Regionalization, Decomposition, and Information Mining for Precipitation: A Novel Framework Based on Multivariate SOM Clustering and Multiscale Features.

Precipitation regionalization, serving as a foundation for extrapolating information from gauged to ungauged sites, contributes to a comprehensive understanding of the spatial distribution of precipitation. However, existing studies have focused mainly on precipitation, neglecting the influence of climate background and meteorological circulation. Moreover, there is a lack of specialized analysis of regionalization results. This study proposes a novel approach to precipitation regionalization that considers covariates in the circulation field and periodic features. The method utilizes Barnett–Preisendorfer canonical correlation analysis coupled with principal component analysis (BPCCA) to select covariates and multivariate self-organizing map (SOM) clustering for preliminary regionalization. Wavelet decomposition is further used for regional feature analysis. The methodology is empirically applied to analyze summer precipitation in Shanxi Province, identifying homogeneous regions characterized by diverse spatiotemporal distributions. The results successfully identified 12 distinct regions of precipitation, effectively capturing the influence of topographic and atmospheric factors. According to the periodic and trend characteristics of each region at different time scales, we merged the division results on the three frequencies and six regions were ultimately differentiated. Specifically, a decreasing trend was observed in the southern and southeastern parts of Shanxi, as well as in the western part of the Lüliang Mountains. They had a significant 4-yr period mode, and the decreasing trend was more significant in the southern region. In contrast, northern Shanxi showed no trend and a significant 8-yr period mode. This proposed method presents an effective strategy for enhancing precipitation regionalization and extracting valuable information from the circulation field and multiple frequencies. [ABSTRACT FROM AUTHOR]