Frequency analysis of extreme rainfall over the Japanese archipelago by leveraging gauge-adjusted radar and satellite estimates

Abstract

With the increasing frequency of extreme rainfall and difficulty in capturing their spatial variability by rain gauges, especially over regions with complex topographic and climatic features like Japan, there is a growing interest in using radar and satellite-derived precipitation estimates for building intensity–duration–frequency (IDF) curves. Such a statistical method provides references for studying extreme rainfall statistics at varying spatiotemporal scales. Yet, a comparative evaluation of the efficiency of these datasets to derive reliable IDFs over the Japanese archipelago is lacking. This study compares IDFs developed from four rain gauge-adjusted remotely sensed datasets (RSDs) derived from weather radars (R/A) and satellites (GsMAP_G, CMORPH_G, and IMERG_G) with those created from long-term rain gauge records (i.e., reference IDFs) for evaluating their reliabilities, particularly for 10- and 100-year return periods. A Bayesian Generalized Extreme Value model was adopted to estimate the IDFs. The results show that the four RSDs exhibited varying performances in deriving reliable IDFs within the 5–95% confidence bounds of the reference IDFs depending on rainfall duration and climate settings. While R/A performed better for one-hour rainfall return levels, particularly in the cold climate with warm summers, satellite-driven RSDs, especially IMERG_G, showed better performances for long-duration rainfall return levels. The performance discrepancies were primarily linked to the differences in rainfall estimation and bias adjustment schemes rather than the differences in record lengths and spatial resolutions among the tested RSDs. Thus, it is crucial to consider the rainfall duration and climate settings when using the derived IDFs for diverse practical applications.