ABSTRACT
Time-series Interferometric Synthetic Aperture Radar (InSAR) technology, renowned for its high-precision, wide coverage, and all-weather capabilities, has become an essential tool for Earth observation. However, the quality of the interferometric baseline network significantly influences the monitoring accuracy of InSAR technology. Therefore, optimizing the interferometric baseline is crucial for enhancing InSAR's monitoring accuracy. Surface vegetation changes can disrupt the coherence between SAR images, introducing incoherent noise into interferograms and reducing InSAR's monitoring accuracy. To address this issue, we propose and validate an optimization method for the InSAR baseline that considers changes in vegetation coverage (OM-InSAR-BCCVC) in the Yuanmou dry-hot valley. Initially, based on the imaging times of SAR image pairs, we categorize all interferometric image pairs into those captured during months of high vegetation coverage and those from months of low vegetation coverage. We then remove the image pairs with coherence coefficients below the category average. Using the Small Baseline Subset InSAR (SBAS-InSAR) technique, we retrieve surface deformation information in the Yuanmou dry-hot valley. Landslide identification is subsequently verified using optical remote sensing images. The results show that significant seasonal changes in vegetation coverage in the Yuanmou dry-hot valley lead to noticeable seasonal variations in InSAR coherence, with the lowest coherence in July, August, and September, and the highest in January, February, and December. The average coherence threshold method is limited in this context, resulting in discontinuities in the interferometric baseline network. Compared with methods without baseline optimization, the interferometric map ratio improved by 17.5% overall after applying the OM-InSAR-BCCVC method, and the overall inversion error RMSE decreased by 0.5 rad. From January 2021 to May 2023, the radar line of sight (LOS) surface deformation rate in the Yuanmou dry-hot valley, obtained after atmospheric correction by GACOS, baseline optimization, and geometric distortion region masking, ranged from -73.87 mm/year to 127.35 mm/year. We identified fifteen landslides and potential landslide sites, primarily located in the northern part of the Yuanmou dry-hot valley, with maximum subsidence exceeding 100 mm at two notable points. The OM-InSAR-BCCVC method effectively reduces incoherent noise caused by vegetation coverage changes, thereby improving the monitoring accuracy of InSAR.
ABSTRACT
The morphological changes in mountain glaciers are effective in indicating the environmental climate change in the alpine ice sheet. Aiming at the problems of single monitoring index and low prediction accuracy of mountain glacier deformation at present, this study takes Meili Mountain glacier in western China as the research object and uses InSAR technology to construct the mountain glacier deformation time series and 3D deformation field from January 2020 to December 2021. The relationship between glacier deformation and elevation, slope, aspect, glacier albedo, surface organic carbon content, and rainfall was revealed by grey correlation analysis. The GA-BP neural network prediction model is established from the perspective of multiple factors to predict the deformation of Meili Mountain glacier. The results showed that: The deformation of Meili Mountain glacier has obvious characteristics of spatio-temporal differentiation; the cumulative maximum deformation quantity of glaciers in the study period is -212.16 mm. After three-dimensional decomposition, the maximum deformation quantity of glaciers in vertical direction, north-south direction and east-west direction is -125.63 mm, -77.03 mm, and 107.98 mm, respectively. The average annual deformation rate is between -94.62 and 75.96 mm/year. The deformation of Meili Mountain glacier has a gradient effect, the absolute value of deformation quantity is larger when the elevation is below 4500 m, and the absolute value of deformation quantity is smaller when it is above 4500 m. The R2, MAPE, and RMSE of the GA-BP neural network to predict the deformation of Meili glacier are 0.86, 1.12%, and 10.38 mm, respectively. Compared with the standard BP algorithm, the prediction accuracy of the GA-BP neural network is significantly improved, and it can be used to predict the deformation of mountain glaciers.
Subject(s)
Ice Cover , Snow , Climate Change , Neural Networks, Computer , TechnologyABSTRACT
In complex mountainous areas where earthquakes are frequent, landslide hazards pose a significant threat to human life and property due to their high degree of concealment, complex development mechanism, and abrupt nature. In view of the problems of the existing landslide hazard susceptibility evaluation model, such as poor effectiveness and inaccuracy of landslide hazard data and the need for experts to participate in the calculation of a large number of evaluation factor weight classification statistics. In this paper, a combined SBAS-InSAR (Small Baseline Subsets-Interferometric Synthetic Aperture Radar) and PSO-RF (Particle Swarm Optimization-Random Forest) algorithm was proposed to evaluate the susceptibility of landslide hazards in complex mountainous regions characterized by frequent earthquakes, deep river valleys, and large terrain height differences. First, the SBAS-InSAR technique was used to invert the surface deformation rates of the study area and identified potential landslide hazards. Second, the study area was divided into 412,585 grid cells, and the 16 selected environmental factors were analyzed comprehensively to identify the most effective evaluation factors. Last, 2722 landslide (1361 grid cells) and non-landslide (1361 grid cells) grid cells in the study area were randomly divided into a training dataset (70%) and a test dataset (30%). By analyzing real landslide and non-landslide data, the performances of the PSO-RF algorithm and three other machine learning algorithms, BP (back propagation), SVM (support vector machines), and RF (random forest) algorithms were compared. The results showed that 329 potential landslide hazards were updated using the surface deformation rates and existing landslide cataloguing data. Furthermore, the area under the curve (AUC) value and the accuracy (ACC) of the PSO-RF algorithm were 0.9567 and 0.8874, which were higher than those of the BP (0.8823 and 0.8274), SVM (0.8910 and 0.8311), and RF (0.9293 and 0.8531), respectively. In conclusion, the method put forth in this paper can be effectively updated landslide data sources and implemented a susceptibility prediction assessment of landslide disasters in intricate mountainous areas. The findings can serve as a strong reference for the prevention of landslide hazards and decision-making mitigation by government departments.