Improved cloud detection in AVHRR daytime and night-time scenes over the ocean

Abstract

Accurate cloud detection is a requirement of many geophysical applications that use visible and infrared satellite data (e.g. cloud climatologies, multichannel sea surface temperature (MCSST)). Unfortunately, a significant source of residual error in such satellite-based products is undetected cloud. Here, a new, computationally efficient cloud detection procedure for both daytime and night-time Advanced Very High Resolution Radiometer (AVHRR) data is developed. It differs substantially from our prior related work. First, a new clustering procedure is used, which produces more homogeneous and distinct clusters than those produced by either our previous work or the ISODATA algorithm of Ball and Hall. Second, the input information vector is reduced in size, incorporates both radiance and spatial components and each component is normalized. These changes improve the clustering/subsequent classification, tend to decrease execution time, and simplify post-processing of the classified (cloud, clear ocean) data to remove any residual outliers. Third, the enhanced performance makes possible the use of a multipass procedure which is very effective in identifying the complex multilayer cloud structures common in satellite data. Validation with independent lidar observations confirms the accuracy of the new procedure. Marine low stratiform clouds (LSCs-fog, stratus and stratocumulus) are also detected effectively. This advance is important because LSCs are a major source of residual cloud contamination in contemporary sea surface temperature (SST) products. Finally, the method is sufficiently general that it can be adapted to other sensors (e.g. the Along-Track Scanning Radiometer (ATSR), the Moderate Resolution Imaging Spectroradiometer (MODIS)).