Can We Build Useful Statistical Model for Extreme Mei-yu Rainfall Prediction Based on the Causal Relations with Known Large-scale Modes?

In 2020, Mei-yu rainfall around the Yangtze River Valley region reached a historical high, causing many casualties and severe economic losses. Due to the significant impact of extreme Mei-yu rainfall, the accurate prediction of extreme Mei-yu rainfall plays an important role in decision-making and policy development for disaster risk reduction and mitigation strategies. Despite many advancements that have been made in climate modelling, models often struggled to simulate extreme rainfall. This is typically related to their coarse resolution and inability to fully represent necessary processes on scales relevant to extreme precipitation. On the other hand, climate models are able to simulate much better large-scale modes of variability. Given that extreme Mei-yu rainfall is known to be controlled by various large-scale modes such as the Western North Pacific Subtropical High (WNPSH) and the South Asian High (SAH), it may be possible to construct statistical models using the indices of large-scale modes to predict extreme Mei-yu rainfall using their causal relationship. In this presentation, we demonstrate that useful causality-guided statistical models can be constructed, which predict extreme Mei-yu rainfall, using indices of known large-scale modes. In particular, we highlight how causality-guided statistical models can be constructed using high temporal resolution predictors that have higher performance than their low temporal resolution counterparts. This is related to the ability to capture the necessary variabilities in modulating extreme Mei-yu rainfall. Furthermore, this approach can be used to identify spatially consistent, key large-scale modes that control regional extreme Mei-yu rainfall. The implication and potential use of this approach is also discussed.