Open AccessCapability of Sentinel-1 Synthetic Aperture Radar polarimetric change detection for burned area extraction in South Kalimantan, Indonesia
Author(s) -
Syam’ani
Publication year - 2020
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/500/1/012004
Subject(s) - synthetic aperture radar , thresholding , remote sensing , polarimetry , extraction (chemistry) , correlation coefficient , change detection , backscatter (email) , environmental science , geology , computer science , artificial intelligence , mathematics , optics , physics , scattering , image (mathematics) , chemistry , statistics , telecommunications , chromatography , wireless
Burned area extraction from optical imageries often has a major problem, that is the presence of atmospheric particles. A potential alternative for burned area extraction is using SAR imageries, those are relatively free of atmospheric interference. The purpose of this research was to explore change detection capabilities of Sentinel-1 SAR polarimetry for burned area extraction. The imagery used is dual-polarized Sentinel-1 (VV,VH). A number of polarimetric transformation methods are applied for the purpose of change detection. Those are, single cross-polarized log ratio, single co-polarized log ratio, dual polarized combination log ratio, dual polarized multiple log ratio, and dual polarized ratio log ratio. For comparison, the Relativized Burn Ratio (RBR) method was applied to the Sentinel-2 MSI imagery. The Otsu thresholding is then applied to separate the burned area and the unburned area. The results of the research showed that the single cross-polarized log ratio (ln(σ 0 VHt1/σ 0 VHt2)) transformation method was the most accurate method. This method has an overall accuracy of 88.7665% (Kappa 0.7567). It is more accurate than Sentinel-2 RBR, which has an overall accuracy of 81.8470% (Kappa 0.6383). Cross validation between Sentinel-1 SAR change detection and Sentinel-2 RBR does not show a significant correlation. The highest correlation coefficient achieved is only 0.25. This shows that burned area extraction between change detection from SAR imageries and RBR from optical imageries has a different mechanism. SAR change detection tends to detect changes in surface roughness, while NIR-based RBR tends to extract changes in leaf chlorophyll conditions.