Historical patterns of regional extreme precipitation change in observations and earth system models

Accurate projections of future extreme precipitation are important for informing risk management and climate change adaptation. However, projections of future extreme precipitation are relatively uncertain, especially at the regional scale, in part because of uncertainty in how well extreme precipitation processes are captured by earth system models. In current research, we are analyzing different metrics of historical extreme precipitation change to determine where historical observations and earth system models agree and disagree. Simultaneously, we are developing machine learning-based pattern classification methods to determine what processes have driven historical changes in extreme precipitation. Our initial case study over the U.S. Midwest region demonstrates how neural networks can be used to distinguish between dynamic vs. thermodynamic drivers of extreme precipitation change. We are now applying this approach globally to understand the processes driving extreme precipitation change in different regions, and how these processes compare between observations and models.