Generative prediction of severe convection events using satellite imagery and diffusion-u-net architectures

Severe convection events pose significant hazards and demand accurate prediction for effective mitigation. Traditional forecasting methods often struggle with the high resolution and specificity required for such phenomena. This study leverages generative modeling to enhance the prediction of severe convection using satellite imagery. In this work, we purpose an Integrated Convection Events Prediction Framework (ICEPF) that leverages advanced satellite imagery and cutting-edge machine learning techniques to enhance the prediction and analysis of severe weather phenomena. The ICEPF combines a diffusion-based generative model for predicting future satellite imagery, a fine-tuned ESRGAN model for enhancing image resolution, and a MobileNet-SegNet-based semantic segmentation model for identifying cumulus clouds. By integrating these models, our framework offers a powerful tool for forecasting severe convection with improved accuracy and detail. Our results indicate that combining generative diffusion models with advanced segmentation techniques can provide a robust framework for forecasting severe weather conditions, offering potential enhancements over traditional meteorological methods.