| Summary: | 碩士 === 國立中央大學 === 遙測科技碩士學位學程 === 107 === The National Disaster Management Agency of Indonesia (2016) recorded 2,425 incidents of land movement disaster during 2011 to 2015, with locations occurring in various parts of Indonesia. In the South Coast of Java Island, Pacitan where located in East Java is one of the most heavily damaged area, during the tropical cyclones, Cempaka, from 27 to 30 November 2017, and induced floods in the lowland area and landslides in the mountainous area. For landslide and flood detection, satellite data is effective to be applied for larger area with economic cost. Among many kinds of satellite resources, synthetic aperture radar (SAR) has less limitation operating in cloudy conditions, which is considered a very useful characteristic for landslide and flood rapid mapping during cloudy condition. With applying SAR data, few studies have focused on the detection of flood and landslide at once, considering their similar backscattering characteristics which are normally lower and difficult to be distinguished. However, this study proposed a method which analyzes the multi-temporal SAR backscattering to investigate the difference between flood and landslide in time domain.
This study focuses on availability of Sentinel-1 C-Band SAR imagery to detect the landslide and flooded area for Cempaka event. The time series of Sentinel-1 were pre-processed to analyze the backscatter change over the landslide and flooded area. Then, the SVM (Support Vector Machine) classifier was applied to map landslide and flooded areas. The accuracy assessment shows that the best classification result is obtained when combining both six VV and six VH polarization time-series data (twelve-bands in SVM classification). The overall accuracy achieves 81.42% and kappa coefficient 0.51. The result indicates the applicability of the proposed method for landslide and flood detection.
|
|---|
