DISCUSSION
The significance of N distribution and processes in the shallow and deep subsurface. Figure 6 suggests that the subsurface N concentration contrast (or Cratio) arising from different land uses predominantly control export patterns. The representatives of shallow and deep waters, such as soil and groundwater, have been shown before to correlate well with stream nitrate concentration across different land uses (Sudduth et al., 2013). The export pattern dependence on concentration contrast echoes the decades-long End Member Mixing Approach (EMMAS) that uses end-member concentrations to infer stream water chemistry (Hooper et al., 1990). The spatial data synthesis across the U.S. continent from this work enables the generalization of this idea across diverse climate, geology, and land use conditions. The physics-based watershed reactive transport modeling facilitates the derivation of the general equation that can be used to estimate export patterns based on streamflow concentrations or measured shallow and deep water concentrations.
These shallow versus deep physical contrasts originate from chemical and physical weathering in pristine sites (Brantley et al., 2017). In urban and agriculture sites, these contrasts arise from human engineering efforts with tile drains, impervious surfaces, and water pipes that modify environments (Grimm et al., 2008). The concentration differences in shallow and deep waters arise not only from different subsurface distribution of N source but also from different processes. In agricultural lands, nitrate concentrations in shallow waters are high not only because abundant N sources leach nitrate; it is also because denitrification is limited with the presence of abundant O2. In deeper groundwater, with limited O2, denitrification can occur and reduce nitrate (Kolbe et al., 2019).
Under broad conditions, the relative magnitude of shallow versus deep water concentrations may hinge on soil properties and geologic structure. These subsurface structures determine shallow and deep connectivity, recharge and water table depth (Brantley et al., 2017), as well as local biogeochemical conditions (e.g., anoxic condition, organic carbon availability) that control the extent of denitrification and nitrate removal (Kolbe et al., 2019; Tesoriero et al., 2015). Miller et al. (2017) showed that nitrate export exhibits a dilution pattern in Tomorrow River, a site with permeable sand and gravel, but a flushing pattern in Duck Creek with low-permeability clayey soil clayey soils developed from glacial tills. The overwhelming convergence toward highb values in agriculture lands, however, indicates that the concentration contrasts in shallow versus deep waters are the predominant control of export patterns.
In urban watersheds, we observe varied export patterns and a large number of sites have higher concentrations in deep water compared to shallow water. This is possibly caused by underground leaky sewage and pipes that can contaminate groundwater and surface water (Divers et al., 2013; Lerner et al., 1999; Pennino et al., 2016). A large number of urban watersheds also exhibit chemostatic patterns, potentially due to the co-occurrence of both shallow N sources (e.g., lawn fertilizer, pet waste, atmospheric deposition, automobile emission) and deeper underground input from buried sewage and septic systems. Kaushal et al. (2014) and Newcomer Johnson et al. (2014) suggest that urban N export can be influenced by the degree of hydrologic connectivity associated with impervious surface, stormwater infrastructure and sewage pipe, and stream restoration. Point source discharge from wastewater treatment plant (WWTP) can also increase nutrient loading (Luthy et al., 2015), potentially contributing to varied export patterns. In arid and semiarid regions with small urban streams, high nitrate concentration from WWTP can dominate the base flow at the dry time and become diluted under high flow conditions, resulting in dilution pattern (Marti et al., 2010). In fact, urban watersheds are complicated as the groundwater-soil-surface water interactions are modified by the level of urbanization and management, and non-point sources from leaky infrastructure and chronic groundwater contamination (Kaushal & Belt, 2012).
Limitations, simplifications, and uncertainties . The conceptual model in Figure 1 and the general b equation in Figure 6 emphasize two end members in shallow and deep zones and are meant to build a simple relationship between export patterns and major components of water contributing to the stream. Here we lump the waters into shallow and deep waters components to illustrate the first-order control. Such simplification is often necessary in practice. In urban watersheds, for example, the impervious surface often contributes to large surface runoff in storms. Surface runoff however typically dominates as short pulses in early stages of storm events often resulting in an overall limited contribution of surface runoff to annual discharge (e.g., 11% by Pellerin et al. (2008). These temporarily large contributions are followed by rapid subsurface flow (via underground stormwater pipes) with much longer duration. Field studies typically have incomplete information (e.g., hydrograph separation, isotopic signature) of contributing flow paths. It is therefore often necessary to lump different water sources into major compartments (Barnes & Raymond, 2010), in order to capture average behaviors.
With its simplicity, the model does not take into account specifics of individual sites, which can lead to deviations from the b curve in Figure 6. For example, in some places, N is distributed and processed in a way that demands more than two end members (Cowie et al., 2017; Miller et al., 2017). The model also does not explicitly account for N removal in streams. Instream removal depends on a wide variety of parameters including local climate and seasonality, landscape structure (e.g., topography, hyporheic zone), and biogeochemical conditions (e.g., nitrate legacy, stream carbon source) (Dodds et al., 2002; Hill, 1996; Mulholland et al., 2008; Vidon & Hill, 2004). Significant instream N removal can lead to underestimation of Csw and Cdw but to a different extent. As Csw is estimated under high flow conditions where instream N removal is often not efficient, it may not be heavily influenced by instream processes (Dodds et al., 2002). The estimation of deep water concentrations however is based on low flow conditions where instream N removal can be highly effective such that Cdw may be underestimated. This possible different extent of underestimation can shift Cratio (= Csw / Cdw) and b values to higher values, moving data points toward the top right end of the d b curve. This may be the reason that some data points from Ag and Mixed fall in the right-hand side of the b curve in Figure 6.
Inferring shallow and deep water chemistry from stream chemistry. This study indicates that we can infer shallow and deep water chemistry from streamflow chemistry under different flow regimes. This is important as detailed subsurface characterization and concentration measurements are often limited to only a few long-term monitoring sites in developed countries (Brantley et al., 2018; Gran et al., 2019), although we often claim an era of “big data”. As shown in Figure 6, few intensively measured watersheds have both soil water and groundwater chemistry measurements. The extrapolation of shallow and deep water chemistry from high flow and low flow stream chemistry therefore enables estimation of water chemistry (without digging holes in the ground) to infer possible export patterns and loads. This approach can be applied not only for nitrate, but also for other solutes in general.
N export under changing climate. The agriculture sites have increased nitrate concentrations in shallow and deep waters by 15-20 times and 9-16 times, respectively, compared to undeveloped sites (Figure 3a). In fact, extensive tile drainage networks in agricultural lands (e.g., 80% of the landscape in the Midwest) shortcut the shallow water directly to stream, lowers water table (Blann et al., 2009; King et al., 2015). In effect, these draining tiles decrease vertical connectivity to deeper aquifers (Association, 2018) and therefore increase concentration contrasts between shallow water (e.g., soil water and tile drainage water) and deep groundwater, inducing pronounced flushing patterns (Figure 6). The flushing pattern indicates that export is sensitive to large hydrological events such as flooding, which has been predicted to intensify as the pace of climate change accelerates (Prein et al., 2017).
On the other hand, if high nitrate in shallow water is redirected more to the deeper subsurface via higher vertical connectivity, the water will reroute through longer flow paths, enhancing transformation via denitrification and nitrate removal (Kolbe et al., 2019). The extent of such transformation will depend on local conditions. The longer and slower groundwater flow paths however will also lead to decade-to century-scale time lags between anthropogenic N inputs and riverine outputs (Sebilo et al., 2013; Van Meter & Basu, 2015). This can present significant challenges for balancing nitrogen budget (e.g., the ‘missing’ N in mass-balance) (Boyer et al., 2002) and management effectiveness (e.g., land-use management, N-loading reduction) (Sebilo et al., 2013; Van Meter et al., 2017). In addition, this downward N flux also raises tantalizing questions about long-term subsurface structure and functioning. How and how much do these man-made, high nutrient levels elevate carbon effluxes (Zamanian et al., 2018), accelerate weathering processes, and alter watershed functioning (Kaushal et al., 2013; Perrin et al., 2008)? These broader earth system responses can have far-reaching impacts on carbon and nutrient cycles at the global scale.