An improved cloud detection algorithm for application to SST retrieval

Undetected clouds are believed to contribute significantly to the errors in retrieved sea surface temperature (SST.) The current SST cloud detection algorithm employed by NOAA on AVHRR data uses a combination of channel and channel ratio thresholds and of spatial variance tests. The algorithm parameters are tuned to provide a very low probability of passing a cloud (low error rate) at the expense of classifying many clear scenes as cloudy (high false alarm rate.) We have developed a new multispectral SST cloud detection algorithm that works on single pixel data. When compared to the algorithm used in the AVHRR Pathfinder Matchup Database, this new algorithm has a lower error rate and lower false alarm rate, providing more data points with higher accuracy. We plan to test further improvements to the algorithm, including tuning the parameters plus the inclusion of a spatial variance test, to further reduce the false alarm rate.