Spatiotemporal Classification of Short-Duration Heavy Rainfall in Beijing Using K-Shape Clustering

Abstract

Understanding the spatiotemporal dynamics of short-duration heavy rainfall (SDHR) is critical for urban flood management. This study applies the K-shape clustering algorithm to classify 105 SDHR events in Beijing (2009–2021) using hourly rainfall data. Compared to K-means and DTW, K-shape prioritizes temporal shape alignment, crucial for capturing phase-shifted rainfall patterns. Three clusters emerged: (1) localized moderate-intensity events (13.3% of events) peaking at noon (11:00–14:00 LST) in western/southeastern regions, with weak burstiness (44.3% stations peak within 0–1 h) and moderate spatial variability (Cv = 1.08); (2) highly variable, intense urban rainfall (47.6% of events) characterized by rapid burstiness (72.5% stations peak within 0–1 h) and extreme spatial heterogeneity (Cv = 1.21), concentrated in central urban areas with peak intensities >130 mm/h; (3) prolonged heavy rainfall (39.1% of events) lasting >6 h, featuring significant accumulation (mean > 50 mm/day) in northeastern plains. The framework identifies high-risk zones (e.g., Cluster 2’s urban flash floods) and informs adaptive drainage design (e.g., prolonged resilience for Cluster 3). This study highlights the necessity of combining statistical metrics with domain expertise for robust SDHR classification and provides insights for urban flood management, emphasizing targeted strategies for different rainfall patterns.