Highlights

  • Terrestrial N inputs and hydrology control patterns of atmospheric nitrate export

  • Forested lands exported less atmospheric nitrate than more agricultural and developed lands

  • The concept of kinetic N saturation can be applied to interpret atmospheric nitrate patterns across heterogenous watersheds

Introduction

Deposition of atmospheric nitrate (NO3−Atm) has increased dramatically worldwide during about the past 150 years (Galloway and others 2004). Despite declines in recent decades in some regions (Tørseth and others 2012; Li and others 2016), deposition remains elevated and contributes to the eutrophication and acidification of terrestrial and aquatic ecosystems globally (Galloway and others 2003; Kemp and others 2005; Clark and Tilman 2008). The specific impacts of NO3−Atm partially depend on whether it is processed (incorporated into the terrestrial nitrogen cycle) or exported unprocessed to surface waters. Terrestrial processing of NO3−Atm can provide longer-term storage (that is, assimilation) or removal (that is, denitrification), whereas stream export can have more immediate impacts, such as exacerbating nutrient pollution of downstream waters (Howarth and others 2000). Understanding the factors controlling the relative amounts of NO3−Atm that are processed versus exported to streams is needed to evaluate potential impacts on affected ecosystems.

Landscape properties represent a potentially dominant factor regulating the proportion of (unprocessed) NO3−Atm deposition that is exported in streamwater. NO3−Atm occurs across broad spatial extents (Driscoll and others 2001) and thus impacts diverse landscapes. Different land uses (for example, forest, agriculture, developed) are commonly associated with generalizable patterns of streamwater NO3− export (Jordan and others 1997; Groffman and others 2004; Kaushal and others 2008) that can partially be attributed to variable amounts and sources of nitrogen (N) inputs (Lovett and Goodale 2011), differing rates of key N cycling processes and/or alterations of hydrologic flowpaths (Sudduth and others 2013)—all of which could influence processing of NO3−Atm. For example, the conceptual kinetic N saturation model suggests that ecosystem N losses, including streamwater export, occur when rates of inputs (for example, from deposition, fertilizer) exceed sinks at various temporal scales (for example, vegetative uptake, immobilization; Lovett and Goodale 2011). This model was developed and has been applied to understand N deposition effects on streamwater NO3− export from predominantly forested watershed (Eshleman and others 2013), but it may be applicable to NO3−Atm processing and export from mixed land use watersheds with elevated N input rates (Eshleman and Sabo 2016). However, prior research into watershed cycling of NO3−Atm has focused primarily on predominantly forested or alpine watersheds (for example, Tsunogai and others 2010; Fang and others 2015; Osaka and others 2016; Bourgeois and others 2018a; Bourgeois and others 2018b; Sebestyen and others 2019) where deposition represents the primary input of N and streamwater NO3 export is generally low. Thus, the relative importance of potential controls on NO3−Atm dynamics associated with variable land uses and elevated, non-deposition N inputs is unclear (Burns and others 2009; Tsunogai and others 2016). Assessing the potential effects of land use on the fate of NO3−Atm requires accurate accounting of streamwater NO3−Atm export across watersheds with varied N sources, magnitudes of NO3− export, and hydrologic conditions, but this remains a major challenge.

Many prior studies have used δ18O values of NO3− in streamwater to distinguish atmospheric and terrestrial fractions (Kendall and others 1995; Burns and Kendall 2002; Burns and others 2009; Kaushal and others 2011). This approach takes advantage of NO3−Atm having elevated δ18O values (~ 60–90‰) relative to NO3− of terrestrial origin (δ18O ≅  − 15—+ 15‰; Kendall and others 2007; Michalski and others 2012). However, interpretation of δ18O as a tracer of NO3−Atm is complicated by many factors. For example, NO3− consumption (plant or microbial uptake, denitrification) can elevate the δ18O values of residual NO3− resulting in potentially overlapping ranges of δ18O values of terrestrial and atmospheric NO3− (Böttcher and others 1990; Kendall and others 2007). Additionally, dilution of the δ18O NO3−Atm signal is likely to be greatest in watersheds with high loads of streamwater NO3− export relative to atmospheric inputs (that is, agricultural watersheds), which, when combined with the large range of terrestrial δ18O values, can obscure the NO3−Atm signal. These complications are mitigated by an increasingly used tracer of NO3−Atm, triple oxygen isotopes of NO3−:

$$ \Delta^{17} {\text{O}} = \left( { \frac{{1 + \delta^{17} {\text{O}}}}{{(1 + \delta^{18} {\text{O}})^{\beta } }} - 1} \right) \times 1000 $$
(1)

where δ = (Rsample/Rreference)—1, R = ratio of heavy to light isotope, and β \(\cong\) 0.52 (Michalski and others 2003). The Δ17O value of terrestrial NO3− is ≅ 0‰ (Kendall and others 2007), and relative to δ18O, the Δ17O values of NO3−Atm (~ 20–30‰ in the mid-latitudes; Tsunogai and others 2010; Rose and others 2015; Tsunogai and others 2016; Bourgeois and others 2018b; Nelson and others 2018) in residual NO3− change minimally during biological processing (Young and others 2002; Michalski and others 2004; Kendall and others 2007). Furthermore, dilution of Δ17O values of NO3−Atm can occur, but the small range of Δ17O values for terrestrial NO3− (~ 0‰) allows for more accurate quantification of NO3−Atm, even in watersheds with high rates of streamwater NO3− export relative to deposition.

Measurements of nitrate Δ17O only allow for quantification of unprocessed NO3−Atm because they “trace” NO3− produced in the atmosphere. Terrestrial N cycling (immobilization, assimilation, mineralization, and nitrification) only retains the N atom of NO3−. Thus, there is a distinction between processing and retention (that is, proportion of NO3− inputs that are exported in streamwater on an annual basis) of deposited NO3−Atm. For example, a NO3−Atm molecule could theoretically be deposited, undergo terrestrial N cycling (that is, become immobilized, mineralized, then nitrified), and be exported as NO3− in streamwater a short time later (that is, days or weeks post-deposition), and would be considered processed (that is, the molecule would have a Δ17O \(\cong\) 0) but not retained. Thus, the fraction of NO3−Atm deposition that is processed represents the upper limit of watershed retention (that is, NO3−Atm processing ≥ retention).

Estimates of mean annual streamwater nitrate- Δ17O and NO3−Atm loads, which are not equivalent, provide a useful framework for assessing the relative rates of watershed-scale NO3− consumption (denitrification, immobilization, or assimilation) and addition (nitrification, fertilization; Figure 1). The relative rates of these processes affect both NO3−Atm and total NO3− (NO3−Total) cycling and streamwater export across diverse land uses. The difference in the mass of NO3−Atm deposited and exported in streams is caused by NO3− consumption processes along hydrologic flowpaths, which do not alter the Δ17O value of the residual NO3− (Böttcher and others 1990; Michalski and others 2004; Kendall and others 2007). Reduction in the Δ17O value of deposited NO3− is caused by the addition of new microbially or synthetically sourced nitrate or dilution by existing terrestrial NO3− (for example, synthetic fertilizer, nitrification) with Δ17O ≅ 0‰ (Kendall and others 2007) encountered along hydrologic flowpaths. This framework for assessing relative rates of watershed-scale NO3− consumption and addition is primarily possible because of the unique triple oxygen isotopic tracer of NO3−Atm, but also due to the widespread deposition of NO3−Atm across watersheds (Driscoll and others 2001) and the relative mobility of NO3− (Chapin and others 2011). By quantifying the processing and export of NO3−Atm across watersheds with varied land use, we use this framework to assess watershed-scale N cycling dynamics.

Figure 1
figure 1

Framework for interpreting variation in Δ17O- NO3− and NO3−Atm concentrations. These indicators provide different, yet complementary information about watershed-scale N cycling processes. Δ17O of nitrate is equal to the fraction of NO3−Atm (red circles) relative to NO3−Total (black circles, sum of NO3−Terr and NO3−Atm) multiplied by the Δ17O of deposition. Left panel) Addition of NO3−Terr or dilution of NO3−Atm by NO3−Terr decreases the Δ17O of a “reservoir” of NO3− by increasing NO3−Total along hydrologic flowpaths prior to export in streamwater, which is illustrated by the increasing area of the black outlined circle relative to NO3−Atm (red square). Addition of NO3−Terr does not change the concentration of NO3−Atm (area of the red circle). Right panel) NO3− consumption (for example, denitrification, assimilation, immobilization) processes reduce the concentration of NO3−Atm from deposition, along hydrologic flowpaths, before eventual export in streamwater, but does not change the Δ17O value of residual nitrate (indicated by the constant area of the black outlined circle relative to NO3−Atm, the red circle). NO3− consumption is a mass-dependent fractionation process and therefore does not alter the Δ17O (result of mass-independent fractionation processes) of the NO3− reservoir.

Full size image

Here we ask the following questions: How do terrestrial N inputs and land use influence the cycling and surface water export of NO3−Atm at the watershed scale? More specifically, what is the relationship between terrestrial N inputs, the proportion of major land use categories (forest, agriculture, and developed) in watersheds and NO3−Atm concentrations, yields, and processing efficiency (that is, fraction of NO3−Atm deposition that is processed prior to surface water export)? To address these questions, we measured Δ17O values of NO3− on 832 stream samples collected during both baseflow and stormflow conditions from 14 watersheds of varied land use in the Chesapeake Bay watershed in eastern North America during a two-year period. We hypothesize that predominantly forested watersheds with lower terrestrial N inputs will have lower NO3−Atm concentrations and yields, and higher processing efficiency, than watersheds that are predominantly agricultural and/or developed with higher rates of terrestrial N inputs. If our results show that increased NO3−Atm concentrations are positively related to terrestrial N inputs, it would provide support for extending the kinetic N saturation conceptual model to NO3−Atm streamwater export across varied land use watersheds.

Materials and Methods

Study Sites and Field Methods

To assess land use effects on NO3−Atm dynamics across a range of hydroclimatological conditions, streamwater samples were collected from 14 watersheds varying in size (160–127,900 ha), dominant land use (96% forest to 70% developed), and mean annual temperature and precipitation (Table 1). Streamwater grab samples (120–1000 mL) were collected both regularly (2 samples per month) and irregularly during storm events (~ 6–10 samples per year; n = 57–65 total samples per watershed) from the outlets of 14 gaged watersheds within the Chesapeake Bay basin from October 2015–September 2017 (that is, water years 2016 and 2017; Figures S1 and S2). Samples were collected in pre-cleaned polypropylene bottles and kept on ice for 2–4 h before being refrigerated until they were then processed in the laboratory within 24–48 h. Sampling across a range of hydrological conditions (Figure S2) was done to more fully capture streamwater NO3−Atm variation and to improve accuracy of estimated NO3−Atm loads. Estimated loads of many other streamwater constituents (total nitrogen, total NO3−, total phosphorus, and so on) are more accurate when samples are collected over a range of hydrological conditions (Sprague 2001). Daily stream discharge data were obtained from U.S. Geological Survey records for ten of the study watersheds. Stream discharge in the other four watersheds (Table 1) was measured by the authors using comparable stream gaging practices. These practices involve development of a rating curve (log–log regression of discharge vs. stage) for each station and computation of mean daily discharge based on hourly stage data from a digital water level recorder. Weekly precipitation samples during water year (WY) 2017 were obtained from three National Atmospheric Deposition Program (NADP) sites (PA00, MD99, and MD08) bounding the study watersheds (Figure S1). Precipitation NO3− concentration and isotope data are summarized in the Supporting Information (SI). Land use percentages were calculated from the 2016 National Land Cover Database; agricultural land represents the sum of both cultivated crop and pasture/hay land classes (Homer and others 2020). Mean watershed slope (m/m) was obtained using the U.S. Geological Survey StreamStats program (USGS 2016).

Table 1 Watershed Attributes
Full size table

Laboratory Methods

Stream samples were filtered (0.45 µm) and frozen within 48 h of collection. Stream NO3− and nitrite (NO2−) concentrations were measured using flow-injection colorimetric analysis (Lachat Quickchem 8000 FIA +). Weekly precipitation NO3− concentration data were provided by the NADP Central Analytical Laboratory (NADP 2021).

Triple oxygen isotopes (16O, 17O, and 18O) of stream and precipitation NO3− were measured using a Thermo Delta V+ isotope ratio mass spectrometer (Bremen, Germany) via the denitrifier method (Sigman and others 2001; Casciotti and others 2002) with thermal decomposition (at 800° C) of N2O to N2 and O2 at the Central Appalachians Stable Isotope Facility (Kaiser and others 2007). NO2− is denitrified using this method as well, but NO2− concentrations in stream and precipitation samples were low relative to NO3− (NO2−/( NO2− +  NO3−) mean = 0.01, range = 0.00–0.11). Measured oxygen isotope ratios were calibrated to international reference standards USGS 34 (δ17O = − 14.8‰, δ18O = − 27.9‰) and USGS 35 (δ17O = 51.5‰, δ18O = 57.5‰; Böhlke and others 2003) measured throughout sample analysis in equal concentrations to samples (ranging from 100–200 nmol depending on sample NO3− concentration). Analytical precision of Δ17O values of NO3− was 0.5‰ (1σ) as determined by repeated measurements (n ≅ 200) of international reference standard USGS 32 (mean measured Δ17O ≅ − 0.2‰) and laboratory reference standard “Chile NO3−” (Duda Energy 1sn 1 lb. Sodium Nitrate Fertilizer 99+ % Pure Chile Saltpeter from Amazon.com; mean measured Δ17O ≅ 19.7‰) made during runs associated with these streamwater samples over 3+ years. Additionally, streamwater sample replicates were analyzed (n = 60) and had a pooled standard deviation of 0.5‰. Accuracy of Δ17O was tracked using repeated measurements of IAEA-N3 (n = 19, µ = − 0.1‰, 1 σ  = 0.5‰) and closely agreed with published values of − 0.2‰ (Michalski and others 2002; Böhlke and others 2003).

Quantification of Unprocessed Atmospheric NO3 − in Streams and Uncertainty Estimation

Mean streamwater nitrate- Δ17O (\(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\)) for each watershed was calculated over the entire study period to provide an aggregate estimate of watershed response. Analytical uncertainty of individually measured Δ17O samples was incorporated into \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) by sampling with replacement (that is, bootstrapping) from a probability density function that incorporated both normal and uniform distributions (additional details are provided in the SI). \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) was used to calculate the mean percentage of unprocessed NO3−Atm in stream samples using Eq. 2:

$$ \% {\text{NO}}_{{3\;{\text{Atm}}}}^{ - } = \frac{{\left( {\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }} - \Delta^{17} {\text{O}}_{{{\text{Terr}}}} } \right)}}{{\overline{{\left( {\Delta^{17} {\text{O}}_{{{\text{Precip}}}} - \Delta^{17} {\text{O}}_{{{\text{Terr}}}} } \right)}} }} \times 100 $$
(2)

where \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Precip}}}} }}\) = mean Δ17O of wet NO3− deposition during WY2017, and Δ17OTerr =  Δ17O of terrestrially sourced NO3−. We assumed that the annual mean isotopic composition of NO3− in precipitation did not significantly differ between WY2016 and WY2017. Data from a three-year record in the mid-latitudes (inter-annual range = 1.5‰) suggest this assumption is reasonable (Tsunogai and others 2016). Uncertainty in % NO3−Atm from all three parameters in Eq. 2 and was estimated using bootstrapping methods. Values for each parameter in Eq. 2 were randomly sampled from distributions that accounted for analytical uncertainty (\(\Delta^{17} {\text{O}}_{{{\text{Stream}}}}\)), natural intra-annual variation (\(\Delta^{17} {\text{O}}_{{{\text{Precip}}}}\)), and potential variability in β values (Δ17OTerr) during mass-dependent fractionation processes (for example, nitrification, denitrification) that could generate non-zero Δ17O values not attributable to NO3−Atm (Young and others 2002; Kaiser and others 2007). This approach resulted in a distribution of % NO3−Atm that was then used to propagate uncertainty (that is, sample from this distribution with replacement) through additional calculations. The Δ17O value of terrestrial NO3− is commonly assumed to be exactly 0‰ (Sabo and others 2016; Tsunogai and others 2016; Nakagawa and others 2018; but see Rose and others 2015), but previous studies reported negative values 3–4 times beyond the standard deviation of instrument uncertainty (Rose and others 2015; Yu and Elliott 2018) suggesting that β values are not necessarily stable during complex N cycling reactions and/or Δ17O of terrestrial NO3− is not always equal to 0‰. Our approach attempts to account for some of these yet unquantified effects that may cause Δ17O of terrestrial NO3− to deviate from 0‰ by allowing β to vary from 0.51–0.53. Additional details of uncertainty estimation and propagation are provided in the SI.

We acknowledge that natural, or “organic”, NO3− fertilizers (for example, mined from desert deposits and classified as organic) can have Δ17O > 0‰ (Michalski and others 2015). No data on application of this NO3− fertilizer use exist for our watersheds, although it represents a minor percentage (< 0.01%) of N fertilizer applied nationally since ~ 1970 (Böhlke and others 2009). Mean annual flow-weighted concentrations and yields of NO3−Atm were quantified using Eq. 3:

$$ {\text{NO}}_{{3{\text{Atm}}}}^{ - } = \% {\text{NO}}_{{3{\text{Atm}}}}^{ - } \times {\text{NO}}_{{3{\text{Total}}}}^{ - } $$
(3)

where NO3−Total = either annual flow-weighted concentrations (mg N L−1) or yields (kg N ha−1) of NO3−Total.

Daily NO3−Total loads (LNO3, kg d−1) were computed using Weighted Regression on Time, Discharge, and Season-Kalman Filter (WRTDS-K; (Zhang and Hirsch 2019). Models were calibrated using the entire period of record for NO3−Total (11–33 years). The use of the entire record ensured that model coefficients were representative of a greater range of hydroclimatological conditions than was realized in two water years. Estimated daily loads of NO3−Total were summed for WY2016–2017, normalized by watershed area and divided by the period of record (2 years) to compute annual average yields (kg N ha−1 y−1). Flow-weighted annual mean concentrations were calculated by dividing annualized loads by annual discharge for WY2016–2017. NO3−Total uncertainty (annual concentrations and yields) was estimated using block bootstrapping methods and are detailed in the SI. NO3−Atm uncertainty (concentrations and yields) incorporated both NO3−Total and % NO3−Atm uncertainty through bootstrapping, or sampling with replacement from distributions of both NO3−Total and % NO3−Atm.

NO3 − Deposition

Grids of NO3− in wet deposition were generated using NO3− concentration data and point precipitation data from NADP and gridded precipitation data from the PRISM Climate Group for WY2016–2017 (PRISM Climate Group 2004). Interpolated surfaces of monthly precipitation-weighted NO3− were generated using inverse distance weighting and then multiplied by PRISM precipitation data to produce water year NO3− deposition grids. Watershed-scale mean NO3− wet deposition was computed as the areal average of deposition within the watershed boundary.

Processing Efficiency of Atmospheric NO3 −

Processing efficiency (PE), defined as the percentage of deposited NO3− that is incorporated into the terrestrial N cycle (that is, Δ17O is reset to \(\cong\) 0‰) prior to stream export, which builds on a similar metric as Barnes and others (2008), was calculated as:

$$ {\text{PE}} = \left( {1 - \frac{{{\text{NO}}_{{3_{{{\text{Atm}}}} }}^{ - } \left( {{\text{kg}}\;{\text{N}}\;{\text{ha}}^{ - 1} \;{\text{yr}}^{ - 1} } \right)}}{{{\text{NO}}_{{3_{{{\text{Precip}}}} }}^{ - } \left( {{\text{kg}}\;{\text{N}}\;{\text{ha}}^{ - 1} \;{\text{yr}}^{ - 1} } \right)}}} \right) \times 100 $$
(4)

NO3− in wet deposition was used for this calculation. It has previously been assumed that dry NO3− deposition is similar in magnitude to wet NO3− deposition (Lovett and Lindberg 1993; Boyer and others 2002; Grigal 2012; Eshleman and Sabo 2016), which implies that PE values are uniformly underestimated across all watersheds. Scenarios in which this assumption may be violated are presented in the SI. PE uncertainty was estimated from bootstrapped distributions of NO3−Atm yield.

Terrestrial N Inputs

Rates of terrestrial N inputs (in kg N ha−1 y−1) to watersheds were obtained from the Chesapeake Bay Program Chesapeake Assessment and Scenario Tool (Chesapeake Bay Program 2020). Estimates of terrestrial N inputs are made at the county scale and assigned to specific land uses (for example, developed, agriculture). These inputs were aggregated to the watershed scale by calculating the percentage of each land use in different counties for all study watersheds.

Statistical Analyses

Weighted least squares regression (dependent variables weighted by 1/σ , where σ = standard deviation) of mean annual Δ17O values, NO3−Atm concentrations, and PE to land use percentages and terrestrial N input rates was used to estimate slopes because of the non-uniform error in y-values (Bevington and Robinson 2003). The coefficient of determination (r2) was used to assess regression fit, and r2 values are reported as the median of all bootstrapped replicates. Significance of linear regression slopes was determined via bootstrapping at α = 0.05; reported p-values are the proportion of 10,000 slope estimates that are either greater than or less than zero (depending on the direction of the relationship). Welch’s ANOVA was used, due to heterogeneity of variances, to compare means (that is, Δ17O, NO3−Atm) between individual watersheds, watersheds grouped by dominant land use and rates of terrestrial N inputs, and across flow conditions (McDonald 2009). All statistical analyses were performed in R (R Development Core Team 2019).

Results

Mean annual precipitation NO3− concentrations ranged from 0.140–0.160 mg N L−1 and wet NO3− deposition ranged from 1.47–1.77 kg N ha−1 y−1 during WY 2016–2017 (Table S1). Annual areal mean precipitation depth ranged from 101–123 cm (Table S1). Δ17O values of precipitation NO3− ranged from 16.4–29.3‰ with elevated values in the winter and lower values in the summer (Figure S3) and a depth-weighted annual mean (± standard error) of 25.2‰ ± 0.3‰.

In individual streamwater samples, NO3−Total concentrations ranged from 0.001–5.139 mg N L−1 and yields of NO3−Total ranged from 0.60–11.64 kg N ha−1 y−1 (Figure 2; Table S2). Values of Δ17O in individual stream samples ranged from − 1.0–5.0‰, corresponding to % NO3−Atm from 0–21% (Figure 2), and δ18O ranged from − 11.5–14.8‰ (Figure S4). NO3−Atm concentrations in individual samples, calculated using NO3−Total and Δ17O, ranged from 0–0.267 mg N L−1. Averaged over the entire study period (WY2016–2017), \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) ranged from 0.2–1.3‰ across watersheds, representing 1–5% NO3−Atm, and mean flow-weighted NO3−Atm concentrations ranged from 0.007–0.062 mg N L−1 (Table S2). Yields of NO3−Atm ranged from 0.03–0.30 kg N ha−1 yr−1, comprising 1.4–5.8% of total NO3− (NO3−Total) loads in study watersheds during WY2016 and 2017 (Table S2).

Figure 2
figure 2

Box and whisker plots of a NO3− concentrations, b Δ17O- NO3−, c unprocessed atmospheric NO3− percentages, and d unprocessed atmospheric NO3− concentrations. Watersheds are colored and grouped by general land use category: predominantly forested (> 80% forested), mixed agriculture/forest (> 25% both forested and agriculture), and predominantly developed (> 70% developed). Lines in boxes indicate median, upper and lower hinges represent 25 and 75th quartile, whiskers extend 1.5 × inter-quartile range, points beyond this range are plotted individually, and notches in boxes represent ~ 95% confidence interval of median. Asterisk denotes single watershed with significantly different mean from all others.

Full size image

Watershed land use percentage was a statistically significant linear predictor of nearly all NO3−Atm metrics. A higher percentage of agricultural land use was found to predict lower values of \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) and PE (r2 = 0.24, p < 0.0001 for \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\); r2 = 0.15, p = 0.0687 for PE) and higher mean annual flow-weighted NO3−Atm concentrations (r2 = 0.17, p < 0.05; Figure 3). These relationships were generally opposite for forested land use: after removing an outlier (GWN, our most developed watershed), higher percentages of forested land use predicted higher values of \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) (r2 = 22, p < 0.005) and lower mean annual flow-weighted NO3−Atm concentrations (r2 = 0.30, p < 0.0005) and PE (r2 = 0.30, p < 0.0005).

Figure 3
figure 3

Scatter plots of land use percentages and mean annual Δ17O, NO3−Atm, and processing efficiency. Solid line is weighted least squares regression line, dashed lines are bootstrapped 95% confidence intervals, r2 is median of all bootstrapped replicates. Regressions with % developed land use should be interpreted with caution as only one watershed contained > 20% of this land use type.

Full size image

Rates of terrestrial N inputs ranged from 1.3–64.9 kg N ha−1 y−1 averaged over calendar years 2016–2017 (Table 1). Unsurprisingly, terrestrial N input rates were strongly positively correlated with agricultural (r = 0.96) and negatively correlated with forested land use (r = − 0.78) and thus exhibit similar statistical relationships with NO3−Atm related metrics. Elevated terrestrial N input rates predicted lower \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) and PE (r2 = 0.25, bootstrapped p-value < 0.0001 for \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\), r2 = 0.20, p = 0.012 for PE) and higher NO3−Atm (r2 = 0.23, p = 0.010; Figure 5).

Discussion

Using our results from watersheds with varied land use and our framework for interpretation (Figure 1), we present a conceptual model of proposed controls on NO3−Atm dynamics (Figure 4). In this model, elevated rates of terrestrial N inputs relative to NO3− consumption allow proportionally more NO3−Atm to bypass processing and be exported in surface water. This imbalance between terrestrial N inputs and consumption additionally results in elevated NO3−Total concentrations, lowering the Δ17O and % NO3−Atm of streamwater NO3−. Generally, watersheds with appreciable agricultural land use (> 35%) are associated with elevated terrestrial N inputs (for example, from fertilizer), resulting in higher NO3−Atm concentrations with lower PE. Conversely, predominantly forested watersheds have lower terrestrial N inputs, with an inferred approximate balance between inputs and consumption, resulting in much of the deposited NO3− being processed (high PE) and thus NO3−Atm export being low. Impervious surfaces in developed portions of watersheds are an additional control on streamwater NO3−Atm patterns. These surfaces likely promote the rapid routing of deposited NO3−Atm to channels, especially during storm events, and decrease the potential for biologic processing.

Figure 4
figure 4

Conceptual model presenting the effects of land use on NO3−Atm (red circles) dynamics. Δ17O (ratio of red to yellow circles) and NO3−Atm concentrations and fluxes (represented by number of red circles in streamwater) are altered between deposition and export in streamwater by rates NO3− addition (purple arrow) and consumption processes (green arrow), respectively. Imbalances between relative rates of NO3− addition and consumption (agricultural land uses), hydrologic bypassing of biotic retention mechanisms (developed land uses), and tight cycling of NO3− and similar rates of addition and consumption processes (forested land uses) are proposed as the land use effects on observed patterns of NO3−Atm dynamics (stream export and watershed processing efficiency). Symbols courtesy of the Integration and Application Network, University of Maryland Center for Environmental Science (ian.umces.edu/symbols/).

Full size image

Elevated rates of terrestrial N inputs to watersheds associated with land use patterns decrease \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\), increase mean annual NO3−Atm concentrations, and decrease PE (Figure 5).

Figure 5
figure 5

Scatter plots of terrestrial N input rates and mean annual Δ17O, NO3−Atm, and processing efficiency. Solid line is weighted least squares regression line, dashed lines are bootstrapped 95% confidence intervals, and r2 is median of all bootstrapped replicates. Points are colored by dominant land use.

Full size image

One could argue that the relationships described between land use, terrestrial N input rates, and various metrics of NO3−Atm dynamics result from multiplying relatively similar Δ17O values by variable NO3−Total concentrations. However, these metrics (\(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\),NO3−Atm concentrations, and PE) account for multiple sources of uncertainty, including analytical uncertainty of Δ17O, β values (Eq. 1), Δ17O end-members (both terrestrial and atmospheric), and annual NO3−Total concentrations and yields. As such, our methods represent an improvement in uncertainty quantification relative to previous research using Δ17O values to quantify NO3− sources in streamwater (Tsunogai and others 2014; Rose and others 2015; Sabo and others 2016; Tsunogai and others 2016; Nakagawa and others 2018). The multiple sources of uncertainty in \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\), NO3−Atm concentrations, and PE were propagated and incorporated into linear regressions with land use and terrestrial N inputs. Accounting for this uncertainty reduced r2 values (reported as the median r2 of 10,000 bootstraps) and increased p-values (reported as the proportion of 10,000 bootstrap slopes either greater or less than zero, depending on the specific regression) relative to simple linear regression, yet nearly all relationships between land use and terrestrial N inputs with \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\), NO3−Atm concentrations, and PE remain significant (Figures 3 and 5). Thus, we argue that these results are a manifestation of biologic controls on NO3−Atm dynamics and can be interpreted as an extension of the kinetic N saturation conceptual model.

Our results suggest that biologic sinks of NO3− (that is, NO3− consumption) can be overwhelmed by high rates of N inputs, allowing proportionally more NO3−Atm to bypass processing and be exported in surface waters. This idea extends kinetic N saturation (Lovett and Goodale 2011) to streamwater NO3−Atm export and from forested to non-forested watersheds, while building on previous work applying traditional N saturation “stages” (Ågren and Bosatta 1988; Aber and others 1989) to understanding streamwater NO3−Atm export (Rose and others 2015; Nakagawa and others 2018). We note, however, that our extension of the kinetic N saturation conceptual model focuses on processing of NO3−Atm while past work primarily focused on retention of atmospherically deposited N. We are also focusing on inputs (deposited NO3−) that move through watersheds to a specific sink (streamwater export) without biological transformation. Nonetheless, kinetic N saturation focuses on rates of both inputs and sinks and proposes that N saturation effects, including increased leaching of NO3− to surface water, are only realized when rates of inputs exceed those of sinks. Our framework for interpretation (Figure 1) can be used to infer the role of both inputs (terrestrial N inputs) and sinks (NO3− consumption) on NO3−Atm export at the watershed scale.

Large terrestrial N inputs associated with agricultural land use allow more NO3−Atm to be exported and reduce PE (Figures 3 and 5). An imbalance between N inputs (for example, fertilizer) and demand for NO3− (for example, crop uptake, denitrification) creates an accumulation of NO3− in soils and groundwater. NO3− accumulation in agricultural systems is aligned with research suggesting that N supplies in excess of demand shift soils to NO3− dominated “economies”, as there is less competition for N and nitrifying microorganisms thrive (Schimel and Bennett 2004; Booth and others 2005). The large N inputs combined with the relative mobility of NO3− compared to reduced or organic N forms (Chapin and others 2011) ultimately results in increased export of NO3− in surface waters. The imbalance between N inputs and NO3− demand does not imply that NO3− consumption is reduced; rather, rates may even be greater in watersheds with larger terrestrial N inputs—for example, denitrification rates are generally higher in fertilized agricultural soils compared to non-fertilized soils (Barton and others 1999; Hofstra and Bouwman 2005). For a given NO3− consumption rate, however, a larger reservoir of NO3− (for example, more NO3− in groundwater and soil) available for consumption along hydrologic flowpaths likely allows proportionally more NO3−Atm to escape consumption and be exported in surface waters.

In predominantly forested watersheds with lower terrestrial N input rates, it is more likely that inputs and consumption are closer to unity on an annual basis resulting in lower NO3−Atm concentrations and yields, and higher PE. Reduced rates of N inputs likely contributed to NO3− consumption processes imparting a seasonal signal on NO3−Atm concentrations, similar to previous research on streams with low NO3−Total concentrations (Figure S5; Barnes and Raymond 2010; Tsunogai and others 2014; Rose and others 2015; Sabo and others 2016; Hattori and others 2019). Mean NO3−Atm concentrations were about 1.7 × higher in the dormant than growing season in watersheds with lower terrestrial N inputs rates (< 40 kg N ha−1 y−1, > 75% forested land use; ANOVA, p < 0.001), whereas concentrations were not significantly different between seasons in watersheds with higher terrestrial N input rates (> 40 kg N ha−1 y−1, < 52% forested land use; Figure S5). This result likely reflects higher rates of biologically-mediated NO3− consumption processes during the growing (warmer) season. For example, forest canopies can process up to 90% of NO3−Atm during the growing season, severely reducing the potential for NO3−Atm streamwater export (Inoue and others 2021). It is likely that rates of NO3− consumption also increase during the growing season in watersheds with elevated terrestrial N input rates, but that the amount of NO3− consumed is small relative to the total NO3− present, making it difficult to decipher the signal. One factor that may confound the interpretation of intra-annual NO3−Atm concentrations is the seasonal pattern of Δ17O values of NO3− in precipitation. Seasonal patterns in Δ17O values of NO3− in precipitation were similar across all monitoring sites in our study (Figure S3), however, suggesting that this effect would have been consistent across all watersheds.

Our results, combined with others using Δ17O values of NO3− to quantify NO3−Atm, supports the application and extension of kinetic N saturation to NO3−Atm dynamics: annual flow-weighted NO3−Atm concentrations are positively related to NO3−Total concentrations across 56 watersheds from five publications (our watersheds: r2 = 0.66, p < 0.001; others r2 = 0.25, p < 0.001; Figure S6). The magnitude of these relationships is slightly different between our study and others possibly due to differences in sampling frequency, which ranged from quarterly (4 per year; Tsunogai and others 2016) to weekly (Rose and others 2015), range of hydrologic conditions sampled (for example, baseflow only, baseflow and stormflow sampling), load estimation methods (Rose and others 2015; Tsunogai and others 2016; Nakagawa and others 2018) and/or watershed size (Sabo and others 2016), making it challenging to uncover potential causes of these differences in magnitude. Watersheds in these studies additionally represent diverse land uses (forested, urban, agricultural, mixed) and span NO3− deposition gradients (wet = 1.5–2.4 kg N ha−1 y−1, wet + dry = 3.3–6.4 kg N ha−1 y−1). Despite these methodological and physical differences, the direction of the relationships between NO3−Atm and NO3−Total is the same. Unfortunately, we do not have estimates of terrestrial N inputs for those watersheds included in the ancillary publications, but streamwater NO3−Total concentrations are a reasonable proxy of watershed-scale N inputs. NO3−Total concentrations integrate watershed-scale rates of both N inputs and sinks, and elevated NO3−Total concentrations suggest that inputs exceed sinks, allowing proportionally more NO3−Atm export in streamwater.

Large terrestrial N input rates result in the dilution of \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\). This dilution effect is clearly evident in our results: \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) is negatively related with terrestrial N inputs (r2 = 0.25, p < 0.001, Figure 5) and agricultural land use (r2 = 0.24, p < 0.0001, Figure 3), even after removing an outlier with high leverage (after GWN removal: r2 = 0.22, p < 0.005). This interpretation follows the implicit assumption that N inputs and storage are in the form of NO3−. We do not have data to differentiate N forms (for example, ammonium, organic N) of inputs at our study sites. Thus, we assume that the ratio of NO3− to total N of terrestrial N inputs and storage is similar across all watersheds. Reduction or dilution of Δ17O between deposition and streamwater export assumes mixing of both NO3−Atm (Δ17O \(\cong\) 25‰) and NO3−Terr (Δ17O \(\cong\) 0‰) along hydrologic flowpaths. The negative linear relationship between terrestrial N input rates and \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) indicates mixing is likely occurring in all watersheds, with one exception: GWN, our most developed watershed.

Impervious surfaces in developed portions of watersheds can exert hydrologic controls on Δ17O values, NO3−Atm concentrations, and PE. Overland runoff from impervious surfaces, if hydrologically connected to channels, provides a mechanism by which precipitation and dissolved substances within (for example, NO3−Atm) can be directly routed to channels and streams (Brabec and others 2002; Tsunogai and others 2016). Direct routing of water to streams effectively short-circuits terrestrial processing that either removes NO3−Atm (for example, denitrification) or dilutes Δ17O (for example, nitrification). This impervious area effect likely contributed to both the high \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\), NO3−Atm concentrations and yields, and reduced PE in GWN (Table S2). Impervious surface effects were most apparent during storm events: GWN was the only watershed in which both Δ17OStream and NO3−Atm were significantly higher during storm events relative to baseflow (Figure S7). Our results, while derived from a single watershed, provide additional evidence supporting studies that measured elevated NO3−Atm using either δ18O (Burns and others 2009; Hall and others 2016; Yang and Toor 2016) or Δ17O (Riha and others 2014; Tsunogai and others 2016) in developed watersheds.

Measurements of Δ17O- NO3− highlight the challenges of using δ18O alone for source apportionment in mixed land use watersheds. Terrestrial N inputs associated with agricultural activities include fertilizer, some of which may be synthetic NO3− fertilizer. This is plausibly supported by δ18O of streamwater NO3−; mean annual δ18O was positively correlated with agricultural land use in our watersheds (p < 0.0001, r2 = 0.19; Figure S8). Synthetic NO3− fertilizer is formed from tropospheric O2 and inherits a δ18O signature of ~ 24‰ (Michalski and others 2015). Alternatively, the relationship between mean annual δ18O and agricultural land use could be interpreted as the result of increased denitrification in agricultural areas, which can increase the δ18O of residual NO3− (Böttcher and others 1990; Kendall and others 2007). These competing interpretations demonstrate one of the difficulties in using δ18O alone to quantify streamwater NO3−Atm in mixed land use watersheds; it is impossible to assign a specific δ18O NO3−Terr end-member to watersheds with multiple sources of NO3−Terr. The use of Δ17O as a tracer is also limited in watersheds with large inputs of terrestrial N that result in elevated NO3−Total streamwater export relative to NO3−Atm deposition. For example, in a hypothetical watershed with 10 kg N ha−1 y−1 NO3−Total streamwater export, 1 kg N ha−1 y−1 of NO3−Atm deposition and a PE = 80%, \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) would only equal 0.5‰. As the ratio of NO3−Atm deposition to NO3−Total streamwater export decreases, \(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\) also decreases for a constant PE, making it increasingly difficult to detect NO3−Atm in streamwater regardless of the isotopic tracer (Figure S9).

In conclusion, land use influenced all metrics of NO3−Atm dynamics (\(\overline{{\Delta^{17} {\text{O}}_{{{\text{Stream}}}} }}\),NO3−Atm concentrations and yields, PE). Insights into watershed-scale, land -use specific processes affecting NO3−Atm were possible through measurements of Δ17O, a conservative tracer of NO3−Atm, on streamwater samples collected under a range of hydrologic conditions across numerous watersheds. Agricultural land use with elevated rates of terrestrial N inputs was associated with increased streamwater export of NO3−Atm relative to predominantly forested watersheds. Large terrestrial N inputs in agricultural lands overwhelmed N sinks and allowed proportionally more NO3−Atm to escape consumption (denitrification, assimilation, immobilization) and be exported in surface waters. Development in watersheds likely increased NO3−Atm export due to hydrologic connectivity of overland flowpaths that bypass potential biological processing, supporting previous NO3−Atm research in developed watersheds. Accordingly, future changes to land use patterns and rates of terrestrial N inputs to watersheds will likely increase (that is, urbanization, increased fertilizer application rates) or decrease (that is, reforestation of agricultural lands, reduced fertilizer application rates) the fraction of deposited NO3−Atm that is exported in streamwater that directly contributes to nutrient pollution of downstream ecosystems.