Abstracts

Bayesian Deep Learning for Weather and Climate Extremes

Rishikesh Yadav

Weather and climate extremes such as heavy rainfall, heatwaves, floods, droughts, and tropical storms are becoming increasingly frequent and severe under a changing climate. Accurate prediction of these rare but high-impact events remains a major scientific challenge due to their complex spatiotemporal dynamics, nonlinearity, and inherent uncertainty. Bayesian deep learning provides a powerful framework for modeling such extremes by combining the representation learning capability of deep neural networks with probabilistic uncertainty quantification. This study explores Bayesian deep learning algorithms for forecasting weather and climate extremes using large-scale atmospheric and environmental datasets, e.g., ERA5 reanalysis datasets. The framework integrates neural network architectures with Bayesian inference techniques to capture predictive uncertainty, improve robustness, and provide reliable probabilistic forecasts. Special emphasis is placed on extreme-value behavior, tail-risk estimation, and uncertainty-aware predictions that are essential for climate risk assessment and decision-making.