The Feasibility Assessment of Quick Landslide Identification Methods After Hazards with Sentinel-1 SAR Imagery

In remote sensing of landslide investigation, the interpretation of optical image is the main method at present. However, when a disaster occurs, it is very difficult to obtain images without cloud coverage. For example, Typhoon Lubi in August 2021 and Typhoon Nissa in 2022 caused many landslides and road interruptions. However, due to the cover of clouds and fog, it was impossible to obtain satellite images in time to judge the scale of the disaster. Unmanned vehicles are also affected by weather factors, which greatly increases the risk of flight. Therefore, it is extremely necessary to develop disaster identification methods that are not affected by weather.

In this study, the long electromagnetic waves of synthetic aperture radar (SAR) are not affected by cloud and fog to develop a landslide detection model for radar images. The reference range of the location and scale of the landslide can be obtained under bad weather conditions to make up for the weather limitations when evaluating the scope of the disaster with optical images.

In this study, the NDSI&RVID method is used as the index for the identification and interpretation of the landslide area, and the analysis and discussion of the landslide area is carried out in combination with multi-time series and different orbital data. The effect of landslide identification is improved by three methods: single-sequence identification and interpretation stacking, multi-time-series index stacking, and multi-time-series image stacking. Among them, better interpretation results can be achieved by stacking multiple time-series images. It is recommended to use the number of 4 images before the disaster and 1 image after the disaster for data interpretation. Although the image pixel classification effect still needs to be improved, the identification rate for landslides of more than 10 hectares can reach more than 90%. In the absence of optical images, it has considerable reference value.