Vertical profiling of precipitation using passive microwave observations : the main impediment and a proposed solution

Several methods have been proposed to train microwave radiometers to retrieve precipitation rates estimated by a radar which observed the same location at the same time. These radar-trained passive-microwave algorithms differ in the quantities that are estimated: some estimate the vertically-integrated liquid water, while others estimate the near-surface precipitation. Since it is no more or less credible to estimate the rain rate at the surface than it is to estimate the rain rate at any discrete altitude, it is particularly interesting to quantify to what extent it is indeed feasible to estimate vertical profiles of precipitation from a passive microwave radiometer, what the obstacles are, and what vertical resolution would be achievable. To that end, we selected five study regions, and started by quantifying the vertical variability of rainfall as derived from the Tropical Rainfall Measuring Mission (TRMM) radar. Two cases emerged: a monsoon-like case where the first principal component of the vertical precipitation accounts for about 90% of the variability, and a Mediterranean-like case where the first principal component accounts for about 80% of the variability. A Bayesian approach was applied to the TRMM Microwave Imager measurements co-located with the radar profiles. For the monsoon-like regions, it produced estimates of rain rates at 250 meter vertical increments which compared well with the TRMM radar estimates. For the Mediterranean-like regions, the retrieval errors were very large. We therefore proceeded to identify the main reason for the failure of the straightforward training method. It turns out to be the unknown signature of the sea surface in the portion of the beam that does not contain precipitation. In the problematic Mediterranean case, our original straightforward approach can still be applied to measurements which do not suffer from this identifiable partial-beam-filling. For measurements that do, we derive a filtering approach to neutralize the variability of the partial surface signature and thus overcome the problem.