Semi-automatic Landslide Detection Using Google Earth Engine, a Case Study in Poi Village, Central Sulawesi

Abstract

Fast and accurate landslide detection is important for landslide early warning systems. However, data available from local authorities and news reports vary in accuracy (time and location). In this work, we present a new method for identifying landslides, based on Google Earth Engine (GEE) and time-series analysis of Sentinel-2 optical satellite images. The method uses vegetation loss as a proxy for disturbance caused by earthquake-related landslides, and applies a change detection algorithm to compute the Normalized Different Vegetation Index (NDVI) and Relative Different NDVI (rdNDVI). As a test case, we applied this approach to the area of Palu, Central Sulawesi, which was hit by a major earthquake on September 28, 2018. Using time series data from 2015 to 2020, we were able to accurately capture the massive landslide in Poi Village caused by this earthquake. Using GEE had many advantages: the process is semi-automatic, fast and versatile, and the boundaries of the landslide zones can be auto-generated. In addition, the analysis does not require expensive high-resolution data. Our results demonstrate the potential of this new method to produce landslide inventories in a fast, accurate and low-cost manner.