Day 4

Detailed paper information

Back to list

Paper title Expert interpretation of SAR amplitude derived products: an application to event landslide mapping
Authors
  1. Michele Santangelo CNR-IRPI Speaker
  2. Mauro Cardinali CNR-IRPI
  3. Francesco Bucci CNR-IRPI
  4. Federica Fiorucci CNR-IRPI
  5. Alessandro Cesare Mondini CNR-IRPI
Form of presentation Poster
Topics
  • D1. Managing Risks
    • D1.01 Satellite EO for Geohazard Risks
Abstract text Landslides triggered by intense and prolonged rainfalls occur worldwide and cause extensive and severe damages to structures and infrastructures and loss of life. Obtaining even coarse information on the location of triggered landslides during or immediately after an event can increase the efficiency and efficacy of emergency response activities, possibly reducing the number of victims. In most cases, however, in the immediate aftermath of a meteorological triggering event, optical post-event images are unusable due to cloud cover. The increasing availability of images acquired by satellite Synthetic Aperture Radar (SAR) sensors overcome this limitation, because microwaves do not interact with water vapour. In the literature it has been shown that C-band Sentinel-1 SAR amplitude images allow the detection of known event landslides in different environmental conditions. In this work we explore the use of such images to map event landslides.

SAR backscatter products are generally represented by a grey tone matrix of backscatter values mainly influenced by (i) the projected local incidence angle, (ii) surface roughness, and (iii) the dielectric constant, used as a proxy for soil moisture. Similarly to optical images, landslides modify the local tone, texture, pattern, mottling and grain of the grey tone matrix. Therefore we refer to a “radar backscatter signature” of event landslides as the combination of these three main components which can reveal the occurrence of a landslide in radar amplitude products. Interpreters use such features to infer the occurrence of event landslides (landslide detection), and to delineate landslide borders (landslide mapping), similarly to what is done for optical post-event images. In this study, four expert photo-interpreters have defined interpretation criteria of SAR amplitude (i) post-event images of the backscatter coefficient (i.e. β₀, the radar brightness coefficient) and of the (ii) derived images of change computed as the natural logarithm of the ratio between the post- and pre-event images (i.e., ln(β₀post/β₀pre)). Interpretation criteria build on the well-established ones usually applied to optical images. Different criteria were defined to interpret images of change, where clusters of pixels of changes pop out from the salt and pepper matrix (i.e. anomalies). Such changes can be caused by several different phenomena, including slope failures, snowmelt, rainfall, vegetation cuts, among others. Interpreters identify areas where the change has not been random, and decide whether the cluster is a landslide based on the shape of the cluster. The risk of incurring in morphological convergences (i.e. ambiguities in the interpretation) is higher if change images are examined alone. Often, use of ancillary data such as Digital Elevation Models can help exclude possible erroneous interpretations.

The same team of image interpreters mapped two large event landslides. The first is a rock slide - debris flow - mudflow occurred in Villa Santa Lucia, Los Lagos Region, Chile on 16 December 2017. The second is a rock slide occurred in early August 2015 in the Tonzang region, Chin Division, Myanmar. The landslide maps were prepared on a total of 72 images for the Chile test case and 54 for the Myanmar test case. Images included VV (vertical transmit, vertical receive) and VH (vertical transmit, horizontal receive) polarisation, ascending and descending acquisition geometries, multilook processing, adaptive and moving window filters, post-event images and images of change. For the Chile test case, interpreters mapped the event landslide on an optical post-event image before mapping on SAR images, whereas in Myanmar it was done in the end. Maps obtained from SAR aplitude derived products were quantitatively compared to the maps prepared on post-event optical images, assumed as benchmark, by using a geometrical matching index. Despite the overall good agreement between the SAR- and optical-derived landslide maps, locally, errors can be due to geometrical distortions, and speckling-like effects. In this experiment, polarisation played an important role, while filtering was less decisive. Results of this study proved that Sentinel-1 C-band SAR amplitude derived products can be exploited for preparing accurate maps of large event landslides, and that they should be further tested to prepare event inventories. Other SAR bands and resolutions should be tested in different environmental conditions and for different types and sizes of landslides. Application of rigorous and reproducible interpretation criteria to a wide library of test cases will strengthen the capability of expert image interpreters of using such images to produce accurate landslide maps in the immediate aftermath of triggered landslide events worldwide or even train automatic classification systems.