Unraveling the spatial heterogeneity of pre-summer extreme rainfall regimes in South China using multi-dimensional joint clustering

Abstract

Abstract The temporal evolutions of extreme rainfall (ER) events are highly dynamic. In addition, the destructive potential of ERs is fundamentally governed by multi-dimensional metrics, including rainfall accumulation, event duration, and intra-event rain-rate evolution. However, classical ER clustering methods are based on one-dimensional metrics, therefore suffering from critical limitations and one-sidedness in characterizing ER behaviors. In this work, a novel three-stage multi-dimensional clustering framework based on intra-event rainfall dynamics is proposed to accurately capture the characteristics of regional ER behaviors. This framework identifies three distinct pre-summer ER regimes in South China (SC). Cat-1 ER occurred in mountainous northeastern SC is predominantly contributed by prolonged-duration events with flat intra-event variations and low rainfall intensities. These events are typically associated with strong synoptic forcing. Cat-3 ER concentrated in coastal areas are dominated by short-duration events with a steep intra-event evolution, intense rain rate, and mid-event peaks. These events are likely related to warm-sector rainfall under weak synoptic forcing. Cat-2 ER centered at northwestern SC and coastal areas displays a mixed characteristics of Cat 1 and Cat 3, featuring both synoptic-driven and warm-sector mechanisms. As an extension and complement to traditional ER clustering methods, this framework provides a physically interpretable, objectively categorized spatial map of the pre-summer ER morphology in SC, advancing our understanding of the heterogeneity and synoptic drivers underlying major ER regimes, and offering critical scientific references for targeted flood mitigation and early warnings in the region.