Abstract:The existing fire scar detection methods based on multi-temporal analysis only used the difference in either backscattering intensity or polarimetric characteristics between the pre- and post-fire PolSAR data to calculate the difference image (DI), and then applied an object-based image analysis approach (OBIA) to generate the fire scar binary map. These methods all ignored the polarimetric correlation between both the pre- and post-fire PolSAR data. And many parameters of the OBIA method were determined empirically. Therefore, this study proposes a new detection method, which integrates the Hotelling–Lawley trace (HLT) statistic with a hierarchical automatic image segmentation method (GGKI-MRF). The HLT can simultaneously capture changes in both polarimetry and intensity, which helps to distinguish between burned and unburned areas. And the GGKI-MRF method is developed by combining the generalized Gaussian-based Kittler–Illingworth (GG-KI) thresholding with Markov Random Field (MRF) model, which has a good performance in reducing the isolate points (false alarms) and the omission errors. The experimental results from multi-temporal and polarimetric PALSAR PolSAR data show that the proposed method can achieve high detection accuracy (i.e., the kappa coefficient of 0.81 and the overall accuracy up to 0.92).

The Automatic Detection of Fire Scar in Alaska Using Multi-Temporal PALSAR Polarimetric SAR Data