Seasonal Stream Flow Forecasting and Decision Support in Central Texas

A decision support model based on stream flow ensemble forecasts has been developed for the Lower Colorado River Authority in Central Texas, and predictive skill is added to climatology-based forecasts by conditioning the ensembles on observable climate indicators. These indicators include stream flow (persistence), soil moisture, and large-scale recurrent patterns such as the El Nino-Southern Oscillation, Pacific Decadal Oscillation, and the North Atlantic Oscillation. In the absence of historical soil moisture measurements, the Variable Infiltration Capacity (VIC) Retrospective Land Surface Data Set is applied. Strong correlation between observed runoff volumes and runoff volumes simulated by the (uncalibrated) VIC model indicates the viability of this approach. Following correlation analysis to screen potential predictors, a Bayesian procedure for updating ensemble probabilities is outlined, and various skill scores are reviewed for evaluating forecast performance. Verification of the ensemble forecasts using a resampling procedure indicates a small but potentially significant improvement in forecast skill over climatology that could be exploited in seasonal water management decisions. Future work involves evaluation of seasonal soil moisture forecasts, further evaluation of annual flow forecasts, incorporation of climate forecasts in reservoir operating rules, and estimation of the value of the forecasts.