Using Generative Models to Produce Realistic Populations of UK Windstorms
This work addresses the problem of realistic windstorm simulation for meteorological research and risk assessment, but it is incremental as it compares existing generative models without introducing a new method.
This study evaluated four generative models for simulating UK windstorms using historical data, finding that each had distinct strengths and limitations in replicating spatial and statistical characteristics, with an ensemble approach suggested for improved reliability.
This study evaluates the potential of generative models, trained on historical ERA5 reanalysis data, for simulating windstorms over the UK. Four generative models, including a standard GAN, a WGAN-GP, a U-net diffusion model, and a diffusion-GAN were assessed based on their ability to replicate spatial and statistical characteristics of windstorms. Different models have distinct strengths and limitations. The standard GAN displayed broader variability and limited alignment on the PCA dimensions. The WGAN-GP had a more balanced performance but occasionally misrepresented extreme events. The U-net diffusion model produced high-quality spatial patterns but consistently underestimated windstorm intensities. The diffusion-GAN performed better than the other models in general but overestimated extremes. An ensemble approach combining the strengths of these models could potentially improve their overall reliability. This study provides a foundation for such generative models in meteorological research and could potentially be applied in windstorm analysis and risk assessment.