Advancing Watershed Legacy Nitrogen Modeling to Improve Global Water Quality

Abstract

Despite widespread implementation of watershed nitrogen reduction programs across the globe, nitrogen levels in many surface waters remain high. Watershed legacy nitrogen storage, i.e., the long-term retention of nitrogen in soils and groundwater, is one of several explanations for this lack of progress. However scientists and water managers are ill-equipped to estimate how legacy nitrogen moderates in-stream nitrogen responses to land conservation practices, largely because modeling tools and associated long-term monitoring approaches to answering these questions remain inadequate. We demonstrate the need for improved watershed models to simulate legacy nitrogen processes and offer modeling solutions to support long-term nitrogen-based sustainable land management across the globe.

Graphical Abstract

Introduction

Decades-long global intensification of land development from population growth, and particularly agricultural expansion for food production, has generated excess nitrogen (N) leaching to soils and groundwater systems. (1,2) Excess inputs of N to landscapes are primarily derived from overapplication of N-based fertilizer and manure and, in some areas of the world, atmospheric deposition from stationary (e.g., industrial) and mobile (e.g., vehicle) emissions. Once integrated into soils and groundwater, a considerable portion of this surplus N is eventually transported to the world’s surface waters. (3,4)

More than 30 years of effort have aimed to reduce the amount of excess soil and groundwater N reaching downstream surface waters, e.g., rivers, lakes, and estuaries, through programs such as the European Nitrates Directive (5) and the Mississippi River/Gulf of Mexico Hypoxia Task Force. (6) Despite widespread implementation of these programs, N levels in many surface waters across the globe remain high. (7,8)

The emergence of the watershed legacy N storage concept in the past several decades is one key explanation for this lack of N reduction progress. (9,10) Legacy N storage, or simply “legacy N”, is the long-term accumulation and storage of N in the landscape resulting from biogeochemical transformations of N and slow hydrological transport, both within and through soil and groundwater systems. Limited information regarding the lags between conservation practice implementation and water quality responses confounds monitoring and research into the efficacy of these efforts. Therefore, confronting the global legacy N issue via improved modeling approaches, alongside incentivizing agricultural non-point source N reductions (11) and optimizing conservation practices under diverse socio-political and economic factors, (12) is a major water quality management challenge.

Legacy N dynamics occur over years to decades, leaving water managers frustrated: these broad and variable time frames often preclude attainment of N-based water quality goals over short, i.e., less than decadal, time scales. (13−17) Specifically, this means that legacy N prevents land conservation practices, such as those related to agriculture, e.g., optimizing fertilizer application timing, implementing conservation tillage practices, and planting cover crops, from producing meaningful water quality improvements. These legacy N challenges have therefore galvanized calls for improved land management approaches, policy actions, and strategic water quality monitoring. (18,19)

Solutions to N-related water quality issues will emerge only if we reduce losses of N from the landscape to deep soils, shallow aquifers, and groundwater, and directly to surface waters. This requires improved methods (models and tools) for understanding how legacy N obscures the interactions among inputs of N in the landscape, conservation practices to reduce losses of soil N to the groundwater, and the extent of N-based water quality responses, i.e., N loads and concentrations in downstream surface waters (Figure 1). Therefore, concerted global efforts integrating legacy N dynamics into watershed-scale methods that project the water quality effects of current and future land management planning methods are critical. (18) The information produced by these efforts will afford the necessary solutions and advances to quantify and communicate to the public realistic expectations of management-related water quality results; however, such approaches are gravely lacking, and water managers and research scientists need a shared framework for how to address this methodological issue.

Figure 1.

Linked land-to-water components involving legacy N processes and management, and examples of each of these linked components.

Watershed models provide this framework and are the primary tools for estimating how legacy N delays downstream water quality changes in response to land conservation practices. (20,21) These models can have different structures, but all use the watershed as a system within which to represent water and nutrient budgets and processes affecting streamflow and water chemistry. However, nearly all existing and widely used watershed models provide a limited representation of reality. Therefore, they lack a full range of dynamic legacy N-related component processes (Figure 1), essentially ignoring legacy N or making broad and erroneous assumptions about its existence. This general lack of legacy N consideration in modeled N and water balances often leads to misguided conservation practice implementation and water quality improvements falling short of their goals. (13,22)

Here, we detail the immediate need for improved watershed models to predict how, and to what extent, legacy N will affect aquatic N responses to land conservation practices. Building on recently developed legacy N models (e.g., refs (13) and (23−25)), we further chart a path forward to address current deficiencies in modeling watershed legacy N dynamics (20) that require urgent attention. We specifically propose ways to strengthen watershed models and modeling practices to improve critical analyses for long-term N-based sustainable land management across the globe. Although legacy N can occur within surface waters, the atmosphere, and the biosphere, we focus on N legacies in the soil and groundwater systems.

Need for Improved Watershed Legacy N Modeling

Mitigation of legacy N reaching surface waters becomes more probable if we can accurately project the extent of legacy N storage and removal rates, in addition to hydrological transport lags, in soils and groundwater. (9) Water managers require these data to understand how legacy N will moderate the water quality responses to conservation efforts and to provide the public with reasonable expectations of these results. However, long-term monitoring of legacy N processes is limited. Therefore, watershed models that incorporate legacy N storage, transformation, and transport processes are essential for managing N in surface waters. While few modeling methods exist to address legacy N issues, two primary approaches have emerged in practice: empirical models and process-based watershed models.

Empirical models are a broad classification of models developed using measured data to derive relationships between sets of variables (see the Supporting Information). They are useful screening tools for (1) locating where legacy N storage sources may exist within a watershed or regional river basin or (2) identifying a range of years within which watershed N inputs may reach downstream surface waters, e.g., from 4–7 years (26) to several decades. (15) We provide an overview of these models for watershed legacy N analyses in the Supporting Information (Key Gaps and Components in Watershed Legacy N Modeling).

Here, however, we specifically focus our discussion on watershed-scale process-based models, herein termed watershed models. These models use the principles of physics, such as conservation of mass and energy, to represent key watershed hydrological and biogeochemical processes via a suite of partial differential equations. While some limitations to process-based models exist (see the Supporting Information), because of their basis in first physical principles, these watershed models may be used to project beyond the temporal scales and extents of established (monitored and measured) data. They can also address legacy N storage and lags at finer resolutions than annual times steps, e.g., days and months. The refined temporal resolution in model outputs helps elucidate the sub-annual legacy N controls on aquatic N responses to management. This is important because in many watersheds, legacy N issues may emerge during particular seasons or storm events. (27−29)

The number of process-based watershed model approaches to N legacy and lags is slowly increasing, (23,30,31) incorporating both groundwater and soil N storage and transport processes. However, their application in estimating legacy N impacts on the response of aquatic N to land management practices has been limited to the past decade (examples in Table S1).

Reimagining Watershed Legacy N Models

The temporal disconnect between landscape-based conservation practices and N-based water quality responses in downstream surface waters is now evidenced across multiple decades of research. This disjointed landscape-to-surface water response results, in large part, from legacy N dynamics. These processes need to be integrated into watershed models projecting water quality responses to N land management practices to sustainably manage the world’s aquatic resources (Figure 2).

Figure 2.

Minimum collective legacy N-related dynamics required to improve watershed models for legacy N simulations. These include additions and modifications in the soil profile, shallow aquifer, deep groundwater, and hydrological transport to the nearest surface water. Some widely used watershed models (e.g., SWAT) already integrate many processes into the soil profile (e.g., N accumulation and soil N mineralization) and shallow aquifer (see the asterisk). These processes, however, need refinements and more testing to ensure they are accurately represented. However, other shallow aquifer and all deep groundwater legacy N dynamics are typically absent in most watershed models. While hydrological flowpaths are also present, the potential for decadal lags in horizontal (to the stream) and vertical (through the soil and groundwater system) flows is also atypical in current watershed models and needs to be incorporated.

Process-based watershed modeling approaches are promising for projecting how legacy N mediates land conservation practice effects, as well as climate change interactions with them, on downstream N levels. In fact, in the past several years, calls for the incorporation of N legacy storage and time lags into watershed modeling approaches have become more pronounced. (18,19) Watershed models provide foundational model structures that can theoretically be modified to project time-stepped physically based links among land management practices, legacy N storage and transport, and changes in N-based water quality. However, most existing process-based watershed models continue to have structural constraints (20) that lead to inaccuracies in model outputs that are unaccounted for (see the Supporting Information). Moreover, a clear path charting the course for modeling legacy N’s role in delaying conservation practices’ water quality improvements is lacking.

While solutions to the legacy N issue, such as recovery and reuse of legacy N, are evolving, (18,19) we specifically focus on key advances needed in watershed legacy N modeling. These developments will help improve the understanding and management of linkages among conservation practice timing, magnitude, and location to downstream water quality changes. Because process-based watershed models are the tool for projecting conservation practice impacts on downstream waters, these advancements are critical. We call for the following improvements to watershed-scale legacy N models (Figure 2):

Improved Legacy N-Related Process Representation

Few watershed models incorporate all necessary watershed legacy N dynamics, including legacy N soil and groundwater storage and hydrological transport lags, into future land management projections. This is often due to a combination of factors, such as epistemic uncertainty in system processes, a focus on system-representative processes specific to the location where the model was developed or specific to the hypothesis (or management question) that it was developed to test, and decreased computational efficiencies when models are overparameterized. Therefore, modifications to future modeling efforts should at minimum focus on improving soil N storage and transformations based on dynamic seasonal N inputs, soil N leaching to the groundwater system, shallow and deep aquifer storage and N removal processes, and hydrological transport lags, including travel times of the legacy source signal (Figure 2). However, it is important to balance model fidelity (the degree to which a model faithfully represents processes) and resource expenditures for such legacy N improvements. (32)

Biogeochemical Processes

For specifically addressing biogeochemical processes influencing legacy N, many watershed models include accumulation of N in soil layers. However, it is important to refine or make these additions when not present, as well as including time-stepped sub-annual N inputs, for capturing legacy N soil storage in watershed models (Figure 2). These models should also represent soil organic nitrogen (SON) mineralization, a process that may contribute to legacy N, in addition to excess fertilizer inputs. While SON is integrated into some watershed models used for legacy N questions (e.g., ELEMeNT (13,31)), improvements in processes related to soil N accumulation (e.g., carbon–nitrogen soil interactions (23)) are needed. In addition, a groundwater denitrification model component, or module, could simulate losses of N that may otherwise be retained in the mass balance.

Hydrological Processes

Soil water transport functions that move water laterally (e.g., shallow subsurface flow) and vertically (leaching) to the groundwater system based on physical hydrological and hydraulic processes should be continually improved in process-based watershed models simulating the effects of potential land conservation practices. In this way, we can better estimate the partitioning and tracking of legacy N stores and their transport functions following conservation practice implementation. Watershed models also need better-formulated model structures and numerics to represent (1) propagation of N sources on the landscape to legacy N in surface waters and (2) appropriate lags in N leaching from soils to the shallow aquifer and to the deep groundwater. This would improve estimates linking hydrological N transport and N travel times downstream. Promising dynamic approaches have been developed in recent years (e.g., refs (23), (31), and (33)) yet need to evolve and expand. In addition, appropriate groundwater residence times (e.g., more than 5 years) and transport to the nearest surface water and storage of nitrate (and other forms of N) (Figure 2) are needed in watershed models used for long-term legacy N storage estimates. Linking these processes to those within riparian and hyporheic zones, which can also attenuate N for long periods, (22) would optimize results.

Finally, to best estimate the effects of conservation practices and N mitigation measures in a changing climate, water balances and transport processes need to be improved alongside the watershed N budget. Often, watershed model improvements are heavily weighted toward biogeochemistry or hydrology, but not both. However, simultaneous evolution of both watershed model components is important because legacy N emerges from both biogeochemical retention in soils and groundwater as well as slow subsurface water transport. For example, in areas with increased frequencies and extents of drought, high-magnitude flushing of nitrates from soils (34) may be more likely to occur during storm events, which are becoming more frequent in some areas across the globe. These intra-annual variations could mask the effects of conservation practices during drier, lower-flow periods throughout a year, and annual estimates of legacy N may not reflect these intra-annual variabilities.

Artificial Drainage Integration

The presence of tile drains in agricultural landscapes accelerates flows and minimizes legacy N conditions. (22,35) Therefore, to correctly source and manage excess N on the landscape, properly accounting for artificial drainage in watershed legacy N models is imperative. Tiles and drainage ditches can all affect the timing of water quality responses to land conservation measures. As a result, integrating and expanding artificial drainage into watershed models for legacy N simulations (at model-appropriate spatial resolutions), as well as improving monitoring of water table depths, is imperative for developing adequate model outputs.

Uncertainty Estimates

Simulating legacy N remains a relatively new science; therefore, it is critical to integrate uncertainty when presenting modeled outputs. (36) Model inputs, parameter values, model structure, and simulated outputs are all potential contributors to model uncertainty. At minimum, process-based watershed models integrating legacy N processes should demonstrate the breadth of potential model output uncertainty, i.e., uncertainty around the simulated model output that reflects all forms of model uncertainty. In this way, it is feasible to broadly evaluate how integrating legacy N stores and fluxes improves model projections of future land management impacts on water quality, and to verify the range of potential outcomes based on the modeling effort. (16,17) Ultimately, however, also addressing model input, parameter, and structural uncertainties is the end goal.

Integrating Readily Accessible Data Sets

Legacy N accumulation occurs over long periods, often decades, in many watersheds, rendering limited data for model verification. This means that most necessary process improvements in Figure 2, even when included in a model, are traditionally not verified by additional data. For example, many watershed models currently simulate N accumulation in soils; however, this process is not calibrated or verified to measured data, often because these data are not available. Making use of available satellite-based spatial data sets would be advantageous, even if the data measured do not match the model parameter units (sometimes called “soft data” verification). One example is integrating remotely sensed soil moisture data into a watershed legacy N model to verify dynamic soil water storage model simulations, (37) which are critical to modeling soil N processes affecting legacy N storage. Similarly, groundwater processes currently absent in watershed models used for legacy N questions may be developed and verified by integrating surficial aquifer data or publicly available groundwater well nitrate observations.

Dynamic Model Coupling

Because long-term data for verifying legacy N models are limited, dynamic coupling of watershed models, i.e., linking individual models together, (23,30) will ultimately improve our scientific understanding of legacy N dynamics. Innovations in coupling empirical and process-based approaches mark significant progress in this area, such as linking N source maps with groundwater travel times to detect groundwater delivery of N to surface waters. (38) Another viable method is coupling a watershed model to a groundwater transport model, e.g., via MODFLOW, (39) which simulates groundwater flow through aquifers, or lagged watershed travel time distributions. (23)

In this way, we can move beyond model structural limitations regarding (1) the length of time N can be stored in the soils or groundwater (e.g., groundwater N residence times in SWAT are typically capped at 500 days) and (2) the lack of legacy source signal tracking and propagation through the soils and groundwater. However, model coupling is often data intensive, and when two process-based models are linked, model complexity can increase considerably. Therefore, model coupling should be considered where data are available to appropriately parametrize the models and where financial and human resources are available to support model computational needs.

In addition, although potentially computationally expensive, ensemble modeling, i.e., using multiple watershed models or multiple model representations within a particular watershed model structure, could help develop multiple lines of evidence for optimizing water quality goals affected by N legacies. Ensemble modeling affords an “envelope” of output uncertainties around N loads and concentrations in response to dynamic legacy N processes interacting with climate change and conservation practices.

Practices for Improving Legacy N Modeling

Outside of model structural advancements, improving watershed legacy N modeling practices will ensure more advantageous conservation practice selection for sustainable management of aquatic systems. The concepts of watershed modeling communities of practice (an engaged group of modelers and practitioners focused on model improvements) and participatory modeling (the process of involving citizens in model development) (40,41) are becoming widely accepted. In addition to advancements in these areas for watershed legacy N modeling, we suggest two key approaches that need continued refinement: improved monitoring of legacy N processes and appropriate model selection for projecting how legacy N mediates the downstream water quality impacts of conservation practices.

Improved Monitoring

There is a lack of diverse long-term monitoring to better understand current legacy N issues and to improve models for projecting legacy N effects on downstream water quality. Improved long-term monitoring of N and water levels in soils, shallow aquifers, groundwater, and associated stream networks in response to land management practices is therefore critical for understanding hydrological and biogeochemical N lags associated with legacy N. This may be realized by before-and-after studies, e.g., measuring soil, groundwater, and downgradient stream nitrate concentrations prior to conservation practice implementation and then monitoring these levels for decades after implementation. In addition, integrating this information into the next generation of watershed models would improve simulation of interacting conservation practices, climate change, and legacy N dynamics. (19,20,36)

Appropriate Model Selection

Model selection for legacy N questions is often mismatched for the size of the watershed or the temporal scale of research or management questions. Water managers and modelers need to consider the size of the watershed (and associated water and N retention capacities) in concert with the management goal or research question to select the most appropriate legacy N watershed model.

Not all legacy N research and management questions, however, require highly complex models. For example, small watersheds (e.g., <15 km2) with limited soil and groundwater storage capacities respond more rapidly to landscape-scale changes, such as conservation practices, than those with greater N and water storage volumes. (35,42) In these cases, highly complex process-based watershed models may not be needed. Monitored data and an empirical model, such as a net anthropogenic nitrogen inputs-based model, a simple mass balance approach for estimating human-generated N inputs to watersheds, with the addition of a lagged N delay component, (43,44) could be appropriate.

However, complex process-based models are most often required because of the spatial scale of regional and national N management efforts, along with the need to understand processes driving legacy N dynamics. Controlling surplus N in large river basins involves implementing conservation practices across broad spatial extents, e.g., ranging from the ∼11 600 km2 Chesapeake Bay Watershed (United States) (45) to the ∼2.13 million km2 Baltic Sea drainage area. (46) In these systems, N stores and fluxes vis-à-vis legacy N dynamics are important, (17,47) as is the capacity to project more realistic water quality targets based on inclusion of legacy N-related processes in the model.

Moving Forward

Legacy N dynamics are one of the biggest determinants of decadal lags in downstream water quality responses to land conservation practice implementation. Currently, a consistent set of watershed modeling practices for capturing these dynamics is lacking across the globe. However, this methodological framework is needed to realistically set goals for land conservation practices to improve N-based water quality. A primary path toward sustainably managing N in global waters is therefore to build capacity in watershed models to project how legacy N dynamics interact with land management practices and modify their efficacy for years to come. We suggest these advances include modifying watershed models used for legacy by integrating (1) the primary biogeochemical and hydrological dynamics that are important for legacy N, (2) artificial drainage to account for the rapid transport of N to streams and rivers, (3) readily available spatial data sets for improved legacy N model verification, (4) uncertainty estimates to bound legacy N projections, and (5) dynamic coupling of watershed and groundwater models related to legacy N processes. We also call for improved long-term monitoring networks for legacy N stores and fluxes and practical, measured model selection processes.

Synopsis.

Legacy nitrogen watershed modeling approaches need to be improved to sustainably manage global water quality.