Evaluation of Machine Learning and Deep Learning Models for Extreme Climate Event Forecasting

One of the greatest challenges in this era is climate change which calls for the better capacity of accurate prediction and adaptation tools as mitigation tools. This study investigates the suitability of artificial intelligence (AI) to be applied in climate modeling by evaluating accuracy of two machine learning (ML) and one deep learning (DL) models. In particular, this paper chooses extreme events datasets wherein the traditional General Circulation Models (GCMs) often have difficulty in prediction or analysis. In particular, the study deals with the rare but critical cases (extreme temperature, tornados and anomalous meteorological phenomena in general) using historical climate data and dedicated feature engineering methods. The selected models include Random Forest, XGBoost, and a Long Short-Term Memory (LSTM) network, and they are tested on a curated dataset containing various extreme weather events. Comparative analyses of strengths and weaknesses of each approach judge predictive accuracy, computational efficiency, and generalization to novel conditions. Results show that, compared to the classical ML approach, the deep learning lacks when testing with the same amount of data records. However, DL approach shows great improvements in capturing spatiotemporal dependencies with increasing dataset size and making early warning systems more robust. This work emphasizes how machine and deep learning may play a role in supporting GCMs, particularly in respect to forecast of complex extreme events, by identifying areas in which AI can increase the strength or resolve of existing frameworks. These findings provide a strong rationale for incorporating AI methods into climate research and policy decision-making, thereby enabling data-driven, adaptive strategies aimed at reducing uncertainties in climate risk modeling and mitigating climate threats.