A scenario-based stochastic programming model for water supplies from the highland lakes

A scenario-based, multistage stochastic programming model is developed for the management of the Highland Lakes by the Lower Colorado River Authority (LCRA) in Central Texas. The model explicitly considers two objectives: (1) maximize the expected revenue from the sale of interruptible water while reliably maintaining firm water supply, and (2) maximize recreational benefits. Input data can be represented by a scenario tree, built empirically from a segment of the historical flow record. Thirty-scenario instances of the model are solved using both a primal simplex method and Benders decomposition, and results show that the first-stage (‘here and now’) decision of how much interruptible water to contract for the coming year is highly dependent on the initial (current) reservoir storage levels. Sensitivity analysis indicates that model results can be improved by using a scenario generation technique which better preserves the serial correlation of flows. Ultimately, it is hoped that use of the model will improve the LCRA's operational practices by helping to identify flexible policies that appropriately hedge against unfavorable inflow scenarios.