Robust Time-Series InSAR Deformation Monitoring by Integrating Variational Mode Decomposition and Gated Recurrent Units

Continuous and large-scale surface deformation monitoring is critical for the comprehension of natural hazards and environmental changes. This can be facilitated by time-series interferometric synthetic aperture radar (TS-InSAR), which provides unprecedented spatial and temporal resolution. However, the original TS-InSAR measurements, being a superposition of trend, seasonal, and noise signals, often suffer from outlier and annual seasonal variations due to the influences of atmospheric delay, especially in coastal and mountainous areas, resulting in skewed monitoring if neglected. To address these issues, an integration method of variational mode decomposition and gated recurrent unit (VMD-GRU) is proposed in this study to enhance the robustness of continuous large-scale surface deformation monitoring. The VMD decomposes low-frequency trend, specific-frequency seasonal, and high-frequency noise components from the original TS-InSAR data via frequency-domain variational optimization first. Then, by eliminating the seasonal component decomposed by VMD from the original time series, the time series is reconstructed, effectively removing the influence of annual seasonal variations. Subsequently, GRU is utilized to further eradicate noise from the reconstructed time series, mitigating the influence of outliers and noise, thereby yielding a trend component that intuitively reflects surface deformation. Experiments on physical-based synthetic and real-world datasets demonstrate that the proposed VMD-GRU outperforms the existing methods. By introducing the frequency priors, the proposed method significantly enhances the robustness and accuracy of continuous large-scale surface deformation monitoring, providing a more reliable understanding of natural hazards and environmental changes.