Science Service System

Summary of Proposal GEO1749

TitleSAR change detection methodologies for event landslide mapping
Investigator Mondini, Alessandro - CNR - National Research Council, IRPI
Team Member
Dr. Notarnicola, Claudia - Eurac, Institute for applied Remote Sensing
Dr. Furuta, Ryoichi - RESTEC, Research & Development Department
Dr Pizziolo, Marco - Regione Emilia Romagna, SGSS
SummaryThe use of high and very-high resolution images and the exploitation of innovative change detection techniques are important support for many application domains, such disaster mapping and risk mitigation. This project aims at investigating the use of multi-sensor (both opticaland SAR) RS imagery for detecting and mapping event landslides in different geo-environmental conditions. This implies the accomplishment of the following specific objectives: the assessment of RS data acquired from different sensors and at different scales. Particular attention will be devoted to the use of the X-band SAR data in order to understand potentials and limitations.The investigation and development of advanced, robust and customized change detection techniques and algorithms for addressing the rapid mapping of event landslides. The comparison of the products derived at different scales and from different sensors, with the aim to derive useful guidelines for their integration in the perspective of an integrated monitoring framework. The first phase of the project will regard an extensive state-of-the-art analysis regarding the most recent and advanced change detection techniques applied to high resolution SAR and optical images. Particular attention will be devoted to the specific chosen application, that is detection and monitoring of landslide and the related specific conditions. A part will also be dedicated to the identification and relationship understanding of the features of the investigated RS signals (both SAR and optical) more sensitive to the detection of landslide at different spatial and temporal scales. Contextual information (e.g., the local topography, available inventory maps) will be considered. Both field reference measurements and theoretical backscattering/radiative transfer models will be exploited. Specific strategies will be adopted and developed for the adaptation to the characteristics and challenges of the specific test sites under consideration (e.g densely vegetated areas) at different spatial-temporal scales. Different techniques will be considered and the different product derived will be compared among each other and with reference information derived from ground measurements and known ground information. The aim is to derive indications about the accuracy of the estimates derived from RS and hints about possible synergy among multiple RS products and field measurements. In this phase traditional correlation and accuracy metrics will be exploited. Under well-defined ground conditions, e.g bare soils or lightly vegetated soils, the recognition of some landslides might become really problematic: the landslide spectral fingerprints left on the territory and trapped by the satellite images are or can be very similar to the fingerprints of the surrounding stable areas. There, models for the estimation of the pre-event rainfall flow and infiltration, even exploiting soil moisture maps, will support the mapping of an event by defining the area prone to the instability. The expected general outcomes of this project will be the following: a set of products at different spatial scales and derived from different techniques associated with known landslide events in the selected test areas, an analysis on strengths, weaknesses and possible synergies among multi-scale RS products and synergy of different sensors (SAR and optical) in the optic of their combined exploitation for a more accurate and comprehensive study of processes. With specific regard to the use of TerraSAR-X SAR imagery, the expected results of the activity will be: the analysis and the assessment of TerraSAR-X X-band SAR data for the detection and mapping of landslide under different conditions, the analysis of the potential and limitations of X band SAR for the mapping and monitoring of natural processes in synergy with other RS data and auxiliary information.
Detailed reportThe list of the expected activities proposed in the previous report included: 1) pre-processing all the new images 2) look for situations where pre- and post-event (were the event is represented by a new landslide) images are available, 3) Change detection analysis according to the method proposed in the previous paragraph (log ratio imaging) 4) Statistical characterization of the fingerprints of the different classes (changed and unchanged areas) 5) landslide map preparation and quality assessment 6) analysis of the limitations In case of unavailability of a new reliable test case in the new acquired images, consider the possibility to recreate a synthetic pre-event image, combining the information of land cover present before the event obtained from an optical image. All the obtained images have been pre-processed and prepared according to the procedure proposed in the previous report. Unfortunately, in the temporal window covered by the stack, no new event (population of) landslides occurred and, as consequence, the change detection method (correctly) did not highlight any significant change. We tried to characterize the significance of the different distribution of sigma nought inside and outside landslides only using post event images, but that, at the catchment scale (geomorphological requirement for event landslide mapping) resulted not discriminant, or at least not enough to run standard classification methods for landslide identification and mapping. Small but significat differences were obtained only at the slope scale where huge landslides occurred. It then resulted impossible to prepare a landslide inventory map, and furthermore to carry out an analysis of the limitations. After having obtained some good preliminary results using Sentinel 1, our conclusion is that the possibility to obtain routinely event inventory maps using TerraSAR-X is to be planned a priori with a continuous acquisition program over the selected areas. The preparation of the inventories using CD approaches using the archive can only be occasionally successful. We also tried to reconstruct synthetic pre-event images but we gave up believing the process too uncertain.

Back to list of proposals

DLR 2004-2016