CONCEPTS AND QUESTIONS Nitrogen fluxes from the landscape are 37 controlled by net anthropogenic nitrogen inputs and by climate Robert Howarth1*, Dennis Swaney1, Gilles Billen2, Josette Garnier2, Bongghi Hong1, Christoph Humborg3, Penny Johnes4, Carl-Magnus Mörth3, and Roxanne Marino1 The flux of nitrogen (N) to coastal marine ecosystems is strongly correlated with the “net anthropogenic nitrogen inputs” (NANI) to the landscape across 154 watersheds, ranging in size from 16 km2 to 279 000 km2, in the US and Europe. When NANI values are greater than 1070 kg N km–2 yr–1, an average of 25% of the NANI is exported from those watersheds in rivers. Our analysis suggests a possible threshold at lower NANI levels, with a smaller fraction exported when NANI values are below 1070 kg N km–2 yr–1. Synthetic fertilizer is the largest component of NANI in many watersheds, but other inputs also contribute substantially to the N fluxes; in some regions, atmospheric deposition of N is the major component. The flux of N to coastal areas is controlled in part by climate, and a higher percentage of NANI is exported in rivers, from water- sheds that have higher freshwater discharge. Front Ecol Environ 2012; 10(1): 37–43, doi:10.1890/100178 (published online 15 Jul 2011) Excessive amounts of nitrogen (N) represent the other agricultural inputs, could not be discerned. Today,largest pollution problem in coastal marine waters. the relative contribution of sources to coastal N pollution Human activity has increased N inputs by 10- to 15-fold remains uncertain in many cases, in part because no in many regions, but has had little effect in others (NRC direct approaches for making such evaluations exist. In 2000; Howarth et al. 2005, 2011). Nitrogen derives from such instances, models provide the only robust assess- many sources, and different sources of N dominate in dif- ment tool (EPA–SAB 2008). ferent areas. Over 20 years ago, Peierls et al. (1991) In one paper from a 1994 workshop, Howarth et al. demonstrated a correlation between human population (1996) examined the flux of N from large watershed density and nitrate fluxes in very large rivers and sug- regions to the North Atlantic Ocean in the context of gested that sewage was the primary cause, with perhaps a the N inputs to the landscape from human activity. contribution from atmospheric deposition. At the coarse Inputs considered were use of synthetic N fertilizer, N fix- scale they applied, other drivers, including fertilizer and ation associated with agricultural crops, atmospheric deposition of oxidized N (NOy), and the net movement In a nutshell: of N into or out of the region in human food and animal feeds. We termed the sum of these inputs the “net anthro- • Nitrogen (N) pollution is one of the primary threats to the eco- logical integrity of estuaries and other coastal marine ecosys- pogenic nitrogen inputs”, or NANI. At the coarse scale of tems large regions surrounding the North Atlantic Ocean, the • Although synthetic fertilizer is the main source of N pollution average multi-year flux of N transported in rivers to the in many areas, other sources – such as atmospheric deposition North Atlantic was well correlated with NANI. and the movement of N in food and animal feeds – contribute, Alexander et al. (2002) compared many models for esti- and are sometimes dominant • N fluxes in rivers to coastal ecosystems increase as the “net mating N fluxes in large river basins and concluded that a anthropogenic nitrogen inputs” (NANI) to the landscape simple model that predicts N flux as a linear function of increase NANI was one of the most accurate, with low bias and • NANI provides a powerful approach for estimating these N error as compared with those of more complicated mod- fluxes and for determining the major sources of N pollution in els. This simple model has since been used to estimate the the landscape total riverine N flux from the global landscape to the world’s oceans (Galloway et al. 2004; Boyer et al. 2006). 1Department of Ecology and Evolutionary Biology, Cornell University, NANI does not include sewage or animal wastes Ithaca, NY *(howarth@cornell.edu); 2UMPC Université Paris 6 and because these are simply flows of N that originate from CNRS, UMR Sisyphe, Paris, France; 3Baltic NEST Institute, Stock- other sources already included in NANI. Similarly, the holm Resilience Centre, Stockholm, Sweden; 4Aquatic Environments only atmospheric input considered is NOy deposition, Research Centre, School of Human and Environmental Sciences, which in the temperate zone originates largely from fossil- University of Reading, Whiteknights, Reading, UK fuel combustion and is therefore a new input of N to the © The Ecological Society of America www.frontiersinecology.org N fluxes from watersheds R Howarth e t al. 38 (a) (b) (c) NE US watersheds SE US watersheds Western US watersheds Lake Michigan watersheds French and Belgian watersheds Swedish watersheds UK watersheds Figure 1. Maps showing the distribution of the watersheds included in our analysis (a) in the US and (b) in Europe. The watersheds in the UK are shown both in the European map and (c) in the more detailed map of the UK. landscape. Deposition of ammonia is excluded, given that (Howarth et al. 2006; Schaefer and Alber 2007; Schaefer most of the ammonia in the atmosphere is deposited near et al. 2009; Han and Allan 2008) for estimates of area, the site of emission to the atmosphere (ie within the same average discharge, average temperature, and riverine region) and originates from agricultural sources already total N flux (see WebTable 1), and for 3 out of 4 of the included in NANI (Howarth et al. 1996, 2006). input terms for NANI: synthetic fertilizer, N fixation in The NANI approach, or the closely related approach of agroecosystems, and the net input of N in human food considering total N inputs (TNI, which is equivalent to and animal feeds. These data generally come from the NANI plus natural N fixation), has been applied in many county scale. To estimate the fourth input term – NOy regions, including the northeastern US (Alexander et al. deposition – we used output from the US Environmental 2002; Boyer et al. 2002; Howarth et al. 2006), the south- Protection Agency’s Community Multi-scale Air Quality eastern US (Schaefer and Alber 2007), many of the (CMAQ) system rather than the NOy deposition esti- watersheds on the west coast of the US (Schaefer et al. mates reported in the original papers. CMAQ is an emis- 2009), and watersheds in Michigan (Han and Allan sions-based model that predicts total oxidized deposition, 2008). In all of these cases, riverine N fluxes were well including gases across the US, at a grid of 36 km × 36 km correlated with NANI (or TNI), but the percentage of (www.cmaq-model.org/). the N inputs exported in rivers varied among the regions. The European watersheds included 25 in France and Several of these previous studies suggested that the frac- Belgium, 30 in the UK, and 36 in Sweden (WebTable 1). tion of NANI exported in riverine flows is related to cli- The French and Belgian basins included the Seine, matic variables, including precipitation, temperature, and Somme, and Scheldt watersheds and 22 nested sub- freshwater discharge. However, the conclusions of these basins; these basins and the approach used for estimating studies often contradicted one another. We hypothesized the NANI terms are described in Billen et al. (2009). The that the influence of climate on the relationship of NANI budgets for the UK watersheds were constructed NANI and riverine N flux might become clearer if a through government (Department for Environment, larger set of watersheds from a diversity of regions were Food and Rural Affairs) statistics on food and feed considered. Here, we report on such a study, one that import/export for the UK, and UN Food and Agriculture includes 154 US and European watersheds. Organization (FAO) and UK statistics on precipitation, discharge, climatic variables, riverine N flux, and the N n Data sources content of food and feed consumed in the UK, following the approach outlined by Boyer et al. (2002). Background Our analysis included watersheds in the US, France, data for the UK watersheds were derived from a range of Belgium, the UK, and Sweden (Figure 1). The watersheds sources, including research reports for the UK Environ- varied considerably in size, from 16 km2 to 279 000 km2. ment Agency, and published studies (see Web- The US watersheds included 16 in the Northeast (Boyer References). For all Swedish watersheds, we used agricul- et al. 2002; Howarth et al. 2006), 12 in the Southeast tural statistics obtained from the Statistiska-Centralbyrån (Schaefer and Alber 2007), 17 in the West (Schaefer et al. for 1995 (www.scb.se). We constructed food and feed 2009), and 18 in the Upper Midwest (Han and Allan budgets following Boyer et al. (2002), using the statistical 2008). For the US watersheds, we used published data agricultural data together with FAO statistics. Fertilizer www.frontiersinecology.org © The Ecological Society of America R Howarth et al. N fluxes from watersheds use data were obtained from Eurostat (a) 39 (http://epp.eurostat.ec.europa.eu/). 8000 Riverine N flux, climatic, and atmos- pheric deposition data were collected 6000 y = 0.24x + 66.5 2 from the Baltic Environmental Data- r = 0.66 base (http://nest.su.se/models/bed. 4000 htm). For all the European water- US sheds, we derived deposition esti- 2000 France/Belgium mates from the European Monitoring UK and Evaluation Programme’s model, Sweden0 an emissions-based model similar to 0 5000 10 000 15 000 20 000 25 000 30 000 CMAQ, using a grid of 50 km × 50 NANI (kg N km–2 yr–1) km. For the other NANI terms, esti- mates were generally based on the (b) finest scale of administrative govern- 10 000 Piecewise linear: ment unit for which information was y = 0.156x + 111, NANI < 1070 available, roughly equivalent to 1000 y = 0.25x + 278, NANI ≥ 10702 county-scale data in the US. r = 0.67 For all watersheds included in this USFrance/Belgium paper, the riverine N fluxes reported 100 UK are multi-year averages, usually for 6 Linear: y = 0.23x + 1342 Swedenr = 0.60 or 7 years. The NANI estimates come 10 from a single-year period within those 100 1000 10 000 100 000 6 or 7 years. Note that NANI gener- NANI (kg N km–2 yr–1) ally does not vary greatly over short time intervals (Hong et al. 2011). Figure 2. The flux of N from the landscape in rivers is significantly and highly correlated with NANI on both (a) linear (P = 2 × 10–37) and (b) log–log (P = 3 × 10–32) scales n Riverine N flows and NANI across the 154 watersheds. In the log–log plot, we explored a possible threshold break point in the function, fitting two line segments with the break point determined by Riverine N flux from the 154 water- minimizing the sum of squared deviations using the Solver add-on in Microsoft Excel. sheds is significantly correlated with This piecewise linear fit suggests a threshold response at a NANI value of approximately NANI on both linear and log–log 1070 kg N km–2 yr–1, with the slope of the line above this threshold being virtually the scales (Figure 2, a and b). The slope same as for the linear fit in (a). The slopes of these relationships indicate that, on of the regression on the linear scale average, approximately 25% of NANI is exported from the landscape to coastal oceans, (Figure 2a) indicates that, on aver- at least for the values of NANI greater than 1070 kg N km–2 yr–1. age, approximately 25% of NANI is exported in the rivers included in this study. The slope for 2002). A better understanding of the fate of non-river- the single-line fit in the log–log plot is less, but we also exported NANI is critical if we are to predict how sinks explored a threshold response in the log–log relationship and fluxes may change in the future as a result of climate by using a piecewise linear fit. The existence of a thresh- change, land-use change, and saturation of some sinks. old might indicate some saturation process at the water- The NANI approach was originally developed for very shed scale, as was previously observed at smaller scales for large regions (such as the entire northeastern US from inputs of N from atmospheric deposition to forests (Aber Maine through Virginia, or the entire Mississippi River et al. 2003) and for fertilizer inputs to agroecosystems basin), and has subsequently been applied to smaller – (Howarth et al. 2005; Billen et al. 2007). The piecewise but generally still large – watersheds (Alexander et al. linear fit to the log–log relationship suggests a threshold 2002; Howarth et al. 2006; Schaefer and Alber 2007; Han response at a NANI value of approximately 1070 kg N and Allan 2008; Schaefer et al. 2009). For several reasons, km–2 yr–1, with the slope of the line above this threshold one might expect the approach to be more robust at being virtually the same as for the linear fit in Figure 2a larger spatial scales and to break down below some and indicating that 25% of NANI is exported in rivers. threshold watershed size. For example, cross-boundary At lower levels of NANI, the percentage of NANI transfer of ammonia in the atmosphere is small relative to exported appears to be less than 25%. other NANI terms at large spatial scales but becomes The fate of NANI that is not exported in rivers – some increasingly important at smaller scales (Howarth et al. 75% on average at higher NANI levels – remains poorly 2006). Also, the NANI approach is presumably most known. For the northeastern US, the best available evi- robust when watersheds are large relative to the scale of dence suggests that some is retained in soils and forest input data. For NOy deposition in the US, this spatial biomass, but more is denitrified (van Breemen et al. scale for input data is 1296 km2, and for some other data in © The Ecological Society of America www.frontiersinecology.org Riverine TN flux Riverine TN flux (kg N km–2 yr–1) (kg N km–2 yr–1) N fluxes from watersheds R Howarth et al. 40 (a) Riverine N flux as a function of fertilizer inputs for all watersheds Even when such data are available, they are difficult to extrapolate to the 8000 US y = 0.24x + 259.8 watershed scale. We estimated the France/Belgium r2 = 0.69 natural rate of N fixation from the 6000 UK regression between evapotranspira- Sweden tion and fixation developed by 4000 Cleveland et al. (1999) for a global dataset on N fixation. Evapotrans- 2000 piration for our watersheds was esti- 0 mated as the difference between pre- 0 5000 10 000 15 000 20 000 cipitation and freshwater discharge. Fertilizer inputs (kg N km–2 yr–1) Riverine N fluxes from the water- (b) sheds are significantly correlated 2000 with both TNI and NANI (Web- US Sweden Figure 2). The two relationships are 0.0024x extremely similar, and a test of coin-y = 117.7e 2 0.0011x r = 0.66 cidence of the regressions shows no 1000 y = 109.2e r2 = 0.52 significant differences between the two. Given that the TNI approach requires the estimation of a highly uncertain term (ie the natural rate 0 of N fixation) and does not signifi- 0 200 400 600 800 1000 cantly improve the correlation with NOy deposition (kg N km –2 yr–1) riverine N flux, we prefer the NANI approach. Figure 3. (a) Synthetic N fertilizer is often the major term of NANI in watersheds, and fertilizer alone is significantly correlated with the average flux of N in rivers across the 154 n The influence of the individual watersheds (P = 5 × 10–41). (b) The atmospheric deposition of oxidized N (NOy) is an NANI terms important term of NANI in some watersheds; for those watersheds where this deposition equals or exceeds the input of synthetic N fertilizer, deposition is significantly correlated Each NANI component contributes with riverine N fluxes (P = 2 × 10–16 for the US watersheds and P = 7 × 10–5 for the to riverine N flux. For many of the Swedish watersheds). Note that the N in NOy deposition originates largely from the watersheds included here, synthetic combustion of fossil fuels, and also contributes to acid rain. N fertilizer is the single largest input. Not surprisingly, therefore, fertilizer various locations, the scale is even coarser. We searched input alone is significantly correlated with riverine N flux for a size-threshold effect on the utility of the NANI (Figure 3a). More surprising is the finding that agricul- approach by analyzing the goodness of fit between NANI tural N fixation alone (WebFigure 3) and NOy deposition and riverine N flow while stepwise deleting one watershed alone (WebFigure 4) are also correlated with riverine N at a time from the analysis by dropping the smallest flux. The contribution of atmospheric deposition holds remaining watershed at each step (WebFigure 1). The for the entire dataset but becomes quite notable when goodness of fit in the relationship gradually improved as looking at the subset of watersheds for which NOy deposi- watersheds were deleted, but in general we saw few if any tion is greater than fertilizer inputs (Figure 3b). This sharp break points. This suggests that the NANI approach group only includes watersheds in the US and in Sweden; is reasonably robust and predictive, even in watersheds we have fit separate regressions for the watersheds in the that are far smaller than those to which the approach has two countries. Both show an exponential response with usually been applied in the past. For many of the analyses proportionately greater N flux in rivers as deposition in this paper, we concentrate on watersheds greater than increases above 500–900 kg N km–2 yr–1. This is consis- 250 km2. These analyses often show similar statistical tent with the threshold for downstream leakage of N from results when cutoffs of 250 km2, 500 km2, or 1000 km2 are forests receiving atmospheric deposition as described in used, but far less statistically powerful results when water- Aber et al. (2003). However, unlike the forests studied by sheds smaller than 250 km2 are included. Aber et al. (2003), the watersheds in our analysis receive We also explored incorporating TNI by adding the nat- other NANI components. The higher riverine N flux for ural rate of N fixation to NANI. The TNI approach is a given input of N deposition in the US as compared with conceptually attractive, because the mass balance for N that in Sweden is probably the result of these other input terms is more complete (Boyer et al. 2002). NANI components being greater in the US watersheds However, N fixation is difficult to measure, and data for (WebTable 1). particular regions or watersheds are seldom available. The net input of N in food and animal feeds has two www.frontiersinecology.org © The Ecological Society of America Riverine TN flux Riverine TN flux (kg N km–2 yr–1) (kg N km–2 yr–1) R Howarth et al. N fluxes from watersheds relationships with river N flux. This River flux driven by fertilizer 41 net input is positively correlated with inputs, which are greater riverine N flux when the net food where food/feed inputs aregreater River flux driven by and feed term is positive, and is nega- imported food (animal 8000 tively correlated with riverine N flux wastes, sewage) when the net food and feed term is 6000 negative (Figure 4). The positive cor- y = –0.27x + 517 r2 = 0.39 y = 0.41x + 261 relation for positive net inputs in 4000 r2 = 0.60 food and feed is driven by sewage and animal wastes from the imported 2000 food and feeds. The watersheds that 0 have large net negative inputs of N in –15 000 –10 000 –5000 0 5000 10 000 food and feeds (ie positive net Net food/feed (kg N km–2 yr–1) exports) are agricultural regions that export crop products. In these, the Figure 4. The net input of N in food for humans and in animal feeds has a complex export of food and feed is supported relationship with the flux of N in rivers, shown here for watersheds that are larger than by large inputs of synthetic N fertil- 250 km2. For those watersheds with a positive net input of N in food and feed (ie a net izer and/or N fixation. Indeed, over import of food and feeds; green squares), the flux of N in rivers increases as the net the entire dataset of 154 watersheds, import increases. This presumably reflects the influence of animal wastes and human the net food/feed term is negatively sewage. For those watersheds with a negative net input of N in food and feed (ie a net correlated with the sum of fertilizer export of food and feeds; red squares), the riverine N flux increases as the net import of and agricultural N fixation (Web- food and feed becomes more negative (ie the basin exports more). This is probably due to Figure 5). Thus, the negative correla- much greater input of synthetic N fertilizer in the watersheds with the greater export of tion of the net food/feed term with food and feed. Both of the relationships shown are highly significant: P = 1 × 10–14 for riverine N flux is clearly driven by the green squares and P = 6 × 10–8 for the red squares. fertilizer use and N fixation. average riverine N flux that is related to the long-term n The role of climate sinks in the landscape, which are primarily denitrification and accumulation of N in soils and biomass (van In an earlier paper that looked at only 16 northeastern Breemen et al. 2002)? That is, with the larger dataset now US watersheds, the fraction of NANI exported in rivers available, is there an influence of climate on the average was clearly correlated with precipitation and discharge amount of NANI exported over multiple-year periods but not with temperature (Howarth et al. 2006). For a aside from interannual storage and flushing? The answer similar analysis that included both northeastern and is yes: the fraction of NANI exported in long-term aver- southeastern US watersheds, Schaefer and Alber (2007) age riverine N flux is significantly correlated with tem- found that the fraction of NANI exported was correlated perature, precipitation, and discharge (WebFigure 6). with all of these climate variables, but suggested that The explanatory power of the relationships is weak for temperature had the strongest relationship. Conversely, both temperature and precipitation (r2 = 0.03 and 0.11; Schaefer et al. (2009) found no relationship between the P = 0.037 and 9 × 10–5, respectively). However, discharge fraction of NANI exported and any climate variable in is highly correlated with the fraction of NANI exported the western US. in rivers (r2 = 0.41, P = 5 × 10–16; WebFigure 6). The riverine N flux data in this paper and in Howarth One might question whether the relationship between et al. (2006), Schaefer and Alber (2007), and Schaefer et the fraction of NANI exported and discharge is a result of al. (2009) are all averages for multiple years. Other stud- auto-correlation, because discharge information is used to ies have demonstrated that when examining year-by-year estimate riverine N flux. The same question can be raised patterns, the fraction of NANI exported is greater in about studies that demonstrate an influence of discharge years with high discharge and less in years with low dis- on long-term average N flux from watersheds with low charge, but this can be explained as storage of N in the human impact (Lewis et al. 1999; Lewis 2002) or that landscape in dry years followed by flushing in wet years show the relationship of interannual N flux to discharge (McIsaac et al. 2001; Donner and Scavia 2007; Han et al. (McIsaac et al. 1999; Donner and Scavia 2007; Han et al. 2009; David et al. 2010). For watersheds with low inputs 2009; David et al. 2010). In fact, this is not of concern, of anthropogenic N, long-term average riverine N fluxes because the discharge information used to estimate river- are greater in those having higher discharge, but this may ine N fluxes is taken at short time intervals and multi- be the result of differences in rates of natural N fixation plied by the N concentration over the same time interval. (Lewis et al. 1999; Lewis 2002; Howarth et al. 2006). The N concentration generally is not a simple function of Here, we return to the question raised in Howarth et al. this short time discharge, and concentrations can be (2006): is there an influence of climate on the long-term higher or lower at different discharge rates, with different © The Ecological Society of America www.frontiersinecology.org Riverine TN flux (kg N km–2 yr–1) N fluxes from watersheds R Howarth et al. 42 trends in different systems (McDiffett et al. 1989; finding suggests fewer long-term sinks for N in the Bachman et al. 2002). Although annual average dis- landscape as it becomes wetter, with a greater percent- charge itself is indeed correlated with riverine N flux age of NANI exported to coastal waters. Managers across the watersheds in this study (P = 0.007 for the must consider this influence on N pollution in the face slope; WebFigure 7), the relationship has rather little of a changing climate, where in the future some explanatory power (r2 = 0.056). In part this is because regions will become wetter and others drier. of the human domination of the N cycle, which is cap- tured in NANI. Discharge better explains the fraction n Acknowledgements of NANI exported (WebFigure 6) than riverine N flux (WebFigure 7). Funding was supplied in part from the National We used a multiple regression approach to explore Oceanic and Atmospheric Administration (NOAA) the influence of climate on N fluxes in rivers, exclud- through the Coastal Hypoxia Research Program, the ing small watersheds (< 250 km2; WebTable 2). We US Department of Agriculture through the tested models to estimate the riverine N flux based on Agriculture, Energy, and Environment Program at various functions of NANI, discharge, and tempera- Cornell University, and David R Atkinson through an ture, as well as models that either did or did not force endowment given to Cornell to support a professorship the intercept through zero. We followed the guidance awarded to RH. This paper resulted from workshops of Hirsch et al. (1993) and only tested simple regres- held in Sigtuna, Sweden, and Paris, France, funded by sion models based on physically plausible explanations Baltic Nest and Nine-ESF. This is Contribution for relating the climate variable to riverine N flow. #CHRP 138 from the NOAA Coastal Hypoxia The intercept term was never significant in these mod- Research Program. els, and here we show only the models with the zero intercept. Temperature alone as a term was never sig- n References nificant (WebTable 2). On the other hand, NANI Aber J, Goodale C, Ollinger SV, et al. 2003. Is N deposition terms were always significant, as were NANI-discharge altering the N status of northeastern forests? BioScience 53: interaction terms. The most parsimonious of these 375–89. models is based on a NANI term and a separate NANI- Alexander RB, Johnes PJ, Boyer EW, and Smith RA. 2002. A comparison of models for estimating the riverine export of discharge interaction term, with both being highly sig- N from large watersheds. Biogeochemistry 57/58: 295–339. nificant: predicted flux = NANI (0.00035Q + 0.115), Bachman LJ, Krantz DE, and Bohlke J. 2002. Hydrogeologic where Q is the average discharge. The predicted river- framework, ground-water geochemistry, and assessment of N ine N export is highly correlated (r2 = 0.86, P = 1 × yield from base flow in two agricultural watersheds, Kent 10–10) with the measured riverine N flux and is cen- County, Maryland. Cincinnati, OH: US Environmental tered on the 1:1 line (WebFigure 8). For our large set Protection Agency. EPA/600/R-02/008. Billen G, Garnier J, Nemery J, et al. 2007. A long term view of of watersheds in Europe and the US, the multi-year nutrient transfers through the Seine River continuum. Sci average flux of N in rivers can be explained with a Total Environ 275: 80–97. great deal of precision based simply on this NANI-dis- Billen G, Thieu V, Garnier J, and Silvestre M. 2009. Modelling charge model. the N cascade in regional watersheds: the case study of the Seine, Somme and Scheldt rivers. Agr Ecosyst Environ 133: 234–46. n Conclusions Boyer EW, Goodale CL, Jaworski NA, and Howarth RW. 2002. Anthropogenic N sources and relationships to riverine N The NANI approach provides a simple and robust export in the northeastern USA. Biogeochemistry 57/58: method for estimating the flux of N from temperate- 137–69. zone watersheds, including relatively small watersheds, Boyer EW, Howarth RW, Galloway J, et al. 2006. Riverine N export from the continents to the coasts. Global Biogeochem as well as insight on the major sources of N pollution Cy 20: GB1S91; doi:10.1029/2005GB002537. in the landscape. This evaluation of sources can pro- Cleveland CC, Townsend AR, Schimel DS, et al. 1999. Global vide guidance to water-quality managers regarding patterns of terrestrial biological N (N2) fixation in natural where to focus their efforts. For instance, field-scale systems. Global Biogeochem Cy 13: 623–45. agricultural practices should be the main focus for David MB, Drinkwater LE, and McIsaac GF. 2010. Sources of watersheds where N fertilizer dominates NANI, but nitrate yields in the Mississippi River basin. J Environ Qual39: 1657–67. treatment of wastes should be the greater focus for Donner SD and Scavia D. 2007. How climate controls the flux watersheds where the net import of N in food and feeds of N by the Mississippi River and the development of dominates. Although NANI alone is quite predictive hypoxia in the Gulf of Mexico. Limnol Oceanogr 52: 856–61. for estimating riverine N flux, a model that includes EPA–SAB (US Environmental Protection Agency–Science both NANI and discharge increases the precision of Advisory Board). 2008. Hypoxia in the northern Gulf ofMexico: an update by the EPA Science Advisory Board. the estimate. That the average flux of N in rivers Washington, DC: US EPA. Report EPA-SAB-08-003. increases as average multi-year discharge increases has Galloway JN, Dentener FJ, Capone DG, et al. 2004. N cycles: profound implications for managing N pollution. Our past, present, and future. Biogeochemistry 70: 153–226. www.frontiersinecology.org © The Ecological Society of America R Howarth et al. N fluxes from watersheds Han H and Allan JD. 2008. Estimation of N inputs to catch- sheds of the United States. Biogeochemistry 57/58: 375–85. ments: comparison of methods and consequences for river- Lewis WM, Melack JM, McDowell WH, et al. 1999. N yields 43 ine export prediction. Biogeochemistry 91: 177–99. from undisturbed watersheds in the Americas. Han H, Allan JD, and Scavia D. 2009. Influence of climate and Biogeochemistry 46: 149–62. human activities on the relationship between watershed N McDiffett WF, Beidler AW, Dominick TF, and McCrea KD. input and river export. Environ Sci Technol 43: 1916–22. 1989. Nutrient concentration–stream discharge relation- Hirsch RM, Helsel DR, Cohn TA, and Gilroy EJ. 1993. ships during storm events in first-order streams. Statistical analysis of hydrologic data. In: Maidment DR Hydrobiologia 179: 97–102. (Ed). Handbook of hydrology. New York, NY: McGraw-Hill. McIsaac GF, David MF, Gertner GZ, and Goolsby DA. 2001. Hong B, Swaney DP, and Howarth RW. 2011. A toolbox for cal- Net anthropogenic N input to the Mississippi River basin culating net anthropogenic nitrogen inputs (NANI). and nitrate flux to the Gulf of Mexico. Nature 414: 166–67. Environ Modell Softw 26: 623–33. NRC (National Research Council). 2000. Clean coastal waters: Howarth RW, Billen G, Swaney D, et al. 1996. Riverine inputs understanding and reducing the effects of nutrient pollu- of N to the North Atlantic Ocean: fluxes and human influ- tion. Washington, DC: National Academies Press. ences. Biogeochemistry 35: 75–139. Peierls BL, Caraco NF, Pace ML, and Cole JJ. 1991. Human Howarth RW, Ramakrishna K, Choi E, et al. 2005. Nutrient influence on river N. Nature 350: 386–87. management, responses assessment. In: Ecosystems and Schaefer SC and Alber MA. 2007. Temperature controls a lati- human well-being, vol 3. Washington, DC: Island Press. tudinal gradient in the proportion of watershed N exported Howarth RW, Swaney DP, Boyer EW, et al. 2006. The influence to coastal ecosystems. Biogeochemistry 85: 333–46. of climate on average N export from large watersheds in the Schaefer SC, Hollibaugh JT, and Alber M. 2009. Watershed N northeastern United States. Biogeochemistry 79: 163–86. input and riverine export on the west coast of the US. Howarth RW, Chan F, Conley D, et al. 2011. Coupled biogeo- Biogeochemistry 93: 219–33. chemical cycles: eutrophication and hypoxia in coastal van Breemen N, Boyer EW, Goodale CL, et al. 2002. Where did marine ecosystems. Front Ecol Environ 9: 18–26. all the N go? Fate of N inputs to large watersheds in the Lewis WM. 2002. Yield of N from minimally disturbed water- northeastern USA. Biogeochemistry 57/58: 267–93. Is ExEceSsAs ’Nsi tIrsosgeune isn tihne EU.cSo. Elonvgiryonmen su N Trends, Risks, and Solutions serie t:s es E i W # n1 E5 colo Nitrogen is both an essential nutrient and a pollutant, a benefit and a hazard, depending on form, gy location, and quantity. Agriculture, industry, and transportation have spread nitrogen liberally around the planet, with complex and interrelated consequences for ecological communities and human health. This Issue tracks nitrogen through its different chemical forms and biological incarnations as it progresses across economic, environmental, and regulatory bounds. The authors argue for a systematic approach to managing nitrogen and its consequences. ESA’s Issues in Ecology series • reports a consensus of a panel of scientific experts • uses clear, commonly understood language • focuses on issues related to the environment • is intended for scientists, educators, students, and decision makers • is reviewed by external experts for technical content Copies are available in print (order online) and as a PDF (free download) For more information, visit: www.esa.org/issues © The Ecological Society of America www.frontiersinecology.org