STREAM-Sat: A Novel Near-Realtime Quasi-Global Satellite-Only Ensemble Precipitation Dataset

Abstract

Satellite-based precipitation observations can provide near-global coverage with high spatiotemporal resolution in near-realtime. Their utility, however, is hindered by oftentimes large uncertainties that vary substantially in space and time. This problem is particularly pronounced in regions which lack dense ground-based measurements to quantify or reduce such uncertainty. Since this uncertainty is, by definition, a random process, probabilistic representations are needed to advance their operational application. Ensemble methods, in which uncertainty is depicted via multiple realizations of precipitation fields, have been widely used in numerical weather and climate prediction, but rarely in satellite contexts. Creating such an ensemble dataset is challenging due to the complexity of observational uncertainties and the scarcity of “ground truth” to characterize them. In this study, we attempt to resolve these two challenges and propose the first quasi-global (covering all continental land masses within 50°N-50°S) satellite-only ensemble precipitation dataset (STREAM-Sat), derived entirely from NASA's Integrated Multi-SatellitE Retrievals for Global Precipitation Measurement (IMERG) and GPM's radar-radiometer combined precipitation product (2B-CMB). No ground-based measurements are used to generate STREAM-Sat, and it is suitable for near-realtime use without extending the 4-hr latency and 0.1°, 30-min spatiotemporal resolution of IMERG Early. We compare STREAM-Sat against several precipitation datasets, including global satellite-based, rain gage-based, atmospheric reanalysis, and merged products. While our proposed approach faces some limitations and is not universally superior to the comparison datasets in all respects, it does hold relative advantages due to its unique combination of accuracy, resolution, rainfall spatiotemporal structure, latency, and utility in hydrologic and hazard applications.