Forecasting the Future: Building AI-Powered Weather Models with Python in an African HPC Context

Climate change is a lived reality in Africa, manifesting in extreme weather events from droughts to floods. This necessitates improved weather forecasting and climate resilience across the continent. This talk details my participation in the AI WeatherQuest competition, a global initiative challenging participants to develop machine learning models for medium-range weather prediction using reanalysis data. Leveraging my work at the Centre for High Performance Computing (CHPC) and PhD research at UCT’s Department of Oceanography, I will present the end-to-end workflow developed for AI-driven forecasting models. Utilising Python’s robust scientific ecosystem, including xarray, pandas, scikit-learn, and deep learning frameworks like PyTorch. I established a modular pipeline to ingest gridded climate data, engineer meaningful features, and train both baseline and advanced neural network architectures. This workflow was significantly accelerated and optimised on the Lengau supercomputer, and seamlessly integrated with established climate tools like METplus. I will demonstrate how Python facilitated this interaction across scientific data formats, HPC job schedulers, and visualisation tools (cartopy, matplotlib, hvplot), resulting in a workflow that is both scalable and reproducible for the African domain.
Beyond the technical advancements, I will discuss the broader relevance of this work for African weather services, research software communities, and open science advocates. I will also share how this experience aligns with my ongoing mission to democratise complex HPC workflows, particularly through localisation and inclusive educational initiatives at CHPC summer schools and coding events.
This talk is ideal for anyone interested in climate modelling, AI for science, reproducible workflows, or leveraging Python for impactful, real-world challenges within the African context. Attendees can expect a compelling blend of code, computation, and climate science, all contributing to the vision of a more resilient and prepared continent.