Objectives Time-series offset tracking is a critical technique for landslide monitoring with optical images, which has been applied worldwide successfully in the recent years. The continued monitoring of landslide can provide early warning and emergency information for the nature hazard prevention. However, the resolution in the temporal domain is restricted by the single-platform datasets, especially in the case of bad weather.To solve this issue, we use the multi-source optical images, including Sentinel-2 and Landsat 8 to generate the long time-series deformation of landslides. The high resolution of deformation in the temporal domain can help us know more about the development process of the landslide.
Methods The method is executed based on the idea of small baseline subset (SBAS) in the interferometric synthetic aperture radar (InSAR) technique, aiming to reduce the influence of the temporal baseline and noise on the signal noise ratio (SNR). Meanwhile, to obtain time-series deformation of the multi-platform optical images, we use a weighted singular value decomposition (SVD) method to obtain the time-series deformation of the multi-platform optical images. 11 Sentinel-2 datasets and 16 Landsat 8 datasets are used to generate two-dimensional horizontal time-series deformations of the Baige landslide from 2014-11 to 2018-09. Firstly, we calculate the time-series offset tracking of the single platform, i.e., Sentinel-2 and Landsat 8, and the dual-platform by considering the selection of the image pair, search window size, step length, and the threshold of SNR respectively. And a time-series deformation of 27 images are obtained with offset tracking technique of the dual-platform. Secondly, the accuracy of the results for the single-platform and the dual-platforms is evaluated by two commonly used methods: Stable area statistics with a priori knowledge and cross-validation of independent datasets.Then, the developmental stages of the Baige landslide is divided based on the long time-series deformation of the dual-platform, the creep deformation theory, and the tangential angle of the time-series curve.Finally, the relationship between precipitation and deformation is discussed to analyze the causes of the landslide.
Results The results show that: (1)The time-series deformation of the dual-platform can obtain more details compared the results of the single platform. (2)The deformation accuracy of the dual-platform is higher than that of the single platform. For example, the improvements of the dual-platform in the east-west direction and the north-south direction are 3.02% and 5.37% respectively compared to the Sentinel-2. In the Landsat 8 case, the results are 3.61% and 0.40% respectively. During the time span, the maximum accumulated deformation and the deformation rate reach to 42.90 m and 9.06 m/a respectivley. In addition, the deformation accuracy of the Sentinel-2 is higher than that of the Landsat 8 in the similar situation of observations.(3)The pre-event of the Baige landslide can be divided into the initial stage, the uniform stage, the initial acceleration, and the medium-term acceleration. The corresponding average deformation rates of each stage are 7.05 mm/d, 10.27 mm/d, 26.80 mm/d, and 65.79 mm/d respectively. The landslide is going through the medium-term acceleration stage until the 5th, Feb. 2018.(4)the correlation between the deformation and precipitation is 0.88, which indicates that heavy rainfall is one of the major causes for the Baige landslide.
Conclusions Although the deformation resolution of the multi-platform optical image is higher than that of the single platform, the fusion accuracy is limited by the accuracy of different individual platforms caused by the sensitivity for the topography. Hence, in the future work, we will focus on the combined estimation of the multi-platform based on the variance component estimation.