Rules-Based Systems Modeling for Hydropower Forecasting in Multi-Objective Reservoir Systems: Application to California's Central Valley Project

Yash Vijay Amonkar , H.B. Zeff, Eric Mork, Konstantinos Oikonomou, Kelsey A Semrod, Gregory W Characklis

Hydropower from large multi-objective reservoirs and water management projects constitutes the bulk of global reservoir-based generation. Yet accurate forecasting remains challenging because generation is governed not only by hydrology but by complex institutional rules, environmental regulations, and infrastructure constraints. This study demonstrates that a rules-based systems modeling approach, one that explicitly integrates water allocation institutions, operational constraints, and hydrological variability, produces significantly more accurate and financially valuable hydropower forecasts than purely statistical methods. This framework is applied to California's Central Valley Project (CVP) using CALFEWS, an open-source daily timestep rules-based model of the Central Valley, extended here to include the Trinity River Diversion System and its evolving environmental flow regulations. CALFEWS is validated against 20 years of operational hydropower data (2003–2023), achieving r² = 0.84 for monthly CVP hydropower generation. Herein, we develop a 12 month two-stage October–April hydropower forecasting approach, initialized with October storage conditions and historical climatology, that incorporates water year type classification in April. The resulting forecasts achieve high fidelity, nearly matching the skill of simulations with perfect inflow foresight. Compared to statistical baselines, CALFEWS reduces hydropower forecast errors by 40% (RMSE) and hydropower financial losses relative to historic baselines by 20% ($75–115 million over a decade). While demonstrated for the CVP, the framework is directly transferable to any multi-objective reservoir system where institutional rules and competing water demands govern hydropower generation, a condition common across the U.S. and globally.