Machine Learning for Climate and Environmental Sciences: Advances and Challenges
Vast amounts of observational and model output data are available for a variety of climate and environmental science problems. With the tremendous advances in machine learning (ML) especially over the past decade, the hope is that ML methods will be able help make advances on these key scientific problems. In this talk, I will share some progress we have made over the past decade in applying ML methods to problems in climate and environmental sciences, including terrestrial ecosystem modeling towards improving land surface models, multi-task learning for improving climate multi-model ensembles, and sub-seasonal climate forecasting. I will also discuss some key challenges in applying modern ML methods to such scientific problems. Finally, I will briefly discuss the ongoing TorchGeo project, which is combining the power of modern ML with geospatial data.