Identifying Spatio-Temporal Drivers of Extreme Events

The spatio-temporal relations of extreme events impacts and their drivers in climate data are not fully understood and there is a need of machine learning approaches to identify such spatiotemporal relations from data. The task, however, is very challenging since there are time delays between extremes and their drivers, and the spatial response of such drivers is inhomogeneous. In this work, we proposed a first approach and benchmarks to tackle this challenge. Our approach was trained end-to-end to predict spatio-temporally extremes and spatio-temporally drivers in the physical input variables jointly. We assumed that there exist precursor drivers, primarily as anomalies in assimilated land surface and atmospheric data, for every observable impact of extremes. By enforcing the network to predict extremes from spatio-temporal binary masks of identified drivers, the network successfully identified drivers that are correlated with extremes. We evaluated our approach on three newly created synthetic benchmarks where two of them are based on remote sensing or reanalysis climate data and on two real-world reanalysis datasets. Our project leveraged Marvin’s powerful GPU infrastructure to develop the deep neural networks and to do the parallel multi-GPU training on massive climate datasets. For the real-world climate data, we trained the model on the scalable multi-GPUs A100 80GB (SGPU partition) and for the synthetic data, we trained on the (MLGPU partition) with A40 48GB GPUs.

Website: https://hakamshams.github.io/IDE/

Paper: https://proceedings.neurips.cc/paper_files/paper/2024/hash/aa7259c82d642e47d5661f3218cdcad2-Abstract-Conference.html

Poster:  https://github.com/HakamShams/IDEE/blob/main/docs/poster/Shams_Gall.png