Weather hazard interpretation and forecast by an airborne radar

A method for modeling and tracking convective clouds within radar images is presented. An object modeling approach is used, based on the extraction of grayscale skeletons from 2-dimensionnal cross-section of 3-dimensional radar data. Grayscale skeletons are appropriate shape descriptors for nonrigid and heterogeneous objects, in which gray-level local maxima correspond to regions of interest. Linking meta-data to some chosen points of the skeleton enhances the modeling scheme; this provides a good representation of the weather scene in terms of hazards for an aircraft. Skeletons are stored within a graph structure and tracked among successive pictures by means of relaxation labeling processes. The deduced advection field is used to nowcast the clouds evolution. Preliminary results are satisfying concerning advection forecast.