AN APPLICATION OF FRACTAL BOX DIMENSION TO THE RECOGNITION OF MESOSCALE CLOUD PATTERNS IN INFRARED SATELLITE IMAGES

Mesoscale cloud patterns are analyzed through the application of fract al box dimensions. Verification of fractal properties in satellite inf rared images is carried out by computing box dimensions with two diffe rent methods and by computing the fraction of cloudy pixels for two se ts of images: 174 are considered the ''control series,'' and 178 (for verification) are considered the ''test series.'' The main instabiliti es in the behavior of such dimensions are investigated from the simula tion of circles filling space in several spatial distributions. It is shown that the box dimensions are sensitive to the increase of the are a covered and to the spatial organization-that is, the number of cells , the spatial clustering, and the isotropy of the distribution of pixe ls. From a principal components analysis, the authors find six main pa tterns in the cloudiness for the control series. The three main patter ns related to enhanced convection are the massive noncircular spread c loudiness, the highly isotropic distribution of cloud in several cells , and the most circular pattern associated with mesoscale convective c omplexes. The six patterns are separated into a cluster analysis, and the properties of each cluster are averaged and verified for the test series. This method is a simple and skillful procedure to recognize me soscale cloud patterns in satellite infrared images.