Use of Gaussian mixture model for standardized precipitation index in monsoon region

This study analyzed precipitation patterns across South Korea using daily data from 55 Automated Synoptic Observation System stations with over 30 years of records. The study region encompasses diverse climatic zones within the East Asian monsoon system, from coastal areas to mountainous regions, representing the full spectrum of Korea’s temperate monsoon climate variability. This study developed and evaluated a hybrid distribution selection framework for Standardized Precipitation Index (SPI) calculation that addresses the limitations of the conventional single-distribution approach in monsoon climates. This combined kernel density estimation for precipitation pattern detection with selective application of Gaussian Mixture Models (GMM) where multi-peak distributions are identified, compared against traditional Gamma and Weibull approaches across 3–24 month time scales. Results showed systematic multi-peak precipitation patterns at 76.4% of stations for 6-month duration and 69.1% for 24-month, reflecting the complex seasonal characteristics of the East Asian monsoon region. GMM substantially enhanced statistical fit in regions with the highest precipitation complexity (mean Δ[jls-end-space/]AIC improvements of −8 to −9), whereas SPI-15 and SPI-18 incorporating complete annual cycles, exhibited optimal performance with the Weibull distribution. This framework provides the first systematic guidance for distribution selection in monsoon regions, demonstrating that hybrid approaches combining Weibull and GMM distributions substantially improve drought probability assessment compared to conventional single-distribution methods.