Semi-automatic mapping of shallow landslides using free Sentinel-2 images and Google Earth Engine

Abstract. The global availability of Sentinel-2 data and the widespread coverage of cost-free and high-resolution images nowadays give opportunities to map, at a low cost, shallow landslides triggered by extreme events (e.g. rainfall, earthquakes). Rapid and low-cost shallow landslide mapping could improve damage estimations, susceptibility models and land management. This work presents a two-phase procedure to detect and map shallow landslides. The first is a semi-automatic methodology allowing for mapping potential shallow landslides (PLs) using Sentinel-2 images. The PL aims to detect the most affected areas and to focus on them an high-resolution mapping and further investigations. We create a GIS-based and user-friendly methodology to extract PL based on pre- and post-event normalised difference vegetation index (NDVI) variation and geomorphological filtering. In the second phase, the semi-automatic inventory was compared with a benchmark landslide inventory drawn on high-resolution images. We also used Google Earth Engine scripts to extract the NDVI time series and to make a multi-temporal analysis. We apply this procedure to two study areas in NW Italy, hit in 2016 and 2019 by extreme rainfall events. The results show that the semi-automatic mapping based on Sentinel-2 allows for detecting the majority of shallow landslides larger than satellite ground pixel (100 m2). PL density and distribution match well with the benchmark. However, the false positives (30 % to 50 % of cases) are challenging to filter, especially when they correspond to riverbank erosions or cultivated land.