Objectives As a deformation monitoring technique based on surface observation, interferometric synthetic aperture radar (InSAR) has been widely used in the study of surface deformation monitoring for its advantages of high precision and wide range. However, there is still a contradiction between the number of measuring points and the measurement accuracy in the common methods of surface deformation monitoring using InSAR, which limits its application in urban low-coherence areas. Distributed scatterer-InSAR (DS-InSAR) has emerged as a powerful remote sensing technique that significantly increases the number of coherent target pixels in non-urban areas without sacrificing spatial resolution. However, there remains a challenge in balancing the quantity of coherent targets with the accuracy of deformation estimation in conventional DS-InSAR method. To address this issue, we propose a new DS-InSAR method that integrates M-estimation and Bayesian estimation (MB-InSAR), further improving the accuracy of parameter estimation for deformation rates in low-coherence areas.

Methods The proposed method applies a three-arc strategy to construct the persistent scatterer-disctributed scatterer (PS-DS) arcs network, which provides a more accurate prior distribution for the DS estimator. Subsequently, the MB-InSAR method uses the obtained PS point parameter to provide the prior distribution of the parameters for DS solution. The DS points to be solved are divided into points with different interpolation error levels through the velocity interpolation variance threshold. The DS point parameters with small interpolation error are obtained based on weighted Bayesian estimation, and the DS point parameters with large interpolation error are obtained based on M estimation. The robustness of DS parameter estimation is further improved.

Results Quantitative evaluation using simulated data shows that the MB-InSAR method can provide more stable and accurate estimation results than maximum likelihood estimation (MLE) and Bayesian estimation for observation noise and uneven distribution of observation points. The MB-InSAR method improves the accuracy of long time series velocity estimation by 47%, compared with traditional MLE and Bayesian estimation. Moreover, the MB-InSAR method exhibits less sensitivity to PS density and noise. The experiments in Katy and Sienna region, Houston city, USA, indicate that the MB-InSAR method increases the number of measurement points by 160% and 125%, respectively. Futhermore, the velocity obtained by the MB-InSAR method shows strong consistency with GPS observation, and the maximum difference between the MB-InSAR method and GPS is less than 2.5 mm/a.

Conclusions The MB-InSAR method is effective in improving the estimation accuracy for the DS targets, which can shed more light on deciphering the fine-scale deformation characteristics with the complex surface environment.