Abstract Drought is among the most severe hydro-meteorological events, and it is expected to become even more critical and recurrent in multiple regions worldwide. The availability of hydrological information months ahead is crucial for developing effective early warning strategies, which often remain focused on meteorological drought, and to support effective adaptation programs and enhanced preparedness. The Copernicus Emergency Management System (CEMS) provides seasonal ensemble forecasts of river discharge based on the LISFLOOD hydrological model simulations, which are driven by seasonal meteorological ensemble forecasts from the seasonal forecast system version 5 (SEAS5) by the European Centre for Medium-Range Weather Forecasts (ECMWF). In this study, we evaluate the performance of these forecasts in detecting hydrological drought in the period 1991–2022 using the ECMWF reanalysis ERA5 as a proxy for the observational reference. The main results show that the skill of the ensemble mean of the Standardized Streamflow Index (SSI) is high at both 1- and 3-month time horizons; lower-skill forecasts (but still informative) can be achieved up to 6 months ahead. The forecast always outperforms trivial ones done by using the latest available state of the reanalysis, with the highest added value, on average, in the Northern Hemisphere. Seasonality in average river discharge is highlighted as one of the main drivers of the differences in skill between seasons. Additionally, the dominant role of inter-annual variability in initial conditions (soil moisture) and meteorological forcings (precipitation) on forecast skill is observed in summer and spring–summer for the former, and in winter and autumn–winter for the latter. Besides the high skill observed with the ensemble mean, the signal-to-noise ratio (SNR) of the ensemble forecast is identified as key metric to operationally assess and communicate forecast reliability.
García-Gamero et al. (2026) studied this question.