Identification and filtering of rainfall and ground radar echoes using textural features

This study deals with the identification of precipitation and ground echoes in radar images using their textural features. The images were collected by meteorological radars in the regions of Setif (Algeria) and Bordeaux (France). Two kinds of texture-based techniques have been considered, consisting in calculating either the histograms of grey levels or the histograms of their sum and difference. Hence, the first-order probability distributions were found to be sufficient to account for the textural features of radar images. Energy is found to be the textural parameter that clearly distinguishes between precipitation and ground echoes. With both methods, fixed ground echoes and anaprops are efficiently rejected, whereas precipitation echoes are kept almost unchanged. These methods have the advantages of effectiveness and simplicity. The threshold of discrimination is independent of the geographical and climatic conditions in the regions under study. Because the computation time needed by these techniques is small, the radar images can be processed in real-time.