A Spatially Constrained k-means++ Approach for Multi-Disaster Regionalization in China

Abstract

The spatial zoning of disaster-prone areas in mainland China presents a compelling yet complex challenge. Despite significant strides in disaster management and risk reduction, a comprehensive and region-specific disaster zoning that considers spatial contiguity remains largely unexplored. This study seeks to address this gap by applying and comparing four clustering methods: k-means, spatially constrained k-means, k-means++, and spatially constrained k-means++. These methods were evaluated based on their ability to categorize regions by the extent of areas affected by five major disaster types: floods, droughts, low temperatures, typhoons, and hailstorms. The spatially constrained k-means++ algorithm emerged as the most effective, as it addressed spatial discontinuity inherent in disaster zoning and mitigated the initial value problem linked to traditional k-means methods. Using this approach, mainland China was divided into four distinct disaster-prone clusters. Cluster 1, encompassing 25 provinces, exhibited a complex and overlapping hazard profile, while Clusters 2 and 3 (Hebei–Jilin–Xinjiang and Jiangxi–Hubei, respectively) reflected more specific regional disaster patterns. Hainan formed an independent cluster because of its unique typhoon dominance. These spatially coherent zoning results provide a robust foundation for developing differentiated disaster management strategies, from integrated approaches in multi-disaster regions to specialized interventions in areas facing dominant threats.