METHOD FOR LANDSLIDES IDENTIFICATION AT THE SAO PAULO STATE COAST, BRAZIL

Satellite images are an important tool to map natural disaster, mainly debris flow. The Support Vector Machines (SVM) algorithm has been used to classify the natural disaster, obtaining good results, although some images present shadows and mists which difficult the classification. Some enhancements minimize those problems facilitating the classification process. This paper aims to present a method to classify debris flow areas near to an important road of the Sao Paulo State coast, Brazil, using LANDSAT images. Maximum Likelihood Classification (MLC) and SVM algorithms were applied. Due to the shadows the classification points huge debris flow areas. To neutralize the influence of shadows, Normalized Difference Vegetation Index (NDVI) was employed which turns easier to sample the training areas and perform the classification. MLC algorithm cannot be applied in case of a unique band, SVM can. So SVM is performed for the enhancement of classification and better results are observed with the combined methods SVM/NDVI. The overlay of this classification and Digital Terrain Model confirms the coincidence of debris flow event and classification. This method was very effective to the area now studied and may be useful to debris flow mapping