- This event has passed.
Jose González-Abad – Harnessing Machine Learning for Climate Modeling: Towards the Development of a DestinE Climate Emulator
Bio: Jose González-Abad is a postdoctoral researcher at Instituto de Física de Cantabria (CSIC-Universidad de Cantabria) specializing in deep learning applications for climate science. He holds a degree in Statistics, a Master’s in Data Science, and a PhD focused on deep learning for statistical downscaling. His research centers on applying deep learning to emulate Global and Regional Climate Models. He is also interested in interpretability methods and their application to climate modeling. He has contributed to international initiatives such as CORDEX and published in various scientific journals and conferences.
Abstract: Machine learning advances offer transformative opportunities for climate modeling by enabling computationally efficient emulators of high-resolution simulations. This talk presents the development of a climate emulator for the Destination Earth (DestinE) Climate Digital Twin that aims to reproduce key climate outputs at a fraction of traditional computational costs. The approach adapts the Artificial Intelligence Forecasting System (AIFS) framework to capture essential climate properties, including realistic spectra, accurate variable covariances, seasonal cycles, and major modes of natural variability such as the Madden-Julian Oscillation and El Niño-Southern Oscillation, as well as stratospheric phenomena like the polar vortex and Quasi-Biennial Oscillation. Leveraging DestinE Climate DT data, the emulator will enable long simulations at resolutions down to 25 km and facilitate scenario interpolation. The presentation outlines the methodological approach, the challenges of adapting ML models for climate applications, and the anticipated milestones, including progress toward achieving stable multi-year rollouts and capturing key climate variability across sub-seasonal to interannual scales.
For virtual participation, please contact info@cesoc.net