Multi-Temporal Landslide Inventories: Creation from Visual Interpretation of Satellite Imagery

Organised by Laboratory of Geo-information Science and Remote Sensing

Tue 27 August 2019 10:30 to 11:00

Venue Gaia, gebouwnummer 101
Room 2

By Michiel Voermans

Multi-temporal Landslide Inventories (MLI) are essential to increase understanding about evolutionary processes of landsliding and have become a prerequisite for time-variant landslide susceptibility modelling. However, until today only a handful of unique, rich and detailed MLI are available for the long run (>15 years) worldwide. To create more MLI, current knowledge, tools and techniques as used in mono-temporal landslide mapping need to be investigated.

The overall objective of this thesis is to find out how MLI can be created using visual interpretation of satellite imagery. It answers two questions to meet this objective: 1) Which satellite data sources can be used for visual interpretation to create MLI? Through literature research, an overview was obtained on which freely accessible satellite data sources can be used for visual interpretation to create MLI. And 2) For the Collazzone study area in Italy, what are the differences between the existing MLI created by multiple sources and a MLI created by visual interpretation of satellite imagery only? World’s richest MLI contains 17 time-slices of this study area. Using only freely accessible satellite imagery of this study area, a second MLI was created through pseudo-stereoscopy. Resulting MLI was compared with the existing MLI for validation.

Visual interpretation through pseudo-stereoscopy of solely freely accessible satellite imagery identified only 441 out of 1567 known landslides, with total landslide surface area of underidentifications nearly twice as high as of overidentifications. While time-slices of the existing MLI are timed shortly after landslide-triggering events, time-slices created from freely accessible imagery were not. It is therefore expected that, with proper timing by means of purchasing imagery, more satisfactory MLI can be created. The low identification performance contrasts with published work that inventoried landslide types that leave more easily detectable signatures and at landslide prone areas with less spatial complexity than in the Collazzone area. As to date, for such areas and landslides, MLI of 15 years at max can be created. Future research should investigate the potential of methods that exploit parts of the electromagnetic spectrum other than the visual window and of non-heuristic methods for the creation of MLI.