This study aims to analyze time-series measurements encompassing rainstorm events with over a century of datasets to identify rainstorm evolution and dimensional transitions in non-stationarity. Rainstorm events are identified using partial duration series (PDS) to extract changes in rainstorm characteristics, namely maximum intensity (MAXI), duration (D), total rainfall (TR), and average rainfall intensity (ARI), in response to climate change. Ensemble empirical mode decomposition is used for trend filtering and non-stationary identification to explore spatiotemporal insight patterns. Trend models for the first-second-order moments of rainstorm characteristics are used to formulate the identified mean-variance trends using combined multi-dimensional linear-parabolic regression. Best-fitting combinations of various distributions (probability density functions) and trend models for multiple characteristic series are identified based on the Akaike information criterion. We analyze the dimensional transition in rainfall non-stationarity based on sensitivity analysis using PDS to determine its natural geophysical causes. The integrated methodology was applied to the data retrieved from nine weather stations in Taiwan. Our findings reveal rainstorm characteristics of "short D but high rainfall intensity" or "lower MAXI but high TR" across multiple stations. The parabolic trend of the first-order moment (i.e., mean) of ARI, D, and TR appears at the endpoint of the mountain ranges. Areas receiving monsoons and those on the windward plain show a rising parabolic trend in the first- and second-order moments (i.e., mean-variance) characterizing MAXI, implying that the occurrence frequency and magnitude of extreme MAXI increases. Non-stationary transitions in MAXI appear for mountain ranges exposed to the monsoon co-movement effect on both windward and leeward sides. Stations in the plains and rift valleys show upgraded and downgraded transitions in the non-stationary dimensions for D, respectively.
© 2024. The Author(s).