Evaluating machine-learning weather prediction for UK windstorm weather warnings

Machine-learning weather prediction (MLWP) has recently emerged as a powerful new approach, with potential to complement or even replace numerical weather prediction (NWP) within early warning systems. The low computational cost of MLWP models, once trained, offers advantages in terms of the timeliness of forecasts, accessibility by national meteorological services, and energy savings. Yet their ability to capture weather hazards—particularly extreme winds associated with extra-tropical cyclones—remains insufficiently understood. This study provides the first comprehensive evaluation of MLWP performance for dynamically diverse UK windstorms, addressing a critical evidence gap before these models can be responsibly deployed in weather risk management. We analyse eight state-of-the-art MLWP models and benchmark them against reanalysis, NWP analysis products, and an operational NWP ensemble for six contrasting UK cyclones from winter 2023/24. We demonstrate substantial inter-storm and inter-model variability in skill at 24-hour lead time. Forecast accuracy is linked to the underlying atmospheric dynamics and spatial scales of the wind-producing features: wind maxima associated with mesoscale structures are systematically underestimated across all MLWP systems. Significant variability in the representation of mesoscale storm structure is also identified, and the highest-resolution MLWP model (Aurora-0.1) does not deliver uniform improvements across the cyclone set. In-depth analysis of two mature Shapiro–Keyser cyclones reveals a consistent tendency for MLWP systems to smooth sharp frontal gradients and to misrepresent key mesoscale features, including bent-back warm fronts and frontal-fracture regions. These deficiencies are physically consistent with the underestimation of localised extreme winds. Overall, our findings provide insight into the current limitations of MLWP for underpinning wind weather warnings over the UK. While MLWP models are not yet suitable to replace NWP models for severe windstorm prediction, their demonstrated skill in capturing storm tracks and synoptic-scale evolution suggests promising complementary pathways for future integration into national weather forecasting and resilience frameworks.