A Technique for Classifying Uncertain MOD35/MYD35 Pixels Through Meteosat Second Generation-Spinning Enhanced Visible and Infrared Imager Observations

This paper describes a technique that uses the information gathered from the geostationary instrumentation [Meteosat Second Generation (MSG)-Spinning Enhanced Visible and Infrared Imager (SEVIRI)] to investigate the pixels detected as ??uncertain?? by the operational Moderate Resolution Imaging Spectroradiometer (MODIS) cloud-mask algorithm. This technique analyzes the uncertain MODIS areas by using a time series of MSG-SEVIRI images taken at infrared (IR) and visible (VIS) wavelengths. In order to classify the uncertain pixels related to the granules acquired during the daytime and completely included in the high-resolution visible (HRV) image, the spectral and textural features derived from a time series of HRV images are used as inputs in a K-nearest neighbor (K-NN) classifier. For the areas not included in the HRV image and for those acquired during nighttime, the input parameters are determined from a time series of IR/VIS and IR images, respectively. The K-NN classifier detected 52.0%, 48.7%, and 37.0% of the MOD35/MYD35 uncertain pixels analyzed over land and 54.5%, 45.4%, and 49.7% of those analyzed over sea as cloud free, when using HRV, IR, and IR/VIS features as inputs, respectively. Percentages of 39.8%, 31.8%, and 37.3% of the pixels analyzed over land and 40.7%, 47.4%, and 38.0% of those analyzed over sea were classified as cloudy when using HRV, IR, and IR/VIS features as inputs, respectively. The remaining uncertain pixels were classified as low confidence cloudy or cloud free by the K-NN classifier. A set of comparisons was made with cloud-profiling radar/cloud-aerosol lidar with orthogonal polarization 2B-Geometrical Profiling-Lidar product results.