Applying Season-ahead Streamflow Forecasts to Guide Reservoir Allocations for the Highland Lakes in Central Texas

Abstract

Season-ahead reservoir inflow forecasts can assist water managers by anticipating and preparing for extreme conditions, such as droughts. Previous studies have developed statistical streamflow models, which rely on hydrologic persistence or large-scale climate variables. This study incorporates both, including local and global predictor information, in a novel hybrid autoregressiveprincipal component regression framework. A superior combination of predictors is obtained, based on select performance metrics, using both linear and logistic regression models. Application to the Lower Colorado River basin in Texas demonstrates significant overall skill, particularly for the March-June inflow season (correlation = 0.70; RPSS = 0.73) Evaluation of the July-October inflow season also exhibits increased predictive skill over climatology (correlation = 0.63; RPSS = 0.65.) The prediction model outputs are subsequently coupled to a reservoir operations model to compare the impacts of incorporating a streamflow forecast into management policies versus simply applying long-term historical average. The two major reservoirs that provide water for the LCRB, Lake Travis and Lake Buchanan, are treated as a single water supply and a water balance model is created to simulate outflows (e.g. evaporation, water allocations, environmental demands) and actual inflows for the time period 1982-2010. Operating policies from the most recent revision of the Lower Colorado River Authority?s Water Management Plan are used. Allocations are made based on reliability of maintaining combined storage in the reservoir above 1.3 million acre-feet, which is the threshold for meeting extreme drought criteria within the current Water Management Plan. Results indicate some increased benefits with the forecast in both short and long-term operations, especially in response to extreme events.