• VarDA enhances seasonal SWE and inflow forecasts, especially in the snowmelt season. • SEAS5 forecasts outperform ESP by better capturing climate-driven hydrological signals • Assimilating sparse SWE data improves model initialization and early lead-time forecast skill. Seasonal inflow forecasting is critical for optimizing hydropower operations in snow-dominated basins, where climate-induced non-stationarity challenges traditional methods. This study integrates a variational data assimilation (VarDA) scheme into a semi-distributed hydrological model (HBV) to enhance seasonal predictions of inflow, snow cover area, and snow water equivalent in the Manic-5 basin, Québec, Canada. The hydrological model is calibrated and validated using observed hydro-meteorological data, IMS-based snow cover area, and in-situ/ERA5-Land snow water equivalent datasets. A closed-loop hindcasting framework is applied for 2017–2022 under three meteorological forcing scenarios: single perfect forecasts, climatology-based ensemble forecasts with the Extended Streamflow Prediction (ESP), and meteorological data-based ensemble forecasts from ECMWF (European Centre for Medium-Range Forecast) Seasonal Forecasting System version 5 (SEAS5). Results show that variational data assimilation significantly improves both inflow and snow water equivalent forecasts across all scenarios, particularly during the first 1–3 months of lead time. SEAS5 outperforms ESP in capturing climate signals, while ESP retains skill through historical analogs. Forecast skill is evaluated using mean continuous ranked probability score, Brier skill score, and reliability diagrams. Variational data assimilation enhances model initialization through the assimilation of sparse in-situ snow water equivalent data, especially in spring. The findings highlight the potential of combining ensemble-based forecasting with data assimilation to improve seasonal hydrological predictability and support climate-resilient hydropower planning in cold-region basins.
Uysal et al. (Sun,) studied this question.