关键词: PS选点, CR同质像元, KS检验, PSInSAR, 武山-甘谷断裂

Abstract:

The key challenge of PSInSAR(Permanent Scatter Interferometric Synthetic Aperture Radar)lies in the quality of Permanent Scatter(PS)points, which are difficult to extract accurately in fracture zones due to the complex natural ground cover and unique geomorphological environments. In such areas, the inability to reliably extract high-quality PS points limits the application of PSInSAR for monitoring interseismic deformation. To address the problem of high-quality PS point selection in fault zones and improve the effectiveness of PSInSAR technology for monitoring interseismic deformation, this paper presents a comparative study of coherence coefficient and amplitude deviation double thresholds. The study further integrates the Kolmogorov-Smirnov(KS)test and CR homogeneous pixel method, in addition to conventional coherence coefficient point selection techniques. Sentinel-1A SAR images from March 14, 2015, to February 16, 2020, are used as the data source, with the Wushan-Gangu section of the fault zone on the northern edge of the Western Qinling Mountains, near the northeastern edge of the Qinghai-Tibet Plateau, serving as the test area for PSInSAR processing. The quality and reliability of PS point selection using various methods are compared and analyzed.

Two sets of coherence coefficient T_γ and amplitude deviation D_γ double thresholds were tested. The coherence coefficient was set at T_γ=0.5, while the amplitude deviation D_γ was set to 0.5 and 0.3, resulting in 19806 and 2485 PS points, respectively. When the amplitude deviation threshold was lowered, the number of PS points on ridgelines decreased significantly, while there was little change in Gangu County, suggesting poor pixel amplitude stability in mountainous areas. Lowering the threshold eliminated many PS points, retaining only those with high amplitude and stable time series, typically found in hard targets like urban buildings. A KS double sample test was then applied in combination with the double threshold method, with all thresholds coefficient T_γ, amplitude deviation D_γ, and KS test P_γ set at 0.5. This approach yielded 1 313 PS points, showing a significant reduction in PS points on ridgelines and in Gangu County, while urban points became more concentrated on hard targets like buildings. Although the KS test reduced PS points in vegetated areas, it did not fully eliminate noise points. Finally, based on the double threshold results of coherence coefficient T_γ=0.5 and amplitude deviation D_γ=0.3, both the KS test and CR homogeneous pixel selection methods were applied. The CR homogeneous pixel method used a phase difference threshold P_γ=0.5 and a temporal phase stability threshold N_γ=50. This yielded 2 485 PS points for the double threshold method, 133 PS points for the double threshold plus KS test, and 414 PS points for the double threshold plus CR homogeneous pixel method. The latter two methods significantly reduced PS points, with a higher concentration of points in Gangu County, consistent with the expectation that PS points predominantly correspond to hard targets like buildings.

Statistical analysis of the results demonstrated that the combination of the coherence coefficient, amplitude deviation, and CR homogeneous pixel method provided the highest quality PS points, effectively excluding noise points in vegetated areas. The combination of coherence coefficient, amplitude deviation, and KS test ranked second, improving accuracy in urban areas but failing to eliminate noise in vegetated areas. Using Sentinel-1A SAR images and the Wushan-Gangu fault as the test area for time series PSInSAR processing, the accuracy of PS point selection was further verified. A comparative analysis of deformation monitoring results from the three methods revealed that both the KS test and CR homogeneous pixel method improved the accuracy of fault deformation monitoring, with the CR homogeneous pixel method yielding superior results. Monitoring data from 2015 to 2020 showed that the deformation rate of the northern block of the Wushan-Gangu fault ranged from -2 to -0.2mm/a, with an average deformation of approximately -1.7mm/a. In contrast, the southern block exhibited a deformation rate between 0.3 and 0.5mm/a, with an average deformation of about 1.8mm/a The relative average deformation rate between the northern and southern blocks was 0.7mm/a, indicating left-lateral strike-slip movement. Among the three methods, the double threshold plus CR homogeneous pixel method produced PS points with the smallest deformation rate standard deviation, indicating more stable and reliable deformation results.

Key words: PS point selection, CR homogeneous pixel, Kolmogorov-Smirnov test, PSInSAR, Wushan-Gangu Fault