Abstract: MAPVEG performs a first level classification of NASA/JPL AIRSAR data into simple ground cover types. Classification images generated by MAPVEG can be used to separate vegetated from non-vegetated areas, which may be useful in determining the soil moisture in areas free of vegetation. MAPVEG classification can separate forests from the surrounding vegetation, and can be used to identify clear-cut ... areas, for example. Urban areas and potential marshlands are examples of other identifiable classes. MAPVEG images can be used as a first stage in a classification scheme more specific to a user's application. For example, having classified an area as low vegetation in an agricultural region, one might wish to go on to separate the area into different crop types.

The classification algorithm is based on a model fit to the AIRSAR data which provides estimates of the contributions of surface, double-bounce, and volume scatter to the observed polarimetric radar signatures for each of the three wavelengths measured by the AIRSAR. After the model fit has been applied to the data, simple rules (established by training on different ground cover signatures in just three AIRSAR images) are used to classify the results into one of several different ground cover types. The algorithm used is capable of distinguishing between areas with no vegetation cover, areas with low- and medium-level vegetation cover, forested areas, and urban areas. The three vegetated classes of ground cover are further sub-divided into areas where double-bounce scatter is significant and areas where it is not.

The output from MAPVEG consists of a byte image with each class assigned to a different numerical value plus a header which provides information on the image dimensions, a key to the numerical values, and a listing of the percentage of the image assigned to each class. The image is typically much smaller than the original AIRSAR data (e.g. a 330K image from 39Mb of typical AIRSAR data) and can be easily displayed on a PC or Macintosh using standard image processing software tools.