Understanding the Intricacies of Model Biases in Storm-Related Extreme Precipitation Using CMIP6

Abstract

This paper evaluates the biases, defined as deviations from ERA5, in extreme precipitation linked to tropical cyclones (TCs), extratropical cyclones (ETCs), and atmospheric rivers (ARs) in Coupled Model Intercomparison Project Phase 6 (CMIP6) models during 1985–2014. Overall, the multi-model ensemble significantly overestimates the intensity of storm-associated extreme precipitation, with global average biases of 19.1% for TCs, 11.4% for ETCs, and 12.2% for ARs. This corresponds to at least a 57% higher risk of storm-associated extreme precipitation above the 99.9th percentile in the models. The skill score of the CMIP6 multi-model ensemble mean ranks in the mid-range among the models, while high-resolution models typically achieve higher scores. Composite rainfall structure analysis indicates that, although CMIP6 models roughly reproduce the overall structure of extreme precipitation associated with the three storm types, they display significant wet biases with pronounced regional features. These regional differences are primarily influenced by flawed dynamical processes, while moisture conditions play a secondary role.