Forecast-based Attribution of Climate Change Signals in High-Impact Extratropical Cyclones Using AI Weather Models

Extratropical cyclones are among the most damaging weather systems in midlatitudes, yet quantifying the influence of anthropogenic climate change (ACC) on individual storms remains challenging due to the interplay between thermodynamic and dynamical processes. Here, we assess the potential of AI-based weather prediction (AIWP) models to both forecast and attribute ACC signals in two high-impact European extratropical cyclones with contrasting characteristics: storm Ciarán (November 2023), which underwent explosive cyclogenesis and extreme winds, and storm Claudia (November 2025), characterized by an intense atmospheric river.

We evaluate four state-of-the-art AIWP models and benchmark them against the operational forecasts of ECMWF Integrated Forecast System (IFS), finding that all AIWP systems skillfully reproduce the large-scale evolution of both storms several days in advance, albeit with event-dependent performance. Building on this skill, we apply a forecast-based storyline attribution framework in which factual forecasts initialized from ERA5 are compared with counterfactual simulations generated by applying a pseudo–global warming (PGW) perturbation derived from CMIP6 historical simulations to the model initial conditions.

The attribution analysis reveals physically coherent ACC fingerprints across sea-level pressure, low-level winds, moisture, and precipitation. Moisture-related signals are robust across models for both storms, while wind and circulation responses show greater model and event dependence, reflecting the differing levels of robustness in thermodynamic versus dynamic responses to climate change. Precipitation attribution using the ECMWF Artificial Intelligence Forecasting System (AIFS) indicates ACC-driven increases in accumulated rainfall for both events, consistent with near–Clausius–Clapeyron thermodynamic scaling at the regional scale, but with locally larger increases arising from event-dependent dynamical adjustments. These results demonstrate that AIWP models enable fast, event-specific climate change attribution within the forecast window, supporting near–real-time assessments of ACC influences on high-impact extratropical cyclones and opening new opportunities for operational climate services.