Multi-dimensional and long-term time series monitoring and early warning of landslide hazard with improved cross-platform SAR offset tracking method

Multi-dimensional, long-term time series displacement monitoring is crucial for generating early warnings for active landslides and for mitigating geohazards. The synthetic aperture radar (SAR) interferometry method has been widely applied to achieve small-gradient landslide displacement monitoring; however, measuring the landslide displacement with a steep gradient has posed certain challenges. In comparison, the SAR offset tracking method is a powerful tool for mapping large-gradient landslide displacement in both the slant-range and azimuth directions. Nevertheless, there are some limitations in the existing SAR offset tracking approaches: (i) the measurement accuracy is greatly reduced by the extreme topography relief in high mountain areas, (ii) a fixed matching window from expert experience is usually adopted in the calculation of cross-correlation, (iii) estimating the long-term displacements between the SAR images from cross-platforms and with longer spatiotemporal baselines is a challenging task, and (iv) it is difficult to calculate the three-dimensional (3D) landslide displacements using a single SAR dataset. Additionally, only a few studies have focused on how to realize early warning of landslide deformation using SAR measurements. To address these issues, this paper presents an improved cross-platform SAR offset tracking method, which can not only estimate high-precision landslide displacements in two and three dimensions but also calculate long-term time series displacements over a decade using cross-platform SAR offset tracking measurements. Initially, we optimize the traditional SAR offset tracking workflow to estimate high-precision ground displacements, in which three improvements are highlighted: (i) an “ortho-rectification” operation is applied to restrain the effect of topography relief, (ii) an “adaptive matching window” is adopted in the cross-correlation computation, and (iii) a new strategy is proposed to combine all the possible offset pairs and optimally design the displacement inversion network based on the “optimization theory” of geodetic inversion. Next, robust mathematical equations are built to estimate the two-dimensional (2D) and 3D long-term time series landslide displacements using cross-platform SAR observations. The M-estimator is introduced into the 2D displacement inversion equation to restrain the outliers, and the total least squares criterion is adopted to estimate the 3D displacements considering the random errors in both the design matrix and observations. We take the Laojingbian landslide, Wudongde Reservoir Area (China), as an example to demonstrate the proposed method using ALOS/PALSAR-1 and ALOS/PALSAR-2 images. The results reveal that the proposed method significantly outperforms traditional methods. We also retrieve the movement direction of each pixel of the Laojingbian landslide using the proposed method, thus allowing us to understand the fine-scale landslide kinematics. Finally, we capture and analyze the acceleration characteristics of the landslide, perform an early warning of hazard, and forecast the future displacement evolution based on the 3D displacement time series coupled with the physical models of the rocks. This work was supported by the National Natural Science Foundation of China (Grant Nos. 41874005 and 41929001), the Fundamental Research Funds for the Central Universities, CHD (Grant Nos. 300102269712 and 300102269303), and by the China Geological Survey Projects (Grant Nos. DD20190637 and DD20190647). This research was also supported by a Chinese Scholarship Council studentship awarded to Xiaojie Liu (Grant No. 202006560031).