The Soil Water Assessment Tool (SWAT) watershed model is commonly used to simulate the impact of land use and land management changes on water quantity and water quality. It was developed by researchers within the U.S. Department of Agriculture, Agricultural Research Service (USDA, ARS) in the mid-1990s, and the model has undergone continual review and expansion since it was first developed. As a result, the model is extremely well-documented in a detailed user manual and contains over 1000 peer-reviewed journal articles that describe applications and enhancements.

SWAT is a physical model that uses mathematical equations to represent watershed processes such as hydrology, soil erosion, crop growth, and nutrient cycling on the land and in the stream network on a daily time scale. The model is spatially-referenced to a specific watershed or sub-watershed. Within the model, the smallest spatial units are the hydrologic response units (HRUs) which generally have uniform soil type, land use, and slopes.

SWAT model inputs include:

• Soil types – Soil type is a key determinant of how well water infiltrates into the soil. Poorly drained clay soils like those common in the Maumee watershed have low infiltration rates and thus high runoff potential. Many of these soils were once part of a large wetland system. Subsurface drains have been installed on these fields to lower the water table and make them productive for agriculture.
• Elevation data – These data are used in model set up to delineate watersheds, identify flow paths, and define slopes, which are another key determinant of runoff potential.
• Land use data – Land use impacts the manner in which water moves through the landscape, and different land uses (forested, agricultural, urban) produce different amounts of runoff.
• Climate data – Daily temperatures and the amounts and timing of precipitation are key drivers of many watershed processes, including crop growth and runoff.
• Land management actions – Models include information about crop rotations, tillage and fertilizer application methods, and locations of practices like wetlands and riparian buffers.

Agricultural management practices that can be modeled include:

• Crop rotations – such as corn, soybeans, and wheat
• Cover crops – such as cereal rye or tillage radish
• Tillage – timing and intensity of tillage operations, and no-tillage
• Fertilizer application – timing, rate, nutrient content and placement of fertilizers in/on the soil
• Grazing operations
• Filter strips – placement and effectiveness at treating overland flow
• Irrigation
• Subsurface (“tile”) drainage – placement, depth, efficiency
• Wetlands – placement, capacity, and treatment efficiency

As with other model components, each practice has associated mathematical equations that describe how it functions and performs, based on edge of field studies, and improvements are incorporated into the model over time.

Unfortunately, data on the specific locations of conservation practices, implemented for example through U.S. Farm Bill programs, are not released publically due to privacy laws. Therefore, individual modelers are required to make informed decisions about how to incorporate various land management practices into their models, and they often use data and information from a wide variety of sources including the National Agriculture Statistics Survey, other published data, and interviews with farmers to identify the most reasonable locations.

The SWAT model reports crop yields and the amount of water (flow), nutrients, and sediment being delivered to waterways at the field and sub-watershed scales, and at the outlet of the modeled watershed. For the Maumee multiple model project, the outlet was located at Waterville, OH, just east of the City of Toledo because extensive monitoring data for model calibration and confirmation are collected there.

The modeling process includes three main phases, with model set-up and calibration being the most time intensive.

• Model set-up – Input data on soils, elevation, land uses, climate, and land management practices.
• Calibration – Compare model output (e.g., the flow) to observed data. This step is iterative. A modeler will make informed adjustments to model parameters to improve the model fit to observations based on goodness of fit statistics.
• Confirmation or verification – Once the model is calibrated, the model is run with data from another time period or location, testing for statistical agreement between the model output and observations.

Upon completing the three phases, the resultant model is assumed to be a reasonable representation of what is happening in the watershed and can be used to evaluate a range of land management options, such as the impact of improved nutrient management on nutrient loads.

It is unlikely that any two models for the same watershed would be identical, even if they are both reasonable representations of the observations because individual modelers make critical and often different decisions about spatial discretization, input data sources, sub-equations to use, land management actions, model parameterization, and calibration approaches.

We used multiple models and multiple modelers to raise confidence in our results. Models incorporate numerous assumptions and decisions in the modeling process, and using multiple models increases our ability to capture a range of potential outcomes, while also smoothing out extremes that might result from assumptions of an individual modeler. Five of the six models used USDA’s Soil and Water Assessment Tool (SWAT) - a state-of-the-art watershed model commonly used to test how land management actions influence water quality outcomes. Even though they used the same base model, each group’s model was different because of the critical different decisions they make about spatial discretization, input data sources, subroutines to use, land management operations, model parameterization, and calibration approaches. Each of these decisions contributes to real differences between and among SWAT models. The sixth model was a USGS SPAtially Referenced Regressions on Watershed attributes (SPARROW) model.

There are analogous efforts in climate and weather forecasting. Multiple models exist at different spatial resolution, but often use the same underlying equations. It is common practice in climate modeling to consider a multi-model mean, and the Intergovernmental Panel on Climate Change (IPCC) uses multiple climate models to forecast climate.

There may be a temptation to select one model based on “superior performance.” However, there are many ways to evaluate performance including graphical and statistical methods, as well as other measures of performance such as ensuring field-level nutrient export, soil nutrient content, and crop yields are within observed ranges. The true accuracy of the models in representing the baseline condition is not uniquely quantifiable and each model represents a reasonable representation of the real world.

In one of the promising agricultural conservation scenarios, a series of practices including subsurface application of P fertilizer and use of winter cover crops on 1.5 million acres of cropland, as well as the conversion of 30,000 acres of cropland to buffer strips was able to achieve both the total phosphorus and dissolved reactive phosphorus target loads for Lake Erie on average.

AP Correction: AP corrects cropland conversion error in news article

The three scenarios that converted agricultural land to grassland were run to demonstrate how much land would need to be taken out of production to achieve the phosphorus load targets if there were no changes to current agricultural practices. In other words, if you consider the time period (2005-2014), total phosphorus and dissolved reactive phosphorus targets were achieved just 3 of the ten years. Reaching these targets without implementing additional conservation actions would require taking 25-50% of land out of production. These scenarios were run to demonstrate the need for enhanced conservation.

The land conversion scenarios took considerable land out of production, are rather extreme scenarios, and are unlikely to be implemented. These scenarios were included to illustrate how much land would have to be removed from production to achieve the target loads if no additional nutrient management and in-field or edge-of-field (buffer strips) practices were employed. For all other scenarios simulated by the models, the impact on total crop production in the watershed was minor. On average, there was a 0% to 2% loss in production.

Agricultural nonpoint sources of pollution also impact the tributaries that drain to Lake Erie. A high percentage of these streams are impaired and many of these impairments have persisted for decades. These inland water impairments don’t generate the same level of interest in the press as algal blooms in the western basin but they are very important to wildlife.

What we need is a comprehensive picture of these problems – in tributaries and the western basin – to simultaneously address them and find win-win scenarios.

Since 2012 the Wildlife component of the USDA’s Conservation Effects Assessment Project (CEAP) has been supporting a SWAT modeling effort focused on the tributaries of the western Lake Erie basin to provide information on the types and extent of agricultural practices necessary to reduce in-stream nutrients and sediments and restore stream health. This Western Lake Erie Wildlife CEAP is complementary to the Multiple Modeling project and is nearing completion (April 30, 2016). Results of these two projects align in that both have determined that: a) addressing agricultural nonpoint sources will require significant increases in agricultural best management practices; b) nutrient management is critical; and c) targeting is essential. Furthermore, it is anticipated that products from these two efforts can be integrated to help find win-win scenarios.

In addition to the Western Lake Erie Wildlife CEAP, the Western Lake Erie Cropland CEAP project recently released a report on their efforts to quantify the impact of existing conservation practices, and to evaluate the potential impact of future scenarios on field level exports. Their projections are also consistent with the results from our effort and with the CEAP Wildlife project. Their projections show that significant increases in agricultural best management practices will be needed to achieve phosphorus load reductions to Western Lake Erie.

More information: CEAP Cropland Report, chapter five (PDF).

USDA Conservation Effects Assessment Program (CEAP)

CEAP is a multi-agency effort to quantify the environmental effects of conservation practices and programs and develop the science base for managing the agricultural landscape for environmental quality. Project findings will be used to guide USDA conservation policy and program development and help conservationists, farmers and ranchers make more informed conservation decisions.

Assessments in CEAP are carried out at national, regional and watershed scales on cropland, grazing lands, wetlands and for wildlife. The three principal components of CEAP—the national assessments, the watershed assessment studies, and the bibliographies and literature reviews—contribute to building the science base for conservation. That process includes research, modeling, assessment, monitoring and data collection, outreach, and extension education. Focus is being given to translating CEAP science into practice.

More information: nrcs.usda.gov/wps/portal/nrcs/main/national/technical/nra/ceap