Bayesian Algorithm for Rain Detection in Ku-Band Scatterometer Data

Abstract

Ku-band scatterometers are sensitive to rain effects due to their cm-scale radar wavelength. The NSCAT-4DS geophysical model function (GMF) corrects for sea surface temperature (SST), whereas it doesn’t consider rain. Rain causes biases in the retrieved wind fields and to prevent these, quality control (QC) flags play an important role in rain identification. Since horizontal polarization and vertical polarization radar beams have a particular sensitivity to rain clouds, a noticeable difference between the rain-dominated backscatter distribution and the wind-dominated backscatter distribution is observed. Employing a Bayesian approach and exploiting these particular wind and rain backscatter characteristics, the authors propose an algorithm to provide the posterior rain probability for each measurement in a Wind Vector Cell and test the method for the Haiyang-2C scatterometer. In a comprehensive comparison between posterior rain probability, KNMI QC flag and Joss flag, for posterior rain probabilities higher than 0.5, the rejection rate is approximately a quarter of that of the KNMI QC flag with better rain detection behavior. While the Joss flag, the difference between the retrieved wind speed and the two-dimensional variational ambiguity removal analysis wind speed, has the best performance in identifying rain in the sweet swath, it comes at the cost of a higher missing rate. The comparison with ASCAT winds also proves the method’s effectiveness. Posterior rain probability has the best rain identification ability in the nadir swath. A combination of different QC flags should be beneficial and applied in wind retrieval.