SAR (Synthetic Aperture Radar) images can show ground features, even when they were taken at night or in cloudy weather. Therefore, it has been said that SAR images may have a great potential use for disaster remote sensing. In this paper, we used only full polarimetric SAR data taken after the March 11, 2011 earthquake in eastern Japan to evaluate the inundated areas. First, we derived a coherency matrix for each pixel from the scattering matrix of the SAR data. The image was then classified into 4 classes (water, vegetation, field, and urban area) based on the elements of the coherency matrix. After that, the inundated areas were extracted by masking rivers and lakes included in the water area. As a comparison, AVNIR-2 (Advanced Visible and Near Infrared Radiometer-
Ⅱ) image was also classified into the same 4 classes with its inundated areas extracted. Ground validation data was retrieved from Google Earth images. As a result, the overall accuracy and kappa coefficient was 74% and 0.65 respectively for the full polarimetric SAR and 95% and 0.93 respectively for the AVNIR-2. Results indicate that the full polarimetric SAR image after the earthquake was useful enough to estimate inundated areas.
View full abstract