Evapotranspiration and Precipitation Inputs for SWAT Model Using Remotely Sensed Observations.

The ability of numerical models, such as the Soil and Water Assessment Tool (or SWAT), to accurately represent the partition of the water budget and describe sediment loads and other pollutant conditions related to water quality strongly depends on how correctly spatiotemporal variability in precipitation and evapotranspiration (ET) can be described. Nonetheless, water managers and other users often neglect to assess the ability of the models to correctly express temporal variability in the water budget and instead rely on the ability of the models to match annual means when calibrating or validating the models. Using data collected from both corn and soybean fields during a three year period in the South Fork watershed in central Iowa, this study investigated methods for improving the ability of SWAT model to describe both precipitation and ET, the two critical water balance components. By combining high resolution NEXRAD data with rainfall measurements from gauge networks, the accuracy of the precipitation inputs for SWAT, and thus the model representation of the water budget, is greatly enhanced. Likewise, to calculate ET, SWAT requires an estimate of potential ET (PET) that is either determined internally by the model using: i) the Priestly-Taylor; ii) Penman-Monteith; or iii) Hargreaves methods, or iv) calculated by an external source and provided to SWAT as an input. For this study, ET estimates derived from PET were computed internally and compared to spatial distributed ET estimates derived from in-situ observations and MODIS NDVI data.  Reading-in potential ET into SWAT computed from local or regional stations performed the best, while the weather generator input only captured the annual average, but not the variation throughout the growing season affecting biomass and yield estimates.  This research provides insight into the value of remote sensing and field observations for the application of physically-based watershed models, as well as the selection of potential ET inputs to drive the model.