How well does the IMERG satellite precipitation product capture the timing of precipitation events?

Abstract

Precipitation occurs in the form of discrete “events” and the event characteristics (event duration, depth, peak intensity, start/end time) significantly influence the hydrologic response of a basin. Despite this importance, event-based performance of satellite precipitation products has still not been fully investigated to assess limitations in the retrieval algorithms, guide future improvements, and inform hydrologic applications. In this study, we evaluate the precipitation event performance of the GPM IMERG product using as reference the high-resolution ground gauge–radar dataset (GV-MRMS) in the Continental United States (CONUS) at the native IMERG resolution (0.1°×0.1°, 0.5 h), with a primary focus on the detectability and timing of events. Our results show that IMERG generally overestimates the event duration but underestimates the mean event precipitation intensity in the summer, while the opposite is true for winter. This discrepancy is mostly attributed to the under-representation of short-duration intense events in the summer and long-duration moderate events in the winter in IMERG. In terms of the detection of individual events, about 50% of the reference events are properly detected by IMERG, and conversely, 50% of IMERG events do not match a reference event. However, nearly 40% of the missed or false events result from temporal mismatching of less than 3 h between the retrieved and the reference event. The remaining 60% comes from IMERG not detecting an existing event or inventing a nonexistent event. When IMERG successfully detects an event, the average temporal overlap with the reference event is about 70% of its total duration, which mostly stems from the mistiming of IMERG-derived events. IMERG events tend to start, peak, and end earlier than GV-MRMS events, with national mean shifts of −26 min, −17 min, and −7min, respectively. For about only 20% of all situations the starting time of the event is correctly reproduced by IMERG, and the same applies to the peak time and end time. Our results provide guidance for applications of IMERG at sub-daily scales, as well as new insights for the improvement of satellite retrieval algorithms.