Automatic landslide mapping from satellite imagery with a topography-driven thresholding algorithm

10 Abstract— We present an improvement of image classification by 11 “thresholding”, using topographic information to determine 12 multiple thresholds. We devised a two-steps procedure for 13 automatic classification into landslide or no landslide categories of 14 a change-detection map obtained from satellite imagery. 15 Requirements of the proposed procedure are knowledge of the 16 occurrence of a landslide event, availability of a preand post17 event pseudo-stereo image pair and a digital elevation model. The 18 novel feature of the approach is represented by the use of slope 19 units as topographic-aware subsets of the scene within which we 20 apply a multiple thresholding method to classify a landslide class 21 membership tuned on the sole landslide spectral response. The 22 method is fully automatic after site-dependent operations, required 23 only once, are performed, and exhibits improved classification 24 performance with limited training requirements. Our automatic 25 procedure is a step forward towards systematic acquisition of 26 landslide events and real-time landslide mapping from satellite 27 imagery. 28