Simulating internal watershed processes using multiple SWAT models
Graphical abstract
Introduction
Shifts in land use and management have transformed many landscapes, altering water resources and raising environmental concerns among managers and other stakeholders (Guswa et al., 2014). Understanding and predicting the impacts of these changes is key to developing effective mitigation strategies that balance development and environmental goals. Policy- and decision-making often rely on computer model simulations, raising concerns around appropriate model use and prediction confidence. Significant progress in model error analysis and uncertainty estimation has been made; however, models considered to be accurate can have inaccurate representations of dominant processes in a system (Hrachowitz and Clark, 2017). Models varying in structure and parameterization can effectively reproduce similar system behavior—the equifinality concept—allowing for multiple possible internal representations of a system (Beven, 2006). Capturing this process-level information results in more robust hydrologic and land management models and is key for effective decision-making.
Distributed parameter models have become a common tool for hydrologic analysis and watershed management planning. With greater access to spatially variable and fine resolution datasets, the representation of natural systems within these models can be vastly improved (Vieth et al., 2008; Daggupati et al., 2015). The Soil and Water Assessment Tool (SWAT) is a highly parameterized process-based model developed for watershed scale assessment of climate, land management, and land-use change (Neitsch et al., 2011). While originally intended for use on ungauged basins, modelers generally improve SWAT model performance through calibration and validation to streamflow, and potentially water quality, at a single downstream outlet location. To the extent that models are only assessed at the watershed outlet, they can fail to take into account key intra-watershed processes—internal processes which govern global responses—and can thereby lead to successful models that lack realistic system representation (Yen et al., 2014; Daggupati et al., 2015). Yet capturing these processes is important as most management implementation and improvement strategies in agricultural regions occur at the field scale (Diebel et al., 2008; Muenich et al., 2017).
Data capable of validating these internal processes at the necessary scales are rarely available, especially in large watersheds. This use of smaller scale data is important as studies have shown that, while sometimes difficult to incorporate and calibrate to, the use of direct field-scale data can significantly impact field-scale model performance (Merriman et al., 2018; Muenich et al., 2017; Kalcic et al., 2015; Daggupati et al., 2015). Two important sources of field-scale data that have become more available and have been used in upstream validation are remote-sensed data and edge of field (EOF) monitoring. The use of remotely sensed data, capable of effectively describing the internal distribution of key watershed characteristics, such as soil moisture (Rajib et al., 2016) and vegetation (Ma et al., 2019; Rajib et al., 2020), has greatly improved SWAT calibration and performance. EOF monitoring provides valuable site-specific information on management practices, as well as key nutrient transport budgets (Pease et al., 2017; Hanrahan et al., 2019), which can inform soft validation for upland performance. As their availability grows, these data present an opportunity to add to the understanding and refinement of upstream representation and prediction accuracy of hydrologic models.
A lack of data availability, compounded with model framework uncertainties and computational constraints, may prohibit thorough uncertainty characterization of deterministic models such as SWAT. The use of multiple models that vary in model inputs, parameterization, and structure can capture some of the variability that is lost using a single deterministic model, and therefore play a role in uncertainty analysis (Hrachowitz and Clark, 2017). Such ensemble modeling has been used in a variety of SWAT modeling efforts to evaluate watershed water quality and policy-relevant management (Evenson et al., 2020; Martin et al., 2019; Scavia et al., 2017). Model ensembles can help capture uncertainties in model structure, parameterization, input data, and assumptions, thus enabling the exploration of the impacts of these facets on model performance.
The aim of this study was to assess the extent to which watershed hydrologic models, calibrated and validated at the watershed outlet, can capture field-scale dynamics upstream. This was done by developing the latest version of the Maumee River Watershed (MRW) SWAT model, Apostel, and comparing it against EOF monitoring data and two previous MRW SWAT model iterations, Kalcic et al. (2016) and Kujawa et al. (2020). While the Kalcic and Kujawa models varied with respect to management inputs, structure, and parameterization, both achieved acceptable Moriasi et al., 2007, Moriasi et al., 2015 performance metrics at the watershed outlet (Kalcic et al., 2016; Kujawa et al., 2020). Our objective was to assess the ability of the model ensemble to capture upstream field level discharges and nutrient loadings. The results will be discussed in the context of critical differences between models in terms of structure and inputs.
Section snippets
Study area
The Maumee River Watershed (MRW) is the largest watershed draining to Lake Erie. Excess inputs of nutrients from the MRW, especially phosphorus, have increased harmful algal blooms in the Lake's western basin and hypoxia (low oxygen concentrations) in its central basin that threaten both ecosystems and human health (Scavia et al., 2014, Scavia et al., 2016; Michalak et al., 2013; Ohio EPA, 2010, Ohio EPA, 2013). The MRW's impact on the health of regional resources has made it a key target for
Calibration and validation
Final calibration of the Apostel model involved adjustments in 40 parameters (Table S26) and resulted in ‘satisfactory’ or better performance standards (Table 2; Moriasi et al., 2007, Moriasi et al., 2015). Validation for 2000–2004 also achieved ‘satisfactory’ model performance, although sediment PBIAS fell just outside the target range. Primary parameterization differences among the Kalcic, Kujawa, and Apostel models were parameters for tile drainage, soil nutrient initialization and
Newly developed field-scale MRW SWAT model (Apostel)
The Apostel MRW SWAT model provides a platform for incorporating field-specific management-level decisions. While direct field-scale HRU data were not available for this watershed, a more realistic representation of management practices and schedules was implemented near the farm field scale, which is not typical in SWAT models (Karki et al., 2020). The Apostel model performance at the outlet was similar to that of previous iterations of the model, though model run time tripled due to greater
Conclusions
Ensemble modeling has been used to gauge uncertainty and reduce potential biases introduced through decisions made during model development. We used three models, calibrated and validated at the watershed outlet, to assess performance at the farm field scale for capturing field level hydrology. This was done without incorporating true field-scale management data, so that we could test whether baseline model performance was representative across tile-drained agricultural fields. The ensemble
CRediT authorship contribution statement
Anna Apostel: Writing – Original Draft, Methodology, Conceptualization, Data Curation, Formal Analysis, Visualization, Software, Validation
Margaret Kalcic: Funding Acquisition, Project Administration, Conceptualization, Methodology, Writing – Review & Editing, Supervision, Software, Validation
Awoke Dagnew: Methodology, Software
Grey Evenson: Methodology, Software, Writing – Review and Editing
Jeffrey Kast: Methodology, Writing – Review and Editing
Kevin King: Resources, Data Curation, Writing –
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This project was funded by the National Science Foundation Coastal SEES grant OCE-1600012, with additional support from a Harmful Algal Bloom Research Initiative grant from the Ohio Department of Higher Education, and the U.S. Department of Agriculture, Agricultural Research Service Soil Drainage Unit.
References (73)
- et al.
Predicting dissolved reactive phosphorus in tile-drained catchements using modified SWAT model
Ecohydrol. Hydrobiol.
(2019) A manifesto for the equifinality thesis
J. Hydrol.
(2006)- et al.
Using the soil and water assessment tool (SWAT) to model ecosystem services: a systematic review
J. Hydrol.
(2016) - et al.
Subsurface tile drained area detection using GIS and remote sensing in an agricultural watershed
Ecol. Eng.
(2017) - et al.
Manure management at Ohio confined animal feeding facilities in the Maumee River Watershed
J. Great Lakes Res.
(2019) - et al.
Earthworm populations and species dsitributions under no-till and conventional tillage in Indiana and Illinois
Soil Biol. Biochem.
(1997) - et al.
The hydrologic model as a source of nutrient loading uncertainty in a future climate
Sci. Total Environ.
(2020) - et al.
Extended SWAT model for dissolved reactive phosphorus transport in tile-drained fields and catchments
Agric. Water Manag.
(2016) - et al.
Enhancing SWAT with remotely sensed LAI for improved modelling of ecohydrolgical process in subtropics
J. Hydrol.
(2019) - et al.
Total and soluble reactive phosphorus loadings to Lake Erie a detailed accounting by year, basin, country, and tributary
J. Great Lakes Res.
(2016)
A new soil-temperature module for SWAT application in regions with seasonal snow cover
J. Hydrol.
Multi-objective calibration of a hydrological model using spatially distributed remotely sensed/in-situ soil moisture
J. Hydrol.
Assessing and addressing the re-eutrophication of Lake Erie: central basin hypoxia
J. Great Lakes Res.
A multi-model approach to evaluating target phosphorus loads for Lake Erie
J. Great Lakes Res.
Representation of agricultural conservation practices with SWAT
Hydrol. Process.
Estimation of soil cracking and the effect on surface runoff in a Texas Blackland Prairie watershed
Hydrol. Process.
Input/output documentation, version 2012
Texas Water Resources Institute
Vertical stratification of soil phosphorus as a concern for dissolved phosphorus runoff in the Lake Erie Basin
J. Environ. Qual.
Assessment and synthesis of 50 years of published drainage phosphorus losses
J. Environ. Qual.
Sensitivity and identifiability of stream flow generation parameters of the SWAT model
Hydrol. Process.
Impact of model development, calibration and validation decisions on hydrological simulations in West Lake Erie Basin
Hydrol. Process.
Landscape planning for agricultural nonpoint source pollution reduction I: a geographical allocation framework
Environ. Manag.
Linking soil phosphorus to dissolved phosphorus losses in the Midwest
Agricultural and Environmental Letters.
Uncertainty in critical sources area predictions from watershed-scale hydrologic models
J. Environ. Manag.
Comparison of performance of tile drainage routines in SWAT 2009 and 2012 in an extensively tile-drained watershed in the Midwest
Hydrol. Earth Syst. Sci.
Hydrologic modeling to support decision making
Water Resour. Res.
Nutrient balanaces influence hydrologic losses of nitrogen and phosphorus across agricultural fields in northwestern Ohio
Nutr. Cycl. Agroecosyst.
Impact of the numbers of observations and calibration parameter on equifinality, model performance, and output and parameter uncertainty
Hydrol. Process.
HESS opinions: the complementary merits of competing modelling philosophies in hydrology
Hydrol. Earth Syst. Sci.
The role of satellite-based remote sensing in improving simulated streamflow: a review
Water.
Defining soil and water assessment tool (SWAT) hydrologic response units (HRUs) by field boundaries
International Journal of Agricultural and Biological Engineering.
Cited by (26)
Evaluating the longevity of in-stream phosphorus legacies: A downstream cascade of recovery following point source remediation
2024, Science of the Total EnvironmentAn approach to select optimum inputs for hydrological modeling to improve simulation accuracy in data-scarce regions
2023, Journal of Hydrology: Regional StudiesSource appointment at large-scale and ungauged catchment using physically-based model and dynamic export coefficient
2023, Journal of Environmental ManagementAn integrated source apportionment method by incorporating spatial location information and source-transfer-sink simulation
2022, Journal of Cleaner ProductionCitation Excerpt :Krisnayanti et al. (2021) used an approach empirical to estimate runoff from the relationship between rainfall, land use, and soil hydrology groups. Physics-based models can simulate the more detailed processes associated with pollutants than an approach empirical, which can help identify the temporal-spatial transport of pollutants (Apostel et al., 2020). However, the understanding of the impacts of point sources (PS) and non-point sources (NPS) is hindered by the generally poor representation of spatial locations in the existing physics-based models, which resulted in these calculations being typically not available at high spatial resolutions.
The spatial organization of CAFOs and its relationship to water quality in the United States
2022, Journal of HydrologyCitation Excerpt :To account for the nutrient contribution from manure produced by these animal operations in watershed models, modelers often rely on average manure application rates to represent manure management practices in the studied watershed area, neglecting the spatial dynamics of animal agriculture and changes in management practices other than manure application and crop patterns (Wang and Baerenklau, 2015). When solely relying on application rates, models have not shown much improvement when modeling observed TP and TN loads (Apostel et al., 2021; Kast et al., 2021). More information on the location of these animal facilities, their distance to water bodies, and the number of animals they hold may improve the representation of these point and non-point sources of nutrient pollution in models.