Assessment of the interaction of land-cover change on shallow landslide occurrence using an automated object-based approach

Abstract

Land-cover change could considerably lower landslide triggering rainfall thresholds allowing precipitation events with shorter recurrence intervals to initiate shallow landslides. This research focusses on developing an automated, robust and upscalable workflow to quantitatively assess the effect land-cover change has on initiating rainfall induced shallow landslides in the Laternser Valley. Land-cover is classified using four sets of high resolution orthophotos (198x, 2001, 2006, 2009; 0.25 m spatial resolution) by applying an object-based approach with eCognition software. The correlation between land-cover change and landslide occurrence was assessed by analyzing land-cover change trends in the vicinity (< 25 meters) of mapped shallow landslides. The obtained classification accuracy ranges from 76% for 198x to 88% for 2009. The relative area undergoing land-cover change is 18% in the whole Laternser valley and 34% in the vicinity of landslides. Overall land-cover change trends indicate a shift from grassland to forest in the whole Laternser valley. However, in the vicinity of landslides the opposite is observed, namely a shift from forest to grassland and grassland to bare soil. Even though a general vegetation reduction is detected in the vicinity of landslides no correlation between LCC and landslide occurrence could be established yet.