Deriving optimal weather pattern definitions for the representation of precipitation variability over India

Abstract

This study utilizes cluster analysis to produce sets of weather patterns for the Indian subcontinent. These patterns have been developed with future applications in mind; specifically relating to the occurrence of high-impact weather and meteorologically induced hazards such as landslides. The weather patterns are also suited for use within probabilistic medium- to long-range weather pattern forecasting tools driven by ensemble prediction systems. A total of 192 sets of weather patterns have been generated by varying the parameter which is clustered, the spatial domain and the number of weather patterns. Non-hierarchical k-means clustering was applied to daily 1200 UTC ERA-Interim reanalysis data between 1979 and 2016 using pressure at mean sea level (PMSL) and u- and v-component winds at 10-m, 925-hPa and 850-hPa. The resultant weather pattern sets (clusters) were analysed for their ability to represent the main climatic precipitation patterns over India using the explained variation score. Weather patterns generated using 850-hPa winds are among the most representative, with 30 patterns being enough to represent variability within different phases of the Indian climate. For example, several weather pattern variants are evident within the active monsoon, break monsoon and retreating monsoon. There are also several variants of weather patterns susceptible to western disturbances. These weather pattern variants are useful when it comes to identifying periods most susceptible to high-impact weather within a large-scale regime, such as identifying the most flood prone periods within the active monsoon. They hence have potentially many forecasting applications.