Assessing a Decade of Progress: Characterizing Performance Trends in Archived Operational Ensemble Streamflow Forecasts Across the Western US

Hydroclimate variability and the frequency of extreme events are increasing globally, exacerbating water-related risks and disruptions to critical infrastructure, agriculture, energy systems, human health, and ecosystems. One of the most cost-effective adaption strategies is forecast-informed decision-making, which leverages hydrometeorological forecasts to improve water resource management outcomes. The success of forecast-based strategies depends on forecast skill, prompting substantial investment in improving hydrometeorological forecasting systems. However, it remains unclear whether operational forecast skill has improved over time – particularly for ensemble hydrologic forecasts – due in part to the absence of a systematic, scalable framework for evaluating changes in probabilistic forecast performance with limited historical records.

In the United States, the Hydrologic Ensemble Forecast System (HEFS) was established by the National Weather Service in 2014 and provides ensemble streamflow forecasts at more than a thousand locations nationwide. Despite several system enhancements over the past decade and HEFS’s expanding operational role, there has been no comprehensive evaluation of the evolution of operational forecast skill. This study addresses that gap through a retrospective analysis of operational, short- and medium-range (1-14 day ahead) ensemble streamflow forecasts issued by the California Nevada River Forecasting Center at more than 90 locations between water years 2014-2025. We quantify trends in both deterministic and probabilistic skill using a hierarchical Bayesian model designed to improve inference under sparse data. This framework offers a robust, transferable approach for tracking operational forecast performance and supporting the long-term potential of forecast-informed adaptation strategies. Preliminary results suggest that the performance of operational forecasts for high flow events has improved over the past decade – though the degree of improvement depends on the performance metric and forecast lead time examined. There is some evidence that improvements made to the HEFS forecasting system – namely, enhanced data assimilation and meteorological model version updates – may be associated with forecast improvements at short lead times.