Semi-automatic detection of rainfall-triggered landslides in segmented optical satellite images

Abstract

A new method for rapid mapping of recent rainfall-triggered landslides using a pair of multitemporal very-high spatial resolution images acquired by optical sensors is outlined. The algorithm analyzes image segments, obtained from the post-event image, that are used to guide spectral comparisons and shape analysis of merged candidate segments. The algorithm is tested on 14.9 km2 of satellite imagery acquired in Nova Friburgo (Brazil) in 2011. Most of the landslides were detected successfully.