Statistical Error Decomposition of Regional-Scale Climatological Precipitation Estimates from the Tropical Rainfall Measuring Mission (TRMM)

Abstract

Abstract Monthly rainfall estimates inferred from the NASA Tropical Rainfall Measuring Mission (TRMM) satellite contain errors due to discrete temporal sampling and remote spaceborne rain retrievals. This paper develops a regional-scale error model that uses the rain information in the ground data to disentangle the sampling and retrieval errors in the satellite estimate statistically. The proposed method computes a mean rain rate from monthly rainfall statistics for each TRMM rain sensor by subsampling high-resolution ground-based rain data at satellite overpass times. This additional rain-subsampled parameter plays an essential role in the statistical decomposition of the total error distribution into its sampling and retrieval error components. Using the statistical formalism developed in this study, an error analysis was performed on 5 yr of monthly rain estimates produced by the TRMM Microwave Imager (TMI) and precipitation radar (PR) rain sensors aboard TRMM over a quasi 2° × 2° region of the TRMM ground validation (GV) site at Melbourne, Florida. Annual retrieval and sampling error statistics were computed for the TMI and PR using monthly rainfall estimates derived from two independent ground-based sensors: a regional rain gauge network and the Next-Generation Weather Radar (NEXRAD). Subsampled ground-based rainfall estimates produced for the radar and the gauges were highly correlated with the TMI and PR rainfall estimates, and both GV sensors produced relatively consistent error estimates. The PR-to-TMI sampling error ratio was equal to about 1.3, which was in close agreement with prelaunch predications, and the TMI-to-PR retrieval error ratio was about 2.0. For the TMI, a seasonally alternating rainfall bias was also observed that was negative during winter and positive during summer.