Applied Mathematics Colloquium - Daniel Ries

Daniel Ries, Principal R&D Computer Scientist, Sandia National Laboratory

Explaining climatic pathways from stratospheric aerosol injections using feature importance on echo state networks

The effects of climate change are seen regularly around the world. One proposed mitigation strategy is climate intervention through stratospheric aerosol injection (SAI), but the effects of such interventions are not well understood. Data driven models can be used to confirm and validate posited relationships using observed data. Deep learning models have become increasingly popular for climate modeling due to their ability to capture complex relationships. However, these methods typically lack interpretability, thus limiting their usefulness on understanding the impact of SAI. We develop a spatio-temporal feature importance explainability method for echo state networks, which is a computationally efficient neural network designed for temporal data. As an exemplar for an SAI, we apply our model to the 1991 volcanic eruption of Mount Pinatubo, which has been widely studied and been shown to have a significant impact on the earth’s climate. We assess our model’s ability to validate and discover climate pathways using the United States Department of Energy’s Energy Exascale Earth System Model and Modern-Era Retrospective analysis for Research and Applications, Version 2. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.