Rapid Response and Landslide Mapping using Freely Available Synthetic Aperture Radar in Google Earth Engine

Abstract:

Landslides and other natural hazards can cause loss of life and catastrophic damage to infrastructure and natural reserves. The rapid and accurate mapping of landslides is critical for quick emergency response and disaster mitigation. Satellite imagery is typically acquired within days of any disaster event. Unlike optical sensors, synthetic aperture radar (SAR) sensors can penetrate clouds and operate day or night. Although a number of landslide detection methods using SAR have been developed, they require downloading a large volume of data to a local system and specialized SAR processing software and training, which is time consuming and computation intensive. In this work, we use SAR data to identify rainfall triggered landslides in Hiroshima, Japan, a densely vegetated mountainous region, following Typhoon Prapiroon in July 2018. Damage proxy maps (DPMs) were created by quantifying the change in SAR amplitude before and after the rainfall event. We used publicly available images from Copernicus Sentinel-1 satellites and a Shuttle Radar Topography Mission (SRTM) digital elevation model. We processed these data in Google Earth Engine (GEE), a free cloud based online platform. SAR data becomes available in GEE within two days after acquisition and our DPM processing can normally complete the computations in less than a minute. To test the accuracy of our results, we compared our DPMs with the landslide inventory provided by the Geospatial Information Authority of Japan. We determined the area under the curve (AUC) of the receiver operating characteristic (ROC) plot to evaluate the performance of the SAR amplitude difference method and to compare landslide detection with a successive number of stacked SAR images. Our results show the AUC score increases from 63% to 84% with increased number of images used up to two years after the typhoon. We also found that combining ascending and descending data further reduced noise and corrected shadowing and geometric distortion. Our results demonstrate that this SAR amplitude-based change detection method is a rapid and effective procedure to inform emergency response teams and to generate landslide inventories. Not only is this methodology effective for detecting landslides, but it is also applicable for rapidly mapping other natural disasters such as floods and wildfires.