Characterizing TRMM Multi-satellite Precipitation Analysis (MPA) quality at multiple time and space scales using surface data

Over the past several years the Multi-satellite Precipitation Analysis (MPA) system was developed and implemented to create quasi-global estimates of precipitation on a 3-hourly 0.25°x0.25° latitude/longitude grid. The MPA uses precipitation estimates computed for the full complement of satellite-based precipitation sensors, as well as rain gauge analyses in the post-real-time implementation of the MPA. The operational non-real-time version of the MPA is being used to compute the Version 6 TRMM 3B42 3-hour and the 3B43 monthly products. The latest release of the real-time version of the MPA is used to compute estimates that are posted to the World-Wide Web about 10 hours after observation time. The latter lack rain gauge input due to the sparsity of global real-time gauge data. By the time of the conference the non-real-time products will have an 8-year archive, while the real-time will have 4-year archive, although only the last year of real-time uses all of the passive low-orbit microwave satellite data.

It is important to characterize the performance of each MPA product across the range of global climate zones for which the product provides estimates, both as a tool for improving the products and as information for data users. In this first evaluation, the MPA products are compared to a selection of tropical surface data sets, including rain gauges located on Pacific Ocean atolls, siphon rain gauges mounted on Atlas II buoys, and TRMM GV radar estimates (Melbourne, Florida and Kwajalein Island). The goal is to perform the comparisons on the 0.25°, 1°, and 2.5° scales for monthly, daily, and 3-hourly estimates. However, the attributes of the comparison data sets impose limitations. Specifically, the atoll data are only available as daily accumulations, while both the atoll and buoy data sets are sparse enough that individual gauges must be compared to the gridded MPA values. Similar comparisons are performed using the GPCP and Version 5 TRMM estimates to further assess the relative quality of the MPA. Interestingly, for the period 1998 into 2005 the Version 6 3B43 product averaged to monthly 2.5° values displays a –12% bias against all atolls. This value is very close to the corresponding GPCP value for the same time period and considerably more negative than the Version 5 3B43 (albeit for a somewhat shorter record). One goal of these comparisons is to assess the utility of the three data sources, none of which have been exploited previously to the extent that this study attempts.

In addition, some results from parallel comparisons will be summarized, particularly the continental-scale evaluations being carried out in Australia, the United States, and Western Europe under the International Precipitation Working Group.