A new automated method of cloud masking for advanced very high resolution radiometer full‐resolution data over the ocean

Abstract

This paper reports on a new algorithm to remove cloud‐contaminated pixels from daytime and nighttime 1‐km advanced very high resolution radiometer (AVHRR) data. The technique was developed in response to Navy needs to efficiently and accurately eliminate cloud contaminated pixels from real‐time satellite digital images. The remaining “cloud free” sea surface temperature (SST) pixels would be available for analysts to utilize in tracking ocean mesoscale fronts and eddies as well as input to SST and ocean thermal analysis. Initial poor results with existing cloud‐masking techniques led to this effort. The method uses a series of approaches to locate cloud‐contaminated pixels which include (1) use of multiple bands to detect signatures not readily available from single‐channel data, (2) extraction of cloud edge information through local segmentation of the image using the cluster shade texture measure based on the gray level cooccurrence (GLC) matrix and, (3) discrimination of cloud‐free from cloud‐contaminated regions with an area labeling procedure. This technique is evaluated on identical data sets utilizing other experimental and operational cloud algorithms and cloud masks produced through human interpretation. This method, tested over a wide range of conditions and geographical locations, produces accurate and efficient daytime and nighttime SST data sets.