A Comparison of Daily Streamflow Prediction by an Artificial Neural Network and the Soil and Water Assessment Tool (SWAT) in Two Small Watersheds in Central South Texas

The current available rainfall-runoff models vary from the simple lumped data-driven models that purely depend on the observed inputs and outputs of a watershed to the more complex physically-based models representing a system using mathematical equations that describes important physical laws of conservation of mass, energy, and momentum. In this study, the accuracy of streamflow estimated by a data-driven Artificial Neural Network (ANN) and the physically-based Soil and Water Assessment Tool (SWAT) are compared. The models were applied in two small watersheds, one highly urbanized and the other primarily covered with forest and shrub, in the San Antonio Region of central south Texas, where karst geologic features are prevalent. Both models were found to perform very well in the prediction of daily streamflow in the urbanized watershed. However, both models poorly predicted streamflow in the nonurban watershed. Additionally, the ANN model performance significantly improved when a time series autoregressive model structure using historical streamflow data was implemented. The SWAT model achieved very limited improvement through model calibration with the current model structure. This result suggests that the ANN model is more suitable for short-term streamflow forecasting in watersheds heavily affected by karst features where surface water flow is strongly influenced by the complex processes of rapid groundwater recharge and discharge.