AI for extreme weather forecasting: Cyclone prediction using deep learning

This thesis aims to analyse the role of AI in extreme weather forecasting, with a particular focus on Long Short-Term Memory (LSTM)-based and Transformer-based deep learning models for cyclone prediction. It applies sequence-based learning techniques to track storm development using historical data from International Best Track Archive for Climate Stewardship (IBTrACS) and reanalysis data from European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5). The methodology involves data preprocessing, feature scaling, and model parameter optimisation to improve the accuracy of wind speed and cyclone intensity predictions. Despite AI's potential, challenges persist in this domain, including limited data availability, high computational demands, and the necessity for explainable models. This thesis proposes solutions such as explainable AI, hybrid modelling, and the creation of scalable, efficient models tailored for regions with limited resources. Advancing AI-driven solutions for weather forecasting ultimately supports climate resilience efforts and highlights AI's broader role in mitigating environmental challenges.