Experience gained through many rainfall algorithm intercomparison projects has made it clear that there are no suitable ground truth data for validating oceanic rainfall. However, since the retrieval algorithms are based on first principles models, the net uncertainty can be estimated by estimating the uncertainty of the various assumptions in the models. Further, in the case of Level-3 products, the random component of the uncertainty can be measured by subdividing the data appropriately.

We have developed a Level-3 Oceanic rainfall algorithm that can be applied to sensors such as TMI and AMSR-E with window channels at frequencies below the 60 GHz Oxygen band. The philosophy of the algorithm is based on uncertainty estimates for the various components. We have models for the uncertainty introduced by Drop Size Distribution, inhomogeniety of the rainfall over the FOV (i.e. beam filling), and instrument calibration. Having the Precipitation Radar on the satellite has been especially useful for refining the error models. It has been used to refine the freezing level retrieval and the models used for the Drop Size Distribution and Beam Filling. It has provided a basis for uncertainty estimates in all three cases.

The algorithm solves for a zero rain offset which absorbs errors due to calibration at the cold end, surface emissivity, average wind speed and average non-raining cloud . The uncertainty of this offset can be measured and included in the uncertainty calculations. The algorithm produces a weighted average of the rain rates yielded by 6 separate channels of TMI or AMSR-E. The weights are based on the computed uncertainties of each rain rate. With this algorithm we have produced maps of monthly average oceanic rain rates over 5 degree boxes and the associated uncertainties.