THERMAL INFRARED REMOTE VELOCIMETRY OF TIDAL FLOWS IN SALT MARSH ESTUARIES: INTEGRATIVE METHODS FOR MEASURING INUNDATION IN WETLAND 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 Evan T. Heberlein December 2025 © 2025 Evan T. Heberlein ALL RIGHTS RESERVED THERMAL INFRARED REMOTE VELOCIMETRY OF TIDAL FLOWS IN SALT MARSH ESTUARIES: INTEGRATIVE METHODS FOR MEASURING INUNDATION IN WETLAND ECOSYSTEMS Evan T. Heberlein, Ph.D. Cornell University 2025 Sea level rise, increasingly powerful storms, and expanding coastal development have heightened the risks faced by coastal populations and ecosystems around the world. Remote sensing methods can play an increasingly powerful role in help- ing researchers understand risks posed by changing hydrodynamic conditions in estuarine wetlands and the coastal ocean. This dissertation applies field-based thermal infrared surface velocimetry to tidal flows in salt marsh estuaries, demon- strating this velocity measurement method’s ability to quantify complex flows of significance to coastal management and ecosystems. Chapter 1 presents a novel application of a drone-based uncooled thermal infrared sensor to measure tidal surface velocity in an estuary. The local wind speed and direction were used to calculate the wind stress at the water surface relative to the velocity measurement axis, which explained discrepancies between drone-based vs. in-channel acoustic profiler velocity measurements. Chapter 2 demonstrates the use of a cooled ther- mal infrared camera to measure spring tide velocities through the former dike at the mouth of the Herring River Estuary (HRE) at Cape Cod National Seashore in Massachusetts. These measurements are used to develop a stage-velocity rating curve, which can be used to estimate the volume of seawater exchanged through this restriction at a given tidal elevation providing an important baseline for post- restoration tidal exchange targets. Chapter 3 applies the same imaging method to beach overwash tidal flooding at Duck Harbor in the upper HRE, which has caused significant ecological and biogeochemical changes inland from the beach breach. Tidal velocity and depth are remotely measured from thermal infrared images to estimate flow rate through the wide, shallow overwash channel. Chapter 4 builds from the ecological impacts of tidal flooding to investigate more fundamen- tal relationships between vegetation distribution and inundation, using statistical modeling to analyze patterns of soil moisture content and soil organic matter in ephemeral wetlands. BIOGRAPHICAL SKETCH Evan T. Heberlein grew up in St. Paul, Minnesota and attended Whitman Col- lege in Walla Walla, Washington, where he majored in Biology & Environmental Studies and minored in Politics, graduating in May 2016. He studied abroad on a wildlife ecology program in Tanzania and Kenya with the School for Field Studies in Fall 2013, and wrote an undergraduate honors thesis about the ecological and biogeochemical effects of diking and salt marsh drainage in the Herring River Es- tuary at Cape Cod National Seashore, where he was an intern in Summer 2014. After college he worked as an Assistant Scientist for Terracon Consultants in Min- nesota conducting field sampling at environmental remediation sites. He received his master’s in Environmental Science & Management from the Bren School at the University of California, Santa Barbara in June 2021, where he studied water resources management, coastal and marine resources management, and environ- mental data science, and co-chaired the Environmental Justice Club. He joined Cornell’s Civil & Environmental Engineering Department in August 2021 in the Environmental Fluid Mechanics & Hydrology group, and conducted a research award at the Pacific Northwest National Laboratory Marine and Coastal Research Lab in Sequim, Washington during the final year of his Ph.D. iii ACKNOWLEDGEMENTS I gratefully acknowledge the financial support of the United States Geological Survey 104g National Competitive Grants Program, administered by the New York State Water Resources Institute; the Cornell Atkinson Center for Sustainability Academic Venture Fund; and the United States Department of Energy Office of Science Graduate Student Research Award. I would also like to acknowledge the assistance of my many collaborators in data collection, analysis, and logistical support: • Carpinteria Salt Marsh Reserve fieldwork, November 2021 and May 2022: Dr. Marc Mayes, Dr. Bryn Morgan, Professor Kelly Caylor, Dr. Andrew Brooks, Evert Vermeer, Lili Prahl • Cape Cod National Seashore fieldwork—Herring River Estuary, July 2022 and May 2023: Andrew Epps, Camille Blevins, Petra Zuniga, Professor Kevin Befus, Professor Michelle Hummel, Dr. Kasra Naseri • Cape Cod National Seashore fieldwork—Duck Harbor Beach, July-August 2023: Dr. Katherine Castagno, Dr. Meagan Eagle, Tim Smith • Archbold Biological Station fieldwork, April 2024: Professor Jed Sparks, Professor Warren G. Abrahamson, Dr. Betsie Rothermel, Rachel Fedders, Dan Petticord Thank you to Cape Cod National Seashore Aquatic Ecologist Dr. Sophia Fox for giving me the opportunity as an undergraduate intern to volunteer on data collection in the Herring River Estuary one day a week during Summer 2014. 11 years later that opportunity has profoundly shaped the trajectory of my career as a scientist, and I hope this dissertation can contribute to the important work of restoring the Herring River Estuary. iv Thanks to Dr. Nick Ward at Pacific Northwest National Lab for allowing me the time and flexibility to complete this dissertation during my time there. A special thanks to Dr. Seth Schweitzer for writing the endlessly useful IRQIV.jl library and for the many hours spent in meetings and the field helping me learn how to do thermal infrared velocimetry. And finally, a heartfelt thank you to my advisor Professor Todd Cowen for bringing an environmental data scientist with a BA in biology into a high-level fluid mechanics group and supporting my success every step of the journey. v TABLE OF CONTENTS Introduction 1 1 Wind stress effects on drone-based thermal infrared surface ve- locimetry measurements of tidal flow in an estuary 5 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.1.1 Image-based velocimetry techniques . . . . . . . . . . . . . . 5 1.1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.2.1 Study Site . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.2.2 Acoustic profiler in-situ velocity measurements . . . . . . . . 13 1.2.3 Drone-based image collection . . . . . . . . . . . . . . . . . 17 1.2.4 Georeferencing and stabilization of drone images . . . . . . . 20 1.2.5 Space-time image velocimetry (STIV) . . . . . . . . . . . . . 21 1.2.6 Tide and wind conditions . . . . . . . . . . . . . . . . . . . 24 1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 1.3.1 Uncertainty associated with time synchronization between velocity measurements . . . . . . . . . . . . . . . . . . . . . 30 1.3.2 Coordinate system and sign of velocity measurements . . . . 31 1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 1.4.1 Physical causes of time-varying velocity differences . . . . . 34 1.4.2 Measurement uncertainty and data quality . . . . . . . . . . 36 1.4.3 Implications for image-based surface velocity measurements . 40 1.5 Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 1.6 Open Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 1.7 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2 Developing a stage-discharge rating curve for an impounded es- tuary using thermal infrared surface velocimetry 43 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2.1 Study site . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 2.2.2 Field campaign . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.2.3 Image Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 55 2.2.4 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 2.3.1 Developing rating curves . . . . . . . . . . . . . . . . . . . . 61 2.3.2 Hydrograph comparison to model outputs . . . . . . . . . . 64 2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 vi 3 Remote measurements of flow rate, bathymetric change and sedi- ment fluxes through a tidal beach overwash channel using thermal infrared imagery 68 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.2.1 Study site . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.2.2 Field campaign . . . . . . . . . . . . . . . . . . . . . . . . . 71 3.2.3 Infrared velocimetry analysis . . . . . . . . . . . . . . . . . . 72 3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.3.1 Flow rate estimate from surface velocimetry . . . . . . . . . 77 3.3.2 Elevation surveys . . . . . . . . . . . . . . . . . . . . . . . . 78 3.3.3 Acoustic measurement of flow rate at confluence . . . . . . . 79 3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4 Is a historical vegetation map a useful predictor of dry season soil water fraction in seasonal ponds? 85 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.1.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4.2.1 Study Site . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4.2.2 Experimental design . . . . . . . . . . . . . . . . . . . . . . 91 4.2.3 Soil Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . 94 4.2.4 Mixed effects modeling . . . . . . . . . . . . . . . . . . . . . 95 4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.3.1 Correlation between soil water fraction and pond type . . . 96 4.3.2 Soil organic matter . . . . . . . . . . . . . . . . . . . . . . . 100 4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 A Chapter 2 Appendix: Velocity measurement uncertainty 106 A.1 River-side comparison with ADV velocities . . . . . . . . . . . . . . 109 A.2 River-side acceleration . . . . . . . . . . . . . . . . . . . . . . . . . 113 A.3 Bay-side mass conservation . . . . . . . . . . . . . . . . . . . . . . . 118 B Chapter 3 Appendix: Space-time diagram contrast enhancement methods 122 C Chapter 4 Appendix: Mixed effects model supporting informa- tion 126 vii LIST OF TABLES 1.1 Micasense Altum infrared microbolometer band specifications. . . . 17 1.2 Wind velocity and direction average and standard deviation (σ) values from the CSMR meteorological station during each flight. . . 25 1.3 AquaDopp surface-extrapolated and drone-based STIV velocity mea- surements, and two metrics of wind effects on surface velocity: angle between wind direction and measured flow direction, and the wind stress component parallel to the flow direction along the x-axis dur- ing each flight. Note that the in-channel surface-extrapolated tidal velocity changed directions between flights A and B but the STIV velocity measurement direction remained constant. . . . . . . . . . 32 4.1 Pond types: the major vegetation wetland species used to classify seasonal ponds at ABS by hydroperiod, arranged from driest to wettest species. Adapted from [Abrahamson, Warren G. et al., 1984]. 91 4.2 ANOVA results for measured soil water content at pond center across all vegetation types. . . . . . . . . . . . . . . . . . . . . . . 97 4.3 Pairwise Tukey test results for measured soil water content at pond center between adjacent vegetation types on the ABS scale. All non-adjacent pairwise differences were statistically significant at the P < 0.001 level. . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.4 Mixed model intercept and slope values for modeled soil water con- tent by weight fraction. Significance codes reflect difference be- tween model estimate and 0. Full raw coefficient estimates listed in Table C.1 in Appendix C. . . . . . . . . . . . . . . . . . . . . . 98 4.5 Mixed effects model ANOVA table for soil water content including degrees of freedom. All listed effects were statistically significant at the P < 0.001 level. . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.6 Estimated marginal mean (EMM) contrasts for modeled soil water content at pond center between adjacent pond types. . . . . . . . . 100 4.7 Estimated marginal means of linear trends between adjacent pond types for modeled soil water content across all transect distances. . 100 C.1 Raw coefficient estimates for mixed effects model model described in 4.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 viii LIST OF FIGURES 1.1 Aerial overview of Carpinteria Salt Marsh Reserve and weather station location. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2 Beam orientations relative to coordinate system in the x− z plane of the Nortek AquaDopp acoustic velocity profiler. Note x is pos- itive in the downstream, outgoing tide direction, and h denotes water depth. Inset formula shows calculation of streamwise veloc- ity U from the oblique beam velocities in the j-th vertical bin. . . . 14 1.3 Experimental schematic showing orientation and surface extent of AquaDopp measurement beams in blue relative to drone imagery FOV in red. Actual drone FOV included portions of both channel banks for each flight. Figure not to scale. . . . . . . . . . . . . . . 18 1.4 Three-band RGB drone image collected during flight C with the x-axis located directly over the AquaDopp velocity measurement head. The positive x-direction denotes tidal outflow, while x < 0 denotes tidal inflow. The translucent yellow region shows the lo- cation of u-velocity measurements along the x-axis: both the im- age pixel coordinates sampled in STIV and the surface extent be- tween AquaDopp beams 1 and 3. Images were rotated so that the AquaDopp x-axis was parallel to pixel columns in images. Ground control points visible during this flight are circled in red. True north relative to the rotated image is shown at top left, with the spatial scale at bottom right. An orange recovery float is visible to the right of x-axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.5 Comparison of image stabilization methods for flight C. Each point shows the respective method’s estimated horizontal displacement of the drone from the previous image based on a stationary ground location (GCP or the previous image’s corner pixel locations) in meters. Displacements were calculated as distance magnitude from the previous timestep without accounting for direction, and means were subtracted from each time series for comparison. . . . . . . . 22 1.6 Tide and wind conditions during the experiment (conducted May 2022). Top: Relative tidal elevation (black solid) and 10-minute av- eraged surface-extrapolated tidal velocity from the deployed AquaDopp (Usurfex, red dashed) over two experiment days. Red dashed ve- locity zero-axis highlights the sign of tidal velocity denoting flow direction. Bottom: Average wind speed (black solid) and direction (red dashed) over the same time period. Timing of five 10-minute drone flights denoted by blue vertical lines on both plots, letters next to lines denote flight. . . . . . . . . . . . . . . . . . . . . . . . 26 ix 1.7 Rotated infrared thermal image from flight C corresponding to the same timestep as the RGB image in Figure 1.4, upsampled to the resolution of the visible light bands. Pseudocolor scale shows dif- ferences in relative pixel intensity in the thermal infrared band, not converted to real temperature. Location of the x-axis sampled by STIV overlying the AquaDopp is shown in red. An animated ver- sion of this figure showing the stabilized x-axis location is available in the supporting information. . . . . . . . . . . . . . . . . . . . . 28 1.8 Space-time diagram for flight C, assembled by sampling one spatial column of pixels from the x-axis in each image during the 10-minute hover timeseries. 20 ϕ angles detected from the space-time diagram are shown at their centered locations within the overlapping square windows of the diagram which were analyzed to derive each angle. 29 1.9 Surface-extrapolated velocity (Usurfex, blue line) from the AquaDopp running-averaged over the same duration as 20 flight C STIV ve- locity measurements (USTIV , red points) generated from the the 20 detected ϕ angles shown on the space-time diagram in Figure 1.8 according to Equation 1.2. . . . . . . . . . . . . . . . . . . . . . . 30 1.10 Parallel wind stress (τ||) effect on calculated difference (U∆) between STIV velocity measurement (USTIV ) and surface-extrapolated ve- locity measurement (Usurfex) for five flights. Individual boxplots summarize 20 time-synchronized comparisons between the two ve- locity measurements, showing median, first and third quartile box, largest non-outlier whiskers based on interquartile range, and any individual outliers. The vertical axis intercept U∆ = 0 denotes that the two velocity measurements were equal. . . . . . . . . . . . . . . 33 2.1 Aerial view of Chequessett Neck Road Dike during a flood tide, annotated with red arrows showing the flow direction allowed by the three culverts. Note the aerated flow ejected from the one bidirectional culvert. Image credit: Merrily Cassidy, Cape Cod Times [Bragg, 2014] . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.2 Section view of bidirectional southeast culvert within Chequessett Neck Road Dike illustrating important flow control structures. Fig- ure not to scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.3 USGS tide gauge elevation on either side of the dike and calculated elevation head across the dike during the field experiment (July 13-15, 2022). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 x 2.4 River-side (left) and bay-side (right) pseudocolor georectified FOVs for IR-QIV, showing GCPs used for georeferencing (red points) and shadow transect for profile measurement of streamwise surface ve- locity (red line). Both images captured at night, illustrating de- tailed thermal surface features for pattern tracking present in the flow. Pseudocolor scale represents raw, uncorrected pixel intensity. Axes units are UTM x,y coordinates (m). . . . . . . . . . . . . . . 58 2.5 Stage-velocity rating curves for the river-side of the dike showing average velocity vs. elevation head every five minutes from 3:25 to 7:30 on July 14, 2022. Error bars show 95% confidence interval for each set of cross-culvert surface velocity measurements, derived using the Student’s t-distribution (n = 101 for each timestep). . . . 62 2.6 Stage-velocity rating curves for the bay-side of the dike showing average velocity vs. elevation head every five minutes from 21:00 on July 14 to 4:00 on July 15, 2022. Error bars show 95% con- fidence interval for each set of cross-culvert surface velocity mea- surements, derived using the Student’s t-distribution (n = 101 for each timestep). The bay-side culvert entrance becomes submerged when the tidal stage is greater than 1.55 m; surface velocities mea- sured during this period (shown in red) do not reflect the velocity through the culvert and were not curve-fit. A three-hour gap in data collection when the culverts were submerged is visible on the left side of the figure. . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.7 Predicted discharge hydrograph from IR-QIV-derived river and bay rating curves in solid blue, plotted alongside the two models’ pre- dicted discharge values during the field experiment in dashed pink (Delft3D) and green (hydraulic mass balance). . . . . . . . . . . . 64 2.8 Hydrograph derived from river and bay rating curves in Figures 2.5 and 2.6 evaluated for mass conservation between ebb and flood tidal stages during the period of the study, corrected for constant discharge from the Herring River. Flow rate shown in blue on left axis, cumulative volume for each tidal phase shown in red on right axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.1 The Herring River Estuary outlined in blue at left, with location of Duck Harbor subwatershed labeled and washover fan inset imagery collected by Planet Labs during 2023 field campaign. The approx- imate camera deployment location on August 1-2 is shown on the south side of the washover fan in the inset. . . . . . . . . . . . . . 70 xi 3.2 Reconstructed tidal water surface elevation based on predicted tidal minima and maximum during August 1 overwash event shown as blue line, shifted to georeferencing vertical datum. Three ground control point elevations used to calculate offset between tidal and georeferencing datum shown as colored points at top. The green point represents the water surface elevation when overwash was first observed, and the blue dot shows the WSE in Figure 3.3. . . . . . 73 3.3 Orientation of evenly placed horizon-parallel sampling lines for pat- tern tracking in red, distributed along white transect across over- wash channel during flow, with waves on the water surface shown in pseudocolor infrared image from approximately 10:22 PM, August 1, 2023. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.4 Left: space-time diagram assembled from temporal samples of in- terpolated pixel intensity along 8th sampling line, labeled with an- gles of wave (red) and bulk (orange) velocity signals identified in autocorrelation integral; Center: normalized autocorrelation inte- gral plotted by angle, illustrating identified wave (red) and bulk velocity (orange) peaks plotted as angles on space-time diagram; Right: histogram of pixel intensities in space-time diagram after pixel-wise median subtraction. . . . . . . . . . . . . . . . . . . . . 75 3.5 Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. The camera was located on the left side of the cross-channel transect as shown in Figure 3.1. . . . . . . . . . . . . . . . . . . . . 78 3.6 Elevation profiles showing change in bed elevation across overwash channel mouth after three measured overwashes. Data collected by Katherine Castagno. . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.7 AquaDopp amplitude response colormap and identified WSE in red over the course of deployment at the confluence between the Duck Harbor overwash channel and the Herring River. . . . . . . . . . . 80 3.8 One hour running-averaged AquaDopp flow rate for Duck Harbor channel confluence with Herring River. Positive values indicate flow from Duck Harbor into Herring River, negative values indicate water from the river backing up Duck Harbor channel. . . . . . . . 81 4.1 Northern set of four ponds with transects highlighted in red. Re- droot ponds in blue (R1 north, R2 south), Maidencane ponds in purple (M1 north, M2 south). . . . . . . . . . . . . . . . . . . . . . 93 4.2 Southern set of four ponds with transects highlighted in red. Hy- pericum ponds in green (H1 east, H2 west), Broomsedge ponds in yellow (B1 west, B2 east). . . . . . . . . . . . . . . . . . . . . . . . 94 xii 4.3 Soil water content by weight fraction along normalized transect distance grouped by pond type (color), with presence or absence of standing water shown as shape. Trendlines show linear fit by pond type with 95% confidence interval. Transect distance of 0 denotes pond center, 1 denotes pond edge. . . . . . . . . . . . . . . . . . . 97 4.4 Relationship between soil water content and soil organic content, both by fractional weight. Red trendline shows linear fit to all data with 95% confidence interval: Adj. R2 = 0.7134, Intercept = -0.0288, Slope = 0.4872, P < 0.001 . . . . . . . . . . . . . . . . . 101 A.1 Histogram of all vertical (i) and horizontal (j) pixel displacements for subwindows along river-side cross-culvert velocity profile. Hor- izontal axis scale set by IR-QIV pattern tracking algorithm maxi- mum search radius of ±48 pixels. . . . . . . . . . . . . . . . . . . . 106 A.2 Percentage of NaN’s induced by IR-QIV pattern tracking algorithm for subwindows along cross-culvert velocity profile for river-side measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 A.3 Percentage of NaN’s induced by IR-QIV pattern tracking algorithm for subwindows along cross-culvert velocity profile for bay-side mea- surements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 A.4 Proportion of cross-culvert velocity profile sampled for pre- and post-submergence periods of tidal inflow to the culvert. Submerged portion in center not shown as velocities were not included in the rating curve calculation. . . . . . . . . . . . . . . . . . . . . . . . . 109 A.5 Thermal infrared image used for IR-QIV analysis and inset photo showing deployment location of ADV on river-side of two-directional culvert relative to front of pier between culverts. . . . . . . . . . . 110 A.6 Left vertical axis: 1 Hz-averaged U (light blue) and 5-minute cen- tered moving mean streamwise velocity (dark blue) measured by an ADV deployed on the river side of the dike over a three hour period during ebb tide. Right vertical axis: USGS tide gauge ele- vation head hz across the dike (red) used to time-synchronize ADV data. Positive values indicate outflow on both axes. . . . . . . . . 111 A.7 Scatterplot comparing time-synchronized velocity measurements be- tween the ADV and nearest IR-QIV subwindow sampled for rating curve velocity. The inner color gradient for each point shows the timestamp after 3:25 on July 14, 2022, and the outer color gradi- ent shows the approximate depth of the ADV measurement volume beneath the water surface at each measurement time. . . . . . . . . 112 A.8 Locations of IR-QIV subwindows sampled to calculate acceleration in the central culvert (upper red points) and nearest the ADV (lower red points). The closest subwindows on the upstream side of the red shadow transect are shown as black points. . . . . . . . . . . . 114 xiii A.9 Values of U du dx calculated by backward finite difference for the cen- tral culvert (upper line of subwindows in Figure A.8), with color coded timesteps matching the central point color in Figure A.7. x=0 denotes the location of the last subwindow upstream of the shadow transect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 A.10 Values of U du dx calculated by backward finite difference for the ADV location (lower line of subwindows in Figure A.8), with color coded timesteps matching the central point color in Figure A.7. x=0 denotes the location of the last subwindow upstream of the shadow transect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 A.11 Scatterplot comparing time-synchronized velocity measurements be- tween the ADV and acceleration-adjusted IR-QIV subwindow sam- pled for rating curve velocity. The inner color gradient for each point shows the timestamp after 3:25 on July 14, 2022, and the outer color gradient shows the approximate depth of the ADV mea- surement volume beneath the water surface at each measurement time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 A.12 Vector field of flow into bay-side culvert annotated with mass con- servation evaluation scheme. The widest apart streamlines that reach from the culvert to the far edge of the image are shown in red. Surface velocities normal to the transverse profiles are shown using the color gradient. Fence posts above the culvert entrance are visible along the upper edge of the image. . . . . . . . . . . . . 119 A.13 Left axis: absolute value of surface flux at four transverse profiles for timesteps with two valid streamlines (shown as points connected by colored solid lines), following scheme in Figure A.12. Profile 3X/4 is farthest from the culvert entrance. Data collection con- tinued at this location during flood tide until 4:00 on July 15 but only the streamline orientations at the points shown here allowed for the calculations described above. Right axis: ratio of distance from internal sluicegate to culvert entrance Ls (3 m) divided by water depth to bottom of sluicegate hs shown as dashed black line 120 B.1 Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. Contrast-enhancement pre-processing of space-time diagrams was implemented before slope detection, with values below the 50th percentile pixel intensity coerced to zero. . . . . . . . . . . . . . . . 123 xiv B.2 Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. Contrast-enhancement pre-processing of space-time diagrams was implemented before slope detection, with values below the 80th percentile pixel intensity coerced to zero. . . . . . . . . . . . . . . . 124 C.1 Ranked predictions vs. residuals for mixed effects model described in 4.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 xv INTRODUCTION Sea level rise, increasingly powerful storms, and expanding coastal development have heightened the risks faced by coastal populations and ecosystems around the world. Remote sensing methods can play an increasingly powerful role in helping researchers understand risks posed by changing hydrodynamic conditions in estu- arine wetlands and the coastal ocean. Upon arriving at Cornell with this interest in mind, I sought out a collaboration with contacts from an undergraduate intern- ship at Cape Cod National Seashore in Massachusetts, where the Herring River Estuary is currently undergoing a significant tidal restoration via dike removal. The reintroduction of unrestricted tides at this site constitutes a large-scale field experiment in accelerated sea level rise, which has provided a unique opportunity to document significant tidal inundation events using remote velocimetry or veloc- ity measurement methods during multiple field campaigns in this rapidly changing ecosystem. This dissertation, which follows the papers option, is organized into four chapters summarizing four field campaigns that sought to demonstrate the ca- pabilities of thermal infrared remote sensing to measure tidal inundation, as well as quantify the effects of inundation on wetland ecosystems. In chapter 1: Wind stress effects on drone-based thermal infrared surface ve- locimetry measurements of tidal flow in an estuary, I sought to gain familiarity with thermal infrared surface velocimetry using a simple one-dimensional pattern tracking algorithm to measure a relatively undisturbed open-channel tidal flow in a research reserve. However, the use of a drone-based thermal imager presented a novel set of capabilities and challenges both within our group’s surface velocimetry experience and in the larger field of research. Analyzing these data also presented an opportunity to document the effect of wind on remote surface velocity mea- surements, which has been of interest to our group since taking this method to the 1 field. This chapter built a strong foundation of skills in combining measurements from image datasets with contemporaneous data from other sources like profilers and weather stations, and this paper was published in Water Resources Research in September. Building from this experience learning thermal infrared velocimetry methods in chapter 1, I wanted to apply these methods to questions of interest at my primary study site. Chapter 2: Developing a stage-discharge rating curve for an impounded estuary using thermal infrared surface velocimetry, is based on my first field campaign in the Herring River Estuary during summer 2022, which involved measuring high-velocity overnight tidal flows through the former dike at the mouth of the river using a thermal infrared camera and a two-dimensional pattern tracking algorithm. This campaign’s primary goal was to develop a stage-discharge rating curve for the former dike at the mouth of the Herring River, which can be used to estimate the volume of seawater exchanged through this restriction at a given tidal elevation. This rating curve serves to quantify the pre-restoration tidal con- dition which previously posed significant measurement challenges and provides an important baseline for post-restoration tidal exchange targets, and our manuscript detailing these results will soon be submitted to Coastal Engineering. During my field campaign for chapter 2, I observed the mortality of a large low-lying forested area in the upper Herring River Estuary caused by tidal flood- ing through a dune system breach, which formed during a 2021 winter storm that had carved a new pathway for seawater to enter the estuary during high tides. The Duck Harbor Beach breach created a significant new dynamic in the hydrology and salinity regime of the study system, and with the dike reconstruction underway in 2023 I sought to measure this new tidal flow. Chapter 3, Remote measurements of flow rate, bathymetric change and sediment fluxes through a tidal beach overwash 2 channel using thermal infrared imagery combines the one-dimensional space-time image velocimetry pattern tracking method used in chapter 1 and the ground- based oblique perspective thermal camera used in chapter 2 to quantify the flow rate during overwash events at Duck Harbor. This field campaign presented an opportunity to combine my growing knowledge of wave physics and coastal engi- neering with my interest in beach morphodynamics as well as the ecological and biological effects of tidal flooding. This field campaign’s results remotely resolve both the flow velocity and depth of the overwash channel, helping to constrain the magnitude of this difficult to measure flow, and these results will be submitted to JGR: Oceans. While the salt-killed Pitch Pine trees at Duck Harbor had been mulched by the time I conducted my fieldwork for chapter 3, native salt marsh vegetation had already begun recolonizing areas behind the beach breach. Observing this ecolog- ical impact of tidal flooding and its implications for the changing biogeochemistry of the Herring River Estuary increased my interest in fundamental patterns of in- undation and vegetation distribution in wetlands. During Cornell’s Florida Field Course through the Ecology and Evolutionary Biology department, I conducted field research for chapter 4: Is a historical vegetation map a useful predictor of dry season soil water fraction in seasonal ponds? This study used a vegetation map of Archbold Biological Station in Florida compiled in the early 1980s to design a study of soil water content and soil organic matter in ephemeral wetlands that had been mapped by vegetation type. My findings showed a robust hierarchy of soil moisture and soil carbon across four different pond types identified by their dominant plant species over 40 years ago. This research setting helped me gain experience in statistical modeling and identifying trends in soil carbon across wet- land ecosystems, and our manuscript with author of the original map, Professor 3 Warren G. Abrahamson, is currently in review in Wetlands. My progression from comprehending remote sensing methods for measuring tidal inundation, to addressing significant research questions with these methods in challenging measurement conditions at an active restoration site, to expanding my scope to the ecological and biogeochemical effects of different inundation regimes demonstrates my growth as a scientist during my PhD. 4 CHAPTER 1 WIND STRESS EFFECTS ON DRONE-BASED THERMAL INFRARED SURFACE VELOCIMETRY MEASUREMENTS OF TIDAL FLOW IN AN ESTUARY 1.1 Introduction Remote measurements of water velocities are a powerful tool for observing surface and bulk characteristics of environmental flows. Drones, also known as unmanned aerial vehicles (UAV) or small uncrewed aircraft systems (sUAS), are a flexible and easily-deployable platform for a variety of airborne imaging sensors capable of making water velocity measurements [Eltner et al., 2020, Perks et al., 2016]. Drone-mounted cameras capture nadir images with minimal perspective distor- tion, enabling synoptic and spatially-explicit velocity measurements over areas of flowing surface waters such as rivers and estuaries. However, additional work is needed to test the effect of environmental factors like wind on the measurement results of this field velocimetry methodology relative to more conventional wa- ter velocity measurement instruments, such as acoustic Doppler current profilers (ADCPs). This experiment was therefore designed to compare drone-mounted surface velocimetry measurements relative to in-channel acoustic velocity profiler measurements of a tidally-forced and wind-affected flow. 1.1.1 Image-based velocimetry techniques The pattern tracking techniques used here to remotely measure surface water ve- locities from aerial time series imagery are rooted in particle image velocimetry (PIV). PIV is a well-developed laboratory technique to quantify water velocities in digital images by tracking illuminated particles, producing time-varying velocity 5 vectors of small spatially-averaged regions across the entire field of view (FOV) [e.g., Cowen and Monismith, 1997]. Large-scale PIV (LSPIV) is a primarily field- based application of PIV tracking algorithms to derive two-dimensional distributed surface velocities from oblique images, which are orthorectified to physical units [e.g., Eder et al., 2023, Muste et al., 2014]. These remote velocity measurement techniques do not require the introduction of instruments such as acoustic velocity profilers into the water body, which avoids disturbing the flow being measured and keeps researchers out of harm’s way under unsafe conditions, such as floods. Large-scale PIV often requires the presence or introduction of trackable surface features, such as wood particles [Biggs et al., 2022, Kim et al., 2008], ice [Et- tema et al., 1997], foam [Al-mamari et al., 2019], or suspended sediment [Legleiter and Kinzel, 2020] to produce accurate surface velocity measurements, and this re- quirement has given rise to two alternative field-based pattern tracking techniques, which are combined in this study. Space-time image velocimetry (STIV) consists of a one-dimensional pattern tracking technique where intensities along lines of pixels parallel to the flow direction are assembled into a space-time diagram with axis units of distance (along the pixel line) and time (from one image timestep to the next). The slope of pixel intensity patterns visible in the space-time diagrams rep- resents the average velocity along one particular path in the image. This technique can be useful if insufficient surface features exist for two-dimensional large-scale PIV algorithms to be accurate [Fujita et al., 2019], especially with visible light imagery where surface features might not be evenly distributed. However, some detectable surface texture is required to derive accurate measurements from any velocimetry method using visible light imagery, including STIV. This requirement highlights a significant advantage of using thermal images for surface velocime- try. Infrared quantitative image velocimetry [IR-QIV — Schweitzer and Cowen, 6 2021] uses a thermal infrared camera with sufficiently-precise thermal resolution (<0.1◦C) to detect naturally-occurring temperature variations on the surface of flowing water, providing a trackable signal across the water surface and removing the need to artificially seed the flow. These subtle temperature variations on the water surface or “skin” are a signature of surface processes such as atmospheric heat exchange and wave breaking, as well as bottom-generated turbulence in rela- tively shallow water bodies like estuaries [Brumer et al., 2016]. Thermal infrared cameras are well-suited to detect these temperature variations due to the low pen- etration of thermal infrared radiation in water, on the order of 10−5 m in depth [Zappa et al., 2003]. Infrared imagery is particularly useful when flows of interest, such as snowmelt-driven floods [Fujita, 2017] or high tides, occur overnight, be- cause sunlight is not necessary for infrared sensors to accurately detect the surface water temperature. Thermal infrared image quality is highest and thermal sur- face velocimetry methods are therefore most accurate when the air temperature is lower than the water temperature [Legleiter et al., 2024a], but it is typically not permissible to fly a drone at night when these conditions often occur. This study builds on a variety of past work using thermal infrared imagery to measure tidal flows and wave dynamics in estuaries and the coastal ocean, where the undulating free surface is affected by tides, currents, and wind waves which have historically posed measurement challenges [Chickadel et al., 2009, Dugan et al., 2014, Laxague and Zappa, 2020, Schwendeman and Thomson, 2015, Sutherland and Melville, 2015]. In open channels, surface velocity measurements have also been used to infer characteristics of bulk velocity and channel discharge throughout the water column [Levesque and Oberg, 2012], leveraging the physical relationship between surface turbulence and boundary-layer turbulence induced by bed shear over variable bathymetry [Johnson and Cowen, 2016]. 7 The utility of drones as surface velocimetry tools has grown rapidly in the last decade. Consumer-grade visible light cameras mounted on a drone have been used to measure water surface velocity using seeded paper pellets and naturally- occurring algae [Tauro et al., 2015], and quantify two-dimensional surface hydro- dynamics using introduced dye tracers in branching salt marsh channels [Pinton et al., 2020]. Uncooled thermal infrared microbolometer cameras have been in- corporated into drone systems to conduct large-scale PIV on open channel flow in rivers [Eltner et al., 2021, Kinzel and Legleiter, 2019], detect groundwater discharge to the coastal ocean [Lee et al., 2016], and track infrared-emitting seed particles [Thumser et al., 2017]. Hot water can also be introduced into environmental flows as a thermal tracer that is detectable with a microbolometer camera [Lin et al., 2019], although introducing additional water at a different temperature could the- oretically change the characteristics of the flow being measured. Cooled thermal infrared cameras are now becoming light and power-efficient enough to be flown on multirotor drones for velocimetry applications measuring open-channel flow [Kinzel et al., 2024, Legleiter et al., 2024a], yielding images with better thermal resolution and less noise than uncooled microbolometer cameras. All image velocimetry methods extract physical velocity measurements from images via a calibration process relating the locations of features in the images to real-world physical coordinates of those features. In field-based image velocimetry applications like large-scale PIV, this process often involves georeferencing features visible in the camera field of view using GPS, which also locates velocity measure- ments within a coordinate system on the earth’s surface. A common technique involves the use of ground control points (GCPs), which are targets designed to be visible to the camera. By surveying GCPs using GPS, the relationship between the physical and image coordinates of these GCPs can be inferred and used to 8 account for the camera’s orientation and perspective distortion, providing physical coordinates for each pixel in the image [Holland et al., 1997]. GCP-based geo- referencing methods can be used to infer two-dimensional physical length scales in images as in this study, or three-dimensional relationships between objects in the field of view can be derived and used to process oblique imagery. General recommendations for drone-based velocimetry from [Detert, 2021] dictate that at least four GCPs are evenly distributed along the channel banks with some GCPs located near the corners of the drone-based imager’s field of view, where radial distortion causes the highest spatial uncertainty in a georeferenced image from a nadir viewing angle [Schweitzer and Cowen, 2022]. An alternative to GCP-based georeferencing of images collected from a stationary camera is structure from mo- tion [Tomasi and Kanade, 1992], which can derive 3D spatial scales for a landscape or structure from a time series of 2D images taken from a moving camera on a drone. With regard to images of moving water collected from a stationary drone, ready-to-use image stabilization toolboxes have been developed as a pre-processing step for velocimetry [e.g., Pizarro et al., 2022]. The growing use of drones for surface velocimetry has coincided with increasing access to software tools for implementing this approach [e.g., Legleiter and Kinzel, 2023], highlighting the need to understand wind’s effect on the output from these tools. Of the surface velocimetry research efforts that have addressed potential impacts of wind stress, these have primarily focused on the impact of wind on the presence of trackable surface texture and thresholds for valid results [Ansari et al., 2023, Brumer et al., 2016, Fairley et al., 2024, Lewis et al., 2018, Muste et al., 2005], or flight safety and stability of drone imagery [Koutalakis and Zaimes, 2022, Pearce et al., 2020], rather than the deviation of remote surface velocity field mea- surements from co-located, contemporaneous in-situ near-surface measurements 9 due to wind stress. Additionally, attempts to conduct thermal surface velocimetry of tidal flow from a drone in an estuarine setting have not been documented in the literature. 1.1.2 Objectives Our study investigates wind stress effects on remote surface water velocity measure- ments, which reflect both the channel’s bulk flow velocity and the wind-driven sur- face boundary layer. Measuring potential differences between average in-channel velocity and surface velocity will help identify conditions when surface-based in- ferences of bulk flow parameters might require particular care or caution. A small estuary presents an ideal setting for this experiment due to relatively consistent sea breeze winds and predictable tides, enabling observations of tidal flow both with and against the wind direction. Tidal flow velocity measurements were col- lected with a thermal infrared camera mounted on a drone. Drone flights were conducted immediately adjacent to a scientific-grade meteorological station, which presented an opportunity to assess the impact of wind stress on remote surface velocity measurements relative to contemporaneous and co-located in-channel ve- locity measurements from an acoustic velocity profiler. This study also presents a novel demonstration of this velocimetry method applied to a tidal flow, as we are not aware of previously published remote thermal surface velocities collected from an aerial drone in an estuary. 10 1.2 Materials and Methods 1.2.1 Study Site The study was conducted at Carpinteria Salt Marsh Reserve (CSMR), a 93 ha protected tidal wetland in Santa Barbara County, California, USA, owned and managed by the University of California Natural Reserve System (Figure 1.1). The marsh receives seasonal runoff from Franklin and Santa Monica Creeks, and daily tidal exchange through a roughly 30 m wide inlet from the Santa Barbara Channel. While discharge is not continuously monitored on the creeks entering the salt marsh, the neighboring larger watershed immediately to the southeast (Carpinteria Creek) has a USGS gauging station that showed zero discharge during this study [USGS, 2022a]. Therefore, all flow measured during this experiment was assumed to be tidally-driven. This experiment consisted of five repeated drone flights conducted during a 24-hour period starting at 12:00 Pacific time on May 23, 2022. Many California estuaries open and close seasonally due to sediment trans- port and hydrological regimes: regional beaches are exposed to high wave energy and stream discharge in coastal watersheds has high seasonal variance [Clark and O’Connor, 2019]. Despite its protection as a research reserve, channel modifica- tions are occasionally implemented in CSMR for management purposes. CSMR underwent a dramatic ecological disturbance following the Thomas Fire in Decem- ber 2017, when destructive debris flows deposited sediment over large areas of the marsh, resulting in a 222% increase in bare soil immediately following the debris flows and potential longer-term impacts to marsh type and plant diversity [Silva et al., 2022]. Emergency dredging of marsh channels to prevent the adjacent town of Carpinteria from flooding was undertaken in CSMR after the January 2018 de- 11 Figure 1.1: Aerial overview of Carpinteria Salt Marsh Reserve and weather station location. 12 bris flows and again in spring 2023, after high winter rainfall increased mountain erosion and led to the deposition of large amounts of sediment in the marsh, which blocked tidal creek channels. Demonstrating the potential of drone-based surface velocimetry in a salt marsh estuary is an important secondary outcome of this study, as this method could ad- dress some of the surface hydrology measurement challenges in these ecosystems. The geomorphology and ecological function of salt marshes is largely determined by hydrodynamic conditions and sediment supply, and these physical characteristics are significantly impacted by submerged vegetation and even marsh herbivore pop- ulation dynamics [Elmer, 2014, Friedrichs and Perry, 2001, Neumeier and Amos, 2006]. Due in part to these dynamic biophysical interactions, salt marshes often exhibit complex branching hydrology, resulting in a wide range of hydrodynamic conditions and stage-dependent tidal transport pathways [Fagherazzi et al., 2008], which can limit the utility of point velocity measurements. Short-term salt marsh hydrodynamics are primarily driven by tidal elevation and resulting pressure gradi- ents between two-directional tidal flow and freshwater inputs [Young et al., 2016], which over time influence marsh elevation through erosion and deposition of sed- iment [Voulgaris and Meyers, 2004]. In light of the complex spatial and temporal dynamics of salt marsh hydrology and associated ecological processes, an easily- deployable and mobile remote sensing system that can measure surface velocity presents a powerful tool for characterizing short-term hydrodynamic conditions across many channels in a given marsh. 1.2.2 Acoustic profiler in-situ velocity measurements The experiment consisted of repeated stationary drone flights over the same stretch of tidal channel where a Nortek AquaDopp profiler was deployed on the chan- 13 Figure 1.2: Beam orientations relative to coordinate system in the x− z plane of the Nortek AquaDopp acoustic velocity profiler. Note x is positive in the down- stream, outgoing tide direction, and h denotes water depth. Inset formula shows calculation of streamwise velocity U from the oblique beam velocities in the j-th vertical bin. nel bottom, collecting “water-truth” velocity measurements to compare with the remotely-sensed surface velocities measured from the drone. The AquaDopp con- tinuously recorded streamwise (along-channel) u-velocities (in the x-direction) at 1 Hz in vertical bins of ∆z=1.5 cm. The instrument was fitted with a head consist- ing of three beams aligned in the same plane in the along-channel direction, with beams 1 and 3 diverging from the vertical (beam 2) by 60◦ in opposite streamwise (x) directions as seen in Figure 1.2. The vertical profile of the x and z directed components of the velocity vector (u and w, respectively) is measured by assuming horizontal homogeneity (i.e., stream- wise homogeneity) and using the beam geometry to calculate the two components of the velocity field in each range-gated bin elevation. The bin height was set at 15 mm and the data recorded in beam coordinates for post-processing. The water depth h in the tidal channel, which fluctuated with the tidal stage, was identifi- 14 able via a high amplitude response in B2, since the free surface strongly reflects the acoustic signal from the AquaDopp back towards the instrument. Therefore, the free surface at each timestep was defined as the highest amplitude response bin in the vertical beam (B2). Using the geometry of the AquaDopp head, the corresponding surface bin in the oblique beams (B1 and B3) and the physical dis- tance between the beams at the water surface was identified. The AquaDopp data were processed in 12-minute intervals, coinciding with each flight hover time ± a one-minute buffer. AquaDopp data collected on May 23 before and during high tide, as shown in Figure 1.6, exhibited phase-wrapping in beam coordinates, meaning B1 and B3 would suddenly change sign for brief periods. Phase-wrapping is a somewhat com- mon issue with acoustic velocity profilers, for example if the beam pulse timing interval is too short, and beam data were phase-unwrapped by manually correcting these sign changes in beam coordinates. Phase-unwrapped beam data were then processed with an Adaptive Gaussian Window (AGW) filter [Cowen and Moni- smith, 1997, Cowen, 2006], which retained at least 95% of data in both oblique beams across all flights. Then u and w velocities were calculated from B1 and B3 according to the equation shown in Figure 1.2. Velocity data collection on May 24 occurred during a falling tide. These data exhibited a 12.2% weaker average B2 amplitude response and more consistent fluctuations in the beam signal than on May 23, likley a result of smaller or fewer scattering particles suspended in the water column, but no phase-wrapping. On May 24 there were obvious non-physical deviations in beam velocity magnitude between B1 and B3, such as B1 velocity jumping by 50-100% relative to B3 or vice-versa, likely a result of the lower beam amplitudes and hence weaker signal-to-noise ratio. Threshold velocity filtering was implemented in beam coordinates to remove these spurious velocities, followed by 15 processing with the AGW filter. This ensured that the AGW filter could accu- rately and robustly estimate the standard deviation of the signal of interest before the u and w velocities were calculated. Remote measurements of free surface velocity can be used to estimate the depth-averaged channel velocity using the velocity index k = Ub/Usurf , the ratio of depth-averaged to surface velocity. The velocity index is an important component of calculating volumetric discharge from surface velocity measurements [Levesque and Oberg, 2012]. This ratio was derived using experimental data from [Hulsing, 1967] by [Rantz, 1982], who determined that a velocity index of k=0.85 for natural channels produced errors under ±5 % of volumetric discharge. Due to the common use of the velocity index to make inferences about bulk flow from remote free surface velocity measurements [e.g., Johnson and Cowen, 2017, Kim et al., 2008, Puleo et al., 2012], our study inverted the velocity index and applied it to the depth-averaged, time-dependent vertical profile of streamwise velocity measured by the AquaDopp, to derive a surface-extrapolated velocity (Usurfex) based on this in-channel data: Usurfex = < U(t, z) > k (1.1) This relationship reflects the depth-averaged velocity estimates that would be made in a study using only remote surface velocimetry measurements with a velocity index, i.e. that the depth-averaged velocity is 85% of the surface velocity. Equation 1.1 also provides a more robust estimate of the idealized surface velocity in the absence of wind than AquaDopp velocity measurements in the bin nearest the surface, which often suffers from noise and dropouts in the data collected for this study. 16 Table 1.1: Micasense Altum infrared microbolometer band specifications. Resolution 160×120 pixels Field of view 57◦×44.3◦ Thermal sensitivity <50 mK Thermal accuracy +/-5 K Wavelength range 8-14 µm 1.2.3 Drone-based image collection The drone used for this experiment was a DJI Matrice 600 Pro that hovered at approximately 30 m over the tidal channel. The drone was fitted with a Gremsy S1 gimbal carrying a Micasense Altum six-band imager, which includes a thermal infrared microbolometer sensor (specifications in Table 1.1). Five repeated drone flights were launched from the end of the access road just west of the weather station adjacent to the combined channel of Franklin and Santa Monica Creeks shown in Figure 1.1. Imagery of tidal flow was collected from the hovering drone for approximately 10 minutes per flight at this point along the channel. The spatial coverage of the drone field of view relative to the in-channel AquaDopp is shown in Figure 1.3. The Micasense imageprocessing toolbox in Python was used to align the six imagery bands, correct for lens distortion, and upsample the lower-resolution thermal band to the resolution of the visible bands (2064x1544 pixels) [McAllister et al., 2024]. As shown in Table 1.1, the Micasense Altum thermal band has a relatively small number of pixels, leading to a very coarse spatial resolution (19.6 cm per pixel on average during this experiment). However, patterns detected on the wa- ter surface visible in the thermal infrared drone images taken during each flight still presented a trackable signal for the STIV pattern tracking algorithm used for image velocimetry (Section 2.5). Since STIV only requires relative differences in intensity or value between pixels in any type of imagery to produce trackable pat- 17 Figure 1.3: Experimental schematic showing orientation and surface extent of AquaDopp measurement beams in blue relative to drone imagery FOV in red. Actual drone FOV included portions of both channel banks for each flight. Figure not to scale. terns, thermal infrared pixel intensities were not converted to temperature in this experiment. STIV was conducted for each flight using a subset of each thermal in- frared image in the flight time series that was located directly above the in-channel AquaDopp, thereby measuring water velocity in the same area of the channel us- ing both the drone-based and in-channel methods to enable direct comparisons between the two. To create the unified x-axis aligning pixel columns in the drone images with the AquaDopp’s planar beam orientation shown in Figure 1.3, the location and orientation of the AquaDopp were identified in an RGB image assembled from the aligned bands of the the first stationary image in each flight. This image was rotated by an angle which aligned the vertical pixel columns in the image with the AquaDopp x-axis, and all subsequent images in each flight’s time series were rotated by this angle. The location of the AquaDopp head in the rotated image was set as the x coordinate origin for each flight, as shown in Figure 1.4. 18 X - + N 10 m Figure 1.4: Three-band RGB drone image collected during flight C with the x- axis located directly over the AquaDopp velocity measurement head. The positive x-direction denotes tidal outflow, while x < 0 denotes tidal inflow. The translu- cent yellow region shows the location of u-velocity measurements along the x-axis: both the image pixel coordinates sampled in STIV and the surface extent between AquaDopp beams 1 and 3. Images were rotated so that the AquaDopp x-axis was parallel to pixel columns in images. Ground control points visible during this flight are circled in red. True north relative to the rotated image is shown at top left, with the spatial scale at bottom right. An orange recovery float is visible to the right of x-axis. 19 1.2.4 Georeferencing and stabilization of drone images Five GCPs consisting of rocks wrapped in aluminum foil for thermal band visibil- ity were placed along the banks of the tidal channel at CSMR, and their center positions were averaged over 10 minutes per GCP with an InertialSense GPS. To georeference the aligned and rotated images, the GCPs visible in each flight were identified and the ratio of the distance between GCPs in pixel space (from the first stationary image in the flight) and physical space (from the GPS measurements) was averaged for all GCP pairs to derive a pixels-per-meter ratio for each flight, which was assumed constant across the image due to the nadir viewing angle. The location of the aligned AquaDopp and STIV x-axis was stabilized from image to image by tracking the displacement of a GCP from image to image caused by drone motion during the hover. The x coordinate origin overlying the AquaDopp was corrected for each image using the displacement of the GCP from the previous frame in the hover. Therefore, the STIV coordinate origin location was maintained over the AquaDopp head despite any drone movement. The performance of this GCP displacement method to correct for drone motion was compared to a method using the GPS data recorded when each photo was taken [Pirelli, 2019], shown in Figure 1.5. The GPS method estimates the displacement of the four image corners in physical space due to drone motion between each image in the time series, and these four corner displacements were averaged to one displacement value at each timestep. The DJI drone’s standard internal GNSS unit was set to GPS lock mode for all flights, but no real-time kinematic (RTK) base connection was set up. GPS x, y accuracy ranged from 0.78-1.02 m over this flight, and this range is of the same order of magnitude as the random noise exhibited in the GPS signal in Figure 1.5. In the absence of RTK-level GPS accuracy during a drone flight, this comparison indicates that the GCP-based drone motion tracking 20 method used herein is potentially more robust to the effects of small drone motions than the GPS method, directly correcting space-time diagram origin coordinates based on the apparent motion of a GCP from image to image. This accuracy is reflected in the standard deviation of the drone displacements calculated by the GCP method (2.9 cm, or 1.8 pixels before conversion) compared with the much larger standard deviation of displacements from the GPS method (49.6 cm), suggesting that the GPS position is dominated by random error while the GCP displacement approach appears more reflective of the drone’s actual drift. 1.2.5 Space-time image velocimetry (STIV) We chose STIV for velocimetry analysis because (1) Microbolomter infrared mea- surements generally exhibit only modest signal-to-noise ratio when imaging water in the environment, and STIV is one of the most robust quantitative imaging ap- proaches to determining the mean velocity field from relatively noisy imagery [Lu et al., 2023, Torres-Rua, 2017]; (2) the AquaDopp’s acoustic measurement plane was oriented in an along-stream direction with the beam orientation denoted as the yellow line in Figure 1.4, and the one-dimensional STIV method most closely recreates this measurement axis from the drone images. The extent of the x-axis sampled from images in STIV was determined by calculating the physical distance between the diverging B1 and B3 AquaDopp beams at the water depth identified in the vertical B2 beam during each flight using the geometry of the AquaDopp head shown in Figure 1.2. Then this physical distance between the AquaDopp beams was converted to pixels using the pixels- per-meter ratio calculated for each flight using the GCPs. The length of the x-axis is therefore a function of water depth: when the water depth was greater, the oblique AquaDopp beams diverged farther before reaching the surface, and more 21 24 08:00 24 08:05 24 08:10 Time n May 24, 2022 −0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Di sp la ce m en t ( m ) [ m ea n su bt ra ct ed ] Dr ne displacement calculated by GPS Dr ne displacement calculated by GCPs Figure 1.5: Comparison of image stabilization methods for flight C. Each point shows the respective method’s estimated horizontal displacement of the drone from the previous image based on a stationary ground location (GCP or the previous image’s corner pixel locations) in meters. Displacements were calculated as dis- tance magnitude from the previous timestep without accounting for direction, and means were subtracted from each time series for comparison. 22 of the x-axis fell between the beams, so a larger set of pixels along the x-axis was sampled for STIV. The pixel extent of the x-axis was centered on the AquaDopp head coordinate x = 0, and this column of pixels was appended to the space-time diagram at each timestep from the same pixel location in the rotated and stabilized image time series. The mean pixel intensity was subtracted from each individual thermal image before the column of pixels along the STIV x-axis was added to the space- time diagram to correct for changes in the microbolometer camera’s temperature, which can cause changes to the global pixel intensity from image to image. The Micasense imageprocessing toolbox was used to import and align the six-band imagery [McAllister et al., 2024]. This workflow was designed to ensure the most precise possible comparison between remote and in-situ velocity measurements rather than to estimate a synoptic surface velocity field. The pixel intensity gradient across the space-time diagram was measured by integrating the two-dimensional autocorrelation in a polar coordinate system; the maximum of this integral (ϕ) represents the angle of highest autocorrelation, which can be converted to a velocity using the pixels per meter ratio and framerate. These gradient features’ angles represent velocities based on the units of the space-time diagram axes — distance traveled over a specific time: USTIV = Sx St tanϕ (1.2) where Sx is the physical length scale per pixel (1.6 cm/pix for flight C), St is the time scale per pixel (or framerate−1, 1.34 s/pix for flight C), and ϕ is the maximum of the two-dimensional autocorrelation integral [Fujita et al., 2019]. The tangent of the angle ϕ gives a distance to time ratio or velocity in pixels, which is converted to meters per second by multiplying by the distance and time scales 23 Sx/Sy. This process was repeated in 20 overlapping square sections of the space- time diagram, which were evenly distributed along the entire time axis to resolve changes in the STIV velocity measurement during each flight. The dimensions of these square sections were set by the x-axis pixel length scale for each flight’s space-time diagram, ensuring that velocity calculations were made along the same spatial scale as the in-channel Usurfex velocity measurements. This project’s implementation of STIV was designed to enable as direct a com- parison as possible with the in-situ AquaDopp measurements, rather than demon- strating an optimized example of this velocimetry method. To make synoptic surface velocity measurements, many STIV search lines would be sampled across images of the water surface, and each search line’s orientation could be optimized to lie parallel to the local flow direction [Han et al., 2021, Legleiter et al., 2024b]. In our project, the STIV x-axis orientation was simply set by the orientation of the AquaDopp’s x-axis in the tidal channel, and only one search line was sampled be- cause the primary objective was a localized comparison between the remote STIV and in-situ AquaDopp measurements. Several user-friendly applications and tool- boxes exist for surface velocimetry using drone imagery [e.g., Hydro-STIV used by Biggs et al., 2022, Legleiter and Kinzel, 2023], and our goal in this project was to assess the need to consider wind stress effects on the surface velocity fields produced by these existing software tools. 1.2.6 Tide and wind conditions CSMR maintains a meteorological station that measures tidal elevation and wind speed and direction every three seconds (0.3 Hz), recording the average, minimum, maximum and standard deviation of these metrics in 10-minute intervals. Tidal elevation is measured using a pressure transducer in the tidal channel bed adjacent 24 Table 1.2: Wind velocity and direction average and standard deviation (σ) values from the CSMR meteorological station during each flight. Flight Wind U10 (m/s) Wind σU10 (m/s) Wind angle from N Wind angle σ A 2.30 0.34 242◦ 7◦ B 1.53 0.31 225◦ 10◦ C 1.62 0.36 286◦ 17◦ D 1.80 0.33 227◦ 10◦ E 2.46 0.30 233◦ 9◦ and connected to the meteorological station tower. The tidal stage over the course of the experiment is shown in the top plot of Figure 1.6. Meteorological station wind data come from an anemometer positioned 9.75 m above ground level, with wind direction angles referenced to true north [UCNRS, 2022] and defined as the direction from which the wind originates. Average wind speeds were less than 5 m/s for all flights during the experiment, as shown in the bottom plot of Figure 1.6. The location of the meteorological station relative to the experiment location is shown in Figure 1.1. The drone was launched and operated within roughly 50 meters of the meteorological station during this study. This experiment was conducted on two mostly cloudy days at the start of coastal California’s summer low cloudiness season, which is caused by persistent temperature inversions associated with the season of maximum lower tropospheric stability [Clemesha et al., 2016]. All flights were conducted during sea breeze conditions, with average winds coming off the Santa Barbara Channel from an approximately southwest-west (225◦-270◦) direction (Table 1.2). Data from the meteorological station were used to calculate surface wind stress during each flight. Because the elevation of the meteorological station anemometer is within 0.25 m of 10.00 m from the ground, and the meteorological station is located on the bank of the tidal channel approximately one meter above the water 25 May 23, 12:00 May 23, 18:00 May 24, 00:00 May 24, 06:00 May 24, 12:00 2022 1.2 1.4 1.6 1.8 2 2.2 T id a l w a te r le v e l (m ) -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 T id a l w a te r v e lo c it y ( m /s ) A B C D E May 23, 12:00 May 23, 18:00 May 24, 00:00 May 24, 06:00 May 24, 12:00 Date/time 2022 0 0.5 1 1.5 2 2.5 3 3.5 1 0 -m in . a v g . w in d s p e e d ( m /s ) 0 60 120 180 240 300 360 1 0 -m in . a v g . w in d d ir e c ti o n ( °) A B C D E Figure 1.6: Tide and wind conditions during the experiment (conducted May 2022). Top: Relative tidal elevation (black solid) and 10-minute averaged surface- extrapolated tidal velocity from the deployed AquaDopp (Usurfex, red dashed) over two experiment days. Red dashed velocity zero-axis highlights the sign of tidal ve- locity denoting flow direction. Bottom: Average wind speed (black solid) and direction (red dashed) over the same time period. Timing of five 10-minute drone flights denoted by blue vertical lines on both plots, letters next to lines denote flight. 26 level (which fluctuates on the order of one meter with the tidal stage as shown in Figure 1.6), the meteorological wind speed was used directly as an approximation for U10. The wind drag coefficient at the water surface C10 was determined using a least square fit for U10 < 5 m/s from [Wüest and Lorke, 2003]: C10 = 0.0044U−1.15 10 (1.3) The parallel component of the wind stress, τ||, at the water surface relative to the x-axis was calculated using this drag coefficient and the anemometer wind speed estimate for U10: τ|| = ρC10U 2 10 cos θ (1.4) where ρ is the density of seawater (1026 kg/m3) and θ is the angle between the x-axis and the average wind direction. τ|| is positive in the same direction as the tidal flow and negative in the opposite direction. The orientation of the rotated aligned images relative to north and the wind direction was determined using the physical locations of GCPs visible in each set of rotated images. 1.3 Results Trackable thermal signatures were present in the images collected during each flight. Figure 1.7, a typical image, shows pixel intensity variations on the water surface that are advected along the channel from image to image in the time series. An animation showing the time series of thermal images used for pattern tracking is also available in the supporting information. 27 10 m Figure 1.7: Rotated infrared thermal image from flight C corresponding to the same timestep as the RGB image in Figure 1.4, upsampled to the resolution of the visible light bands. Pseudocolor scale shows differences in relative pixel intensity in the thermal infrared band, not converted to real temperature. Location of the x-axis sampled by STIV overlying the AquaDopp is shown in red. An animated version of this figure showing the stabilized x-axis location is available in the supporting information. 28 08 :00 :06 08 :00 :52 08 :01 :38 08 :02 :24 08 :03 :09 08 :03 :55 08 :04 :41 08 :05 :26 08 :06 :13 08 :06 :58 08 :07 :43 08 :08 :29 08 :09 :15 08 :10 :00 Time on Ma 24, 2022 0 0.317 0.633 0.95 1.27 1.58 1.9 2.22 2.53 2.85 Di st an ce a lo ng x -a xi s ( m ) Figure 1.8: Space-time diagram for flight C, assembled by sampling one spatial col- umn of pixels from the x-axis in each image during the 10-minute hover timeseries. 20 ϕ angles detected from the space-time diagram are shown at their centered loca- tions within the overlapping square windows of the diagram which were analyzed to derive each angle. For each flight, a space-time diagram was assembled by sampling pixels along the x-axis overlying the measurement beams of the AquaDopp, which were added to the diagram at each timestep. A sloping gradient was clearly visible across space- time diagrams for all five flights, as shown in Figure 1.8. 29 07:58 08:00 08:02 08:04 08:06 08:08 08:10 08:12 Time (HH:mm) May 24, 2022 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 U v e lo c it y ( m /s ) STIV velocity Running time-averaged surface-extrapolated velocity Figure 1.9: Surface-extrapolated velocity (Usurfex, blue line) from the AquaDopp running-averaged over the same duration as 20 flight C STIV velocity measure- ments (USTIV , red points) generated from the the 20 detected ϕ angles shown on the space-time diagram in Figure 1.8 according to Equation 1.2. 1.3.1 Uncertainty associated with time synchronization be- tween velocity measurements To assess the difference between the drone-based and in-channel velocity measure- ments during each flight, twenty overlapping STIV velocity measurements (USTIV ) from each flight’s space-time diagram were compared with twenty corresponding surface-extrapolated velocity (Usurfex) measurements from the AquaDopp averaged over the time duration represented by each STIV measurement. A time series of these two velocity signals is shown in Figure 1.9. The AquaDopp was synchronized to computer time during pre-deployment programming. The Altum camera’s internal clock was not synchronized to computer time on May 23. Hover period timestamps and image framerate during flights A and 30 B were derived from the drone’s internal GPS log by identifying the period of constant maximum drone elevation and near-zero velocity, and dividing the number of stationary images by this GPS time duration. Flight B’s GPS-derived start and end timestamps are within one to three seconds of hover times noted in the field using cell phone time. On May 24 a meteorological mast was mounted on the drone which triggered recording of GPS-derived accurate timestamps for each image collected during flights C, D and E. Image timestamps allowed for a direct comparison between individual STIV velocity measurements and corresponding periods of the AquaDopp surface-extrapolated velocity signal. Nortek, the AquaDopp manufacturer, defines the uncertainty as 1% of the measured velocity value ±0.5 cm/s for the AquaDopp HR model used in this study [NortekAS]. Based on the velocity range for flight C shown in Figure 1.9 this would yield a maximum uncertainty of ±1.49 cm/s. 1.3.2 Coordinate system and sign of velocity measurements For streamwise water velocities in this coordinate system, positive velocities denote outflow from the estuary, and negative velocities denote tidal inflow. The differ- ence in velocity U∆ between the STIV velocity measurement USTIV and surface- extrapolated AquaDopp velocity measurement Usurfex was calculated accordingly: U∆ = sign(U) ∗ |USTIV − Usurfex| (1.5) where sign(U) is the sign function: sign(U) =  1 if |USTIV | > |Usurfex| −1 if |USTIV | < |Usurfex| (1.6) USTIV and Usurfex had the same sign for all flights except for flight B: the surface-extrapolated AquaDopp velocity measurement direction had switched to 31 outgoing from flight A, but the STIV velocity measurement direction remained incoming, as shown in Table 1.3. For flight B, |USTIV | < |Usurfex|, and this flight had the largest velocity difference due to the difference in sign between the two velocities. The effect of wind stress parallel to the x-axis on the difference U∆ between STIV velocity measurements and surface-extrapolated AquaDopp velocity measurements is shown in Figure 1.10. Table 1.3: AquaDopp surface-extrapolated and drone-based STIV velocity mea- surements, and two metrics of wind effects on surface velocity: angle between wind direction and measured flow direction, and the wind stress component parallel to the flow direction along the x-axis during each flight. Note that the in-channel surface-extrapolated tidal velocity changed directions between flights A and B but the STIV velocity measurement direction remained constant. Flight AquaDopp U surfex (m/s) Drone USTIV (m/s) Wind angle from U surfex direction Parallel wind stress τ|| (Pa) A -0.56 -0.61 −42◦ 6.9 B 0.49 -0.12 123◦ -3.6 C 0.57 0.35 180◦ -6.8 D 0.66 0.29 124◦ -4.2 E 0.49 0.36 127◦ -5.9 1.4 Discussion Wind direction relative to the tidal flow direction played a clear role in the sign and magnitude of the drone-based STIV velocity measurements relative to the surface- extrapolated in-channel AquaDopp velocity measurements. Winds were approxi- mately out of the west (270◦) or southwest (225◦) off the Santa Barbara Channel during all flights (Table 1.2), roughly aligning with the direction of incoming tide in the channel where the experiment was conducted. During flight A when the par- allel wind stress (τ||) was aligned with the direction of the incoming tide, the mean 32 -8 -6 -4 -2 0 2 4 6 8 Parallel wind stress || (Pa) -0.8 -0.6 -0.4 -0.2 0 0.2 V e lo c it y d if fe re n c e U ( m /s ) A B C D E Figure 1.10: Parallel wind stress (τ||) effect on calculated difference (U∆) between STIV velocity measurement (USTIV ) and surface-extrapolated velocity measure- ment (Usurfex) for five flights. Individual boxplots summarize 20 time-synchronized comparisons between the two velocity measurements, showing median, first and third quartile box, largest non-outlier whiskers based on interquartile range, and any individual outliers. The vertical axis intercept U∆ = 0 denotes that the two velocity measurements were equal. 33 STIV velocity measurement was higher than the surface-extrapolated in-channel AquaDopp velocity measurement, as shown in Table 1.3. For all other flights, when the parallel wind stress was opposite to the tidal outflow, the mean STIV velocity measurement was lower than the surface-extrapolated velocity. In fact flight B, which occurred soon after the tidal flow had changed direction from incoming to outgoing as shown in Figure 1.6, actually appeared to still have an incoming tidal velocity direction in the STIV surface measurement. This difference in measured flow direction is an important result for remote measurements of surface velocity with relatively slow (<1 m/s) or unsteady flows when wind stress is a factor. As shown in Figure 1.10, the velocity differences (U∆) between the two measurement methods can be of the same order as the surface-extrapolated “water-truth” in- channel velocities measured by the AquaDopp (Usurfex), representing a significant potential source of error if the remote STIV velocity measurement was assumed to be indicative of the bulk flow velocity under these wind and flow conditions. 1.4.1 Physical causes of time-varying velocity differences The range of individual velocity measurement differences (U∆) across the five flights shown in Figure 1.10 are on the order of 5-50 cm/s, much larger than the AquaDopp’s maximum uncertainty for flight C of 1.5 cm/s. This range could be partially explained by differing timescales of unsteadiness between the the in- channel flow velocity and the wind speed and direction. The Usurfex timeseries in Figure 1.9 is running-averaged over a timescale greater than four minutes to match the time-averaging inherent to the STIV velocity measurement method, while the non-averaged Usurfex velocity signal fluctuates within the same or greater bounds on the order of seconds. Similarly, the met station wind data are 10-minute average values, and the wind speed and direction standard deviation (σ) values in Table 34 1.2 reflect the unsteadiness of surface wind stress during each flight. While local wind stress likely has some effect on water velocity throughout the water column, fluctuations in surface wind stress are decoupled from tidal flow dynamics, result- ing in changing velocity differences during a single flight. Wind duration and fetch, which varies greatly with wind direction at the study site, have a high impact on the wind-driven transport of water surfaces [Veron and Melville, 2001]. The 10- minute average wind data available for this study therefore limits what wind stress calculations are feasible over shorter timescales within a given 10-minute flight, such as estimating the wind-driven water surface velocity component. Similar to the decoupled unsteadiness of wind stress and tidal velocity during a single flight, wind and tidal flow directions also differ across diurnal timescales vis- ible between flights. For example, flight B has a significantly larger U∆ magnitude in Figure 1.10 than the four other flights due to the difference in direction between the two velocity measurements. This outlier could be explained by flight B’s timing relative to the tidal cycle and diurnal wind patterns typical of coastal regions. For much of the six hour period preceding flight B, the wind had been blowing roughly onshore at 2.5-3 m/s, while the tide was steadily rising (Figure 1.6), resulting in a tidal flow roughly aligned with the prevailing wind. This alignment between the afternoon sea breeze direction and rising tidal flow direction could have contributed to the high STIV velocity measurement magnitude for flight A relative to the four other flights shown in Table 1.3. When the tide crested between flights A and B, the depth-averaged flow direction reversed while the wind direction remained relatively constant, and this upstream surface momentum flux due to wind stress would have taken time to dissipate. It appears that during flight B, which was conducted less than an hour after high tide, the wind was still blowing the surficial skin layer of the water column upstream while the rest of the water column had 35 switched directions to an outgoing tide. Presumably, an overnight outgoing tide could be similarly accelerated by aligned wind stress of an offshore nighttime land breeze, highlighting the potential constructive or destructive relationship between wind direction and tidal velocity depending on how these two diurnal cycles align on a given day. This finding further emphasizes the complex interacting processes that modulate surface wind stress effects in the intertidal zone, as past studies have shown that expected air to sea momentum fluxes caused by surface wind stress in coastal areas can deviate significantly from open ocean parameterizations due to other nearshore processes such as horizontal current shear, depth-limited wave breaking, and decreased wave celerity in shallow water [e.g., Ortiz-Suslow et al., 2015]. 1.4.2 Measurement uncertainty and data quality Tradeoffs between the spatial and temporal scales of drone imagery used for STIV analysis shaped this study’s methods. Using flight C as a typical example, the characteristic length scale of the area shown in drone images during the hover was 23.7 m in the streamwise direction (along the short image axis) and 31.6 m in the cross-channel direction (along the long image axis), using the pixel to meter conversion derived from the GCP coordinates. As shown in Figure 1.7, the region of pixels sampled in STIV relative to the overall image dimensions is small, and a lower hover altitude would have enabled higher resolution space-time diagrams with a larger number of pixels in the direction of the flow and smaller pixel footprints. The conservative choice of a 30 m target hover altitude was based on uncertainty regarding the water depth and AquaDopp beam divergence distance, as well as a lack of a live feed from the Altum camera. The frame rate of the collected images was somewhat variable, as the Altum camera was set to record at 1 Hz 36 while its recommended maximum frame rate is 0.67 Hz. As a result, the imaging frequency was limited by the write speed of the camera hardware, with a typical framerate of approximately 0.75 Hz across the five flights. Therefore, a lower hover altitude could have introduced other measurement challenges by increasing the pixel length scale of apparent pattern motion between frames, demonstrating the tradeoff between spatial resolution and frame rate that bounded this experiment’s methods. The second major reason likely explaining the range in velocity differences within individual flights in Figure 1.10 is the angular sensitivity of the STIV method applied to data collected at a slower-than-optimal framerate. Due to the pixel length of the x-axis that overlaid the AquaDopp measurement beams at the surface (192 pixels for flight C), it was not feasible to resolve the two-dimensional autocorrelation’s polar integral in increments smaller than one degree, as some frac- tional increments of a degree did not intersect with any pixels in this relatively small grid. Despite the drone flying conservatively high, the upsampled thermal band pixel footprint was still a reasonably high-resolution 1.6 cm/pix. Combined with the the Altum’s sub-1 Hz framerate, this fine spatial scale yielded large apparent motion of thermal features along the x-axis from image to image. This produced steeper than desired pixel intensity gradients in the space-time diagrams, resulting in at least one polar integral maxima (ϕ) at the steepest possible angle (±89◦) in flights A, C and E. Angles this steep are less than ideal for STIV velocity mea- surements due to the sensitivity of the tangent function at near-right angles, since tan±90◦ is asymptotic to infinity, and STIV’s velocity estimation error is lowest for angles ϕ = 45◦ [Fujita et al., 2019, Legleiter et al., 2024b]. In Figure 1.9, the four STIV velocity measurement increments correspond to ϕ=86-89◦, highlighting this method’s lack of precision at steep angles. However, the trend described by 37 the individual STIV velocity measurements does follow the surface-extrapolated velocity with peaks and valleys in the same parts of the flight C time series, and the same was true for flight E within the same 4◦ range of ϕ angles. Flight A in particular would have benefited from a faster framerate, as ϕ was ±90◦ for all STIV measurement windows, with the second highest integral maxima at -89◦. Therefore, the underestimate ϕ=-89◦ was assumed for all STIV velocity measurements. However, this underestimate still produced a larger USTIV velocity than the contemporaneous Usurfex velocity, resulting in a positive U∆ value for flight A despite this flight occurring during a strong peak in the Usurfex incoming tidal velocity (Figure 1.6). The actual STIV velocity measurement magnitude was almost certainly larger than the -0.61m/s measurement at this framerate (Table 1.3). A more accurate USTIV measurement would likely make U∆ more positive and push the spread of velocity comparisons for flight A higher in Figure 1.10, further strengthening the positive relationship between τ|| and U∆ visible in this figure. Therefore, the STIV velocity measurements for flight A exhibit a higher uncertainty than other flights, but the clear signs of negative directional bias in this flight’s STIV analysis maintain a strong causal relationship between wind stress and apparent surface flow velocity based on the overall results of this study. Differences in data quality were apparent across the tidal cycle for both drone images and AquaDopp velocity measurement data. In terms of thermal resolution of trackable features in the imagery, differences in image quality between flights are influenced by a number of environmental factors affecting the imaging sensor such as air temperature and wind [Legleiter et al., 2024a, Lewis et al., 2018], as well as the thermal signal on the water surface. For example, the scale of turbulent thermal structures can vary significantly across the tidal cycle [Zappa et al., 2003], which could present a challenge in successfully implementing this 38 method in an unbiased and consistent way at any tidal stage. Differences in infrared signal-to-noise ratio could also have contributed to spurious peaks in the two- dimensional autocorrelation polar integral for some flights’ space-time diagrams, despite normalizing the integral by the number of pixels within each degree bin. The integral maxima (ϕ) for flights D and E were largely biased to pixel rows and columns due to apparent fixed pattern noise, with the resulting 0◦ or 90◦ angles representing non-physical velocities of zero or infinity. This bias was eliminated by constraining the valid range of ϕ angles for flights D and E to ±1-89◦, and the resulting ϕ time series fluctuated between 84-89◦, similar to the range in flights B and C which did not require such bias correction. In summary, there were two sources of right-angle bias across three flights, with flight A requiring bias correction due to an insufficient framerate for the velocities measured, and flights D and E requiring bias correction due to fixed pattern noise in the space-time diagram. Changes in acoustic velocity profiler echo intensity between ebb and flood tide have been observed in other studies due to differences in passive scalar concentration [Dugan and Piotrowski, 2012]. Differences in the intensity of the AquaDopp signal response were apparent between the two deployments in this experiment which occurred at different tidal stages, with an average normalized amplitude response of 55.6% on May 23 during the rising and cresting tide, and an average response of 43.4% on May 24 during the falling tide. This lower amplitude response on day two could partially explain observed relative noisiness of May 24th’s AquaDopp data in this study. 39 1.4.3 Implications for image-based surface velocity mea- surements Measurements near the water surface using in-situ instruments, such as acoustic velocity profilers, are challenging. Acoustic velocity measurements near the free surface often exhibit a reflectance band of acoustic signal interference which can lead to less accurate measurements [González et al., 1996], and placing instruments in the channel also disturbs the flow being measured. Remote sensing velocimetry techniques, such as the drone-based STIV method presented here, keep people out of harm’s way and can produce distributed surface velocity measurements over a much wider area than acoustic current profilers. However, if the surface veloc- ity signal being measured is significantly contaminated by wind stress, incorrect conclusions might be drawn. For example, applying a velocity index of 0.85 to the USTIV velocity measured from the drone during flight B would yield a depth- averaged velocity of -0.11 m/s, denoting an incoming tide. However, in the tidal elevation in Figure 1.6 as well as the sign of the AquaDopp Usurfex velocity mea- surement relative to flight A, it is clear that flight B occurred after the tidal flow direction had turned to outgoing. Therefore, one could easily draw the wrong conclusion from flight B’s STIV results alone, and this appears to be due to the surface wind stress in the incoming tide direction. Furthermore, the substantial spread of individual U∆ values within multiple flights in Figure 1.10 shows that decoupled surface wind stress fluctuations and bulk velocity fluctuations present another potential source of error in estimating bulk velocity from surface velocity. Importantly, across multiple flights the velocity differences (U∆) between the two measurement methods are of the same order as the surface-extrapolated “water- truth” in-channel velocities measured by the AquaDopp (Usurfex), and in one of five test flights the measured direction of the compared velocities was opposite. These 40 results demonstrate that surface wind stress relative to the magnitude of bulk flow velocity is an important factor in accurately estimating bulk velocity and discharge from remote surface water velocity measurements. This finding is a reminder that differences between remote sensing velocity measurements and in-channel acous- tic velocity measurements reflect the different physical processes observed by each method; acknowledging this difference is critical, as remote sensing techniques are increasingly used to infer in-situ velocities. Ongoing technological advances such as drone-mounted wind speed anemometers and growing access to velocime- try software tools present additional methods to further explore the focus of this study. Additional studies of wind stress effects on surface velocimetry applications are necessary to develop methods that account and correct for wind stress when making remote measurements of water velocity in the environment. 1.5 Acronyms • ADCP: Acoustic Doppler current profiler • PIV: Particle image velocimetry • FOV: Field of view • GCP: Ground control point • STIV: Space-time image velocimetry • IR-QIV: Infrared quantitative image velocimetry • CSMR: Carpinteria salt marsh reserve • AGW: Adaptive Gaussian window • RGB: Red-green-blue • RTK: Real-time kinematic 41 1.6 Open Research Publicly-accessible data used in this project such as UC Natural Reserve System meteorological station data [UCNRS, 2022] and USGS stream gauge data [USGS, 2022a] are cited in the bibliography, along with published software implemented in analysis [Cowen, 2006, McAllister et al., 2024, Pirelli, 2019]. Code and data needed to reproduce results from this study can be found in a dedicated Cornell University Library eCommons Repository [Heberlein et al., 2025]. 1.7 Acknowledgments This work was performed (in part) at the University of California Natural Reserve System Carpinteria Salt Marsh Reserve [UCNRS, 2023], and we are grateful to CSMR Director Dr. Andrew Brooks for his support of this project. We grate- fully acknowledge the support of the Cornell Atkinson Center for Sustainability, the United States Geological Survey (G23AP00025-00) and Cornell University in supporting Mr. Heberlein, Dr. Schweitzer and Prof. Cowen. Grants from the Zegar Family Foundation (SB200109, SB220237) are gratefully acknowledged as a source of support for Dr. Morgan and Prof. Caylor. 42 CHAPTER 2 DEVELOPING A STAGE-DISCHARGE RATING CURVE FOR AN IMPOUNDED ESTUARY USING THERMAL INFRARED SURFACE VELOCIMETRY 2.1 Introduction Impoundments that limit or exclude tidal flows into coastal marshes have been created around the world for centuries, with purposes ranging from historic land reclamation for agriculture and flood prevention [Segeren, 1983], to 20th-century transportation infrastructure, insect control and habitat modification [Portnoy, 1999]. While there is likely no precise figure for the extent of impounded coastal wetlands globally, estimates for the percentage of wetlands with some form of tidal restriction along the Atlantic coast of the United States range from 11% to >50% [Barbier et al., 2011, Kroeger et al., 2017, Montague et al., 1987, Soukup and Port- noy, 1986]. Tidal impoundment in salt marshes has severe hydrological, ecological and biogeochemical impacts on these ecosystems. Hydrologically, impoundment effectively reduces the tidal prism or volume of water that comes in and out of the salt marsh with the daily tide – the closer an impoundment is to the mouth of the estuary and the smaller the opening in the impoundment, the smaller the tidal prism and associated fluxes of salinity and sediment from the ocean into the estuary [Kidd et al., 2015]. A reduced tidal prism has many biogeochemical ef- fects, including pore space collapse and marsh surface elevation subsidence caused by the oxidation of drained sediments due to decreased tidal inundation [Port- noy and Giblin, 1997], which in turn can mobilize acidic compounds from these soils and drastically lower estuarine pH [Soukup and Portnoy, 1986]. Oxidation of these sediments, combined with reduced input of sulfate-rich seawater, can shift 43 the carbon balance of these ecosystems from a sink in a healthy salt marsh to a methane source when impounded, as methanogens outcompete sulfate reduc- ers under sulfate-poor, oxic conditions [Sanders-DeMott et al., 2022]. Decreased salinity also changes flocculation dynamics and associated sediment deposition pat- terns within the estuary [Abolfazli and Strom, 2023]. Ecologically, salt marshes are susceptible to invasion by Phragmites australis and other plant species that outcompete native salt marsh vegetation when salinity is artificially lowered via impoundment [Roman et al., 1984]. Impoundment often degrades fish habitat by reducing tidal circulation which leads to oxygen depletion, but ecological effects can vary widely based on how the impoundment alters salinity and water depth relative to pre-impoundment conditions [Montague et al., 1987]. Hydrodynamic- ecological feedbacks have been documented in salt marshes, with small changes in flow regime or species distribution producing nonlinear responses [Neumeier and Amos, 2006, Valiela and Fox, 2008]. Restoration of impounded salt marshes can therefore convey many ecological benefits, including improved bird and juvenile fish habitat [Gregory Shriver et al., 2004, Teo and Able, 2003], coastal resilience via wave attenuation and shoreline stabilization by native vegetation [Shepard et al., 2011], and increased carbon se- questration in inundated salt marsh sediments [Kroeger et al., 2017, Settelmyer et al., 2019]. In recent decades, legislation requiring removal of small impound- ments to improve endangered fish habitat has restored habitat connectivity and unrestricted tidal exchange in many small coastal watersheds [Roegner et al., 2010]. While many tidal barrier removal projects have been completed, tidal reintroduc- tion creates profound changes in the artificial freshwater wetlands that have typ- ically developed behind salt marsh impoundments. Native salt marsh vegetation can become reestablished quickly after tidal restoration [Karberg et al., 2018], but 44 barrier removal also often results in the mortality of freshwater plant species estab- lished in the artificially-fresh areas behind the barrier [Wang et al., 2022], which can sometimes require additional landscape-scale ecological restoration activities [Smith, 2007]. The effect of high tides and sea level rise on restored marshes must also be accounted for and mitigated if necessary, and hydraulic structures designed to dynamically limit tidal inundation are perhaps ironically an important compo- nent of some salt marsh restoration projects [Masselink et al., 2017, Rankin et al., 2023]. Hydrodynamic changes in a salt marsh estuary’s connection to the ocean are the causal mechanism of ecological degradation via tidal impoundment and restoration via barrier removal. Therefore, understanding the hydrodynamics of flow through tidal impoundments is an essential component of quantifying and restoring a degraded salt marsh estuary’s former tidal regime. 2.2 Methods The primary objective of this project was to quantify the pre-restoration baseline tidal prism of a diked estuary at Cape Cod National Seashore (Massachusetts, USA) by developing a stage-discharge rating curve for the tidal impoundment. A rating curve is a graph that relates volumetric discharge to a range of water levels, or stages, at a specific location along a channel. This relationship allows for the estimation of channel discharge based solely on stage measurements. Rating curves developed at gauging stations that continuously monitor river stage have been used for over a century by hydrological research organizations, such as the U.S. Geological Survey, to derive continuous records of river discharge from stage data [McCallum et al., 2025]. Stage-discharge rating curves assume a stationary, monotonic relationship be- tween stage and discharge - each time a river reaches a given stage, discharge will 45 be the same, and as stage increases, discharge increases. Both of these assumptions are violated in tidal systems with bidirectional flow, where a given stage can pro- duce zero discharge one day (such as at the peak of a high tide) and both upstream or downstream flows the next day (preceding or following a higher high tide). Stan- dardized methods to measure river discharge in tidally-affected reaches typically rely on integrated velocity measurements at gauging stations to derive discharge estimates [Ruhl and Simpson, 2005], because flow direction and magnitude cannot be uniquely inferred by stage alone in tidal systems. These methods use a velocity index value or function to estimate the bulk velocity of the channel from the con- tinuous velocity measurement at the gauging station [Johnson and Cowen, 2017]. However, it is possible to resolve discharge in tidal estuaries without velocity data by separating the non-stationary river inputs from the stationary effects of tidal harmonic constituents [Jay and Kukulka, 2003]. More recent studies using this approach have estimated ungauged river discharge based on tidal harmonic signal perturbations in historic tide gauge data [Moftakhari et al., 2013] and calculated discharge from multiple water level time series distributed along a tidally-affected river using wavelet analysis [Lee et al., 2021, Moftakhari et al., 2016]. Other re- cent publications have leveraged information embedded in the stage time series, such as the stage rate of change or nonlinear dynamics over time, to determine the discharge magnitude and direction [Jones et al., 2019, Kearney et al., 2017]. The presence of tidal impoundments in estuaries further complicates the cal- culation of discharge by creating a phase lag in the tidal signal upstream of the impoundment [Bjerklie et al., 2013]. While tidal influence in most dammed river systems does not extend upstream of the lowest dam, flood and drainage control infrastructure closer to the mouth of estuaries can dramatically alter tidal circu- lation patterns in these ecosystems, causing significant changes in water levels, 46 salinity, sediment transport, and habitat distribution [Ysebaert et al., 2016]. De- veloping measurement methodologies for tidal discharge in impounded estuaries is an important step toward understanding the hydrology and hydrodynamics of these systems and optimizing infrastructure design and operation, particularly in light of sea level rise and increasing anthropogenic impacts in estuaries globally [Kennish, 2019]. 2.2.1 Study site The Herring River is a second-order tidal watershed that covers approximately 19 km2, lying primarily within Cape Cod National Seashore in Massachusetts, USA [Befus et al., 2023]. This system has been impounded at its mouth since 1909 by Chequessett Neck Road Dike. Since multiple widespread fish kills in the river during the 1980’s, extensive ecological study and restoration planning by the National Park Service (NPS) and partners have culminated with an extensive tidal restoration project that is currently being implemented [Roman et al., 1995, Smith et al., 2020, Soukup and Portnoy, 1986]. This restoration project, and the decades of extensive data collection that preceded it, present an excellent opportunity to develop measurement techniques and experimental frameworks for studying the hydrodynamic and biogeochemical effects of salt marsh impoundments. Impoundment in the Herring River Estuary (HRE) has had widespread negative ecological impacts over the last century. Former natural shellfish beds within the estuary saw significant decline after impoundment, and the accumulation of fecal coliform due to reduced tidal flushing and degraded water quality in the HRE has necessitated regular seasonal closures of lucrative commercial oyster aquaculture operations in downstreamWellfleet Bay [Portnoy and Allen, 2006]. Marsh platform elevation upstream of the dike has subsided by approximately 1 m relative to 47 unaltered salt marshes below the dike [Portnoy and Giblin, 1997], due in part to decreased sediment deposition on the marsh platform during high tides [Portnoy, 1999]. Former salt marsh areas have also been extensively invaded by Phragmites australis, with much higher methane emissions from these areas than from un- impounded reference sites [Sanders-DeMott et al., 2022]. Fish populations also have declined significantly upstream of the dike relative to Wellfleet Bay, due to decreased tidal range and acidic, low-oxygen conditions caused by impoundment [Roman et al., 1995]. Chequessett Neck Road Dike is an earthen dam structure spanning approxi- mately 150 m across the mouth of the river, with three culverts located in the center allowing limited tidal exchange. These culverts each measure 1.8-2.2 m in width and 1.5 m in internal height (Figure 2.2), and on the bay side of the dike these openings can become completely submerged during the highest tides. The center and northwest culverts are fitted with hanging flapper gates which block almost all incoming tidal flow, only allowing outflow. The third culvert on the southeast side of the dike has an internal sluicegate stuck partially open at a height of 48.5 cm [Befus et al., 2023], so all incoming tidal flow passes through a cross-sectional area of approximately 1 m2 [Portnoy and Allen, 2006]. This structure only allows tidal inflows to reach approximately 1 km upstream of the dike, and results in tidal asymmetry where incoming flood tides last approximately two hours longer than outgoing ebb tides [Huntington et al., 2021]. In its impounded state, the HRE was characterized by very small river discharge relative to the tidal prism and a very narrow impoundment opening relative to width of estuary - both ratios of an order between 1:10 and 1:100. These characteristics severely decreased the tidal range on the upstream side of the dike, creating conditions of very low velocity across the HRE’s tidal reaches with rapid acceleration to high velocities through 48 the dike culverts. Because tidal forcing was at least an order of magnitude greater than river inflow to the HRE, the elevation head across the dike generated by tidal phase lag was hypothesized to be the primary driver of discharge. Cape Cod lies at the southern end of the Gulf of Maine, parts of which exhibit a macrotidal range; however, most of Cape Cod has a tidal range between 2-4 m, which indicates a mesotidal regime [Hammar-Klose et al., 2003]. Average tidal range at the mouth of the HRE is approximately 2.1 m on the bay side and 0.5 m on the river side of the dike [Portnoy and Allen, 2006]. During our study, bay-side tidal range greater than 3 m was measured during the spring tides observed. Figure 2.1: Aerial view of Chequessett Neck Road Dike during a flood tide, an- notated with red arrows showing the flow direction allowed by the three culverts. Note the aerated flow ejected from the one bidirectional culvert. Image credit: Merrily Cassidy, Cape Cod Times [Bragg, 2014] 49 Past efforts to measure Herring River Estuary flow rate From 2015-2017, the U.S. Geological Survey (USGS) undertook a campaign to mea- sure volumetric tidal discharge and associated loading rates for nutrient species in the Herring River. This field study adopted an index velocity approach to measur- ing the complex volumetric discharge through the dike, which acts as a funnel that rapidly accelerates flow velocities as water approaches the structure on the ebb or flood tide. USGS measured cross-channel average bulk velocity throughout the water column using an Acoustic Doppler Current Profiler (ADCP) along transects on both sides of the dike. This bulk velocity metric was compared to a surface index velocity measurement made near the dike culverts on the river side using a radar sensor, which relies on signal reflections off of waves traveling at the water surface velocity. A regression was derived to predict the bulk ADCP velocity mea- surement from the surface index velocity measured by the radar. However, this study did not publish a stage-discharge rating curve for the dike due to measure- ment challenges, including high flood tide velocities > 5m/s on the river side of the dike which disrupted the radar’s surface velocity signal, and shifting cross-sectional velocity profiles due to the convergent flow [Huntington et al., 2021]. This study did record daily mean discharge measurements from this two year period, yielding an average flow rate of 1.95 m3/s during ebb tide and -1.89 m3/s during flood tide over the study; 11.9% of days during the study did not list flow rates due to data dropout periods longer than two hours during measurement [Huntington and Spaetzel, 2020]. 2.2.2 Field campaign This field campaign in the Herring River Estuary took place in July 2022, coin- ciding with a projected high tide of 3.8 m on July 15, with a following low tide 50 of -0.4 m. This campaign was timed with a spring tide to observe flow through Chequessett Neck Road Dike under as large a range of tidal elevations as possible, producing the widest range of stages for the stage-discharge rating curve. Our pri- mary velocity measurement method was to collect a continuous thermal infrared imagery time series of flow through the dike for analysis in a surface velocimetry workflow described below. Remote velocimetry was the method chosen for velocity measurements due to the capability of such methods to resolve synoptic velocities throughout the camera’s field of view (FOV), and further motivated by the chal- lenging hydrodynamic measurement conditions documented during the previous USGS campaign [Huntington et al., 2021]. Image collection We used a FLIR SC8300 cooled thermal infrared camera for IR-QIV applications. This camera detects mid-wave infrared radiation (3-5 µm) using a 1.05-megapixel (1344x784) indium-antimonide (InSb) sensor, which is cooled to cryogenic temper- atures (approximately 80°K) to reduce thermal noise. The FLIR camera used a 17 mm focal-length lens and is mounted inside a fan-cooled weatherproof metal enclosure when deployed for extended periods of time in field conditions. Image acquisition is controlled by a field computer which sets camera parameters through FLIR software and records image files to external storage. The camera’s mounting orientation should be as close to a nadir (straight down or orthogonal) orientation as possible relative to the water’s free surface to reduce perspective distortion, but this ideal orientation is not typically feasible in field settings, so images are often recorded from a fixed oblique angle. We mounted the FLIR camera and enclosure using an adjustable bracket on support poles for a tide gauge and solar panel located on either side of the dike above the tide gate openings. The camera was mounted as high as possible on 51 each pole such that the field of view (FOV) could observe the widest area of the water surface across the mouth of the flowing culverts on either side of the dike (Figure 2.2). The camera was controlled using a field computer, and the computer time was synchronized immediately before the field campaign, resulting in accurate image timestamps. 3.0m 0. 49 m Internal fixed sluicegate* Bidirectional culvert section view Herring River Wellfleet Bay .3 m Channel bed 3. 0m 1. 5m Culvert opening Channel bed Length = 20.4m, bed slope = 0.0067 (z = 15cm) 1. 8m 2.4m 1. 5m Culvert opening *both one-directional (outflow only) flapper gate culverts have the same geometry and gate placement within culvert Figure 2.2: Section view of bidirectional southeast culvert within Chequessett Neck Road Dike illustrating important flow control structures. Figure not to scale. Georeferencing data collection The camera FOV included portions of the dike structure above and adjacent to the three culverts through which tidal flow occurred, and ground control points (GCPs) were placed throughout this portion of the field of view and their physical location was recorded. GCPs consisted of 30 cm2 plywood square targets wrapped 52 in aluminum foil, a low emissivity material that is clearly visible in the thermal infrared band. We placed a high emissivity tape on the center of these targets to maximize contrast and provide a point for georeferencing each target. Physical locations of GCPs were initially surveyed in using an Inertial Sense GPS antenna connected to a Microsoft Surface tablet with a real-time kinematic (RTK) connec- tion to the nearest MassDOT RTK base station in Truro. We recorded the GPS signal for 10 minutes at each GCP location to observe potential variability in the survey location and the consistency of the RTK correction signal. Additionally, a board with foil tape intervals and the GPS antenna was extended from the dike farther into the center of the camera field of view, and the GPS location of the board was recorded for 10 minutes in each orientation in which it was deployed. These additional temporary GCP locations overlying the water surface were in- tended to reduce error in reprojecting IR-QIV results to physical coordinates by creating a more uniform distribution of GCPs throughout the images on either side of the dike. Initial efforts to georectify the infrared images collected on either side of the dike using this RTK GPS survey data were unsuccessful due to considerable ver- tical uncertainty in the survey data, likely caused by poor hotspot reception and subsequent dropouts in the RTK base station connection. In May 2023 during a follow-up site visit, bolts on the fence above the dike culverts, which were visible in the original images, were located and surveyed using a Leica Total Station. Geo- rectification of both the river- and bay-side images was successful by using these fence survey locations as GCPs, with the addition of one board orientation from the original GPS survey. An important component of deriving accurate measure- ments from remote surface velocimetry is measuring the water surface elevation relative to the camera. We used a thermistor with a pressure transducer to log 53 water depth during image collection, and measured its vertical distance from the camera on either side of the dike. This pressure data was corrected for atmospheric pressure using meteorological data from a nearby USGS eddy covariance flux tower [Sanders-Demott et al., 2022]. USGS operated two tide gauges recording water level on either side of the dike from 2017 to 2022 [USGS, 2022b]. The river gauge location was on a piling in the Herring River approximately 60 m upstream of the dike, and the bay gauge was on the pole used to mount the camera directly above the outlet of the southeast bi- directional culvert. These gauges recorded water level in synchronized five minute intervals, and were surveyed to the same datum, with the difference between the gauges at each measurement time (hz = hriver − hbay) denoting the elevation head across the dike (Figure 2.3). Positive hz (hriver > hbay) is associated with positive discharge from the estuary, while negative hz (hriver < hbay) denotes negative discharge or tidal inflow to the estuary. The offset between this tide gauge datum and the thermistor’s pressure transducer was calculated by matching the tidal signals, so that the the same tide gauge water surface elevation could be used for both georeferencing and to calculate discharge from the resulting rating curves. Shifting the source of water surface elevation data from the thermistor pressure transducer, which was deployed only during our field campaign, to the USGS tide gauges allows simple extrapolation of flow rate from our rating curve for the five year period preceding dike reconstruction, and also allows the calculation of real- time elevation head across the dike which is hypothesized to primarily control the flow rate. 54 Figure 2.3: USGS tide gauge elevation on either side of the dike and calculated elevation head across the dike during the field experiment (July 13-15, 2022). 2.2.3 Image Analysis Infrared quantitative image velocimetry (IR-QIV) uses thermal infrared images to detect naturally-occurring temperature variations on the surface of flowing water, providing a trackable signal across the water surface [Schweitzer and Cowen, 2021]. These subtle temperature variations on the water surface or “skin” are a signature of surface processes such as atmospheric heat exchange and wave breaking as well 55 as bottom-generated turbulence in relatively shallow water bodies like estuaries [Brumer et al., 2016]. Thermal infrared cameras are well-suited to detect these temperature variations due to the low penetration of thermal infrared radiation in water, on the order of 10−5 m in depth [Zappa et al., 2003]. Based on this physical understanding of surface temperature, fixed-perspective thermal infrared imagery has been used to quantify surface hydrodynamics in the coastal ocean driven by waves, wind and currents on an undulating surface, which has historically presented measurement challenges [Laxague and Zappa, 2020, Sutherland and Melville, 2015]. Infrared imagery is particularly useful when flows of interest, such as snowmelt- driven floods [Fujita, 2017] or high tides, occur overnight, because sunlight is not necessary for infrared sensors to accurately detect the surface water temperature. To measure the velocity field in the time series of infrared images using IR-QIV, we create a continuous spatial grid of small image sections called subwindows that are sampled from successive pairs of images in the time series [Schweitzer and Cowen, 2021]. The movement or displacement of the thermal pixel intensity patterns in each subwindow is resolved from the first image in the pair to the second using a pattern tracking algorithm called minimum quadratic difference (MQD). MQD identifies the closest match to the pixel intensity pattern in a subwindow in the first image in the pair, within a defined search radius around the subwindow location in the second image, and the displacement of the pixel intensity pattern between images represents the surface velocity for that subwindow at the timestep between images [Gui and Merzkirch, 2000]. In order to convert the velocity field from the image time series to physical units, we derived a custom georeferencing system for both the river-side and bay-side camera deployment locations. This process involves relating the locations of each GCP in pixel space and physical space, and solving the resulting system of equations to find the camera’s location, 56 orientation and optical parameters [Holland et al., 1997]. A full 24-hour tidal cycle was monitored on each side of the dike, beginning on the river side on the afternoon of July 13, 2022 and ending on the bay side the morning of July 15. Therefore roughly half of the imagery collected on either side of the dike was of water exiting the culverts, and this flow condition was char- acterized by high velocities and significant air entrainment. This challenging flow measurement environment corroborated Huntington et al.’s observations of surface velocities > 5m/s at the outflow side of the dike [Huntington et al., 2021]. Such velocities are too high to accurately measure at the framerate we used, and the issue of estimating the volume of air entrained in this flow presents a potentially significant source of error in calculating tidal discharge. However, inflow to the culverts on both sides of the dike was characterized by a detailed thermal feature signal across the water surface, which moved at a slower velocity than the aerated dike outflow, and visibly accelerated while approaching the culvert entrances. Im- ages collected during tidal flow into the observed culverts, that is during ebb tide on the river side of the dike and during flood tide on the bay side, were therefore suitable for IRQIV analysis. On both sides of the dike, the thermal reflection or shadow of the culvert roof on the water surface was used as a boundary for optimal pattern tracking con- ditions, as this fixed reflection feature could skew the magnitude of velocimetry measurements very near or just inside the culvert entrance. To mitigate the effects of this reflection, the IR-QIV image pairs were analyzed backwards, calculating the displacement of surface features from the second to first frame in each pair. This approach ensured that IR-QIV subwindows located very near the dike en- trance were tracking pattern motion away from instead of towards the potential interference of the thermal shadow. The resulting mean vector field horizontal (U) 57 and vertical (V ) components in image coordinates were reversed before convert- ing to physical units, to derive a vector field representing time-forward velocities. This physical velocity vector field was interpolated along a transect following the the thermal shadow (hereafter: shadow transect) and evaluated at 100 evenly- distributed points within the vector field to derive a cross-culvert surface velocity profile at the closest valid measurement distance to the culvert entrance (Figure 2.4). Figure 2.4: River-side (left) and bay-side (right) pseudocolor georectified FOVs for IR-QIV, showing GCPs used for georeferencing (red points) and shadow transect for profile measurement of streamwise surface velocity (red line). Both images captured at night, illustrating detailed thermal surface features for pattern tracking present in the flow. Pseudocolor scale represents raw, uncorrected pixel intensity. Axes units are UTM x,y coordinates (m). Our goal in analyzing the continuous image time series collected on either side of the dike was to determine a characteristic surface flow velocity through the dike at each elevation head observed between the two USGS tide gauges, which both recorded water level measurements at synchronized five-minute intervals. This velocity was measured using the MQD tracking algorithm to calculate pattern 58 displacements between 101 sequential image pairs, covering roughly 11 seconds at the 9 Hz framerate. A given set of image pairs was temporally centered around the frame with the closest timestamp to a given tide gauge measurement. Pixel displacements in the vertical (i) and horizontal (j) image dimensions calculated by the MQD algorithm were filtered in time using an adaptive gaussian window (AGW) filter on each individual subwindow’s displacements [Cowen, 2006], and displacements were filtered jointly, such that any filtered value in one direction was accompanied by filtering in the other. After filtering, pixel displacements in each dimension were converted to physical units (m/s) and coordinates (UTM) using the georectification parameters determined using the GCPs as described above. The resulting vectors between the two dimensions of displacement were then temporally averaged across this set of images to determine a filtered mean velocity vector field in physical space. The cross-culvert velocity profile derived using IR-QIV was stored at each five- minute tide gauge interval from the first interval of flow into the culvert until the interval where recording stopped. The resulting data were used to fit a stage- velocity curve to the mean of the cross-culvert velocity measurements at each five-minute time interval, using a second-order polynomial fit. These polynomial functions thus provided a predicted surface velocity at a given elevation head hz(t). To derive discharge estimates Q(t)[m 3 s ] from this stage-velocity relationship, the predicted surface velocity at a given hz was multiplied by the cross-sectional area of the culvert(s) at the current tide gauge water depth on the inflowing side of the dike, and the resulting discharge estimate was multiplied by a velocity index k = 0.9 commonly used for concrete channels [Rantz, 1982]. 59 2.2.4 Models In addition to the USGS water level monitoring and the IR-QIV analysis, two mod- els have been developed and applied for the Herring River Estuary using a simple mass balance approach and discretized hydrodynamic simulation. The models pro- vide independent estimates of bidirectional discharge across the dike that can be compared with the IR-QIV analysis. Hydraulic mass balance model The hydraulic mass balance model simulates a net water flux across the dike hy- draulic infrastructure and a water level in the Herring River using hydraulic and momentum equations for flow through the sluice and flap gates driven by a pre- scribed tidal water level in Wellfleet Harbor [Befus et al., 2023]. The mass balance model uses a stage-storage relationship for the diked portion of Herring River to convert from changes in water volume (i.e., volume divided by density for mass) to stage. The most optimized model parameterization was run for the July 13-15, 2022 field experiment using a 10-minute time step and tidal water levels measured at the NOAA Boston tide gauge (site id: 8443970). Depth-averaged hydrodynamic model The Delft3D hydrodynamic model was applied in a depth-averaged manner through- out the Herring River Estuary to understand water level and flow dynamics inside the dike. Discharge through the dike culverts was evaluated using this Delft3D model based on tide gauge data during the field campaign as a second modeling point of reference relative to field measurements. 60 2.3 Results While we identified a 20Hz target framerate for image collection based on prelim- inary estimates of flow velocity and FOV length scale, hardware constraints and near-continuous image collection for 48 hours caused this target framerate to fluc- tuate. These fluctuations would have created challenges in analyzing the image time series, so a slower stable framerate of 9Hz was used for the majority of image collection. The actual framerate was pulled from the image time series for each measurement to ensure accurate calculation of the velocity time component. As shown in Figure 2.3, the shape of hz is generally symmetrical in time during flood tide (hz < 0), but asymmetrical during ebb tide (hz > 0). When the tide shifts to outflow from the estuary hz > 0, the elevation head rises quickly following its same trajectory from before the zero-crossing to the maximum head value for the tidal cycle. After peaking, hz appears to decline more slowly as long as the river water level remains above the bay water level, before rapidly declining towards an incoming flow (hz < 0) as the rising tide on the bay side of the dike reaches the river side water elevation. This asymmetrical elevation head signal reflects the damping and phase-lag of the tidal cycle upstream of the dike, and indicates that tidal outflow dynamics may change before and after the peak of hz. 2.3.1 Developing rating curves A separate stage-velocity curve for each side of the dike was constructed from velocimetry observations of flow going into the dike on that side. Therefore the ebb-tide (outgoing) curve is based on velocities measured on the river-side of the dike, and the flood-tide (incoming) curve is based on bay-side data. 61 River-side ebb tide Figure 2.5: Stage-velocity rating curves for the river-side of the dike showing av- erage velocity vs. elevation head every five minutes from 3:25 to 7:30 on July 14, 2022. Error bars show 95% confidence interval for each set of cross-culvert surface velocity measurements, derived using the Student’s t-distribution (n = 101 for each timestep). On the river-side of the dike, there was a clear change in the trend of the stage- velocity curve before and after the peak value of hz, with a rapid rise in velocity from hz = 0 to the maximum elevation head, followed by a slower decline. There- fore, we split the data before and after the peak elevation head for the tidal cycle, fitting separate second-order polynomials to the rising and falling legs of hz > 0 (Figure 2.5). Bay-side flood tide The bay-side of the dike included the added challenge of experiencing an unre- stricted full tidal regime during a spring tide event, which led to submergence 62 of the culvert openings at high tide. Surface velocities measured during this pe- riod are not representative of bulk velocities through the fully submerged culvert (Figure 2.6). Figure 2.6: Stage-velocity rating curves for the bay-side of the dike showing average velocity vs. elevation head every five minutes from 21:00 on July 14 to 4:00 on July 15, 2022. Error bars show 95% confidence interval for each set of cross-culvert surface velocity measurements, derived using the Student’s t-distribution (n = 101 for each timestep). The bay-side culvert entrance becomes submerged when the tidal stage is greater than 1.55 m; surface velocities measured during this period (shown in red) do not reflect the velocity through the culvert and were not curve- fit. A three-hour gap in data collection when the culverts were submerged is visible on the left side of the figure. Hysteresis is visible in both rating curves, with the same elevation head pro- ducing different velocities depending on whether the elevation head is growing or shrinking. 63 2.3.2 Hydrograph comparison to model outputs A discharge hydrograph was produced using the tide gauge water surface elevations on either side of the dike to derive velocity estimates from the rating curves in Figures 2.5 and 2.6 and estimate discharge. This hydrograph shows lower discharge in both directions is shown relative to the model outputs in Figure 2.7. Figure 2.7: Predicted discharge hydrograph from IR-QIV-derived river and bay rating curves in solid blue, plotted alongside the two models’ predicted discharge values during the field experiment in dashed pink (Delft3D) and green (hydraulic mass balance). The balance of inflow and outflow in the hydrograph generated using the two rating curves was calculated, with an average inflow from the Herring River of 0.34 m3 s subtracted [Befus et al., 2023]. These rating curves balance mass to approxi- mately 10% during the study period as shown in Figure 2.8. 64 Figure 2.8: Hydrograph derived from river and bay rating curves in Figures 2.5 and 2.6 evaluated for mass conservation between ebb and flood tidal stages during the period of the study, corrected for constant discharge from the Herring River. Flow rate shown in blue on left axis, cumulative volume for each tidal phase shown in red on right axis. 2.4 Discussion The mass balance across the tidal phases shown in the Figure 2.8 hydrograph is a noteworthy result, since the rating curves in Figures 2.5 and 2.6 used to produce this hydrograph were derived from two separate image datasets collected on op- posite sides of the dike with distinct georeferencing systems and flow conditions observed. This result demonstrates that infrared surface velocimetry can be used to constrain surface velocity in challenging flow measurement conditions, and if channel geometry is known as in this system, this velocimetry method can be 65 used to estimate discharge as well. Discharge measurements through the Herring River dike have historically proved challenging [Huntington et al., 2021], and our application of IR-QIV to estimate discharge through this structure may help in- form discharge targets for ongoing tidal restoration in the Herring River Estuary. Additional velocimetry analysis incorporating acoustic doppler velocimeter (ADV) velocity data is included in Appendix A. There exists a non-stationary relationship between stage and discharge in tidal systems, with different bidirectional discharge values occurring at the same water surface elevation depending on whether the tide is ebbing or flooding [Kearney et al., 2017]. The hysteresis visible in the rating curves in Figures 2.5 and 2.6 reflects the complexity of bidirectional tidal flow in impounded systems such as the Herring River, with different relationships between stage and velocity depending on whether the elevation head across the dike is growing or shrinking. The range of elevation heads driving the different velocity values shown in these plots can occur at different water levels, since this head value is set by the relative difference in water surface elevation between the upstream and downstream sides of the dike. It follows that a given elevation head cannot be tied to a singular discharge in this system, because different water surface elevations yield different cross-sectional areas inside the dike culverts, thereby producing different discharge values at the same relative elevation head. Building on the fact that velocity scales as the square of hydraulic head under the energy equation, we assume second-order polynomial rating curves adequately represent stationary relationships between elevation head and velocity across a range of absolute water surface elevations on either side of the dike. This assumption of a stationary head-velocity relationship is challenging to further interrogate with the dataset collected during this field campaign, which 66 only recorded one tidal cycle on each side of the dike. While the dike creates particularly challenging flow measurement conditions, it also provides a barrier to momentum transfer between the Herring River and Wellfleet Bay. With flows approaching the dike from either side essentially approaching zero velocity before rapidly accelerating through the culverts, the elevation head across the dike is the predominant driver of the flows observed. Therefore, the assumption of stationarity in the head-velocity relationship employed here may be more valid than in a free- flowing tidal system, and is certainly more defensible than assuming a stationary head-discharge relationship in this system. While both models simulate much larger discharges in either direction than the rating curve-based hydrograph in Figure 2.7, the net effect over a tidal cycle is limited as the larger-magnitude positive and negative discharges largely cancel out over a single cycle. Head loss due to a hydraulic jump in the sluicegate culvert is not reflected in the models, which were optimized for water levels rather than discharge [Befus et al., 2023], in part because no published discharge measurements for the Herring River dike existed prior to this measurement campaign. Therefore, it is perhaps unsurprising to see such wide variability between the models and the rating curve hydrograph in Figure 2.7, and cumulative discharge over time may be a better metric of discharge accuracy across these predictive methods. Ultimately, the field-based result of far more limited discharge relative to model predictions may indicate that the Herring River dike restricts tidal flow more than previously thought, increasing the impact of its replacement with adjustable tide gates which is currently underway. 67 CHAPTER 3 REMOTE MEASUREMENTS OF FLOW RATE, BATHYMETRIC CHANGE AND SEDIMENT FLUXES THROUGH A TIDAL BEACH OVERWASH CHANNEL USING THERMAL INFRARED IMAGERY 3.1 Introduction Tidal overwash is a coastal inundation process resulting from the overtopping of coastal dunes, leading to the formation of channels through breaches formed in the dune system. The morphological effects of overwash events due to hurricanes and significant wave events affecting coastal barrier systems have been well doc- umented in the literature [Durán et al., 2016, Mathew et al., 2010, Ritchie and Penland, 1988], but direct observations of overwash hydrodynamics using surface velocimetry or other methods are sparse [e.g., Holland et al., 1991]. Many overwash channels naturally close, as sediment transport into the low point formed in the breach re-accretes the channel’s base elevation which decreases the frequency of overwash events and further erosion [Donnelly et al., 2006]. High tide overwashes not associated with storm events are likely to become more frequent, extensive and widespread due to sea level rise. The increase in erosion and salinity caused by overwash events can have significant impacts on coastal ecosystems, including groundwater salinization [Anderson Jr., 2002] and dune vegetation distribution [Fahrig et al., 1993]. Initial storm-driven waves that breach the dune system and subsequent over- wash events transport sediment inland, creating depositional features called washover fans which are characterized by high rates of sediment transport [Shaw et al., 2015]. Over time, washover fans play an important role in the landward movement 68 of transgressive barrier islands, providing mobilized sediment with a pathway to more protected depositional areas, effectively helping barrier islands maintain re- silience to sea level rise [Rodriguez et al., 2020]. Overwash events are not uncom- mon on Cape Cod, with a 2013 storm creating a breach in the dunes at Ballston Beach on the Atlantic side of the Cape and sending seawater down the Pamet River into Cape Cod Bay, effectively crossing the entire width of the peninsula [Mckay and Rupp, 2022]. The distinctive shape of Cape Cod is shaped by multiple littoral cells, which are segments of the coast exhibiting self-contained sediment circulation patterns driven by waves, tides, and currents [Berman, 2011]. The lack of established field methods to measure the hydrodynamics of overwash and its potentially severe impacts to coastal ecosystems motivate our efforts to quantify seawater and sediment fluxes through a beach overwash channel exposed to Cape Cod’s dynamic sediment transport regime. 3.2 Methods This field campaign observed three successive overnight overwash events at Duck Harbor Beach on Cape Cod, Massachusetts between July 31 and August 2, 2023. 3.2.1 Study site Duck Harbor Beach lies within Cape Cod National Seashore, backed by a low dune system which divides a former salt marsh subwatershed of the Herring River Estuary from Cape Cod Bay. In early 2021, overtopping waves from a winter storm breached this dune system creating an overwash channel, which carries sea- water across the beach and into formerly forested areas during high tides, causing widespread forest mortality [Castagno et al., 2025]. Duck Harbor is near a nodal 69 point between littoral cells along Cape Cod Bay, with diverging wind and wave patterns causing sediment transport away from this part of the coastline [Berman, 2011], potentially providing some explanation for why the overwash channel at Duck Harbor has persisted and grown in the years since its formation. Quanti- fying seawater flow rate through the overwash channel is of significant interest in the context of the ongoing Herring River Estuary restoration, which is restoring tidal inflow via a significant dike replacement project at the river mouth. This tidal overwash flow carries with it significant amounts of beach sediment that have formed a washover fan extending inland from the beach, visible in Figure 3.1. Duck Harbor presents an excellent location to develop overwash measurement method- ologies because this site’s flow is so predictable based only on tidal elevation, while in most other locations where overwash has been observed it is typically caused by hazardous storm conditions precluding measurements during the event itself. Figure 3.1: The Herring River Estuary outlined in blue at left, with location of Duck Harbor subwatershed labeled and washover fan inset imagery collected by Planet Labs during 2023 field campaign. The approximate camera deployment location on August 1-2 is shown on the south side of the washover fan in the inset. 70 3.2.2 Field campaign After passing through the breached line of dunes behind the beach, overwashes entering the Duck Harbor subwatershed of the Herring River Estuary continue flowing downhill via a drainage channel, meeting the main stem Herring River approximately 1.5 km inland of the beach. We placed a Nortek AquaDopp current profiler in the Duck Harbor channel just upstream of the confluence and it recorded water level and velocity every 30 seconds for 10 days, capturing the entire cycle of overwash dynamics. The location of this profiler and a cross-channel elevation transect were surveyed using a real-time kinematic (RTK) GPS unit to enable estimation of the Duck Harbor overwash channel’s discharge to the Herring River main stem. High tide overwashes were expected at Duck Harbor over several nights of predicted spring tides beginning July 31, 2023. As part of a longer-term monitor- ing effort to document the extent of the evolving washover fan at Duck Harbor [Castagno et al., 2025], daily RTK GPS elevation surveys were conducted begin- ning on July 31 before the first overwash event of the month. This survey included the measurement of ground control points (GCPs) used for georeferencing, which consisted of fenceposts driven into the beach and wrapped with reflective tape for increased visibility in infrared imagery. The bottom and top of each fencepost GCP were surveyed daily to account for erosion or movement. At the time of the field campaign, the washover fan exhibited two distinct channels, visible as darker areas within the washover fan in Figure 3.1, separated by a higher point hosting some remnant vegetation that was only submerged at the peak of the highest tides. There was no available vantage point where a camera could be located with both channels fully visible. On July 31, the narrower northern channel was monitored from the channel bank, and on August 1-2 the 71 wider southern channel was monitored from a low dune south of the channel, with significant overwash flows observed all three nights. Monitoring consisted of continuous thermal infrared image collection of the overwash flow at a target image acquisition frequency of 20 Hz. We used a FLIR SC8300 cooled thermal infrared camera for IR-QIV applica- tions. This camera detects mid-wave infrared radiation (3-5 µm) using a 1.05- megapixel (1344x784) indium-antimonide (InSb) sensor, which is cooled to cryo- genic temperatures (approximately 80 K) to reduce thermal noise. The FLIR camera was fitted with a 17 mm focal-length lens and is mounted inside a fan- vented weatherproof metal enclosure when deployed for extended periods of time in field conditions. Image acquisition is controlled by a field computer which sets camera parameters through FLIR software and records image files to external stor- age. Roughly two hours of data were collected each night from the beginning of the overwash to after the tide had crested. 3.2.3 Infrared velocimetry analysis The collected images were georeferenced using the surveyed location of the GCPs from the day of data collection, by relating the locations of each GCP in pixel space and physical space, and solving the resulting system of equations to find the camera’s location and orientation [Holland et al., 1997]. Due to the low vantage point of the camera, images were collected at a very oblique angle to capture all of the overwash channel—13◦ below horizontal on August 1. Accurate velocimetry measurements require an accurate water surface elevation (WSE), which would have been difficult to measure directly in the very shallow overwash channel using traditional methods like a pressure transducer. Since the GCP stakes were placed throughout the camera’s field of view (FOV) in the channel, the base elevation of 72 the stakes was used to determine the WSE at the time the rising tidal elevation reached each GCP, identified visually. Then intermediate WSEs were inferred using a cosine function fit to the predicted tidal minima and maxima for the days of the study [NOAA, 2023], and the offset between this continuous water level time series and the georeferencing vertical datum was averaged across three different timesteps when the rising tide intersected with three different GCPs as shown in Figure 3.2. Figure 3.2: Reconstructed tidal water surface elevation based on predicted tidal minima and maximum during August 1 overwash event shown as blue line, shifted to georeferencing vertical datum. Three ground control point elevations used to calculate offset between tidal and georeferencing datum shown as colored points at top. The green point represents the water surface elevation when overwash was first observed, and the blue dot shows the WSE in Figure 3.3. To process the images collected on August 1, a cross-channel transect was defined in the images between two GCPs on opposite sides of the south channel 73 with vertical elevations within 3 cm, separated by 38.6 m. Along this transect, 20 intermediate points were evenly spaced in physical units, and lines of pixels extending seaward from these points with physical lengths of 5.0 m were defined as sampling locations for pattern tracking as shown in Figure 3.3. This area of the image time series was sampled from 201 successive images collected at 20 Hz, covering approximately 10 seconds of actual time. Each pixel’s median temporal pixel intensity was subtracted from each pixel in the time series of the sampled image area to remove the effect of stationary features like rocks disrupting the flow signal. Then the sampled image area was interpolated using a parabolic function, and 201 evenly-spaced sub-pixel intensities were pulled from each sampling line in each image and assembled into square synthetic images of pixel intensities along each line over time called a space-time diagram, shown at left in Figure 3.4. Figure 3.3: Orientation of evenly placed horizon-parallel sampling lines for pattern tracking in red, distributed along white transect across overwash channel during flow, with waves on the water surface shown in pseudocolor infrared image from approximately 10:22 PM, August 1, 2023. 74 The space-time image velocimetry (STIV) method was used to extract veloc- ities along each sampling line by finding the two-dimensional autocorrelation of each space-time diagram and integrating it in polar coordinates, with the max- imum of the polar integral representing the angle of the dominant flow in the space-time diagram [Fujita et al., 2019]. Space-time diagrams assembled from this dataset exhibited two slope patterns, and flow in the time series of images was clearly affected by shoaling waves coming off the bay as well as the underlying overwash flow velocity. Therefore, the steeper (higher-velocity) slopes, which had higher pixel intensities and were therefore identified as the peak of the autocorrela- tion integral, were classified as the wave velocity, while the flatter (lower-velocity) slopes represented the bulk velocity. The secondary peak was identified by finding the highest local maxima in a five-degree running-average of the autocorrelation integral between the global peak at the wave velocity and 0, as shown in Figure 3.4. Figure 3.4: Left: space-time diagram assembled from temporal samples of inter- polated pixel intensity along 8th sampling line, labeled with angles of wave (red) and bulk (orange) velocity signals identified in autocorrelation integral; Center: normalized autocorrelation integral plotted by angle, illustrating identified wave (red) and bulk velocity (orange) peaks plotted as angles on space-time diagram; Right: histogram of pixel intensities in space-time diagram after pixel-wise median subtraction. This method was used to determine the bulk flow velocity along each sampling line along with the channel depth at that location, which can be inferred from the 75 shallow-water approximation for current-affected wave speed c: c = U0 + √ gh → h = (c− U0) 2 g (3.1) where g is gravitational acceleration, U0 is the overwash flow velocity and h is the water depth [Dean and Dalrymple, 1991]. Correcting the apparent wave speed for the Doppler shift induced by the overwash flow velocity is necessary to yield an accurate depth estimate. This derivation of both depth and velocity from remote measurements enables the computation of flow rate through the overwash channel, by taking the normal component of the bulk velocity calculated at each sampling line relative to the cross-channel transect, and multiplying it by the wave-derived water depth at each sampling line and the channel width. For sampling lines where both the wave and bulk velocities could not be resolved from the space-time diagram, the nearest valid velocity was extrapolated to a linear depth fit from the nearest valid depth to the edge of the channel defined by the GCPs. Even if the wave velocity slope was correctly identified by the STIV algorithm, it is not possible to accurately estimate depth from this method without also identifying the bulk velocity slope to correct the wave speed before calculating the water depth. Additionally, erroneously detected slopes along pixel rows and columns, denoting zero or infinite velocity respectively, were excluded from the velocity calculation along with bulk velocity slopes denoting outflow and derived depths greater than 1 m, neither of which were observed in the field. Finally, the normal surface velocity at each sampling line was multiplied by a velocity index k of 0.85, which provides an estimate of the ratio between surface and depth-averaged bulk velocity in natural channels [Johnson and Cowen, 2017]. 76 Velocity, depth, and flow rate uncertainty The process described above was repeated in 50 overlapping time steps over three minutes of data at the WSE shown in blue in Figure 3.2, and the average flow rate was derived by summing the averaged valid depth and velocity measurements for each sampling line. The distribution of up to 50 valid velocity and depth measurements at each sampling line was bootstrapped 1000 times with replacement [Efron and Tibshirani, 1994], with each bootstrap iteration drawing the same index from the velocity and depth distributions to preserve the paired nature of these measurements. Then the 1000 depth and velocity bootstrap measurements at each sampling line were aggregated into 1000 bootstrap estimates of flow rate. The 95% confidence interval for each sampling line’s depth and velocity as well as total flow rate were derived from these bootstrap distributions. 3.3 Results 3.3.1 Flow rate estimate from surface velocimetry Using the baseline analysis method described previously produced a flow rate es- timate of 3.3 m3/s ± 18%. In addition to the bootstrap confidence interval, the ratio of the secondary bulk velocity peak, shown in orange in the central panel of Figure 3.4, to the minimum between the bulk velocity and primary wave peak was averaged across all STIV calculations as shown at the top of Figure 3.5. 77 Figure 3.5: Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. The camera was located on the left side of the cross-channel transect as shown in Figure 3.1. Additional efforts to implement contrast enhancement as a pre-processing step for space-time diagrams before slope detection produced the same estimated flow rate and confidence interval of 3.3 ± 0.6 m3/s, with additional details and figures shown in Appendix B. 3.3.2 Elevation surveys The daily RTK GPS elevation surveys conducted between overwash events followed set transects near the perimeter of the washover fan in addition to visiting the GCPs. The difference in vertical profiles for the transect across the entire mouth 78 of the overwash channel covering over 96 m is shown in Figure 3.6. Figure 3.6: Elevation profiles showing change in bed elevation across overwash channel mouth after three measured overwashes. Data collected by Katherine Castagno. The average decline in channel elevation across the profile in Figure 3.6 is 6 cm; extrapolating this figure across the 5 m by 38.6 m area where STIV analysis was conducted in Figure 3.3 yields a total sediment volume mobilized of 11.6 m3 over three overwash events. This result highlights the significant volume of sediment being mobilized during these events which leads to the expansion of the washover fan shown in Figure 3.1. 3.3.3 Acoustic measurement of flow rate at confluence Analysis of the AquaDopp profiler data from the Duck Harbor channel at the confluence with the mainstem Herring River revealed significant fluctuations in the WSE at the confluence. Deployed on the bottom of the channel, the AquaDopp measures velocity at set vertical distances or bins, and the bin with the strongest amplitude response was indicative of acoustic reflections off the water surface. The bin with the maximum amplitude response running-averaged over a 30 minute 79 window was taken to be the water surface to mitigate noise in the amplitude signal, and the WSE was derived based on the corresponding distance from the AquaDopp in the maximum bin, which is shown in red in Figure 3.7. Figure 3.7: AquaDopp amplitude response colormap and identified WSE in red over the course of deployment at the confluence between the Duck Harbor overwash channel and the Herring River. The bins at and below the WSE depth were considered valid velocities at each timestep, and the flow rate was derived by computing each valid bin’s cross-channel area based on its elevation relative to the cross-channel elevation survey at the AquaDopp deployment location. Then the resulting area was multiplied by the streamwise velocity in each bin, and this depth-varying vertical flow rate profile was summed to derive a total flow rate at each timestep as shown in Figure 3.8. 80 Figure 3.8: One hour running-averaged AquaDopp flow rate for Duck Harbor channel confluence with Herring River. Positive values indicate flow from Duck Harbor into Herring River, negative values indicate water from the river backing up Duck Harbor channel. These data indicate that significant volumes of seawater are reaching the Her- ring River due to the Duck Harbor overwashes, with the peak flow rate in the Duck Harbor channel during this period approximately 50% of the overwash flow rate measured at the beach. This fraction of the total overwash volume reaching the main stem, coupled with the multi-day lag of the highest flow rates after the highest spring tides on August 1 and 2, highlight potential insights into the high degree of groundwater connectivity in this system. Note that the elevated water depth at the end of the AquaDopp deployment in Figure 3.7 was not accompanied by as significant an increase in the flow rate in Figure 3.8, this was the only signif- icant precipitation event during the AquaDopp deployment which did not create as much directional discharge out of Duck Harbor as the earlier overwash events. 81 3.4 Discussion Overall, these results demonstrate the utility of STIV to constrain overwash flow rate without in situ measurements. The different space-time diagram contrast enhancement methods detailed in Appendix B yielded the same discharge as in Figure 3.5, but the apparent stability of this contrast enhancement method did not produce significantly smaller confidence intervals for individual sampling lines. While the rapidly changing depth, bathymetry and wave dynamics in the overwash channel could explain some of this high temporal variance across the channel over longer timescales, short-term unsteadiness and noise in the data are more likely causes of the high uncertainty reflected in the three minutes of data shown in Figure 3.5. The need to detect slopes for both wave and bulk velocity signals in each space-time diagram to correct the wave speed for current velocity and derive a depth estimate renders the depth variable particularly sensitive to noise and error. Additionally, sampling lines in the far field, toward the right side of Figure 3.5, yielded fewer successful slope detections and thus more repetitive bootstrap resampling. This relative lack of far-field data could be caused by more obstacles to waves and flow in the the shallow areas near the edge of the overwash channel, larger pixel footprints blurring the velocity signals in the space-time diagram, or noisier and weaker thermal signal in this more oblique area of the infrared image time series. However, the highest velocities and depths are generally clustered near the center of the overwash channel as would be expected in open-channel flow, and this region shown in darker blues in Figure 3.5 contributes most to the total flow rate. The finding of significant sediment transport from the repeated elevation sur- veys illustrates the dynamic growth of the washover fan, suggesting that there may be differences in the channel bathymetry between August 1 and 2 that could 82 be detected remotely using this cross-channel STIV method. Documenting the flow velocities mobilizing this sediment will enable analyses of critical shear stress, grain size distribution and the morphological trajectory of the washover fan. These elevation survey data provide a valuable ground-truth for validating the channel bathymetry detected remotely during each overwash. The lag between the highest spring tides and the peak discharge from the over- wash channel to the confluence in Figure 3.8 appears to be approximately two days, with the largest overwashes occurring August 1 and 2 and the highest flow rate at the confluence recorded August 3 and 4. Flow rate in the overwash channel at the confluence dropped to roughly zero and even briefly reversed between these high flow rate events, suggesting that the overwash may be moving through the Duck Harbor watershed as a pulse of seawater that takes days to reach this confluence 1.5 km inland, with water from the Herring River backing up the channel between pulses. The very consistent semidiurnal signal in the water surface elevation shown in Figure 3.7 even before the largest overwash pulses arrive in Figure 3.8 further supports tidally-influenced groundwater dynamics in the watershed. Integrating these findings with ongoing reactive transport groundwater modeling efforts in Duck Harbor will help improve understanding of how overwash salinity is traveling throughout the upper Herring River Estuary, with salt-stressed vegetation being observed as far upstream as Bound Brook since overwashes began. This method of deriving overwash depth and velocity remotely using STIV presents a powerful tool to measure seawater and sediment fluxes as high-tide overwashes not associated with storms become more frequent and widespread due to sea level rise. While running field campaigns during storms large enough to cause storm surge overwash is likely an unsafe proposition, STIV could also be implemented to measure storm-driven overwash using video from an automated 83 fixed location such as a traffic camera, Argus tower, or surf webcam. Integrat- ing remote discharge measurements into coastal biogeochemical studies presents another application of this method, which essentially measures a flux of salt into brackish or freshwater ecosystmes by constraining a total seawater volume flux during an overwash event. As overwash impacts become more widespread, addi- tional applications of STIV to remotely estimate overwash flow rate are likely to arise. 84 CHAPTER 4 IS A HISTORICAL VEGETATION MAP A USEFUL PREDICTOR OF DRY SEASON SOIL WATER FRACTION IN SEASONAL PONDS? 4.1 Introduction Documenting the spatial distribution of vegetation has been a core area of study for many ecologists since the birth of this discipline, and ecologists have used a wide array of classification methods to derive ecological relationships and produce vegetation maps since the late 19th century [Whittaker, 1978]. These early ecolog- ical methods sought to integrate the complex relationships between different plant species into a cohesive theory of ecosystems, the characteristics of which are largely prescribed by climatic conditions and soil type [Tansley, 1935]. One practical early method of vegetation classification, which was more concerned with the relative relationships between plant species in a given ecosystem than within a rigid global evolutionary hierarchy, was classification by dominant character-species that serve as indicators of environmental conditions in the areas where each species is domi- nant [Whittaker, 1962]. While this method was somewhat subjective and superfi- cial, it was rooted in the widely useful idea that these indicator species’ dominance could be used as a proxy for environmental conditions by analyzing a habitat and interpreting a community of individual plants’ response to the conditions found there [Clements, 1920]. Modern wetland delineation methods rely on dominant indicator plant species to determine wetland boundaries crucial to wetland management, regulatory com- pliance, informed decision-making, and protection of plants or animals of special concern, and these methods are rooted in early aquatic plant ecology studies. These 85 early studies determined how the distribution of different aquatic vegetation species correlated with environmental and water quality parameters such as soil drainage [Livingston, 1905], wave climate [Pearsall, 1917], turbidity [Warming, 1925], and mineral content [Curtis, 1959]. Other studies observed how aquatic plant suc- cession modified the physical environment to create new habitat for other plant species: via submerged vegetation species decreasing the depth of lakes and ponds which allowed emergent species to colonize areas where the water was previously too deep [Cowles, 1901], or by certain plants in fragmented interdunal wetlands creating conducive conditions for the dispersed seeds of other plant species to ger- minate there [Gleason, 1926]. These early species-based spatial surveys of aquatic vegetation provide valuable information about historical environmental conditions within the surveyed ecosystems, but contemporary data are needed to understand the extent of what these historical resources can tell researchers today. More recent studies have shown that the spatial variability of vegetation zones in wetlands develops in response to inundation duration, frequency and timing, to- gether referred to as hydroperiod. In experimental studies, different plant species exhibit different tolerances and preferences for the duration of flooding, which sorts species into different ecological niches by inundation duration [Campbell et al., 2016]. The duration of flooding is often related to wetland topography, with niches falling into different elevation ranges. Along riverbanks for example, different wetland plant species occupy zones at different distances from the river, from a typically inundated low zone along the banks to a higher zone that is only occasionally inundated during floods [Reinecke et al., 2015]. Hydroperiod variabil- ity across large wetland complexes with heterogeneous hydrology produces a rich spatial variety of wetland types and plant species [Murray-Hudson et al., 2015]. In temporary or ephemeral wetlands, seasonal and inter-annual variability of precip- 86 itation causes variation in the hydroperiod, which increases wetland biodiversity by creating growing conditions for a wider variety of species [Casanova and Brock, 2000]. The decades-old practice of wetland delineation is based on identifying dif- ferent hydrophitic vegetation zones that inhabit different hydroperiod niches, and mapping different soil types that influence plant species distribution and hydrology [Laboratory, 1987]. 4.1.1 Objectives Historical vegetation maps can be powerful tools to inform current research ques- tions and guide experimental design, but developing methods that both evaluate and extrapolate these maps’ on-the-ground relevance today is essential to max- imize the potential contributions of these historical resources. Our study was inspired by a 1984 vegetation map of seasonal ponds, which defined a pond type classification system based on the dominant wetland plant species at each pond’s center, using this species as a proxy for the hydroperiod. This extensive historical record of vegetation distribution primarily based its classification system on years of empirical observations and systematic plant surveys at the study site, as well as depth to water table measurements. Our study addresses the following objectives to determine this map’s representativeness and utility today: 1. Determine whether species distributions documented on the original map have changed due to potential perturbations to these vegetation communities over the past four decades. We hypothesize that dominant wetland plant species will have remained the same. 2. Enhance the quantitative granularity of the historic pond classification sys- tem by systematically measuring soil moisture gradients within and between pond types. We hypothesize higher soil moisture at the center of ponds and in the wetter 87 pond types, and lower soil moisture near the edge of ponds and in the drier pond types. 3. Expand this analysis to also include soil organic matter calculations for all soil water fraction samples. We hypothesize that samples with higher soil water fraction will have higher soil organic content. 4. Update the original study using modern statistical methods to more robustly quantify the relationship between pond type and soil moisture present at the study site. 4.2 Methods This study was designed and conducted during a two-week graduate field ecology course in April 2024, with site selection, fieldwork and sample processing conducted in a span of approximately four days. The basis for this study’s methodology was a 40-year-old vegetation map of a research station in Florida, USA which delineated major vegetation types in seasonal ponds and surrounding areas as coded polygons [Abrahamson, Warren G. et al., 1984]. This map was hand-drafted by Edwin Rivera using sectioned aerial photos labeled in the field, with a scale of 1:2400 (1”:200’). 4.2.1 Study Site Archbold Biological Station (ABS) is a research institute located at the southern end of Florida’s Lake Wales Ridge. ABS preserves intact Florida sand scrub habi- tat, which is characterized by evolutionary adaptations to water scarcity on the very well-drained soils of this ecosystem, and the sandy topography of the research station hosts many shallow, seasonally-dry ponds. During interglacial sea level 88 highstands over the last 23 million years the Lake Wales Ridge was some of the only emergent shoreline on the Florida Platform, and these former sandy islands now form a ridge up to 95m above current mean sea level [Bostick et al., 2022]. Ancient shoreline features such as paleo dunes and beaches dominate the relatively hilly topography of the Lake Wales Ridge [Abrahamson et al., 1984], which hosts the highest point in peninsular Florida at Sugarloaf Mountain approximately 160 km north of ABS. The many seasonal ponds at ABS bear some resemblance to interdunal wet- lands or dune slacks; aeolian transport causes these wetlands to form in low spots where the water table intersects the surface, and dune slacks are found along sandy coastlines in the United Kingdom [Stratford et al., 2013], Western Australia [Se- meniuk and Semeniuk, 2011], the Laurentian Great Lakes [DeVries-Zimmerman et al., 2021, Philben et al., 2024] and elsewhere. As dune slacks mature, vegeta- tion succession occurs and these wetlands begin to develop soil from accumulated biomass [Sýkora et al., 2004]. While the ABS seasonal ponds are also underlain by differing soil types associated with different vegetation communities, the ele- vation of the flatwood vegetation zone where the ponds are located would likely have been seafloor during the Pliocene [Givens et al., 1984], so a combination of benthic, aeolian and pedogenic processes presumably shaped the unique variety of seasonal ponds found at ABS. The intensive vegetation survey of ABS illustrated in the map used for this study was begun by James Layne in the late 1960s, and continued by Warren Abrahamson and Ann Johnson from 1976 to 1981 [Abrahamson, Warren G. et al., 1984]. This research campaign cataloged all the seasonal ponds at ABS, which range in diameter from tens to hundreds of meters and cover 10% of the research station’s area. Hydroperiod varied dramatically across the ponds based on their 89 depth relative to the water table elevation, underlying soils, surface area and pre- cipitation. A common zonal pattern in vegetation species observed across many ponds was used to derive a species-based pond type classification system for pond depth and flooding frequency, shown in Table 4.1. Abrahamson’s group compre- hensively cataloged the spatial distribution of vegetation species, and documented differing depths to the water table in 20 individual ponds representing four of the pond types, finding the deepest water table in the driest Broomsedge pond type and surface water in the wettest Maidencane pond type. The authors also cal- culated soil water fraction at three ponds, finding roughly 5.5 times higher soil moisture in one Maidencane pond than in two Hypericum ponds across multiple seasonal measurements. The Abrahamson study did not investigate spatial differ- ences in soil water fraction within individual ponds or across more than two pond types in the classification system. It appears that soil organic matter (SOM) was measured at the site but these data may not have been published, and this study also found no clear trend between other soil nutrients (P, K, Ca and Mg) or pH and plant species distribution. During this campaign Abrahamson’s group produced the vegetation map which forms the basis of this study [Abrahamson, Warren G. et al., 1984]. Further research conducted at ABS during this period highlighted the importance of fire in maintaining stable vegetation community zones across this xeric and fire-adapted ecosystem [Myers, 1985]. Florida scrub plant species’ fire adaptations include heat- or smoke-stimulated seed germination and resprouting from surviving tissues, and a nearly four decade-long study monitoring prescribed fire impacts on vegetation at ABS found significant increases in species richness, evenness and diversity due to repeated prescribed burning over this period, with no invasive plant species becoming established in study areas [Abrahamson et al., 2021]. 90 Table 4.1: Pond types: the major vegetation wetland species used to classify seasonal ponds at ABS by hydroperiod, arranged from driest to wettest species. Adapted from [Abrahamson, Warren G. et al., 1984]. Common name Latin name Wetness rank USDA wetland indicator status Broomsedge Andropogon brachystachyus 6 FAC Cutthroat grass Panicum abscissum 5 FACW Hypericum Hypericum edisonianum 4 OBL Spartina Spartina bakeri 3 FACW Redroot Lachnanthes caroliniana 2 OBL Maidencane Panicum hemitomon 1 OBL 4.2.2 Experimental design Eight seasonal ponds were chosen from two blocks of ABS with equivalent recent fire histories, specifically blocks that were under prescribed burn treatment in 2023 with fire return intervals in the 6-9 year range [Main, Kevin N. and Menges, Eric S., 1997], to control for fire as a confounding variable affecting soil moisture and organic matter. The seasonal ponds at ABS are classified in six categories based on the historically-predominant vegetation type found at the wettest area in the center of each pond. These dominant central vegetation types thus roughly correspond to the proportion of the year the different categories of ponds hold water. Two ponds from each of the following vegetation categories were chosen for this study: Broomsedge (Andropogon brachystachyus, driest species), Hypericum edisonianum (third-driest species), Redroot (Lachnanthes caroliniana, second-wettest species) and Maidencane (Panicum hemitomon, wettest species). While the historical map does not show soil type, the three driest pond types selected are located within the scrubby flatwoods vegetation type, characterized by well-drained sandy soil and higher topography, while the Maidencane ponds are located within the flatwoods vegetation type which has poorly-drained soil and lower topography [Abrahamson, Warren G. et al., 1984]. 91 These four pond types were used as ordered factors, with a hypothesized soil water fraction gradient increasing from Broomsedge ponds to Maidencane ponds; adjacent pond type in this study therefore refers to sampled pond types which are adjacent on this ordered wetness scale rather than physically adjacent ponds. A transect was laid out from the deepest area of each pond out to the edge of the pond just beyond the first upland vegetation encountered, either saw palmetto (Serenoa repens) or various woody shrub species for all ponds sampled, shown in red in Figures 4.1 and 4.2. 92 Figure 4.1: Northern set of four ponds with transects highlighted in red. Redroot ponds in blue (R1 north, R2 south), Maidencane ponds in purple (M1 north, M2 south). 93 Figure 4.2: Southern set of four ponds with transects highlighted in red. Hyper- icum ponds in green (H1 east, H2 west), Broomsedge ponds in yellow (B1 west, B2 east). 4.2.3 Soil Sampling For each pond, a belt transect of 2 meters in width was used to characterize vegeta- tion community changes from the center of the pond to the upland area surrounding the pond. The presence or absence of water was also noted along this transect. The resulting zones formed by water presence and vegetation type were used to de- termine 3-5 distinct sediment sampling locations at the approximate center of each zone along each pond’s transect. At each sampling location, two soil cores were collected using a hand corer of 30cm length and an inner diameter of 2.5cm. Each soil core was measured for length and characterized for soil type, photographed and then placed in a sealed plastic bag. The transect location and water depth were also recorded for each sampling location. Transect distances were normalized from zero to one across the eight ponds by dividing the sampling distance along the transect by the total transect length, with zero denoting the center of the pond and one denoting the end of each transect in the upland vegetation areas. 94 Fresh soil samples were transferred to paper bags, weighed and placed in a drying oven at 60◦C for 65-67 hours. Dried samples were weighed to determine soil water fraction by mass, sieved through a 2mm screen and subsampled into 20mL borosilicate glass vials, which were weighed before ashing in a muffle furnace at 450◦C for 16 hours. Ashed samples were again weighed to determine loss on ignition (LOI) of organic material [Santos et al., 2020, Kasozi et al., 2009]. 4.2.4 Mixed effects modeling Mixed effects models are able to identify statistical correlations across non-independent data such as samples collected from the same pond in this study, and these models have been used in similar studies to identify the relative nutrient removal efficiency of different wetland plants [Frenzel et al., 2024], and determine the effect of plant community composition on wetland methane fluxes [Schultz et al., 2011]. A mixed effects statistical model was set up using the ’lme4’ package in RStudio to determine how soil water fraction varies with respect to transect distance and pond type, the two of which were considered as fixed effects [Bates et al., 2024]. Additionally, since samples were collected in two ponds of each vegetation type, pond code was included as a random effect to account for possible variation in soil water content between individual ponds of the same type. An interaction in the pond type and transect distance fixed effects was included in the model to allow the effect of transect position on soil water content to vary across pond types. Random slopes across individual ponds were not included due to the sparsity of the collected data, so the modeled slope for each pond type represents the aggregate of the two individual ponds of each vegetation type that were sampled. 95 4.3 Results All eight ponds studied still had the coded species present at pond center, confirm- ing the null hypothesis that the dominant species in each pond had not changed from the original map. For both the historical map and our study, ponds that hosted more than one vegetation type in different zones were classified by the dominant species in the wettest, most central area of the pond. Pond-center soils were predominantly silty sand in the Broomsedge and Hyper- icum ponds, sandy silt in the Redroot ponds and silt in the Maidencane ponds. Stratified soils with a 1cm silt layer on top of sand were visible in the central sam- ple from one Hypericum pond, and a 1cm layer of sphagnum moss was sampled on top of the soil at the center of one Redroot pond. Soils in upland areas adjacent to ponds at the end of each transect were consistently sand or silty sand. 4.3.1 Correlation between soil water fraction and pond type Soil water fraction showed a very clear trend across all four pond types, with soil water fraction at pond center decreasing from the wettest to driest pond type, with all four pond types exhibiting decreasing soil water fraction from the center towards the pond edge (Figure 4.3). 96 Figure 4.3: Soil water content by weight fraction along normalized transect distance grouped by pond type (color), with presence or absence of standing water shown as shape. Trendlines show linear fit by pond type with 95% confidence interval. Transect distance of 0 denotes pond center, 1 denotes pond edge. An analysis of variance (ANOVA) on measured soil water fraction revealed significant differences (P < 0.001) in central samples across pond types (Table 4.2). The central location was chosen for this ANOVA because the largest range in measured soil water content was found at a normalized transect distance of 0, intermediate transect measurements are not directly comparable between ponds (because they occur at different transect distances), and the upland vegetation found at a transect distance of 1 was not expected to vary greatly between pond types being located outside of the pond. Table 4.2: ANOVA results for measured soil water content at pond center across all vegetation types. Term Degrees of freedom F statistic P value 1 Pond type 3.00 45.34 <0.001 2 Residuals 12.00 97 To investigate statistically significant pairwise relationships between pond types, Tukey testing was conducted for all pairs of pond types (Table 4.3). These results showed no significant difference in pond center soil water fraction between the two driest pond types, and significant differences in pond center soil water fraction in both pairwise comparisons between the three wetter pond types. Table 4.3: Pairwise Tukey test results for measured soil water content at pond center between adjacent vegetation types on the ABS scale. All non-adjacent pairwise differences were statistically significant at the P < 0.001 level. Comparison Mean Difference 95% CI Significance 1 Hypericum-Broomsedge 0.15 -0.01–0.30 2 Redroot-Hypericum 0.22 0.07–0.38 ** 3 Maidencane-Redroot 0.20 0.35–0.04 * Mixed effects modeling Incorporating the collected data into a mixed effects model enables a more robust comparison of the soil water fraction trends observed within and between pond types. The apparent variation in slopes between pond types shown in Figure 4.3 supports the inclusion of the interaction term between pond type and transect distance in the mixed model formula. Modeled slope and intercept estimates for each pond type are shown in Table 4.4. Table 4.4: Mixed model intercept and slope values for modeled soil water content by weight fraction. Significance codes reflect difference between model estimate and 0. Full raw coefficient estimates listed in Table C.1 in Appendix C. Broomsedge Hypericum Redroot Maidencane Intercept 0.13 ± 0.03*** 0.17 ± 0.04*** 0.39 ± 0.04*** 0.57 ± 0.04*** Slope -0.06 ± 0.04 -0.17 ± 0.05** -0.27 ± 0.05*** -0.33 ± 0.05*** The water fraction model had a marginal R2 of 0.904 which only accounts for the fixed effects of pond type and transect location, and a conditional R2 value of 0.913, which incorporates both fixed effects and the random pond effect. Slopes 98 were steepest for the wettest Maidencane pond type, denoting the largest difference in soil water fraction from pond center to edge, and slopes were flattest for the driest Broomsedge pond type, confirming our hypothesized soil water fraction gradient within and between pond types. ANOVA test F-statistics for each effect included in the model show that tran- sect distance averaged across all pond types had the largest effect size, and ANOVA p-values indicate that soil water fraction is significantly different between pond types averaged over transect distance as well (Table 4.5). These results indicate that the strongest predictor of soil water fraction in the ABS ponds is actually the sample location within a given pond type rather than the type of pond sampled. The statistical significance of the interaction between transect distance and pond type indicates that the slopes of soil water fraction along transect distance are sig- nificantly different between pond types. A plot of ranked predictions vs. residuals for this mixed effects model is shown in Appendix C (Figure C.1). Table 4.5: Mixed effects model ANOVA table for soil water content including degrees of freedom. All listed effects were statistically significant at the P < 0.001 level. Mean square Numerator df Denominator df F-statistic Pond type 0.23 3 16.82 76.90 Transect distance 0.56 1 45.81 189.12 Interaction 0.04 3 45.81 15.02 Post-hoc model testing The ‘emmeans’ package was used to contrast the estimated marginal mean (EMM) of the mixed effects model’s prediction for soil water fraction at pond center be- tween pond types [Lenth et al., 2024], and these pairwise differences were statisti- cally significant (P < 0.01) between all adjacent pond types (Table 4.6). To evaluate differences in modeled slope relating transect distance to soil wa- 99 Table 4.6: Estimated marginal mean (EMM) contrasts for modeled soil water content at pond center between adjacent pond types. Pond type comparison Difference in EMM Degrees of freedom 1 Hypericum - Broomsedge 0.17 ± 0.04** 16.15 2 Redroot - Hypericum 0.22 ± 0.04*** 16.55 3 Maidencane - Redroot 0.18 ± 0.04*** 19.48 ter fraction for each pond type, estimated marginal means of linear trends were contrasted between adjacent pond types (Table 4.7). Table 4.7: Estimated marginal means of linear trends between adjacent pond types for modeled soil water content across all transect distances. Pond type comparison Slope difference estimate SE df T-statistic 1 Hypericum - Broomsedge -0.17 ± 0.129** 0.13 46.04 -3.21 2 Redroot - Hypericum -0.11 ± 0.126 0.13 46.08 -2.03 3 Maidencane - Redroot -0.06 ± 0.125 0.13 46.13 -1.08 These slope estimates indicate that the significant difference in modeled slope identified in the model ANOVA’s interaction term (Table 4.5) is only statistically significant (P < 0.01) between the two driest adjacent pond types, Broomsedge and Hypericum. This finding suggests that the more significant differences in pond center soil water content EMM between adjacent pond types shown in Table 4.6, which effectively represent the y-intercept for each pond type’s soil water content slope, drive more of the overall separation between soil water fraction profiles for each pond type evident in the collected data (Figure 4.3). 4.3.2 Soil organic matter SOM measured in the same soil cores used for soil water content analysis was positively correlated with soil water fraction (R2 = 0.71) as shown in Figure 4.4, and this positive relationship is statistically significant (P < .001), confirming our hypothesis that soil samples with higher soil water fraction will have higher SOM. 100 Figure 4.4: Relationship between soil water content and soil organic content, both by fractional weight. Red trendline shows linear fit to all data with 95% confidence interval: Adj. R2 = 0.7134, Intercept = -0.0288, Slope = 0.4872, P < 0.001 4.4 Discussion Based on the soil water fraction data collected during this study and the derived mixed effects model, the 1984 ABS wetness scale for dominant pond vegetation type successfully predicts relative differences in dry season soil water fraction across the four pond types sampled and tested in this study, particularly at the center of each pond. This finding is statistically significant and suggests that the four-decade old vegetation map used to design this study is still quite useful in formulating research questions and hypotheses related to soil moisture at this study site. The finding of a larger soil water fraction effect size for transect distance rather than pond type in Table 4.5 indicates that the soil water fraction gradient is actually stronger within pond types than between pond types. This finding suggests that 101 within-pond hydroperiod variability due to differences in pond bathymetry or soil type from pond center to edge could be the primary determinant of vegetation species distribution in the seasonal ponds. This result is a significant contribution to the original work of [Abrahamson et al., 1984] and others to develop the pond classification system at ABS. The consistency of vegetation type in each pond’s center relative to the his- torical map reflects the protection of ABS as an ecological research reserve. One potential cause of vegetation change over recent decades could have been the robust prescribed burning program at ABS which was accounted for in pond selection, although prescribed burning at this site has been recorded dating back to 1967 or earlier [Abrahamson et al., 1984], well before the historical map was published. Invasive feral hogs’ impact on vegetation could also have caused plant community changes in the studied ponds, and hog rooting impacts were observed in 2.8% of nearly 3500 monthly water level readings from ABS’ long-term monitoring set of seasonal ponds collected between 2018 and 2024. While free-ranging domestic swine have been present in Florida for centuries and substantial feral swine popu- lations were well-established in the state by the 1980s, their population continues to increase [Hernández et al., 2018]. The strong correlation between pond type and central soil water fraction evi- dent in the data and model results in section 4.3.1 suggest that pond type could predict additional ecosystem characteristics associated with soil water fraction, and that these correlations could exist at a larger scale beyond the eight ponds sampled in this study. Future work on the ABS pond classification system could explore the causal basis for the soil water fraction trends identified here by collect- ing data on the physical factors that may influence seasonal pond hydroperiod and thus dominant vegetation species: factors such as pond depth, area and bottom 102 profile and soil type likely play a role in determining soil water fraction and thus vegetation distribution. Similarly, the species in question could be grown in meso- cosms or other laboratory experiments that examine the tolerance of each species to different soil water fraction levels, which would establish probable ranges of soil water fraction for ponds of each vegetation type that could be compared to field data. There is a statistically significant positive correlation between SOM and soil water fraction, with 71% of the observed trend in SOM explained by soil water fraction (Figure 4.4). This positive relationship illustrates a potential effect of the hydroperiod gradient within and across the different pond types at ABS, with wetter pond zones potentially building more organic-rich soil and accumulating more carbon. This trend could be explained by potential increases in primary pro- ductivity and higher rates of organic matter accumulation [Sahrawat, 2004], or the inhibition of organic matter oxidation in more frequently-flooded ponds [Qiu et al., 2023]. In wetland soils, elevated soil water content slows down decomposition by creating anaerobic conditions, and rewetting of dried wetland soils can dramati- cally reduce greenhouse gas fluxes [Hao et al., 2025]. Hydroperiod therefore likely exerts a strong influence on soil organic matter in the subtropical soils of ABS, with wetter pond zones’ longer hydroperiods reducing decomposition and promot- ing the accumulation of SOM. The significant relationship identified between SOM and soil moisture presents a potential opportunity to expand on past work at ABS by documenting differential rates of carbon cycling across different pond types. However, much more data would be required to demonstrate the carbon dynam- ics hypothesized herein. While wetter soils typically foster anaerobic conditions which have been shown to enhance carbon sequestration by inhibiting oxidation of SOM [Bridgham et al., 2006, e.g.,], the accumulation of SOM is mediated by 103 a complex set of biological, chemical and physical factors [Schmidt et al., 2011]. Since the measurements presented here represent SOM accumulation on annual timescales far outside the scope of this study’s field campaign, and this accumula- tion is partially mediated by biological processes which likely vary across the four different pond vegetation types, additional causal analyses of SOM data based solely on soil water content were not pursued in this study. Additionally, investigating the temporal dynamics of pond water levels and soil water fraction would require more than one sampling effort at each pond. The timing of this study near the end of the spring dry season at ABS could have resulted in the widest variation in soil water fraction across the four pond types as well as below average values. Sampling during a wet period or drought could have yielded more similar soil moisture values across the ponds samples. This study did observe marginal increases and subsequent declines in water level for three out of five gauged pond measurements after a small rain event (<1cm), but repeated sampling would be necessary to determine the annual dynamics and long- term trends in soil water fraction across these four pond types. ABS monitors a set of seasonal ponds which have graduated staff gauges installed in their lowest points, and water level readings are made on these ponds approximately monthly. Unfortunately, the incorporation of ABS’s long-term pond monitoring data into this study’s analysis was challenging because two of the four pond types sampled in this study (Broomsedge and Redroot) are not represented among the set of gauged ponds where water levels are monitored. While soils are well-mapped at ABS, soil type was not factored into the pond selection criteria for this study, which prioritized identical recent fire history and proximity of selected ponds to other replicates for ease of access in the limited time available to conduct fieldwork during the field course. It is also likely that 104 certain pond types preferentially occur in certain soil types at ABS, for example if two ponds of similar shape and depth had soil types with different drainage characteristics that supported different vegetation species which prefer that level of drainage. This likelihood is supported by the soil characterizations described in the results, which found sandier soils in the drier ponds and siltier soils in the wetter ponds. Therefore, it may not be feasible to implement a truly randomized comparison of pond type and soil type at ABS, as the wettest pond type likely does not occur in the most well-drained soil type, nor the driest pond type in soils that retain water. However, the study’s effort to isolate the relationship between pond type and soil moisture through this sampling scheme may limit the conclusions that can be drawn about soil moisture gradients in other areas of ABS or the Lake Wales Ridge with different soil types or fire return intervals. Overall this study has shown that decades-old vegetation maps are useful in developing scientific questions and designing meaningful studies, and that these studies can add to the ecological understanding these maps represent. Similar to inferring long-term vegetation and climate trends from ancient pollen preserved in Lake Annie’s sediments at ABS [Watts, 1980], high-quality vegetation maps will have persistent utility in documenting successional change and climate shifts as long as these resources are preserved and accessible to future scientists. This study’s results can be used to approximate dry season soil water fraction gradients within and between these four pond types at ABS and describe potential trends in SOM. While addressing more complex scientific questions with historical veg- etation maps will inevitably require more involved data collection and analysis, these valuable resources should not go overlooked in efforts to investigate tempo- ral trends in land cover or understand the spatial heterogeneity of environmental conditions that persist today. 105 APPENDIX A CHAPTER 2 APPENDIX: VELOCITY MEASUREMENT UNCERTAINTY Instantaneous pixel displacements calculated in the IR-QIV subwindows along the cross-culvert velocity profile showed relatively few dropouts or spurious mea- surements throughout the dataset. Histograms of these pixel displacements were generally symmetrical with relatively few outliers and displacements within the maximum search radius set for the IR-QIV algorithm. All calculated pixel displace- ments for subwindows along the cross-culvert velocity profile used to construct the river-side rating curve are shown in Figure A.1. Figure A.1: Histogram of all vertical (i) and horizontal (j) pixel displacements for subwindows along river-side cross-culvert velocity profile. Horizontal axis scale set by IR-QIV pattern tracking algorithm maximum search radius of ±48 pixels. 106 The proportion of invalid results returned by the pattern tracking algorithm for these subwindows, denoted as NaN (not a number), was assessed for each surface velocity measurement (Figures A.2 & A.3). All timesteps exhibited 5% or less spurious measurements. Figure A.2: Percentage of NaN’s induced by IR-QIV pattern tracking algorithm for subwindows along cross-culvert velocity profile for river-side measurements. 107 Figure A.3: Percentage of NaN’s induced by IR-QIV pattern tracking algorithm for subwindows along cross-culvert velocity profile for bay-side measurements. Additionally, the bay side flow into the single bi-directional culvert exhibited recirculating eddies at the edge of the culvert during certain times, which produced areas of positive (outflowing) velocities along the cross-culvert velocity profile that did not actually represent outflow from the estuary during this period of flood tide. Therefore these areas of positive velocity were excluded from the cross-culvert velocity profile and average velocity calculations (Figure A.4). While velocities were measured in the same pixel locations at all timesteps, the overall physical length of the culvert velocity profile changes with the water level relative to the camera, with higher water levels corresponding to a shorter cross-channel velocity measurement line. 108 Figure A.4: Proportion of cross-culvert velocity profile sampled for pre- and post- submergence periods of tidal inflow to the culvert. Submerged portion in center not shown as velocities were not included in the rating curve calculation. A.1 River-side comparison with ADV velocities A Nortek Vector acoustic Doppler velocimeter (ADV) was deployed on the river side of the dike at the entrance of the bidirectional culvert (Figure A.5), mea- suring x-y-z velocities at 16 Hz (velocity range 7 m/s). The ADV mounting was adjusted multiple times to ensure submergence with changing water depth, proper head orientation, and secure attachment during high-velocity aerated tidal inflows through this gate. Inflow ADV data exhibit an extremely noisy high velocity signal, highlighting the challenges of measuring water velocity exiting the dike culverts. Outflow data exhibit more coherent velocity signals, but biomass debris caught on the ADV at various points disrupting measurements. 109 Figure A.5: Thermal infrared image used for IR-QIV analysis and inset photo showing deployment location of ADV on river-side of two-directional culvert rela- tive to front of pier between culverts. ADV data were time synchronized by referencing field notes regarding ADV deployment, retrieval and adjustment times against the USGS tide gauge data, revealing a time offset of roughly 2 hours which was corrected (Figure A.6). The sign of ADV velocity measurements was reversed to match the discharge velocity sign, since the x-arm of the ADV was pointed upstream. The ADV’s pressure sensor, which is located at the base of the main instrument body approximately 26 cm above the velocity measurement volume, was corrected for atmospheric pressure using USGS flux tower meteorological data [Sanders-Demott et al., 2022]. 110 Figure A.6: Left vertical axis: 1 Hz-averaged U (light blue) and 5-minute centered moving mean streamwise velocity (dark blue) measured by an ADV deployed on the river side of the dike over a three hour period during ebb tide. Right vertical axis: USGS tide gauge elevation head hz across the dike (red) used to time-synchronize ADV data. Positive values indicate outflow on both axes. The x-component of the ADV velocity time series was averaged over the same 11 second periods as the IR-QIV measurements taken during the coherent outflow period shown in Figure A.6 beginning at 3:25 - every five minutes synchronized with the tide gauge measurements. These average ADV streamwise velocities were compared with the normal component of the corresponding IR-QIV surface velocity vector nearest to the ADV deployment location (approximately 20 cm upstream) in the cross-channel measurement profile (Figure A.7). 111 Figure A.7: Scatterplot comparing time-synchronized velocity measurements be- tween the ADV and nearest IR-QIV subwindow sampled for rating curve velocity. The inner color gradient for each point shows the timestamp after 3:25 on July 14, 2022, and the outer color gradient shows the approximate depth of the ADV measurement volume beneath the water surface at each measurement time. These results indicate that the ADV velocity measurements were significantly higher than the IR-QIV measurements over this time period, roughly double at any given timestep. Figure A.7 also indicates that the difference between ADV and IR-QIV velocity measurements increased as the ADV measurement depth got closer to the surface; these measurements took place over a period of increasing velocity and decreasing water surface elevation, so the increasing difference could be a product of increasing velocities with a constant proportional difference. The low end of the depth color scale in Figure A.5 represents the shallowest depth the 112 ADV pressure sensor could measure in its deployment orientation (approximately 26 cm), so some of the measurements at this depth were les than 26 cm below the surface. While the ADV was fixed in place, the deployment depth effectively changed by nearly a meter due to the changing water surface elevation as shown in the outer circle color in Figure A.7, and these data do not indicate a vertical velocity profile with maximum velocities below the surface. A.2 River-side acceleration While the results in figure A.7 suggest a significant discrepancy between the ADV and IR-QIV velocity measurements, these measurement locations are not precisely co-located, with the IR-QIV results taken from a subwindow approximately 20 cm upstream of the ADV. To measure acceleration between the shadow transect where IR-QIV velocity measurements were sampled in constructing the rating curve and the true culvert entrance, the product of streamwise velocity and acceleration U du dx was calculated for two streamwise lines of subwindows (Figure A.8). The longer of these subwindow lines (8 subwindows) was oriented near the center of the central culvert, and the shorter line (6 subwindows) terminated at the ADV location. 113 Figure A.8: Locations of IR-QIV subwindows sampled to calculate acceleration in the central culvert (upper red points) and nearest the ADV (lower red points). The closest subwindows on the upstream side of the red shadow transect are shown as black points. Backwards finite difference was used to approximate du dx using the velocity change vs. the distance between subwindows; the backwards scheme excludes the farthest subwindow from the culvert and resolves acceleration at the closest subwindow to the culvert entrance in each line. The spatial distribution of U du dx along the two lines of subwindows was therefore centered at the last subwindow up- stream of the shadow transect (Figures A.9 & A.10). U du dx was calculated over the same time period of comparison between IR-QIV and ADV velocity measurements shown in Figure A.7. 114 Figure A.9: Values of U du dx calculated by backward finite difference for the cen- tral culvert (upper line of subwindows in Figure A.8), with color coded timesteps matching the central point color in Figure A.7. x=0 denotes the location of the last subwindow upstream of the shadow transect. 115 Figure A.10: Values of U du dx calculated by backward finite difference for the ADV location (lower line of subwindows in Figure A.8), with color coded timesteps matching the central point color in Figure A.7. x=0 denotes the location of the last subwindow upstream of the shadow transect. The central culvert data shown in Figure A.9 exhibits rapid acceleration at the last subwindow before the culvert entrance, particularly in the last hour of the comparison time window when hz was past its peak in Figure A.6. The ADV location data shown in Figure A.10 show a similar period of rapid acceleration, but at the beginning of the time series when velocities were lowest as shown by the same color points in Figure A.7, with U du dx falling to values near or less than zero at later timesteps. This apparent deceleration at the culvert entrance near the ADV is likely caused by spurious surface vectors immediately next to the ADV as the instrument’s affect on the surface flow signature grew at higher velocities. 116 These results indicate significant acceleration between the shadow transect where IR-QIV velocity measurements are sampled for the rating curve and the entrance to the culvert at certain times during the tidal cycle. Taking the U du dx values from the closest subwindow to the central culvert en- trance (x >0.6 m in Figure A.9) at each timestep in Figure A.7 and applying the acceleration component to the IR-QIV velocity measurement over the distance to the ADV, we can estimate and correct for the potential acceleration between the IR-QIV and ADV velocity measurement locations. This corrective addition somewhat improves the agreement between the IR-QIV and ADV velocity mea- surements (Figure A.11). 117 Figure A.11: Scatterplot comparing time-synchronized velocity measurements be- tween the ADV and acceleration-adjusted IR-QIV subwindow sampled for rating curve velocity. The inner color gradient for each point shows the timestamp after 3:25 on July 14, 2022, and the outer color gradient shows the approximate depth of the ADV measurement volume beneath the water surface at each measurement time. A.3 Bay-side mass conservation The field of view shown in the bay-side imagery was roughly aligned with the incoming tidal flow, such that much of the approaching flow passed through the entire image as it converged into the culvert. This perspective allowed for an eval- uation of mass conservation between streamlines that traversed the entire image (Figure A.12). 118 Figure A.12: Vector field of flow into bay-side culvert annotated with mass con- servation evaluation scheme. The widest apart streamlines that reach from the culvert to the far edge of the image are shown in red. Surface velocities normal to the transverse profiles are shown using the color gradient. Fence posts above the culvert entrance are visible along the upper edge of the image. The average vector field across the full image was calculated at each five minute interval corresponding to a tide gauge measurement during the flood tide observed. This vector field was numerically integrated to solve for streamlines that pass from the far edge of the image into culvert entrance. 13 out of 85 timesteps during the seven hours of flood tide observed had at least two streamlines that met this criteria. Transverse profiles perpendicular to one of the streamlines were generated at locations 25%, 50% and 75% of the longer image axis X away from the culvert 119 edge of the image. The normal component of the vector field was calculated along each transverse profile, and the average velocity at each profile < Unormal > was multiplied by the length of the profile Lp to derive a surface flux [m2/s] for each transverse profile at each of these timesteps (Figure A.13). Figure A.13: Left axis: absolute value of surface flux at four transverse profiles for timesteps with two valid streamlines (shown as points connected by colored solid lines), following scheme in Figure A.12. Profile 3X/4 is farthest from the culvert entrance. Data collection continued at this location during flood tide until 4:00 on July 15 but only the streamline orientations at the points shown here allowed for the calculations described above. Right axis: ratio of distance from internal sluicegate to culvert entrance Ls (3 m) divided by water depth to bottom of sluicegate hs shown as dashed black line These results indicate that the surface mass flux is variably conserved between the far field transverse profiles and the culvert. Surface flux at the culvert, shown in blue, was not highest among the transverse profiles at any of the timesteps eval- uated. The culvert entrance surface flux ranged between 15-96% of the maximum 120 surface flux between the three other transverse profiles at any given timestep, with an average of 47% of the maximum surface flux at the culvert entrance. There ap- pears to be no clear relationship between mass conservation across the transverse profiles and the sluicegate submergence depth, so these results do not suggest a backwash eddy behind the sluicegate. Georeferencing uncertainty is highest in the far field away from the culvert entrance, due in part to the location of GCPs used for georectification on the dike in the near field. The georectified field of view shown in Figure A.12, in which the camera was located near the upper right corner, exhibits limited perspective distortion due to the relatively near-nadir viewing angle. If this field of view is underestimating perspective distortion because of georeferencing error, the far field in the lower left of the image would actually represent a larger spatial scale, which would in turn increase the velocities and length scales of the transverse profiles in this region. Therefore the larger surface fluxes measured at the far field transverse profiles X/2 and 3X/4 are likely not the result of georeferencing error, and could actually be underestimates. 121 APPENDIX B CHAPTER 3 APPENDIX: SPACE-TIME DIAGRAM CONTRAST ENHANCEMENT METHODS In an effort to constrain the confidence interval of individual sampling lines’ depth and velocity measurements, contrast enhancement algorithms were applied to the space-time diagrams before slope detection. Noting the wave and bulk veloc- ity slopes’ higher pixel intensities in Figure 3.4 relative to the background, values below the median pixel intensity were coerced to zero. This pre-processing of the space-time diagrams yielded virtually the same flow rate estimate and confidence interval, shown in Figure B.1. 122 Figure B.1: Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. Contrast-enhancement pre- processing of space-time diagrams was implemented before slope detection, with values below the 50th percentile pixel intensity coerced to zero. The reduced ratio of the bulk velocity peak to the inter-peak minium from the baseline case to the 50th percentile cutoff motivated the testing of a more aggressive contrast enhancement threshold. In the third test case the 80th percentile was implemented as the space-time diagram pixel intensity minimum, which once again produced the same stable flow rate and confidence interval of 3.3 ± 0.6 m3/s as shown in Figure B.2. 123 Figure B.2: Cross-channel velocity (top) and depth (bottom) profiles with boot- strapped 95% confidence intervals shown as error bars, with the average flow rate and confidence interval noted above the depth plot. Contrast-enhancement pre- processing of space-time diagrams was implemented before slope detection, with values below the 80th percentile pixel intensity coerced to zero. Notably, the results in Figures B.1 and B.2 produce the same flow rate of 3.3 ± 0.6 m3/s as in the original analysis in Figure 3.5. Averaging across the 22 sampling lines, the bootstrapped discharge estimates are relatively stable despite high variance in detected depth and velocity along each sampling line as evidenced by the large 95% confidence intervals above. Next steps for this analysis will involve starting from this method development timestep when the width of the inundated channel is clearly visible in the imagery, and analyzing the rest of the imagery collected on August 1 in three-minute incre- ments, adjusting the WSE using the continuous curve fit in Figure 3.2. The WSE 124 and the wave-derived depth profile will be used to estimate the edge location of the overwash channel at each timestep, excluding sampling lines that fall outside the channel at lower water levels, and extrapolating the larger width of the chan- nel beyond the initial transect at peak water levels. Our ultimate goal with this dataset is to derive a comprehensive flow rate for Duck Harbor at a given tidal water level by combining observations from the north overwash channel on July 31 with those from the south channel on August 1 and 2, effectively yielding a stage-discharge relationship for the entire washover fan channel system. 125 APPENDIX C CHAPTER 4 APPENDIX: MIXED EFFECTS MODEL SUPPORTING INFORMATION −0.04 0.00 0.04 0 20 40 60 Ranked prediction R e s id u a ls Figure C.1: Ranked predictions vs. residuals for mixed effects model described in 4.3.1 126 Table C.1: Raw coefficient estimates for mixed effects model model described in 4.3.1 Term Estimate Std. error t value p value 1 Broomsedge (null intercept) 0.13 0.06 2.12 0.10 2 Hypericum (intercept) 0.17 0.08 2.04 0.11 3 Redroot (intercept) 0.39 0.08 4.62 0.01 4 Maidencane (intercept) 0.57 0.08 6.74 0.00 5 Broomsedge * transect distance (null slope) -0.06 0.10 -0.60 0.58 6 Hypericum * transect distance (slope) -0.17 0.14 -1.20 0.30 7 Redroot * transect distance (slope) -0.28 0.14 -1.98 0.12 8 Maidencane * transect distance (slope) -0.33 0.14 -2.33 0.08 127 BIBLIOGRAPHY Ehsan Abolfazli and Kyle Strom. Salinity Impacts on Floc Size and Growth Rate With and Without Natural Organic Matter. Journal of Geophys- ical Research: Oceans, 128(7):e2022JC019255, 2023. ISSN 2169-9291. doi:10.1029/2022JC019255. Warren G. Abrahamson, Ann F. Johnson, James N. Layne, and Patricia A. Per- oni. Vegetation of the Archbold Biological Station, Florida: An Example of the Southern Lake Wales Ridge. Florida Scientist, 47(4):209–250, 1984. ISSN 0098-4590. Warren G. Abrahamson, Christy R. Abrahamson, and Matthew A. Keller. Lessons from four decades of monitoring vegetation and fire: Maintaining diversity and resilience in Florida’s uplands. Ecological Monographs, 91(2):e01444, 2021. ISSN 1557-7015. doi:10.1002/ecm.1444. Abrahamson, Warren G., Ann F. Johnson, and James N. Layne. Archbold Biolog- ical Station Vegetation Map, 1984. Mahmood M. Al-mamari, Sameh A. Kantoush, Sohei Kobayashi, Tetsuya Sumi, and Mohamed Saber. Real-Time Measurement of Flash-Flood in a Wadi Area by LSPIV and STIV. Hydrology, 6(1):27, March 2019. ISSN 2306-5338. doi:10.3390/hydrology6010027. William P. Anderson Jr. Aquifer Salinization from Storm Overwash. Journal of Coastal Research, 18(3):413–420, 2002. ISSN 0749-0208. S. Ansari, C. D. Rennie, E. C. Jamieson, O. Seidou, and S. P. Clark. RivQNet: Deep Learning Based River Discharge Estimation Using Close-Range Water Sur- 128 https://doi.org/10.1029/2022JC019255 https://doi.org/10.1002/ecm.1444 https://doi.org/10.3390/hydrology6010027 face Imagery. Water Resources Research, 59(2):e2021WR031841, 2023. ISSN 1944-7973. doi:10.1029/2021WR031841. Edward B. Barbier, Sally D. Hacker, Chris Kennedy, Evamaria W. Koch, Adrian C. Stier, and Brian R. Silliman. The value of estuarine and coastal ecosys- tem services. Ecological Monographs, 81(2):169–193, 2011. ISSN 1557-7015. doi:10.1890/10-1510.1. Douglas Bates, Martin Maechler, Ben Bolker [aut, cre, Steven Walker, Rune Haubo Bojesen Christensen, Henrik Singmann, Bin Dai, Fabian Scheipl, Ga- bor Grothendieck, Peter Green, John Fox, Alexander Bauer, Pavel N. Kriv- itsky (shared copyright on simulate.formula), Emi Tanaka, and Mikael Jagan. Lme4: Linear Mixed-Effects Models using ’Eigen’ and S4, July 2024. K. M. Befus, A. P. D. Kurnizki, K. D. Kroeger, M. J. Eagle, and T. P. Smith. Forecasting Sea Level Rise-driven Inundation in Diked and Tidally Re- stricted Coastal Lowlands. Estuaries and Coasts, May 2023. ISSN 1559-2731. doi:10.1007/s12237-023-01174-1. Greg Berman. Longshore Sediment Transport Cape Cod Massachusetts. Technical report, Woods Hole Sea Grant, 2011. Hamish John Biggs, Brendon Smith, Martin Detert, and Hamish Sutton. Surface image velocimetry: Aerial tracer particle distribution system and techniques for reducing environmental noise with coloured tracer particles. River Research and Applications, 38(6):1192–1198, 2022. ISSN 1535-1467. doi:10.1002/rra.3973. David M. Bjerklie, Kevin O’Brien, and Ron Rozsa. A one-dimensional diffusion analogy model for estimation of tide heights in selected tidal marshes in Con- necticut. Technical report, January 2013. 129 https://doi.org/10.1029/2021WR031841 https://doi.org/10.1890/10-1510.1 https://doi.org/10.1007/s12237-023-01174-1 https://doi.org/10.1002/rra.3973 Kyle W. Bostick, Shelley A. Johnson, and Martin B. Main. Florida’s Geological History. Fact Sheet WEC189, Department of Wildlife Ecology and Conservation, UF/IFAS Extension, Gainesville, FL, July 2022. Mary Ann Bragg. New bridge part of Herring River project. Cape Cod Times, 2014. Scott D. Bridgham, J. Patrick Megonigal, Jason K. Keller, Norman B. Bliss, and Carl Trettin. The carbon balance of North American wetlands. Wet- lands, 26(4):889–916, December 2006. ISSN 1943-6246. doi:10.1672/0277- 5212(2006)26[889:TCBONA]2.0.CO;2. Sophia E. Brumer, Christopher J. Zappa, Steven P. Anderson, and John P. Dugan. Riverine skin temperature response to subsurface processes in low wind speeds. Journal of Geophysical Research: Oceans, 121(3):1721–1735, 2016. ISSN 2169- 9291. doi:10.1002/2015JC010746. Daniel Campbell, Paul A. Keddy, Michaelyn Broussard, and Tiffany B. McFalls- Smith. Small Changes in Flooding Have Large Consequences: Experimental Data from Ten Wetland Plants. Wetlands, 36(3):457–466, June 2016. ISSN 1943-6246. doi:10.1007/s13157-016-0754-7. Michelle T. Casanova and Margaret A. Brock. How do depth, duration and frequency of flooding influence the establishment of wetland plant com- munities? Plant Ecology, 147(2):237–250, April 2000. ISSN 1573-5052. doi:10.1023/A:1009875226637. Katherine A. Castagno, Eliza J. Fitzgerald, Kaitlyn Button, Petra Zuñiga, Tommy Tucker, Timothy Smith, and Mark Borrelli. Washover fan deposits resulting from perigean spring tides: An example from Cape Cod, Mas- 130 https://doi.org/10.1672/0277-5212(2006)26[889:TCBONA]2.0.CO;2 https://doi.org/10.1672/0277-5212(2006)26[889:TCBONA]2.0.CO;2 https://doi.org/10.1002/2015JC010746 https://doi.org/10.1007/s13157-016-0754-7 https://doi.org/10.1023/A:1009875226637 sachusetts, USA. Geomorphology, 470:109536, February 2025. ISSN 0169-555X. doi:10.1016/j.geomorph.2024.109536. C. Chris Chickadel, Alexander R. Horner-Devine, Stefan A. Talke, and Andrew T. Jessup. Vertical boil propagation from a submerged estuarine sill. Geophysical Research Letters, 36(10), 2009. ISSN 1944-8007. doi:10.1029/2009GL037278. Ross Clark and Kevin O’Connor. A systematic survey of bar-built estuaries along the California coast. Estuarine, Coastal and Shelf Science, 226:106285, October 2019. ISSN 0272-7714. doi:10.1016/j.ecss.2019.106285. Frederic Edward Clements. Plant Indicators: The Relation of Plant Communities to Process and Practice. Carnegie Institution of Washington, 1920. ISBN 978- 0-598-34455-7. Rachel E. S. Clemesha, Alexander Gershunov, Sam F. Iacobellis, A. Park Williams, and Daniel R. Cayan. The northward march of summer low cloudiness along the California coast. Geophysical Research Letters, 43(3):1287–1295, 2016. ISSN 1944-8007. doi:10.1002/2015GL067081. E. A. Cowen and S. G. Monismith. A hybrid digital particle tracking velocimetry technique. Experiments in Fluids, 22(3):199–211, January 1997. ISSN 1432-1114. doi:10.1007/s003480050038. Edwin A. Cowen. Agw filter.m [Software], 2006. Henry Chandler Cowles. The Physiographic Ecology of Chicago and Vicinity; A Study of the Origin, Development, and Classification of Plant Societies (Concluded). Botanical Gazette, 31(3):145–182, March 1901. ISSN 0006-8071. doi:10.1086/328089. 131 https://doi.org/10.1016/j.geomorph.2024.109536 https://doi.org/10.1029/2009GL037278 https://doi.org/10.1016/j.ecss.2019.106285 https://doi.org/10.1002/2015GL067081 https://doi.org/10.1007/s003480050038 https://doi.org/10.1086/328089 John T. Curtis. The Vegetation of Wisconsin: An Ordination of Plant Communi- ties. University of Wisconsin Pres, November 1959. ISBN 978-0-299-01943-3. Robert George Dean and Robert A. Dalrymple. Water Wave Mechanics for Engi- neers and Scientists. Number 2 in Advanced Series on Ocean Engineering. World scientific, Singapore New Jersey London [etc.], 1991. ISBN 978-981-02-0420-4. M. Detert. How to Avoid and Correct Biased Riverine Surface Image Velocime- try. Water Resources Research, 57(2):e2020WR027833, 2021. ISSN 1944-7973. doi:10.1029/2020WR027833. Suzanne J. DeVries-Zimmerman, Brian Yurk, Kathleen M. Fast, Alexandra Don- aldson, and Edward C. Hansen. Waxing and waning slacks: The changing ecohy- drology of interdunal wetlands/slacks in a Lake Michigan coastal dune complex during rising Lake Michigan-Huron levels. Journal of Great Lakes Research, 47 (6):1565–1580, December 2021. ISSN 03801330. doi:10.1016/j.jglr.2021.09.001. Chantal Donnelly, Nicholas Kraus, and Magnus Larson. State of Knowledge on Measurement and Modeling of Coastal Overwash. Journal of Coastal Research, 224:965–991, July 2006. ISSN 0749-0208, 1551-5036. doi:10.2112/04-0431.1. J. P. Dugan and C. C. Piotrowski. Measuring currents in a coastal inlet by ad- vection of turbulent eddies in airborne optical imagery. Journal of Geophysical Research: Oceans, 117(C3), 2012. ISSN 2156-2202. doi:10.1029/2011JC007600. J. P. Dugan, S. P. Anderson, C. C. Piotrowski, and S. B. Zuckerman. Air- borne Infrared Remote Sensing of Riverine Currents. IEEE Transactions on Geoscience and Remote Sensing, 52(7):3895–3907, July 2014. ISSN 1558-0644. doi:10.1109/TGRS.2013.2277815. 132 https://doi.org/10.1029/2020WR027833 https://doi.org/10.1016/j.jglr.2021.09.001 https://doi.org/10.2112/04-0431.1 https://doi.org/10.1029/2011JC007600 https://doi.org/10.1109/TGRS.2013.2277815 Ruth Durán, Jorge Guillén, Antonio Ruiz, José A. Jiménez, and Enric Sagristà. Morphological changes, beach inundation and overwash caused by an extreme storm on a low-lying embayed beach bounded by a dune system (NW Mediter- ranean). Geomorphology, 274:129–142, December 2016. ISSN 0169-555X. doi:10.1016/j.geomorph.2016.09.012. Mat́ıas Eder, Leticia Tarrab, Román Martino, Leandro Masso, Antoine Pata- lano, Mat́ıas Ragessi, Gerardo Hillman, Andrés Rodriguez, and Mariana Pagot. LSPIV measurements to assess the impact of a bridge on a weakly undulat- ing flow. Journal of Applied Water Engineering and Research, 11(4):512–522, October 2023. ISSN null. doi:10.1080/23249676.2023.2173676. Bradley Efron and R. J. Tibshirani. An Introduction to the Bootstrap. Chapman and Hall/CRC, New York, May 1994. ISBN 978-0-429-24659-3. doi:10.1201/9780429246593. Wade H. Elmer. A Tripartite Interaction Between Spartina alterniflora, Fusar- ium palustre, and the Purple Marsh Crab (Sesarma reticulatum) Contributes to Sudden Vegetation Dieback of Salt Marshes in New England. Phytopathology®, 104(10):1070–1077, October 2014. ISSN 0031-949X. doi:10.1094/PHYTO-08- 13-0219-R. A. Eltner, D. Mader, N. Szopos, B. Nagy, J. Grundmann, and L. Bertalan. Us- ing thermal and RGB UAV imagery to measure surface flow velocities of rivers. The International Archives of the Photogrammetry, Remote Sensing and Spa- tial Information Sciences, XLIII-B2-2021:717–722, June 2021. ISSN 1682-1750. doi:10.5194/isprs-archives-XLIII-B2-2021-717-2021. Anette Eltner, Hannes Sardemann, and Jens Grundmann. Technical Note: Flow velocity and discharge measurement in rivers using terrestrial and unmanned- 133 https://doi.org/10.1016/j.geomorph.2016.09.012 https://doi.org/10.1080/23249676.2023.2173676 https://doi.org/10.1201/9780429246593 https://doi.org/10.1094/PHYTO-08-13-0219-R https://doi.org/10.1094/PHYTO-08-13-0219-R https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-717-2021 aerial-vehicle imagery. Hydrology and Earth System Sciences, 24(3):1429–1445, March 2020. ISSN 1027-5606. doi:10.5194/hess-24-1429-2020. Robert Ettema, Ichiro Fujita, Marian Muste, and Anton Kruger. Particle-image ve- locimetry for whole-field measurement of ice velocities. Cold Regions Science and Technology, 26(2):97–112, October 1997. ISSN 0165-232X. doi:10.1016/S0165- 232X(97)00011-6. Sergio Fagherazzi, Muriel Hannion, and Paolo D’Odorico. Geomorphic structure of tidal hydrodynamics in salt marsh creeks. Water Resources Research, 44(2), 2008. ISSN 1944-7973. doi:10.1029/2007WR006289. Lenore Fahrig, Bruce Hayden, and Robert Dolan. Distribution of Barrier Island Plants in Relation to Overwash Disturbance: A Test of Life History Theory. Journal of Coastal Research, 9(2):403–412, 1993. ISSN 0749-0208. Iain Fairley, Nicholas King, Jason McIlvenny, Matthew Lewis, Simon Neill, Ben- jamin J. Williamson, Ian Masters, and Dominic E. Reeve. Intercomparison of surface velocimetry techniques for drone-based marine current characterization. Estuarine, Coastal and Shelf Science, 299:108682, April 2024. ISSN 0272-7714. doi:10.1016/j.ecss.2024.108682. Philip Frenzel, Olivia Brunings, Karen J. Esler, David le Maitre, and Alanna J. Rebelo. How Well Do Endemic Wetland Plant Species Perform in Wa- ter Purification? Wetlands, 44(8):124, November 2024. ISSN 1943-6246. doi:10.1007/s13157-024-01871-y. Carl T. Friedrichs and James E. Perry. Tidal Salt Marsh Morphodynamics: A Synthesis. Journal of Coastal Research, pages 7–37, 2001. ISSN 0749-0208. 134 https://doi.org/10.5194/hess-24-1429-2020 https://doi.org/10.1016/S0165-232X(97)00011-6 https://doi.org/10.1016/S0165-232X(97)00011-6 https://doi.org/10.1029/2007WR006289 https://doi.org/10.1016/j.ecss.2024.108682 https://doi.org/10.1007/s13157-024-01871-y I. Fujita, Y. Notoya, K. Tani, and S. Tateguchi. Efficient and accurate estimation of water surface velocity in STIV. Environ Fluid Mech, 19(5):1363–1378, October 2019. ISSN 1573-1510. doi:10.1007/s10652-018-9651-3. Ichiro Fujita. Discharge Measurements of Snowmelt Flood by Space-Time Image Velocimetry during the Night Using Far-Infrared Camera. Water, 9(4):269, April 2017. ISSN 2073-4441. doi:10.3390/w9040269. Kerry T. Givens, James N. Layne, Warren G. Abrahamson, and Susan C. White- Schuler. Structural Changes and Successional Relationships of Five Florida Lake Wales Ridge Plant Communities. Bulletin of the Torrey Botanical Club, 111(1): 8–18, 1984. ISSN 0040-9618. doi:10.2307/2996205. H. A. Gleason. The Individualistic Concept of the Plant Association. Bulletin of the Torrey Botanical Club, 53(1):7–26, 1926. ISSN 0040-9618. doi:10.2307/2479933. Juan A González, Charles S Melching, and Kevin A Oberg. Analysis of open- channel velocity measurements collected with an acoustic doppler current pro- filer. RIVERTECH 96, 1996. W Gregory Shriver, Thomas P Hodgman, James P Gibbs, and Peter D Vickery. Landscape context influences salt marsh bird diversity and area requirements in New England. Biological Conservation, 119(4):545–553, October 2004. ISSN 0006-3207. doi:10.1016/j.biocon.2004.01.016. L. Gui and W. Merzkirch. A comparative study of the MQD method and several correlation-based PIV evaluation algorithms. Experiments in Fluids, 28(1):36– 44, January 2000. ISSN 1432-1114. doi:10.1007/s003480050005. Erika S. Hammar-Klose, Elizabeth A. Pendleton, E. Robert Thieler, and S. Jeffress 135 https://doi.org/10.1007/s10652-018-9651-3 https://doi.org/10.3390/w9040269 https://doi.org/10.2307/2996205 https://doi.org/10.2307/2479933 https://doi.org/10.1016/j.biocon.2004.01.016 https://doi.org/10.1007/s003480050005 Williams. Coastal vulnerability assessment of Cape Cod National Seashore to sea-level rise. Technical Report 2002-233, U.S. Geological Survey, 2003. Xiaolin Han, Kebing Chen, Qiang Zhong, Qigang Chen, Fujun Wang, and Danxun Li. Two-Dimensional Space-Time Image Velocimetry for Surface Flow Field of Mountain Rivers Based on UAV Video. Front. Earth Sci., 9, June 2021. ISSN 2296-6463. doi:10.3389/feart.2021.686636. Yuefeng Hao, Jiafu Mao, Charles M. Bachmann, Forrest M. Hoffman, Gerbrand Koren, Haishan Chen, Hanqin Tian, Jiangong Liu, Jing Tao, Jinyun Tang, Lingcheng Li, Laibao Liu, Martha Apple, Mingjie Shi, Mingzhou Jin, Qing Zhu, Steve Kannenberg, Xiaoying Shi, Xi Zhang, Yaoping Wang, Yilin Fang, and Yongjiu Dai. Soil moisture controls over carbon sequestration and greenhouse gas emissions: A review. npj Clim Atmos Sci, 8(1):1–14, January 2025. ISSN 2397-3722. doi:10.1038/s41612-024-00888-8. Evan T. Heberlein, Marc Mayes, Bryn E. Morgan, Kelly K. Caylor, Seth A. Schweitzer, and Edwin A. Cowen. Data From: Wind stress effects on drone- based thermal infrared surface velocimetry measurements of tidal flow in an estuary [Dataset], 2025. Felipe A. Hernández, Brandon M. Parker, Cortney L. Pylant, Timothy J. Smyser, Antoinette J. Piaggio, Stacey L. Lance, Michael P. Milleson, James D. Austin, and Samantha M. Wisely. Invasion ecology of wild pigs (Sus scrofa) in Florida, USA: The role of humans in the expansion and colonization of an invasive wild ungulate. Biol Invasions, 20(7):1865–1880, July 2018. ISSN 1573-1464. doi:10.1007/s10530-018-1667-6. K. Todd Holland, Robert A. Holman, and Asbury H. Sallenger. Estimation of Overwash Bore Velocities Using Video Techniques. In Coastal Sediments ’91: 136 https://doi.org/10.3389/feart.2021.686636 https://doi.org/10.1038/s41612-024-00888-8 https://doi.org/10.1007/s10530-018-1667-6 Proceedings of a Specialty Conference on Quantitative Approaches to Coastal Sediment Processes, pages 489–497, Seattle, WA, 1991. ASCE. K.Todd Holland, R.A. Holman, T.C. Lippmann, J. Stanley, and N. Plant. Prac- tical use of video imagery in nearshore oceanographic field studies. IEEE Journal of Oceanic Engineering, 22(1):81–92, January 1997. ISSN 1558-1691. doi:10.1109/48.557542. Harry Hulsing. Measurement of Peak Discharge at Dams by Indirect Method. U.S. Government Printing Office, 1967. Thomas G Huntington and Alana B Spaetzel. Tidal daily discharge and quality assurance data supporting an assessment of water quality and discharge in the Herring River, Wellfleet, Massachusetts, November 2015 &ndash; Septem- ber 2017, 2020. Thomas G. Huntington, Alana B. Spaetzel, John A. Colman, Kevin D. Kroeger, and Robert T. Bradley. Assessment of water quality and discharge in the Her- ring River, Wellfleet, Massachusetts, November 2015 to September 2017. USGS Numbered Series 2020-5120, U.S. Geological Survey, Reston, VA, 2021. David A. Jay and Tobias Kukulka. Revising the paradigm of tidal analysis – the uses of non-stationary data. Ocean Dynamics, 53(3):110–125, September 2003. ISSN 1616-7228. doi:10.1007/s10236-003-0042-y. E. D. Johnson and E. A. Cowen. Remote monitoring of volumetric dis- charge employing bathymetry determined from surface turbulence met- rics. Water Resources Research, 52(3):2178–2193, 2016. ISSN 1944-7973. doi:10.1002/2015WR017736. 137 https://doi.org/10.1109/48.557542 https://doi.org/10.1007/s10236-003-0042-y https://doi.org/10.1002/2015WR017736 E. D. Johnson and E. A. Cowen. Remote determination of the velocity index and mean streamwise velocity profiles. Water Resources Research, 53(9):7521–7535, 2017. ISSN 1944-7973. doi:10.1002/2017WR020504. Allan E. Jones, Amber K. Hardison, Ben R. Hodges, James W. McClelland, and Kevan B. Moffett. An expanded rating curve model to estimate river discharge during tidal influences across the progressive-mixed-standing wave systems. PLoS One, 14(12):e0225758, December 2019. ISSN 1932-6203. doi:10.1371/journal.pone.0225758. Jennifer M. Karberg, Karen C. Beattie, Danielle I. O’Dell, and Kelly A. Omand. Tidal Hydrology and Salinity Drives Salt Marsh Vegetation Restoration and Phragmites australis Control in New England. Wetlands, 38(5):993–1003, Octo- ber 2018. ISSN 1943-6246. doi:10.1007/s13157-018-1051-4. G. N. Kasozi, P. Nkedi-Kizza, and W. G. Harris. Varied Carbon Content of Organic Matter in Histosols, Spodosols, and Carbonatic Soils. Soil Sci- ence Society of America Journal, 73(4):1313–1318, 2009. ISSN 1435-0661. doi:10.2136/sssaj2008.0070. William S. Kearney, Giulio Mariotti, Linda A. Deegan, and Sergio Fagherazzi. Stage-discharge relationship in tidal channels. Limnology and Oceanography: Methods, 15(4):394–407, 2017. ISSN 1541-5856. doi:10.1002/lom3.10168. Michael J. Kennish. Estuaries: Anthropogenic Impacts. In Encyclopedia of Coastal Science, pages 803–811. Springer, Cham, 2019. ISBN 978-3-319-93806-6. doi:10.1007/978-3-319-93806-6 140. Ian M. Kidd, Andrew Fischer, Shuhong Chai, and Jenny A. Davis. A scenario- based approach to evaluating potential environmental impacts following a tidal 138 https://doi.org/10.1002/2017WR020504 https://doi.org/10.1371/journal.pone.0225758 https://doi.org/10.1007/s13157-018-1051-4 https://doi.org/10.2136/sssaj2008.0070 https://doi.org/10.1002/lom3.10168 https://doi.org/10.1007/978-3-319-93806-6_140 barrage installation. Ocean & Coastal Management, 116:9–19, November 2015. ISSN 0964-5691. doi:10.1016/j.ocecoaman.2015.06.016. Y. Kim, M. Muste, A. Hauet, W. F. Krajewski, A. Kruger, and A. Bradley. Stream discharge using mobile large-scale particle image velocimetry: A proof of concept. Water Resources Research, 44(9), 2008. ISSN 1944-7973. doi:10.1029/2006WR005441. Paul J. Kinzel and Carl J. Legleiter. sUAS-Based Remote Sensing of River Discharge Using Thermal Particle Image Velocimetry and Bathymet- ric Lidar. Remote Sensing, 11(19):2317, January 2019. ISSN 2072-4292. doi:10.3390/rs11192317. Paul J. Kinzel, Carl J. Legleiter, and Christopher L. Gazoorian. Reach-Scale Map- ping of Surface Flow Velocities from Thermal Images Acquired by an Uncrewed Aircraft System along the Sacramento River, California, USA. Water, 16(13): 1870, January 2024. ISSN 2073-4441. doi:10.3390/w16131870. Paschalis Koutalakis and George N. Zaimes. River Flow Measurements Utilizing UAV-Based Surface Velocimetry and Bathymetry Coupled with Sonar. Hydrol- ogy, 9(8):148, August 2022. ISSN 2306-5338. doi:10.3390/hydrology9080148. Kevin D. Kroeger, Stephen Crooks, Serena Moseman-Valtierra, and Jianwu Tang. Restoring tides to reduce methane emissions in impounded wetlands: A new and potent Blue Carbon climate change intervention. Sci Rep, 7(1):11914, September 2017. ISSN 2045-2322. doi:10.1038/s41598-017-12138-4. Environmental Laboratory. Corps of Engineers wetlands delineation manual. Wet- lands Research Program Technical Report Y-87-1, U.S. Army Corps of Engi- neers, Waterways Experiment Station, January 1987. 139 https://doi.org/10.1016/j.ocecoaman.2015.06.016 https://doi.org/10.1029/2006WR005441 https://doi.org/10.3390/rs11192317 https://doi.org/10.3390/w16131870 https://doi.org/10.3390/hydrology9080148 https://doi.org/10.1038/s41598-017-12138-4 Nathan J. M. Laxague and Christopher J. Zappa. Observations of mean and wave orbital flows in the ocean’s upper centimetres. Journal of Fluid Mechanics, 887: A10, March 2020. ISSN 0022-1120, 1469-7645. doi:10.1017/jfm.2019.1019. Eunhee Lee, Heesung Yoon, Sung Pil Hyun, William C. Burnett, Dong-Chan Koh, Kyoochul Ha, Duk-jin Kim, Yongcheol Kim, and Ki-mook Kang. Unmanned aerial vehicles (UAVs)-based thermal infrared (TIR) mapping, a novel approach to assess groundwater discharge into the coastal zone. Limnology and Oceanogra- phy: Methods, 14(11):725–735, 2016. ISSN 1541-5856. doi:10.1002/lom3.10132. Myungjin Lee, Younghoon Yoo, Hongjun Joo, Kyung Tak Kim, Hung Soo Kim, and Soojun Kim. Construction of rating curve at high water level considering rainfall effect in a tidal river. Journal of Hydrology: Regional Studies, 37:100907, October 2021. ISSN 2214-5818. doi:10.1016/j.ejrh.2021.100907. Carl J. Legleiter and Paul J. Kinzel. Inferring Surface Flow Velocities in Sediment-Laden Alaskan Rivers from Optical Image Sequences Acquired from a Helicopter. Remote Sensing, 12(8):1282, January 2020. ISSN 2072-4292. doi:10.3390/rs12081282. Carl J. Legleiter and Paul J. Kinzel. The Toolbox for River Velocimetry using Images from Aircraft (TRiVIA). River Research and Applications, 39(8):1457– 1468, 2023. ISSN 1535-1467. doi:10.1002/rra.4147. Carl J. Legleiter, Paul J. Kinzel, Michael Dille, Massimo Vespignani, Uland Wong, Isaac Anderson, Elizabeth Hyde, Chris Gazoorian, and Jennifer M. Cramer. Mapping River Flow from Thermal Images in Approximately Real Time: Proof of Concept on the Sacramento River, California, USA. Remote Sensing, 16(24): 4746, January 2024a. ISSN 2072-4292. doi:10.3390/rs16244746. 140 https://doi.org/10.1017/jfm.2019.1019 https://doi.org/10.1002/lom3.10132 https://doi.org/10.1016/j.ejrh.2021.100907 https://doi.org/10.3390/rs12081282 https://doi.org/10.1002/rra.4147 https://doi.org/10.3390/rs16244746 Carl J. Legleiter, Paul J. Kinzel, Frank L. Engel, Lee R. Harrison, and Gregory Hewitt. A two-dimensional, reach-scale implementation of space-time image velocimetry (STIV) and comparison to particle image velocimetry (PIV). Earth Surface Processes and Landforms, 49(10):3093–3114, 2024b. ISSN 1096-9837. doi:10.1002/esp.5878. Russell V. Lenth, Balazs Banfai, Ben Bolker, Paul Buerkner, Iago Giné-Vázquez, Maxime Herve, Maarten Jung, Jonathon Love, Fernando Miguez, Julia Pi- askowski, Hannes Riebl, and Henrik Singmann. Emmeans: Estimated Marginal Means, aka Least-Squares Means, December 2024. Victor A. Levesque and Kevin A. Oberg. Computing Discharge Using the Index Velocity Method. Technical Report 3–A23, U.S. Geological Survey, 2012. Quinn W. Lewis, Evan M. Lindroth, and Bruce L. Rhoads. Integrating un- manned aerial systems and LSPIV for rapid, cost-effective stream gaug- ing. Journal of Hydrology, 560:230–246, May 2018. ISSN 0022-1694. doi:10.1016/j.jhydrol.2018.03.008. D. Lin, J. Grundmann, and A. Eltner. Evaluating Image Tracking Approaches for Surface Velocimetry With Thermal Tracers. Water Resources Research, 55(4): 3122–3136, 2019. ISSN 1944-7973. doi:10.1029/2018WR024507. Burton Edward Livingston. The Relation of Soils to Natural Vegetation in Roscom- mon and Crawford Counties, Michigan. Botanical Gazette, 39(1):22–41, 1905. ISSN 0006-8071. Jianghuai Lu, Xiaohong Yang, and Jianping Wang. Velocity Vector Estimation of Two-Dimensional Flow Field Based on STIV. Sensors, 23(2):955, January 2023. ISSN 1424-8220. doi:10.3390/s23020955. 141 https://doi.org/10.1002/esp.5878 https://doi.org/10.1016/j.jhydrol.2018.03.008 https://doi.org/10.1029/2018WR024507 https://doi.org/10.3390/s23020955 Main, Kevin N. and Menges, Eric S. Station Fire Management Plan. Technical report, Archbold Biological Station, 1997. Gerd Masselink, Mick E. Hanley, Anissa C. Halwyn, Will Blake, Ken Kingston, Thomas Newton, and Mike Williams. Evaluation of salt marsh restora- tion by means of self-regulating tidal gate – Avon estuary, South Devon, UK. Ecological Engineering, 106:174–190, September 2017. ISSN 0925-8574. doi:10.1016/j.ecoleng.2017.05.038. Sojan Mathew, Robin G.D. Davidson-Arnott, and Jeff Ollerhead. Evolution of a beach–dune system following a catastrophic storm overwash event: Greenwich Dunes, Prince Edward Island, 1936–2005. Can. J. Earth Sci., 47(3):273–290, March 2010. ISSN 0008-4077, 1480-3313. doi:10.1139/E09-078. Justin McAllister, and-viceversa, Löıc Dutrieux, fdarvas, Mateusz Jochym, Daniel- sLouis, rrowlands, taftj, and lloyd5389. Micasense/imageprocessing [Software]. MicaSense, Inc., February 2024. Brian McCallum, Chad Wagner, and Claire Bunch. Centennial Stream- gages | U.S. Geological Survey. https://www.usgs.gov/mission-areas/water- resources/science/centennial-streamgages, April 2025. S. K. Mckay and Byron R. Rupp. Preliminary Scoping of Watershed Planning: Pamet River, Massachusetts. Technical report, September 2022. H. R. Moftakhari, D. A. Jay, S. A. Talke, T. Kukulka, and P. D. Bromirski. A novel approach to flow estimation in tidal rivers. Water Resources Research, 49 (8):4817–4832, 2013. ISSN 1944-7973. doi:10.1002/wrcr.20363. H. R. Moftakhari, D. A. Jay, and S. A. Talke. Estimating river dis- charge using multiple-tide gauges distributed along a channel. Journal of 142 https://doi.org/10.1016/j.ecoleng.2017.05.038 https://doi.org/10.1139/E09-078 https://doi.org/10.1002/wrcr.20363 Geophysical Research: Oceans, 121(4):2078–2097, 2016. ISSN 2169-9291. doi:10.1002/2015JC010983. Clay L. Montague, Alexander V. Zale, and H. Franklin Percival. Ecological effects of coastal marsh impoundments: A review. Environmental Management, 11(6): 743–756, November 1987. ISSN 1432-1009. doi:10.1007/BF01867242. M. Murray-Hudson, P. Wolski, L. Cassidy, M. T. Brown, K. Thito, K. Kashe, and E. Mosimanyana. Remote Sensing-derived hydroperiod as a predictor of flood- plain vegetation composition. Wetlands Ecol Manage, 23(4):603–616, August 2015. ISSN 1572-9834. doi:10.1007/s11273-014-9340-z. M. Muste, A. Hauet, I. Fujita, C. Legout, and H. C. Ho. Capabilities of Large- scale Particle Image Velocimetry to characterize shallow free-surface flows. Advances in Water Resources, 70:160–171, August 2014. ISSN 0309-1708. doi:10.1016/j.advwatres.2014.04.004. Marian Muste, Jörg Schöne, and Jean-Dominique Creutin. Measurement of free-surface flow velocity using controlled surface waves. Flow Mea- surement and Instrumentation, 16(1):47–55, March 2005. ISSN 0955-5986. doi:10.1016/j.flowmeasinst.2004.08.003. Ronald L. Myers. Fire and the Dynamic Relationship between Florida Sandhill and Sand Pine Scrub Vegetation. Bulletin of the Torrey Botanical Club, 112(3): 241–252, 1985. ISSN 0040-9618. doi:10.2307/2996539. Urs Neumeier and Carl L. Amos. The influence of vegetation on turbulence and flow velocities in European salt-marshes. Sedimentology, 53(2):259–277, 2006. ISSN 1365-3091. doi:10.1111/j.1365-3091.2006.00772.x. NOAA. 8446613 Wellfleet, MA, 2023. 143 https://doi.org/10.1002/2015JC010983 https://doi.org/10.1007/BF01867242 https://doi.org/10.1007/s11273-014-9340-z https://doi.org/10.1016/j.advwatres.2014.04.004 https://doi.org/10.1016/j.flowmeasinst.2004.08.003 https://doi.org/10.2307/2996539 https://doi.org/10.1111/j.1365-3091.2006.00772.x NortekAS. Aquadopp Profiler 1 MHz. D. G. Ortiz-Suslow, B. K. Haus, N. J. Williams, N. J. M. Laxague, A. J. H. M. Reniers, and H. C. Graber. The spatial-temporal variability of air-sea momentum fluxes observed at a tidal inlet. Journal of Geophysical Research: Oceans, 120 (2):660–676, 2015. ISSN 2169-9291. doi:10.1002/2014JC010412. Sophie Pearce, Robert Ljubičić, Salvador Peña-Haro, Matthew Perks, Flavia Tauro, Alonso Pizarro, Silvano Fortunato Dal Sasso, Dariia Strelnikova, Sal- vatore Grimaldi, Ian Maddock, Gernot Paulus, Jasna Plavšić, Dušan Pro- danović, and Salvatore Manfreda. An Evaluation of Image Velocimetry Tech- niques under Low Flow Conditions and High Seeding Densities Using Unmanned Aerial Systems. Remote Sensing, 12(2):232, January 2020. ISSN 2072-4292. doi:10.3390/rs12020232. W. H. Pearsall. The Aquatic and Marsh Vegetation of Esthwaite Water. Journal of Ecology, 5(3/4):180–202, 1917. ISSN 0022-0477. doi:10.2307/2255656. Matthew T. Perks, Andrew J. Russell, and Andrew R. G. Large. Technical Note: Advances in flash flood monitoring using unmanned aerial vehicles (UAVs). Hy- drology and Earth System Sciences, 20(10):4005–4015, October 2016. ISSN 1027- 5606. doi:10.5194/hess-20-4005-2016. Michael Philben, Suzanne DeVries-Zimmerman, Grace Behrens, Eli Kane, and Christian Lundy. Iron availability regulates carbon accumulation and methane production in a Lake Michigan interdunal wetland. Wetlands, 44(8):115, October 2024. ISSN 1943-6246. doi:10.1007/s13157-024-01870-z. Daniele Pinton, Alberto Canestrelli, and Luca Fantuzzi. A UAV-Based Dye- Tracking Technique to Measure Surface Velocities over Tidal Channels and Salt 144 https://doi.org/10.1002/2014JC010412 https://doi.org/10.3390/rs12020232 https://doi.org/10.2307/2255656 https://doi.org/10.5194/hess-20-4005-2016 https://doi.org/10.1007/s13157-024-01870-z Marshes. Journal of Marine Science and Engineering, 8(5):364, May 2020. ISSN 2077-1312. doi:10.3390/jmse8050364. Luigi Pirelli. Camera calculator.py [Software], August 2019. Alonso Pizarro, Silvano F. Dal Sasso, and Salvatore Manfreda. VISION: VIdeo StabilisatION using automatic features selection for image velocime- try analysis in rivers. SoftwareX, 19:101173, July 2022. ISSN 2352-7110. doi:10.1016/j.softx.2022.101173. John W. Portnoy. Salt Marsh Diking and Restoration: Biogeochemical Implica- tions of Altered Wetland Hydrology. Environmental Management, 24(1):111– 120, July 1999. ISSN 1432-1009. doi:10.1007/s002679900219. John W. Portnoy and Jenny R. Allen. EFFECTS OF TIDAL RESTRIC- TIONS AND POTENTIAL BENEFITS OF TIDAL RESTORATION ON FECAL COLIFORM AND SHELLFISH-WATER QUALITY. shre, 25(2): 609–617, August 2006. ISSN 0730-8000, 1943-6319. doi:10.2983/0730- 8000(2006)25[609:EOTRAP]2.0.CO;2. John W. Portnoy and A.E. Giblin. Effects of historic tidal restrictions on salt marsh sediment chemistry. Biogeochemistry, 36(3):275–303, March 1997. ISSN 1573-515X. doi:10.1023/A:1005715520988. Jack A. Puleo, Thomas E. McKenna, K. Todd Holland, and Joseph Calan- toni. Quantifying riverine surface currents from time sequences of thermal infrared imagery. Water Resources Research, 48(1), 2012. ISSN 1944-7973. doi:10.1029/2011WR010770. Ruonan Qiu, Ge Han, Siwei Li, Feng Tian, Xin Ma, and Wei Gong. Soil moisture dominates the variation of gross primary productivity during hot drought in 145 https://doi.org/10.3390/jmse8050364 https://doi.org/10.1016/j.softx.2022.101173 https://doi.org/10.1007/s002679900219 https://doi.org/10.2983/0730-8000(2006)25[609:EOTRAP]2.0.CO;2 https://doi.org/10.2983/0730-8000(2006)25[609:EOTRAP]2.0.CO;2 https://doi.org/10.1023/A:1005715520988 https://doi.org/10.1029/2011WR010770 drylands. Science of The Total Environment, 899:165686, November 2023. ISSN 0048-9697. doi:10.1016/j.scitotenv.2023.165686. R-core. TukeyHSD, December 2019. Caleb Rankin, Troy Gaston, Mahmood Sadat-Noori, William Glamore, Jason Mor- ton, and Anita Chalmers. Innovative tidal control successfully promotes salt- marsh restoration. Restoration Ecology, 31(7):e13774, 2023. ISSN 1526-100X. doi:10.1111/rec.13774. Saul Edward Rantz. Measurement and Computation of Streamflow. U.S. Depart- ment of the Interior, Geological Survey, 1982. M. K. Reinecke, C. A. Brown, K. J. Esler, J. M. King, M. T. Kleynhans, and M. Kidd. Links Between Lateral Vegetation Zones and River Flow. Wetlands, 35(3):473–486, June 2015. ISSN 1943-6246. doi:10.1007/s13157-015-0634-6. William Ritchie and Shea Penland. Rapid dune changes associated with overwash processes on the deltaic coast of South Louisiana. Marine Geology, 81(1):97–122, June 1988. ISSN 0025-3227. doi:10.1016/0025-3227(88)90020-5. Antonio B. Rodriguez, Ethan J. Theuerkauf, Justin T. Ridge, Beth M. VanDusen, and Stephen R. Fegley. Long-term washover fan accretion on a transgressive barrier island challenges the assumption that paleotempestites represent indi- vidual tropical cyclones. Sci Rep, 10(1):19755, November 2020. ISSN 2045-2322. doi:10.1038/s41598-020-76521-4. G. Curtis Roegner, Earl W. Dawley, Micah Russell, Allan Whiting, and David J. Teel. Juvenile Salmonid Use of Reconnected Tidal Freshwater Wetlands in Grays River, Lower Columbia River Basin. Transactions of the American Fisheries Society, 139(4):1211–1232, July 2010. ISSN 0002-8487. doi:10.1577/T09-082.1. 146 https://doi.org/10.1016/j.scitotenv.2023.165686 https://doi.org/10.1111/rec.13774 https://doi.org/10.1007/s13157-015-0634-6 https://doi.org/10.1016/0025-3227(88)90020-5 https://doi.org/10.1038/s41598-020-76521-4 https://doi.org/10.1577/T09-082.1 Charles T. Roman, William A. Niering, and R. Scott Warren. Salt marsh vegeta- tion change in response to tidal restriction. Environmental Management, 8(2): 141–149, March 1984. ISSN 1432-1009. doi:10.1007/BF01866935. Charles T. Roman, Richard W. Garvine, and John W. Portnoy. Hydrologic modeling as a predictive basis for ecological restoration of salt marshes. Environmental Management, 19(4):559–566, July 1995. ISSN 1432-1009. doi:10.1007/BF02471967. Catherine A. Ruhl and Michael R. Simpson. Computation of discharge using the index-velocity method in tidally affected areas. Technical Report 2005-5004, 2005. K. L. Sahrawat. Organic matter accumulation in submerged soils. Advances in Agronomy, 81:169–201, 2004. Rebecca Sanders-Demott, Meagan J Eagle, Kevin D Kroeger, Thomas W Brooks, Jennifer A O’keefe Suttles, Sydney K Nick, and Adrian C Mann. Carbon dioxide and methane fluxes with supporting environmental data from coastal wetlands across Cape Cod, Massachusetts (ver 2.0, June 2022), 2022. Rebecca Sanders-DeMott, Meagan J. Eagle, Kevin D. Kroeger, Faming Wang, Thomas W. Brooks, Jennifer A. O’Keefe Suttles, Sydney K. Nick, Adrian G. Mann, and Jianwu Tang. Impoundment increases methane emissions in Phragmites-invaded coastal wetlands. Global Change Biology, 28(15):4539–4557, 2022. ISSN 1365-2486. doi:10.1111/gcb.16217. Aline de Camargo Santos, J. Mabry McCray, Samira H. Daroub, Diane L. Rowland, Shangning Ji, and Hardev Sandhu. Nitrogen Assessment of Shallow Florida 147 https://doi.org/10.1007/BF01866935 https://doi.org/10.1007/BF02471967 https://doi.org/10.1111/gcb.16217 Histosols. Communications in Soil Science and Plant Analysis, 51(14):1916– 1929, August 2020. ISSN 0010-3624. doi:10.1080/00103624.2020.1798990. Michael W. I. Schmidt, Margaret S. Torn, Samuel Abiven, Thorsten Dittmar, Georg Guggenberger, Ivan A. Janssens, Markus Kleber, Ingrid Kögel-Knabner, Johannes Lehmann, David A. C. Manning, Paolo Nannipieri, Daniel P. Rasse, Steve Weiner, and Susan E. Trumbore. Persistence of soil organic matter as an ecosystem property. Nature, 478(7367):49–56, October 2011. ISSN 1476-4687. doi:10.1038/nature10386. Rachel Schultz, Sarah Andrews, Lindsay O’Reilly, Virginie Bouchard, and Serita Frey. Plant Community Composition More Predictive than Diversity of Carbon Cycling in Freshwater Wetlands. Wetlands, 31(5):965–977, October 2011. ISSN 1943-6246. doi:10.1007/s13157-011-0211-6. Seth A. Schweitzer and Edwin A. Cowen. Instantaneous River-Wide Water Surface Velocity Field Measurements at Centimeter Scales Using Infrared Quantitative Image Velocimetry. Water Resources Research, 57(8):e2020WR029279, 2021. ISSN 1944-7973. doi:10.1029/2020WR029279. Seth A. Schweitzer and Edwin A. Cowen. A Method for Analysis of Spatial Uncer- tainty in Image Based Surface Velocimetry. Frontiers in Water, 4, 2022. ISSN 2624-9375. Michael Schwendeman and Jim Thomson. Observations of whitecap coverage and the relation to wind stress, wave slope, and turbulent dissipation. Journal of Geophysical Research: Oceans, 120(12):8346–8363, 2015. ISSN 2169-9291. doi:10.1002/2015JC011196. 148 https://doi.org/10.1080/00103624.2020.1798990 https://doi.org/10.1038/nature10386 https://doi.org/10.1007/s13157-011-0211-6 https://doi.org/10.1029/2020WR029279 https://doi.org/10.1002/2015JC011196 W. A. Segeren. Introduction to Polders of the World. Water International, January 1983. doi:10.1080/02508068308686006. C. A. Semeniuk and V. Semeniuk. Dune slacks in Western Australia. Journal of the Royal Society of Western Australia, 94(4):503–532, December 2011. ISSN 0035922X. S. Settelmyer, E. Swails, and J. Eaton. Herring River Carbon Project Feasibility Study. Technical report, National Estuarine Research Reserve System Science Collaborative, 2019. John Shaw, Yao You, David Mohrig, and Gary Kocurek. Tracking hurricane- generated storm surge with washover fan stratigraphy. Geology, 43(2):127–130, February 2015. ISSN 0091-7613. doi:10.1130/G36460.1. Christine C. Shepard, Caitlin M. Crain, and Michael W. Beck. The Protective Role of Coastal Marshes: A Systematic Review and Meta-analysis. PLOS ONE, 6(11): e27374, November 2011. ISSN 1932-6203. doi:10.1371/journal.pone.0027374. Germán D. Silva, Dar A. Roberts, Joseph P. McFadden, and Jennifer Y. King. Shifts in Salt Marsh Vegetation Landcover after Debris Flow Deposition. Remote Sensing, 14(12):2819, January 2022. ISSN 2072-4292. doi:10.3390/rs14122819. David R. Smith, Mitchell J. Eaton, Jill J. Gannon, Timothy P. Smith, Eric L. Derleth, Jonathan Katz, Kirk F. Bosma, and Elise Leduc. A decision framework to analyze tide-gate options for restoration of the Herring River Estuary, Mas- sachusetts. USGS Numbered Series 2019-1115, U.S. Geological Survey, Reston, VA, 2020. Stephen M. Smith. Removal of Salt-killed Vegetation during Tidal Restoration of a New England Salt Marsh: Effects on Wrack Movement and the Establishment 149 https://doi.org/10.1080/02508068308686006 https://doi.org/10.1130/G36460.1 https://doi.org/10.1371/journal.pone.0027374 https://doi.org/10.3390/rs14122819 of Native Halophytes. Ecological Restoration, 25(4):268–273, December 2007. ISSN 1543-4060, 1543-4079. doi:10.3368/er.25.4.268. Michael A. Soukup and John W. Portnoy. Impacts from Mosquito Control-induced Sulphur Mobilization in a Cape Cod Estuary. Environmental conservation, 13 (1):47–50, 1986. C. J. Stratford, N. S. Robins, D. Clarke, L. Jones, and G. Weaver. An ecohydrolog- ical review of dune slacks on the west coast of England and Wales. Ecohydrology, 6(1):162–171, 2013. ISSN 1936-0592. doi:10.1002/eco.1355. Peter Sutherland and W. Kendall Melville. Field Measurements of Surface and Near-Surface Turbulence in the Presence of Breaking Waves. Journal of Phys- ical Oceanography, 45(4):943–965, April 2015. ISSN 0022-3670, 1520-0485. doi:10.1175/JPO-D-14-0133.1. Karle V. Sýkora, Jeroen C.J.M. van den Bogert, and Frank Berendse. Changes in soil and vegetation during dune slack succession. Journal of Vegetation Science, 15(2):209–218, 2004. ISSN 1654-1103. doi:10.1111/j.1654-1103.2004.tb02256.x. A. G. Tansley. The Use and Abuse of Vegetational Concepts and Terms. Ecology, 16(3):284–307, 1935. ISSN 0012-9658. doi:10.2307/1930070. Flavia Tauro, Christopher Pagano, Paul Phamduy, Salvatore Grimaldi, and Mau- rizio Porfiri. Large-Scale Particle Image Velocimetry From an Unmanned Aerial Vehicle. IEEE/ASME Transactions on Mechatronics, 20(6):3269–3275, Decem- ber 2015. ISSN 1941-014X. doi:10.1109/TMECH.2015.2408112. S. L. H. Teo and K. W. Able. Habitat use and movement of the mummichog (Fundulus heteroclitus) in a restored salt marsh. Estuaries, 26(3):720–730, June 2003. ISSN 0160-8347. doi:10.1007/BF02711983. 150 https://doi.org/10.3368/er.25.4.268 https://doi.org/10.1002/eco.1355 https://doi.org/10.1175/JPO-D-14-0133.1 https://doi.org/10.1111/j.1654-1103.2004.tb02256.x https://doi.org/10.2307/1930070 https://doi.org/10.1109/TMECH.2015.2408112 https://doi.org/10.1007/BF02711983 Philipp Thumser, Christian Haas, Jeffrey A. Tuhtan, Juan Francisco Fuentes- Pérez, and Gert Toming. RAPTOR-UAV: Real-time particle tracking in rivers using an unmanned aerial vehicle. Earth Surface Processes and Landforms, 42 (14):2439–2446, 2017. ISSN 1096-9837. doi:10.1002/esp.4199. Carlo Tomasi and Takeo Kanade. Shape and motion from image streams un- der orthography: A factorization method. Int J Comput Vision, 9(2):137–154, November 1992. ISSN 1573-1405. doi:10.1007/BF00129684. Alfonso Torres-Rua. Vicarious Calibration of sUAS Microbolometer Temperature Imagery for Estimation of Radiometric Land Surface Temperature. Sensors, 17 (7):1499, July 2017. ISSN 1424-8220. doi:10.3390/s17071499. UCNRS. Carpinteria [Dataset], 2022. UCNRS. Carpinteria Salt Marsh Reserve Archives - Natural Reserve System. https://ucnrs.org/reserves/carpinteria-salt-marsh-reserve/, May 2023. NRCS USDA. The PLANTS database, 2025. USGS. Carpinteria C NR Carpinteria CA [Dataset], 2022a. USGS. Herring R at Chequessett Neck RD at Wellfleet, MA, 2022b. Ivan Valiela and Sophia E. Fox. Managing Coastal Wetlands. Science, 319(5861): 290–291, January 2008. doi:10.1126/science.1153477. Fabrice Veron andW. Kendall Melville. Experiments on the stability and transition of wind-driven water surfaces. Journal of Fluid Mechanics, 446:25–65, November 2001. ISSN 1469-7645, 0022-1120. doi:10.1017/S0022112001005638. George Voulgaris and Samuel T. Meyers. Temporal variability of hydrodynam- ics, sediment concentration and sediment settling velocity in a tidal creek. 151 https://doi.org/10.1002/esp.4199 https://doi.org/10.1007/BF00129684 https://doi.org/10.3390/s17071499 https://doi.org/10.1126/science.1153477 https://doi.org/10.1017/S0022112001005638 Continental Shelf Research, 24(15):1659–1683, October 2004. ISSN 0278-4343. doi:10.1016/j.csr.2004.05.006. Wenzhi Wang, Peipei Zhang, Hongxia Zhang, Charlotte Grossiord, Stephanie C. Pennington, Matthew J. Norwood, Weibin Li, Alexandria L. Pivovaroff, Laura Fernández-de-Uña, Riley Leff, Steven B. Yabusaki, Scott Waichler, Vanessa L. Bailey, Nicholas D. Ward, and Nate G. McDowell. Severe declines in hydraulic capacity and associated carbon starvation drive mortality in seawater exposed Sitka-spruce (Picea sitchensis) trees. Environ. Res. Commun., 4(3):035005, March 2022. ISSN 2515-7620. doi:10.1088/2515-7620/ac5f7d. Eugenius Warming. Oecology of Plants: An Introduction to the Study of Plant- communities. Oxford University Press, 1925. W. A. Watts. The Late Quaternary Vegetation History of the Southeastern United States. Annual Review of Ecology and Systematics, 11:387–409, 1980. ISSN 0066-4162. Robert H. Whittaker. Classification of natural communities. Bot. Rev, 28(1): 1–239, January 1962. ISSN 1874-9372. doi:10.1007/BF02860872. Robert H. Whittaker. Approaches to Classifying Vegetation. In Robert H. Whit- taker, editor, Classification of Plant Communities, pages 1–31. Springer Nether- lands, Dordrecht, 1978. ISBN 978-94-009-9183-5. doi:10.1007/978-94-009-9183- 5 1. Alfred Wüest and Andreas Lorke. Small-Scale Hydrodynamics in Lakes. Annual Review of Fluid Mechanics, 35(1):373–412, 2003. doi:10.1146/annurev.fluid.35.101101.161220. 152 https://doi.org/10.1016/j.csr.2004.05.006 https://doi.org/10.1088/2515-7620/ac5f7d https://doi.org/10.1007/BF02860872 https://doi.org/10.1007/978-94-009-9183-5_1 https://doi.org/10.1007/978-94-009-9183-5_1 https://doi.org/10.1146/annurev.fluid.35.101101.161220 David L. Young, Brittany L. Bruder, Kevin A. Haas, and Donald R. Webster. The hydrodynamics of surface tidal flow exchange in saltmarshes. Estuar- ine, Coastal and Shelf Science, 172:128–137, April 2016. ISSN 0272-7714. doi:10.1016/j.ecss.2016.02.006. Tom Ysebaert, Dirk-Jan van der Hoek, Rick Wortelboer, Jeroen W. M. Wijsman, Marijn Tangelder, and Arno Nolte. Management options for restoring estuar- ine dynamics and implications for ecosystems: A quantitative approach for the Southwest Delta in the Netherlands. Ocean & Coastal Management, 121:33–48, March 2016. ISSN 0964-5691. doi:10.1016/j.ocecoaman.2015.11.005. Christopher J. Zappa, Peter A. Raymond, Eugene A. Terray, and Wade R. McGillis. Variation in surface turbulence and the gas transfer velocity over a tidal cycle in a macro-tidal estuary. Estuaries, 26(6):1401–1415, December 2003. ISSN 0160-8347. doi:10.1007/BF02803649. 153 https://doi.org/10.1016/j.ecss.2016.02.006 https://doi.org/10.1016/j.ocecoaman.2015.11.005 https://doi.org/10.1007/BF02803649 Introduction Wind stress effects on drone-based thermal infrared surface velocimetry measurements of tidal flow in an estuary Introduction Image-based velocimetry techniques Objectives Materials and Methods Study Site Acoustic profiler in-situ velocity measurements Drone-based image collection Georeferencing and stabilization of drone images Space-time image velocimetry (STIV) Tide and wind conditions Results Uncertainty associated with time synchronization between velocity measurements Coordinate system and sign of velocity measurements Discussion Physical causes of time-varying velocity differences Measurement uncertainty and data quality Implications for image-based surface velocity measurements Acronyms Open Research Acknowledgments Developing a stage-discharge rating curve for an impounded estuary using thermal infrared surface velocimetry Introduction Methods Study site Field campaign Image Analysis Models Results Developing rating curves Hydrograph comparison to model outputs Discussion Remote measurements of flow rate, bathymetric change and sediment fluxes through a tidal beach overwash channel using thermal infrared imagery Introduction Methods Study site Field campaign Infrared velocimetry analysis Results Flow rate estimate from surface velocimetry Elevation surveys Acoustic measurement of flow rate at confluence Discussion Is a historical vegetation map a useful predictor of dry season soil water fraction in seasonal ponds? Introduction Objectives Methods Study Site Experimental design Soil Sampling Mixed effects modeling Results Correlation between soil water fraction and pond type Soil organic matter Discussion Chapter 2 Appendix: Velocity measurement uncertainty River-side comparison with ADV velocities River-side acceleration Bay-side mass conservation Chapter 3 Appendix: Space-time diagram contrast enhancement methods Chapter 4 Appendix: Mixed effects model supporting information