An AVHRR multiple cloud-type classification package

Using imagery from NOAA's Advanced Very High Resolution Radiometer (AVHRR) orbiting sensor, one of the authors (RLB) earlier developed a probabilistic neural network cloud classifier valid over the world's maritime regions. S ince then, the authors have created a database of nearly 8000 16 x 16 pixel cloud samples (from 13 Northern Hemispheric land regions) independently cl assified by three experts. From these samples, 1605 were of sufficient qual ity to represent 11 conventional cloud types (including clear). This databa se serves as the training and testing samples for developing a classifier v alid over land. Approximately 200 features, calculated from a visible and a n infrared channel: form the basis for the computer vision analysis. Using a 1-nearest neighbor classifier, meshed with a feature selection method usi ng backward sequential selection, the authors select the fewest features th at maximize classification accuracy. In a leave-one-out test, overall class ification accuracies range from 86% to 78% for the water and land classifie rs, with accuracies at 88% or greater far general height-dependent grouping s. Details of the databases, feature selection method, and classifiers, as well as example simulations, are presented.