Increasingly, Data Availability Limits Model Predictive Capacity: the Western Lake Erie Basin, a Case Study

Recent algal blooms in Western Lake Erie Basin (WLEB) have renewed scientific community's interest in developing process based models to better understand and predict the drivers of eutrophic conditions in the lake. At the same time, in order to prevent future blooms, farmers, local communities and policy makers are interested in developing spatially explicit nutrient and sediment management plans at various scales, from field to watershed. These interests have fueled several modeling exercises intended to locate "hotspots" in the basin where targeted adoption of additional agricultural conservation practices could provide the most benefit to water quality. The models have also been used to simulate various scenarios representing potential agricultural solutions. The Soil and Water Assessment Tool (SWAT) and its sister model, the Agricultural Policy Environmental eXtender (APEX), have been used to simulate hydrology of interacting land uses in thousands of scientific studies around the world. High performance computing allows SWAT and APEX users to continue to improve and refine the model specificity to make predictions at small-spatial scales. Consequently, data inputs and calibration/validation data are now becoming the limiting factor to model performance. Water quality data for the tributaries and rivers that flow through WLEB is spatially and temporally limited. Land management data, including conservation practice and nutrient management data, are not publicly available at fine spatial and temporal scales. Here we show the data uncertainties associated with modeling WLEB croplands at a relatively large spatial scale (HUC-4) using site management data from over 1,000 farms collected by the Conservation Effects Assessment Project (CEAP). The error associated with downscaling this data to the HUC-8 and HUC-12 scale is shown. Simulations of spatially explicit dynamics can be very informative, but care must be taken when policy decisions are made based on models with unstated, but implicit assumptions. As we interpret modeling results, we must communicate the spatial and temporal scale for which the model was developed and at which the data is valid. When there is little to no data to enable appropriate validation and calibration, the results must be interpreted with appropriate skepticism.