Recent advances in data-driven weather and climate modelling have challenged the traditional dominance of physics-based numerical prediction systems. Machine-learning-based models, such as AIFS, have demonstrated competitive or superior performance for a range of standard forecast metrics. However, these evaluations primarily focus on large-scale variables under typical conditions, while the representation of local, high-impact extreme events remains insufficiently explored.This study investigates the ability of data-driven models to represent extreme precipitation and to generalise beyond the conditions encountered during training. A graph neural network model, trained using the Anemoi framework, is developed based on output from a convection-permitting (4 km) regional climate simulation with the ALARO model over Western Europe. Model performance is quantified using precipitation return levels derived from annual maxima across multiple accumulation durations, ranging from hourly to multi-day timescales. This experimental design enables a systematic comparison between data-driven and physics-based representations of extremes.To explicitly assess generalisation, a second model is trained on a modified dataset in which the upper tail of the precipitation distribution is selectively masked. By comparing return levels from both data-driven models with those from the reference simulation, we evaluate the extent to which such models can reproduce extremes that are absent from the training data.This framework provides a controlled setting to assess the robustness of AI-based models for high-impact precipitation events and their ability to generalise to unseen conditions. This is particularly relevant for extreme precipitation, which is inherently underrepresented in training data and is expected to evolve under climate change, potentially leading to conditions outside the historical training distribution.
Mesoscale models serve as a critical bridge between urban microclimate dynamics and larger atmospheric processes simulated by earth system models, enabling a deeper understanding of weather and climate phenomena in cities. Long-term regional projections with high spatiotemporal resolution are currently lacking which limits comprehensive research addressing climate change effects in cities. Therefore, a high-resolution European climate dataset was constructed for the 2011-2021 period by downscaling ERA5 reanalysis data to 2.5 km, with an intermediate step at 12.5 km, using the ALARO-1 regional climate model coupled online with the SURFEX land surface model. By focusing on city resolving scales, the dataset allows to investigate how extreme weather events are manifested in cities. Next to facilitating detailed investigations into extreme weather events, the dataset serves as a training set to construct a computational cheap machine learning emulator. The emulator, a U-Net model, downscales surface temperature, wind and relative humidity simultaneously from 12.5 km to 2.5 km permitting us to efficiently downscale EURO-CORDEX projections to city-resolving scales. This novel approach enables the creation of an ensemble of urban-scale climate projections, enhancing our ability to evaluate the impact of different climate scenarios on European cities.Applications include investigating the interplay between heatwaves and the urban heat island effect based on the historical dataset. Furthermore, the climate projections at high spatiotemporal resolution will permit future impact studies on heritage and health. Hence, this research provides valuable insight into urban planning and resilience strategies.