Evaluating the application of deep-learning ensemble sea level and storm surge forecasting in the Baltic Sea

The sea level predictions, which are done with state-of-the-art hydrodynamic models, often suffer from various uncertainties which are related to errors in atmospheric and open boundary forcing fields as well as in physical flux parametrizations. The errors in atmospheric prediction, and regional extreme phenomena like storm surges are due to models' approximated descriptions of the physical environments (e.g coupling with waves, atmosphere, ice, runoff), but also due to the stochastic nature of weather prediction, which is often treated as a single deterministic forecast in many applications. Moreover, the submission of correct warning alerts is subject to conversion to local reference level. The application of data-driven AI models can offer a promising addition to the classic hydrodynamic models in addressing these challenges. In the present study the sea level deep-learning forecasting model HIDRA2, demonstrates promising capabilities for forecasting storm surges in numerous coastal stations across the eastern coast of the Baltic Sea. Our comprehensive assessment of HIDRA2's includes intercomparison with the sea level forecasts predicted with the regional configuration of the NEMO 4.0 hydrodynamics model. Moreover the probabilistic storm surge forecast from the ensemble sea level predictions allows us to identify and refine the best sub-set of ensembles for accurately predicting storm surges. This case study will play an important role in guiding decision-making processes regarding the integration of deep-learning methodologies into the operational phase of sea level prediction, particularly in the Baltic Sea region.

The EU funds this work under the agreement DE_330_MF between ECMWF and Météo-France. The on-demand capability proposed by the Météo-France led international partnership is a key component of the weather-induced extremes digital twin, which ECMWF will deliver through different phases of Destination Earth, launched by the European Commission.