Abstract ID: 135
Sub-seasonal drought forecasting in the European Alps with EFAS data in a machine-learning-aided hybrid approach
Lead Author: Annie Y.-Y. Chang
ETH Zurich, Switzerland
Keywords: Hybrid Forecasting, Drought, EFAS, Streamflow, Alpine region
Abstract: The European Flood Awareness System (EFAS) has been in operation since 2012 providing flood risk overviews for Europe up to 15 days in advance. More recently, it has also run long-range hydrological outlooks for sub-seasonal to seasonal horizons. In this study, we turn our focus to this increased anticipation capability and how it may contribute to EFAS predictability of drought events in support to trans-national operational services. More specifically, our study area focuses on Alpine catchments. In recent years, the European Alpine space has experienced several unprecedented low-flow conditions and drought events. As many economic sectors in the region depend heavily on sufficient water availability, such as hydropower production, navigation and transportation, agriculture, and tourism, it is important for decision-makers to have early warnings of drought tailored to their needs and geographical conditions.
Our goal is to investigate how we can adopt the 46-day sub-seasonal EFAS forecasts in a hybrid forecasting system to more accurately inform decision-makers on low flow conditions and their spatio-temporal evolution across the Alpine space. Our case study area comprises 139 catchments located in the European Alpine space. We apply the machine learning algorithm Temporal Fusion Transformer (TFT), which combines LSTM (Long Short-term Memory) networks for local processing of known inputs and temporal attention layers for learning long-term dependencies within the forecast horizon. We incorporate features including European weather regime forecasts, streamflow climatology, hydropower proxies, and spatial relationships among forecast stations to improve the predictability of low flow. We focus on evaluating the duration, deficit, and magnitude of the drought events, as well as the ensemble forecast reliability using metrics such as the Continuous Ranked Probability Skill Score (CRPSS).
The outcome of this study will shine a light on the application of the EFAS forecasts in sub-seasonal hydrological drought predictions. This study will contribute to evaluating the potential of these forecasts to provide useful information for achieving skillful early warnings, while also supporting regional and local water resource management.
Co-authors:
Konrad Bogner (WSL, Birmensdorf, Switzerland)
Maria-Helena Ramos (Université Paris-Saclay, INRAE, HYCAR, Antony, France)
Shaun Harrigan (ECMWF, Reading, UK)
Daniela I.V. Domeisen (ETH Zürich, Zürich, Switzerland & University of Lausanne, Lausanne, Switzerland)
Massimiliano Zappa (WSL, Birmensdorf, Switzerland)
