Interferometric Phase Optimization Method for Mountainous Regions Considering Geometric Decorrelation of Distributed Scatterers

Distributed scatter interferometric synthetic aperture radar (DS-InSAR) technology has been extensively employed for surface deformation monitoring, with phase optimization as a pivotal step. Currently, phase optimization techniques utilize the statistical intensity distribution of pixels to select homogeneous pixels. Pixels with low temporal intensity stability are excluded from consideration, avoiding their involvement in the phase optimization. However, it is noteworthy that distinguishing between homogeneous and heterogeneous pixels becomes more challenging in mountainous areas. Additionally, pixels with low stability are affected not only by thermal or environmental noise but also by the influence of local incidence angles, causing ground deformation beyond the Maximum Detectable Deformation Gradient (MDDG) of InSAR, resulting in geometric decorrelation. These pixels are often erroneously classified as noise and discarded. Nevertheless, these pixels contain rich and crucial deformation information, indicating disaster risks. Therefore, optimizing the phase of these pixels is essential.

This paper introduces a method for interferometric phase optimization of distributed scatterers in mountainous regions, considering geometric decorrelation (GD-DS). Using real InSAR differential interferometric phases as a basis, the study simulates interferometric phase datasets with rich spatiotemporal features, ensuring the correlation between simulated GD-DS phases and MDDG. Subsequently, K-means clustering is applied to segment the MDDG map, with resulting connected regions representing homogeneous pixels with similar local incidence angles. Convolutional denoising training is performed on homogeneous pixels using the generative adversarial network model (Pix2pix), and the trained model is then applied to real interferometric phase images. The proposed strategy and method are successfully applied to interferometric phase optimization in the Jishishan region of Gansu Province, China. Compared to traditional methods, the new approach demonstrates superior phase optimization performance, particularly in the case of GD-DS. Discussion and analysis of the spatial correlation between GD-DS and MDDG in the real experimental area confirm that introducing MDDG as a reference to optimize GD-DS is a key factor in improving phase optimization. Furthermore, the computational time of the new method is significantly reduced compared to traditional methods.