Intercomparison and Validation of GPCP One-Degree Daily Rainfall Estimates with Surface-based Radar Estimates in the Mississippi River Basin

The Global Precipitation Climatology Project (GPCP) has been producing truly global daily precipitation estimates at 1-degree resolution (1DD) for the period of January 1997 - to date by combining data from different satellite sensors. Quantification of the errors in such estimates, through direct comparisons with independent high-quality surface-based rainfall estimates, is essential to provide a quantitative confidence level on the estimates and to further improve the quality of such estimates. The purpose of this study is to perform extensive quantitative assessment of the GPCP-1DD estimates by comparing them with independent radar-based rainfall estimates over the Mississippi river basin, for the period 1997-2000. The archive of radar-based estimates used, conveniently referred to as the MRB hereafter, is suitable to perform validation studies in view of its high space-time resolution, large areal coverage, and relatively long time series. The Mississippi river basin covers 435 one-degree grids. The authors used several comparison statistics which fall into unconditional continuous statistics, conditional (quasi-)continuous statistics, and categorical statistics. The analyses include annual and seasonal time periods. The results are presented for each 1-degree grid and for a combination of these grids in certain regions. Highlights of the results are as follows. (1) There is a very good correspondence between the GPCP and MRB estimates both spatially and temporally, as evidenced by correlations in excess of 0.90 mostly. (2) The GPCP has a very high skill in discriminating rainy from nonrainy days as reported by the MRB estimates, at a large range of rainfall thresholds. (3) There is a systematic difference between the two estimates, with the GPCP overestimating at almost all locations and seasons. Both the bias and the root-mean-square difference consistently increase with increasing rain rate. The bias varies from 0.09 to 7.21 mm day-1, at the MRB-estimated means of 0.10 and 17.26 mm day-1, respectively. The bias-adjusted root-mean-square difference varies from 0.22 to 5.87 mm day-1, at the MRB-estimated means of 0.10 and 17.26 mm day-1, respectively. (4) The GPCP estimates for winter show slightly larger systematic error and less rain presence/absence diagnosis skill compared to other seasons. Finally, although the errors in the MRB estimates are expected to be relatively small, the authors acknowledge that the MRB estimates are subject to error sources, and they describe how the statistics of the GPCP-MRB difference are partitioned into the GPCP error and the MRB error components.