RESULTS
Nitrate concentrations under different land uses. Nitrate data from 228 sites across the contiguous United States were analyzed (See Detailed Methods). This includes 147 sites from the National Water-Quality Assessment (NAWQA) project (Spahr et al., 2010) and 71 sites from the Water Quality Watch (WQW) (https://waterwatch.usgs.gov/wqwatch/) within the United States Geological Survey (USGS), and 10 sites from intensively monitored sites via research networks in the U.S.. Higher stream nitrate concentrations generally cluster in the Midwestern “Corn Belt” that is dominated by agricultural and mixed lands, whereas undeveloped and urban sites consistently have lower nitrate concentrations (Figure 2). The soil water and groundwater nitrate concentrations (Figure S1, 81 sites) estimated from stream concentrations generally follow the spatial pattern of stream nitrate concentration, indicating the penetration of nitrate into deeper groundwater.
Statistical boxplot (Figure 3) confirmed that stream nitrate concentrations (median and mean in mg/L) decrease from Agriculture (3.2, 4.0), to Mixed (1.4, 2.5), to Urban (0.58, 0.94), and to Undeveloped (0.16, 0.26). The deep water concentrations (mg/L) have a sequence similar to stream water concentrations, decreasing from Agriculture (2.9, 3.0) to Urban (1.2, 1.7) to Undeveloped land (0.13, 0.25). These values are close to the national median groundwater concentration (mg/L) of agricultural land (3.1), urban land (1.4), and major aquifer wells (0.56) (Burow et al., 2010). Compared to pristine sites, agriculture and urbanization have elevated nitrate level by 3 to 20 times. Agriculture and Mixed lands also show much higher shallow water concentrations (Csw) compared to deep water concentrations (Cdw), whereas Undeveloped lands exhibit the opposite trend. The Urban sites are the only land type with Cdw> Csw. Stream and soil water concentrations correlate strongly with the percentage of agricultural land with the correlation Cstream = 0.28e0.031 × %Ag (R2 = 0.54) and Csw = 0.35e0.034 × %Ag (R2 = 0.54), respectively. Deep nitrate concentrations do not correlate well with agricultural area fraction (R2 = 0.25), indicating that other factors might be important in determining nitrate in deep water (Figure S3).
Estimated versus measured concentrations in shallow and deep waters. Co-located measurements of stream, soil water, and groundwater concentrations are rare, except in a few intensively measured sites. End-member estimation such as hydrograph separation and chemical mixing analysis are often used to quantify baseflow and quick flow contribution (Raffensperger et al., 2017) and solute concentrations in end-members (Miller et al., 2017). To validate the estimation of Csw and Cdwusing stream concentration at the 95th and 5th percentiles, we compare estimation from this work (Figure 3a and Figure S1) against literature data and the National Ground-Water Monitoring Network (NGWMN) database (see Detailed Methods). Results show that the linear relationship between Cstream and estimated Csw from this work (slope = 0.9, R2 = 0.94) is close to the linear relationship estimated for 94 sites of diverse land uses covering 4 orders of magnitude (Sudduth et al., 2013) (Figure 4a), suggesting a strong correlation between Cswand Cstream. The agricultural sites dominate the top right corner with high Csw and Cstream. Comparison of estimated and measured groundwater concentrations as a proxy for Cdw exhibits a consistent spatial pattern from closely located sites in the NGWMN (Figure S5). Most estimated and measured Cdw falls on the 1:1 line in Figure 4b, indicating a close match (Figure 4b). The relatively large error bars of measured Cdw may be due to varying sampling depths in different wells.
Nitrate export patterns under different land uses. Figure 5 (and Figure S2) showed that although different export patterns (i.e., multiannual scale C-Q relationship) occur in all land use conditions, more agricultural sites have high b values and more urban sites have low b values. In other words, flushing patterns dominate in agriculture and mixed lands. This challenges the existing perception that agricultural lands typically have chemostatic or biogeochemical stationary patterns (Basu et al., 2010; Basu et al., 2011; Thompson et al., 2011). Diamond and Cohen (2018) also found that a greater agricultural land cover was not associated with chemostasis patterns. For urban watersheds, chemostasis and dilution patterns are most commonly observed but flushing also occurs. In pristine sites, both chemostasis and flushing patterns are common.
Concentration contrasts in shallow and deep waters drive C-Q patterns. To explain nitrate export patterns under different land use conditions, we resort to a recently-developed, process-based watershed reactive transport model BioRT-Flux-PIHM (Zhi et al., 2019) to explore the drivers of C-Q patterns. The model was set up first to reproduce hydrological and nitrate data in Conewago Creek, a watershed with ~ 47% agricultural land in the Chesapeake Bay. The model has three major water components contributing to the stream: surface runoff, soil water, and deep water. The shallow water combines surface runoff and shallow soil interflow water, and the deep water is the groundwater that interacts with the stream. Annually the shallow and deep water comprise 80.7% and 19.3% of the total discharge, respectively, although at daily scale the deep water can vary from 1% under dry conditions to 99% during large storm events. The hydrology aspect of the model captured the general trend that shallow water dominates stream discharge at high flow and deep water predominates at low flow. The model reproduced the daily discharge dynamics that was highly responsive to precipitation, as well as daily concentration range, temporal trends, and C-Q patterns (Figure S6). The slope bfrom the C-Q data is 0.13, and the annual concentrations of shallow water (Csw) and deep water (Cdw) were 3.8 and 2.1 mg/L, respectively, yielding a Cratio of 1.8.
To cast the model to broader conditions, 500 Monte-Carlo simulations were run using the hydrology and reaction kinetic conditions from the base case and variable soil N concentrations that yield the representative range of shallow and deep waters in the four land use types in Figure 1. The model reproduced the range of soil water and groundwater concentrations and Cratio in Figure 2 (see Methods and Figure S7 for C-Q relationships in four cases representing four different land uses). In each simulation case, the model outputs of concentrations and discharge were used to calculate b values and the annual average Csw, Cdw, and Cratio (Figure S8). Model results from the 500 cases (overlapping gray circles, Figure 6) collapse to an “S” shaped curve (\(b=\frac{\delta_{b}\ C_{\text{ratio}}}{C_{\text{ratio},\ 1/2}\ +\text{\ C}_{\text{ratio}}\ }+b_{\min}\)) that illustrates the consistent dependence of b values on Cratio, where \(\delta_{b}=1.66\) is the difference between maximum b (\(b_{\max}=0.73\)) and minimum b(\(b_{\min}=-0.93\)) and \(C_{\text{ratio},\ 1/2}=0.80\) is the concentration ratio when b = ½ (\(b_{\max}+b_{\min}\)). This modeled “S” curve explained the b values from 81 sites with available Cratio data. The curve shows that most agricultural and mixed lands have high b values whereas urban and undeveloped lands have lower b values.