Snow cover classification using millimeter-wave radar imagery

Millimeter-wave radar imagery at 95 and 215 GHz shows great promise in discriminating between dry, wet, and refrozen snow cover on the ground. A study was carried out to explore the feasibility of simulated millimeter-wave radar images to classify various snow types, and to predict the percent extent of each type. A simple empirical model derived from actual backscatter measurements was used to compute the backscatter coefficient as a function of frequency, incidence angle, surface roughness and (wet snow) moisture content. Copolarized and crosspolarized images of 100 pixels × 100 pixels extent of 1 meter resolution were generated. The relevant spatial statistics and probability distributions of snow surface roughness and moisture content were estimated based on ground truth data gathered during various snow backscatter measurement programs. Images were generated for various combinations of dry/wet/refrozen snow percentages within the overall image. Using a training set of 20 pixels × 20 pixels for each snow cover type, a Bayesian classifier was employed to classify each pixel based on the mean and standard deviation of the images of the training sets. Overall mis-classification errors of less than 12% were achieved, although in most cases, the errors were less than 8% especially when substantial amounts of wet snow was present.