Using Machine Learning to Analyze Physical Causes of Climate Change: A Case Study of U.S. Midwest Extreme Precipitation

While global warming has generally increased the occurrence of extreme precipitation, the physical mechanisms by which climate change alters regional and local precipitation extremes remain uncertain, with debate about the role of changes in the atmospheric circulation. We use a convolutional neural network (CNN) to analyze large-scale circulation patterns associated with U.S. Midwest extreme precipitation. The CNN correctly identifies 91% of observed precipitation extremes based on daily sea level pressure and 500-hPa geopotential height anomalies. There is evidence of increasing frequency of extreme precipitation circulation patterns (EPCPs) over the past two decades, although frequency changes are insignificant over the past four decades. Additionally, we find that moisture transport and precipitation intensity during EPCPs have increased. Our approach, which uses deep learning visualization to understand how the CNN predicts EPCPs, advances machine learning as a tool for providing insight into physical causes of changing extremes, potentially reducing uncertainty in future projections.