Comparison of Short-Term Streamflow Forecasting using Stochastic Time Series, Neural Networks, Process-Based, and Bayesian Models

Abstract Streamflow forecasts are essential for water resources management. Although there are many methods for forecasting streamflow, real-time forecasts remain challenging. This study evaluates streamflow forecasts using a process-based model (Soil and Water Assessment Tool-Variable Source Area model-SWAT-VSA), a stochastic model (Artificial Neural Network -ANN), an Auto-Regressive Moving-Average (ARMA) model, and a Bayesian ensemble model that utilizes the SWAT-VSA, ANN, and ARMA results. Streamflow is forecast from 1-8 d, forced with Quantitative Precipitation Forecasts from the US National Weather Service. Of the individual models, SWAT-VSA and the ANN provide better predictions of total streamflow (NSE 0.60-0.70) and peak flow, but underpredicted low flows. During the forecast period the ANN had the highest predictive power (NSE 0.44-0.64), however all three models underpredicted peak flow. The Bayesian ensemble forecast streamflow with the most skill for all forecast lead times (NSE 0.49-0.67) and provided a quantification of prediction uncertainty.