Identification of Landslide Scars in the Region of the Serra do Mar, São Paulo State, Brazil, Using Digital Image Processing and Spatial Analysis Tools

Abstract

The objective of the present study, developed in a mountainous region in Brazil where many landslides occur, is to present a method for detecting landslide scars that couples image processing techniques with spatial analysis tools. An IKONOS image was initially segmented, and then classified through a Batthacharrya classifier, with an acceptance limit of 99%, resulting in 216 polygons identified with a spectral response similar to landslide scars. After making use of some spatial analysis tools that took into account a susceptibility map, a map of local drainage channels and highways, and the maximum expected size of scars in the study area, some features misinterpreted as scars were excluded. The 43 resulting features were then compared with visually interpreted landslide scars and field observations. The proposed method can be reproduced and enhanced by adding filtering criteria and was able to find new scars on the image, with a final error rate of 2.3%.