AbstractRainfall is a critical input variable of statistical streamflow forecasting models at subseasonal to seasonal time scales. This study presents a framework for evaluating the utility of a high-resolution experimental winter seasonal climate reforecasts for Florida (CLIFF) in improving streamflow forecasts. The CLIFF forecasts were coproduced through a scientist–stakeholder group of the Florida Water and Climate Alliance. The framework consists of a statistical streamflow generation model, four different sets of rainfall inputs, and distinct metrics for evaluating the resulting streamflow forecasts. The four sets of rainfall inputs include rainfall climatology, observed rainfall, NOAA-based seasonal rainfall forecasts, and CLIFF-based rainfall forecasts. Because NOAA ensemble precipitation forecasts were not available in this study, NOAA-based categorical precipitation outlooks were postprocessed via a hidden Markov chain model to obtain the corresponding NOAA-based seasonal rainfall forecasts. Streamflow forecasts based on rainfall climatology served as a reference. Different evaluation metrics, including Spearman correlation, mean absolute percent error (MAPE), and rank probability skill score (RPSS), were employed to evaluate model performance. The framework was demonstrated for streamflow forecasts for two rivers in the southwest of Florida, serving as a major source of a regional water supply agency. A retrospective streamflow forecasting model was designed for the dry season [November, December, January, and February (NDJF) months] for each of the 20 years from 2000 to 2019. Results revealed that CLIFF-based streamflow forecasts are a promising alternative to NOAA-based forecasts. Deterministic streamflow forecasts based on CLIFF rainfall have a smaller mean absolute percent error (MAPE) compared with the NOAA-based streamflow forecasts. Although NOAA-based probabilistic streamflow forecasts outperformed CLIFF-based probabilistic streamflow forecasts for the winter forecasting periods of November, December, and January, the latter forecasts performed better for the forecasting period of February. Thus, the two probabilistic forecasts are complementary. Although the results are limited to the study area, it has general application for evaluating the utility of different rainfall forecasts in providing deterministic/probabilistic streamflow forecasts.