20): 11713-11726. https://doi.org/10.1002/ece3.6807  

 

Riley, S.C. and E.S. Dunlop. 2016. “Misapplied survey data and model uncertainty 

result in incorrect conclusions about the role of predation on alewife population 

dynamics in Lake Huron: a comment on.” Canadian Journal of Fisheries and Aquatic 

Sciences 73(5), 860–864. https://doi.org/10.1139/cjfas-2015-0237 

 

Riley, S.C., Roseman, E.F., Nichols, S.J., O’Brien, T.P., Kiley, C.S., and J.S. 

Schaeffer. 2008. “Deepwater demersal fish community collapse in Lake 

Huron.” Transactions of the American Fisheries Society 137(6): 1879–1890. 

https://doi.org/10.1577/T07-141.1  

 

Robson, D., and D. Chapman. 1961. “Catch curves and mortality rates.” 

Transactions of the American Fisheries Society 90(2): 181–189. 

https://doi.org/10.1577/1548-8659(1961)90[181:CCAMR]2.0.CO;2  

 

Rogers, T.L., Bashevkin, S.M., Burdi, C.E., Colombano, D.D., Dudley, P.N., 

Mahardja, B., Mitchell, L., Perry, S., and P. Saffarinia. 2024. “Evaluating top 

down, bottom-up, and environmental drivers of pelagic food web dynamics 

along an estuarine gradient.” Ecology 105(4), e4274. 

https://doi.org/10.1002/ecy.4274  

 

Rose, K.A., Cowan, J.H., Winemiller, K.O., Myers, R.A., and R. Hilborn. 2001. 

“Compensatory density dependence in fish populations: importance, 

controversy, understanding and prognosis.” Fish and Fisheries 2(4): 293–327. 

https://doi.org/10.1046/j.1467-2960.2001.00056.x  

 

Rudstam, L.G., Brooking, T.E., Krueger, S.D., Jackson, J.R., and L. Wetherbee. 

2011. “Analysis of compensatory responses in land‐locked alewives to walleye 

predation: a tale of two lakes.” Transactions of the American Fisheries Society 

140(6): 1587–1603. https://doi.org/10.1080/00028487.2011.639265  

https://doi.org/10.1002/ece3.6807
https://doi.org/10.1080/00028487.2011.639265


 

66 

 

Rudstam, L.G., Lathrop, R.C., and S. Carpenter. 1993. “The rise and fall of a 

dominant planktivore: direct and indirect effects on zooplankton.” Ecology 

74(2): 303–319. https://doi.org/10.2307/1939294  

 

Rudstam, L.G., Parker-Stetter, S.L., Sullivan, P.J., and D.M. Warner. 2009. 

“Towards a standard operating procedure for fishery acoustic surveys in the 

Laurentian Great Lakes, North America.” ICES Journal of Marine Science 

66(6): 1391–1397. https://doi.org/10.1093/icesjms/fsp014  

 

Rudstam, L.G., Schaner, T., Gal, G., Boscarino, B.T., O’Gorman, R., Warner, 

D.M., Johannsson, O.E., and K.L. Bowen. 2008. “Hydroacoustic measures of 

Mysis relicta abundance and distribution in Lake Ontario.” Aquatic Ecosystem 

Health & Management 11(4): 355–367. https://doi.org/10.1080/14634980802539708  

 

Rudstam, L.G., Watkins, J.M., Currie, W., and K.E. Bowen. 2025. Sub-indicator: 

Mysis. State of Great Lakes 2022 Technical Reports. 

 

Shapiro, J., Lamarra, V., and M. Lynch. 1975. “Biomanipulation and ecosystem 

approach to lake restoration”, in: Brezonik, P.L., Fox, J.L. (Eds.), Proceedings of a 

Symposium on Water Quality Management through Biological Control, Gainesville, 

pp. 85-96.  

 

Sharma, S., Filazzola, A., Nguyen, T., Imrit, M.A., Blagrave, K., Bouffard, D., 

Daly, J., Feldman, H., Feldsine, N., Hendricks-Franssen, H.-J., Granin, N., 

Hecock, R., L’Abée-Lund, J.H., Hopkins, E., Howk, N., Iacono, M., Knoll, 

L.B., Korhonen, J., Malmquist, H.J., Marszelewski, W., Matsuzaki, S.S., 

Miyabara, Y., Miyasaka, K., Mills, A., Olson, L., Peters, T.W., Richardson, 

D.C., Robertson, D.M., Rudstam, L.G., Wain, D., Waterfield, H., 

Weyhenmeyer, G.A., Wiltse, B., Yao, H., Zhdanov, A., and J.J. Magnuson. 

2022. “Long-term ice phenology records spanning up to 578 years for 78 lakes 

around the Northern Hemisphere.” Scientific Data 9: 318. 

https://doi.org/10.1038/s41597-022-01391-6  

 

Sharma, S., Richardson, D.C., Woolway, R.I., Imrit, M.A., Bouffard, D., Blagrave, 

K., Daly, J., Filazzola, A., Granin, N., Korhonen, J., Magnuson, J., 

Marszelewski, W., Matsuzaki, S.S., Perry, W., Robertson, D.M., Rudstam, 

L.G., Weyhenmeyer, G.A., and H. Yao. 2021. “Loss of ice cover, shifting 

phenology, and more extreme events in Northern Hemisphere lakes.” Journal 

https://doi.org/10.2307/1939294
https://doi.org/10.1093/icesjms/fsp014


 

67 

of Geophysical Research: Biogeosciences 126(10): e2021JG006348. 

https://doi.org/10.1029/2021JG006348  

 

Smith, S.H., 1970. “Species interactions of the alewife in the Great Lakes.” 

Transactions of the American Fisheries Society 99(4): 754–765. 

https://doi.org/10.1577/1548-8659(1970)99%3C754:SIOTAI%3E2.0.CO;2  

 

Snyder, R.J. and T.M. Hennessey. 2003. “Cold tolerance and homeoviscous 

adaptation in freshwater alewives (Alosa pseudoharengus).” Fish Physiology 

and Biochemistry 29(2): 117-126. 

https://doi.org/10.1023/B:FISH.0000035920.60817.11  

 

Stein, R.A., DeVries, D.R. and J.M. Dettmers. 1995. “Food-web regulation by a 

planktivore: exploring the generality of the trophic cascade hypothesis.” 

Canadian Journal of Fisheries and Aquatic Sciences 52(11): 2518-2526. 

https://doi.org/10.1139/f95-842  

 

Stewart, D.J., and F.P. Binkowski. 1986. “Dynamics of consumption and food 

conversion by Lake Michigan alewives: an energetics-modeling synthesis.” 

Transactions of the American Fisheries Society 115(5): 643–661. 

https://doi.org/10.1577/1548-8659(1986)115%3C643:DOCAFC%3E2.0.CO;2  

 

Stewart, T.J. and W.G. Sprules. 2011. “Carbon-based balanced trophic structure 

and flows in the offshore Lake Ontario food web before (1987–1991) and after 

(2001–2005) invasion-induced ecosystem change.” Ecological Modelling 

222(3): 692-708. https://doi.org/10.1016/j.ecolmodel.2010.10.024  

 

Stewart, T.J., Sprules, W.G., and R. O’Gorman. 2009. “Shifts in the diet of Lake 

Ontario alewife in response to ecosystem change.” Journal of Great Lakes 

Research 35(2): 241–249. https://doi.org/10.1016/j.jglr.2008.12.010  

 

Tunney, T., McCann, K., Lester, N., and B.J. Shuter. 2012. “Food web expansion 

and contraction in response to changing environmental conditions.” Nature 

Communications 3(1): 1105. https://doi.org/10.1038/ncomms2098  

 

Van Donk, E., M.P. Grimm, R.D. Gulati, P.G.M. Heuts, W.A. de Kloet, and L. Van 

Liere. 1990. “First attempt to apply whole-lake food-web manipulation on a large 

scale in the Netherlands.” Hydrobiologia 200(1), 291-301. 

https://doi.org/10.1007/BF02530347  

https://doi.org/10.1016/j.ecolmodel.2010.10.024
https://doi.org/10.1016/j.jglr.2008.12.010
https://doi.org/10.1038/ncomms2098


 

68 

 

Vander Zanden, M.J., and Y. Vadeboncoeur. 2002. “Fishes as integrators of benthic 

and pelagic food webs in lakes.” Ecology 83(8): 2152–2161. 

https://doi.org/10.1890/0012-9658(2002)083[2152:FAIOBA]2.0.CO;2  

 

Wang, R.W., Rudstam, L.G., Brooking, T.E., Snyder, D.J., Arrigo, M.A., and E.L. 

Mills. 2010. “Food web effects and the disappearance of the spring clear water 

phase in Onondaga Lake following nutrient loading reductions.” Lake and 

Reservoir Management 26(3): 169–177. 

https://doi.org/10.1080/07438141.2010.504070 

 

Walsworth, T.E., Hansen, A.G. and J.M. Lepak. “Alternating ecosystem states 

driven by an invasive fish in a life-history intraguild predation system.” 

Canadian Journal of Fisheries and Aquatic Sciences 82: 1-15.  

https://doi.org/10.1139/cjfas-2025-0012  

 

Warner, D.M., Kiley, C.S., Claramunt, R.M., and D.F. Clapp. 2008. “The influence 

of alewife year-class strength on prey selection and abundance of age-1 

chinook salmon in Lake Michigan.” Transactions of the American Fisheries 

Society 137(6): 1683–1700. https://doi.org/10.1577/T07-130.1  

 

Warner, D.M., Rudstam, L.G., and R.A. Klumb. 2002. “In situ target strength of 

alewives in freshwater.” Transactions of the American Fisheries Society 

131(2): 212–223. 

https://doi.org/10.1577/15488659(2002)131%253C0212:ISTSOA%253E2.0.CO;2 

 

Warren, L.D., Honsey, A.E., Bunnell, D.B., Collingsworth, P.D., Hondorp, D.W., 

Madenjian, C.P., Warner, D.M., Weidel, B.C. and T.O. Hӧӧk. 2024. 

“Synchrony of alewife, Alosa pseudoharengus, year-class strength in the Great 

Lakes region.” Canadian Journal of Fisheries and Aquatic Sciences 81(10): 

1456-1467. https://doi.org/10.1139/cjfas-2023-0322  

 

Watkins, J.M., Rudstam, L.G., Mills, E.L., and M.A. Teece. 2012. Coexistence of 

the native benthic amphipod Diporeia spp. and exotic dreissenid mussels in the 

New York Finger Lakes. Journal of Great Lakes Research 38(2): 226–235. 

https://doi.org/10.1016/j.jglr.2012.02.001  

 

Wells, L. 1970. “Effects of alewife predation on zooplankton populations in Lake 

Michigan.” Limnology and Oceanography 15(4): 556–565. 

https://doi.org/10.1577/T07-130.1
https://doi.org/10.1139/cjfas-2023-0322


 

69 

https://doi.org/10.4319/lo.1970.15.4.0556     



 

70 

Tables 

 

Table 1: Benthic species abundance estimated from surveys conducted before and 

after alewife Alosa pseudoharengus collapse in Keuka Lake, New York, USA. 

ANOVA results showed changes in Diporeia spp. densities across years (F(2,56) = 34.7, 

p < 0.01), with Tukey’s HSD indicating declines from pre- to post-alewife collapse 

(2010-2019; p < 0.01) and increases by 2025 (p = 0.023). Zebra mussel (Dreissena 

polymorpha) densities also declined (ANOVA F(2,56) = 3.86, p = 0.027), with lower 

densities observed post-alewife collapse (2010-2019 and 2010-2025; Tukey’s HSD 

both p < 0.05). No change was observed in quagga mussel (Dreissena bugensis) 

densities among years (ANOVA F(2,56) = 2.17, p = 0.12). Chironomid densities 

decreased marginally between pre- and post-alewife collapse from 2010-2019 

(ANOVA F(1,36) = 4.02, p = 0.053), while oligochaete densities did not differ (ANOVA 

F(1,36) = 0.79, p = 0.38). Species-group densities (individuals/m2) are given as the mean 

(standard error).  

 

Benthic species Pre-alewife 

collapse (2010) 

Post-alewife 

collapse (2019) 

Post-alewife 

collapse (2025) 

Dreissena bugensis 580 (262) 135 (42) 267 (91) 

Dreissena polymorpha 172 (93) 0 (0) 2 (2) 

Diporeia spp. 704 (91) 70 (15) 273 (36) 

Chironomidae 117 (45) 31 (9) N/A 

Oligochaeta 192 (48) 136 (40) N/A 

 

  



 

71 

Table 2: Predator-prey consumption rates in Keuka Lake, New York, USA and Otsego 

Lake, New York, USA associated with alewife (Alosa pseudoharengus) collapse. 

Calculations are provided as mean (range), based on fish densities estimated from fall 

hydroacoustic surveys. Alewife were considered extirpated from Otsego Lake after 

2013. Mysids (Mysis diluviana) are present only in Keuka Lake. 

 

Study system Scenario Years Alewife per top 

piscivore 

Mysid per alewife 

Otsego Lake Pre-alewife 

collapse 

1998-2009 738 (56-3,209) N/A 

Otsego Lake Post-alewife 

collapse 

2010-2012 9 (4-14) N/A 

Keuka Lake Pre-alewife 

collapse 

2000-2011 30 (2-52) 1,040 (863-1,216) 

Keuka Lake Post-alewife 

collapse 

2016-2023 5 (5-5) 18,988 (14,8940-

23,036) 

 

 



 

72 

Table 3: Lake trout (Salvelinus namaycush) consumption exceeds alewife (Alosa 

pseudoharengus) production, suppressing alewife recovery following their population 

collapse in Keuka Lake, New York, USA. Fish biomass estimates are derived from 

hydroacoustic density data and average fish size (mass, g) from netting surveys. In 

survey years without alewife mass, length (mm) was converted to mass using length-

weight parameters from FishBase (Froese and Pauly 2025;  

<https://www.fishbase.se/summary/1583> accessed 20 October 2025). For simplicity, 

lake trout consumption is assumed to consist entirely of alewife. All estimates are in 

situ unless otherwise noted.  

 

aQ/B = Consumption to biomass ratio. 

bP/B = Production to biomass ratio for age-0 alewife from Stewart and Sprules (2011). 
cZ = Instantaneous mortality rate for age-1+ alewife from catch curve analysis (Appendix S1: Figures 

S5, S6). 
dLake trout life history parameters from Fish Base (Froese and Pauly 2025;  

<https://www.fishbase.se/summary/248> accessed 20 October 2025). 
eHydroacoustic densities from the 2023 survey include introduced top predator walleye (Sander vitreus) 

in the epilimnion (< 20m depths). 

  



 

73 

Figures 

 

Figure 1. Long-term fisheries surveys revealed alewife (Alosa pseudoharengus) 

population collapse after 2015 in Keuka Lake, New York, USA and after 2010 in 

Otsego Lake, New York, USA, with population recovery failure in both study 

systems. (a, b) Alewife density (fish / hectare) estimated from hydroacoustic surveys 

declined by an order of magnitude (Keuka Lake: aggregated small targets assumed to 

represent young-of-year (age-0) alewife and medium targets assumed to represent 

yearling-and-older (age-1+) alewife; Otsego Lake: forage fish estimates assumed to be 

all alewife ages). Across gear types, (c, d) alewife catch (fish / net) declined to no 



 

74 

alewife caught after 2007 in Keuka Lake and after 2010 in Otsego Lake cold-water 

surveys designed to sample lake trout (Salvelinus namaycush), (e, f) forage gillnet net 

catches in Keuka Lake declined significantly to small numbers of alewife after 

collapse, while gillnet surveys ceased after alewife collapse in Otsego Lake, and (h) 

no alewife were caught after 2011 in Otsego Lake long-term trap net surveys designed 

to capture spawning adult alewives. Alewife are presently considered extirpated from 

Otsego Lake. (g) An adult alewife from their introduced landlocked range. Gray boxes 

show periods of alewife decline (light gray) and alewife collapse (dark gray). Photo 

credit: alewife (Alosa pseudoharengus) provided by Brad Hammers, New York State 

Department of Environmental Conservation. 

  



 

75 

 

Figure 2. Time series of mean annual (June-September) water quality in Keuka Lake, 

New York, USA (shaded blue points) and Otsego Lake, New York, USA (shaded 

yellow points). (a, b) Water clarity, indicated by Secchi disk depth (m), increased in 

both study lakes over the time series (Generalized additive model GAM, Keuka Lake 

edf = 1.8, deviance explained = 43%, p < 0.01; Otsego Lake edf = 2.7, deviance 

explained = 73.4%, p < 0.01). (c, d) Epilimnetic nitrate concentrations increased in 

Otsego Lake during the alewife collapse period (GAM, Keuka Lake edf = 1.0, 

deviance explained = 1.5%, p = 0.5; Otsego Lake edf = 2.2, deviance explained = 

32.8%, p = 0.03). By contrast, epilimnetic (e, f) total phosphorus concentrations 

(GAM, Keuka Lake edf = 2.0, deviance explained = 17.3%, p = 0.13; Otsego Lake edf 



 

76 

= 1.0, deviance explained = 51.6%, p < 0.01) and (g, h) chlorophyll a concentrations 

(GAM, Keuka Lake edf = 2.6, deviance explained = 65.1%, p < 0.01; Otsego Lake edf 

= 2.4, deviance explained = 83.8%, p < 0.01) decreased in Otsego Lake across the 

time series, while they generally decreased in Keuka Lake from the early 1990s up to 

the late 2000s as alewife declined. A red dashed horizontal line indicates the long-term 

average. Gray boxes show periods of alewife (Alosa pseudoharengus) decline (light 

gray) and alewife collapse (dark gray) while black vertical lines indicate the arrival of 

Dreissena spp., zebra mussels (dotted lines) and quagga mussels (dashed lines). 

  



 

77 

 

Figure 3. Epilimnetic volumetric biomass (µg dry weight / L) of zooplankton 

estimated from vertical net tows increased during and immediately after alewife 

(Alosa psuedoharengus) collapse in both study systems. (a) In Keuka Lake, mean total 

zooplankton biomass increased marginally from 12.6 µg / L pre-alewife collapse in 

1998-2013 to 21.2 µg / L post-alewife collapse in 2014-2024 (Welch’s two-sample t-

test, t(13) = 1.6, p = 0.1). (b) In Otsego Lake, mean total zooplankton biomass 



 

78 

decreased from 89.2 µg / L pre-alewife collapse in 2002-2010 to 42.5 µg / L post-

alewife collapse in 2011-2024 (Welch’s two-sample t-test, t(17) = -5.2, p < 0.01). In 

both panels, gray boxes indicate periods of alewife decline (light gray) and alewife 

collapse (dark gray) while black vertical lines indicate the arrival of Dreissenid spp., 

zebra mussels (dotted lines) and quagga mussels (dashed lines). Zooplankton species-

group ‘Pred.Clad.Cerco’ includes predatory cladocerans Leptodora kindtii and 

Cercopagis pengoi. Note, y-axis scales differ between plots. 

  



 

79 

 

Figure 4. Zooplankton size increased after alewife (Alosa pseudoharengus) collapsed. 

(a, b) Size frequency distributions shifted from smaller zooplankton pre-alewife 

collapse to larger zooplankton post-alewife collapse in both lakes. (c, d) Boxplots 

(median and interquartile range) of Daphnia spp. total length (mm) revealed a shift to 

larger sizes immediately after alewife collapse (Welch’s two sample t-test, Keuka 

Lake t(16) = -3.6, p < 0.01; Otsego Lake t(11) = -7.5, p < 0.01). In Keuka Lake, Daphnia 

retrocurva dominated in earlier years and Daphnia pulicaria in later years. Yet, (e, f) 

Daphnia spp. abundance (mean annual density individuals / m3) did not change after 

alewife collapse in Keuka Lake (Welch’s two-sample t-test, t(17) = -0.32, p = 0.8) 

while abundance declined significantly in Otsego Lake (Welch’s two-sample t-test, 

t(7.7) = 5.8, p < 0.01). We removed years 1992-1997 for Keuka Lake size comparisons 

due to known issues with measuring tablets. Inset photo credits: Daphnia retrocurva 

(left) provided by Christopher Marshall, Cornell University; Daphnia pulicaria (right) 

provided by Joseph Connolly, Cornell University.  



 

80 

 

Figure 5. Whole-water column netting in Keuka Lake, New York, USA after alewife 

collapse shows high mysid (Mysis diluviana) abundance (October 2023 survey, blue; 

June 2024 survey, yellow) revealed high mysid abundances in Keuka Lake, New York, 

USA after the alewife (Alosa pseudoharengus) collapse. (a) Hydroacoustic-based 

mysid densities (gold points are east-west transect means as the sampling unit; shaded 

area is ± 1 SE) increased by 26% after alewife collapse (dark gray shaded area). (b) 

October 2023 average mysid biomass (g/m2 dry weight) exceeded 2023 Great Lakes 

estimates (data provided by Great Lakes National Program Office, US Environmental 

Protection Agency): Erie (summer), Huron (September), Michigan (August), Ontario 

(September), and Superior (summer). (c) Length-frequency distributions indicate a 1-

year life cycle (weighted random sample: 50 measured individuals/site; Oct-2023 ten 

stations; Jun-2024 four stations). (d) Mysid density (number/m2) increased marginally 

with depth (GAM, edf = 1.0, deviance explained = 37%, p = 0.09). Great Lakes 

samples were collected at > 60 m depths where mysids are rare in shallower waters; 

however, Keuka Lake has a maximum depth of 57 m. Inset photo credit: mysid (Mysis 

diluviana) provided by Alexander Koeberle, Cornell University.  



 

81 

 

Figure 6. Top predator lake trout (Salvelinus namaycush) and walleye (Sander 

vitreus) response to alewife (Alosa pseudoharengus) decline (light gray boxes) and 

collapse (dark gray boxes) in Keuka Lake, New York, USA and Otsego Lake, New 

York, USA. (a, b) Hydroacoustic data show declining predator fish abundances with 

alewife collapse. Large targets represent adult lake trout (pre-2016 in Keuka Lake, 

pre-2000 in Otsego Lake), shifting to a mixture of walleye and lake trout. (c, d) In 

Keuka Lake, lake trout catch rates appeared to increase up to alewife decline then 

decrease (but not statistically significant) across three periods: 1979-1994 before 

alewife decline, 1997-2011 alewife decline, and 2016-2022 post-alewife collapse 

(ANOVA, F(2, 11) = 2.78, p = 0.11), but post-hoc tests suggest a marginal decline after 



 

82 

alewife collapsed (Tukey’s HSD, p = 0.09). In Otsego Lake, lake trout catch differed 

marginally across three periods: 1994-2006 before alewife decline, 2008-2010 alewife 

decline, and 2012-2022 post-alewife collapse (ANOVA, F(2, 12) = 3.83, p = 0.052), 

with a marginal decline from before alewife decline to post-alewife collapse (Tukey’s 

HSD, p = 0.08). (e) Adult lake trout survival in Keuka Lake remained stable over time 

(GAM, edf = 1.0, deviance explained = 10%, p = 0.29). (f) Walleye catch in Otsego 

Lake was high during alewife collapse but declined after stocking ceased in 2014. (g, 

h) Lake trout growth varied by lake and size class. In Keuka Lake, lake trout growth 

increased significantly for small fish (GAM, edf = 2.8, deviance explained = 70.9%, p 

= 0.01), decreased significantly for medium fish (GAM, edf = 1.0, deviance explained 

= 75.2%, p < 0.01), and no large fish were caught after alewife collapse (GAM, edf = 

1.0, deviance explained = 19.6%, p = 0.23). In Otsego Lake, lake trout growth 

remained stable for small fish (GAM, edf = 2.0, deviance explained = 30.8%, p = 

0.21), but decreased significantly for medium fish (GAM, edf = 2.8, deviance 

explained = 72.3%, p < 0.01) and large fish (GAM, edf = 1.0, deviance explained = 

49.9%, p = 0.01) after alewife collapse. In both panels, small lake trout are 300-

499mm, medium lake trout are 500-649mm, and large lake trout are 650-799mm 

standard length. In Otsego Lake panels, vertical dot-dash lines indicate a period of 

walleye stocking (2000-2014) to control the alewife population. 

  



 

83 

 

Figure 7. Time series of winter ice cover from 1970-2025. Each winter season is 

labeled by the year it concludes, e.g., winter 2024-2025 is designated as 2025. Keuka 

Lake ice cover data were unavailable, so we evaluated comparable lakes in the region, 

including: (a) maximum percent ice cover for Lake Ontario in the Great Lakes, (b) 

December-March winter growing degree days (GDD) from Penn Yan Airport weather 

station near Keuka Lake, and (c) ice cover duration for Lake Ontario basin inland 

lakes Cazenovia Lake and Oneida Lake, New York. (d) We also evaluate ice cover 

duration for the study system Otsego Lake. Dashed lines indicate the long-term mean 

for each lake, and gray boxes indicate periods of alewife (Alosa pseudoharengus) 

decline (light gray) and alewife collapse (dark gray).   



 

84 

CHAPTER 2 

WHOLE-LAKE ACOUSTIC TELEMETRY TO EVALUATE SURVIVAL OF 

STOCKED JUVENILE FISH  

 

Abstract: 

Estimates of juvenile survival are critical for informing population dynamics and the 

ecology of fish, yet these demographic parameters are difficult to measure. Here, we 

demonstrate that advances in animal tracking technology provide opportunities to 

evaluate survival of juvenile tagged fish. We implemented a whole-lake telemetry 

array in conjunction with small acoustic tags (including tags < 1.0g) to track the fate 

of stocked juvenile cisco (Coregonus artedi) as part of a native species restoration 

effort in the Finger Lakes region of New York, USA. We used time-to-event modeling 

to characterize the survival function of stocked fish, where we infer mortality as the 

cessation of tag detections. Survival estimates revealed distinct stages of juvenile cisco 

mortality including high immediate post-release mortality, followed by a period of 

elevated mortality during an acclimation period. By characterizing mortality over 

time, the whole-lake biotelemetry effort provided information useful for adapting 

stocking practices that may improve survival of stocked fish, and ultimately the 

success of the species reintroduction effort. The combination of acoustic technology 

and time-to-event modeling to inform fish survival may have wide applicability across 

waterbodies where receiver arrays can be deployed at scale and where basic 

assumptions about population closure can be satisfied. 

 

Introduction 

Mortality is a key process governing population growth and features heavily in 

fisheries conservation and management, informing species ecology, harvest limits, and 



 

85 

stocking programs [1, 2]. Yet direct estimation of mortality is difficult to achieve, 

particularly with conventional marking approaches such as mark-recapture for small 

or juvenile fishes [3, 4]. Juvenile fish survival is believed to be an important driver of 

population dynamics, where mortality through this critical period influences the 

trajectory of cohorts through later adult life stages [5, 6]. Here, we demonstrate that 

contemporary biotelemetry technology and time-to-event modeling unlock new 

opportunities for survival modeling, even with small fishes. We applied whole-lake 

acoustic telemetry to track the fate of re-introduced juvenile cisco (Coregonus artedi) 

equipped with small transmitters (0.6g and 3.5g) in the Finger Lakes region of New 

York State, USA. 

An important application of biotelemetry technology is to elucidate survival of 

juvenile stocked fish [7, 8, 9]. Releases of hatchery-reared fish are widely utilized to 

augment existing fish populations or to reestablish formerly extirpated or depleted 

populations [1]. Assessment of the fate of stocked fish, however, is typically lacking, 

and this information is needed to determine if augmentation efforts meet long-term 

conservation and management goals [3, 10, 11, 12]. For example, mortality can 

provide information useful to manage fishery harvest rates and to assess whether 

stocking efforts can lead to a self-sustaining population [1, 13]. Due to cost and 

logistical constraints, hatchery programs typically focus on juvenile life stage 

production, necessitating survival assessment approaches that can be applied to small 

fishes, such as small telemetry devices coupled with spatially extensive receiver arrays 

[7]. Stocking success, often evaluated as the contribution of hatchery fish to enhance 

wild fish stocks or the fitness of fish, is difficult to measure, and net positive effects of 

hatchery stocking have long been disputed [13]. Nonetheless, increased availability of 

approaches to estimate survival can in turn help hatchery managers evaluate and adapt 

stocking programs. 



 

86 

Biotelemetry uses electronic transmitters (hereafter referred to as ‘tags’) to track the 

movement and fate of study animals [14, 15]. Technological advances in biotelemetry 

include tag miniaturization (e.g., tags < 1.0g), longer tag and receiver battery life, and 

increased tag detection ranges [16, 17]. Thus, while tag sizes previously restricted 

biotelemetry applications to relatively large specimens that can accommodate large 

telemetry tags with minimal adverse effects on survival, contemporary tags have been 

sufficiently miniaturized to provide ecological insight to a diverse age and size range 

of both freshwater and marine fish species [7, 18, 19]. Currently, the smallest 

biotelemetry tags are < 0.22g injectable acoustic tags [20].  

Coupled with spatially extensive acoustic receiver arrays, acoustic tags provide the 

opportunity to estimate demographic parameters such as mortality [21]. If arrays are 

sufficiently dense to continuously detect tagged individuals, the fate of tagged fish can 

be determined. Assuming that tagged fish move throughout their environment and in 

systems closed to immigration and emigration, mortalities are reflected by two 

detection signals. First, the cessation of tag detections, within the period of active tag 

battery life, can indicate a fish perished in regions outside the detection range of 

receivers. Second, continuous tag detections at a single receiver for a repeat time 

interval can indicate a fish perished within the detection range of a receiver [9]. 

Acoustic tags and spatially extensive receiver arrays can therefore yield so called 

‘time-to-event’ data whereby both the status of a tagged fish (alive or dead) and the 

time of mortality since release can be inferred [22, 23]. Time-to-event data enable 

powerful mortality modeling approaches that can characterize the survival function of 

tagged fish [24]. The survival function provides information on the time path of 



 

87 

mortality and can accommodate the influence of subject-level covariates [25]. Thus, in 

addition to estimating survival up to any point in time, time-to-event modeling can be 

used to understand the influence of biological and environmental covariates on 

survival. 

In recent years, native coregonine (Coregonus spp.) restoration has been attempted by 

fishery scientists and managers to improve ecosystem integrity throughout the 

Laurentian Great Lakes basin of North America [26, 27, 28]. In the Lake Ontario 

basin, cisco are a pelagic cold-water species that was historically a major component 

of the native forage fish assemblage and had high commercial, cultural, and 

recreational significance to coastal communities [29]. Cisco populations have sharply 

declined throughout the basin due to a combination of fishing pressure, habitat and 

water quality degradation, and competition from non-native aquatic species [30, 31, 

32]. Some ongoing cisco restoration efforts involve reintroductions or augmentations 

from hatchery stocking programs; however, post-release survival rates are not well-

described for this species in the Great Lakes region [32, 33]. This lack of survival 

information on stocked cisco currently hampers both the ability to assess stocking 

success and to adjust stocking practices to achieve conservation, restoration, or 

management targets for this species. 

In this study, we track the fate of hatchery-reared juvenile cisco reintroduced into 

Keuka Lake, New York, USA. The fate of a subset of released fish equipped with 

acoustic tags were monitored through an extensive lake-wide array of acoustic 

receivers which autonomously logged detections as tagged fish swam within the 

detection range of a receiver. We infer that the cessation of tag detections on the array 



 

88 

for a given tagged fish indicates a mortality event. Our objectives were to: 1) 

Implement a whole-lake scale passive acoustic telemetry array to assess the fate of 

acoustically-tagged fish, 2) Analyze the individual fates of tagged cisco using time-to-

event models that characterize the time path of survival for released fish, and 3) 

Interpret stocked fish survival information to support assessment and adaptation of 

stocking efforts to restore extirpated cisco in Keuka Lake. Our results indicate 

mortality events could be clearly discerned for juvenile cisco using tag detections, 

meeting conditions for time-to-event survival modeling. This study suggests that 

contemporary acoustic biotelemetry may have wide applicability in estimating 

survival of small fish in waterbodies with sufficient receiver coverage and that meet 

basic assumptions about population closure. 

 

Methods  

Study system 

Keuka Lake is an inland waterbody located in the Finger Lakes region of New York, 

USA and is within the Lake Ontario basin of the Laurentian Great Lakes. Known for 

its distinct ‘y-shape’, Keuka Lake has a 57m maximum depth, 31m mean depth, 3km 

maximum width, and a total surface area of 46.88km2 (Fig. 1) [34]. Keuka Lake also 

contains a popular native lake trout (Salvelinus namaycush) fishery. In addition to 

providing an important fishery resource in Keuka Lake, the lake trout population is 

unique within the region as it is self-sustained through ‘wild’ recruitment, whereas 

most other lake trout populations in the region are supported with hatchery 

augmentation. Thus, maintaining lake conditions to conserve lake trout recruitment in 



 

89 

Keuka Lake is a management priority by New York State Department of 

Environmental Conservation (NYSDEC). Cisco were historically the dominant prey 

fish species for predators like lake trout, however competition with introduced, non-

native alewife (Alosa pseudoharengus) and rainbow smelt (Osmerus mordax) and 

degraded water quality from high nutrient input led to their extirpation in Keuka Lake 

by the 1990s. In recent years, non-native fish populations have declined, and water 

quality has improved from nutrient management as the lake is undergoing 

oligotrophication. Resource managers hypothesize that future limnological conditions 

favor cisco over non-native alewife, and therefore seek to bolster fish community 

resilience by reintroducing cisco to restore this native species [28]. 

 

Acoustic-tagging and hatchery releases 

From 2019 to 2020, 296,979 juvenile cisco were stocked from regional hatcheries into 

Keuka Lake across multiple release cohorts (Table 1). We tagged and released 210 

juvenile cisco of two age classes (10 months post hatching, hereafter referred to as 

age-0; and 19 or 22 months post hatching, hereafter referred to as age-1) with Juvenile 

Salmon Acoustic Telemetry System (JSATS) Acoustic Micro Transmitters (AMT; 

Fig. 2; Table 1). These are small acoustic tags that emit uniquely coded sound pulses 

on pre-programmed frequency and time intervals to track individual fish across a fixed 

acoustic receiver array [7]. Age-0 fish were surgically implanted with 0.6g tags, and 

age-1 fish were implanted with 3.5g tags. All experimental protocols for sampling and 

handling of fish were approved by U.S. Geological Survey and the New York State 

fish collections review office. Fish collections occurred under Scientific License to 



 

90 

Collect or Possess Permit #2977 and were in accordance with all fish handling, animal 

care, and ethical use requirements governing fish collection permits as granted by 

NYSDEC. Additionally, all methods for this study are reported in accordance with 

Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. Acoustic 

tag implantations followed protocols described by McKenna et al. [33] that would lead 

to low to no tag-related mortality, rather than strictly following a 2% tag size to body 

size rule. This protocol included holding all tagged fish onsite for at least 14 days 

before release to ensure tag retention and to monitor their post-surgery survival and 

condition. We did not observe any mortalities or tag expulsions among study fish prior 

to their release into Keuka Lake, consistent with findings by McKenna et al. [33]. All 

acoustic transmitters were programmed with a 20-second transmission interval (e.g., 

‘ping rate’). At this ping rate, we conservatively anticipated a minimum expected 

battery lifespan of 262 days for the 0.6g tags and 1,218 days for the 3.5g tags (80% of 

full battery life according to manufacturer specifications). See SI 1.1 for further 

information on surgery procedures and tag specifications. 

 

Acoustic receiver array 

We configured 24 acoustic receivers throughout Keuka Lake to form a lake-wide 

telemetry array to maximize spatial coverage for detecting post-release movements 

and the fate of tagged cisco after stocking (Fig. 1). We deployed acoustic receivers 

(Lotek-brand Wireless Hydrophone System 4250 Series), rated to 400m depths, with 

four lithium primary D-cell batteries to increase operation lifespan. Each receiver was 

serviced every three to four months to download autonomously logged data of tag 



 

91 

detections and to refresh lithium batteries. Prior to releasing tagged fish, we quantified 

receiver efficiency by assessing the detection of study tags at increasing distances 

from moored receivers in the lake. We found that acoustic receivers had a linear 

diameter detection range of approximately 200 to 300m (see SI 1.2 for further acoustic 

receiver specifications, range testing, and receiver array details).   

Acoustic receiver moorings were anchored on the lake bottom for the duration of the 

study, with all receiver units facing upward in the water column (Fig. S1). Because 

cisco are a pelagic cold-water species, we expected tagged fish to primarily inhabit 

deep waters that range from below a thermocline, which forms between 9-11m during 

summer months, to depths near the bottom of Keuka Lake (fish could occupy depths 

up to approximately 57m) [31, 35]. Hence, we did not suspect deflection or 

interference of tag transmissions from seasonal thermal strata due to their habitat 

preference [36]. Additionally, three acoustic receivers were stationed midway in each 

branch to increase coverage for offshore and nearshore movements of tagged cisco 

between lake branches. We deployed a receiver upstream (stream wetted width = 20m) 

within the Keuka Lake outlet and a receiver downstream (stream wetted width 100m) 

at the confluence of Seneca Lake to determine if any tagged fish emigrated from the 

study system (Fig. S1). All tagged and untagged cisco in this study were stocked 

offshore via boat in the northwestern branch (Fig. 1). 

 

Detection filtering  

As pelagic schooling fish, cisco move widely [37]. In Keuka Lake, we expected that 

surviving cisco would be capable of moving among branches at a lake-wide scale and 



 

92 

at deep depths where receivers were moored. Thus, we assumed that if a tagged fish 

was alive, it would be detected on the whole-lake acoustic receiver array; in contrast, 

we inferred a mortality event when a tagged fish ceased to be detected by any receiver 

across the array. All datafiles downloaded from acoustic receivers were first processed 

with acoustic data processing software (Lotek WHS Host software; V1.8.3942.3) to 

remove detections that did not correspond with a fish tag identification (hereafter tag 

ID), i.e., false detections [9, 38]. Subsequent detection datafiles were post-processed 

using a combination of custom filtering scripts, detection functions (‘GLATOS’ 

package [39] in program R; version 4.2.2, R Core Team 2022), and manual inspection 

of each tagged fish’s data to obtain detection histories. 

To distinguish true positive detections from spurious detections on receivers, we 

removed detections that 1) did not correspond to a known tag ID, 2) occurred past the 

expected tag battery lifespan, 3) failed to meet the minimum requirements for 

consecutive detections (required ≥ two detections on the same receiver ≤ 4-minutes 

apart), and 4) occurred outside a 20-second transmission rate window (±2-seconds at 

each 20-second multiple within 4-minutes). These criteria, including the 4-minute 

interval, were determined from the tag transmission rate and receiver detection range, 

inferred swimming behavior of this species, as well as detection criteria from studies 

with similar telemetry equipment [39, 40, 41, 42]. We retained 19.3% of total 

detections using these criteria for a final dataset of true tag detections. To visualize 

detections across the receiver array, we collapsed individual detections into discrete 

‘detection events’ for each tagged fish to account for arrival and departure at a 

receiver location [39]. Detection histories for each fish revealed when tagged fish 



 

93 

ceased to be detected, i.e., died outside the detection range of receivers. Once a tagged 

cisco went undetected, we inferred mortality as a function of its last known time 

detected and a detection interval. For all tagged cisco observed on the receiver array, 

we calculated the average time between observed detections for individual fish. 

Mortality, or inferred time-to-death, was therefore calculated as the last time detected 

on the array plus half the average time interval between tag detections. For further 

information on our detection filtering approach for this study see SI 1.3.   

 

Modeling assumptions 

Lake-wide acoustic receiver deployment provides the opportunity to track tagged 

cisco across their lifespan as ‘time-to-event’ data, where both the outcome of interest 

(death or censorship) and the event time (days after release) are known [25, 43]. For 

time-to-event models of biotelemetry data, we assumed that the status of an individual 

animal is always known and that survival probabilities are equal for censored and 

uncensored individuals [43, 44]. To meet assumptions for subsequent survival 

analyses, we inferred that all tagged fish have an equal probability of detection across 

the receiver array (no systematic detection heterogeneity). Further, we assumed that 

the acoustic-transmitter burden does not affect survival of tagged fish, as supported by 

hatchery cisco survival and tag retention trials [33] such that survival estimates from 

tagged fish represent population-level survival rates. Lastly, some tagged fish could be 

consumed by a piscivorous predator, and the tag could therefore be at large for some 

time period in the predator’s stomach before being excreted. Estimated predator tag 

retention time varies by tag size, and false detection rates have been reported [9]. We 



 

94 

assumed that the two different sized tags, which were ≤ 3.5g and used in this study, 

would likely be excreted within one to three days. This is consistent with existing 

literature of comparable tag sizes [38, 45, 46] and would thus have a low impact on 

inferred survival times.        

To meet assumptions for tracking the lifespan of tagged study fish, we assume Keuka 

Lake is a closed system. Because cisco were extirpated and stocked fish did not have 

sufficient time to reach reproductive maturity, we assumed no cisco other than those 

stocked entered the system. Closed population status was confirmed by a lack of tag 

detections at receivers positioned in the exit point (Keuka Outlet; Fig. S1). Combined, 

the lack of additions and non-mortality related losses imply that estimated mortality 

rates reflect true cisco survival, uncontaminated by immigration, emigration, or 

recruitment.  

 

Survival analysis 

A major challenge in quantifying survival rates of stocked fish is detecting individuals 

over the period of their life history under study. For example, conventional mark-

recapture approaches for estimating apparent survival such as Cormack-Jolly-Seber 

models [47, 48] necessitate repeated captures of individuals and require estimation of 

both detection processes and survival processes [8, 21]. A spatially extensive whole-

lake acoustic array coupled with continuously transmitting tags provides time-to-event 

data whereby status is inferred by tag detections (alive) or cessation of detections 

(dead). Accordingly, probability of detection across the receiver array is equal to one 

because the fate of tagged fish is known. 



 

95 

We used two time-to-event modeling approaches to evaluate survival of stocked 

juvenile cisco. First, we visualized empirical summaries of survival outcomes using 

Kaplan-Meier curves [49]. Subsequently, we applied Cox proportional hazards models 

to characterize the survival function of juvenile stocked cisco and evaluate whether 

subject-level covariates influence fish survival [25, 50]. The primary output of Cox 

proportional hazards modeling is a survival function, S(t), which provides predicted 

survival rates up to a specified time, t, and which can incorporate covariate effects on 

survival if available. Through survival functions, time-to-event model approaches such 

as Cox proportional hazards models provide the ability to estimate survival up to any 

desired time while also providing information on the shape of the survival rate over 

the study period. Right censorship is a common feature of time-to-event data, whereby 

the observation period on a given subject terminates prior to that individual 

experiencing mortality [21, 25]. Time-to-event models can accommodate censorship 

events; however, no fish in this study survived to their expected tag battery lifespan or 

past the end of acoustic receiver monitoring. Thus, right censorship events did not 

contribute to our survival estimates. 

 We considered a suite of covariates that could affect juvenile stocked cisco 

survival rates. Stocked fish covariate data were collected before release and included 

total length (mm), mass (g), and age-at-release (age-0 or age-1). We used the mean 

cohort mass and total length for one fish with missing size measurements from Fall 

2019 (age-0 fish). We also estimated Fulton’s body condition factor (K) as 100,000 × 

mass × length-3 [51]. Although survival rates may vary by sex, juvenile cisco cannot 

be reliably sexed from visual observation, so this information was not available for 



 

96 

modeling. We tested for collinearity among variables and found positive correlation 

between age, length, and mass. Consequently, we explored models with age or size 

(mass or length) metrics, but not both as an interaction. We did not detect correlation 

between age and K. Length and mass were measured at release and, along with K, 

were not treated as time-varying covariates for survival analyses. We explored the 

support for covariate effects using multi-model inference based on Akaike’s 

Information Criterion adjusted for small sample size (AICc) [52]. Thus, we fit 17 total 

Cox proportional hazards models to the time-to-event dataset of tagged fish released 

from October 2019 to October 2020, including release year, age-at-release, length, 

mass, and K, along with interaction terms that allowed for testing size effects within 

age cohorts. 

 

Results 

Whole-lake biotelemetry 

 Tagged cisco moved widely throughout Keuka Lake, with detections 

confirmed at all receivers in the whole-lake acoustic array. Cisco exhibited preference 

for some regions of Keuka Lake, with the most tagged fish activity found in the 

northwestern branch (near the stocking site) to the receiver gate midway in the south 

branch of the lake (Fig. 1). A lower frequency of tag detections was observed in the 

northeastern branch and at the confluence in the middle of the lake (Fig. 1). The mean 

time between successive detections for tagged fish at large was 1.92 days (se ± 0.17; 

range 0-10.4 days). Empirical Kaplan-Meier survival curves indicate high initial 

mortality at release and sustained high mortality before a few individuals transitioned 



 

97 

to a long-term low mortality regime (Fig. 3a-c). Age-1 cisco show substantially higher 

initial and long-term survival compared to age-0 stocked cisco (Table 2). The 

maximum observed survival of tagged age-1 fish was estimated as 405 days after 

release, whereas maximum observed survival of tagged age-0 fish was estimated as 

152 days after release.  

 

Time-to-event survival models 

The candidate model set investigated the effects of individual-level covariates on 

survivorship. We did not detect any structural deficiencies in fitted Cox models, 

indicating that these regressions satisfied goodness-of-fit testing (see SI 2.1) [23, 25]. 

Visual inspection of Schoenfeld and Martingale residuals of each covariate tested in 

our most general Cox proportional hazards models only showed minor deviations from 

β = 0, and deviance residuals failed to indicate highly influential observations. We 

therefore proceeded with multi-model inference to evaluate the strength of support for 

each candidate Cox survival models.  

Multi-model inference results highlighted the importance of age-at-release on juvenile 

cisco survival. The top AICc-ranked Cox model included a single covariate: age-at-

release (Table 3; k = 1, AICc 1753.95, and 42% AICc weight). The next top-ranked 

model within 2 ΔAICc [52], includes the covariates year and age-at-release (k = 2, 

1.87 ΔAICc, 16% AICc weight). Additionally, we found moderate support for a model 

that included age and condition (k = 2, 2.03 ΔAICc, 15% AICc weight). Our three top 

AICc-ranked models account for 76% of total model set support by AICc weight 

(Table 3). Combined across the model set, age-at-release had > 99% relative AICc 



 

98 

variable importance, followed by year (28%), condition (28%), length (13%), and 

mass (<1%) (Table 4). The top AICc-ranked Cox model, Survival ~ Age, reveals 

significant differences in predicted survival, 𝑆̂, between tagged fish released at age-0 

versus at age-1 (Log-rank test for the model p <0.01, χ2 = 85.16, 1 df; Fig. 4). For 

example, the estimated coefficient, 𝛽̂, for age in this model and the corresponding 

hazard ratio, 𝑒𝛽̂, revealed that tagged age-1 fish experience 75% less mortality than 

tagged age-0 fish (𝛽̂age-1 = -1.39, 𝑒𝛽̂ = 0.25, and 95% confidence limits = 0.18, 0.34). 

While stocking year and condition factor featured in the next top AICc-ranked models 

and had moderate relative AICc variable importance, neither variable was found to be 

statistically significant (p > 0.05) in any of the fitted models in which they were 

included. This indicates that although we observed minor differences in predicted 

survival given release year and condition factor, age-at-release still emerges as a top 

predictor of fish survival. 

 

Stocked juvenile cisco survival dynamics 

Cox survival models indicate long-term survival of few juvenile cisco after initial 

release into Keuka Lake. Predicted survival from top AICc-ranked Cox models 

indicate age-1 cohorts have higher immediate post-release survival (time (t) < 1-day) 

and longer overall survival than age-0 cohorts (Fig. 4). For example, survival curves 

predict high initial mortality within the first day of release (t < 1 day), particularly 

with age-0 cisco which experienced 77% initial mortality (𝑆̂(𝑡=1)= 0.23), in contrast to 

30% initial mortality (𝑆̂(𝑡=1)= 0.70) with age-1 cisco (Table 2; Fig. 4). From the top-



 

99 

ranked Cox model, Survival ~ Age, the probability of an age-1 juvenile cisco 

surviving a full year (i.e., 𝑆̂(𝑡=365)) was 0.02 (95% confidence limits = <0.001, 0.099), 

whereas the annual survival rate predicted for stocked age-0 cisco asymptotes at zero.  

 

Discussion 

Whole-lake biotelemetry coupled with contemporary tag technology enables novel 

opportunities to estimate survival of juvenile and small fish released into natural 

aquatic systems. In Keuka Lake, time-to-event models provided survival estimates for 

stocked cisco and generated insights needed to assess the efficacy of native species 

reintroduction efforts. By evaluating the time path of survival for stocked cisco, time-

to-event data provide insight into fine-scale temporal patterns in survival beyond 

approaches that assess an overall survival rate over a fixed time interval such as 

through binomial regression [23, 25]. For example, Cox models fitted to stocked cisco 

data in Keuka Lake reveal that juvenile fish mortality was highest during the initial 

post-stocking period (within one day of release), and that longer term survival was 

dependent on age-at-release. Survival models indicated that mortality within the first 

day after release was particularly significant for age-0 fish (<25% survival) compared 

to age-1 stocked fish (70% survival). These results provide fishery managers with an 

indication of the effect of using different sized hatchery fish on the likely success of 

restoring a native cisco population.  

The high mortality upon release of juvenile cisco observed in Keuka Lake may stem 

from a suite of potential factors, including avian predation, predation by lake trout (an 

abundant top predator in the lake), and physiological stress. Avian predation was 



 

100 

witnessed by project biologists at stocking, and fish predators were observed on 

hydroacoustic units from boats during stocking. Local anglers captured two lake trout 

with two depredated tagged cisco found in stomach contents shortly after stocking, 

providing anecdotal but direct evidence of high potential for piscivore predation (SI 

3.1; Fig. S6). Reduced survival from handling, transport, and stocking of hatchery-

reared fish is well-documented [1, 13], and these factors could also have contributed to 

mortality from physiological stress for stocked cisco in this study. Tag burden could 

have also contributed to mortality (see SI 1.1); however, we followed protocols 

predicted to lead to low tagging-related mortality according to a study design to 

evaluate juvenile cisco response to JSATS tag implantation [33]. Further, we 

monitored tagged fish at the hatchery prior to release and did not find evidence of poor 

condition or mortality from tagging, suggesting that tag burden may not have been a 

significant source of stocking mortality. Moving forward, we anticipate opportunities 

to use predation sensor tags [9, 38] or to assess survival outcomes for cohorts released 

under alternative stocking practices that could help disentangle and quantify specific 

mortality drivers for stocked juvenile fish. 

We found that juvenile cisco in both age classes survived the initial release then 

experienced an acclimation period of elevated mortality, with a midpoint of this 

acclimation period of approximately t = 30 days after release for age-0 fish and 

approximately t = 50 days for age-1 fish. Although no age-0 fish survived greater than 

six months after release, we found approximately 15% survival rate to six months in 

age-1 fish, with only one age-1 fish surviving more than one year after stocking (2% 

annual survival rate). While this is the first study to yield survival estimates for this 



 

101 

species with time-to-event data, comparable studies elsewhere in the Great Lakes 

region of North America estimate low survival rates of other coregonine species such 

as bloater (Coregonus hoyi). For example, population modeling and trawl surveys 

estimate <20% initial apparent survival of stocked juvenile bloater [27] and low initial 

apparent survival (≤42%) of tagged age-1 bloater through acoustic telemetry in Lake 

Ontario [53]. From our survival analysis, few individuals survived to the end of the 

study, and therefore it is unlikely that enough cisco from those experimental releases 

survived to reproductive maturity, which would be required to establish a self-

sustaining population in Keuka Lake. Nonetheless, mortality rates for both age classes 

appeared to decrease and stabilize after approximately t = 60 days. This suggests that 

if fish survive through an acclimation period of high mortality, remaining individuals 

might enter a natural mortality regime allowing for some long-term survival. 

Time-to-event survival estimation provides insight into predictors of tagged fish 

mortality by constructing models that investigate the association of subject-level 

covariates with estimated hazard rates for mortality. Our top-ranked Cox proportional 

hazards models predict higher initial survival and longer-term survival for age-1 

stocked fish than for age-0 stocked fish. These results may indicate that there is a size 

threshold for stocked juvenile cisco needed to escape high rates of predation from 

birds and fishes [54, 55]. Additionally, large fish may have higher energy reserves to 

persist through physiological stress at release, as well as to survive through a period of 

famine as stocked fish adapt to lake foraging conditions [56, 57, 58]. Such factors 

could have limited the ability of juvenile cisco, particularly age-0 fish, in this study to 

acclimate to lake depths, navigate to suitable habitats, or find suitable prey sources for 



 

102 

feeding [29, 53]. Future experiments could thus include stocking large juvenile cisco 

as a possible mechanism for predation avoidance via predator gape limitation and to 

survive through acclimation of the lake environment.  

Post-release survival results for stocked juvenile cisco in Keuka Lake highlight the 

importance of evaluating stocking outcomes through the fate of hatchery-reared fish 

inferred from acoustic telemetry. We found that a small proportion of stocked age-0 

cisco survived through early stages after their release, comparable to similar findings 

of acoustic-tagged juvenile bloater [53]. Results indicate that age-1 cisco experience 

higher survival than age-0 stocked fish, which aligns with results from other studies 

[59, 60]. Rearing pelagic schooling fish, such as cisco, to age-1 or greater requires 

significant hatchery space and is costly at a sufficiently large scale. Instead, survival 

curves estimated for juvenile stocked fish in Keuka Lake suggest that alternative 

release practices may improve chances of cisco restoration success. For example, 

applying estimates from the top AICc-ranked fitted Cox model (Age-0 cisco 𝑆̂(𝑡=1)= 

0.23, and assuming a fixed long-term survival rate of the upper 95% confidence limit 

𝑆̂(𝑡=180)= 0.004), a 25% reduction in first day post-release mortality would lead to a 

predicted 83% increase in survival of age-0 fish to six months. Reducing initial 

mortality through alternative stocking practices shows promise in increasing the 

number of fish that survive through high mortality periods and into a long-term 

survival regime to potentially reach reproductive maturity [1, 13].   

In addition to survival, detection histories can also provide insight into movement and 

habitat use of tagged fish. Sections of the lake between receivers lacked detection 

coverage, and thus fine-scale spatial insights were beyond the scope of this study. 



 

103 

Generally, we found that tagged cisco surviving past initial release moved widely 

throughout the lake with preference by both age classes for the west and south 

branches of Keuka Lake. Tagged age-0 fish that survived t ≥ 30 days (n = 4) were 

detected on 8.3 receivers on average (range 4, 14) and age-1 fish that survived t ≥ 50 

days were detected on 7.9 receivers on average (range 1, 19), which demonstrates the 

spatial ecology utility of acoustic receiver arrays. Tagged cisco that survived high 

post-release mortality to an acclimation period therefore exhibited their expected high 

movement behavior [31, 37]. 

Survival estimates from this study can be used to adapt stocking practices and improve 

the probability of successfully achieving management goals for cisco restoration [28]. 

Our approach provides continuous survival monitoring from the onset of this 

reintroduction effort, as opposed to evaluating success after multiple years of stocking 

via detection or capture of surviving adults. In Keuka Lake, acoustic telemetry 

revealed that survival rates are too low under current stocking practices to re-establish 

native forage populations, thus informing updates to stocking practices to reduce 

stocked fish mortality. Managers sought stocking age-0 cisco after the lake de-

stratified in October to allow lake trout to distribute throughout the water column, 

thereby overwhelming predators with numerous prey targets [27]. Several hatchery 

release practices may show promise in reducing initial mortality of stocked juvenile 

fish. Net pen stocking – where fish are held for a period of weeks in situ in enclosed 

pens and provided feed to assist in transition to novel environmental conditions – may 

assist juvenile cisco in averting high initial predation, facilitate acclimation to lake 

conditions, and transition to wild defensive behaviors such as schooling [61]. For 



 

104 

example, in both marine and freshwater fisheries, net pen-stocking has been used for 

stocking several Salmonine species including Pacific salmon (Oncorhynchus spp.) [1]. 

In addition to net pen-stocking, stocking at multiple release sites and at night may 

improve survival outcomes by assisting stocked fish in avoiding predator gauntlets. 

Previous studies have shown that cisco have a predator-driven diel migration [26, 53], 

which could potentially be leveraged to mitigate juvenile cisco overlap with predatory 

birds and fish through nighttime releases but requires investigation. 

This study shows that acoustic telemetry technology can track the fate of tagged fish 

in a deep inland lake. Our approach may be well-suited for supporting fisheries 

research and conservation efforts in such inland lake settings which are widely 

distributed across many landscapes. While this technology may be scalable to 

waterbodies of similar depth and area, it may be unfeasible to deploy arrays at 

sufficient density to meet survival modeling assumptions used here in larger lakes or 

marine settings. In settings that fail to meet closed population assumptions (e.g., 

tagged fish can leave the area monitored by a receiver array), mortality and apparent 

survival estimates are often calculated from tracking electronically tagged fish through 

a series of spatially coordinated acoustic receivers whereby directional swim rates 

between receivers, constant tag transmissions, or movement off the array are 

incorporated along with detection probability [24, 53, 62, 63]. Nevertheless, telemetry 

receiver coverage has continued expanding in larger lake settings such as the Great 

Lakes, USA and Canada and Lake Champlain, USA as well as marine settings along 

the Pacific and Atlantic Coasts, USA and Canada (e.g., Atlantic Cooperative 

Telemetry Network, Pacific Ocean Shelf Tracking project) and Australia (e.g., 



 

105 

Integrated Marine Observing System in Australia) [8, 64]. As acoustic technology 

advances, receiver network coverage will continue to provide finer spatial and 

temporal coverage of life history stages of tagged fish [21]. 

Results from the Keuka Lake whole-lake biotelemetry study demonstrate that novel 

acoustic technology coupled with time-to-event modeling provided valuable estimates 

of survival rates of tagged juvenile fish. While long-term monitoring requires 

maintenance of acoustic receiver arrays for extended periods, time-to-event models are 

well suited to such data, accommodating censorship events that may arise from field 

challenges. For example, time-to-event models can produce unbiased survival 

estimates even in the presence of ‘interval’ censor events. Such events can arise if fish 

perish during windows of time when acoustic receiver arrays may be inoperable or 

inaccessible due to field conditions or equipment failures. While previous tag size 

restricted telemetry applications to relatively larger fish, small acoustic tags now have 

sufficient battery life to monitor juvenile fish outcomes for periods of months or 

longer, as demonstrated in this study. In addition to tag miniaturization, emerging 

biotelemetry technology can capture additional information useful for disentangling 

sources of mortalities, such as depth and temperature [14, 15], acceleration [65, 66], 

and depredation [38] data. We anticipate growing application for biotelemetry to 

estimate fish survival in systems for which whole-waterbody acoustic receiver array 

coverage with known population closure can be maintained for periods spanning entire 

life cycles.  

 

Acknowledgements 



 

106 

The authors would like to thank multiple collaborating partners from various State and 

Federal agencies for their ongoing work on this project. We thank S. Robb, B. Carson, 

A. Thoms, B. Minges, and M. Sanderson (NYSDEC); K. Osika, A. Haley, K. Healy, 

K. Towner, A. Day, and C. Wlasniewski (NYSDEC Bath Hatchery); W. Evans, D. 

Domachowske, E. Stoddard, and M. Ferron (NYSDEC Oneida Hatchery); G. Mackey 

and J. Krause (USGS-Tunison); S. Schlueter (USFWS); M. Moss and M. Economos 

(NY Cooperative Fish and Wildlife Research Unit); and L. Rudstam and E. Cooch 

(Cornell University). We also thank C. Holbrook and J. Bergman for their valuable 

feedback on acoustic tag detection criteria, filtering, and data processing. A. Chalkman 

and P. Wigglesworth provided constructive guidance for Lotek-brand acoustic 

telemetry equipment. The project team thanks Keuka Lake Association, local anglers, 

and local marinas for their continued support for cisco restoration in Keuka Lake. We 

thank T. Brown and K. Fitzpatrick (Cornell University) for their helpful feedback on 

this manuscript. Lastly, we thank A. Honsey (USGS) who provided comments on an 

earlier draft that improved this piece. This research was funded by New York State 

Department of Environmental Conservation using Federal Aid Sport Fish Restoration 

Funds from “Grant F64-R”. Any use of trade, firm, or product names is for descriptive 

purposes only and does not imply endorsement by the U.S. Government. 

 

  



 

107 

References  

 

1. Brown, C. & R. L. Day. The future of stock enhancements: lessons for 

hatchery practice from conservation biology. Fish Fish. 3, 79-94 (2002). 

 

2. Hilborn, R. & C. J. Walters. Quantitative Fisheries Stock Assessment: Choice, 

Dynamics, and Uncertainty. Springer Science & Business Media (2013). 

 

3. Pine, W. E., Pollock, K. H., Hightower, J. E., Kwak, T. J. & J. A. Rice. A 

review of tagging methods for estimating fish population size and components of 

mortality. Fisheries. 28, 10 (2003). 

 

4. Sandford, M., Castillo, G., & Hung, T. C. A review of fish identification 

methods applied on small fish. Rev. Aquac. 12, 542-554 (2020). 

 

5. Beamish, R. J. & C. Mahnken. A critical size and period hypothesis to explain 

natural regulation of salmon abundance and the linkage to climate and climate change. 

Prog. Oceanogr. 49, no. 1-4: 423-437 (2001). 

 

6. Michel, C. J., Notch, J. J., Cordoleani, F., Ammann, A. J. & E. M. Danner. 

Nonlinear survival of imperiled fish informs managed flows in a highly modified river. 

Ecosphere, 12(5), e03498 (2021). 

 

7. McMichael, G. A., et al. The juvenile salmon acoustic telemetry system: a new 

tool. Fisheries. 35, 9-22 (2010). 

 

8. Dudgeon, C. L. et al. Integrating acoustic telemetry into mark–recapture 

models to improve the precision of apparent survival and abundance estimates. 

Oecologia. 178, 761–772 (2015).  

 

9. Klinard, N. V. & J. K. Matley. Living until proven dead: addressing mortality 

in acoustic telemetry research. Rev. Fish Biol. Fish. 30, 485-499. (2020). 

 

10. Galbreath, P. F., Bisbee, M. A., Dompier, D. W., Kamphaus, C. M. & T. H. 

Newsome. Extirpation and tribal reintroduction of coho salmon to the interior 

Columbia River Basin. Fisheries 39, 77-87 (2014).  

 

11. Stevens, J. R., Kocik, J. F. & T. F. Sheehan. Modeling the impacts of dams and 

stocking practices on an endangered Atlantic salmon (Salmo salar) population in the 

Penobscot River, Maine, USA. Can. J. Fish. Aquat. Sci. 76, 1795-1807 (2019). 

 

12. Hickerson, B. T., Grube, E. R., Mosher, K. R. & A. T. Robinson. Successful 

restoration of a native fish assemblage in the Blue River, Arizona. N. Am. J. Fish. 

Manag. 41, 746-756 (2021). 

 



 

108 

13. Araki, H. & C. Schmid. Is hatchery stocking a help or harm: Evidence, 

limitations and future directions in ecological and genetic surveys. Aquaculture, 308. 

S2-S11 (2010). 

 

14. Cooke, S. J. et al. Remote bioenergetics measurements in wild fish: 

opportunities and challenges. Comp. Biochem. Physiol. Part A: Mol. Integr. 

Physiol. 202, 23– 27 (2016). 

 

15. Crossin, G. T. et al. Acoustic telemetry and fisheries management. Ecol. 

Appl. 27, 4 (2017).  

 

16. Donaldson, M. R. et al. Making connections in aquatic ecosystems with 

acoustic telemetry monitoring. Front. Ecol. Environ. 12, 565-573 (2014). 

 

17. Matley, J. K., et al. Global trends in aquatic animal tracking with acoustic 

telemetry. Trends Ecol. Evo. 37, 79-94 (2022). 

 

18. Woodley, C. M. et al. Surgically implanted JSATS microacoustic transmitters 

effects on juvenile Chinook salmon and steelhead tag expulsion and survival. Battelle, 

Pacific Northwest National Laboratory, PNNL-20857. U.S. Army Corps of Engineers, 

Portland District; 57 pp. (2011). 

 

19. Cooke, S.J. et al. Tracking animals in freshwater with electronic tags: past, 

present and future. Anim. Biotelemetry 1, 5 (2013). 

 

20. Deng, Z. et al. An injectable acoustic transmitter for juvenile salmon. Sci. Rep. 

5, 8111 (2015). 

 

21. Lees, K. J., MacNeil, M. A., Hedges, K. J. & N. E. Hussey. Estimating 

demographic parameters for fisheries management using acoustic telemetry. Rev. Fish 

Biol. Fish. 31, 25-51 (2021). 

 

22. Heupel, M. R., Semmens, J. M. & A. J. Hobday. Automated acoustic tracking 

of aquatic animals: scales, design and deployment of listening station arrays. Mar. 

Freshw. Res. 57, 1-13 (2006). 

 

23. Rose, C. et al. Survival outcome patterns revealed by deploying advanced tags 

in quantity: Pacific halibut (Hippoglossus stenolepis) survivals after release from trawl 

catches through expedited sorting. Can. J. Fish. Aquat. Sci. 76, 12 (2019). 

 

24. Villegas‐Ríos, D., Freitas, C., Moland, E., Thorbjørnsen, S. H. & E. M. Olsen. 

Inferring individual fate from aquatic acoustic telemetry data. Methods Ecol. Evol. 11, 

1186-1198 (2020).  

 



 

109 

25. Hosmer, D. W., Lemeshow, S. & S. May. Applied Survival Analysis: 

Regression Modeling of Time-to-event Data. Wiley-Interscience 2nd ed. (2008).  

 

26. Stockwell, J. D., et al. A synthesis of cisco recovery in Lake Superior: 

implications for native fish rehabilitation in the Laurentian Great Lakes. N. Am. J. 

Fish. Manag. 29, 626-652 (2009). 

 

27. Weidel, B. C. et al. Results of the collaborative Lake Ontario bloater 

restoration stocking and assessment, 2012–2020. J. Great Lakes Res. 48, 371-380 

(2022). 

 

28. Bunnell, D. B. et al. A science and management partnership to restore 

coregonine diversity to the Laurentian Great Lakes. Environ. Rev. e-First 

https://doi.org/10.1139/er-2022-0109 (2023). 

 

29. Brown, T. A. et al. Contemporary spatial extent and environmental drivers of 

larval coregonine distributions across Lake Ontario. J. Great Lakes Res. 48, 359-370 

(2022). 

 

30. Schmidt, S. N., Harvey, C. J. & M. J. Vander Zanden. Historical and 

contemporary trophic niche partitioning among Laurentian Great Lakes coregonines. 

Eco. Appl. 21, 888-896 (2011). 

 

31. George, E. The history and ecology of Cisco Coregonus artedi in the 

Laurentian Great Lakes. Aquat. Eco. Health Manag. 22, 280-293 (2019). 

 

32. Rook, B. J., Hansen, M. J. & C. R. Bronte. How many ciscoes are needed for 

stocking in the Laurentian Great Lakes? J. Fish Wild. Manag. 13, 28-49 (2022). 

 

33. McKenna Jr, J. E., Sethi, S. A., Scholten, G. M., Kraus, J. & M. Chalupnicki. 

Acoustic tag retention and tagging mortality of juvenile cisco Coregonus artedi. J. 

Great Lakes Res. 47, 937-942 (2021). 

 

34. Bloomfield, J. A. Lakes of New York State, Volume 1: Ecology of the Finger 

Lakes. New York: Academic Press, Inc. (1978). 

 

35. Dryer, W. R. & J. Beil. Life history of lake herring in Lake Superior. Fish. 

Bull. 63, 493-530 (1964). 

 

36. Kuai, Y. et al. Strong thermal stratification reduces detection efficiency and 

range of acoustic telemetry in a large freshwater lake. Anim. Biotelemetry. 9, 46 

(2021). 

 

37. Milne, S. W., Shuter, B. J. & W.G. Sprules. The schooling and foraging 

ecology of lake herring (Coregonus artedi) in Lake Opeongo, Ontario, Canada. Can. 

https://doi.org/10.1139/er-2022-0109


 

110 

J. Fish. Aquat. Sci. 62, 1210-1218 (2005). 

 

38. Halfyard, E. A. et al. Evaluation of an acoustic telemetry transmitter designed 

to identify predation events. Methods Ecol. Evol. 8, 1063-1071 (2017). 

 

39. Holbrook, C. et al. glatos: A Package for the Great Lakes Acoustic Telemetry 

Observation System. (2019). <https://gitlab.oceantrack.org/GreatLakes/glatos>.  

 

40. Algera, D. A. et al. Stranded kokanee salvaged from turbine intake 

infrastructure are at low risk for reentrainment: A telemetry study in a hydropower 

facility forebay. North Am. J. Fish. Manag. 40, 6 (2020). 

 

41. Tuononen, E. I., Cooke, S. J., Timusk, E. R. & K. E. Smokorowski. Extent of 

injury and mortality arising from entrainment of fish through a very low head 

hydropower turbine in central Ontario, Canada. Hydrobiologia. 849, 407-420 (2020). 

 

42. Bergman, J. N. et al. Tracking the early stages of an invasion with 

biotelemetry: behaviour of round goby (Neogobius melanostomus) in Canada’s 

historic Rideau Canal. Biol. Invasions. 24, 1149-1173 (2022). 

 

43. Miller Jr., R. G. Survival Analysis 2nd ed. Wiley Classics Library, New York. 

(1998). 

 

44. White, G. C. & R. A. Garrott. Analysis of Wildlife Radio-Tracking Data. 

Elsevier. (1990). 

 

45. Schultz, A. A., Kumagai, K. K., & B. B. Bridges. Methods to evaluate gut 

evacuation rates and predation using acoustic telemetry in the Tracy Fish Collection 

Facility primary channel. Anim. Biotelemetry. 3, 1-9 (2015). 

 

46. Daniels, J., Sutton, S., Webber, D., & J. Carr. (2019). Extent of predation bias 

present in migration survival and timing of Atlantic salmon smolt (Salmo salar) as 

suggested by a novel acoustic tag. Anim. Biotelemetry. 7, 1-11 (2019). 

 

47. Jolly G. M. Explicit estimates from capture–recapture data with both death and 

immigration–stochastic model. Biometrika, 52, 225–247 (1965). 

 

48. Seber G. A. A note on the multiple-recapture census. Biometrika. 52, 249–259. 

(1965). 

 

49. Kaplan, E. L. & and P. Meier. (1958). Nonparametric estimation from 

incomplete observations. J. Am. Stat. Assoc. 53, 457-481 (1958). 

 



 

111 

50. Cox, D. R. Regression models and life-tables. J. Royal Stat. Soc. Series B 

(Methodological). 34, 2 (1972). 

 

51. Fulton, T. W. The rate of growth of fishes. Fisheries Board of Scotland, Annual 

Report 22 part 3, 141–241 (1904). 

 

52. Burnham, K. P. & D. R. Anderson. Model Selection and Multimodel Inference: 

A Practical Information-theoretic Approach. Springer 2nd ed. (2002). 

 

53. Klinard, N. V. et al. Post‐stocking movement and survival of hatchery‐reared 

bloater (Coregonus hoyi) reintroduced to Lake Ontario. Freshw. Biol. 65, 6 (2020). 

 

54. Maule, A. G., Schreck, C. B., Bradford, C. S., & B. A. Barton, B. A. 

Physiological effects of collecting and transporting emigrating juvenile chinook 

salmon past dams on the Columbia River. Trans. Am. Fish. 117, 245-261 (1988). 

 

55. Evans, A. F. et al. Cumulative effects of piscivorous colonial waterbirds on 

juvenile salmonids: A multi predator-prey species evaluation. PLoS One. 17(8), 

e0272875 (2022). 

 

56. Schreck, C. B., Olla, B. L. & M. W. Davis. Behavioral responses to stress. Fish 

Stress Health Aqua. 62, 145-170 (1997). 

 

57. Saloniemi, I., Jokikokko, E., Kallio-Nyberg, I., Jutila, E. & P. Pasanen. 

Survival of reared and wild Atlantic salmon smolts: size matters more in bad years. 

ICES J. Mar. Sci. 61, 782-787 (2004). 

 

58. Janhunen, M., Piironen, J., Vainikka, A. & P. Hyvärinen. The effects of 

environmental enrichment on hatchery-performance, smolt migration and capture rates 

in landlocked Atlantic salmon. PLoS One. 16(12), e0260944 (2021). 

 

59. Lorenzen K. Allometry of natural mortality as a basis for assessing optimal 

release size in fish-stocking programmes. Can. J. Fish. Aquat. Sci. 57: 2374–2381 

(2000). 

 

60. Fonken, D. et al. Benefits of stocking fewer, but larger individuals with 

implications for native fish recovery. Can. J. Fish. Aquat. Sci. 80, 439-450 (2022). 

 

61. Rillahan, C., Chambers, M. D., Howell, W. & W.H. Watson III. The behavior 

of cod (Gadus morhua) in an offshore aquaculture net pen. Aquaculture 310, 3-4 

(2011). 

 

62. Faust, M. D. et al. Acoustic telemetry as a potential tool for mixed-stock 

analysis of fishery harvest: a feasibility study using Lake Erie walleye. Can. J. Fish. 



 

112 

Aquat. Sci. 76, 1019-1030 (2017). 

 

63. Cudney, J. L., Bangley, C. W., Dell’Apa, A., Diaddorio, E., & Rulifson, R. A. 

Development of a continental shelf acoustic telemetry array to support behavioral 

research of fish in a high energy ocean environment. Fish. Research. 247, (2022). 

 

64. Krueger, C. C., et al. Acoustic telemetry observation systems: challenges 

encountered and overcome in the Laurentian Great Lakes. Can. J. Fish. Aquat. Sci. 75, 

1755-1763 (2018). 

 

65. Kerstetter, D. W., Luckhurst, B. E., Price, E. & J. E. Graves. Pop-up satellite 

archival tags to demonstrate survival of blue marlin (Makaira nigricans) released from 

pelagic longline gear. Fish. Bull. 101, 939–948 (2003). 

 

66. Nielsen, J. K. et al. Characterizing activity and assessing bycatch mortality of 

Pacific halibut with accelerometer pop-up satellite archival tags. Anim. Biotelemetry. 

6, 10 (2018). 

  



 

113 

Tables 

Table 1: Summary of hatchery-reared juvenile cisco (Coregonus artedi) stocked in 

Keuka Lake, New York, USA from October 2019 to October 2020. Age-0 fish, also 

known as Fall Fingerlings, are 10-months old from hatching whereas age-1 fish, 

known as Yearlings, are between 1 and 2-years old. Tagged cisco mass (g) and total 

length (mm) are shown for each cohort as mean ± SD(range). 

 

Table 2: Summary of the fate of tagged juvenile cisco released into Keuka Lake, New 

York, USA from October 2019 to October 2020.  

 

 

  



 

114 

Table 3: Candidate model set for Cox proportional hazards regressions fit to time-to-

event data from acoustic-tagged juvenile cisco in Keuka Lake. Multimodel inference 

is based on Akaike’s Information Criterion adjusted for small sample sizes (AICc). 

 

 

  



 

115 

Table 4: Relative variable importance values from multi-model inference for Cox 

proportional hazards models fit to tagged juvenile cisco based on Akaike’s 

Information Criterion adjusted for small sample sizes (AICc). 

 

 

  



 

116 

Figures 

 

 

Figure 1. Study location of a whole-lake acoustic telemetry study to track survival 

rates of stocked juvenile fish in Keuka Lake, New York, USA. The solid black circles 

show acoustic receiver placement for whole-lake level array coverage and the 

transparent gray circles show the total number of ‘detection events’ (discrete events of 

individual tag detections) from October 2019 to August 2021. The inset map (bottom 

right) indicates the Finger Lakes region of New York State, USA.   

  



 

117 

 

 

Figure 2. Juvenile cisco (Coregonus artedi) re-introduced from extirpation in Keuka 

Lake, New York, USA. a) Juvenile cisco with a surgically implanted acoustic 

transmitter. b) Small (0.6g) acoustic telemetry transmitters used for age-0 fish in this 

study. Photographs by M. Chalupnicki, USGS. 

  



 

118 

 

 

Figure 3. Kaplan Meier curves for tagged cisco (Coregonus artedi) with estimated 

survival probability (solid lines) and 95% confidence interval (dashed lines) for each 

cohort of stocked fish: a) Fall 2019, b) Fall 2020, and c) Summer 2020. Blue lines 

indicate the age-0 release cohort (Fall 2019, Fall 2020) and orange lines indicate the 

age-1 release cohort (Fall 2019, Summer 2020). The first dates on the x-axes are the 

stocking date for each cohort of released fish.  

 

 

 

 

  



 

119 

 
Figure 4. Estimated Cox proportional hazard model survivorship curves for acoustic-

tagged juvenile cisco stocked into Keuka Lake, New York, USA. (a) Predicted 

survival and 95% confidence limits (shaded areas) by age-at-release from the top 

ranked model (Survival ~ Age; 42% model weight) based on Akaike’s Information 

Criterion as adjusted for small sample size (AICc). (b) Predicted survival from the 

second AICc-ranked Cox model (Survival ~ Year + Age; 17% model weight). 

  



 

120 

CHAPTER 3 

HOW ACCURATELY DOES eDNA REFLECT THE SPATIAL DISTRIBUTION 

OF COLD-WATER FISH? FIELD VALIUDATION FROM A TEMPERATE LAKE 

 

Abstract: 

 

1. Applications of environmental DNA (eDNA) based detection technology to 

evaluate the distribution of aquatic organisms are increasing, yet field validations of 

eDNA are important to measure accuracy in study systems. To successfully apply this 

technology to species conservation, it is critical to understand how both species 

biology and environmental conditions affect the accuracy of inference from eDNA 

detection data. 

2. We implemented a field assessment of the accuracy and spatial resolution of eDNA-

based species distributions for a native cold-water, schooling fish, cisco Coregonus 

artedi, that has been re-introduced to a deep temperate lake. We leveraged a 

combination of acoustic telemetry, providing known spatial locations of tagged fish, 

and lake-wide eDNA sampling to infer their distribution in Keuka Lake, New York, 

USA. Sub-surface (12m and 18m depths) eDNA samples were collected to 

accommodate the diel vertical migration behavior of this fish species. 

3. The results of this study validated the accuracy of positive eDNA detections with 

the distribution of tagged fish to coarse spatial scales. Yet, several fine-scale locations 

revealed a mismatch between eDNA and acoustic telemetry detections, consistent with 

rapid transport of genetic material via lake currents.  



 

121 

4. Empirical measurements of lake currents using drifters found cisco eDNA 

detections could deviate from specimens’ source locations by as much as 3.3 km at 12 

m depth or 1.5 km at 18 m depth over a 24 h transport period. 

5. Our study indicates that accurate species distributions estimated from eDNA 

sampling in lakes may require further understanding of transport mechanisms and 

persistence of environmental genetic material to relate point detections to source 

animal locations. Integrating eDNA sampling with additional data collection of 

species biology and environmental conditions will increase the spatial resolution of 

fish distribution assessments. 

 

1. Introduction 

Environmental DNA (eDNA) based monitoring is a powerful, emerging technology 

for detecting aquatic species and inferring their distributions to support fisheries 

conservation efforts (Deiner et al., 2017; Allendorf et al., 2022). This technique has 

proved valuable because it circumvents the need for the capture and handling of 

organisms, which can be logistically challenging and physiologically stressful for 

study subjects. By contrast, collecting water samples for eDNA analysis is relatively 

easy and noninvasive (Deiner et al., 2017; Sepulveda et al., 2019). Given higher 

detection sensitivity, eDNA-based detection methods show promise for management 

applications of low-density species that are challenging to detect with traditional 

capture methods, including endangered species, early-stage species introductions, and 

migratory species that seasonally occupy managed areas (Furlan et al., 2019; Piggott 



 

122 

et al., 2021; Rojahn et al., 2021; Allendorf et al., 2022). Despite the advantages of 

using eDNA-based detection technology to infer species distributions, how differences 

in habitat use by the species and environmental conditions impact the accuracy and 

spatial resolution of this technique is poorly understood (Barnes et al., 2014). Efforts 

to compare eDNA-based distribution assessments against true physical distributions of 

species in situ are rare but important to evaluate the accuracy of eDNA based species 

distributions (Takahara et al., 2013; Eichmiller et al., 2014; Yamamoto et al., 2016; 

Kessler et al. 2020; Nordstrom et al. 2024). Here, we leverage a species reintroduction 

effort, in which the locations of tracked organisms are known, to evaluate eDNA 

performance in situ within a lake ecosystem. 

Species distributions inferred from eDNA techniques may differ from the physical 

distribution of sampled organisms, because of both 1) the degree to which eDNA 

collection sites differ from the locations host animals shed sampled genetic material, 

and 2) the probability of correctly detecting eDNA in water samples given that eDNA 

is subject to dilution and degradation (Barnes et al., 2014; Deiner and Altermatt, 2014; 

Wilcox et al., 2016; Itakura et al., 2019). To maximize the spatial resolution of eDNA, 

collected field samples must be close in space and time to the target organisms 

shedding DNA, before it has been transported, diluted, or degraded by the 

environment (Wilcox et al., 2016; Guillera-Arroita et al., 2017).   

The location where DNA is shed from an organism and becomes eDNA is affected by 

many aspects of a target species’ biology. Fish have a wide range of depth and 

temperature preferences, movement patterns, and life history strategies that complicate 



 

123 

their eDNA detection probability both spatially and temporally throughout the year 

(Yamanaka and Minamoto, 2016; Sard et al., 2019; Takeuchi et al., 2019). For 

example, pelagic, cold-water fish species occupy habitats across a range of 

metalimnetic depths; thus, the accessibility of shed genetic material may require 

eDNA sampling across multiple depths (Littlefair et al., 2021). In contrast, single 

depth samples may be effective for benthic or shallow-depth dwelling taxa (Allendorf 

et al., 2022). Finally, highly migratory species may transition rapidly between 

locations and habitats, meaning any spatial-temporal mismatch between sample 

collection and true occupancy is likely to confound eDNA-based inferences (Buxton et 

al., 2017; Guillera-Arroita et al., 2017; Levi et al., 2019). The amount of shed genetic 

material also influences eDNA detectability; for example, mesocosm and cage 

enclosure experiments demonstrate that eDNA detection rates increase with fish 

biomass (Barnes et al., 2014). Conversely, eDNA detection rates are expected to 

decrease for rare species that yield low concentrations of eDNA (Sepulveda et al., 

2019).  

The persistence of eDNA after it is shed, but before it is sampled, is affected by 

conditions of the environment. For aquatic organisms, physical mechanisms that 

influence eDNA dynamics include the potential for water currents to transport eDNA 

particles (Deiner and Altermatt, 2014; Barnes et al. 2021), eDNA dilution over time 

(Sansom and Sassoubre, 2017), and eDNA deposition on sediments (Barnes et al., 

2014; Turner et al., 2015). Water temperature, microbial activity, and UV-B also 

affect degradation rates which partially determine eDNA persistence over time 

(Strickler et al., 2015; Tsuji et al. 2017; Kessler et al. 2020). eDNA transport is a 



 

124 

function of the morphology of the sampled waterbody including size, depth, and flow 

regimes, and thus can vary widely in aquatic settings and present challenges in linking 

eDNA detections to underlying species occupancy (Harper et al., 2019; Stewart, 2019; 

Littlefair et al., 2021). For example, studies in lotic systems have reported eDNA was 

detectable as far away as 9km for freshwater mussels (Sansom and Sassoubre, 2017), 

10km for zooplankton (Deiner and Altermatt 2014), and 54km (Nevers et al., 2020) to 

over 100km (Pont et al., 2018) for fish from source DNA locations. Although eDNA 

detections in lentic systems such as lakes are assumed to be less complex than rivers 

(Rees et al., 2014; Jo and Yamanaka, 2022), few studies quantify eDNA transport 

distances in lakes (see Dunker et al., 2016; Ghosal et al., 2018; Bedwell and Goldberg, 

2020).  

Reliable eDNA sampling would consequently account for environmental conditions 

and the detectability of targeted species in survey design. Combined, the degree to 

which these processes affect the accuracy of eDNA based species distribution 

estimates relative to the true underlying physical occupancy of target taxa is not well 

characterized. The goal of this study was to assess the efficacy of eDNA for 

representing the distribution of a previously extirpated fish species, cisco Coregonus 

artedi, reintroduced to Keuka Lake, New York, USA. By leveraging a combination of 

acoustic telemetry, fish abundance, and lake current data, our objectives were to: 1) 

Implement a lake-wide eDNA-based species distribution assessment, 2) characterize 

the physical distribution of the target species from fish equipped with acoustic 

transmitters (hereafter, ‘tagged fish’), 3) characterize water movements as a potential 

physical process affecting eDNA detection accuracy, and 4) quantify the spatial 



 

125 

resolution of eDNA-based species distributions against validated physical species 

distributions. We expected that lake regions with high densities of tagged fish 

detections would correlate with positive eDNA detections; however, we hypothesized 

that the spatial resolution of eDNA-based fish distributions may be impacted by water 

movement, DNA degradation, and dilution when compared against tagged fish 

locations.  

 

2. Methods  

We applied a suite of field efforts to characterize the accuracy with which eDNA 

samples reflect the spatial distribution of fish in Keuka Lake. We used an acoustic 

telemetry array to validate the physical distribution of cisco and compared this against 

eDNA-based survey outcomes. Further, all cisco were recently stocked; thus, we 

combined survival estimates with stocking release numbers to characterize fish density 

in the study system. Finally, we deployed drifters to characterize lake currents as a 

potential transport mechanism of shed genetic material. 

2.1. Study system 

Keuka Lake is in the Lake Ontario basin of the North American Laurentian Great 

Lakes (Figure 1). A narrow but deep temperate lake, Keuka Lake undergoes summer 

thermal stratification and is distinctively Y-shaped with three arms meeting at the 

confluence region (Figure 1) with 3 km maximum width, 57 m maximum depth, and 

4,688 ha total surface area (Bloomfield, 2015).  



 

126 

Cisco, a representative pelagic, cold-water fish, is an opportune species for validating 

their distribution across heterogenous lake habitats. Cisco are native to Nearctic 

regions including the Finger Lakes region of New York, USA where Keuka Lake is 

located (Lindsey and Woods, 1970; Page and Burr, 2011; Figure 1). This species 

formed the historical prey fish base for Keuka Lake; however, managers considered 

cisco extirpated from this system by the mid-1990s due to a combination of decreased 

water quality and competition with introduced alewife Alosa pseudoharengus and 

rainbow smelt Osmerus mordax. To restore the native fish assemblage and improve 

food web resilience underpinning popular sport fisheries, managers from New York 

State Department of Environmental Conservation (NYSDEC) initiated a native species 

reintroduction effort in 2018. All cisco reintroduced into Keuka Lake were stocked as 

juvenile age- 0 (10-months post hatching) or age-1 (19- or 22-months post hatching 

following procedures described by McKenna et al. (2021) and Koeberle et al. (2023). 

Cisco stocking numbers are provided in Table 1 and Table S1. All research protocols 

for sampling fish received ethical approval by U.S. Geological Survey and the New 

York State fish collections review office. Fish collections occurred under Scientific 

License to Collect or Possess Permit #2977 and were in accordance with all fish 

handling, animal care, and ethical use requirements governing fish collection permits 

as granted by NYSDEC. Finally, all methods reported here were in accordance with 

Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. 

 

2.2 eDNA survey design and water sampling 



 

127 

eDNA water samples were collected across Keuka Lake on two occasions that 

bracketed a cisco stocking event. Water samples were first collected during a period of 

low expected fish abundance (21-July-2020; n = 260 stocked fish), prior to the autumn 

stocking event, to quantify baseline eDNA detections from previously surviving 

cohorts of stocked fish (see supplemental text S1.1; Table S1). A second eDNA survey 

was conducted shortly after an autumn stocking event during a period of higher 

expected fish abundance (27-October-2020; n = 204,466 stocked fish Table 1). Keuka 

Lake was thermally stratified during both eDNA surveys. In temperate lakes, cisco 

exhibit diel vertical migration and have temperature preference for 12˚C - 16˚C 

(Rudstam, 1993). Cisco are usually found in the cooler waters below the summer 

thermocline. As a planktivorous fish, however, they may disperse above the 

thermocline to feed on plankton present in cooler surface waters during late autumn 

(George, 2019). To maximize eDNA detection probability, we collected water samples 

at 12 m and 18 m depths with Van Dorn samplers based on temperature and dissolved 

oxygen profiles (see Figure S1 for profiles). In July, the thermocline was at 12 m, 

while in October it had deepened with a weaker temperature gradient. Despite this 

shift, the mixed layer in October was within the cisco’s preferred temperature range < 

16˚C, increasing the likelihood of fish presence at both depths. We expected cisco 

would migrate vertically through the sampled water depths in both sampling periods. 

eDNA sampling locations were spatially allocated in proportion to the surface area of 

the regions of Keuka Lake (three arms and confluence region; Figure 1).  

At each sampling location, a 2 L grab sample was collected at each 12m and 18m 

depth. Each sample was stored on ice in coolers and filtered at onshore processing 



 

128 

stations within four hours of collection. To prevent field contamination, Van Dorn 

samplers were disinfected in a 10% bleach solution for at least 10 minutes and rinsed 

between each deployment. Separate samplers were used for each depth. Field 

personnel also sprayed working surfaces with bleach, changed gloves for each sample, 

and designated a single crew member to focus on water sample collections. Samples 

were filtered through one filter until saturation, or the entire 2 L sample was filtered. If 

filter saturation was reached, the filtration volume was recorded, and the rest of the 

grab sample was discarded. All environmental samples were filtered using a 47 mm 

glass fiber filter (1.5 μm pore size) at shoreline stations. Filters were placed on ice 

immediately after processing, temporarily stored at -20˚C during transport, and then 

held at -80˚C until DNA extraction. In total, we collected 111 environmental water 

samples in the July pre-stocking sampling occasion and 110 environmental water 

samples in the October post-stocking occasion. In addition, 15 field negative controls, 

three negative water sampling controls, and two bottle negative control samples were 

collected during each sampling occasion to test for eDNA contamination (see 

supplemental text S1.2 Field collections).   

 

2.3 eDNA analysis 

All DNA extractions were conducted with the Qiagen DNeasy Blood and Tissue Kit 

(Qiagen Corporation, Valencia, CA) following a slightly modified version of the tissue 

protocol. Modifications included a 55% increase in digestion buffer/proteinase K 

volume and overnight incubation in the Qiagen Lyse and Spin Basket. Extractions 



 

129 

were conducted in seven batches with two negative controls and one positive control 

(DNA sample from American eel Anguilla rostrata) within each batch. A positive 

control was included to verify each extraction batch was carried out with a similar 

extraction efficiency (see supplemental text S1.2 DNA extractions for further details). 

Two eDNA markers were developed to detect cisco mitochondrial DNA based on 

sequence data obtained from Lake Ontario broodstock used for hatchery-raised cisco 

stocked in Keuka Lake: COSP_ND401 and COSP_cytB01 (Table 2). Based on 

sequence similarity, markers are not specific to cisco and will also amplify bloater 

Coregonus hoyi, a deepwater Coregonine species reintroduced to Lake Ontario. 

Bloater are not distributed in the Finger Lakes region which eliminates the risk of false 

positive detections in Keuka Lake (see supplemental text S1.2 Marker development 

and validation).  

Quantitative PCR (qPCR) analysis was performed to detect cisco DNA in 

environmental samples using eight qPCR technical replicates for each marker. Each 

20 µl qPCR reaction for marker COSP_cytB01 contained 1X TaqMan® 

Environmental Master Mix 2.0, 500 nM forward primer, 500 nM reverse primer, 250 

nM probe, and 3 µL of DNA template. Each reaction for marker COSP_ND401 

contained 1X TaqMan® Environmental Master Mix 2.0, 500 nM forward primer, 900 

nM reverse primer, 250 nM probe, and 3 µL of DNA template. Cycling conditions 

were denaturing for 10 minutes at 95 ℃ followed by 45 cycles of 15 second 

denaturing (95 ℃) and 60 second annealing (65 ℃) with a QuantStudio 7 Pro 

Thermalcycler. Samples that amplified with a cycle threshold (Ct) < 45 were specified 

as positive for containing cisco DNA. All samples were assessed for PCR inhibition 



 

130 

prior to analysis with species-specific markers (See supplemental text S1.2 qPCR 

analysis and inhibition tests for further details) 

 

2.4 Acoustic telemetry analysis  

We validated the spatial distribution of the stocked cisco population by monitoring 

tagged fish across a lake-wide telemetry receiver array from 2018 to 2021 (Figure 2a). 

Our eDNA study used a subset of tagged fish (July n = 56 tagged cisco see Table S3; 

October n = 66 tagged cisco see Table 1) detections from this telemetry dataset over 

an approximate two-week window immediately preceding and inclusive of eDNA 

sampling occasions (e.g., 7-July-2020 to 21-July-2020 and 15-October-2020 to 27-

October-2020, respectively) to represent the true underlying physical distribution of 

stocked fish in this closed population. Juvenile cisco were tagged with Juvenile 

Salmon Acoustic Telemetry System Acoustic Micro Transmitters (0.6 g tags for age-0 

fish; 3.5 g tags for age-1 fish). We assumed that tagged cisco movement and behavior 

are representative of the population at large, as validated by an acoustic tagging study 

for this species by McKenna et al. (2021) and extensive, lake-wide acoustic receiver 

coverage described by Koeberle et al. (2023). Additionally, cisco are a schooling fish 

species (Milne et al., 2005), thus tagged fish are assumed to mix and disperse with 

untagged fish upon stocking. Given their schooling behavior, we assumed that the 

spatial occupancy of tagged cisco reflected the spatial occupancy of the population at 

large.   



 

131 

The lake-wide acoustic telemetry array consisted of 24 Lotek-brand WHS 4250 

acoustic receivers deployed throughout Keuka Lake to maximize spatial coverage for 

detecting tagged cisco. Receivers were bottom-mounted and faced upward to detect 

tagged cisco in the pelagic zone. Range testing indicated that acoustic receivers had a 

linear detection radius of approximately 200–300 m. Three additional receivers were 

positioned midway in each lake branch to enhance coverage of offshore and nearshore 

movements. Finally, a closed population assumption was verified with a lack of tag 

detections at receivers placed in the lake outlet to ensure no tagged fish emigration 

(Koeberle et al., 2023). 

2.5 Data analysis 

The whole-lake acoustic telemetry array enabled tracking of tagged cisco as time-to-

event data. As such, both the outcome (e.g., mortality or censoring) and timing 

(number of days after stocking) were known for tagged fish. For survival analyses, we 

assumed that all tagged fish have an equal detection probability across the receiver 

array with no systematic detection heterogeneity. Using the acoustic telemetry dataset, 

we first inferred mortalities of tagged fish to assemble time-to-event data for survival 

modeling. Subsequently, we modeled tagged fish time-to-event data using Kaplan-

Meier curves to characterize survival of stocked fish following their release into 

Keuka Lake. Next, to describe spatial occupancy dynamics of the stocked population 

as inferred from tagged fish, we generated a ‘residence index’, or the proportion of 

time spent by tagged fish at a given receiver location, to estimate the relative 

distribution of fish across the lake. Finally, we inferred the spatial density of cisco in 



 

132 

Keuka Lake over the telemetry monitoring period (preceding and including eDNA 

sampling) by coupling abundance estimates from survival modeling with residency 

index information. All analyses were conducted in program R version 4.3.2 (R Core 

Team 2023). Additional details on the calculations used to estimate fish abundance 

and density from acoustic telemetry are provided in supplemental text S1.3.  

 

2.6 Lake currents survey 

Strong water currents can move shed genetic material away from its source location 

quickly, biasing the spatial resolution of eDNA-based species distributions (Sansom 

and Sassoubre, 2017). To characterize potential eDNA transport from water 

movements, we measured deep water currents in Keuka Lake using Lagrangian 

drifters (hereafter ‘drifters’; see Figure S3) following designs of McCaffrey and 

Koeberle (2024). Each drifter was composed of a low-drag surface float equipped with 

a GPS tracker tethered to a high-drag drogue below the surface set at targeted depths 

(Manley, 2010; Fossette et al., 2012). We deployed six drifters configured to track 

currents at 12 m (n = 3) and 18m (n = 3) sampling depths from 30-September-2022 to 

19-October-2022. Subsequently, drifter locations were processed into current 

velocities, current directionality, and current movement distances across moving 

windows of time (see supplemental text S1.4). Previous studies have shown that 

eDNA can degrade rapidly in aquatic environments, with exponential decreases in 

eDNA concentrations occurring in as few as 24 h to multiple days in cooler waters 

(Sansom and Sassoubre, 2017; Mauvisseau et al., 2022). Further, due to the large 



 

133 

volume of Keuka Lake and expected low densities of cisco present during the study 

from survival estimates, we assumed eDNA would diffuse rapidly. Therefore, we 

selected a conservative window of 0-48 h to quantify lake current movements as 

reflective of potential transport distances of eDNA from source locations.  

 

3. Results  

 

3.1 eDNA surveys 

The novel qPCR markers developed for cisco eDNA, COSP_ND401 and 

COSP_cytB01, successfully detected cisco DNA in Keuka Lake water samples. 

Standard curve analysis demonstrated highly efficient (> 95%) markers with y-

intercept values less than 40 cycles (Figure S2; Table 3). Markers were also highly 

sensitive to target DNA reliably detecting < 1 copy per reaction in the laboratory 

(Table 3). Nine positive environmental samples contained cisco DNA above the limit 

of detection; however, all positive samples were below the limit of quantification 

values for which reliable concentrations could be estimated (Table 3). Thus, eDNA 

assays provided useful presence-absence detection information but eDNA 

concentrations were too low to reliably estimate the quantity of cisco eDNA at 

positive sampling locations. Negative controls demonstrate that our field and lab 

procedures successfully limited sample contamination (see text S2.1 Positive and 

negative control tests). Only one field control (jar blank sample) detected cisco DNA; 

however, this did not affect our conclusions from the eDNA survey as potential 



 

134 

contamination was an isolated event (see supplemental text S2.1 Environmental 

samples and qPCR inhibition tests). None of the 111 eDNA filter samples collected in 

the July pre-stocking, low fish abundance survey contained positive detections for 

cisco. In contrast, nine out of 110 water samples (8.18%) amplified for positive cisco 

eDNA detections in the October post-stocking, higher fish abundance survey (Table 

4). During the October survey, eight positive eDNA detections occurred in the west 

arm and the confluence region of the lake, with one positive eDNA sample in the 

south arm (Figure 2b), indicating preferential distribution for cisco in one arm of the 

lake. Sampling at multiple depths was necessary to detect this pelagic-ranging fish 

species, as six positive eDNA detections occurred at 12m depths, while three positive 

eDNA detections occurred at 18 m depths (Figure 2b). Only one site contained 

positive eDNA detections at both 12 m and 18 m sampling depths (Figure 2b; Table 

4).  

eDNA results suggest that cisco were present at low densities at detection sites. Five 

out of nine positive eDNA samples were detected with both qPCR markers, while four 

out of nine samples were only detected with marker COSP_cytB01. In general, 

samples that amplified for cisco DNA had a low number of positive qPCR lab 

replicates, with 80% of positive samples (marker COSP_ND401) amplifying for n  2 

of the eight lab replicates, and 77.7% of positive samples (marker COSP_cytB01) 

amplifying with one replicate (Table 4). Only one sample amplified for  75% of 

replicates with both markers (Table 4).  

 



 

135 

3.2 Fish abundance and distribution  

The lake-wide distribution of cisco inferred from eDNA detections and from tagged 

fish detections were similar, however we found discrepancies between methods at fine 

spatial scales (Figure 2a, b). eDNA results from the October post-stocking survey 

indicated cisco were predominately distributed in both the west arm and the 

confluence region (Figure 2b), whereas telemetry detections during the same period 

indicated that tagged cisco were restricted only to the west arm (Figure 2a). 

Combining information from the telemetry and eDNA distribution data sets, we 

hypothesize that the Keuka Lake cisco population was primarily restricted to the west 

arm, whereas positive eDNA detections in the confluence resulted from the presence 

of low abundances of cisco or transport of genetic material out of the west arm to the 

adjacent confluence region via lake currents (see section 3.3 Lake currents below).   

Applying survival estimates to stocking numbers, we estimated that during the post-

stocking October survey lake-wide fish abundance 12 days after stocking was 18,588 

(95% CI 8,668, 39,862; Table 1). In contrast, we estimated only 195 fish (95% CI 165, 

229; Table S4) were present lake-wide during the July survey given that a small cohort 

of cisco (n = 260 total fish) were released in early July prior to the eDNA survey (see 

supplemental text S1.3 for further information).  Assuming cisco predominately used 

the west arm, as indicated by both the acoustic telemetry and eDNA surveys, this 

would equate to 0.19 fish/ha (95% CI 0.16, 0.23) during the July pre-stocking survey 

period, when no cisco eDNA detections occurred, and 18.4 fish / ha (95% CI 8.6, 



 

136 

39.5) in the west arm during the October post-stocking survey when nine water 

samples amplified for cisco eDNA (Table S4).  

 

3.3 Lake currents 

Drifter deployments indicated that lake currents are a potential mechanism for eDNA 

transport, and that these movement trajectories can be rapid and complex (Figure 3a-

c). Currents varied consistently between the two sampling depths, whereby the 12 m 

drifters generally moved more than 18 m drifters. Current velocities were high in 

Keuka Lake. The 12 m drifters moved with 0.13 km/h average velocity (0.46 km/h 

average maximum velocity) and 18 m drifters moved with 0.036 km/h average 

velocity (0.37 km/h average maximum velocity). Movements were substantial even at 

short time windows whereby 12m drifters (Figure 3b) moved on average 1.23 km (sd 

± 0.89km) while 18m drifters (Figure 2c) moved on average 0.52 km (sd ± 0.37km) 

over a 12h time interval. Over a maximum 48h time interval we calculated that 12 m 

drifters (Figure 3b) moved on average 2.12 km (sd ± 1.52km) and 18 m drifters 

(Figure 3c) moved on average 1.59 km (sd ± 1.08km) in Keuka Lake. Within lake 

arms, currents generally moved in a north-south orientation, while some current 

movement patterns were more complex, including in the confluence region (Figure 

S4). For example, one drifter deployed at the confluence region moved into adjacent 

lake arms but subsequently returned to the confluence (see Figure 3a).  

 

4. Discussion 



 

137 

eDNA-based survey methods are useful for detecting aquatic species and estimating 

their distributions (Burian et al., 2021). Keuka Lake provided the opportunity to 

compare the use of acoustic telemetry and eDNA to infer the lake-wide, spatially 

heterogenous distribution of a reintroduced pelagic fish. In the present study site, we 

observed a species occupancy gradient from the highest fish densities inferred in the 

west arm, validated by acoustic telemetry and eDNA detections, to rare or no 

abundance at the level of lake regions, with mismatches observed between acoustic 

telemetry and eDNA. This suggests that eDNA-based fish distributions can be 

accurate at coarse spatial scales in lakes. Yet, at higher spatial resolutions, we 

hypothesize that lake currents have the potential to rapidly transport eDNA material 

away from source locations. Lake current data from drifters suggests that the spatial 

mismatch between the true cisco distribution and positive eDNA detections could be 

negligible if eDNA is collected shortly after being shed, or as far as 1.5-2km from the 

source location within 48hrs. Consequently, this field validation demonstrates that 

species distribution assessments may require additional information on eDNA 

transport and persistence to accurately interpret species occupancy at finer spatial 

scales (Wilcox et al., 2016; Sansom and Sassoubre, 2017).  

We expect our results on water movements and the potential for eDNA transport in 

Keuka Lake to apply to other large, deep lakes. Calculated drifter velocities in Keuka 

Lake are within the range of lake current velocities reported in other lentic systems, 

with examples that range from 0.11 ± 0.02 km/hr (Lake Geneva, France and 

Switzerland; Lemmin, 2020), 0.07 – 0.54 km/hr (Markermeer, Netherlands; 

Vijverberg et al., 2011), and 0.13 – 0.54 km/hr (Upper Klamath Lake, Oregon, USA; 



 

138 

Gartner et al., 2007). Similarly, lake current velocities measured in nearby, deeper 

(maximum depth = 188 m) Seneca Lake in the Finger Lakes region, New York, USA, 

are on average 0.55 km/h (2m sampling depth), 0.36 km/hr (10 m depth), and 0.09 

km/hr (30 m depth) (McCaffrey and Koeberle, 2024). We hypothesize that the 

currents detected in Keuka Lake using drifters were likely generated by seiches. 

Similar internal wave-driven currents were documented in Kempenfelt Bay, Lake 

Simcoe, Canada, an embayment of comparable size (3 km width, 15 km length, 42 m 

depth) to Keuka Lake (Flood et al., 2020). While seiches in the essentially linear-

shaped Finger Lakes (Figure 1) may limit dispersal, Keuka Lake is distinctively y-

shaped. This morphology could similarly lead to a mixing and dilution effect 

described by Flood et al. (2020), particularly when water is close to the confluence 

region. Similar lake current velocities and complex advection demonstrate that 

temperate lakes can have significant water movement, and thus managers may need to 

consider currents when assessing the spatial resolution of eDNA detections in lakes.  

Field measurements also demonstrated that lake currents can be complex and 

multidirectional. We found that drifter locations generally increased over time from 

their geographic deployment origin, however, average distances for multiple drifters 

appeared to reach an asymptote near 36 h (Figure 3). Thus, if shed DNA material is 

assumed to remain within a packet of water at depth, then the spatial extent eDNA can 

be transported away from its source location by currents may depend both on 

directional transport over shorter time frames and periodic lake processes like an 

internal seiche over longer time frames (Ahrnsbrak, 1974, Fricker and Nepf, 2000). 

Furthermore, lake currents can vary across depths (McCaffrey and Koeberle, 2024), so 



 

139 

a target species’ life history may require understanding eDNA transport at multiple 

water layers. For example, cisco exhibit diel movements throughout the mid-water 

column, hence the differences in current directionality and velocity we found at 12m 

and 18m water layers (Figure 3, Figure S4) could further diffuse an eDNA signal 

across lake depths (Littlefair et al. 2021). Thermocline depth, which varies throughout 

the year, can also influence the detection efficiency of acoustic-tagged fish in 

thermally stratified lakes (Wells et al. 2021; Kuai et al. 2021). Incorporating a species’ 

ecology in eDNA survey design is important to ensure accurate presence-absence data 

for monitoring the status of a fish population.      

Comparing the abundance and physical distribution of tagged fish to eDNA-inferred 

occupancy provides insight into the efficacy of eDNA as a fishery monitoring tool in 

lakes. First, our results quantify practical lower limits to eDNA detection techniques 

when species exist at low density. In the October survey, estimated post-stocking 

densities in the west arm were comparable to wild Coregonine densities reported in 

similar temperate lakes, including cisco in North America (Rudstam et al., 1987) and 

vendace Coregonus albula in Europe (Jurvelius et al., 1984; Axenrot and Degerman, 

2016). No cisco eDNA was detected using 2 L water samples in Keuka Lake at the 

July pre-stocking survey when estimated cisco abundance was < 200 fish (≈ 0.19 fish 

/ha in the west arm; Figure S7), so sampling for rare fishes with eDNA may require 

larger water collections to account for low detection probability of genetic material 

(Thompson, 2004; MacKenzie et al., 2005; Ficetola et al., 2015; Furlan et al., 2019; 

Sepulveda et al., 2019). Second, the ability to characterize the spatial distribution of a 

pelagic fish species required sampling at multiple depths. While we detected cisco 



 

140 

eDNA at 12 m or 18 m sampling depths, positive detections at both depths only 

occurred at one location. These results corroborate previous eDNA studies that found 

sampling at multiple depths was necessary for detecting fish in thermally stratified 

lakes (Littlefair et al., 2021). Third, our results emphasize that data on water currents 

are important for correct interpretation of eDNA-inferred species distributions in 

lakes. This point is well recognized in lotic environments (see Rees et al., 2014; 

Shogren et al., 2017; Nevers et al., 2020; Pont 2024), but we found substantial water 

movements in a lentic system that were consistent with discrepancies between 

telemetry- and eDNA-based cisco distributions. The drifters used in this study are 

relatively inexpensive (< £236.09 GBP or $300 USD) and easy to deploy (McCaffrey 

and Koeberle, 2024), and empirical current data can be used to characterize the spatial 

resolution and potential directional bias of eDNA source locations. Our approach 

requires information or assumptions about the expected time window over which 

DNA persists (e.g. 24-48 h) and thus improved in situ dilution and degradation models 

would further inform water current eDNA transport dynamics.   

In the future, we anticipate a growing body of research that leverages emerging 

technology to quantify biotic and abiotic relationships between eDNA detections and 

their proximity to a known source location (Wilcox et al., 2016; Littlefair et al., 2021; 

Kessler et al. 2020). For example, in situ studies testing eDNA persistence and 

dispersal from point source locations in lentic systems have shown promise for 

informing detection dynamics of shed genetic material and can be extended to lotic 

waterbodies like lakes with currents (Pilliod et al., 2014). This includes in situ 

experiments to measure eDNA dispersal rates and distances from live animal 



 

141 

enclosures (Brys et al. 2020), wildlife tracking to mechanistically control and monitor 

eDNA production and release (Kessler et al. 2020), and simulated DNA releases to 

elucidate eDNA transport dynamics (Shogren, et al. 2017). Future experiments could 

also integrate 3-dimensional acoustic receiver systems or depth-sensing tags to 

provide further insight into vertical habitat use of tracked fish (Littlefair et al. 2021). 

In the present study, we surveyed the distribution of mitochondrial DNA molecules 

and made inferences about the distribution of cisco from which those molecules 

originated, however future studies could explore diffusion of different eDNA states 

such as intracellular, extracellular DNA, or nuclear DNA (Jo and Minamoto, 2021). 

Further, accurate assessments of spatial distributions from eDNA may require 

integration of data on fish behavior, movement ecology, and the temporal trajectory of 

a DNA signal over time in survey design (Bista et al., 2016; Spear et al., 2021). Lastly, 

we anticipate opportunities to couple field-based measurements of lake currents and 

point source eDNA deployment experiments with hydrodynamic models that account 

for dilution, advection, settling, and reaction rate processes (Wain and Rehmann 

2010). Such models could improve predictions of eDNA transport and detection 

probability, enhancing species habitat assessments for fishery management 

applications (Harrison et al., 2019). 

Given additional biological and environmental insights, eDNA survey designs for lake 

ecosystems may need to evaluate tradeoffs between spatial coverage and detection 

resolution, depending on the spatial scale of management objectives (Burian et al., 

2021; Allendorf et al., 2022). eDNA based detection technology for understanding 

species distributions can be useful but complex, and empirical field-based validations 



 

142 

such as this study are critical to the successful application of this technology to aquatic 

conservation moving forward. 

 

Acknowledgements 

The authors thank Steve Robb, Bree Minges, Ariel Thoms, Pete Austerman, and Ben 

Carson (NYSDEC); Sarah Rubenstein, Jeremy Kraus, Gregg Mackey (USGS-

Tunison); Kimberly Fitzpatrick, Liam Zarri, Angela Fuller, Melanie Moss, Mandy 

Economos, Kelly Perkins, and Evan Cooch (Cornell University). We also thank Carl 

St. John, Rick Koeberle, Mike Ashdown, and members of the Shack-Sethi and 

Limnology lab groups. Le Moyne College provided funding for three drifters used in 

this study. Lastly, we thank Janelle Morano, Liam Zarri, Zena Casteel, and Lars 

Rudstam (Cornell University) for their helpful feedback which greatly improved the 

quality of this manuscript.  

 

Data Availability Statement 

All data sets used for this research are publicly available. eDNA and acoustic 

telemetry data are available from Zenodo: https://doi.org/10.5281/zenodo.16794444. 

Lake current data are available from Zenodo: 

https://doi.org/10.5281/zenodo.12775323. 

 

Research Funding 



 

143 

New York State Department of Environmental Conservation, Grant Number: F64-R.   



 

144 

References  

Ahrnsbrak, W. F. 1974. “Some additional light shed on surges.” Journal of 

Geophysical Research 79, no. 24: 3482-3483. 

https://doi.org/10.1029/JC079i024p03482  

 

Allendorf, F.W., Funk, W.C., Aitken, S.N., Byrne, M. and G. Luikart. 2022. 

Conservation and the Genomics of Populations. Oxford University Press. 3rd  

Edition.  

 

Axenrot, T., and E. Degerman. 2016. “Year-class strength, physical fitness and 

recruitment cycles in vendace (Coregonus albula).” Fisheries Research 173: 61 

69. https://doi.org/10.1016/j.fishres.2015.03.017 

 

Barnes, M. A., Turner, C. R., Jerde, C. L., Renshaw, M. A., Chadderton, W. L., and 

D.M. Lodge. 2014. “Environmental conditions influence eDNA persistence in 

aquatic systems.” Environmental Science & Technology 48, no. 3: 1819-1827. 

https://doi.org/10.1021/es404734p 

 

Barnes, M. A., Chadderton, W. L., Jerde, C. L., Mahon, A. R., Turner, C. R., and 

D.M. Lodge. 2021. “Environmental conditions influence eDNA particle size 

distribution in aquatic systems.” Environmental DNA 3, no. 3: 643-653. 

https://doi.org/10.1002/edn3.160 

 

Bedwell, M. E. and C.S. Goldberg. 2020. “Spatial and temporal patterns of 

environmental DNA detection to inform sampling protocols in lentic and lotic 

systems.” Ecology and Evolution 10, no. 3: 1602-1612. 

https://doi.org/10.1002/ece3.6014 

 

Bista, I., Carvalho, G. R., K. Walsh, et al. 2016. “Annual time-series analysis of 

aqueous eDNA reveals ecologically relevant dynamics of lake ecosystem 

biodiversity.” Nature Communications 8: 14087. 

https://doi.org/10.1038/ncomms14087 

 

Bloomfield, J. A. (Ed.). 2015. Lakes of New York State: Ecology of the Finger Lakes 

(Vol. 1). Elsevier. 

 

Brys, R., Haegeman, A., D. Halfmaerten, et al. 2021. “Monitoring of spatiotemporal 

occupancy patterns of fish and amphibian species in a lentic aquatic system 

using environmental DNA.” Molecular Ecology 30, no. 13: 3097-3110. 

https://doi.org/10.1111/mec.15742 

 

Burian, A., Mauvisseau, Q., Bulling, M., Domisch, S., Qian, S., and M. Sweet. 2021. 

“Improving the reliability of eDNA data interpretation.” Molecular Ecology 

Resources 21, no. 5: 1422-1433. https://doi.org/10.1111/1755-0998.13367 



 

145 

 

Buxton, A. S., Groombridge, J. J., Zakaria, N. B., and R.A. Griffiths. 2017. “Seasonal 

variation in environmental DNA in relation to population size and 

environmental factors.” Scientific Reports 7, no. 1: 46294. 

https://doi.org/10.1038/srep46294 

 

Deiner, K. and F. Altermatt. 2014. “Transport distance of invertebrate environmental 

DNA in a natural river.” PloS one 9, no. 2: e88786. 

https://doi.org/10.1371/journal.pone.0088786 

 

Deiner, K., Bik, H. M., E. Mächler, et al. 2017. “Environmental DNA metabarcoding: 

Transforming how we survey animal and plant communities.” Molecular 

Ecology 26, no. 21: 5872-5895. https://doi.org/10.1111/mec.14350 

 

Dunker, K. J., Sepulveda, A. J., R. L. Massengill, et al. 2016. “Potential of 

environmental DNA to evaluate northern pike (Esox lucius) eradication efforts: 

an experimental test and case study.” PloS one 11, no. 9: e0162277. 

https://doi.org/10.1371/journal.pone.0162277 

 

Eichmiller, J. J., Bajer, P. G., and P.W. Sorensen. 2014. “The relationship between the 

distribution of common carp and their environmental DNA in a small lake.” 

PloS one 9, no. 11: e112611. https://doi.org/10.1371/journal.pone.0112611 

 

Ficetola, G. F., Pansu, J., A. Bonin, et al. 2015. “Replication levels, false presences 

and the estimation of the presence/absence from eDNA metabarcoding data.” 

Molecular Ecology Resources 15, no. 3: 543-556.  

https://doi.org/10.1111/1755-0998.1233 

 

Flood, B., Wells, M., Dunlop, E. and J. Young. 2020. “Internal waves pump waters in 

and out of a deep coastal embayment of a large lake.” Limnology and 

Oceanography 65, no. 2: 205-223. https://doi.org/10.1002/lno.11292 

 

Fossette, S., Putman, N. F., Lohmann, K. J., Marsh, R., and G. C. Hays. 2012. “A 

biologist’s guide to assessing ocean currents: a review.” Marine Ecology 

Progress Series 457: 285-301. https://doi.org/10.3354/meps09581 

 

Fricker, P. D., and H. M. Nepf. 2000. “Bathymetry, stratification, and internal seiche 

structure.” Journal of Geophysical Research: Oceans 105, no. C6: 14237 

14251. https://doi.org/10.1029/2000JC900060 

 

Furlan, E. M., Gleeson, D., Wisniewski, C., Yick, J., and R. P. Duncan. 2019. “eDNA 

surveys to detect species at very low densities: A case study of European carp 

eradication in Tasmania, Australia.” Journal of Applied Ecology 56, no. 11: 

2505-2517. https://doi.org/10.1111/1365-2664.13485 

 



 

146 

Gartner, J. W., Wellman, R. E., Wood, T. M., and R. T. Cheng. 2007. “Water velocity 

and suspended solids measurements by in-situ instruments in Upper Klamath 

Lake, Oregon.” U.S. Geological Survey Open-File Report 2007-1279: viii + 

137 pp. 

 

George, E. 2019. “The history and ecology of Cisco Coregonus artedi in the 

Laurentian Great Lakes.” Aquatic Ecosystem Health & Management 22, no. 3:  

280-293. https://doi.org/10.1080/14634988.2019.1670461 

 

Ghosal, R., Eichmiller, J. J., Witthuhn, B. A., and P. W. Sorensen. 2018. “Attracting 

Common Carp to a bait site with food reveals strong positive relationships 

between fish density, feeding activity, environmental DNA, and sex 

pheromone release that could be used in invasive fish management.” Ecology 

and Evolution 8, no. 13: 6714-6727. https://doi.org/10.1002/ece3.4169 

 

Guillera‐Arroita, G., Lahoz‐Monfort, J. J., van Rooyen, A. R., Weeks, A. R., and R. 

Tingley. 2017. “Dealing with false‐positive and false‐negative errors about 

species occurrence at multiple levels.” Methods in Ecology and Evolution 8, 

no. 9: 1081-1091. https://doi.org/10.1111/2041-210X.12743 

 

Harper, L. R., A. S. Buxton, H. C. Rees, et al. 2019. “Prospects and challenges of 

environmental DNA (eDNA) monitoring in freshwater ponds.” Hydrobiologia 

826: 25-41. https://doi.org/10.1007/s10750-018-3750-5 

 

Harrison, J. B., Sunday, J. M., and S. M. Rogers. 2019. “Predicting the fate of eDNA 

in the environment and implications for studying biodiversity.” Proceedings of 

the Royal Society B 286, no. 1915: 20191409. 

https://doi.org/10.1098/rspb.2019.1409 

 

Itakura, H., Wakiya, R., Yamamoto, S., Kaifu, K., Sato, T., and T. Minamoto. 2019. 

“Environmental DNA analysis reveals the spatial distribution, abundance, and 

biomass of Japanese eels at the river‐basin scale.” Aquatic Conservation: 

Marine and Freshwater Ecosystems 29, no. 3: 361-373. 

https://doi.org/10.1002/aqc.3058 

 

Jo, T., and T. Minamoto. 2021. “Complex interactions between environmental DNA 

(eDNA) state and water chemistries on eDNA persistence suggested by meta 

analyses.” Molecular Ecology Resources 21, no. 5: 1490-1503. 

https://doi.org/10.1111/1755-0998.13354 

 

Jo, T., and H. Yamanaka. 2022. “Meta‐analyses of environmental DNA downstream 

transport and deposition in relation to hydrogeography in riverine 

environments.” Freshwater Biology 67, no. 8: 1333–1343. 

https://doi.org/10.1111/fwb.13920 

 



 

147 

Jurvelius, J., Lindem, T., and J. Louhimo. 1984. “The number of pelagic fish in Lake 

Paasivesi, Finland, monitored by hydroacoustic methods.” Fisheries Research 

2, no. 4: 273-283. https://doi.org/10.1016/0165-7836(84)90030-4 

 

Kessler, E. J., Ash, K. T., Barratt, S. N., Larson, E. R., and M. A. Davis. 2020. 

“Radiotelemetry reveals effects of upstream biomass and UV exposure on 

environmental DNA occupancy and detection for a large freshwater turtle.” 

Environmental DNA 2, no. 1: 13-23. https://doi.org/10.1002/edn3.42 

 

Klymus, K. E., Merkes, C. M., M. J. Allison, et al. 2020. “Reporting limits of 

detection and quantification for environmental DNA assays.” Environmental 

DNA 2, no. 3: 271-282. https://doi.org/10.1002/edn3.29 

 

Koeberle, A. L., Pearsall, W., B. E. Hammers, et al. 2023. “Whole-lake acoustic 

telemetry to evaluate survival of stocked juvenile fish.” Scientific Reports 13, 

no. 1: 18956. https://doi.org/10.1038/s41598-023-46330-6 

 

Kuai, Y., Klinard, N.V., A.T. Fisk, et al. 2021. “Strong thermal stratification reduces 

detection efficiency and range of acoustic telemetry in a large freshwater lake.” 

Animal Biotelemetry, 9, pp.1-13. https://doi.org/10.1186/s40317-021-00270-y 

 

Lemmin, U. 2020. “Insights into the dynamics of the deep hypolimnion of Lake 

Geneva as revealed by long-term temperature, oxygen, and current 

measurements.” Limnology and Oceanography 65, no. 9: 2092-2107. 

 

Lesperance, M.L., Allison, M.J., Bergman, L.C., Hocking, M.D. and C.C. Helbing. 

2021. “A statistical model for calibration and computation of detection and 

quantification limits for low copy number environmental DNA samples.” 

Environmental DNA 3, no. 5: 970-981. https://doi.org/10.1002/edn3.220 

 

Levi, T., Allen, J. M., D. Bell, et al. 2019. “Environmental DNA for the enumeration 

and management of Pacific salmon.” Molecular Ecology Resources 19, no. 3: 

597-608. https://doi.org/10.1111/1755-0998.12987  

 

Lindsey, C. C. and C. S. Woods. 1970. Biology of coregonid fishes. In International 

Symposium on Biology of Coregonid Fishes (1969: University of Manitoba). 

University of Manitoba Press. 

 

Littlefair, J. E., Hrenchuk, L. E., Blanchfield, P. J., Rennie, M. D., and M. E. Cristescu 

2021. “Thermal stratification and fish thermal preference explain vertical 

eDNA distributions in lakes.” Molecular Ecology 30, no. 13: 3083-3096. 

https://doi.org/10.1111/mec.15623 

 

MacKenzie, D. I., Nichols, J. D., Sutton, N., Kawanishi, K., and L. L. Bailey. 2005. 

“Improving inferences in population studies of rare species that are detected 



 

148 

imperfectly.” Ecology 86, no. 5: 1101-1113. https://doi.org/10.1890/04-1060 

 

Manley, T. O. 2010. “Hands-on oceanography: Drifters, drogues, and circulation.” 

Oceanography 23:165–171. doi:10.5670/oceanog.2010.17 

 

Mauvisseau, Q., Harper, L.R., Sander, M. Hanner, R.H., Kleyer, H. and K. Deiner. 

2022. “The multiple states of environmental DNA and what is known about 

their persistence in aquatic environments.” Environmental Science & 

Technology 56, no. 9: 5322-5333. https://doi.org/10.1021/acs.est.1c07638 

 

McCaffrey, L. and A. Koeberle. 2024. “An economical open-source Lagrangian drifter 

design to measure deep currents in lakes.” Water Resources Research 60, no. 

10: 2024WR037429. https://doi.org/10.1029/2024WR037429 

 

McKenna Jr, J. E., Sethi, S. A., Scholten, G. M., Kraus, J., and M. Chalupnicki. 2021.  

“Acoustic tag retention and tagging mortality of juvenile cisco Coregonus 

artedi.” Journal of Great Lakes Research 47, no. 3: 937-942. 

https://doi.org/10.1016/j.jglr.2021.03.020 

 

Milne, S. W., Shuter, B. J. and W.G. Sprules. 2005. “The schooling and foraging 

ecology of lake herring (Coregonus artedi) in Lake Opeongo, Ontario, 

Canada.” Canadian Journal of Fisheries and Aquatic Sciences 62, no. 6:  

1210-1218. https://doi.org/10.1139/f05-030 

 

Nevers, M. B., Przybyla-Kelly, K., Shively, D., Morris, C. C., Dickey, J., and M. N. 

Byappanahalli. 2020. “Influence of sediment and stream transport on detecting 

a source of environmental DNA.” PLoS One 15, no. 12: e0244086. 

https://doi.org/10.1371/journal.pone.0244086 

 

Nordstrom, B., Budd, A., Mitchell, N., Cornish, C., Byrne, M., Kuchling, G. & 

Jarman, S. (2024). Environmental DNA reflects spatial distribution of a rare turtle in a 

lentic wetland assisted colonization site. Environmental DNA, 6(1), p.e507. 

https://doi.org/10.1002/edn3.507 

 

Page, L. M., and B. M. Burr. 2011. Peterson field guide to freshwater fishes of North 

America north of Mexico. Houghton Mifflin Harcourt. 

 

Piggott, M. P., Banks, S. C., Broadhurst, B. T., Fulton, C. J., and M. Lintermans.  

2021. “Comparison of traditional and environmental DNA survey methods for 

detecting rare and abundant freshwater fish.” Aquatic Conservation: Marine 

and Freshwater Ecosystems 31, no. 1: 173-184. https://doi.org/10.1002/aqc.3474 

 

Pilliod, D. S., Goldberg, C. S., Arkle, R. S., and L. P. Waits. 2014. “Factors 

influencing detection of eDNA from a stream‐dwelling amphibian.” Molecular 

Ecology Resources 14, no. 1: 109-116.  



 

149 

https://doi.org/10.1111/1755-0998.12159 

 

Pont, D. 2024. “Predicting downstream transport distance of fish eDNA in lotic 

environments.” Molecular Ecology Resources 24: e13934. 

https://doi.org/10.1111/1755-0998.13934  

 

Pont, D., Rocle, M., A. Valentini, et al. 2018. “Environmental DNA reveals 

quantitative patterns of fish biodiversity in large rivers despite its downstream 

transportation.” Scientific reports 8, no. 1: p.10361. 

https://doi.org/10.1038/s41598-018-28424-8  

 

R Core Team. 2023. R: a language and environment for statistical computing. R 

foundation for statistical computing, Vienna. Available at 

https://www.rproject.org. 

 

Rees, H. C., Maddison, B. C., Middleditch, D. J., Patmore, J. R., and K. C. Gough. 

2014. “The detection of aquatic animal species using environmental DNA–a 

review of eDNA as a survey tool in ecology.” Journal of Applied Ecology 

51, no. 5: 1450-1459. https://doi.org/10.1111/1365-2664.12306 

 

Rojahn, J., Pearce, L., Gleeson, D. M., Duncan, R. P., Gilligan, D. M., and J. 

Bylemans. 2021. “The value of quantitative environmental DNA analyses for 

the management of invasive and endangered native fish.” Freshwater Biology 

66, no. 8: 1619-1629.https://doi.org/10.1111/fwb.13779 

 

Rudstam, L. G., Lathrop R.C., and S.R. Carpenter. 1993. “The rise and fall of a 

dominant planktivore: direct and indirect effects on zooplankton.” Ecology 74, 

no. 2: 303–319. https://doi.org/10.2307/1939294 

 

Rudstam, L. G., Clay, C. S., and J. J. Magnuson, J. J. 1987. “Density and size 

estimates of cisco (Coregonus artedii) using analysis of echo peak PDF from a 

single-transducer sonar.” Canadian Journal of Fisheries and Aquatic 

Sciences 44, no. 4: 811-821. https://doi.org/10.1139/f87-099 

 

Sansom, B. J., and L. M. Sassoubre. 2017. “Environmental DNA (eDNA) shedding 

and decay rates to model freshwater mussel eDNA transport in a river.” 

Environmental Science & Technology 51, no. 24: 14244-14253. 

https://doi.org/10.1021/acs.est.7b05199 

 

Sard, N. M., Herbst, S. J., Nathan, L., Uhrig, G., Kanefsky, J., Robinson, J. D., and K. 

T. Scribner. 2019. “Comparison of fish detections, community diversity, and 

relative abundance using environmental DNA metabarcoding and traditional 

gears.” Environmental DNA 1, no. 4: 368-384. https://doi.org/10.1002/edn3.38 

 

Sepulveda, AJ, Schabacker, J, Smith, S, Al-Chokhachy, R, Luikart, G, and S. J. 

https://doi.org/10.1111/1755-0998.12159
https://doi.org/10.1111/1755-0998.13934
https://doi.org/10.2307/1939294


 

150 

Amish. 2019. “Improved detection of rare, endangered and invasive trout in 

using a new large-volume sampling method for eDNA capture.” 

Environmental DNA 1: 227-237. https://doi.org/10.1002/edn3.23 

 

Shogren, A.J., Tank, J.L., Andruszkiewicz, E., Olds, B., Mahon, A., Jerde, L., and D. 

Bolster. 2017. “Controls on eDNA movement in streams: Transport, Retention, 

and Resuspension.” Scientific Reports 7: 5065.  

https://doi.org/10.1038/s41598-017-05223-1 

 

Spear, M. J., Embke, H. S., Krysan, P. J., and M. J. Vander Zanden. 2021.  

“Application of eDNA as a tool for assessing fish population abundance.” 

Environmental DNA 3, no. 1: 83-91. https://doi.org/10.1002/edn3.94 

 

Stewart, K. A. 2019. “Understanding the effects of biotic and abiotic factors on 

sources of aquatic environmental DNA.” Biodiversity and Conservation 28, 

no. 5: 983-1001. https://doi.org/10.1007/s10531-019-01709-8 

 

Strickler, K. M., Fremier, A. K., and C. S. Goldberg. 2015. “Quantifying effects of 

UV-B, temperature, and pH on eDNA degradation in aquatic microcosms.” 

Biological Conservation 183: 85-92. 

https://doi.org/10.1016/j.biocon.2014.11.038  

 

Takahara, T., Minamoto, T. And H. Doi. 2013. “Using environmental DNA to 

estimate the distribution of an invasive fish species in ponds.” PloS one 8, no. 

2: p.e56584. ttps://doi.org/10.1371/journal.pone.0056584. 

 

Takeuchi, A., Iijima, T., W. Kakuzen, et al. 2019. “Release of eDNA by different life 

history stages and during spawning activities of laboratory-reared Japanese 

eels for interpretation of oceanic survey data. Scientific Reports 9, no. 1: 6074. 

https://doi.org/10.1038/s41598-019-42641-9 

 

Thompson, W. (Ed.). 2004. Sampling rare or elusive species: concepts, designs, and 

techniques for estimating population parameters. Island Press, Washington, 

D.C.  

 

Turner, C.R., Uy, K.L. and R.C. Everhart. 2015. “Fish environmental DNA is more 

concentrated in aquatic sediments than surface water.” Biological 

Conservation 183: 93-102. https://doi.org/10.1016/j.biocon.2014.11.017  

 

Tsuji, S., Ushio, M., Sakurai, S., Minamoto, T., and H. Yamanaka. 2017. “Water 

temperature-dependent degradation of environmental DNA and its relation to 

bacterial abundance.” PLoS One 12, no. 4: e0176608. 

https://doi.org/10.1371/journal.pone.0176608 

 

Vijverberg, T., Winterwerp, J.C., Aarninkhof, S.G.J., and H. Drost. 2011. “Fine 

https://doi.org/10.1002/edn3.23
https://doi.org/10.1007/s10531-019-01709-8
https://doi.org/10.1016/j.biocon.2014.11.017
https://doi.org/10.1371/journal.pone.0176608


 

151 

sediment dynamics in a shallow lake and implication for design of hydraulic 

works.” Ocean Dynamics 61: 187–202.  

https://doi.org/10.1007/s10236-010-0322-2 

 

Wain, D.J. and C.R. Rehmann. 2010. “Transport by an intrusion generated by 

boundary mixing in a lake.” Water Resources Research 46, no. 8: W08517. 

https://doi.org/10.1029/2009WR008391 

 

Wells, M.G., Li, J., B. Flood, et al. 2021. “Speed of sound gradients due to summer 

thermal stratification can reduce the detection range of acoustic fish tags: 

results from a field study in Hamilton Harbour, Ontario.” Canadian Journal of 

Fisheries and Aquatic Sciences 78, no. 3, 269-285. 

https://doi.org/10.1139/cjfas-2020-0078 

 

Wilcox, T. M., McKelvey, K. S., Young, M. K., et al. 2016. “Understanding 

environmental DNA detection probabilities: A case study using a stream 

dwelling char Salvelinus fontinalis.” Biological Conservation 194: 209-216. 

https://doi.org/10.1016/j.biocon.2015.12.023 

 

Yamamoto, S., Minami, K., K. Fukaya, et al. 2016. “Environmental DNA as a 

‘snapshot’ of fish distribution: A case study of Japanese jack mackerel in 

Maizuru Bay, Sea of Japan.” PloS one 11, no. 3: e0149786. 

https://doi.org/10.1371/journal.pone.0149786 

 

Yamanaka, H., and T. Minamoto. 2016. “The use of environmental DNA of fishes as 

an efficient method of determining habitat connectivity.” Ecological Indicators 

62: 147-153. https://doi.org/10.1016/j.ecolind.2015.11.022 

 

 

  



 

152 

Tables 

 

Table 1: Summary of stocked juvenile cisco Coregonus artedi in Keuka Lake, New 

York, USA for the autumn stocking event (15-October-2020). Survival rate (𝑆̂t) and 

abundance (𝑁̂t) are estimated from acoustic telemetry up to the time (t) post-stocking 

eDNA survey (27-October-2020).  

aAcoustic-tagged cisco mass (g) and total length (mm) are shown as mean ± SD(range). 

 

Table 2: Cisco Coregonus artedi qPCR markers. Markers are TaqMan® MGB (minor 

groove binder) with a non-fluorescent quencher. 

 

 

  



 

153 

Table 3: Relative performance metrics for the two cisco Coregonus artedi qPCR 

markers developed for the present study. Standard curves were obtained by serially 

diluting gBlock standard in a 5x dilution series over eight orders of magnitude 

between 31,250 copies and 0.4 copies (Figure S2). Limit of detection (LOD) and limit 

of quantification (LOQ) values (DNA copies) are provided using methods by Klymus 

et al. (2020) and Lesperance et al. (2021). The limit of detection value is the fewest 

number of copies that can be detected in 95% of the sample replicates. The limit of 

quantification value is the fewest number of copies in 95% of replicates to calculate 

the concentration of target DNA in a sample. 

 

 

  



 

154 

Table 4: Positive DNA detections from the post-stocking eDNA survey (27-October-

2020) in Keuka Lake, New York, USA, along with their associated cycle threshold 

(Ct) values and number of replicates that amplified for cisco Coregonus artedi DNA. 

aOut of eight total PCR replicates.  
bSame sampling location. 

  



 

155 

Figures 

 

 

Figure 1. Location of the Keuka Lake cisco (Coregonus artedi, inset photo) tracking 

study system in the Finger Lakes region of New York, USA. The lake has three 

distinct branches (west, east, south) and a confluence region. Cisco inset photo credit: 

image is by E. Edmonson and is freely available for use in the public domain at: 

https://commons.wikimedia.org/wiki/File:Cisco.jpg.   



 

156 

 

Figure 2. Field validation of eDNA-based species distributions in Keuka Lake, New 

York, USA. a) Inferred distribution of stocked juvenile cisco Coregonus artedi from 

fish equipped with acoustic transmitters. Cumulative tagged cisco detections are 

shown from 15-October-2020 (fish stocking event) to 27-October-2020 (eDNA 

survey) on acoustic telemetry receivers as present (shaded triangles) or absent (open 

triangles). b) eDNA survey results on the 27-October-2020 post-stocking sampling 

effort. Positive samples that amplified for cisco DNA are indicated as shaded circles 

(red: 12m sampling depth, blue: 18m sampling depth, yellow: both 12m and 18m 

sampling depth), whereas samples that failed to amplify for DNA are indicated as 

open circles.  Inset photos credit: both the acoustic-tagged juvenile cisco photo (left) 

and Van Dorn water sampler photo (right) are by M. Chalupnicki, United States 

Geological Survey.   



 

157 

 

Figure 3. Lagrangian drifters were deployed in Keuka Lake, New York, USA to 

characterize lake currents from 30-September-2022 to 19-October-2022. a) Example 

of a drifter trajectory (30-September-2022 to 7-October-2022) deployed at 12m depths 

with Euclidean travel distances shown at 24hrs and 48hrs. b-c) Plots summarize mean 

(lines) and standard deviation (shaded areas) Euclidean distances away from start 

points for 12m drifters (red) and 18m drifters (blue) over 12, 24, 36, and 48hr time 

intervals.  



 

 

158 

 

CHAPTER 4 

INTEGRATING ACOUSTIC TELEMETRY AND DEMOGRAPHIC 

MODELING TO INFORM CISCO (COREGONUS ARTEDI) 

RESTORATION IN KEUKA LAKE, NEW YORK 

 

Abstract: 

Stocking is an important conservation tool to restore fish populations. Yet, assessing 

restoration success is often limited by a lack of field-based demographic information 

at low abundance, particularly for juvenile fish. Using outcomes from Cisco 

(Coregonus artedi) reintroductions to Keuka Lake, New York, USA, we demonstrate 

a data-driven approach to assess fish stocking performance and evaluate the likelihood 

of achieving conservation goals. Multistage juvenile survival estimates from acoustic 

telemetry quantified high post-stocking mortality rates across three distinct stages. 

Modeled post-stocking stages include immediate release, acclimation, and long-term 

survival assumed to reflect natural mortality of Cisco in Keuka Lake. High juvenile 

mortality severely limited the probability that stocked fish will reach reproductive 

maturity, and population viability analysis with a Leslie matrix life-stage model 

indicated that re-establishing a Cisco population is unlikely with current stocking 

practices and lake conditions. By contrast, using Cisco life history parameters 

extrapolated from other systems would have resulted in false optimism for restoration 

success. Our results highlight the importance of utilizing in situ demographic 

estimates for designing and implementing conservation stocking efforts.  

 

 



 

 

159 

 

Introduction  

Conservation stocking to recover imperiled fishes is common and population 

assessments are needed to guide and evaluate restoration outcomes (Cochran-

Biederman et al. 2014; Jachowski et al. 2016). Post-stocking evaluation has long been 

recognized as a crucial component of fisheries management for identifying causes of 

success or failure of stocking programs (see Cowx 1994). Management actions 

including hatchery rearing and release practices, stocking abundance, and stocking 

duration are important predictors of reintroduction success and represent key controls 

for fishery managers to improve restoration outcomes (Cochran-Biederman et al. 

2014; Fonken et al. 2023). Despite recent integration of defined success criteria into 

management objectives, species reintroductions frequently have high failure rates 

(Seddon et al. 2007; Armstrong and Seddon 2008). Adaptive management can 

improve long-term restoration success rates, but system-specific information on 

reintroduced populations is rare and difficult to obtain (Bacon et al. 2015; Jachowski 

et al. 2016; Lennox et al. 2021).  

An important tool for designing fishery recovery programs is population 

viability analysis. This approach is well-suited for fish restoration efforts because it 

informs adaptive management by identifying risk factors and quantifying uncertainty 

of long-term persistence (Boyce 1992; Beissinger 2002; Ellner and Fieberg 2003). 

Population viability analysis features a population projection model parameterized 

with vital rates of the modeled species (e.g., Leslie matrix; Leslie 1948) whereby 

recovery or extinction probabilities are calculated for future population trajectories. 

Population growth is often sensitive to specific life history vital rates, therefore 



 

 

160 

 

parameterizing the population projection model with accurate demographic estimates 

is important for generating realistic outcomes (Caswell 2001; Beissinger 2002; White 

et al. 2002). This can be problematic when life history data are unavailable (Boyce 

1992). For example, in fisheries restoration applications sparse empirical data may 

limit the capacity of managers to specify appropriate target stock, hence impeding 

their ability to accurately quantify restoration success.  

Because in situ demographic information on reintroduced fish populations is 

costly and difficult to generate, population analyses often apply vital rates from 

comparable systems with extant populations of the species of interest or by borrowing 

rates from similar species (Jachowski et al. 2016). While this provides a baseline to 

design recovery efforts, this practice could also lead to biases if these borrowed vital 

rates are not reflective of conditions in the study system. In addition, when 

reintroduced, the species may recover slowly or persist at low densities. The 

population may therefore go undetected via traditional netting techniques, limiting 

inference to vital rates despite active monitoring efforts (Thompson 2013; Jachowski 

et al. 2016). Furthermore, relying on borrowed estimates from other systems to design 

stocking programs, without post-stocking monitoring, risks failure to meet restoration 

objectives if conditions are unsuitable for population establishment. Ideally, 

population assessments for species restoration efforts would feature reliable 

demographic estimates across a species’ life history using system-specific data. For 

juvenile fishes, which are often underrepresented with traditional survey methods, this 

information is critical but sparse (Boyce 1992; Munzbergova and Ehrlen 2005).  

Stocked fish can provide valuable data for parameterization of population 



 

 

161 

 

models to inform the design and adaptive implementation of species reintroductions. 

Novel technologies like acoustic telemetry create additional opportunities to overcome 

data limitations in demographic estimates like juvenile survival. Hatchery-raised 

juvenile fish are commonly used to recover or reintroduce populations, yet their 

survival after stocking is often poorly understood (Brown and Day, 2002; Cochran-

Biederman et al., 2014). This is because sampling different life stages requires unique 

fishing gears and surveying distinct habitats, complicating comparisons across age 

groups (Rudstam et al. 1984; Murphy and Willis 1996; Kubečka et al. 2012). Further, 

it is difficult to tag and monitor the movement, growth, and survival of small fish with 

previously available tracking technologies that are focused on larger-bodied and adult-

stage fish (Mitchell et al., 2019; McKenna et al., 2021). To address this, recent 

advances in miniaturized transmitter technology coupled with spatially extensive 

telemetry arrays have provided greater understanding of survival processes of small 

fish (McMichael et al. 2010; Koeberle et al., 2023). In addition, spatially extensive 

arrays can generate time-to-event data, providing opportunities to apply multistage 

survival modeling to improve understanding of mortality rates associated with stocked 

fish (Sethi et al. 2024). This approach is useful for managers to evaluate stocking 

success by distinguishing post-stocking mortality of tagged fish from natural mortality 

of wild fish, providing in situ survival estimates missing from many population 

assessments.  

We demonstrate a quantitative assessment of fish restoration stocking through 

an ongoing native Cisco (Coregonus artedi) reintroduction to Keuka Lake, New York, 

USA conducted by New York State Department of Environmental Conservation 



 

 

162 

 

(NYSDEC). Cisco are a pertinent study species within the Coregonus (Salmonidae: 

Coregoninae) species complex. Anthropogenic-driven Cisco population declines have 

occurred from overfishing, water quality and habitat degradation, and non-native 

species introductions throughout its native range (Stockwell et al. 2009; Eshenroder et 

al. 2016; Bunnell et al. 2024). In North America, efforts to restore coregonines have 

intensified with hatchery augmentation throughout the Great Lakes region. These 

efforts have included basin-wide, international partnerships with State, Federal, and 

Tribal organizations to implement an adaptive management framework to increase 

lake ecosystem resilience (Zimmerman and Krueger 2009; Bunnell et al. 2023). 

Population assessments of Cisco restoration are available for the Great Lakes (Fisch et 

al. 2019; Rook et al. 2021, 2022; Fielder and McDonnell 2024). Nevertheless, 

managers lack population projections for Cisco reintroduced to inland lake settings, 

including Keuka Lake. 

Keuka Lake, an inland lake in the Lake Ontario basin, is a deep, meso-

oligotrophic lake with 4,688 ha total surface area, 3 km maximum width, 57 m 

maximum depth and three distinctive arms (West, South, and East) that meet at the 

Confluence region (Bloomfield 1978; Figure 1). Cisco formed the historical prey fish 

base in Keuka Lake, located close to the southern extent of Cisco distributions in 

North America (Page and Burr 2011). Their population in Keuka Lake declined from 

the 1970s to the early 1990s, likely from predation on larval stages by introduced 

forage fishes Alewife (Alosa pseudoharengus) and Rainbow Smelt (Osmerus mordax) 

(Hrabik et al. 1998; Mrnak et al. 2022). Lake managers considered Cisco extirpated by 

the mid-1990s. Lake Whitefish (Coregonus clupeaformis) were also present but were 



 

 

163 

 

no longer observed after 1988. Since then, Keuka Lake has recently introduced 

piscivorous Walleye (Sander vitreus) confirmed in 2016 and several established non-

native invertebrate species including both zebra and quagga mussels (Dreissena spp.; 

zebra mussels confirmed in 1994, quagga mussels in 2008). Keuka Lake also has an 

abundant native mysid (Mysis diluviana) population, an important prey source for 

Cisco, and wild-reproducing Lake Trout (Salvelinus namaycush) that is the base for a 

popular sport fishery. Improved water quality and a steep decline in Rainbow Smelt 

and Alewife populations occurred from the 2000s to mid-2010s. This observed prey 

fish crash led managers to implement Cisco reintroductions in 2018 to present and to 

cease non-native Brown Trout (Salmo trutta) and Atlantic Salmon (Salmo salar) 

hatchery stocking. Cisco restoration is hypothesized to stabilize Keuka Lake’s mid 

trophic fish assemblage and to improve food web resilience to support the recreational 

Lake Trout fishery. Managers are considering restoration outcomes related to 

achieving a standing stock target abundance of adult Cisco, and achieving a long-term, 

self-sustaining Cisco population. 

This study applies contemporary technology for observing released fish, in situ 

demographic modeling, and system-specific information to evaluate the probability of 

reestablishing a fish population. Specifically, we sought to answer the following 

questions: 1) Are current stocking efforts and practices sufficient to restore the Cisco 

population in Keuka Lake? 2) Will the reintroduced Cisco population be self-

sustaining over the long-term? 3) Is our approach useful for informing adaptive 

stocking practices to improve recovery?  

We obtained juvenile and adult Cisco mortality rates from Keuka Lake using 



 

 

164 

 

whole-lake acoustic telemetry and multistage survival modeling for juveniles and 

using catch curve analysis from historical (pre-extirpation) survey data for adults. We 

then conducted a population viability analysis with system-specific information. We 

assessed the likelihood of achieving the management goals of establishing a minimum 

adult standing stock and establishing a self-sustained population over 50 years. We 

anticipate that our data-driven approach will prove useful for fisheries conservation 

where augmentation efforts focus on stocking juvenile fish to restore populations.  

 

Methods  

Fish reintroductions 

Cisco eggs were collected from adult broodstock captured in the Great Lakes 

(see Table S1 for more information) annually in late November or early December 

(McKenna et al. 2021). Annual stocking occurs the following October at the fall 

fingerling juvenile stage (10 months old from hatching). From October 2018 through 

October 2024, over 450,000 total fall fingerlings have been stocked into Keuka Lake 

(see Table S1 for stocking numbers and fish sizes). Additionally, in 2019 and 2020 a 

smaller number (𝑛 < 2,000) of older juvenile Cisco were stocked as yearling fish 

cohorts (18, 19, or 22 months old from hatching). Hatchery stocking primarily releases 

fall fingerlings, except for the yearling cohorts noted above. Current hatchery 

production has capacity for annual releases of 100,000 fall fingerlings or 2,000 

yearlings. All Cisco cohorts were stocked offshore via boat in the northwestern arm of 

the lake at ~50m depths (see Figure 1).  

In situ demographic modeling 



 

 

165 

 

We leveraged a whole-lake acoustic telemetry dataset provided by NYSDEC 

and described in Koeberle et al. (2023) to estimate juvenile fish survival for 

population-level inferences of stocked fish. A subset of stocked Cisco were equipped 

with small acoustic transmitters (hereafter, tags) and tracked across a whole-lake 

acoustic receiver array from years 2019 to 2021 in Keuka Lake (see S1.1 for acoustic 

telemetry specifications). The system is assumed to be a closed population (i.e. no 

immigration and emigration), validated by movement results from the acoustic 

telemetry experiment (Koeberle et al. 2023). Previous laboratory experiments 

indicated that tag burden did not increase mortality (McKenna et al. 2021), and all 

tagged fish were held in the hatchery to monitor their survival and tag retention two 

weeks prior to their release into Keuka Lake (Koeberle et al. 2023). Thus, we assumed 

that demographic estimates of tagged fish represented the broader population of 

stocked fish. All subsequent analyses were conducted in program R version 4.4.1 (R 

statistical programming, 2024).  

To estimate juvenile Cisco survival, we applied a multistage modeling method 

using time-to-event data analysis procedures from Koeberle et al. (2023) and model 

procedures by Sethi et al. (2024). This technique provided an opportunity to estimate 

post-stocking mortality partitioned into three sequential stages including a ‘straight-to-

death’ mortality period immediately upon release (stage one), a period of elevated 

mortality during stocked fish acclimation (stage two), and finally a longer-term natural 

mortality regime (stage three). This is particularly useful for obtaining an estimate of 

wild juvenile fish survival rates which can be taken as the stage three ‘natural 

mortality’ rate. We conducted multi-model selection with Deviance Information 



 

 

166 

 

Criterion (DIC) by testing covariates on stages and transition times between stages 

including size, age-at-release, and condition. See text S1.2 for multistage modeling 

specifications. 

In situ estimates of adult (age-3+) mortality were calculated from a historical 

dataset of Cisco catches provided by NYSDEC. This dataset pre-dates the Cisco 

collapse with lake-wide gillnet surveys conducted from the 1970s to present (Table 

S2). We constructed netted Cisco age and length distributions from each survey from 

1979 to 1991 with age information estimated from scales (Figures S1, S2). We then 

applied regression-based catch curve analysis with the R package ‘FSA’ (fishR Core 

Team 2024) to estimate adult instantaneous total mortality rates and annual mortality 

rates. Historical angler records do not indicate a Cisco fishery in the lake. Therefore, 

we assume total mortality estimates from catch curve analysis is equivalent to natural 

mortality. Catch data suggests that the last year of Cisco reproduction in Keuka Lake 

occurred in 1983 with a maximum observed age of 10 years across survey years 

(Figure S1). Finally, we excluded gillnet surveys from 1971-1976 from catch curve 

analysis due to a lack of corresponding age data, however we observed higher 

numbers of Cisco catches during this period (Table S2). Because in situ adult 

mortality rates were derived from a period when Cisco may have been in decline, we 

also obtained literature values for adult mortality rates from Cisco populations in 

comparable lakes.  

Population modeling  

For this analysis, we specified a standing stock target of 𝑁 = 1,000 adult 

spawners (age-3+). We also identified a range of adult density targets based on other 



 

 

167 

 

lake systems with extant Cisco populations and considered long-term population 

establishment goals over a 50-year time horizon. To quantify standing stock targets 

and enable population viability analysis projections for the Cisco population in Keuka 

Lake, we developed a stage-based (Lefkovitch 1965) matrix population projection 

model based on the life-cycle graph presented in Figure 2. We assumed that adult 

Cisco spawning would occur in November or December at age-3 (Fisch et al. 2019; 

Gatch et al. 2023). The life cycle was therefore truncated at age-3 with constant annual 

mortality rates (e.g., age-3+ is a terminal adult stage). Annual hatchery and wild 

juvenile survival rates reported here were expressed as a pre-spawning census. For 

example, stocked fall fingerlings released in October were treated as juvenile 1-year-

old fish at release (hereafter, age-1 in our life-cycle graph) while yearlings were 

treated as juvenile (sub-adult) 2-year-old fish at release (hereafter, age-2) with annual 

survival estimates derived accordingly from the top-ranked multistage model. 

Discrete-time forward population projections are based on the product of the 

projection matrix: 

𝑛(𝑡 + 1)  = 𝐀 ∙ 𝑛(𝑡), 

where n(𝑡) is a vector of abundances of each stage at time 𝑡 and A is the female-based 

projection matrix.  

 To account for fixed annual stocking protocols, let 𝐑 =  [𝛿1 𝛿2 … 𝛿𝑘]T, 

where 𝛿𝑥 is the number of released fish within an age class. We then defined a stocked 

matrix model 𝐀′ and a vector of stocked fish n′(𝑡) such that: 

𝑛′(𝑡) =  [ 𝑛(𝑡)
1

] 



 

 

168 

 

The stocked system becomes 

𝑛′(𝑡 + 1)  = 𝐀′ ∙ 𝑛′(𝑡) 

where the stocked matrix is defined as 

𝐀′ =  [
𝐀 𝐑
0 1

] 

with a constant annual rate of juvenile fish released as a single age class, R: 

R = [
0
0
𝛿

], 

such that 

𝐀′  = [

0 0 𝑓3+𝑠0 0
𝑠1 0 0 0
0 𝑠2 𝑠3+ 𝛿
0 0 0 1

], 

where 𝑠1 is wild juvenile survival from age-1 to age-2, 𝑠2 is wild juvenile survival 

from age-2 to age-3, and 𝑠3+ is adult (age-3+) survival. Wild-reproduced age-0 

survival, 𝑠0, is the product of adult fertility rates 𝑓3+ (number of female eggs per fish) 

and 𝑠0, where 𝑠0 =  𝑠𝑒𝑔𝑔 ∙ 𝑠𝑓𝑟𝑦 ∙ 𝑠𝑠𝑓. We specified 𝑠𝑒𝑔𝑔 as the probability of egg hatch 

at 𝑡 = 0, 𝑠𝑓𝑟𝑦 as survival from 𝑡 = 0 to six months, and 𝑠𝑠𝑓 (summer fingerling) as 

survival from six months to one year (Fielder and McDonnell 2024). A cohort of 

stocked fish, 𝛿, enters the population as the product of 𝑁𝑟𝑒𝑙, number of hatchery fish 

released, and 𝑆′, their post-stocking annual survival rate. For stocked yearlings in 

Keuka Lake, the stocked vector (column 4 in matrix 𝐀′), 𝐑𝒀, is: 

𝐑𝒀 = [

0
0

𝑁𝑟𝑒𝑙𝑆′2

1

]. 

For stocked fall fingerlings, the stocked vector, 𝐑𝑭𝑭, can be parameterized as:  



 

 

169 

 

𝐑𝑭𝑭 = [

0
𝑁𝑟𝑒𝑙𝑆′1

0
1

], 

with hatchery fish that survive and enter the wild population as wild age-2 fish. With 

this formulation, fall fingerlings first enter the wild population at time 𝑡 = 1 year (as 

juvenile age-2 fish) when projecting from the initial population (see text S1.3 for more 

information).   

Population scenarios and sensitivity analyses 

  To evaluate the potential effectiveness of stocking on population viability, we 

used a combination of perturbation analysis and numerical simulation to evaluate 

efficacy of different stocking schemes in enhancing the viability of Cisco populations. 

We conducted prospective perturbation analysis of our deterministic baseline matrix to 

identify key life cycle events that determine population growth rate, λ (see text S1.3 

for details). Additionally, we derived the minimum vital rates necessary to maintain a 

minimum stable trajectory by solving for the characteristic polynomial for the matrix 

model where 𝜆 was constrained to 1.0 (representing the condition of zero population 

growth). We parameterized the baseline model for Keuka Lake using in situ juvenile 

survival estimated from the top-ranked multistage time-to-event model and adult 

mortality derived from the catch curve analysis. Remaining vital rates for fecundity 

and age-0 survival (egg hatch through year 1 survival) were derived from coregonine 

studies in North America and Europe (Table S3).  

We also considered stochastic versions of our baseline population models in 

two ways, both using numerical simulation experiments. First, we simulated vital rate 



 

 

170 

 

stochasticity by constructing sets of random matrices to reflect variation driven by 

environmental conditions, and projecting the populations based on a random selection 

of a matrix from this set at each time. Ideally, we would either (1) sample for a 

multivariate distribution where the covariances among parameters were specified, or 

(2) sample for a multivariate distribution where the parameter covariance structure is 

implicit in the annual matrix (Fieberg and Ellner 2001). In the absence of covariance 

estimates among parameters, we instead sampled parameter values from specified 

statistical distributions to generate a set of random matrices, over which individual 

matrices were sampled randomly at each time step. The statistical distributions used 

were specified for literature-derived fecundity and age-0 survival estimates to reflect 

environmentally driven variation of vital rates (Table S3). We then conducted 

bootstrapping from this set of matrices to evaluate the probability of obtaining a 

representative matrix with a positive population growth rate where 𝜆 > 1.0.  

The preceding approach generated a set of random matrices that implicitly 

varied around a multivariate mean population growth rate, which was assumed to be 

stationary over the projections. To account for periodic shifts from this mean growth 

rate in our random matrix approach, we also simulated scenarios with episodic high 

recruitment events typical of pelagic schooling fishes, including Cisco (Cury et al. 

2000; Yule et al. 2006). In coregonines, this ‘boom-and-bust’ recruitment dynamic 

(hereafter, boom recruitment) is hypothesized to be linked to cold winters with 

increased ice cover improving age-0 survival rates (Karjalainen et al. 2015; Myers et 

al. 2015; Stewart et al. 2021; Brown et al. 2022; Marjomӓki et al. 2024). Boom 

recruitment years are observed for extant Cisco populations in the Great Lakes and 



 

 

171 

 

adjacent lakes including Lake Simcoe (Brown et al. 2024), Long Term Ecological 

Network inland lakes in northern Wisconsin (J. Vander Zanden pers. comm. 8 

November 2024), and for Vendace (Coregonus albula) and Whitefish (Coregonus 

lavaretus) populations in northern European lakes (Marjomӓki 2005; Axenrot and 

Degerman 2016; Sarvala et al. 2024). High abundance but infrequent year classes may 

improve the probability of re-establishing a population and are therefore important to 

model (Fielder and McDonnell 2024).  

We incorporated boom recruitment scenarios by constructing a set of 

environmental state matrices with vital rates representative of boom (increased age-0 

survival) and average (current age-0 survival estimates) recruitment years. 

Subsequently, population simulations randomly draw from state matrices with 

probabilities that represent the likelihood of boom recruitment occurrence. The 

periodicity of Cisco recruitment cycles ranges from 4- to 7-years in the Great Lakes 

(Yule et al. 2006; Fisch et al. 2019; Rook et al., 2021). Lake managers specified that 

Keuka Lake historically had cold winters with increased ice extent approximately 

every three years, so we specify population scenarios for both 3-year (similar 

recruitment frequency observed in Lake Ontario and Lake Simcoe; see Brown et al. 

2024) and increased risk of longer 7-year boom cycles. Boom recruitment matrices 

were populated with a +600% magnitude increase in joint survival across age-0 life 

stages (Fielder and McDonnell 2024).  

Hatchery stocking analysis 

To evaluate the success of restoration stocking efforts in achieving standing 

stock density targets, we calculated the minimum stocking density and post-stocking 



 

 

172 

 

survival rates necessary to accrue an adult (age-3+) population abundance of 𝑁 =

1,000 spawners. A scenario-based approach to account for uncertainty in juvenile fish 

survival was implemented to contrast hatchery juvenile survival rates and stocking 

effort under current estimates (median value of posterior distributions from the top-

ranked multistage survival model) and optimistic estimates (upper 95% Credible 

Interval) for fall fingerling (𝑆1
′  in the life-cycle diagram) and yearling (𝑆2

′ ) stocked 

fish. We also explored standing stock target success with scenarios (Table S4) for 

wild juvenile survival (age-1, 𝑆1 and age-2, 𝑆2), specified as: (1) Low: current in situ 

estimates (median value), (2) Medium: upper 95% Credible Interval of in situ 

estimates, and (3) High: optimistic (upper 95% Credible Interval for 𝑆1, equivalent 

survival 𝑆2 and adult 𝑆3+). The High scenario assumes that wild age-2 juvenile fish 

have exceeded a size threshold to escape predation (see length-at-age data; Figure 

S2). Deterministic projections were implemented with 10 years of juvenile Cisco 

stocking with estimates provided for both stocking numbers and equivalent lake-wide 

densities. 

Population viability analysis 

Because of high uncertainty about the fate of stocked fish and Keuka Lake 

ecosystem conditions, managers sought to compare both pessimistic and optimistic 

scenarios to inform decision making. Thus, we explored our set of Low (current), 

Medium, and High juvenile survival scenarios to quantify the likelihood of Cisco re-

establishment in Keuka Lake. Success of this restoration effort was specified through 

management objectives as: (1) a self-sustaining population characterized by 𝜆 ≥ 1.0, 

(2) proportion ≥ 50% of population trajectories over a minimum standing stock size of 



 

 

173 

 

𝑁 = 1,000 adults, and (3) with long-term persistence over a 50-year time horizon with 

projected adult fish densities reflective of comparable lakes with wild coregonine 

populations. Across scenarios, we tested the management effects of stocking age (fall 

fingerling or yearling) and stocking rate (annual number stocked and duration 10 or 20 

years). We also explored the viability of the reintroduced Cisco population by 

considering boom recruitment frequency every 3- or 7-years.  

Population trajectories specific to the boom recruitment analysis were 

simulated with bootstrap sampling from our set of environmental state matrices and 

associated probabilities (e.g., 1/3 likelihood for a 3-year boom recruitment cycle or 1/7 

likelihood for a 7-year boom recruitment cycle) over 𝑛 = 10,000 iterations. We also 

calculated the stochastic growth rate for each scenario. While this approach did not 

explicitly simulate a periodic cycle within a single trajectory, we assumed that 

trajectories across the simulated set are, on average, representative of the effects of an 

environmentally driven boom recruitment cycle on the population growth rate. We 

also assumed that few stocked Cisco have survived to date in 2024, therefore all 

simulations were initialized with no Cisco present in Keuka Lake. Next, we quantified 

the probability of extinction for each scenario, defined as the proportion of simulations 

with a terminal abundance of fewer than 100 adults with replicates (n = 250) to 

average out the periodic cycle of boom recruitment years. A population trajectory was 

considered functionally extinct if fewer than 100 adults persisted. Lastly, we compared 

our environmental state matrices approach to the set of random matrices approach for 

stochasticity by conducting sensitivity analyses to evaluate how λ responded to 

variation of early life history survival (age-0), our wild juvenile survival scenarios, 



 

 

174 

 

and in situ versus literature-based adult mortality rates. Since the Cisco population has 

been reintroduced from extirpation, we assumed that modeled populations experience 

density-independent growth during their recovery and are not subject to fishing 

mortality.  

 

Results 

In situ demographic estimates  

Multistage survival models proved useful for estimating the handling and 

release survival of hatchery-stocked juvenile fish and distinguishing natural mortality 

rates representative of wild-equivalent juvenile fish (Table 1). Our top-ranked 

multistage model indicated that age-at-release was an important predictor of survival 

(Figure 3, see Table S5 for multi-model selection results). Mortality was particularly 

high upon release for fall fingerlings, with only 21% of tagged fish surviving past the 

‘straight-to-death’ period after release (stage one; < 1 day). Stocked yearlings had 

higher initial estimated survival of 78% (e.g., stage one) and annual cumulative 

survival of 2.4% (e.g., joint survival taken as stages one, two, and three). For fish that 

survived stage one, estimated stage two acclimation periods were ~25 days for fall 

fingerlings and ~61 days for yearlings. Finally, wild equivalent discrete annual 

survival rates were estimated from the multistage model as 𝑆1 = 0.004 (95% Credible 

Interval < 0.001, 0.078) for fall fingerlings and 𝑆2 = 0.053 (95% Credible Interval < 

0.001, 0.18) for yearlings. The second-ranked multistage model included the covariate 

length with moderate support. Nonetheless, we inferred that age may be associated 

with a size threshold to escape predation and selected the multistage model associated 



 

 

175 

 

with age over length.   

The historical survey dataset also provided key insight into adult mortality 

rates prior to the Cisco collapse. Our in situ catch curve analysis from historical data 

estimated adult (age-3+) mortality rates of 48.6% annually from years 1979-1991 

(Figure S3; see Table S6 for examples from extant Cisco populations). This was 

parameterized in our population model as annual survival, 𝑆3+ = 0.514. Survey years 

1971-1976, though excluded from catch curves as netted fish were not aged, indicated 

that Cisco were distributed throughout Keuka Lake pre-extirpation (Figure S4).  

Deterministic population model and vital rate sensitivities 

Our augmented matrix provided a convenient method for tracking hatchery 

fish contributions as they enter a wild population and for conducting sensitivity 

analysis. The population growth rate for the deterministic Low scenario (current 

juvenile survival) was 𝜆 = 0.52. The Cisco population also showed negative 

population growth for Medium and High scenarios, with 𝜆 = 0.61 and 𝜆 = 0.72, 

respectively. Using our baseline population projection model, we solved for the stage-

specific survival rates that led to a stable population (i.e., 𝜆 = 1.0) and results 

demonstrated that substantial reductions in juvenile mortality rates are necessary to 

achieve a self-sustained Cisco population in Keuka Lake. Under the High scenario, 

wild age-1 juvenile survival would need to increase from 𝑆1 = 0.004 (95% Credible 

Interval < 0.01, 0.078) to 𝑆1 =  0.44 in average years, or to 𝑆1 = 0.064 in boom 

recruitment years. Perturbation analysis revealed early life stage (age-0) survival and 

adult survival were most influential in population growth rates (Table S7). For the 

deterministic Low scenario, wild survival rates were most sensitive with age-1 survival 



 

 

176 

 

sensitivity = 0.501 and adult (age-3+) survival sensitivity = 0.992. Elasticity analysis 

revealed adult mortality as the most influential stage driving population growth 

(elasticity = 0.988), followed by fertility, 𝐹3+𝑆0, rates (elasticity = 0.016). The 

elasticity of the adult survival parameter decreased (elasticity = 0.462) under the High 

scenario (Table S8). Finally, perturbation analysis revealed that the number of 

hatchery fish stocked, and their post-stocking survival, does not substantively affect λ, 

which was primarily driven by wild recruitment and adult mortality rates for stocked 

juvenile fish that survived to age-3+ and potential wild-reproduced fish (Tables S7, 

S8).  

Hatchery stocking success 

 Our in situ demographic analysis revealed that accruing a standing stock target 

of 𝑁 = 1,000 adult (age-3+) spawners (lake-wide density 0.21 fish/ha) with current 

stocking efforts is not attainable for the given rates (Table 2). For the Low juvenile 

survival scenario, reaching this standing stock target, independent of long-term 

population establishment, required stocking fall fingerlings at densities > 7,500 

fish/ha, two orders of magnitude higher than current hatchery production capabilities. 

Alternatively, stocking yearlings at densities ≥ 4.3 fish/ha achieved the standing stock 

target. For the High scenario, the standing stock target was achieved with fall 

fingerling stocking densities > 19 fish/ha annually (Table 2). This result suggests fall 

fingerling stocking is sufficient under the High scenario, although this scenario 

represents a greater than 30x increase in the survival of these young-of-year fish 

(Table 3). When using hatchery costs reported by NYSDEC (see Table S9), we 

estimated the annual cost to accrue 𝑁 = 100 adult spawners per year with current 



 

 

177 

 

stocking practices is > $5 million USD for fall fingerlings, compared to $829,000 

USD for yearlings. Given low survival and high costs to achieve management 

objectives with stocking fall fingerlings, we proceeded with population viability 

analyses using yearling stocking only.  

Viability of the reintroduced population 

Population viability analysis featuring simulations of environmental 

stochasticity indicated that long-term recovery of the reintroduced Keuka Lake Cisco 

population is unlikely at the derived survival rates (Figure 4). Across stochastic 

scenarios, the modeled population always fell below an extinction threshold of 𝑁 =

100 adults and reached 100% probability of extinction by the long-term time horizon 

at 50 years (Figure 4c). While boom recruitment frequency is not directly managed, 

periodic, strong recruitment events reflect ecological realism observed in many Cisco 

populations and provided insights into the potential future trajectories of the 

reintroduced population in Keuka Lake. The 3-year boom recruitment scenario had a 

stochastic growth rate 𝜆𝑠 = 0.87. An increase in duration between boom recruitment 

years, however, increased the risk of population collapse (𝜆𝑠 = 0.78 for the 7-year 

recruitment scenario). Boom recruitment scenarios increased the population growth 

rate and projected adult fish density compared to deterministic models, though 

conditions were insufficient for long-term recovery.  

Population viability analysis also revealed that even under the most optimistic 

survival and recruitment assumptions, the reintroduced Cisco population was at risk of 

collapse soon after current stocking efforts cease (modeled yearling release of 2,000 

fish/year; Figure 4). Projected populations under optimistic conditions failed to reach 



 

 

178 

 

adult densities reflective of self-sustaining populations (see Table S10). Under the 

most optimistic scenario with a modeled 3-year boom recruitment cycle, the present 

10-year stocking strategy yielded a median 0.24 adults/ha (Figure 4b) at year 10. In 

addition, while 30% of simulated trajectories under the most optimistic scenario 

exceeded the management target of 0.21 adults/ha at year 10, no trajectories exceeded 

this target long-term at year 50 (Figure 4a). Doubling the stocking duration to 20 

years increased the probability of successfully accruing an adult standing stock, yet 

trajectories still showed long-term decline. Projected adult densities increased to ~1.0 

adults/ha (maximum trajectory values) over a longer time horizon by year 39, which 

approached lower density estimates observed in extant Cisco populations (Table S10). 

Although 53% of trajectories exceeded the standing stock target after 20 years of 

stocking, almost all simulated population trajectories still decreased at the cessation of 

stocking, with < 0.01% of trajectories exceeding 0.21 adults/ha at year 50.  

Our random matrix approach to simulate environmental stochasticity of age-0 

vital rates provided further insight into key stages that drive population growth rates 

(Figure 5). Simulations revealed that low wild juvenile survival could be ameliorated 

by higher adult survival. For example, lower adult mortality reflective of literature 

values (see Table S6) increased the probability of drawing a matrix with 𝜆 > 1.0 and 

increased the average population growth rate (𝜆 geometric mean, 𝜆̅𝑔𝑒𝑜𝑚) across the 

sampled set of matrices (𝜆̅𝑔𝑒𝑜𝑚 = 0.86 for 70% literature-based survival, compared to 

𝜆̅𝑔𝑒𝑜𝑚 = 0.69 for 51.4% in situ survival). Nevertheless, simulation results indicated 

that boom recruitment years in the study system were critical for long-term population 

viability, despite higher adult survival estimates. Compared to our environmental 



 

 

179 

 

matrix approach, the only modeled High scenario with positive population growth 

occurred with a 3-year boom likelihood and a literature-based 70% adult survival rate, 

where 𝜆𝑠 = 1.04 (Table S11).      

 

Discussion 

Conservation stocking is an important management tool for fisheries 

restoration, yet the recovery challenges presented in this study underscore the need to 

use system-specific data for accurately assessing population viability. Our study of 

Cisco reintroductions to Keuka Lake found that present fish stocking practices are 

unlikely to achieve management targets. Further, our results from population viability 

analysis using in situ juvenile and adult survival rates highlight that Keuka Lake 

conditions are likely prohibitive for the long-term viability of reestablishing a Cisco 

population. By contrast, our data-driven approach was successful for informing 

adaptive stocking as solely relying on life history parameters extrapolated from other 

systems would have led to false optimism for restoration success. Despite testing a 

range of juvenile survival estimates and a literature-derived adult mortality estimate in 

our population assessment, population recovery remained unlikely, even under the 

most optimistic scenarios. 

Post-stocking survival assessments and population analyses highlight two 

demographic processes that are important for future Cisco recovery efforts in Keuka 

Lake: 1) high mortality of stocked juvenile fish immediately upon release (previously 

difficult to estimate without acoustic telemetry technology) points to the need to 

implement alternative stocking practices, and 2) improved lake conditions for higher 



 

 

180 

 

wild juvenile and adult survival rates are needed to increase the probability of 

establishing a self-sustaining population.  

Empirical estimates from Keuka Lake indicated that current fall fingerling 

stocking rates fail to result in a long-term sustainable population growth rate, and thus 

stocking efforts focused on releasing this younger stage will not achieve spawner 

targets.  High immediate mortality of fall fingerlings was attributed to a combination 

of heavy Lake Trout predation, avian predation, and physiological stress from release 

(Koeberle et al. 2023). While older, larger yearling Cisco are also susceptible to these 

factors, multistage survival modeling revealed that stocking of yearlings increased 

long-term survival and the ability to accrue an adult standing stock. For hatchery 

managers, a more practical approach identified in our stocking analysis could combine 

increased yearling Cisco production from 2,000 fish/year to 3,400 fish/year and 

modified rearing and release practices to improve their stocking-related survival ≈ 4𝑥 

(upper 95% Credible Interval). High juvenile mortality, however, cannot be overcome 

by increasing the number of stocked yearlings alone due to current hatchery 

production constraints. These field-based insights have facilitated adaptive 

management of stocking. For example, in 2025 lake managers plan to hold surplus fall 

fingerlings through winter to explore spring yearling stocking (n = 2,000) and 

potentially stock surplus adults (n = 400) in Keuka Lake (see Table S1). Managers are 

also considering modified rearing or release practices for future cohorts of stocked 

juvenile Cisco.  

System-specific adult survival rates estimated from historical (pre-extirpation) 

Cisco catch data combined with our in situ juvenile survival estimates result in a low 



 

 

181 

 

probability of establishing a self-sustaining population. Under optimistic (High 

juvenile survival) scenarios, we calculated adult annual survival would need to 

improve from in situ 51.4% to > 66% (3-year boom recruitment) or > 76% (7-year 

boom recruitment) to achieve a stable population. Such rates are within the range of 

adult Cisco mortality estimates observed in the Great Lakes basin (see Table S6), 

however their application to Keuka Lake could lead to false optimism for population 

restoration if current Great Lakes conditions for Cisco may be better than in inland 

lakes. Although perturbation analysis revealed adult mortality and recruitment were 

important to the projected population, adult mortality cannot easily be manipulated by 

managers. 

Post-release survival assessments revealed that stocking enough fish to achieve 

adult Cisco density targets in Keuka Lake would require numbers beyond current 

hatchery capacity. This result is consistent with Rook et al. (2021) who found that 

restoring Cisco populations to historic levels in the Great Lakes required stocking fish 

at densities two orders of magnitude higher than present stocking rates. Increased 

mortality of hatchery fish compared to wild fish of equivalent age is well-documented 

in hatchery stocking programs for salmonids (see Brown and Day 2002; Saloniemi et 

al. 2004; Araki et al. 2007; Beamish et al. 2012; Kitada 2020; James et al. 2023). 

Extended time in the hatchery increases the risk of reduced fitness to environmental 

conditions (Brown and Day 2002; Jachowski et al. 2016), which in Keuka Lake may 

be reflected by stocked yearlings spending more time in the second acclimatization 

stage of the multistage model than stocked fall fingerlings. Studies demonstrate 

modified rearing practices can effectively reduce stress and improve fitness of 



 

 

182 

 

hatchery fish, including in-tank structure, e.g., gravel substrate and overhead covers 

(Cogliati et al. 2019), periods of light and dark conditions (Maynard et al. 1995; 

Brown and Day 2002), and varying growth rates, diet composition, and feed particle 

size (Cogliati et al. 2023). Chemical or physical predator cues applied in hatcheries 

also improve predator avoidance behavior (Manassa and McCormick 2012; Wilson et 

al. 2021). While management strategies in North America primarily stock juvenile 

Cisco, techniques such as translocations of adults or in situ placement of early life 

stages (e.g., eggs or larvae) have also been implemented for coregonine restoration in 

Europe (Maitland and Lyle 2012; Adams et al. 2014; Bunnell et al. 2024), and studies 

for salmonids have called for reduced times in hatcheries to minimize the lack of 

natural selection (Lennox et al. 2021). Studies have also found that smaller, younger 

fish acclimated to outdoor ponds before release survive better than older, larger fish 

held in hatchery facilities (Olson et al. 2000). Future coregonine research could 

investigate whether alternative rearing and release strategies impact the survival of 

stocked fish, and thus their recruitment to an adult standing stock. Such strategies 

could then be incorporated into perturbation analyses to evaluate their costs, benefits, 

and effects on model parameters in population assessments (Nichols and Hines 2002).  

The Keuka Lake Cisco reintroduction demonstrates that a combination of 

acoustic telemetry and time-to-event modeling now enables the estimation of juvenile 

fish life history parameters in situ, providing managers with empirical demographic 

information central to the evaluation of stocking efforts. Combined with multistage 

survival modeling, we expect these approaches will prove useful for identifying key 

mortality stages for stocked fish. In Keuka Lake, significant mortality at stocking and 



 

 

183 

 

through acclimation supports testing modified release practices to increase survival 

through these initial stages and to reduce the influence of loading, transport, and 

stocking related stress. For example, net pen acclimation whereby fish are held in nets 

in situ for several days to weeks may prevent high initial predation, promote lake 

acclimatation, and improve the transition to wild defensive behaviors such as 

schooling (Brown and Day 2002; Rillahan et al. 2011). A study using net pens for 

Chinook Salmon (Oncorhynchus tshawytscha) in Lake Ontario found that smolt-to-

lake harvest rates were 1.7 – 2.3 times higher for pen acclimated fish than those 

directly stocked (Connerton et al. 2022). Further, Cisco exhibit predator-driven diel 

migration (Stockwell et al. 2010). Our findings that ≈ 80% of stocked fall fingerlings 

perished within the first several hours of daytime release into Keuka Lake support 

testing whether nighttime stocking could alleviate predation mortality upon stocking 

(Roberts et al. 2009). 

Population viability analysis was important to evaluate the risks and 

uncertainty of restoration scenarios. The failure to achieve positive population growth 

across modeled scenarios indicates that present ecosystem conditions are not suitable 

to restore a Cisco population to Keuka Lake. Positive population growth rates were 

not achieved, even under scenarios reflective of optimistic environmental conditions. 

If strong year classes for this species are linked to ice cover, which has decreased in 

recent decades in the Northern Hemisphere, time between strong recruitment years 

could increase (Sharma et al. 2019; Fielder and McDonnell 2024; Brown et al. 2024). 

This increased duration resulted in a higher extinction risk in our models. Population 

assessment outcomes are also reinforced by the absence of empirical evidence for 



 

 

184 

 

long-term survival and recruitment from netting surveys. Since the onset of 

reintroductions to date, no larval Cisco have been collected in annual spring larval fish 

surveys. The only evidence of multi-year survival is one Cisco netted in July 2022 

during a lake-wide bottom gill net survey and estimated from otolith and scale analysis 

to be 2 years 8 months old (see Figure S5). This age coincides with the October 2020 

fall fingerling release, with N = 6 (95% Credible Interval < 1, 401) fall fingerlings 

estimated to have survived to this time period as age-2+ fish.  

The Cisco population analysis presented here poses several limitations. First, 

wild-derived survival estimates of juvenile Cisco in Keuka Lake were estimated from 

tagged fish. Any tagging effect could therefore bias hatchery estimates which in turn 

are propagated to wild-equivalent estimates. In addition, the sample size of fish 

surviving to the third stage is low. Our simulations also lacked features of 

demographic stochasticity and the Allee effect, processes expected to increase 

extinction rates (Lande and Orzack 1988). While decreased predation through 

schooling has been observed in Cisco (Milne et al. 2005; de Kerckhove et al. 2015), 

our models lacked depensatory mortality through a schooling effect, where survival 

increases with larger populations and schooling effectiveness decreases with smaller 

school size (Clark 1974; Magurran 1990). Future modeling efforts could explore how 

stocking density and fish behavior impact schooling effectiveness and post-release 

mortality of Cisco in Keuka Lake and for other stocked schooling pelagic species. 

Additionally, although adult mortality was important in our perturbation analysis, we 

lack information on how much adult mortality varies annually and the in situ rates 

applied to population modeling were estimated from a time when the Cisco population 



 

 

185 

 

was in decline. Lastly, our analysis focused on demographic processes and did not 

explore lake ecosystem drivers of Cisco establishment. For example, future 

assessments could integrate perturbation analysis with environmental covariates 

important for coregonine fishes such as temperature (Jacobson et al. 2010; Fang et al. 

2012; Stewart et al. 2021), oxythermal habitat and nutrient availability (Jacobson et al. 

2008; Jacobson et al. 2010; Magee et al. 2019), and ice cover duration and extent 

(Karjalainen et al. 2015; Brown et al. 2022) to further identify minimum conditions 

required for achieving positive population growth (Polansky et al. 2024). 

Understanding these conditions is important to achieve management objectives as fish 

reintroductions are often more successful if the mechanisms causing decline are 

identified and addressed (see Mrnak et al. 2025).  

Species reintroduction efforts are challenging and adaptive stocking practices 

that apply in situ demography and population modeling could improve the probability 

of success. We found that an absence of in situ monitoring and over-reliance on 

borrowing life history information extrapolated across systems could lead to false 

optimism of ecosystem conditions for Keuka Lake and thus Cisco restoration success. 

Unsuccessful restoration attempts are underrepresented in peer-reviewed literature yet 

offer insights into future efforts elsewhere (Schaub et al. 2009; Jachowski et al. 2016). 

Our modeling approach elucidated in situ juvenile survival processes previously 

difficult to measure for coregonine species and wild population dynamics that hinder 

the ability to restore a pelagic forage fish species in a temperate lake.   



 

 

186 

 

 

Acknowledgements 

The authors thank multiple collaborating State and Federal agency partners for their 

ongoing work on the Keuka Lake Cisco restoration program. We thank NYSDEC: 

Steve Robb, Bree Minges, Ariel Thoms, Pete Austerman, Matthew Sanderson, and 

Ben Carson, and USGS-Tunison: Jeremy Kraus and Gregg Mackey. We also thank 

Andrew Siebert and Lynn Johnson for their helpful feedback on analyses, Taylor 

Brown, Beth Holbook, Will French, and Jared Myers for their input on Cisco life 

history, and Dave Fielder, Kevin McDonnell, and Ben Rook for their valuable 

suggestions on population models. Lastly, we thank two anonymous reviews for 

providing critical feedback that greatly improved this manuscript. Any use of trade, 

firm, or product names is for descriptive purposes only and does not imply 

endorsement by the U.S. Government. 

 

Data availability statement 

All data and code used for this research are publicly available via Dryad: 

https://doi.org/10.5061/dryad.pzgmsbcz6. 

 

Funding Information 

This research was funded by New York State Department of Environmental 

Conservation using Federal Aid Sport Fish Restoration Funds from “Grant F64-R”. 

  



 

 

187 

 

References 

Adams, C.E., Lyle, A.A., Dodd, J.A., Bean, C.W., Winfield, I.J., Gowans, A.R., 

Stephen, A. and Maitland, P.S. 2014. Translocation as a conservation tool: case 

studies from rare freshwater fishes in Scotland. The Glasgow Naturalist, 26(1), 

pp.17-24. 

 

Araki, H., Ardren, W.R., Olsen, E., Cooper, B. and Blouin, M.S. 2007. Reproductive 

success of captive‐bred steelhead trout in the wild: evaluation of three hatchery 

programs in the Hood River. Conservation Biology, 21(1), pp. 181-190. 

https://doi.org/10.1111/j.1523-1739.2006.00564.x 

 

Armstrong, D. P., and Seddon, P. J. 2008. Directions in reintroduction biology. Trends 

in Ecology & Evolution, 23(1), 20-25. https://10.1016/j.tree.2007.10.003 

 

Axenrot, T., and Degerman, E. 2016. Year-class strength, physical fitness and 

recruitment cycles in vendace (Coregonus albula). Fisheries Research, 173, 61-69. 

https://doi.org/10.1016/j.fishres.2015.03.017 

 

Bacon, P. J., Malcolm, I. A., Fryer, R. J., Glover, R. S., Millar, C. P., and Youngson, 

A. F. 2015. Can conservation stocking enhance juvenile emigrant production in wild 

Atlantic salmon? Transactions of the American Fisheries Society, 144(3), 642-654.  

https://doi.org/10.1080/00028487.2015.1017655 

 

Beamish, R.J., Sweeting, R.M., Neville, C.M., Lange, K.L., Beacham, T.D. and 

Preikshot, D. 2012. Wild chinook salmon survive better than hatchery salmon in a 

period of poor production. Environmental Biology of Fishes, 94(1), pp. 135-148.  

https://doi.org/10.1007/s10641-011-9783-5 

 

Beissinger, S. R. 2002. Population viability analysis: past, present, future. Population 

viability analysis. University of Chicago Press, Chicago, (Ch. 1) pp. 5-17. 

 

Bloomfield, J.A. 1978. Lakes of New York State: Ecology of the Lakes of Western 

New York (Vol. 2). Elsevier. 

 

Boyce, M.S. 1992. Population viability analysis. Annual Review of Ecology and 

Systematics, 23, 481-506. 

 

Brown, C. and Day, R.L. 2002. The future of stock enhancements: lessons for 

hatchery practice from conservation biology. Fish and Fisheries, 3(2), pp. 79-94.  

https://doi.org/10.1046/j.1467-2979.2002.00077.x 

 

Brown, T.A., Sethi, S.A., Rudstam, L.G., Holden, J.P., Connerton, M.J., Gorsky, D., 

Karboski, C.T., Chalupnicki, M.A., Sard, N.M., Roseman, E.F. and Prindle, S.E. 2022. 

Contemporary spatial extent and environmental drivers of larval coregonine 



 

 

188 

 

distributions across Lake Ontario. Journal of Great Lakes Research, 48(2), pp. 359-

370. https://doi.org/10.1016/j.jglr.2021.07.009 

 

Brown, T., Rudstam, L., Sethi, S., Ripple, P., Smith, J., Treska, T., Hessell, C., Olsen, 

E., He, J., Jonas, J., Rook, B., Blankenheim, J., Beech, S., Brown, E., Berglund, E., 

Cook, H.A., Dunlop, E., James, S., Pothoven, S., Amidon, Z., Sweka, J., Carl, D., 

Hansen, S., Bunnell, D.B., Weidel, B., and Honsey, A. 2024. Reconstructing half a 

century of coregonine recruitment reveals species-specific dynamics and synchrony 

across the Laurentian Great Lakes. ICES Journal of Marine Science, 82(2), fsae160. 

https://doi.org/10.1093/icesjms/fsae160 

 

Bunnell, D.B., Ackiss, A.S., Alofs, K.M., Brant, C.O., Bronte, C.R., Claramunt, R.M., 

Dettmers, J.M., Honsey, A.E., Mandrak, N.E., Muir, A.M. and Santucci Jr, V.J. 2023. 

A science and management partnership to restore coregonine diversity to the 

Laurentian Great Lakes. Environmental Reviews, 31(4), pp. 716-738.  

https://doi.org/10.1139/er-2022-0109 

 

Bunnell, D.B., Anneville, O., Baer, J., Bean, C.W., Kahilainen, K.K., Sandström, A., 

Selz, O.M., Vonlanthen, P., Wanzenböck, J. and Weidel, B.C. 2024. How diverse is 

the toolbox? A review of management actions to conserve or restore coregonines. 

International Journal of Limnology, 60, p. 5. https://doi.org/10.1051/limn/2024002 

 

Caswell, H. 2001. Matrix Population Models, Second Edition. Sinauer Associates, Inc. 

Publishers, Sunderland, Massachusetts. 

 

Clark, C. W. 1974. Possible effects of schooling on the dynamics of exploited fish 

populations. ICES Journal of Marine Science, 36(1), pp. 7-14. 

https://doi.org/10.1093/icesjms/36.1.7 

 

Cochran‐Biederman, J. L., Wyman, K. E., French, W. E., and Loppnow, G. L. 2015. 

Identifying correlates of success and failure of native freshwater fish reintroductions. 

Conservation Biology, 29(1), pp. 175-186. https://doi.org/10.1111/cobi.12374 

 

Cogliati, K.M., Herron, C.L., Noakes, D.L. and Schreck, C.B. 2019. Reduced stress 

response in juvenile Chinook Salmon reared with structure. Aquaculture, 504, pp. 96-

101. https://doi.org/10.1016/j.aquaculture.2019.01.056 

 

Cogliati, K.M., Noakes, D.L., Khan, F., Sharpe, C.S. and Schreck, C.B. 2023. 

Producing wild fish phenotypes in hatchery-reared fish. Environmental Biology of 

Fishes, 106(5), pp.1113-1134. https://doi.org/10.1007/s10641-022-01279-9 

 

Connerton, M.J., Lantry, J.R., Bronte, C.R. and Lapan, S.R. 2022. Origin, postrelease 

survival, and imprinting of pen‐acclimated and direct‐stocked Chinook Salmon in 

Lake Ontario. North American Journal of Fisheries Management, 42(3), pp. 713-740.  

https://doi.org/10.1002/nafm.10756 



 

 

189 

 

 

Cowx, I. G. 1994. Stocking strategies. Fisheries management and ecology, 1(1), pp. 

15-30. https://doi.org/10.1111/j.1365-2400.1970.tb00003.x 

 

de Kerckhove, D. T., and Shuter, B. J. 2022. Predation on schooling fish is shaped by 

encounters between prey during school formation using an Ideal Gas Model of animal 

movement. Ecological Modelling, 470, 110008. 

https://doi.org/10.1016/j.ecolmodel.2022.110008 

 

Ellner, S. P., and Fieberg, J. 2003. Using PVA for management despite uncertainty: 

effects of habitat, hatcheries, and harvest on salmon. Ecology, 84(6), pp. 1359-1369.  

https://doi.org/10.1890/0012-9658(2003)084[1359:UPFMDU]2.0.CO;2 

 

Eshenroder, R.L., Vecsei, P., Gorman, O.T., Yule, D., Pratt, T.C., Mandrak, N.E., 

Bunnell, D.B. and Muir, A.M. 2016. Ciscoes (Coregonus, subgenus Leucichthys) of 

the Laurentian Great Lakes and Lake Nipigon. Great Lakes Fishery Commission, Ann 

Arbor, Michigan.  

 

Fang, X., Jiang, L., Jacobson, P.C., Stefan, H.G., Alam, S.R., and Pereira, D.L. 2012. 

Identifying cisco refuge lakes in Minnesota under future climate scenarios. 

Transactions of the American Fisheries Society, 141(6), pp. 1608-1621. 

https://doi.org/10.1080/00028487.2012.713888 

 

Fieberg, J. and Ellner, S.P. 2001. Stochastic matrix models for conservation 

and management: a comparative review of methods. Ecology Letters, 4(3), pp. 244-

266. https://doi.org/10.1046/j.1461-0248.2001.00202.x 

 

Fielder, D.G. and McDonnell, K.N. 2024. Evaluation of potential factors affecting the 

success of Cisco re-introduction and re-establishment in Lake Huron. Ecological 

Modelling, 496, p. 110817. https://doi.org/10.1016/j.ecolmodel.2024.110817 

 

Fisch, N.C., Bence, J.R., Myers, J.T., Berglund, E.K. and Yule, D.L. 2019. A 

comparison of age-and size-structured assessment models applied to a stock of cisco 

in Thunder Bay, Ontario. Fisheries Research, 209, pp. 86-100. 

https://doi.org/10.1016/j.fishres.2018.09.014 

 

Fonken, D.R., Conner, M.M., Walsworth, T.E. and Thompson, P.D. 2023. Benefits of 

stocking fewer but larger individuals with implications for native fish recovery. 

Canadian Journal of Fisheries and Aquatic Sciences, 80(3), pp. 439-450.  

https://doi.org/10.1139/cjfas-2022-0071 

 

Gatch, A.J., Gorsky, D., Weidel, B.C., Biesinger, Z.F., Connerton, M.J., Davis, C., 

Lachance, H. and O'Malley, B.P. 2023. Seasonal habitat utilization provides evidence 

for site fidelity during both spawn and non-spawning seasons in Lake Ontario cisco 

Coregonus artedi. Journal of Great Lakes Research, 49(5), pp. 1045-1058.  



 

 

190 

 

https://doi.org/10.1016/j.jglr.2023.06.008 

 

Hrabik, T. R., Magnuson, J. J., and McLain, A. S. 1998. Predicting the effects of 

rainbow smelt on native fishes in small lakes: evidence from long-term research on 

two lakes. Canadian Journal of Fisheries and Aquatic Sciences, 55(6), pp. 1364-1371.  

https://doi.org/10.1139/f98-032 

 

Jachowski, D. S., Millspaugh, J. J., Angermeier, P. L., and Slotow, R. (Eds.). 2016. 

Reintroduction of fish and wildlife populations. Univ of California Press. 

Jacobson, P.C., Jones, T.S., Rivers, P., and Pereira, D.L. 2008. Field estimation of 

lethal oxythermal niche boundary for adult ciscoes in Minnesota lakes. Transactions of 

the American Fisheries Society 137(5), pp. 1464-1474. https://doi.org/10.1577/T07-

148.1 

 

Jacobson, P.C., Stefan, H.G., and Pereira, D.L. 2010. Coldwater fish oxythermal 

habitat in Minnesota lakes: influence of total phosphorus, July air temperature, and 

relative depth. Canadian Journal of Fisheries and Aquatic Sciences 67(12), pp. 2002-

2013. https://doi.org/10.1139/F10-115 

 

James, S.E., Doherty, B., Cox, S.P., Pearsall, I.A. and Riddell, B. 2023. Size and 

timing of hatchery releases influence juvenile-to-adult survival rates of British 

Columbia Chinook (Oncorhynchus tshawytscha) and coho (Oncorhynchus kisutch) 

salmon. Canadian Journal of Fisheries and Aquatic Sciences, 80(4), pp. 700-718.  

https://doi.org/10.1139/cjfas-2022-0121 

 

Karjalainen, J., Keskinen, T., Pulkkanen, M., and Marjomäki, T. J. 2015. Climate 

change alters the egg development dynamics in cold-water adapted coregonids. 

Environmental Biology of Fishes, 98, pp. 979-991. https://doi.org/10.1007/s10641-

014-0331-y 

 

Kitada, S. 2020. Lessons from Japan marine stock enhancement and sea ranching 

programmes over 100 years. Reviews in Aquaculture, 12(3), pp. 1944-1969. 

https://doi.org/10.1111/raq.12418 

 

Koeberle, A.L., Pearsall, W., Hammers, B.E., Mulhall, D., McKenna Jr, J.E., 

Chalupnicki, M. and Sethi, S.A. 2023. Whole-lake acoustic telemetry to evaluate 

survival of stocked juvenile fish. Scientific Reports, 13(1), p. 18956.  

https://doi.org/10.1038/s41598-023-46330-6 

 

Kubečka, J., Godø, O.R., Hickley, P., Prchalováa, M., Říha, M., Rudstam, L., 

Welcommee, R. 2012. Fish sampling with active methods. Fisheries Resources, 123, 

pp. 1-3. https://doi.org/10.1016/j.fishres.2011.11.013 

 

Lande, R., and Orzack, S. H. 1988. Extinction dynamics of age-structured populations 

in a fluctuating environment. Proceedings of the National Academy of 



 

 

191 

 

Sciences, 85(19), pp. 7418-7421. https://doi.org/10.1073/pnas.85.19.7418 

 

Lefkovitch, L. P. 1965. The study of population growth in organisms grouped by 

stages. Biometrics 21, pp. 1-18. https://doi.org/10.2307/2528348 

 

Lennox, R.J., Alexandre, C.M., Almeida, P.R., Bailey, K.M., Barlaup, B.T., Bøe, K., 

Breukelaar, A., Erkinaro, J., Forseth, T., Gabrielsen, S.E. and Halfyard, E. 2021. The 

quest for successful Atlantic salmon restoration: perspectives, priorities, and maxims. 

ICES Journal of Marine Science, 78(10), pp. 3479-3497. 

https://doi.org/10.1093/icesjms/fsab201 

 

Leslie, P. H. 1948. Some further notes on the use of matrices in population 

mathematics. Biometrika, 35(3/4), pp. 213-245. https://doi.org/10.2307/2332342 

 

Magee, M.R., McIntyre, P.B., Hanson, P.C. and Wu, C.H. 2019. Drivers and 

management implications of long-term Cisco oxythermal habitat decline in Lake 

Mendota, WI. Environmental Management, 63(3), pp. 396-407.  

https://doi.org/10.1007/s00267-018-01134-7 

 

Magurran, A. E. 1990. The adaptive significance of schooling as an anti-predator 

defence in fish. In Annales Zoologici Fennici, 27, pp. 51-66.  

 

Maitland, P.S. and Lyle, A.A. 2013. Ex situ and in situ approaches, including assisted 

reproduction, for the conservation of native species of charr (Salmonidae) and 

whitefish (Coregonidae) in Scotland. International Zoo Yearbook, 47(1), pp. 129-139.  

https://doi.org/10.1111/j.1748-1090.2012.00192.x 

 

Manassa, R.P. and McCormick, M.I. 2012. Social learning and acquired recognition of 

a predator by a marine fish. Animal Cognition, 15(4), pp. 559-565.  

https://doi.org/10.1007/s10071-012-0484-z 

 

Marjomäki, T. J. 2005. Evaluation of different harvest strategies for a vendace 

population with highly variable recruitment: a simulation approach. Boreal 

Environment Research, 10(4), 255. 

 

Marjomäki, T.J., Auvinen, H., Helminen, H., Huusko, A., Huuskonen, H., Hyvärinen, 

P., Jurvelius, J., Karels, A., Sarvala, J., Valkeajärvi, P. and Karjalainen, J. 2024. 

Vendace populations on the life table: between-lake variation and the association 

between early life and mature survival and growth. International Journal of 

Limnology, 60(11). https://doi.org/10.1051/limn/2024011 

 

Maynard, D. J., Flagg, T. A., and Mahnken, C. V. 1995. A review of seminatural 

culture strategies for enhancing the postrelease survival of anadromous salmonids. In 

American Fisheries Society Symposium, vol. 15, pp. 307-314. 

 



 

 

192 

 

McKenna Jr, J.E., Sethi, S.A., Scholten, G.M., Kraus, J. and Chalupnicki, M. 2021. 

Acoustic tag retention and tagging mortality of juvenile cisco Coregonus artedi. 

Journal of Great Lakes Research, 47(3), pp. 937-942. 

https://doi.org/10.1016/j.jglr.2021.03.020 

 

McMichael, G.A., Eppard, M.B., Carlson, T.J., Carter, J.A., Ebberts, B.D., Brown, 

R.S., Weiland, M., Ploskey, G.R., Harnish, R.A. and Deng, Z.D. 2010. The Juvenile 

Salmon Acoustic Telemetry System: A New Tool. Fisheries, 35(1), pp. 9-22.  

https://doi.org/10.1577/1548-8446-35.1.9 

Milne, S. W., Shuter, B. J., and Sprules, W. G. 2005. The schooling and foraging 

ecology of lake herring (Coregonus artedi) in Lake Opeongo, Ontario, Canada. 

Canadian Journal of Fisheries and Aquatic Sciences, 62(6), pp. 1210-1218. 

https://doi.org/10.1139/f05-030 

 

Mitchell, L., Newman, K. and Baxter, R. 2019. Estimating the size selectivity of 

fishing trawls for a short-lived fish species. San Francisco Estuary and Watershed 

Science, 17(1). https://doi.org/10.15447/sfews.2019v17iss1art5 

 

Mrnak, J. T., Sikora, L. W., Zanden, M. J. V., and Sass, G. G. 2023. Applying 

panarchy theory to aquatic invasive species management: a case study on invasive 

rainbow smelt Osmerus mordax. Reviews in Fisheries Science & Aquaculture, 31(1), 

pp. 66-85. https://doi.org/10.1080/23308249.2022.2078951 

 

Mrnak, J.T., Wilkinson, M.V., Sikora, L.W., Feucht, L.M., Mrnak, A.M., Vander 

Zanden, M.J. and Sass, G.G. 2025. Invasive control and native restoration: Directing 

ecosystem transformation through purposeful food web manipulations. Fisheries, 

p.vuae017.https://doi.org/10.1093/fshmag/vuae017 

 

Münzbergová, Z., and Ehrlén, J. 2005. How best to collect demographic data for 

population viability analysis models. Journal of Applied Ecology, 42(6), pp. 1115-

1120. https://doi.org/10.1111/j.1365-2664.2005.01099.x 

 

Murphy, B.R. and Willis, D.W. eds., 1996. Fisheries techniques (2nd ed., p. 732). 

Bethesda, Maryland: American Fisheries Society. 

 

Myers, J.T., Yule, D.L., Jones, M.L., Ahrenstorff, T.D., Hrabik, T.R., Claramunt, 

R.M., Ebener, M.P. and Berglund, E.K. 2015. Spatial synchrony in cisco recruitment. 

Fisheries Research, 165, pp. 11-21. https://doi.org/10.1016/j.fishres.2014.12.014 

 

Nichols, J. D., and J. E. Hines. 2002. Approaches to the direct estimation of λ and 

demographic contributions to λ using capture-recapture data. Journal of Applied 

Statistics, 29(1-4), pp. 539-568. https://doi.org/10.1080/02664760120108809 

 

Nichols, J. D., and Armstrong, D. P. 2012. Monitoring for reintroductions. 

Reintroduction biology: integrating science and management, 223-255. Edited by  



 

 

193 

 

Ewn, J. G., Armstrong, D. P., Parker, K. A., and P. J. Seddon. Blackwell Publishing. 

 

Olson, M.H., Brooking, T.E., Green, D.M., VanDeValk, A.J., Rudstam, L.G. 2000. 

Survival and growth of intensively reared large walleye fingerlings and extensively 

reared small fingerlings stocked concurrently in small lakes. North American Journal 

of Fisheries Management, 20(2), pp. 337-348.  

https://doi.org/10.1577/1548-8675(2000)020%3C0337:SAGOIR%3E2.3.CO;2 

 

Page, L.M. and Burr, B.M. 2011. A field guide to freshwater fishes of North America 

north of Mexico. Boston: Houghton Mifflin Harcourt, 663p. 

 

Polansky, L., Mitchell, L. and Nobriga, M.L. 2024. Identifying minimum freshwater 

habitat conditions for an endangered fish using life cycle analysis. Conservation 

Science and Practice, 6(5), p. e13124. https://doi.org/10.1111/csp2.13124 

 

R Core Team. R: A language and environment for statistical computing. R foundation 

for statistical computing, Vienna. https://www.rproject.org. (2024). 

 

Rillahan, C., Chambers, M.D., Howell, W.H. and Watson III, W.H. 2011. The 

behavior of cod (Gadus morhua) in an offshore aquaculture net pen. Aquaculture, 

310(3-4), pp. 361-368. https://doi.org/10.1016/j.aquaculture.2010.10.038 

 

Roberts, L.J., Taylor, J., Gough, P.J., Forman, D.W. and de Leaniz, C.G. 2009. Night 

stocking facilitates nocturnal migration of hatchery‐reared Atlantic salmon, Salmo 

salar, smolts. Fisheries Management and Ecology, 16(1), pp. 10-13.  

https://doi.org/10.1111/j.1365-2400.2008.00611.x 

 

Rook, B., Hansen, M.J. and Bronte, C.R. 2021. How many ciscoes are needed for 

stocking in the Laurentian Great Lakes? Journal of Fish and Wildlife Management, 

13(1), pp.28-49. https://doi.org/10.3996/JFWM-21-025 

 

Rook, B. J., Hansen, M. J., Goldsworthy, C. A., Ray, B. A., Gorman, O. T., Yule, D. 

L., and Bronte, C. R. 2021. Was historical cisco Coregonus artedi yield consistent 

with contemporary recruitment and abundance in Lake Superior? Fisheries 

Management and Ecology, 28(3), pp. 195-210. https://doi.org/10.1111/fme.12474 

 

Rudstam, L. G., Magnuson, J. J., and Tonn, W. M. 1984. Size selectivity of passive 

fishing gear: a correction for encounter probability applied to gill nets. Canadian 

Journal of Fisheries and Aquatic Sciences, 41(8), pp. 1252-1255. 

https://doi.org/10.1139/f84-151 

 

Saloniemi, I., Jokikokko, E., Kallio-Nyberg, I., Jutila, E. and Pasanen, P. 2004. 

Survival of reared and wild Atlantic salmon smolts: size matters more in bad years. 

ICES Journal of Marine Science, 61(5), pp. 782-787. 

https://doi.org/10.1016/j.icesjms.2004.03.032 



 

 

194 

 

 

Sarvala, J., Helminen, H., Karjalainen, J., Marjomäki, T. J., Forsman, T., and Anttila, 

L. 2024. Long-term decline of whitefish (Coregonus lavaretus) population in the 

boreal lake Pyhäjärvi, southwest Finland, relative to simultaneous abiotic and biotic 

changes. International Journal of Limnology, 60, 16. 

https://doi.org/10.1051/limn/2024009 

 

Schaub, M., Zink, R., Beissmann, H., Sarrazin, F., and Arlettaz, R. 2009. When to end 

releases in reintroduction programmes: demographic rates and population viability 

analysis of bearded vultures in the Alps. Journal of Applied Ecology, 46(1), pp. 92-

100. https://doi.org/10.1111/j.1365-2664.2008.01585.x 

 

Seddon, P. J., Armstrong, D. P., and Maloney, R. F. 2007. Developing the science of 

reintroduction biology. Conservation Biology, 21(2), pp. 303-312.  

https://doi.org/10.1111/j.1523-1739.2006.00627.x 

 

Sethi, S.A., Koeberle, A.L., Poulton, A.J., Linden, D.W., Diefenbach, D., Buderman, 

F.E., Casalena, M.J. and Duren, K. 2024. Multistage time-to-event models improve 

survival inference by partitioning mortality processes of tracked organisms. Scientific 

Reports, 14(1), p. 14628. https://doi.org/10.1038/s41598-024-64653-w 

 

Sharma, S., Blagrave, K., Magnuson, J.J., O’Reilly, C.M., Oliver, S., Batt, R.D., 

Magee, M.R., Straile, D., Weyhenmeyer, G.A., Winslow, L. and Woolway, R.I. 2019. 

Widespread loss of lake ice around the Northern Hemisphere in a warming world. 

Nature Climate Change, 9(3), pp. 227-231. https://doi.org/10.1038/s41558-018-0393-5 

 

Stewart, T.R., Mäkinen, M., Goulon, C., Guillard, J., Marjomäki, T.J., Lasne, E., 

Karjalainen, J. and Stockwell, J.D. 2021. Influence of warming temperatures on 

coregonine embryogenesis within and among species. Hydrobiologia, 848(18), pp. 

4363-4385. https://doi.org/10.1007/s10750-021-04648-0 

 

Stewart, T.R., Vinson, M.R. and Stockwell, J.D. 2021. Shining a light on Laurentian 

Great Lakes cisco (Coregonus artedi): How ice coverage may impact embryonic 

development. Journal of Great Lakes Research, 47(5), pp. 1410-1418. 

https://doi.org/10.1016/j.jglr.2021.07.002 

 

Stockwell, J.D., Ebener, M.P., Black, J.A., Gorman, O.T., Hrabik, T.R., Kinnunen, 

R.E., Mattes, W.P., Oyadomari, J.K., Schram, S.T., Schreiner, D.R. and Seider, M.J. 

2009. A synthesis of cisco recovery in Lake Superior: Implications for native fish 

rehabilitation in the Laurentian Great Lakes. North American Journal of Fisheries 

Management, 29(3), pp. 626-652. https://doi.org/10.1577/M08-002.1 

 

Thompson, W. (Ed.). 2013. Sampling rare or elusive species: concepts, designs, and 

techniques for estimating population parameters. Island Press. 

 



 

 

195 

 

White, G.C., Franklin, A.B. and Shenk, T.M. 2002. Estimating parameters of PVA 

models from data on marked animals. Population viability analysis. University of 

Chicago Press, Chicago, Illinois, USA, (Ch. 9) pp. 169-190. 

 

Wilson, J.C., White, D.P., Detmer, T.M. and Wahl, D.H. 2021. Behavioral response of 

juvenile silver and bighead carp to conspecific and heterospecific alarm cues. 

Biological Invasions, 23(7), pp. 2233-2248. https://doi.org/10.1007/s10530-021-

02502-x 

 

Yule, D. L., Stockwell, J. D., Cholwek, G. A., Evrard, L. M., Schram, S., Seider, M., 

and Symbal, M. 2006. Evaluation of methods to estimate lake herring spawner 

abundance in Lake Superior. Transactions of the American Fisheries Society, 135(3), 

680-694. https://doi.org/10.1577/T05-203.1 

 

Zimmerman, M.S. and Krueger, C.C. 2009. An ecosystem perspective on re‐

establishing native deepwater fishes in the Laurentian Great Lakes. North American 

Journal of Fisheries Management, 29(5), pp. 1352-1371. https://doi.org/10.1577/M08-

194.1 

  



 

 

196 

 

Tables 

 

 

 
 

 

  



 

 

197 

 

 

  



 

 

198 

 

 

  



 

 

199 

 

Figures 

 

 

Figure 1. Map of Keuka Lake, New York, USA and photograph of an acoustic-tagged 

fall fingerling Cisco (Coregonus artedi). The map uses a Latitude-Longitude 

coordinate system with NAD83 datum. Cisco image from M. Chalupnicki, U.S. 

Geological Survey.   

  



 

 

200 

 

 
Figure 2. Life-cycle graph specifying the population model structure for reintroduced 

Cisco (Coregonus artedi) stocking and wild recruitment in Keuka Lake, New York, 

USA. Nodes 1 – 3 represent wild stages and nodes 1’ and 2’ represent hatchery stages. 

𝑁𝑟𝑒𝑙 refers to the number of hatchery fish released into Keuka Lake as either fall 

fingerling (FF) or yearling (Y) stages.  

  



 

 

201 

 

 
Figure 3. Estimated survival curves, 𝑆̂𝑡 , of stocked juvenile cisco (Coregonus artedi) 

for Keuka Lake, New York, USA. Survival estimates are derived from the top-ranked 

multistage model for acoustic-tagged juvenile hatchery fish which includes an age 

effect on stage-specific mortalities (solid lines) and transition times (solid points). This 

model includes a sequential straight-to-death, acclimation, and natural mortality 

stages, and estimated transition times between stages. Equivalent wild juvenile 

survival (dashed lines) is derived from the third stage mortality rate from the 

multistage model.   

  



 

 

202 

 

 
Figure 4. Population viability analysis of the reintroduced Cisco (Coregonus artedi) 

population in Keuka Lake, New York, USA under the most optimistic juvenile 

survival scenario with a 3-year boom recruitment cycle (e.g., 1/3 boom recruitment 

likelihood). This analysis reveals a low probability of establishing a self-sustained 

reintroduced Cisco population over a 50-year time horizon. Here, we illustrate the a) 

population trajectories (10,000 iterations; random subset 100 iterations shown for 

visualization), b) average population densities, and c) and the long-term probability of 

extinction (threshold 𝑁 = 100 age-3+ adults) for simulated Cisco reintroductions in 

Keuka Lake. All population trajectories are initiated with 𝑁0 = 0 fish and include 

annual stocking rates of 2,000 yearlings for 10 or 20 years. While trajectories indicate 

the population increases above a target spawner density of 0.21 fish/ha (dashed gray 

lines), all trajectories crash soon after stocking ceases and the extinction probability 

asymptotes to 100% by the end of the 50-year management horizon.  

  



 

 

203 

 

 
Figure 5. Sensitivity analysis of population growth rates to adult (age-3+) Cisco 

(Coregonus artedi) annual mortality rates estimated as a) in situ 51.4% survival and b) 

literature-based 70% survival. To simulate environmental stochasticity, we 

constructed 2,500 projection matrices from random sampling of probability 

distributions of fecundity and early life history (age-0) survival values. We then 

bootstrap sampled from this set of random matrices (10,000 iterations) to calculate the 

expected population growth rate λ at each time step. Higher adult survival rates 

reflective of literature-derived estimates increase the probability of achieving positive 

population growth, e.g., 𝑃(𝜆 > 1.0), in contrast to in situ survival estimates from the 

study system. 

 

  



 

 

204 

 

A CHAPTER 1 APPENDIX  

Appendix S1 for Chapter 1: Dissecting causes and consequences of alewife (Alosa 

pseudoharengus) collapse in lake ecosystems  

 

S1. Methods 

 

S1.1. Water quality data 

All water quality sampling in Keuka Lake followed standard operating procedures 

from a state-wide citizen science program (CSLAP 2024). We conducted sensitivity 

analyses and developed protocols to identify and remove potential outliers due to 

measurement errors in each water quality time series dataset. This included removing 

water sample concentrations equal to 0 (3.1 % of total observations; we omitted due to 

uncertainty if past 0 values were a true 0 or N/A), replacing values below the limit of 

detection with limit of detection values (3.5 % of total observations), and removing 

those exceeding three standard deviations from the annual mean (0.09 % of total 

observations). These criteria only applied to Keuka Lake data to reduce potential bias 

and spatial variability introduced by single-site anomalies, such as short-term runoff 

events at sampling sites near tributaries. By contrast, we did not remove outliers for 

Otsego Lake due to consistent sampling at a single deep-water site. Years with only a 

single measurement (e.g., 2007 for chlorophyll a in Otsego Lake) were excluded from 

statistical analyses to ensure robustness. All water quality data were then summarized 

as annual (June-September) mean values before generalized additive models (GAM) 

were applied. 

 

S1.2. Predator fish stocking  

 

Table S1: Annual stocking rates for piscivorous lake trout and walleye in Otsego 

Lake. Walleye were stocked from 2000 to 2014. Neither species are stocked in Keuka 

Lake. 

Year No. lake trout (spring 

fingerlings)  

No. walleye (spring 

fingerlings) 

No. walleye (fall 

fingerlings) 

1992 10,000 0 0 

1993 10,000 0 0 

1994 5,000 0 0 

1995 5,000 0 0 

1996 5,000 0 0 



 

 

205 

 

1997 5,000 0 0 

1998 5,000 0 0 

1999 5,000 0 0 

2000 5,000 80,000 0 

2001 5,000 45,000 8,000 

2002 5,000 45,000 8,000 

2003 5,000 4,500 15,000 

2004 5,000 80,000 0 

2005 5,000 40,000 15,000 

2006 5,000 70,000 5,000 

2007 5,000 40,000 0 

2008 5,000 40,000 10,000 

2009 5,000 0 12,000 

2010 5,000 0 40,000 

2011 5,000 0 28,000 

2012 5,000 0 7,500 

2013 2,500 0 10,000 

2014 2,500 0 5,000 

2015 2,500 0 0 

2016 2,500 0 0 

2017 2,500 0 0 

2018 2,500 0 0 

2019 2,500 0 0 

2020 2,500 0 0 

2021 2,500 0 0 

2022 2,500 0 0 

2023 2,500 0 0 

2024 2,500 0 0 

Totals 145,000 444,500 163,500 

 

S1.3. Net survey specifications  

 

Keuka Lake gear specifications: 

Lake trout (cold-water) standard gang gill net assessment (1979-2022): Finger 

Lakes standard gang bottom gill net surveys use 8ft x 350ft multifilament nets with of 

1” (25 ft. panels on each end), and 50ft panels with 1.5”, 2.0”, 2.5”, 3”, 4”, 5” stretch 

mesh size. Each mesh panel is placed in random order except for the two 25 ft. panels 

on each end. Surveys were conducted in July or August lake-wide with nets set 

overnight (n = 26-32 sites). Sampling methodologies for lake trout experienced some 

slight modifications across the time series (Hammers et al. 2014). In 2011 one half of 

the gill nets were monofilament, and from 2016 onwards all nets were switched to 



 

 

206 

 

monofilament. This was because 25ft panels of 1.0” stretch mesh multifilament nets 

were no longer available from suppliers. Nets deployed with a 10 ft. spreader pole to 

provide vertical stability. Multifilament nets were constructed with polyfoam line 

while the bottom is weighted lead core line. Monofilament nets are constructed with 

plastic floats attached to the float line. A total of 16 monofilament and 16 

multifilament Finger Lakes standard gang gill nets were set, 2 each/night, for two 

weeks in mid-July in 2011. Beginning in 2016, monofilament nets were used 

exclusively. Index sites were selected in 1997 which have been used in all gill net 

surveys since (n = 32 total sites). From 2011 onwards, catch rates are also adjusted for 

monofilament nets using the ratio 1.7:1 mono:multi (Austerman and Hammers 2015; 

Hammers 2018). This ratio was applied to all species caught after 2011 (note, no 

alewife were caught in this survey from 2011 to 2022). 

Forage gill net assessment (2019-2022): In September, n = 31 sites (e.g., 31 surface, 

31 bottom) throughout the lake were sampled with gillnets that target the prey fish 

community. Each monofilament net is 21ft. high and 70ft. with seven mesh sizes 

ranging from 0.25” to 1” bar mesh (e.g., floating and sinking nets are both 21 m x 7 m 

with 3 m panels: 6.25 mm, 8.0 mm, 15.0 mm, 18.5 mm, and 25.0 mm bar mesh 

monofilament). Each set (or site) consisted of two nets, one floating and one bottom, 

fished at approximately 45ft. depths. All nets were set overnight.   

Percid gill net assessment (2022): In August, n = 17 sites throughout Keuka Lake 

were targeted for yellow perch for four nights. Survey protocol following the New 

York State Percid Sampling Manual (Forney et al. 1994) guidelines using standardized 

gill nets. Surface water temperature ranged from 77-79 F. All nets were set 

perpendicular or oblique to shoreline in 10-70ft depths and fished overnight. Gill nets 

are 150 ft x 6 ft. with 25 ft. panels of monofilament 1.5”, 2.0”, 3.0”, 3.5”, and 4.0” 

stretch mesh (not placed in random order).  

Hydroacoustic gill net assessment (2000-2023): Net specifications and the number 

set varied by survey year. In 2000, n = 2 vertical nets were used (seven nets set in a 

row considered one unit for analysis, total coverage 3 m width x 20 m depth for the 

longer net and 3m width x 12m depth for the shorter net, mesh panels are 6.25mm, 

8mm, 10mm, 12.5mm, 15mm, 18.5mm, 25.0 mm bar mesh). See Warner et al. (2002) 

for further description of nets used in 2000. Even though the nets covered different 

water depths, few fish were caught below 12 m depths. For years 2007 (n = 1 surface 

and bottom site), 2011 (n = 1 surface site), and 2016 (n = 4 surface, 3 bottom sites) the 

same forage nets described above in the forage assessment (2019-2022) were used. 

Hydroacoustic survey data from 2023 was paired with data from the lake-wide forage 

netting assessments in 2022.  

 

Otsego Lake net specifications: 



 

 

207 

 

Lake trout (cold-water) standard gang gill net assessment (1994-2024): This 

survey is comparable to the lake trout gill net survey in Keuka Lake. Gill nets are 450 

ft. long with green multifilament gang with three 150 ft. x 4 ft. (wide) panels (Swedish 

Exp). Variable mesh sizes are 1.5”, 2.0”, 2.25”, 2.5”, 3.0”, 3.5” stretch mesh. 

Throughout the survey, there are n = 6 standard sampling sites around Otsego Lake 

(long-term dataset).   

Percid gill net assessment (2003-2019): The warm-water percid netting survey are 

ubiquitous for NYSDEC and comparable to percid net surveys in Keuka Lake. These 

are bottom set nets that measure 150 ft. (length), clear monofilament, six 25 ft. x 6 ft. 

(or 8 ft. wide) panels, with variable stretched mesh size 1.5”, 2.0”, 2.5”, 3.0”, 3.5”, and 

4.0". This survey uses n = 10 standard sites around the lake. Note, non-stretched (bar) 

net mesh sizes are 0.75”, 1.0”, 1.25”, 1.5”, 1.75”, and 2.0" while stretched mesh sizes 

are 1.5”, 2.0”, 2.5”, 3.0”, 3.5”, and 4.0". 

Hydroacoustic gill net assessment (1996-2013): The nets here were identical to the 

nets described for year 2000 in Keuka Lake (Warner et al. 2002). Variable small mesh 

forage gill nets were set concurrent with each survey. Each net was composed of seven 

3 m wide panels of different mesh sizes (6.25, 8, 10, 12.5, 15, 18.5, and 25 mm bar 

mesh). 

Trap net monitoring (1989-2024): This survey was started in 1989 with a net set in 

Rat Cove, Otsego Lake to monitor the invasive alewife population (Foster 1990). A 

second site was added at Brookwood Point in 2002. In subsequent years, following 

alewife mitigation control largely attributed to walleye stocking (Cornwell 2005), 

alewife catch in the trap nets decreased and in 2013 the trap net survey became a 

generic all-species survey. Rat Cove has a silty bottom with abundant aquatic plant 

growth while Brookwood Point has a rocky bottom with little to no plant growth 

(Skelton 2022). Indiana style trap nets were set on Mondays throughout the summer at 

both Rat Cove and Brookwood Point. These nets have a box frame with multiple 

hoops attached to the frame, along with two wings and a lead line. Nets were set 

perpendicular to shore, with the lead line coming off towards the shoreline. Trap nets 

were checked from Tuesday through Friday in a normal week throughout the summer. 

Fish caught in the trap nets were counted, a measurement was taken of their total 

length, and fish were released relatively close to where they were caught. 

 

S1.4. Hydroacoustic survey analysis 

Hydroacoustic methods description: 

In Keuka Lake, with year 2000 surveyed with a Simrad EY-M 70kHz unit, and the 

other years with different 120kHz Biosonics units (Table 1). East-west transects, 

spatially allocated across the lake, were surveyed at night in September to October 



 

 

208 

 

each year (Table S2). Otsego Lake followed similar sampling procedures with similar 

equipment with lake-wide surveys conducted in spring and fall (Table S2). 

Data analysis followed the GLSOP for single echo detection and targets larger than -

60dB considered fish. Target strength distributions showed three peaks roughly 

divided by -50dB and -40dB (Table 2). Fish density was first calculated based on 

average target strength of echoes larger than -60dB and then divided into three groups 

roughly corresponding to young-of-year fish (targets -60 to -50dB), yearling and older 

(targets -50 to -40dB), and larger adult fish such as lake trout (targets > -40dB). Total 

fish densities were calculated for each transect using water above 20 m and water 

below 20 m depth and divided into three size groups. Preliminary evaluation of TS 

distributions by depth suggest that most smaller targets were above 20 m.  Lake-wide 

densities were calculated from transect densities by weighing by the length of the 

transect (cluster sampling) as described in Parker Stetter et al. (2009).  

Mysids were also estimated with hydroacoustics in Keuka Lake. First, the depth 

region containing mysids were delineated using the echograms. Mysids are not found 

in the epilimnion as they do not tolerate temperatures above 16 C (Rudstam et al. 

1999) and are also affected by light (Boscarino et al. 2010). Fish echoes were removed 

by using a fish exclusion threshold and masking out fish echoes in the echograms 

following Rudstam et al. (2008a). Mysid target strength was obtained by comparing 

net hauls with acoustic returns in the 2023 data. We applied the same mysid target 

strength to years 2007, 2011, and 2016 with 120 kHz data, although mysid tows were 

lacking in those years. Mysid densities were not estimated from the year 2000 data as 

mysid target strength is not known for that frequency (Rudstam et al. 2008b).  

  



 

 

209 

 

Table S2: Acoustic units and settings used for estimating fish abundances in Keuka 

Lake from 2000 to 2023 and in Otsego Lake 1996-2013. Pulse rate varied between 1-3 

pulses per second. 

   



 

 

210 

 

(a) 

 

(b) 

 

(c)  

 

 

 

 



 

 

211 

 

(d) 

 

Figure S1. Selected echograms from the five hydroacoustic surveys in Keuka Lake 

conducted in years 2000, 2007, 2011, 2017, and 2023 to estimate fish and mysid 

densities. (a, b) September 2000 SV echograms using 70kHz sounder and a color scale 

from -80 to -30 dB (EK80 colors) reflective of the high-density alewife period in 

Keuka Lake. Mysids are difficult to detect at 70 kHz. The echogram in (a) has no 

threshold, while echogram (b) has -68dB Tsu threshold applied for fish analysis. (c, d) 

October 2023 SV echograms with a color scale from -80 to -30 dB (EK80 colors) 

reflective of the low-density alewife period in Keuka Lake. Echogram (c) has no 

threshold applied, while echogram (d) has a -66 dB Tsu threshold applied for fish 

analysis.  

 

S2. Results 

Table S3: Hydroacoustic survey fish density (fish/ha) in three target groups calculated 

from transect data in Keuka Lake. Examples of individual transect data are provided in 

supplemental material. Density numbers are in number of fish per hectare for both 

layers, and the total number of fish in the lake is the sum of the upper 20 m and the 

hypolimnion densities. Size group targets are given as average fish density ± 1 SE 

(number of transects).  

Year Layer Small size 

targets 

Medium size 

targets 

Large size 

targets 

Total 

transect 

length (km) 

Fish density by 

depth 

    

2000 Upper 15 

m 

895 ± 230 (9) 750 ± 417 (9) 19.8 ± 6.5 (9) 11.9 

2000 15-30 m 166 ± 126 (9) 64.5 ± 39.8 (9) 19.4 ± 18.1 (9) 11.9 

2000 Below 30 

m 

0.1 ± 0.05 (7) 7.2 ± 2.6 (7) 13.2 ± 3.6 (7) 10.2 



 

 

212 

 

2007 Upper 20 

m 

682 ± 250 

(12) 

190 ± 59 (12) 32 ± 15 (12) 23.5 

2007 Below 20 

m 

10.8 ± 2.1 

(10) 

5.4 ± 1.0 (10) 10.5 ± 2.9 (10) 21.7 

2011 Upper 20 

m 

597 ± 128 

(10) 

422 ± 366 (10) 14.6 ± 5.4 (10) 12.0 

2011 Below 20 

m 

44.6 ± 33.1 

(10) 

36.9 ± 27.0 (10) 6.7 ± 1.5 (10) 12.0 

2016 Upper 20 

m 

56.7 ± 26.3 

(11) 

11.6 ± 4.5 (11) 9.9 ± 2.0 (11) 29.7 

2016 Below 20 

m 

7.3 ± 2.6 (11) 7.5 ± 2.6 (11) 6.4 ± 1.8 (11) 29.7 

2023 Upper 20 

m 

28.5 ± 13.4 

(15) 

17.7 ± 6.1 (15) 5.9 ± 2.3 (15) 29.7 

2023 Below 20 

m 

5.6 ± 5.8 (15) 3.9 ± 0.7 (15) 6.5 ± 0.8 (15) 29.7 

Total fish density      

2000 All depths 1094 822 52 11.9 

2007 All depths 693 195 43 23.5 

2011 All depths 642 459 21 12.0 

2016 All depths 64 19 16 29.7 

2023 All depths 34 22 12 29.7 

 

S2.1. Alewife results 

 

 

Figure S2. Otsego Lake hydroacoustic survey forage fish densities (assumed to reflect 

alewife) estimated during spring assessments.  

 



 

 

213 

 

 

Figure S3. Keuka Lake alewife size distributions from: (a) cold-water standard gang 

gill net surveys, and (b) forage gill net assessments. Young-of-year alewife were 

mostly too small to be caught in cold-water nets, except in 2007. The small size group 

in forage nets were young-of-year alewife. Large fish were likely age-1 or age-2 in 

2019, with a strong growth response after alewife collapse. We infer no age-1+ fish 

were caught in 2022.   

 

Figure S4. Otsego Lake alewife size distributions from: (a) cold-water standard gang 

gill net surveys, (b) long-term trap net surveys, and (c) hydroacoustic forage gill net 

assessments. Alewife growth rates are more difficult to interpret in Otsego Lake 

compared with Keuka Lake due to differences in the time of year sampled.   

 



 

 

214 

 

 

Figure S5. Keuka Lake alewife (a) length-at-age data from 1982-1991 (n = 235) and 

1997 and 2007 (n = 28). A portion of collected alewife in each survey were aged via 



 

 

215 

 

scale samples by NYSDEC collected in cold-water gill net surveys. Keuka Lake 

alewife survival estimated by the Chapman-Robson method. Alewife were aged using 

scales and collected from the cold-water lake trout standard gill net survey. (b) Years 

1982-1991 correspond to a period of high alewife abundance before their decline, 

whereas (c) years 1997 and 2007 correspond to the alewife decline period. Alewife 

ages are unavailable from survey years 2000 and 2003 (no alewife caught). 2007 was 

the final year alewife were caught in the cold-water gill net survey in Keuka Lake. 

Age data are not available for alewife in Otsego Lake. 

 



 

 

216 

 

 

Figure S6. Alewife catch per unit effort (CPUE; number of fish per net) from Keuka 

Lake forage net assessment surveys with vertical, surface, and bottom set gill nets to 

target the prey fish community. Catch proportions are shown for the periods of (a) pre-

alewife collapse 2000-2011 and (b) post-alewife collapse 2016-2022. Note, only four 

total alewife were caught in 2016 (n = 1 in surface nets, n = 3 in bottom nets). (c) 

Alewife catch from vertical gill net surveys in Otsego Lake, shown as a relative 

proportion (%) by gill net depth (m) with three nets set separately at different depths 

either from the surface downward or from the bottom upward.  



 

 

217 

 

S2.2. Water quality results 

 

 

Figure S7. Hypolimnion water quality plots for Keuka Lake (blue points) and Otsego 

Lake (yellow points). (a, b) nitrate (mg N/L) concentrations (GAM, Keuka Lake edf = 

2.9, deviance explained = 48.1%, p < 0.01, Otsego Lake edf = 2.3, deviance explained 

= 31.5%, p = 0.04), (c, d) total phosphorus (μg P/L) concentrations (GAM, Keuka 

Lake edf = 2.4, deviance explained = 29.3%, p = 0.02, Otsego Lake edf = 1.0, 

deviance explained = 18.8%, p = 0.02), and (e) chlorophyll a (μg/L) concentration 

(GAM, Keuka Lake edf = 2.9, deviance explained = 72%, p < 0.01). Red horizontal 



 

 

218 

 

lines are the long-term mean concentrations, and the vertical lines indicate arrival of 

zebra mussels (dotted) and quagga mussels (dashed). In Keuka Lake, hypolimnion 

samples are collected at ‘deep’ sites at 30 m depth. In Otsego Lake, hypolimnion 

samples are collected during whole-water column surveys at one sampling station, and 

for our analysis, designated at > 12 m depths. Chlorophyll a in Otsego Lake is an 

integrated sample and therefore cannot be separated by depth. 

 

S2.3. Zooplankton analysis results 

 

Figure S8. Monthly (June-September) mollusc veligers and rotifer biomass from 

Otsego Lake. The shaded light gray area is the period of alewife decline (2003-2007), 

and dark gray is alewife collapse (2008-2010). The vertical dotted line indicates the 

arrival of zebra mussels and vertical dashed line indicates the arrival of quagga 

mussels. The trendline indicates a general decrease in rotifer counts over time.  

 

S2.4. Mysid results 

 

Description of Keuka Lake mysid netting and acoustics results: 

The comparison of 10 net tows with acoustic data collected at the same location and 

within one to two hours of the net tows resulted in an average TS of -87.94 dB (1 SE 

0.59 dB, N = 10) for a single mysid and -65.32 dB (0.40 dB, N = 10) per g mysid dry 



 

 

219 

 

mass. The mysid lengths averaged 12.9 mm (standard length) with a range of 10.9 to 

16.1 mm among the 10 net tows. The TS estimates were significantly related to 

average mysid length (r2 = 0.43, N = 10) and ranged from -91.5 to -85.6 dB.  Because 

of the large range in mysid length, the relationship between total acoustic area 

backscattering (Sa in dB) and log10 transformed mysid density was poor (r2 = 0.06, p 

= 0.58, N = 10).  Acoustic backscattering is a better indicator of mass than density, and 

the relationship between log10 transformed mysid mass and Sa was substantially 

stronger (r2 = 0.576, p < 0.01, N = 10). These estimates were close to expectations 

from the theoretical scattering model of Staton and Chu as modified by Holda et al. 

2025, Table S3). We used these average TS estimates to calculate mysid density and 

dry mass in years with acoustic data but no tow data (2007, 2011, 2017).   

 

Table S4: Estimated mysid density and biomass in Keuka Lake, New York, USA using 

hydroacoustics. All estimates are based on a fish exclusion threshold of -60dB (see 

Rudstam et al. 2008a) and an average target strength per mysid of -87.7dB and per 

gram dry weight (g dw) of -65.7dB. Numbers in parenthesis are 1 SE and number of 

transects). Note, net tows are collected in the deepest part of each transect with density 

given as number of mysids / m2 ± 1 SE (number of sampling sites). Acoustic surveys 

are whole transects with density given as mean number of mysids / m2 ± 1 SE (number 

of transects). Biomass estimates are also provided as grams dry weight / m2.  

Year Survey 

length (km) 

Acoustic-

based density  

Acoustic-based 

biomass 

Net-based 

density  

Net-based 

biomass 

2007 22.8 108 ± 16 (11) 0.590 ± 0.089 

(11) 

NA NA 

2011 12.0 95 ± 19 (10) 0.520 ± 0.105 

(10) 

NA NA 

2016 29.7 124 ± 21 (11) 0.678 ± 0.116 

(11) 

NA NA 

2023 29.7 129 ± 14 (15) 0.704 ± 0.077 

(15) 

186 ± 16 

(10) 

1.10 ± 0.11 (10) 

  



 

 

220 

 

S3. Discussion  

 

 

Figure S9. Proportion of lake trout diet with each prey item present collected from 

Keuka Lake, New York, USA. Diet samples for lake trout with (a) total length (TL) < 

350mm, and (b) total length (TL) ≥ 350mm. Bars are color-coded by scenario: pre-

alewife collapse 1997-2011 (blue) or post-alewife collapse 2016-2022 (red). Lake 

trout diet samples were not available for Otsego Lake, New York, USA. (c) Mean 

mass-age-age (g) for adult lake trout in Keuka Lake.   



 

 

221 

 

 

Figure S10. Fish species composition results from 2019 and 2022 lake-wide forage 

net assessment surveys in Keuka Lake, New York, USA. Plots are color-coded by net 

location surface (green) or bottom (blue), and include the species alewife (Alosa 

psuedoharengus), golden shiner (Notemigonus crysoleucas), spottail shiner 

(Hudsonius hudsonius), yellow perch (Perca flavescens), pumpkinseed (Lepomis 

gibbosus), bluegill (Lepomis macrochirus), smallmouth bass (Micropterus dolomieu), 

Cyprinidae (minnow spp.), yellow bullhead (Ameiurus natalis), Atlantic salmon 

(Salmo salar), brook silverside (Labidesthes sicculus), brown bullhead (Ameiurus 

nebulosus), cisco (Coregonus artedi), lake trout (Salvelinus namaycush), rainbow 

trout (Oncorhynchus mykiss), rock bass (Ambloplites rupestris), black crappie 

(Pomoxis nigromaculatus), satinfin shiner (Cyprinella analostana), walleye (Sander 

vitreus), and slimy sculpin (Cottus cognatus). 



 

 

222 

 

 

Figure S11. Fish community changes in Otsego Lake, New York, USA. In the trap net 

survey, (a) relative proportion of catch and (b) catch rate of yellow perch. In the percid 

netting survey, (c) relative proportion of catch and (d) catch rate of yellow perch. The 

dashed vertical lines indicate the period of walleye stocking, while the light gray box 

indicates the period of alewife decline and the dark gray box indicates alewife 

collapse.   

 



 

 

223 

 

 

Figure S12. Time series of Keuka Lake prey fish composition in the cold-water 

standard gang gill net survey (1979-2022). This survey was designed to target lake 

trout; however, cold-water native cisco were last caught in 1994. Introduced alewife 

and rainbow smelt were last caught in this survey in 2007.  

 

Table S5: Metadata for summer 2024 epilimnetic and whole-water column vertical 

tow sampling conducted by NYS Department of Environmental Conservation. Paired 

samples were collected at the confluence of the three arms of Keuka Lake, New York, 

USA. Site depth was 53.3m for each sampling event. 

 

 



 

 

224 

 

 

Figure S13. Keuka Lake zooplankton. (a) Proportion of zooplankton species-group 

biomass below the thermocline. Whole water column tows were collected and 

compared with epilimnetic tows in summer 2024. (b) Corresponding temperature and 

dissolved oxygen vertical profiles. 

 

References 

1. Austerman, P. and B. Hammers. 2015. A comparison of lake trout catch in 

monofilament and multifilament gill nets in Western Finger Lakes. New York State 

Department of Environmental Conservation, Avon, NY. 12 pp. 

2. Boscarino, B.T., Rudstam, L.G., Minson, M.A. and Freund, E.E., 2010. 

Laboratory-derived light and temperature preferences of juvenile mysid shrimp, Mysis 

diluviana. Journal of Great Lakes Research, 36(4), pp.699-706. 

3. CSLAP Sampling Protocol. 2024. New York State Citizens Statewide Lake 

Assessment Program. Available from https://nysfola.org/cslap-protocol-forms/.  



 

 

225 

 

4. Cornwell, M.D. 2005. Re-introduction of walleye to Otsego Lake: Re-

establishing a fishery and subsequent influences of a top predator. Occas. Pa. #40. 

SUNY Oneonta Biological Field Station, SUNY Oneonta. 

5. Foster, J.R. 1990. Introduction of the alewife (Alosa pseudoharengus) into 

Otsego Lake. In 22nd Ann. Rept. (1989). SUNY Oneonta Biol. Fld. Sta., SUNY 

Oneonta. 

6. Hammers, B.E., P. Austerman and D. H. Kosowski. 2014. Summary of 

salmonine monitoring in Keuka Lake, 1997-2009. New York State Department of 

Environmental Conservation, Avon, NY. 48 pp. 

7. Hammers, B.E. 2018. Keuka Lake Salmonine Management Assessment, 2010-

2016 Update. New York State Department of Environmental Conservation, Avon, NY. 

50 pp.  

8. Forney, J.L., L.G. Rudstam, and D.L.Stang. 1994. Percid sampling manual. 

Cornell Biological Field Station and New York State Department of Environmental 

Conservation, New York Federal Aid Study VII, Albany, New York 

9. Rudstam, L. G., Hetherington, A. L., & Mohammadian, A. M. (1999). Effect of 

temperature on feeding and survival of Mysis relicta. Journal of Great Lakes Research, 

25(2), 363-371. 

10. Rudstam, L. G., T. Schaner, G. Gal, B. T. Boscarino, R. O'Gorman, D. M. 

Warner, O. E. Johannsson, and K. Bowen. 2008. Hydroacoustic measures of Mysis 

relicta abundance and distribution in Lake Ontario. Aquatic Ecosystem Health and 

Management 11:355-367. 

11. Rudstam, L. G., F. R. Knudsen, H. Balk, G. Gal, B. T. Boscarino, and T. 

Axenrot. 2008. Acoustic characterization of Mysis relicta at multiple frequencies. 

Canadian Journal of Fisheries and Aquatic Sciences 68:2769-2779. 

12. Warner, D. M., Rudstam, L. G., and Klumb, R. A. (2002). In situ target 

strength of alewives in freshwater. Transactions of the American Fisheries Society 

131:212-223.  



 

 

226 

 

B CHAPTER 2 APPENDIX 

Appendix for Chapter 2: Whole-lake acoustic telemetry to evaluate survival of 

juvenile fish 

 

S1. Methods 

 

S1.1. Acoustic-tagging and hatchery practices 

S1.1.1. Hatchery rearing 

Staff from U.S. Geological Survey (USGS) and New York State Department of 

Environmental Conservation (NYSDEC) collected cisco (Coregonus artedi) brood 

stock annually from Lake Ontario in Chaumont Bay, New York, USA during the 

spawning season (late November to early December) and transported 200-300+ adult 

specimens to USGS Great Lakes Science Center, Tunison Lab of Aquatic Science 

(USGS-Tunison) in Cortland, New York, USA for hatchery propagation. Adults were 

spawned over the next one to three days and eggs were transferred to incubation tubes 

to develop. In total, approximately 300,000-1,000,000 total eggs were collected and 

reared annually at regional Finger Lakes hatcheries including NYSDEC Bath and 

Oneida New York State Fish Hatcheries (Bath and Oneida Hatcheries) and USGS-

Tunison.  

 S1.1.2. Tagging surgeries 

Surgeries to implant acoustic telemetry transmitters (hereafter, ‘tags’) were conducted 

by staff at USGS-Tunison and followed protocol by McKenna et al. [1]. Age-0 cisco 

were surgically implanted with small acoustic tags (0.6g Lotek model L-AMT-5.1B; 

Lotek, Newmarket, Ontario, Canada). Age-1 cisco were implanted with larger acoustic 

tags (3.5g Lotek model L-AMT-8.2). Full tag specifications are available here: 



 

 

227 

 

https://www.lotek.com/wp-content/uploads/2017/10/JSATS-AMT-Series-Spec-

Sheet.pdf). For all tagged juvenile cisco released in this study, fish were placed on the 

surgery table and anesthetized with MS-222 solution over the gills via a peristaltic 

pump (Cole-Palmer Masterflex L/S; 25 mL/minute rate). Physical tags were inserted 

through a small incision in the abdomen, anterior to the left pelvic fin. Wounds were 

closed with a single surgeon’s knot using blue monofilament 4-0 suture (Matrix 

Wizard LLC). Mean surgery time for 0.6g tags in the McKenna et al. (2021) study was 

100.3s (Standard Error, SE, 2.56). For tagged fish released into Keuka Lake, Relative 

Tag Weight (RTW) for age-0 fish (0.6g tag) was mean = 5.6% (range 3.8-10.9%) and 

for age-1 fish (3.5g tag) was mean = 5.8% (range 2.2-10.1%). Further details on 

surgery procedures are available in McKenna et al. [1].  

 S1.1.3. Transport and stocking 

All juvenile cisco (tagged and untagged) were transported to Keuka Lake, New York, 

USA for release in specialized fish transport trucks with six insulated transport tanks 

(275-gallon capacity) in each truck equipped with aeration and oxygen diffusion. Each 

insulated tank could hold up to approximately 7,000 fish per tank and total transport 

time was approximately 1.5 hours from hatchery to the release site. In total, 296, 979 

juvenile cisco were stocked from October 2019 to October 2020 (including 210 tagged 

fish), with the highest number of fish stocked in fall, e.g., 92,225 age-0 fish stocked on 

9-Oct-2019 and 204,466 age-0 fish on 15-Oct-2020). Once on site, fish were directly 

transferred via 20cm stocking hose from the stocking truck into insulated stocking 

tanks (also 275-gallon capacity) positioned on a pontoon boat. One tank from the 

stocking truck was divided between the two stocking tanks on the pontoon boat. 



 

 

228 

 

Aeration was provided during the five-minute boat trip to the stocking site. Fish were 

released via a 20cm standpipe that was removed from the bottom of the stocking tank 

allowing the fish to be discharged directly though the bottom of the boat into the 

water. No hose was needed for this operation. All fish (tagged and untagged) were 

stocked above the receiver location coordinates (within 50m at the surface) in the 

northwestern corner of Keuka Lake. Fish were released manually from the boat using 

this process to help reduce transport stress as fish are not manually netted out of the 

tanks and fish can be released volitionally.  

S1.2. Acoustic receiver array 

Acoustic telemetry receivers (Lotek WHS 4250 Series) were deployed at a whole-lake 

scale throughout the duration of this study (full receiver specifications available here: 

https://www.lotek.com/wp-content/uploads/2017/10/WHS-4250-Spec-Sheet.pdf). 

Receivers were deployed at a lake-wide scale, spaced apart at a mean distance of 

2,968.4m (range 1,353m – 6,087m). Range testing was conducted to place receivers 

that would maximize their detection range, specifically by utilizing three receivers 

staged at pinch points within each of the lake arms. Receivers at these points were 

spaced at a mean distance of 240.5m apart (range 184m – 386m). The average 

sampling depth (range) for acoustic receivers by branch was West = 50m (44-57m), 

Confluence = 35m, South = 43m (32-52m), and East = 30m (16-37m). 

Acoustic receiver range testing (receiver efficiency) was conducted on several days in 

September and October, prior to the release of tagged fish. The stated manufacturer 

range for receivers used in this study was 50-250m. We evaluated efficiency for 

conditions in this study by placing three receivers on the lake bottom with 43 Juvenile 

https://www.lotek.com/wp-content/uploads/2017/10/WHS-4250-Spec-Sheet.pdf


 

 

229 

 

Salmon Acoustic Telemetry System (JSATS) transmitters in mesh bags deployed at 

equal 25m intervals over distances 25-800m from receiver locations. An additional 

test was conducted with one receiver and two tags deployed at incremental distances 

out to 300m. Tags were programmed on a 20-second transmission rate (the same rate 

as used for tagged fish in this study) tied to a moored buoy at the surface at fixed 

distances and left to transmit for 30 minutes. All gear was picked up and the data were 

analyzed, whereby we found a receiver efficiency of approximately 200-300m (linear 

diameter). We found all tags were detected within this range, with an immediate lack 

of detections for tags positioned further than these distances (i.e., a distance threshold 

for tag detections).  

We expected that receiver efficiency would increase at greater depths (closer to the 

receiver), with surface detections being the farthest extent of each receiver’s coverage. 

Further, the water column mixes from late fall through early summer (e.g., no 

stratification), so we do not anticipate detection efficiency would be limited by depth 

during this period [2]. Despite stratification with differences in water temperature and 

density (e.g., the summer thermocline) that may cause detection deflection or 

interference, we expected cisco to primarily inhabit the cooler waters below the 

thermocline during this period.      



 

 

230 

 

 
Figure S1. (Left) Example of an acoustic receiver mooring (using Lotek-brand 

Wireless Hydrophone System (WHS) 4250 Series Juvenile Salmon Acoustic 

Telemetry System (JSATS) Acoustic Autonomous Receivers/Dataloggers. The 

acoustic receiver is anchored approximately one meter from the bottom of the cement 

block and is facing upward to detect acoustic tagged cisco (Coregonus artedi) 

swimming in the water column. Each mooring also contained a cable anchor for 

retrieval and surface buoy for safety and identification. All acoustic receivers across 

the array were deployed with this configuration across Keuka Lake, New York, USA. 

Photo by D. Mulhall. (Right) Location of acoustic receivers placed in the Keuka Lake 

outlet (receiver depth = 1.8m) and nearby Seneca Lake (receiver depth = 1.8m) to 

detect potential emigration of tagged juvenile cisco. The outlet flows from the 

northeastern branch of Keuka Lake eastward into Seneca Lake. No tagged cisco were 



 

 

231 

 

detected on either receiver for the duration of this study. Prior to this study, we did not 

expect cisco to utilize this habitat as the Keuka Lake outlet is shallow and contains a 

natural waterfall barrier midway that likely prohibits fish emigration to Seneca Lake.  

 

S1.3. Detection filtering and assumptions 

S1.3.1. Detection criteria 

All tag detection filtering was conducted using statistical software (program R; R Core 

Team, 2022) with output datafiles from Lotek WHS Host software (V1.8.3942.3) that 

were filtered to only include entries that corresponded to unique fish tag 

identifications. In the software program (program R), we took the following steps to 

validate true tag detections: 

1)  Calculate the time (t) between consecutive tag detections for each fish by 

applying the ‘min_lag’ function from the ‘GLATOS’ R package [3].  

2) False detections were then omitted from the acoustic telemetry dataset based 

on the following criteria: 

a. All detections must occur after time of stocking and before the 

minimum (80%) expected battery lifespan stated by the manufacturer 

(Lotek-brand): 0.6g tag = 262 days and 3.5g tag = 1,218 days. 

i. Maximum tag detections were observed on the acoustic receiver 

array for age-0 fish (0.6g tags) as 151 days (n = 1 tagged fish; 

57.6% of minimum expected battery life), and for age-1 fish 

(3.5g tags) as 403 days (n = 1 tagged fish; 33.1% of minimum 

expected battery life).  

b. A specific function (the ‘false_detections’ function from GLATOS) was 

used to specify consecutive detections from ‘min_lag’ must occur 



 

 

232 

 

within 10 minutes, or 30 times the nominal rate of programmed 

transmitter [3, 4], e.g., 30*20-second nominal delay = 600s. 

c. Final true detections must include ≥ two consecutive tag detections on 

the same receiver within a four-minute window. 

We found that most filtered detections that met this criterion occurred at 20s or 40s 

intervals (within ± 2s). Observed tag detections that occurred < 18s and between 22s 

and 38s were considered false detections and were therefore removed. All remaining 

detections ≤ 10 minutes were retained using the ‘false_detections’ function. We 

compared all final detections from our detection filtering criteria in R to manual 

inspection. Manual inspection of individual tag detection histories included visual 

examination of abacus plots using the ‘abacus_plot’ in GLATOS compared with the 

individual filtered detection files. Abacus plots show individual tagged fish movement 

at receiver sites over time, thus providing visual insight to biological feasibility of 

detections across the receiver array [3].  

Once we obtained our final detection dataset, we then applied the following criteria to 

infer initial fish mortality:  

1) Undetected: Tagged fish that went undetected (e.g., no positive detections at 

any receiver after stocking) were inferred as straight-to-death, or mortality upon 

release, events (t = 0 days).  

2) Detected < 1d: All fish with positive detections exclusively within the first day 

(e.g., 0 < t < 1 day post-release) had detections that only occurred < 2 hours from their 

first detection post-stocking. Because fish were stocked in stages, precise release times 

for individual fish were not recorded at stocking. Therefore, same day detections were 



 

 

233 

 

calculated as a fraction of a day from first to final raw detection in the number of 

minutes (fraction of t = 1 day) in our final detection dataset.  

The final tag detection (age-1 fish released in July 2020) occurred on our acoustic 

receiver array on 14 August 2021, with an average time between detections as 3.3 

days. Our acoustic receiver array was deployed through 2021 with no tagged cisco 

detections occurring between the final tag detection and December 2021. Hence, we 

assumed that no tagged fish were alive in Keuka Lake in our study after 14 August 

2021.   

 

S1.3.2. ‘Detection events’ and ‘detection interval’ 

Given tagged fish could still be at large after their final observed detection on the 

receiver array, we inferred final survival times by adding a detection interval to raw 

final detection times. Average time between detection events was calculated using a 

software package (from the ‘detection_events’ function in GLATOS). To reduce 

single detections in our final detection datafiles into detection events, we used the 

criteria: ≥ four consecutive tag hits within one hour. We then conducted a sensitivity 

analysis to compare no detection interval (zero interval; raw time), half the average 

time between detections (half interval), and the full average time between detections 

(one interval). We selected a half interval and added this calculated time to the final 

raw true detections, rounded to the nearest whole day.  



 

 

234 

 

 

Figure S2. Histogram for calculated average time (days) between detection events for 

all release cohorts of tagged cisco (Coregonus artedi) into Keuka Lake, New York, 

USA (Fall 2019, Summer 2020, and Fall 2020). Detection interval summary statistics: 

min=0 days, median=1.52 days, mean=1.92 days, max=10.4 days. 

 

S2. Results 

 

S2.1. Supporting figures for survival analysis 

 We examined residual plots for the three most general models in our Cox 

proportional hazards model set: (1) Survival ~ Year + Length + Length:Age, (2) 

Survival ~ Year + Mass + Mass:Age, and (3) Survival ~ Year + Age + Condition + 

Condition*Age. We tested hazard assumptions for proportionality based on Hosmer et 

al. [5] and Moore [6]. 



 

 

235 

 

 
Figure S3. Martingale residuals of a null Cox proportional hazards model plotted 

against each covariate: Year, Age-at-release, Length, Mass, and Condition factor [6] 

for tagged cisco (Coregonus artedi) stocked into Keuka Lake, NY, USA from 2018 to 

2020. 

 

 
Figure S4. Schoenfeld residuals for subject-level size and condition covariates, a) 

Length, b) Mass, and c) Condition for tagged cisco (Coregonus artedi) stocked into 

Keuka Lake, NY, USA from 2018 to 2020. 

 



 

 

236 

 

 
Figure S5. Deviance residuals for each of the most general Cox proportional hazard 

models: a) Survival ~ Year + Length + Length:Age, b) Survival ~ Year + Mass + 

Mass:Age, and c) Survival ~ Year + Age + Condition + Condition*Age for tagged 

cisco (Coregonus artedi) stocked into Keuka Lake, NY, USA from 2018 to 2020. 

 

S3. Discussion 

 

S3.1. Evidence of predation 

We observed two instances of tagged cisco that were preyed upon by lake trout 

(Salvelinus namaycush), the top piscivorous predators in Keuka Lake. Local anglers 

caught lake trout with depredated tagged and untagged cisco, with an acoustic 

transmitter in a lake trout caught in October 2019 and a transmitter in a trout caught in 

July 2020. The first lake trout was caught two days after fall stocking, with an acoustic 

transmitter corresponding to an age-0 cisco stocked on 9 October 2019, and detected 

on the acoustic receiver closest to the stocking site on 10 October 2019. An age-1 

cisco was partially digested in a lake trout caught within a week from the 7 July 2020 

stocking, with no confirmed transmitter detections on any acoustic receiver in the lake. 

Both fish had multiple fish in their stomach contents, with a proportion positively 

identified by project biologists as juvenile cisco. Additionally, a lake trout caught by a 

local angler on 17 October 2020 (two days post-stocking) contained ≥ 20 juvenile 

cisco in the stomach contents (no acoustic tags were confirmed inside this trout).  

 



 

 

237 

 

 

Figure S6. Angler caught lake trout (Salvelinus namaycush) on 17 October 2020, two 

days after fall stocking, with several cisco (Coregonus artedi) inside the stomach 

sample (photo provided to M. Chalupnicki from an anonymous angler).  

 

References 

1. McKenna Jr, J. E., Sethi, S. A., Scholten, G. M., Kraus, J. & M. Chalupnicki. 

Acoustic tag retention and tagging mortality of juvenile cisco Coregonus artedi. J. 

Great Lakes Res. 47, 937-942 (2021). 

2. Kuai, Y. et al. Strong thermal stratification reduces detection efficiency and 

range of acoustic telemetry in a large freshwater lake. Anim. Biotelemetry. 9, 46 

(2021). 

3. Holbrook, C. et al. glatos: A Package for the Great Lakes Acoustic Telemetry 

Observation System. (2019). <https://gitlab.oceantrack.org/GreatLakes/glatos>.  

4. Pincock, D. G. and S.V. Johnston. Acoustic telemetry overview. Telemetry 

techniques: a user guide for fisheries research, 305-337. (2012). 

5. Hosmer, D. W., Lemeshow, S. & S. May. Applied Survival Analysis: 

Regression Modeling of Time-to-event Data. Wiley-Interscience 2nd ed. (2008).  

6. Moore, D.F. Applied survival analysis using R. Springer International 

Publishing. (2016). 

  



 

 

238 

 

C CHAPTER 3 APPENDIX 

Appendix for Chapter 3: How accurately does eDNA reflect the spatial distribution 

of cold‐water fish? Field validation from a temperate lake 

 

S1. Methods 

 

S1.1 Survey design 

 

Table S1: Cohorts of juvenile cisco Coregonus artedi released into Keuka Lake, New 

York, USA prior to eDNA surveys (July and October 2020). No acoustic-tagged cisco 

stocked from 2018-2019 were detected on the acoustic telemetry array after June 

2020, thus we inferred that few fish survived up to the July 2020 eDNA survey. All 

cisco were released at the same stocking location in the northwestern arm of Keuka 

Lake (Figure 1; Figure 2). All juvenile cisco stocked in this study were reared in 

regional hatcheries from adult cisco broodstock collected annually in December from 

Chaumont Bay, Lake Ontario, New York, USA. 

Stocking date Age-at-release 

(months) 

Total no. 

fish stocked 

No. tagged 

fish 

17-Oct-2018 10 98,789 40 

19-Jun-2019 18 1,360 22 

9-Oct-2019 10 92,165 60 

9-Oct-2019 22 28 28 

 



 

 

239 

 

 

Figure S1. Temperate profiles (a and b) and dissolved oxygen profiles (c and d) for 

Keuka Lake, New York, USA during the eDNA surveys in July 2020 (low abundance, 

pre-stocking) and October 2020 (high abundance, post-stocking). Profiles were 

measured at three sites in the lake in the west, south, and confluence regions. The 

dashed lines indicate the 12 m and 18 m eDNA sampling depths in July and October.  

 

  



 

 

240 

 

S1.2 eDNA methods 

 

Field collections  

 Our eDNA survey protocol included control samples to test for eDNA 

contamination as follows. Field negative control: clean, empty bottles that were 

opened and filled with distilled water in the field. Once filled, bottles continued 

through the sampling handling process alongside all other field samples to monitor the 

field collection process for contamination. Negative water sampling control: clean, 

empty bottles that were filled with distilled water at the filtration station. Once filled, 

bottles were handled and water filtered using the same process as all field samples. 

These bottles did not travel on boats into the field and were meant to monitor the 

filtration process for contamination. Bottle negative control: clean bottles filled with 

distilled water prior to going to the field. These bottles were transported alongside 

other field samples, but the bottle was never opened in the field. Contents were filtered 

as normal at the filtration station.  These bottles were meant as a control to monitor 

contamination in the bottles and to ensure they were effectively cleaned prior to use, 

i.e. a ‘batch control’ for the cleaning process. 

Van Dorn water samplers were used to collect 2 L of water at each shallow (12 m) or 

deep (18 m) sampling location with standard protocol to avoid sample contamination. 

Separate water samplers were used between shallow and deep sites to allow time to 

decontaminate samplers between sampling depths. Each boat had two disinfecting 

totes, one filled with 10% bleach solution to disinfect Van Dorn bottles and lines, the 

other filled with dilute (one tablespoon of dish detergent per tote) soap to rinse 



 

 

241 

 

disinfected water samplers prior to deployment. Van Dorn samplers were disinfected 

prior to each deployment, typically during transit times between sites, but at a 

minimum of 10 minutes in the 10% bleach tote was required followed by rinsing in the 

dilute-soap tote prior to deploying a sampler to collect water. A spray bottle was used 

with 10% bleach solution to spray working surfaces used to stage and collect water 

samples upon arrival to a site. One crew member focused on water sample collection 

to minimize the number of staff handling water samples. Personnel used a new pair of 

rubber gloves for each water collection effort (labeling of the sample jar, deployment 

of the water sampler, and decanting the sampler bottle to the sample jar).  

All environmental samples were filtered through a 47mm glass fiber filter (1.5 µm 

pore size) at filtering stations along the shoreline. Most of the water samples were 

filtered at 2L with no filters clogged during processing. Therefore, none of the jars 

required being passed through multiple filters and all samples were filtered with just 

one filter. In total, five samples (of the 263 total, controls and environmental samples 

combined) were collected at less than 2L, around 1.5 – 1.6 L, with one sample only 

collecting 0.65 L due to heavy siltation from the bottom of a shallow sampling 

location. Note, n = 4 water samples less than 2 L were collected from the pre-stocking 

July eDNA survey for which we expected low cisco abundance. Only n = 1 sample 

was filtered less than 2 L (1.2 L) during the post-stocking October survey and was 

collected at the southern end of the South arm, where no tagged cisco or positive 

eDNA detections occurred. Overall, instances of partial filtration were rare with 

largely clear waters in Keuka Lake, and thus we assumed impacts from water samples 

collected with < 2 L would be negligible for the overall study. Once filtered, 



 

 

242 

 

individual filters were stored in separate labeled vials and then all filter sample vials 

were placed in a cooler with ice, temporarily stored at -20˚C, then transported on ice 

packs to U.S. Fish and Wildlife Service (USFWS) Northeast Fishery Center (NEFC) 

in Lamar, Pennsylvania, USA, where they were stored at -80˚C until DNA extractions 

were performed. 

 

DNA extractions 

All DNA extractions, marker development, and sample processing for this study were 

conducted following protocol by USFWS NEFC. In total, all 263 filter samples from 

July and October 2020 were included in the subsequent analyses. Filter samples were 

first extracted using the Qiagen DNeasy Blood and Tissue Kit (Qiagen Corporation, 

Valencia, CA) using a modified extraction protocol following that of the USFWS 

eDNA Quality Assurance Project Plan for filter samples (USFWS, 2020). Samples 

were eluted with 200 µl of Buffer AE. All filter extractions were conducted in a 

dedicated DNA extraction room with mechanical controls and laboratory hoods to 

maintain a clean, contamination-free work environment. One negative extraction 

control included only the digestion buffer, included to check for reagent-based 

contamination in each batch. The second negative control was reagents with a dry 

sterile filter. The single positive extraction control was a sterile filter with a 

standardized amount of an American eel Anguilla rostrata DNA sample extracted 

from a fin clip. The positive extraction control was included to verify each extraction 

batch was carried out with a similar extraction efficiency. All DNA extracts were 

stored at -20°C until quantitative PCR analysis. 



 

 

243 

 

 

Marker development and validation 

USFWS NEFC conducted extensive validation before use in environmental samples, 

including a standardized set of specificity and sensitivity tests. Mitochondrial 

sequence alignments of target species as well as non-target species (other Coregonine 

species and non-target species known to exist in Keuka Lake based on a list of other 

species present were carried out using Geneious (version 11.1, Biomatters Inc., San 

Diego, CA) to identify appropriate areas for primer and probe placement. Conserved 

areas of select mitochondrial genes were identified within cisco and bloater Coregonus 

hoyi (not present in Keuka Lake) sequences while at the same time offering sequence 

divergence with all other non-target species. This initial in silico marker development 

work identified four appropriate areas (ND1, ND4, ND5, and cytB) for primer and 

probe placement.  

The four potential markers were then tested during an in vitro phase of the validation 

protocol. In vitro testing included: 1) temperature optimization, 2) primer and probe 

concentration optimization, and 3) marker specificity testing. Temperature, primer, 

and probe concentration optimization were designed to ensure the markers operate at 

the highest PCR efficiency and sensitivity possible. Marker specificity testing was 

conducted to ensure that the markers designed during in silico development are 

specific to the intended target species by using tissue-derived DNA samples obtained 

from both the target species (cisco and bloater) and non-target species. During 

specificity testing, tissue-derived DNA from multiple individuals of cisco and bloater 



 

 

244 

 

(including cisco obtained from the same genetic source as those stocked in Keuka 

Lake) and multiple individuals of non-target species sourced from watersheds near 

Keuka Lake and other locations was analyzed to test for amplification with the four 

potential markers (Table S2). During PCR, 10 picograms of all target and non-target 

DNA samples were tested. Sensitivity testing was conducted to compare the relative 

sensitivity of each marker and ensure they were operating at high efficiency for 

analysis of environmental samples. Two of the initial four markers were dropped from 

further evaluation after the initial in vitro testing. 

 Sensitivity testing was conducted on the remaining two markers (COSP_ND401 and 

COSP_cytB01) by diluting cisco gBlock standard (synthetic DNA) in a 5x dilution 

series with the following test levels; 31,250 copies, 6,250 copies, 1,250 copies, 250 

copies, 50 copies, 10 copies, 2 copies, and 0.4 copies. Standard curve analysis (Figure 

S2) was run in a replicate of 10 and the resulting data provided several useful metrics 

to determine marker sensitivity, PCR efficiency, the slope and y-intercept of each 

standard curve generated, as well as amplification success at very low target DNA 

concentration. 

 

qPCR analysis 

Reaction conditions for the COSP_ND401 qPCR analysis were: 500 nanomolar (nM) 

primer, 250 nM qPCR probe, and 1x concentration of TaqMan® Environmental 

Master Mix 2.0 (Applied Biosystems™, Waltham, MA). Reaction conditions for 

COSP_cytB01 were the same as COSP_ND401 except the reverse primer 



 

 

245 

 

concentration was 900 nM. Reaction conditions for AME1 (marker used only for 

positive extraction control analysis) were 500 nM each primer, 125 nM qPCR probe, 

and 1x concentration of TaqMan® Environmental Master Mix 2.0. Each probe 

(Applied Biosystems™, Waltham, MA). AME1 was a TaqMan® MGB (minor groove 

binder) quenched with a non-fluorescent quencher on the 3’ end and labeled on the 5’ 

end with either a 6-FAM or VIC fluorophore. qPCR reactions were run in 20 µl 

volumes and included 17 µl of master mix/primer/probe mixture and 3 µl of DNA 

template. All qPCR reactions were analyzed on a QuantStudio 7 Pro Thermalcycler 

(Thermo Fisher Scientific, Waltham, MA). Cycling conditions were based on the 

manufacturer’s recommendations and carried out for 45 cycles.  

Two synthetic double-stranded gBlock® DNA standards (Integrated DNA 

Technologies, Coralville, IA) were used to serve as positive qPCR control material 

during amplification of environmental and control samples: one for the two cisco 

markers and the second for the American eel assay (AME1) that was used to analyze 

only the positive extraction controls. Finally, fin-clip origin DNA from cisco, bloater, 

and American eel was included as additional qPCR positive control material for each 

marker analyzed.   

For each individual environmental and field control sample collected, eight PCR 

replicates were performed for each marker. The negative and positive DNA extraction 

controls were also assayed in eight replicates per marker. The qPCR positive controls 

were run in duplicate. Finally, negative qPCR controls were also included on each 

plate in quadruplicate per marker. Following PCR, Ct (cycle threshold or the cycle at 



 

 

246 

 

which fluorescence signal rises above the background fluorescence level of a given 

qPCR assay) values were calculated. Each plate was checked for absence of 

amplification in negative control reactions and positive amplification in positive 

controls. Any environmental sample that resulted in Ct values < 45 was considered 

positive for cisco DNA. Cisco qPCR marker specifications as well as limit of 

detection values and limit of quantification values following Klymus et al. (2020) and 

Lesperance et al. (2021) are reported in the main text Tables 2 and 3.  

 

Inhibition tests 

PCR inhibition was tested by running triplicate PCR reactions for all DNA extracts 

using the TaqMan® Exogenous IPC (Internal Positive Control) Reagents Kit (Applied 

Biosystems™, Waltham, MA) following the manufacturer’s recommendations. qPCR 

IPC reactions were run in 20ul volumes and included 17ul of master mix/primer/probe 

mixture and 3ul of DNA template. All qPCR reactions were analyzed on an 

QuantStudio 7 Pro PCR thermal cycler (Applied Biosystems™, Waltham, MA). 

Cycling conditions were based on manufacturer recommendations and carried out for 

40 cycles. Reactions were considered inhibited if Ct values were delayed by cycled 

threshold values of >1.   

  



 

 

247 

 

Table S2: Species, sample locations, and PCR results for the specificity testing during 

marker validation for cisco Coregonus artedi markers COSP_ND401 and 

COSP_cytB01 developed for this study. Ten picograms of each DNA sample were 

added to an individual well of a PCR plate. If a sample positively amplified with each 

marker a “+” is indicated and the corresponding cycle threshold value (Ct) is provided 

in parentheses. Note, only positive samples are included below.  

 

  



 

 

248 

 

 

Figure S2. Standard curves for the two markers (top: COSP_ND401, and bottom: 

COSP_cytB01) selected for detection of cisco Coregonus artedi eDNA from Keuka 

Lake, New York, USA. Sensitivity metrics are listed below the standard curve for 

each marker. Standard curves were generated by serially diluting gBlock standard in a 

5x dilution series over eight orders of magnitude between 31,250 copies and 0.4 

copies. Each concentration was replicated 10x. No points are plotted for the 0.4 copy 

concentrations since those PCR reactions did not amplify. 

 

 

S1.3 Acoustic telemetry analysis 

 

Age-1 cisco stocking (July 2020) 



 

 

249 

 

 A small cohort of juvenile cisco (22-months old from hatching; n = 260 total 

fish) were released into Keuka Lake, New York USA on 7-July-2020 prior to the pre-

fall stocking eDNA survey. Details are available in Table S3 below. 

Table S3: Summary of the fate of age-1 fish during the eDNA surveys, inferred from 

acoustic telemetry. Size ranges for acoustic-tagged fish (n = 56 tags) were total length 

(mm) 184.4 ± 9.6 (165–204) and mass (g) 53.5 ± 10.6 (34.4–75.7).  

eDNA survey Time, t, from 

stocking event 

Survival, 𝑺̂t, (95% 

CI) 

Abundance, 𝑵̂t, 

(95% CI) 

21-July-2020 14 days 0.73 (0.62, 0.86) 195 (165, 229) 

    

27-Oct-2020 98 days 0.21 (0.13, 0.35) 57 (35, 94) 

 

Fish abundance and density estimation  

 Time-to-event survival modeling of inferred mortalities of tagged fish was 

used to estimate lake-wide cisco abundance (Koeberle et al., 2023). We first obtained 

Kaplan-Meier survival estimates (𝑆̂t) as follows (Kaplan and Meier, 1958): 

𝑆̂𝑡 = ∏ (𝑡
𝑖=1

𝑛𝑖 − 𝑑𝑖

𝑛𝑖
), (1) 

where time, t, is the number of days after stocking for each cohort of fish, ni is the 

number of tagged fish at risk of mortality at the start of interval i, and di are the 

number of fish that died during time interval i. The standard error of survival estimates 

was derived as:  

𝑆𝐸̂(𝑆̂𝑡) = ∏ (𝑡
𝑖=1

𝑑𝑖

𝑛𝑖(𝑛𝑖−𝑑𝑖)
), (2) 



 

 

250 

 

We then estimated lake-wide fish abundance on 21-July-2020 (eDNA survey) and 27-

October-2020 (eDNA survey) as the product of survival rates and the number of total 

fish stocked in each cohort:  

𝑁̂ =  𝑆̂𝑡𝑛0, (3) 

where abundance, 𝑁̂, is the estimated total number of fish in Keuka Lake at a specified 

time, t, and 𝑛0 is the number of fish stocked at t = 0. Additionally, 95% confidence 

intervals (95% CI) were calculated for estimates of abundance:  

95% 𝐶𝐼 =  1.96 × 𝑆𝐸̂(𝑆̂𝑡)𝑛0, (4) 

To infer fish distributions when we sampled for eDNA, we calculated a residency 

index (IR), modified from Holbrook et al. (2019) and Kraft et al. (2023) to estimate 

local fish abundance at each acoustic receiver across our whole-lake telemetry array, 

given by:  

𝐼𝑅 =  𝑁̂  ×  
𝐿

𝑇
 (5) 

For tagged fish, the numerator of the ratio, L, is the sum of each detection event (time 

in minutes) observed at a given receiver. T is the sum of time observed at any receiver 

during a given time interval. Lastly, we estimated fish density (𝐹𝐷̂) using IR from each 

receiver as: 

𝐹𝐷̂ =  
𝐼𝑅

𝐴 
 (6) 

Area, A, was calculated by dividing the total surface area of Keuka Lake into 

respective regions (Figure 1).  



 

 

251 

 

 

S1.4 Lake currents survey 

 

Drifter design 

 

Figure S3. Lagrangian drifter example used in this study to characterize water 

currents in Keuka Lake, New York, USA. The design consists of a 1.2 m x 1.2 m 

multi-panel drogue (left) attached below a 0.61 m float (right) via a nylon tether 

configured to 12 m or 18 m sampling depths. A small GPS unit was placed inside the 

top cap of the float (see below the solar beacon) to autonomously record coordinates, 

programmed on a 10-minute transmission window. This model is an economical 

drifter designed specifically for sampling in lakes (McCaffrey and Koeberle, 2024).  

 

Drifter data analysis 

For our drifter study of currents in Keuka Lake, we targeted sampling two consecutive 

one-week periods and paired drifter deployment locations to sample each depth 

(Figure S3). Similar lake conditions to the fall 2020 eDNA survey were targeted for 



 

 

252 

 

fall 2022, primarily with the criteria: 1) A comparable timeframe with similar wind 

conditions and air temperatures; 2) sampling before the thermocline breakdown; and 

3) similar water temperature profiles. We left drifters deployed for an additional five 

days (15-October-2022 to 19-October-2022) with a persistent southerly wind event to 

track their movements during a period of expected stronger lake currents.  

We first evaluated the empirical drift paths of each drifter to validate their movement 

over the course of the sampling periods. To analyze distance travelled, we 

standardized drifter tracks with the deployment location as a common origin (0, 0), 

then calculated Euclidean distance and bearing angle using the haversine method (van 

Brummelen, 2013; McCaffrey and Koeberle, 2024). The full survey (30-September-

2022 to 19-October-2022) included two consecutive 7-day time periods and an 

additional 5-day time period. From this dataset, we specified a total of n = 18 drifter 

deployments (n = 10 drifters at 12m depths; n = 8 drifters at 18m depths) included in 

the subsequent distance analysis. If a drifter became anchored in shallower waters 

during the sampling period, we moved it offshore and treated this occasion as a new 

deployment. We plotted eight out of twelve drifter trajectories which sampled for a 

full 7-day deployment without anchoring in Figure S4 below as representative 

examples of the maximum extent of distances traveled over the full sampling period. 



 

 

253 

 

 

Figure S4. Empirical trajectories of drifters that sampled for a full 7-day period (30-

Sept-2022 to 7-Oct-2022; 7-Oct-2022 to 14-Oct-2022) in Keuka Lake, New York, 

USA. Here, for visualization, we include example trajectories that sampled 

continuously and excluded drifters that were redeployed due to anchoring. Trajectories 

for all drifter deployments are available in McCaffrey and Koeberle (2024).   

 



 

 

254 

 

We calculated the extent of drifter movement assuming a conservative estimate of 48h 

for eDNA persistence in Keuka Lake. Brief time lags in GPS locations occurred in 

several drifter tracks; thus, we applied a rolling time window approach to analyze 

drifter distances using program R (version 4.3.2; R Core Team 2023).   

 

Steps for mean distance calculations (Figures 3b-c): 

1) Bin 10-minute GPS timestamps into 1hr periods by calculating the mean 

position (latitude, longitude) within sequential 1h time steps for each drifter 

deployment (n = 18).  

2) For each drifter deployment, specify a rolling time (t) window with t = {0, 12, 

24, 36, 48} hours (sequential intervals). 

3) For the length of each deployment [0, n], calculate the Euclidean distance 

(using the haversine formula) between drifter coordinates, i, and i + t, from i0 to in-t.  

4) Calculate the mean distance and standard deviation of each 12m and 18m 

depth drifter for each t. 



 

 

255 

 

 

 

Figure S5. Boxplots for mean distances in Figure 3b-c for 12 m and 18 m drifters. 

We provide median (solid line), mean (dashed line), and sample size for the number of 

rolling window intervals applied for each time period.  

 

 

 

 



 

 

256 

 

S2. Results 

 

S2.1 eDNA results 

Positive and negative control tests 

All positive and negative extraction and qPCR controls performed as expected, with 

amplification of positive DNA-based controls, and no amplification detected in any of 

the three types of negative controls. These results indicate that the qPCR amplification 

process was successful during each reaction, and no contamination during the 

extraction or amplification process was detected. 

 

Environmental samples 

We did not find significant correlations between Ct values and sampling depth, 

distance from stocking location, and number of positive lab replicates (Figure S6). 

Field controls indicated sampling procedures performed well in limiting 

contamination. All but one of the negative field control samples correctly failed to 

amplify for cisco eDNA. The one jar blank control that amplified for cisco DNA was 

sample CISC-20-254 (Bartron et al. 2022; Koeberle 2025), collected during the 27-

Oct-2020 survey in the South branch of Keuka Lake. Positive detections in the South 

arm of jar blank sample CISC-20-254 and environmental sample CISC-20-220 (one 

out of nine total environmental samples that amplified for cisco DNA) were filtered at 

the same station by the same personnel. We infer, however, that it is unlikely the 

positive environmental sample CISC-20-220 resulted from contamination. Rather, 



 

 

257 

 

positive jar blank sample CISC-20-254 represented an isolated event because this 

control sample 1) had mean Ct = 41.46 (higher than most positive environmental 

samples, indicating a late-stage detection during amplification), 2) amplified with one 

of the two qPCR markers (marker COSP_cytB01), 3) was collected at a sampling site 

that failed to yield positive environmental samples (and was approximately 2km 

further south than the only other positive eDNA sample in the South arm), and 4) no 

other field controls failed from this study. Field contamination is unlikely to have 

influenced positive environmental samples with lower (earlier stage detection) Ct 

values, including the only positive eDNA detection that occurred in the South arm 

(environmental sample CISC-20-220; Ct = 39.57). In addition, jar blank sample CISC-

20-254 contamination from lab handling is unlikely given the negative lab controls. 

Lastly, positive detections in the West arm and Confluence regions were not 

influenced by the positive field control sample, as both regions represent different 

sampling batches with separate collection and processing control samples that were 

free of cisco DNA.   

 

qPCR inhibition tests  

Inhibition testing was conducted on eight separate PCR runs for all 263 filter samples. 

Of all the filter samples tested for PCR inhibition, three samples resulted in delayed 

amplification of control DNA and were categorized as inhibited resulting in an 

inhibition rate across all sites of 1.1%. The three samples that resulted in inhibition 

were CISC-20-117, CISC-20-150, and CISC-20-155 (Bartron et al. 2022; Koeberle 



 

 

258 

 

2025). The three inhibited DNA samples were treated using a Zymo Inhibitor 

Removal Column (Zymo Research), retested to verify PCR inhibition was relieved 

and analysis continued for these samples. 

 

Figure S6. Outputs from tests with cycle threshold (Ct) values from samples that 

amplified for cisco Coregonus artedi DNA with marker COSP_cytB01. All water 

samples that were positive for DNA amplified using marker COSP_cytB01 (9/9 

positive samples), in contrast with marker COSP_ND401 (5/9 total samples). a) Mean 

Ct plotted by sample depth (12m or 18m). b) Linear regression of mean Ct ~ 

Euclidean distance from stocking site (y = 38.427 + 0.211β, R2 = 0.25, p = 0.168). c) 

Linear regression of mean Ct ~ Number replicates that amplified for DNA (y = 40.09 - 

0.13β, R2 = 0.01, p = 0.72). 

 



 

 

259 

 

S2.2 Acoustic telemetry results 

No tagged fish stocked in previous cohorts (before 2020) were detected during this 

study and survival models revealed that cisco abundance prior to fall stocking was 

likely close to zero besides the age-1 cohort stocked on 7-July-2020 (Koeberle et al., 

2023). From the period between stocking and the eDNA survey in July, tagged age-1 

fish appeared to disperse over a wider range than age-0 fish, with detections that 

occurred in the West and East arms (Figure S7). Nonetheless, tagged age-1 fish 

potentially indicate site preference of the broader population for the west arm as they 

were detected on the same receivers in the October period as tagged age-0 fish 

(Figure S7). 

 

Table S4: Estimated cisco Coregonus artedi densities, 𝐹𝐷̂ (fish/hectare), within the 

West arm (Figure 1; total area 1,009ha) of Keuka Lake, New York USA inferred from 

acoustic telemetry receivers. We divided the west arm into four equal regions, each 

with surface area = 252.25 ha. Acoustic receivers are anchored on the lake bottom and 

face upwards in the water column. 

Acoustic 

Receiver 

Latitude Receiver 

depth (m) 

Percent tag 

detections (L/T) 

𝑭𝑫̂ (95% CI) 

(fish/ha) 

1 42.583712 46.0 0.57 42.0 (19.6, 90.1) 

2 42.568592 57.0 0.272 20.0 (9.3, 43.0) 

3 42.556775 51.8 0.104 7.7 (3.6, 16.4) 

4E 42.526266 44.2 0.054 4.0 (1.9, 8.5) 

 

 

 

 

 



 

 

260 

 

 

Figure S7. Summary of age-1 cisco Coregonus artedi distribution. a) Study system 

Keuka Lake, New York USA. b) No cisco DNA was detected in the pre-stocking 

eDNA survey conducted on 21-July-2020. c) Cumulative presence-absence of tagged 

cisco (n = 56 tagged age-1 fish) from release on 7-July-2020 to the eDNA survey on 

21-July-2020. Note, three receivers failed to log data in July compared with October. 

d) Cumulative presence-absence of surviving tagged age-1 fish from 15-October-2020 

(stocking event) to 27-October-2020 (fall eDNA survey).   



 

 

261 

 

Any use of trade, firm, or product names is for descriptive purposes only and does not 

imply endorsement by the U.S. Government. 

 

References 

1. Bartron, M. L., et al. (2022). Returning native fish communities to inland 

ecosystems of the Northeast: Coregonine restoration in Keuka Lake: eDNA results for 

detection of Cisco (Coregonus artedi). USFWS Northeast Fishery Center, Lamar, PA. 

Final Completion Report. Available via Zenodo: 

https://doi.org/10.5281/zenodo.16794444. 

2. Holbrook, C., et al. (2019). glatos: A Package for the Great Lakes Acoustic 

Telemetry Observation System. <https://gitlab.oceantrack.org/GreatLakes/glatos>. 

3. Kaplan, E. L. and P. Meier. (1958). Nonparametric estimation from incomplete 

observations. Journal of the American statistical association, 53(282), 457-481. 

4. Klymus, K. E., et al. (2020). Reporting limits of detection and quantification 

for environmental DNA assays. Environmental DNA, 2(3), 271-282. 

5. Koeberle, A. L, et al. (2023). Whole-lake acoustic telemetry to evaluate 

survival of stocked juvenile fish. Scientific Reports 13, 18956. 

6. Koeberle, A. (2025). How Accurately Does eDNA Reflect the Spatial 

Distribution of Cold-Water Fish? Field Validation from a Temperate Lake [Data set]. 

In Freshwater Biology. Zenodo: https://doi.org/10.5281/zenodo.16794444. 

7. Kraft, S., et al. (2023). Residency and space use estimation methods based on 

passive acoustic telemetry data. Movement Ecology, 11(1), 12. 

8. Lesperance, M. L. et al. (2021). A statistical model for calibration and 

computation of detection and quantification limits for low copy number environmental 

DNA samples. Environmental DNA, 3(5), 970-981. 

9. McCaffrey, L. and A. Koeberle. (2024) An economical open-source 

Lagrangian drifter design to measure deep currents in lakes. Water Resources 

Research, 60(10).   

10. R Core Team. (2023). R: a language and environment for statistical 

computing. R foundation for statistical computing, Vienna. Available at 

https://www.rproject.org. 

11. U.S. Fish and Wildlife Service (2020). Quality Assurance Project Plan. 

12. van Brummelen, G.R. (2013). Heavenly Mathematics: The Forgotten Art of 

Spherical Trigonometry. Princeton University Press. ISBN 9780691148922.



 

 

262 

 

D CHAPTER 4 APPENDIX 

 

Appendix for Chapter 4: Integrating acoustic telemetry and demographic modeling 

to inform cisco (Coregonus artedi) restoration in Keuka Lake, New York 

 

S1. Methods 

 

S1.1 Fish stocking 

 

Description of acoustic telemetry study: We accessed an acoustic telemetry dataset 

described in detail by Koeberle et al. (2023) and Sethi et al. (2024). In this study, a 

subset of stocked juvenile cisco (Coregonus artedi) were equipped with small 

electronic acoustic transmitters (hereafter, “tags”) and monitored across a whole-lake 

acoustic telemetry receiver array (see Table S1 for stocking and tagging numbers). 

Acoustic tags used for this study were Juvenile Salmon Acoustic Telemetry System 

(JSATS) Acoustic Micro Transmitters (AMT). Fall fingerling cisco were surgically 

implanted with 0.6g Lotek model L-AMT-5.1B (Lotek-brand, Newmarket, Ontario, 

Canada) acoustic tags and yearling cisco were implanted with larger 3.5g Lotek model 

L-AMT-8.2 acoustic tags. All tagging procedures for the Keuka Lake study followed 

protocol described by McKenna et al. (2021). Tags were programmed on a 20 second 

transmission interval. A total of n = 24 Lotek-brand Wireless Hydrophone System 

4250 Series acoustic receivers were deployed throughout Keuka Lake to track the 

movement and survival of tagged fish. Acoustic receivers autonomously stored 

standard movement data (e.g., tag detections at receiver locations). To yield time-to-

event data for multistage survival modeling, we inferred that a tagged fish mortality 

event occurred on the date after which an individual subject’s tag was no longer 

detected across the telemetry array. Further acoustic telemetry specifications, 

including a map of acoustic receiver locations, and time-to-event data processing and 

analysis are available in Koeberle et al. (2023).  



 

 

263 

 

Table S1: Summary of juvenile cisco (Coregonus artedi) stocked in Keuka Lake, NY, 

USA. Acoustic-tagged cisco mass (g) and total length (mm) are shown for each cohort 

as mean ± SD (range). Fish are stocked by New York State Department of 

Environmental Conservation (NYSDEC) with all cisco rearing from 2024-onwards 

occurring at NYSDEC Bath Fish Hatchery in Bath, NY, USA, along Cold Brook 

tributary, less than 10 km southwest of the Keuka Lake inlet. All cisco broodstock was 

collected from Chaumont Bay in Lake Ontario, NY, USA from 2019-2023. From 

2024-onwards, cisco broodstock is collected from U.S. Fish and Wildlife Service 

(USFWS) Jordan River National Fish Hatchery, Elmira, MI, USA.  

Stocking 

date 

Age-at-

release 

(months) 

Total no. 

fish stocked 

No. 

tagged 

fish 

Mass (g) Total 

Length 

(mm) 

17-Oct-

2018a 

10 98,789c,d 40 17.5 ± 9.8 

(6.1-44.4) 

129.0 ± 

20.7 (97-

177) 

19-Jun-

2019a 

18 1,360c,d,e 22 76.8 ± 14.1 

(54.6-103.0) 

207.4 ± 

11.5 (187-

226) 

9-Oct-2019 22 28e 28 111.1 ± 24.0 

(46.7–156.6) 

221.9 ± 15.

6 (183–

252) 

9-Oct-2019 10 92,165c,d 60 9.5 ± 1.8 

(5.5–13.9) 

111.1 ± 6.6 

(98–136) 

7-July-

2020 

19 204e 56 53.5 ± 10.6 

(34.4–75.7) 

184.4 ± 9.6 

(165–204) 

15-Oct-

2020b 

10 202,250c,d 66 12.7 ± 1.7 

(9.9–16.0) 

116.8 ± 5.4 

(105–128) 

20-Oct-

2021 

10 36,290c,d N/A N/A ~89.0 – 

102.0 

18-Oct-

2022 

10 36,190c,d N/A N/A N/A 

23-Oct-

2023 

10 28,161c,d N/A N/A N/A 

23-Oct-

2024 

10 109,080c,f N/A N/A ~102.0 

March 

2025 

15+ 2,000c,f,g N/A N/A N/A 

May 2025 48+ 400e,h N/A N/A N/A 
aCohort of tagged fish excluded from acoustic telemetry analysis due to incomplete receiver coverage. 
bFinal cohort of stocked fish for whole-lake acoustic telemetry analysis (Koeberle et al. 2023). 
cNYSDEC Bath Fish Hatchery, Bath, NY, USA, dNYSDEC Oneida Fish Hatchery, Oneida, NY, USA, 
eUSGS Great Lakes Science Center Tunison Laboratory of Aquatic Science, Cortland, NY, USA 

(McKenna et al. 2021).  
fFirst year stocking with cisco eggs sourced from USFWS Jordan River National Fish Hatchery. 
gAnticipated spring yearling release.  
hAnticipated adult (age-4+) release.    



 

 

264 

 

S1.2 In situ demographic modeling  

  

Description of multistage time-to-event model: Our candidate multistage time-to-

event model set consisted of a suite of 1-, 2-, and 3-stage survival models (n = 16 total 

models). We explored covariates on individual survival stages and transition times 

between stages, including the covariates release age (fall fingerling or yearling), total 

length (mm), and Fulton’s condition factor (K) (Fulton 1904). Stage one referred to as 

‘straight-to-death’ was specified with a 1-day (24hr) time interval to reflect post-

stocking mortality. All candidate models were implemented in a Bayesian framework 

using JAGS with the R package ‘rjags’ (Plummer 2003, 2021) following procedures 

specified by Sethi et al. (2024) including three MCMC chains, 1,000 iterations for the 

adaptation, 5,000 iterations for the burn-in and posterior sampling, and a thin rate of 

20. Model performance was evaluated within a multimodal information criterion 

framework using Deviance Information Criterion (DIC) that included a model 

complexity term (Spiegelhalter et al. 2014). Discrete juvenile survival estimates were 

derived from the top-ranked multistage model via DIC and the median and highest 

density intervals (HDI; 95% Credible Intervals) were calculated for posterior 

distributions. Finally, all population-level estimates were calculated by decaying the 

total number of fish stocked in each cohort by time-specific multistage survival 

estimates.  



 

 

265 

 

Table S2: Historical cisco (Coregonus artedi) catch available from lake-wide gillnet 

surveys in Keuka Lake, New York, USA. Data provided by NYSDEC.  

 

Survey 

year 

Month Net type Age data 

available 

Lengthd (mm) 

mean (range) 

Total 

cisco 

catch (n) 

Survey 

used for 

catch 

curve 

1971 Sept Standard 

ganga 

No 257 (191-371) 51 No 

1974 Sept Standard 

gangb 

No 262 (165-328) 25 No 

1975 Sept Standard 

gangb 

No 282 (188-373) 63 No 

1976 Sept Standard 

gangb 

No 294 (185-368) 89 No 

1979 Aug Lake 

Trout 

surveyc 

Yes 312 (242-367) 30 Yes 

1982 July Lake 

Trout 

surveyc 

Yes 323 (266-390) 37 Yes 

1983 Aug Lake 

Trout 

surveyc 

No 328 (295-350) 14 No 

1985 Aug Lake 

Trout 

surveyc 

Yes 340 (295-390) 22 Yes 

1988 Aug Lake 

Trout 

surveyc 

Yes 333 (300-280) 18 Yes 

1991 Aug Lake 

Trout 

surveyc 

Yes 358 (335-395) 3 Yes 

1994e Aug Lake 

Trout 

surveyc 

No N/A 3 No 

aStandard gang gillnet survey with panels of 1.5”, 2”, 2.75”, 3”, and 3.5” mesh size (net dimensions 

N/A). 
bStandard gang gillnet survey with panels of 1.5”, 2”, 2.5”, 3”, and 3.5” mesh size (net dimensions 

N/A). 
cLake Trout (Salvelinus namaycush) bottom gillnet survey using 8ft x 350ft multifilament nets with of 

1” (25ft panels on each end), and 50ft panels each of 1.5”, 2.0”, 2.5”, 3”, 4”, 5” mesh size. The survey 

is conducted lake-wide with nets set overnight.  
dTotal length for 1979-1991 data; Total length likely for 1971-1976 but unconfirmed.  

eFinal year with cisco observed in catches. Standard Lake Trout survey has continued every three years 

on average to present.  



 

 

266 

 

 
Figure S1. Age distributions for netted cisco (Coregonus artedi) in Keuka Lake, New 

York, USA from years 1979 to 1991 (𝑛 = 110 total fish). Note, the final caught cisco 

(𝑛 = 3) in 1994 lack length and age information so were excluded from the catch 

curve analysis. All catch data provided by NYSDEC. 

 



 

 

267 

 

 
Figure S2. Length-at-age data from gillnet catches (years 1979-1991) with data 

provided by NYSDEC, and acoustic-tagged juvenile fish (years 2018-2020) from an 

acoustic telemetry study (Koeberle et al. 2023). 

 

  



 

 

268 

 

S1.3 Population modeling 

 

Description of stocked stage-based matrix model: Here we detail the derivation of 

the stocked or augmented matrix approach used for hatchery-reared cisco (Coregonus 

artedi) reintroductions to Keuka Lake, New York, USA. Future population trajectories 

are the product of the projection matrix A and a vector of abundances n(𝑡) at each 

stage at time, t. We also add a vector for annual fish releases (i.e. hatchery stocking) 

R, therefore the non-augmented population is defined as: 

 

𝑛(𝑡 + 1) = 𝐀 ∙ 𝑛(𝑡) + 𝐑 

 

= [

0 𝑠𝑎𝑚2 𝑠𝑎𝑚3

𝑠0 0 0
0 𝑠𝑎 𝑠𝑎

] ∙ [

𝑛1

𝑛2

𝑛3

] + [
0
0
𝛿

] 

 

= [

𝑠𝑎𝑚2𝑛2 + 𝑠𝑎𝑚3𝑛3

𝑠0𝑛1

𝑠𝑎𝑛2 + 𝑠𝑎𝑛3 + 𝛿
] 

 

Matrix parameters correspond to the Keuka Lake cisco life-cycle graph (Figure 2), 

where 𝑠𝑗 is juvenile survival, 𝑠𝑎 is adult survival, and 𝑚3+ is adult fertility (the 

product of number of eggs per fish and proportion that survive to sj). We define an 

augmented matrix to account for annual additions of released fish that survive and 

enter the wild population, an approach we believe is useful for fisheries conservation 

and management applications that utilize fish stocking. The augmented matrix with 

annual stocking is: 

 

𝑛′(𝑡 + 1) = 𝐀′ ∙ 𝑛′(𝑡) 

= [

0 𝑠𝑎𝑚2 𝑠𝑎𝑚3 0
𝑠0 0 0 0
0 𝑠𝑎 𝑠𝑎 𝛿
0 0 0 1

] ∙ [

𝑛1

𝑛2

𝑛3

1

] 

 

= [

𝑠𝑎𝑚2𝑛2 + 𝑠𝑎𝑚3𝑛3

𝑠0𝑛1

𝑠𝑎𝑛2 + 𝑠𝑎𝑛3 + 𝛿
1

] 

 

Since: 

𝑛′(𝑡) =  [ 𝑛(𝑡)
1

] 

 

then 𝑛(𝑡 − 1) = 𝑛′
1:𝑘−1(𝑛 + 1), e.g., this is equivalent to dropping the terminal 1 

term from the 𝑛′(𝑡 + 1) vector above. Thus, 

 



 

 

269 

 

𝑛(𝑡 + 1) = [

𝑠𝑎𝑚2𝑛2 + 𝑠𝑎𝑚3𝑛3

𝑠0𝑛1

𝑠𝑎𝑛2 + 𝑠𝑎𝑛3 + 𝛿
] 

 

which is equivalent to the non-augmented system.  

 

The augmented matrix modeling approach is also useful for conducting perturbation 

analysis, which evaluates the rate at which the population growth rate, 𝜆, responds to 

changes in a vital rate parameter. This is calculated from the implicit derivative of the 

characteristic polynomial of the projection matrix with respect to a given parameter, 

evaluated at the value 𝜆 corresponding to the dominant real eigenvalue of the 

projection matrix model. Using the preceding example matrix A: 

 

𝐀 = [

0 𝑠𝑎𝑚2 𝑠𝑎𝑚3

𝑠0 0 0
0 𝑠𝑎 𝑠𝑎

] 

 

with the characteristic polynomial specified as: 𝜆3 − 𝑠𝑎𝜆2 − 2𝑠𝑎𝑚2𝑠0𝜆 − 𝑠𝑎
2𝑚2𝑠0. 

The augmented matrix A’ is: 

 

𝐀′ = [

0 𝑠𝑎𝑚2 𝑠𝑎𝑚3 0
𝑠0 0 0 0
0 𝑠𝑎 𝑠𝑎 𝛿
0 0 0 1

] 

 

with the characteristic polynomial: (1 − 𝜆)(𝜆3 − 𝑠𝑎𝜆2 − 2𝑠𝑎𝑚2𝑠0𝜆 − 𝑠𝑎
2𝑚2𝑠0).  

 

The characteristic polynomial of the augmented matrix A′ is simply the characteristic 

polynomial of the non-augmented matrix A, multiplied by (1 − 𝜆). This is because 

augmentation adds an eigenvalue = 1 to the original matrix A. Note, for non-

augmented matrices where 𝜆 < 1, augmentation will make the second-largest real 

eigenvalue the projected growth rate for the modeled population. Perturbation analysis 

will not differ between the augmented matrix and the original matrix. This is because 

the multiplication of a function by a constant – in this case, (1 − 𝜆) – does not modify 

the rate of change of the function with respect to any of the variables.  

Alternative parameterization of augmentation: For released fall fingerlings in 

Keuka Lake, the augmentation vector (column 4 in matrix 𝐀′), 𝐑𝒇𝒇, can also be 

parameterized as: 

 

𝐑𝑭𝑭 = [

0
0

𝑁𝑟𝑒𝑙𝑆′1𝑆2

1

] 

 

with the number of released hatchery fish (𝑁𝑟𝑒𝑙) that survive two years (𝑆′1𝑆2) and 



 

 

270 

 

enter the wild population as age-3 adults. With this formulation, fall fingerlings enter 

the wild population at time 𝑡 = 2 years (as adults). The augmentation vector may also 

include stocking both age classes at the same time using 𝐑𝑭𝑭 and 𝐑𝒀 

parameterizations from the main text.   

 

Note, we parameterize the stocking vector for the Keuka Lake cisco population model 

with annual post-stocking survival rates from the multistage model (joint survival 

across stages one, two, and three). Alternatively, the stocking vector could also be 

parameterized with stage-based survival rates estimated from the multistage model.    



 

 

271 

 

S1.4   Vital rates and modeling scenarios 

 

Table S3: Vital rates applied to population projection modeling for cisco (Coregonus 

artedi) reintroduced to Keuka Lake, New York, USA. We specified parameter 

estimates, variation, and their sources where applicable (e.g., in situ or literature 

derived). Survival rates are parameterized as a proportion and are expressed as 

survival through a specified life history stage.  

 

 

  



 

 

272 

 

Table S4: In situ vital rates used for population modeling scenarios. Juvenile survival 

rates are annual values estimated from the multistage time-to-event survival model, 

e.g., joint stages one, two, and three for hatchery-origin fish or derived stage three for 

wild fish. Note, average fecundity and joint age-0 survival estimates (literature-based) 

are retained from Table S3 for each scenario. 

 

 

 

  



 

 

273 

 

S2. Results  

 

S2.1 Demographic modeling results  

 

Table S5: Results of a candidate model set for multistage survival models (Sethi et al. 

2024) fit to time-to-event data of acoustic-tagged juvenile cisco (Coregonus artedi) in 

Keuka Lake, New York, USA. Multimodel inference is based on Deviance 

Information Criterion (DIC). Across the constructed model set, all models within the 

DIC 95% confidence set included three stages of survival (Table 1). Model 

convergence was sufficient with Gelman-Brooks-Rubin convergence statistics 𝑅̂  <
1.1 for all fitted parameters. The second and third ranked models included size 

covariates length (DIC = 24.6%) and condition factor (DIC = 17.6%). While this 

indicates moderate support for a size effect, survival rates were best predicted by 

release age so subsequent estimates were derived from fall fingerling and yearling 

survival curves.  

 

No. 

stages 

Model description (covariate, stage 

estimated) 

DIC ΔDIC DIC 

Weight 

3 Age, Stage 2 length estimated 1144.9 0.0 0.578 

3 Length, Stage 2 length estimated  1146.605 1.705 0.246 

3 Condition, State 2 length estimated  1147.276 2.376 0.176 

2 Length, Unit interval 1 1168.334 23.434 <0.01 

2 Age, Unit Interval 1 1213.825 68.925 <0.01 

3 No covariates, Stage 2 length estimated 1219.264 74.364 <0.01 

2 Length, Stage 1 length estimated  1224.353 79.453 <0.01 

2 No covariates, Unit interval 1 1230.982 86.082 <0.01 

2 Condition, Unit interval 1 1231.215 86.315 <0.01 

2 Age, Stage 1 length estimated 1298.833 153.933 <0.01 

2 No covariates, Stage 1 length estimated  1333.585 188.685 <0.01 

2 Condition, Stage 1 length estimated  1334.729 189.829 <0.01 

1 Length 1630.209 485.309 <0.01 

1 Age 1783.673 638.773 <0.01 

1 Condition 1824.264 679.364 <0.01 

1 No covariates 1825.225 680.325 <0.01 

 



 

 

274 

 

 
Figure S3. Catch curves reconstructed from historical cisco (Coregonus artedi) gillnet 

catches (𝑛 = 110 total adult cisco) in Keuka Lake, New York, USA. We include 

catches from 1979-1991 with known age data. Following Ogle (2013), we conducted 

linear regression-based catch curve analysis, excluding age classes with 𝑛 < 1 fish 

(e.g., age-9 and age-10). Further, we fit catch curves at the peak catch at age-5, 

excluding age-4 fish (𝑛 = 85 fish in final analysis; solid points). This approach 

estimates adult instantaneous mortality 𝑍 = 0.666 (𝑠𝑒 0.06) and annual adult 

mortality 𝐴 = 0.486 (𝑠𝑒 𝑁𝐴; 𝑆 = 0.514) from the natural log of catch data. We 

therefore applied in situ adult annual survival rates to population models specified as 

𝑆3+ = 0.514.  



 

 

275 

 

Table S6: Review of adult cisco (Coregonus artedi) annual mortality rates with extant 

populations. Minnesota inland lake dataset provided by Minnesota Department of 

Natural Resources (MN DNR) for catch curve analysis from survey years 2016-2023 

(Ten Mile Lake 2,056ha surface area, 63m max depth; Carlos Lake 1,054 ha surface 

area, 50m max depth; Elk Lake 122ha surface area, 28m max depth). Given the 

sample sizes of aged cisco in MN lakes, we applied the Chapman-Robson maximum 

likelihood-based estimator (Chapman and Robson 1960) for adult survival following 

procedures by Smith et al. (2012) and Ogle (2013). 

 

 

  



 

 

276 

 

S2.2 Population modeling results  

 

 
Figure S4. Distribution of cisco (Coregonus artedi) caught in gillnet surveys 

aggregated from years 1971-1976 in Keuka Lake, New York, USA (see Table S2 for 

more information on catch methods). Black points indicate sampling locations without 

any netted cisco. The map uses a Latitude-Longitude coordinate system with NAD83 

datum.  

  



 

 

277 

 

Table S7: Results from lower-level perturbation analysis with the sensitivity (the 

effect on the population growth rate λ to an absolute change in a model parameter) and 

elasticity (proportional change) of the baseline Low scenario (𝑆1 = 0.004, 𝑆2 =
0.053) augmented matrix model. Lower elements refer to one or more elements of the 

population projection matrix that are products of other elements such as age-0 survival 

(the joint product of egg, fry, and summer fingerling survival). 

 

Parameter Notation Sensitivity Elasticity 

Wild Age-1 S1 0.501 0.004 

Wild Age-2 S2 0.038 0.004 

Wild adult S3+ 0.992 0.988 

Fecundity F3+ 0.000 0.004 

Egg hatch Segg  0.034 0.004 

Fry Sfry 0.033 0.004 

Summer fingerling Ssf 0.036 0.004 

N stocked (yearlings) Nrel Y 0.000 0.000 

Hatchery age-2 S’2 0.000 0.000 

 

Table S8: Results from lower-level perturbation analysis with the sensitivity and 

elasticities of the optimistic High scenario (𝑆1 = 0.078, 𝑆2 = 𝑆3+ = 0.514) 

augmented matrix model. 

 

Parameter Notation Sensitivity Elasticity 

Wild Age-1 S1 1.642 0.179 

Wild Age-2 S2 0.249 0.179 

Wild adult S3+ 0.641 0.462 

Fecundity F3+ 0.000 0.179 

Egg hatch Segg  2.155 0.179 

Fry Sfry 2.096 0.179 

Summer fingerling Ssf 2.287 0.179 

N stocked (yearlings) Nrel Y 0.000 0.000 

Hatchery age-2 S’2 0.000 0.000 

  



 

 

278 

 

Table S9: Total annual hatchery costs of NYSDEC Bath Fish Hatchery, Bath, NY 

utilized in a simple cost-benefit analysis following methods by Fonken et al. (2022). 

Here, we calculated the cost per number of spawners produced, assuming linearity 

between hatchery costs and the number of fish reared. Juvenile cisco (Coregonus 

artedi) represent 6.7% of the total fish raised in the Bath Fish Hatchery. Given the 

expenses below, we calculated the cost to raise fall fingerlings is $0.64 per fish (for 

100,00 fish). Hatchery production is limited to 2,000 yearlings. Assuming a linear 

increase in cost we thus estimate the cost to raise yearlings is $1.28 per fish. All data 

provided by NYSDEC.  

 

Expense Amount (USD) 

Total feed $87, 441 

Utilities $21, 803 

Full-time staff $216, 872 

Miscellaneous expenses  $85, 841 

  



 

 

279 

 

Table S10: Results from a literature-based review to identify adult coregonine 

(Coregonus spp.) densities observed in comparable extant populations.  

 

Lake Species Estimated 

density 

adults 

(fish/ha) 

Method Citation 

Lake Superior 

(contemporary) 

C. artedi 18.96 – 54.58 Stock-recruit 

model 

Rook et al. 

(2021) 

Lake Superior 

(historical) 

C. artedi 7.49 – 194.45  Stock-recruit 

model 

Rook et al. 

(2021) 

Inland lakes 

Trout, 

Muskellunge, and 

Mendota, 

Wisconsin, USA 

C. artedi 1.8 – 558.4 

(87-1,551 

totala) 

Hydroacoustics, 

gill net surveys 

Rudstam, Clay, 

and Magnuson 

(1987); 

Rudstam et al. 

(1993) 

Inland lakes Ten 

Mile, Carlos, and 

Elk, Minnesota, 

USA 

C. artedi 378 – 1,692 

(range 182 – 

2,613)b 

Hydroacoustics, 

gill net surveys 

B. Holbrook, 

MN DNR (pers. 

comm. 8 

October 2024)  

Lake Paasivesi, 

Finland 

C. albula 7.8 – 37.1 

(260-530 

totala) 

Hydroacoustics, 

gill net surveys 

Jurvelius, 

Lindem and 

Louhimo (1984) 

Lakes Vänern, 

Vättern and 

Mälaren, Sweden 

C. albula  150 – 400 

(range 1 – 

1,300a) 

Hydroacoustic, 

midwater trawl 

surveys 

Axenrot and 

Degerman 

(2016) 
aDerived from total estimates across year classes. 
bDensity estimates provided for age-1+ fish from survey years 2013-2023. 



 

 

280 

 

Table S11: Results from the Keuka Lake cisco (Coregonus artedi) population 

viability analysis. For simulated populations, we calculate the stochastic growth rates, 

𝜆𝑠, across modeled scenarios. All simulated populations that apply in situ juvenile and 

adult survival rates indicate negative population growth rates where 𝜆𝑠 < 1.0. 

 

Wild Scenario1 Adult 

survivalb  

Boom 

likelihood 

𝝀𝒔 95% CI 

Low 51.4% 1/3 0.519 (0.519, 0.520) 

Low 70% 1/3 0.703 (0.703, 0.703) 

Low 51.4% 1/7 0.518 (0.518, 0.518) 

Low 70% 1/7 0.702 (0.702, 0.702) 

Medium 51.4% 1/3 0.707 (0.705, 0.709) 

Medium 70% 1/3 0.841 (0.839, 0.842) 

Medium 51.4% 1/7 0.654 (0.651, 0.656) 

Medium 70% 1/7 0.797 (0.794, 0.798) 

High 51.4% 1/3 0.868 0.861, 0.873) 

High 70% 1/3 1.041c (1.037, 1.046) 

High 51.4% 1/7 0.783 (0.780, 0.786) 

High 70% 1/7 0.952 (0.948, 0.956) 
aWild-equivalent juvenile survival: Low (in situ) (𝑆1 = 0.004, 𝑆2 = 0.053; Medium (upper 95% 

Credible Interval) 𝑆1 = 0.078, 𝑆2 = 0.18; High (age-2 equivalent to age-3+ survival) 𝑆1 = 0.078, 𝑆2 =
 𝑆3+. 
bAdult annual survival 0.514 from in situ catch curve analysis, adult annual survival 0.70 from literature 

estimates. 
cIndicates scenario where reintroduced cisco population is expected to grow as 𝜆𝑠 > 1.0. 



 

281 

 

 
 

Figure S5. Evidence of long-term survival of stocked hatchery fish. A 220mm total 

length and 99g mass cisco (Coregonus artedi) was netted on 22-July-2022 (645 days 

post-stocking) during a standard gillnet survey targeting Lake Trout (Salvelinus 

namaycush) in Keuka Lake, New York, USA. Otolith and scale analysis revealed this 

specimen was age-2+ (31 months old) and therefore inferred to have been stocked on 

15-October-2020 as a fall fingerling (Table S1). In September 2019, only one juvenile 

cisco was netted in a forage net survey (likely stocked as a yearling in June 2019, 

Table S1). No cisco were netted in the following forage net survey in September 2022. 

Photograph by D. Mulhall, NYSDEC.   



 

282 

 

References 

 

Axenrot, T., and E. Degerman. 2016. Year-class strength, physical fitness and 

recruitment cycles in vendace (Coregonus albula). Fisheries Research, 173(1), pp. 61-

69. https://doi.org/10.1016/j.fishres.2015.03.017 

 

Chapman, D. G. and D. S. Robson. 1960. The analysis of a catch curve. Biometrics, 

16(3), pp. 354-368. https://doi.org/10.2307/2527687 

 

Fielder, D. G., and K. N. McDonnell. 2024. Evaluation of potential factors affecting 

the success of cisco re-introduction and re-establishment in Lake Huron. Ecological 

Modelling, 496, 110817. https://doi.org/10.1016/j.ecolmodel.2024.110817 

 

Fisch, N. C., et al. 2019. A comparison of age-and size-structured assessment models 

applied to a stock of cisco in Thunder Bay, Ontario. Fisheries Research, 209, pp. 86-

100. https://doi.org/10.1016/j.fishres.2018.09.014 

 

Fonken, D. R., et al. 2023. Benefits of stocking fewer but larger individuals with 

implications for native fish recovery. Canadian Journal of Fisheries and Aquatic 

Sciences, 80(3), pp. 439-450. https://doi.org/10.1139/cjfas-2022-0071 

 

Freeberg, M. H., Taylor, W. W., and R. W. Brown. 1990. Effect of egg and larval 

survival on year‐class strength of lake whitefish in Grand Traverse Bay, Lake 

Michigan. Transactions of the American Fisheries Society, 119(1), pp. 92-100. 

https://doi.org/ 10.1577/1548-8659(1990)119[0092:EOEALS]2.3.CO;2 

 

Fulton, T. W. 1904. The rate of growth of fishes. Fisheries Board of Scotland. Annual 

Report 22, pp. 141–241. 

 

Jurvelius, J., Lindem, T., and J. Louhimo. 1984. The number of pelagic fish in Lake 

Paasivesi, Finland, monitored by hydroacoustic methods. Fisheries Research, 2(4), pp. 

273-283. https://doi.org/10.1016/0165-7836(84)90030-4 

 

Koeberle, A. L., et al. 2023. Whole-lake acoustic telemetry to evaluate survival of 

stocked juvenile fish. Scientific Reports, 13(1), 18956. 

https://doi.org/10.1038/s41598-023-46330-6 

 

McKenna Jr, J. E., et al. 2021. Acoustic tag retention and tagging mortality of juvenile 

cisco Coregonus artedi. Journal of Great Lakes Research, 47(3), pp. 937-942. 

https://doi.org/10.1016/j.jglr.2021.03.020 

 

Ogle, D. H. 2013. fishR Vignette – Catch Curve Estimates of Mortality. Retrieved 

from https://derekogle.com/fishR/examples/oldFishRVignettes/CatchCurve.pdf.  

 

Plummer M. 2021. rjags: Bayesian Graphical Models using MCMC. R package 



 

283 

 

version 4–12. https://CRAN.R-project.org/package=rjags.  

 

Plummer M. 2023. JAGS: A program for analysis of Bayesian graphical models using 

Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed 

Statistical Computing, Vienna, Austria.  

 

Pratt, T. C., and S. C. Chong. 2012. Contemporary life history characteristics of Lake 

Superior deepwater ciscoes. Aquatic Ecosystem Health & Management, 15(3), pp. 

322-332. https://doi.org/10.1080/14634988.2012.708639 

 

Rudstam, L. G., Clay, C. S., and J. J. Magnuson. 1987. Density and size estimates of 

cisco (Coregonus artedi) using analysis of echo peak PDF from a single-transducer 

sonar. Canadian Journal of Fisheries and Aquatic Sciences, 44(4), pp. 811-821. 

https://doi.org/10.1139/f87-099 

 

Rudstam, L. G., Lathrop, R. C., and S. R. Carpenter. 1993. The rise and fall of a 

dominant planktivore: effects on zooplankton species composition and seasonal 

dynamics. Ecology 74, pp. 303-319. https://doi.org/10.2307/1939294 

 

Sethi, S. A., et al. 2024. Multistage time-to-event models improve survival inference 

by partitioning mortality processes of tracked organisms. Scientific Reports, 14(1), 

14628.https://doi.org/10.1038/s41598-024-64653-w 

 

Spiegelhalter, D. J., et al. 2014. The Deviance Information Criterion: 12 Years on. 

Journal of the Royal Statistical Society Series B: Statistical Methodology, 76(3), pp. 

485-493. https://doi.org/10.1111/rssb.12062 

 

Smith, M. W., et al. 2012. Recommendations for catch-curve analysis. North 

American Journal of Fisheries Management, 32(5), pp. 956-967. 

https://doi.org/10.1080/02755947.2012.711270 

 

Stone, U. B. 1938. Growth, habits, and fecundity of the ciscoes of Irondequoit Bay, 

New York. Transactions of the American Fisheries Society, 67(1), pp. 234-245. 

https://doi.org/ 10.1577/1548-8659(1937)67[234:GHAFOT]2.0.CO;2 

 

Van Oosten, J. 1956. Tabular Data on Whitefish Submitted to National Research 

Council. National Research Council, Washington D.C (1956), p. 3. 

 

Yule, D. L., et al. 2008. How systematic age underestimation can impede 

understanding of fish population dynamics: lessons learned from a Lake Superior 

cisco stock. Transactions of the American Fisheries Society, 137(2), pp. 481-495. 

https://doi.org/10.1577/T07-068.1