Predicting Subseasonal Hydrological Droughts for Swiss Lakes and Large Rivers – Combining Mesoscale Hydrological EPS and Machine Learning Approaches

Although Switzerland is not commonly associated with the occurrence of droughts, in recent years, Switzerland has experienced several unprecedented drought events. Considering that many sectors in Switzerland depend heavily on its water resources, hydropower production, navigation and transportation, agriculture, and tourism, it is important to establish a reliable warning system for early drought recognition. Drought forecast at subseasonal timescales, particularly the onset of a drought event, remains a challenge which is linked to the limited skill of subseasonal meteorological forecasts especially in Europe. The goal of this research is to develop a model to produce skillful subseasonal prediction of low-flows in large river basins and water levels of the major lakes in Switzerland. The envisaged methodology combines monthly hydro-meteorological forecast outputs from the hydrological model PREVAH (Precipitation-Runoff-Evapotranspiration HRU model) with machine learning algorithms. An operational setup of PREVAH has been previously implemented for Switzerland with meteorological forcing from 51 ensemble members and 32 days lead time from the operational extended-range prediction system of the European Centre for Medium-Range Weather Forecasts (ECMWF). Although the PREVAH forecasts are considered semi-idealized (assuming natural flow conditions) and they do not go through an in-depth calibration process, they provide a robust representation of the hydrological processes at the catchment level. The proposed machine learning model is expected to mimic the flow routing mechanism and match PREVAH forecasts from its 300 catchments with measured streamflow and lake level in river basins. The proof-of-concept will focus on the river Aare until the station of Brügg-Aegerten, downstream of the lake of Biel. The findings of this work will highlight the potential of directly linking mesoscale hydro-meteorological forecasts with streamflow and providing subseasonal low-flow predictions in an operational set-up.