Rapid Detection of Earthquake-triggered Landslides from Satellite Radar

Abstract

Triggered landslides pose a major risk following large earthquakes in mountainous areas and disrupt emergency response efforts. If information on the spatial distribution of these landslides can be generated quickly enough, it is therefore invaluable to emergency response coordinators. At present, this information is commonly generated manually from optical satellite imagery, which is labour-intensive and can be delayed or left incomplete due to cloud cover. This means a complete picture of triggered landsliding is often unavailable within the time frame of the emergency response. Alternatively, empirical models can predict landslide probability based on factors such as shaking intensity and slope steepness within hours of an earthquake, but these models are static in time and not always reliable as they do not contain any observations of landslides.

Satellite radar offers a third alternative method of generating landslide information for emergency response. These data can be acquired through cloud and are now available within days of any continental earthquake. Radar data are sensitive to landslides, which alter the scattering properties of the Earth’s surface, so could be used to generate all-weather information on landslide spatial distributions within days of an earthquake. Satellite radar data are routinely used to generate other products for emergency response, but for landslide detection, the testing and development of radar methods is not yet sufficiently advanced for them to be widely applied. In this thesis, I present new methods of landslide detection based on satellite radar coherence, a measure of the level of noise in an interferogram that reflects physical changes in the Earth’s surface such as landslides. I carry out systematic testing of new and existing methods of coherence-based landslide detection across four case study earthquakes using data from two satellite radar sensors, allowing identification of which method is preferable depending on the data available after an earthquake. Finally I experiment with combining empirical models and radar coherence methods. Overall, I demonstrate that useful information on landslide intensity can be generated within two weeks of an earthquake using satellite radar data, and that the addition of these data to empirical models can significantly improve their performance.