Separating Climate Signals with Machine Learning

Abstract

Focal AreaWe use machine learning (ML) for the attribution of seasonal and extreme precipitation to different signals in the climate system. We compare ML-based attribution methods to traditional detection and attribution (D&A) techniques. Our project aligns with Focal Area 3 of the AI for Earth System Predictability Call for White Papers: "insight gleaned from complex data (both observed and simulated)." Science ChallengeTo better prepare for future changes in the water cycle, it is vital to understand the regional effects of anthropogenic climate change and internal climate variability on precipitation. With existing methods, there is uncertainty regarding the regional patterns of these climate processes. We propose using machine learning to decompose seasonal total and extreme precipitation (calculated by Risser et. al., 2019) into components caused by each of the following signals: anthropogenic forcing, internal variability, and modes of natural climate variability (such as El Niño/ Southern Oscillation, Atlantic Multidecadal Oscillation, and the Pacific North American teleconnection).