Clustering Quantile Regression-Based Drought Trends in Taiwan

Drought is a normal, recurring climatic feature and occurs in all climatic zones. Imbalanced water availability induced by droughts has far-reaching and adverse impacts both on human lives and natural environments. This study aims to summarize temporal and spatial drought variations in Taiwan by combining quantile regression and cluster analysis. Three-monthly rainfall series covering the 1947–2012 period for 12 rainfall stations are used in this study. Quantile regression is applied to 3-month SPI, drought duration, drought severity, and drought frequency series for exploring temporal drought trends at different quantiles. Various quantile slopes for these 12 stations are then analyzed by hierarchical agglomerative clustering algorithm to detect regional variation patterns. The results show considerable spatial diversity over Taiwan. Stations along east coast are prone to more severity due to declined SPI trends associated with increasing drought duration and severity. Positive SPI slope associated with decreasing drought duration and severity are noted at stations located in the west and lead to lessened droughts. However, temporal variations in drought-duration and drought-severity series are insignificant at most quantiles and stations. In addition, a distinct behavior is found in drought frequency since severe droughts may not accompany frequent droughts.