Development of a fine-scale North American precipitation analysis for retrospective and operational applications

Accurate, reliable, and fine-scale precipitation data are essential for land surface and hydrologic modeling. Such data is critical for many applications, including drought and flood monitoring and assessments, water resource management, and hydroclimatic research. Unfortunately, the meteorology and hydrology communities currently lack a high spatio-temporal resolution (~hourly at cloud-resolving ~1 km spatial scales) gridded precipitation analysis over North America with both: 1) a long, consistent retrospective archive; and 2) real-time availability for operations. Existing products are available at coarse spatial resolutions limiting their use for local applications. They also suffer from inconsistencies in their data records due to factors such as observational gaps and discontinuities in gauge datasets and changes in modeling systems. To overcome these limitations, we developed a fine-scale (1 km) precipitation analysis in both retrospective and near real-time over North and Central America, including Alaska, Hawaii, and Puerto Rico, by leveraging high-quality gauge, satellite, and model datasets through advanced data assimilation methods. Specifically, we employed the following precipitation datasets: 1) daily precipitation gauge observations over North America; 2) the NASA Integrated Multi-SatellitE Retrievals for GPM (IMERG) half-hourly retrospective and real-time precipitation retrievals; 3) data from the NASA Modern Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) and the NASA Goddard Earth Observing System Forward Processing for Instrument Teams (GEOS FP-IT); and 4) ECCC Canadian Precipitation Analysis (CaPA) retrospective and real-time precipitation analyses. We assimilated daily precipitation amounts at 4 km of IMERG and CaPA into the MERRA-2 (GEOS FP-IT for real-time analyses) precipitation using an optimal interpolation technique based on residual kriging and the Bratseth algorithm. Then the precipitation analysis at 4 km was (1) downscaled to a fine-scale resolution using a deep learning super resolution algorithm within the cloud based analytical framework for precipitation research (CAPRI) and (2) disaggregated into hourly using IMERG and MERRA2 datasets. The robustness of the developed precipitation analysis was demonstrated by comparisons against ground measurements and other remote sensing products such as the NOAA Stage IV precipitation, the Parameter elevation Regression on Independent Slopes Model PRISM, and the North American Land Data Assimilation System phase 2 NLDAS-2 using the extended triple collocation methods which allows computing the unknown errors associated with datasets without a reference dataset.