Benchmarking analog and ensemble-based seasonal forecasting strategies for water management in the Upper Rio Grande basin

In the southwestern US, declining runoff efficiencies driven by a warming climate have undermined the skill of seasonal water supply forecast (WSF) methods used for reservoir management by local to federal agencies. Seasonal water allocations are often based on deterministic inflow sequences, derived by matching historical streamflow traces (analogs) to statistical WSF volumes; yet model-based ensemble streamflow forecasting offers a compelling alternative. We evaluate this alternative through a systematic, hindcast-based benchmarking assessment, applying process-based hydrologic modeling to predict streamflow for US Bureau of Reclamation system inflow points in the Upper Rio Grande (URG). We demonstrate the viability of model-based prediction of disaggregated WSFs for guiding reservoir planning compared to existing analog-based practices, using the Ensemble Streamflow Prediction (ESP) technique to develop a 49-year dataset of April 1st hindcasts. Across thirteen URG forecast points, the bias-corrected ESP mean sequences consistently improved hydrograph shape over analog-based sequences, with a median KGE increase of +0.09. For peak flow characteristics, performance was broadly comparable. These results show that ESP-based predictions of seasonal inflow shape are a compelling option for reservoir management where analog-based methods are still used. This study also presents an early implementation of the SUMMA-mizuRoute framework for regional water modeling and seasonal ESP.