Establishing the timings of rainfall-triggered landslides using Sentinel-1 satellite radar data
Résumé
Heavy rainfall can trigger thousands of landslides, which have a significant effect on the landscape and can pose a hazard to people and infrastructure. Inventories of rainfall triggered landslides are used to improve our understanding of the physical mechanisms that cause the event, in assessing the impact of the event and in the development of hazard mitigation strategies. Inventories of rainfall-triggered landslides are most commonly generated using optical or multispectral satellite imagery, but such imagery is often obscured by cloud-cover associated with the rainfall event. Cloud-free optical satellite images may not be available until several weeks following an event. In the case where rain falls over a long period of time, for example during the monsoon season or successive typhoon events, the timing of the triggered landslides is usually poorly constrained. This lack of information on landslide timing limits both hazard mitigation strategies and our ability to model the physical processes behind the triggered landsliding.Satellite radar has emerged recently as an alternative source of information on landslides. The removal of vegetation and movement of material due to a landslide alters the scattering properties of the Earth"s surface, thus giving landslides a signal in satellite radar imagery. Satellite radar data can be acquired in all weather conditions, and the regular and frequent acquisitions of the Sentinel-1 constellation, could allow landslide timing to be constrained to within a few days. Satellite radar data has been successfully used in detecting the spatial distribution of landslides whose timing is known a-priori (for example those triggered by earthquakes). Here we demonstrate that time series of Sentinel-1 satellite radar images can also be used to achieve the opposite: the identification of landslide timing for an event whose spatial extent is known.We analyse radar coherence and amplitude times series to identify changes in the time series associated with landslide occurrence. We compare pixels within each landslide with nearby pixels outside each landslide that have been identified to be similar in pre-rainfall Sentinel-1 and Sentinel-2 imagery. We test our methods on rainfall-triggered landslides in Nepal and Japan, both of which are mountainous countries that experience regular heavy rainfall events that are often obscured by cloud cover in optical satellite imagery.