Sub-seasonal streamflow predictions by combining numerical weather models and re-analysis data in alpine catchments

The European Center for Medium-Range Weather Forecasts (ECMWF) has presented in 2017 its latest seasonal forecasting system, SEAS5, available at 1° spatial resolution and daily timestep. More recently, in 2019, the ERA5 reanalysis dataset was released, replacing ERA Interim in providing climatic variables at a finer spatial and temporal resolution (30 km and hourly respectively). The use of such numerical weather predictions and re-analysis data has increased following the need for skills in planning water resources and preventing hydrogeological risk, as demanded by policy makers, energy stakeholders and public authorities. In this work, we apply at a sub-seasonal timescale the ECMWF-SEAS5 hindcast dataset to assess its prediction skills in the upper Adige river basin in the Eastern Italian Alps. The classical Extended Streamflow Prediction (ESP) framework was designated as a benchmark to assess ECMWF scores over the reference, a model simulation calibrated and validated on the runoff observed from 16 sub-basins and size spanning from 50 to 6900 km². Before application, ECMWF was downscaled and adjusted to the ERA5 re-analysis data by means of a Quantile Mapping (QM) technique. The analysis was conducted over 23 hindcast years from 1993 to 2016 exploiting the semi-distributed basin-scale hydrological model (ICHYMOD). We showed that the sub seasonal QPF-based forecasts have advantages over the ESP, although, generally their skill deteriorates in lead times after day 15. Moreover, ECMWF predictions better perform during early-spring snowmelt and late summer. During late spring and early summer, the forecast skills of the two frameworks vary from basin to basin depending on specific features and lead times.