Item

Abstract

Spatiotemporal landslide inventories of large areas are important for the understanding of regional landslide process dynamic and thus a prerequisite for probabilistic hazard and risk assessment. We developed a (semi-)automated tool for spatiotemporal landslide mapping using satellite time series data of various multispectral sensors. The approach comprises automated multi-sensor pre-processing strategies and multi-temporal change detection methods. Landslides are identified based on the temporal variations of landslide-related surface cover changes, mostly comprised by sudden vegetation cover destruction and longer-term post-failure revegetation. In combination with DEM-derivatives this multi-temporal change detection approach allows landslide identification of different sizes, shapes, and in different stages of development (e.g. fresh failures and reactivations of existing landslides) under varying natural conditions. This paper presents the application of the approach to three different scenarios. In Kyrgyzstan a monitoring of recent (2009–2016) landslide occurrence and a retrospective analysis of long-term (1986–2013) regional landslide dynamic was performed. The approach was also applied to Nepal to analyze landslide occurrence (2011–2015) triggered by the 2015 Gorkha earthquake and by the monsoon seasons before and after. The study sites are 12000, 2500, and 625 km2 for the monitoring and long-term analysis in Kyrgyzstan and for Nepal respectively. The derived multi-temporal landslide inventories contain several thousands of landslides, ranging in size from 100 m2 to 2.8 km2. These data records allowed comprehensive analysis of spatiotemporal variations in landslide occurrence, revealing distinct spatial and temporal hotspots of landslide activity in all of the regions.