Deep Learning for Operational High-Resolution Nowcasting in Switzerland Using Graph Neural Networks

Abstract: Recent advances in neural weather forecasting have shown significant potential for accurate short-term forecasts. However, adapting such gridded approaches to smaller, topographically complex regions like Switzerland introduces computational challenges, especially when aiming for high spatial (1 km) and temporal (10 minutes) resolution. This paper presents a Graph Neural Network (GNN)-based approach for high-resolution nowcasting in Switzerland using the Anemoi framework and observational inputs. The proposed model combines surface observations with selected past and future numerical weather prediction (NWP) states, enabling an observation-guided interpolation strategy that enhances short-term accuracy while preserving physical consistency. We evaluate the method on multiple surface variables and compare it against operational high-resolution NWP (ICON) and nowcasting (INCA) baselines. The results show that the GNN model consistently outperforms traditional approaches in lead times up to 12 hours, especially for wind and precipitation. A comprehensive verification procedure, including spatial skill scores, event-based evaluation, and blind tests with professional forecasters, demonstrates the operational relevance of the approach for mountainous domains.