Project Overview

This research investigates the predictive skill of state-of-the-art deterministic AI weather models for extreme events through the lens of Uncertainty Quantification (UQ). By applying input perturbations—varying from Gaussian noise to Hemispheric Centered Bred Vectors and Huge Ensembles—we evaluate how these data-driven models respond and whether they can provide reliable probabilistic forecasts for rare, high-impact events like the 2022 Pakistan floods.

Papers

On the Predictive Skill of Artificial Intelligence-based Weather Models for Extreme Events using Uncertainty Quantification
Published November 21, 2025
This study demonstrates that input perturbations can extend deterministic models towards probabilistic forecasting, paving the way for reliable AI-driven early warning systems.

Can Artificial Intelligence Global Weather Forecasting Models Capture Extreme Events? A Case Study of the 2022 Pakistan Floods
Presented December 2025
A focused case study comparing three deterministic data-driven models (FourCastNet v2/SFNO, GraphCast, and FuXi) against the ECMWF Integrated Forecasting System Ensemble.

Conference Presentations

NeurIPS 2025 Workshop on Tackling Climate Change with Machine Learning
December 7, 2025, San Diego, USA
Poster: “Can Artificial Intelligence Global Weather Forecasting Models Capture Extreme Events? A Case Study of the 2022 Pakistan Floods”

3rd Workshop on Machine Learning for the Earth System
August 25, 2025, Bonn, Germany
Oral Presentation: “Can AI weather models capture rare events? A case study of the 2022 Pakistan floods”