Detecting cloud contamination in passive microwave satellite measurements over land

Remotely sensed brightness temperatures from passive observations in the microwave (MW) range are used to retrieve various geophysical parameters, e.g. near-surface temperature. Cloud contamination, although less of an issue at MW than at visible to infrared wavelengths, may adversely affect retrieval quality, particularly in the presence of strong cloud formation (convective towers) or precipitation. To limit errors associated with cloud contamination, we present an index derived from stand-alone MW brightness temperature observations, which measure the probability of residual cloud contamination. The method uses a statistical neural network model trained with the Global Precipitation Microwave Imager (GMI) observations and a cloud classification from Meteosat Second Generation-Spinning Enhanced Visible and Infrared Imager (MSG-SEVIRI). This index is available over land and ocean and is developed for multiple frequency ranges to be applicable to successive generations of MW imagers. The index confidence increases with the number of available frequencies and performs better over the ocean, as expected. In all cases, even for the more challenging radiometric signatures over land, the model reaches an accuracy of >= 70% in detecting contaminated observations. Finally an application of this index is shown that eliminates grid cells unsuitable for land surface temperature estimation.