Global Near-Real-Time Precipitation by Optimally Merging Gauge, Satellite, and Atmospheric Model Data

Accurate and timely precipitation data are essential for monitoring and forecasting of floods and droughts. Yet, a fully global near-real-time (NRT) product that simultaneously provides good performance in convection- and snow-dominated regions as well as in densely gauged regions is still lacking. We developed a NRT variant of the recently released historic Multi-Source Weighted-Ensemble Precipitation (MSWEP) dataset. MSWEP-NRT provides gap-free, fully global precipitation estimates with three-hourly temporal and 0.1° spatial resolution, by optimally merging gauge observations (GHCN-D and GSOD), satellite estimates (CMORPH, GSMaP, IMERG, and TMPA 3B42RT), and atmospheric model outputs (GDAS-Anl and JRA-55). To produce MSWEP-NRT, we first merge the satellite- and model-based data using weighted averaging, with weights based on the performance of the respective data sources at surrounding gauges. Next, the merged data are cumulative distribution function matched to the historic MSWEP dataset, resulting in a consistent precipitation record from 1979 until the near present. Finally, daily gauge corrections are applied, using a novel scheme that accounts for differences in the UTC boundary of the 24-hour accumulation period of gauge reports. To account for latency differences among the data sources and potential disruptions in data availability, MSWEP-NRT estimates less than seven days old are progressively upgraded to include any new data that have become available. The product has a latency of approximately four hours. An independent evaluation using NRT gauge observations demonstrates the efficacy of our merging approach.