Adequacy of using a 1/3-degree Special Sensor Microwave Imager dataset to estimate climate-scale rainfall

recently, monthly rainfall products using the National Oceanic and Atmosphe ric Administration National Environmental Satellite, Data, and Information Service Office of Research and Applications Special Sensor Microwave Imager (SSM/I) rainfall algorithm have been generated on a global 2.5 degrees X 2 .5 degrees grid. The rainfall estimates are based on a subsampled set of th e full-resolution SSM/I data, with a resulting spatial density of about one -third of what is possible at SSM/I's highest spatial resolution. The reduc tion in the spatial resolution was introduced in 1992 as a compromise dicta ted by data processing capabilities. Currently, daily SSM/I data processing at full resolution has been established and is being operated in parallel with the subsampled set. Reprocessing of the entire SSM/I time series based on the full-resolution data is plausible but requires the reprocessing of over 10 yr of retrospective data. Because the Global Precipitation Climatol ogy Project is considering the generation of a daily 1 degrees x 1 degrees rainfall product, it is important that the effects of using the reduced spa tial resolution be reexamined. In this study, error due to using the reduced-resolution versus the full-re solution SSM/I data in the gridded products at 2.5 degrees and 1 degrees gr id sizes is examined. The estimates are based on statistics from radar-deri ved rain data and from SSM/I data taken over the Tropical Ocean and Global Atmosphere Coupled Ocean-Atmosphere Response Experiment (TOGA COARE) radar site. SSM/I data at full resolution were assumed to provide rain estimates with 12.5-km spacing. Subsampling with spacings of 25, 37.5 (which correspo nds to the present situation of 1/3 degrees latitude-longitude spatial reso lution). and 50 km were considered. For the instantaneous 2.5 degrees x 2.5 degrees product, the error due to subsampling, expressed as a percentage o f the gridbox mean, was estimated using radar-derived data and was 6%, 10%, and 15% at these successively poorer sampling densities. For monthly avera ged products on a 2.5 degrees X 2.5 degrees grid, it was substantially lowe r: 3%, 4%, and 7%, respectively. Subsampling errors for monthly averages on a 1 degrees X 1 degrees grid were 8%, 16%, and 23%, respectively. Estimate s based on SSM/I data at full resolution gave errors that were somewhat lar ger than those from radar-based estimates. It was concluded that a rain pro duct of monthly averages on a 1 degrees x 1 degrees grid must use the full- resolution SSM/I data. More work is needed to determine how applicable thes e estimates are to other areas of the globe with substantially different ra in statistics.