Towards thunderstorm projections for Canada using machine learning and HighResMIP climate model simulations

Thunderstorms lead to lightning, strong winds, hail, and heavy rainfall, which pose risks to life, property, and ecosystems. Projecting their occurrence in future climates is critical yet challenging due to their localized nature and the computational demands of models capable of explicitly representing convective processes. Although outputs from a small number of continental-scale simulations are now available, the high computational expense of convection-permitting models has limited their application to smaller domains using a limited set of models and experimental designs.

This seminar explores a complementary approach, integrating observed lightning data, reanalysis data, and high-resolution global climate model projections using machine learning (ML) techniques. A modeling framework based on an ensemble of classical ML methods that link observed lightning occurrence and convective storm indices from the ERA5 reanalysis is used to analyze and predict thunderstorm occurrence over Canada. Through training and validation across multiple decades (1999–2023), the ML models have demonstrated skill in predicting thunderstorm occurrence, even under conditions of extreme class imbalance.

Trained models are used in conjunction with HighResMIP climate model simulations to make future thunderstorm projections for various global warming levels (1°C to 4°C above pre-industrial), offering insights into the spatial and temporal evolution of convective activity in Canada. The impact of model equifinality – where distinct modelling configurations lead to similar historical predictions – on future projections of thunderstorm occurrence are explored, as are "perfect model" approaches to evaluate robustness under future climate conditions.

Bio:

Dr. Alex Cannon is a Research Scientist with the Climate Research Division of Environment and Climate Change Canada. His research focuses on climate extreme and future projections of climate at regional scales, with an emphasis on societally relevant climate variables. He is an Adjunct Professor in Atmospheric Science at the University of British Columbia and an Adjunct Professor in the School of Earth and Ocean Sciences at the University of Victoria. He is an Editor-In-Chief of Atmosphere-Ocean, an associate editor of Advances in Statistical Climatology, Meteorology and Oceanography, and a member of the editorial advisory board of Computers & Geosciences. He is a past member of the AMS Committee on Artificial Intelligence Applications to Environmental Science.