Uncertainty Estimates for Retrievals of Oceanic Rainfall from AMSR-E

The paucity of suitable ground truth for oceanic rainfall retrievals forces the use of physically based algorithms. While there are still many advantages to physically based algorithms, we are still left with the problem of validation. Building an error model is the only approach left for meaningful uncertainty estimates for the retrievals. The rainfall retrieval community has considered the primary sources of error and now preliminary estimates are available for the most important ones and many of the less important ones as well. These errors depend on many factors including the rainfall intensity, the freezing level, the frequency, view angle and polarization of the observations, as well as instrument performance and satellite geometry. These dependencies are too complex to permit a simple error estimate. The error model must be integrated with the retrieval algorithm. This has been done for rainfall estimates from AMSR-E. Although there is still a need to refine the individual contributions to the total error, we have adequate understanding to permit a reasonable estimate of the net uncertainty of the rain rate for individual pixels (Level-2) and for 5° x 5° x 1 month boxes (Level-3).