| The southwestern mountainous area is a seismic zone and a high-risk area for landslides and geological disasters in China.The increasing frequency of human engineering activities over the recent years has further enhanced the risk of geological disasters.To address this,the country has invested a large amount of resources in the treatment of geological disasters,achieving significant results.However,there are still some large-scale landslides occurring at high altitudes,causing malignant events such as blocking rivers,burying villages,and damaging roads and tunnels.Therefore,achieving the early identification of large potential landslides and transforming landslide disaster prevention from passive to active is an important task.Field surveying is considered as of the most effective and accurate means for the early identification of potential landslides is field surveying.However,this method can be costly,time-consuming,and poses safety risks.In recent years,the identification of potential landslides based on remote sensing technology has received much attention.With the advancement of technology,the application of multi-source remote sensing images has gradually become highly systematic,forming advanced systems.However,high-precision images are expensive and not conducive to the rapid identification of potential landslides over a large area.Based on previous research,this article employs an existing landslide disaster database in the study area and adopts techniques such as landslide dynamic susceptibility assessment,SBAS-In SAR(Small Baseline Subset Interferometric Synthetic Aperture Radar)analysis,multi-source remote sensing interpretation,UAV(unmanned aerial vehicle)applications,and field verification to achieve the rapid identification from a face to a point and from a point to an individual.Based on these techniques,a chain identification system for the early recognition of large potential landslides in seismic mountainous areas is established.The key research content and results of this paper are as follows:1.A chain-based method for identifying potential large landslides in earthquake-prone mountainous areas is proposed.This approach is a rapid identification method for large potential landslides that progresses layer by layer and from surface to point,and is designed for a large research area.The method includes four identification steps: i)the dynamic susceptibility analysis of landslides(first-level identification);ii)the comprehensive overlay analysis and identification of large potential landslides(second-level identification);ii)the identification of geomorphological features of large potential landslides(third-level identification);and iv)the investigation and verification of large potential landslides(fourth-level identification).This creates a chain-based identification system for large potential landslides.By effectively combining and improving traditional landslide area analysis and the complex identification process of individual landslides,the proposed method addresses the difficulties of early identification,slow identification speed,large workload,and the high costs of identifying large potential landslides associated with previous approaches.The potential large landslide chain-based identification method was applied in the research area of the Maoxian segment in the upper reaches of the Minjiang River according to the four-level identification steps,successfully identifying three large potential landslides.2.The identification of large potential landslides in the research area from surface to point was achieved by combining the results of dynamic susceptibility analysis with SBAS-In SAR.The dynamic susceptibility analysis of landslides introduces the cumulative deformation of SBAS-In SAR as dynamic conditional factors based on traditional susceptibility analysis.A total of 16 conditional factors were selected as the conventional conditional factors,taking into account the geological conditions of the terrain,as well as the topography,hydrology,lithology,structure,earthquakes,and human engineering activities in the research area.Each of the dynamic conditional and conventional conditional factors were modeled separately to obtain dynamic susceptibility analysis results across time.The results revealed the very high susceptibility and high susceptibility areas to increase each year,accounting for 43.54% in 2017 and increasing to 50.73% in 2021.This paper adopts an improved data mining method,namely,the Mahalanobis Distance Support Vector Machine(M-SVM)based on metric learning,to develop easy-to-use software for dynamic susceptibility analysis that can be used to rapidly perform multi-temporal dynamic susceptibility analysis.To characterize the susceptibility value change rate of each evaluation unit,we established a susceptibility gradient change level map.The higher the susceptibility value growth rate,the more likely it is to trigger a landslide.By obtaining the susceptibility value change level from this map,potential landslide individuals can be quickly identified to assist in completing the second-level identification.3.The second level identification of potential landslides was rapidly performed by overlaying the latest dynamic susceptibility analysis results,susceptibility gradient change level map,and SBAS-In SAR deformation rate map.A total of 18 potential landslides were identified through this step.The susceptibility analysis results distinguished the key identification areas and the susceptibility gradient change level map identified potential landslides with no significant deformation on the SBAS-In SAR deformation rate map.Furthermore,the SBAS-In SAR deformation rate map identified potential landslides with no significant increase in susceptibility values.These three methodological steps complemented each other and through their comprehensive analysis,large potential landslides were identified by maximizing the key identification process.4.Based on multi-source high-precision remote sensing images,a targeted interpretation was conducted on the results of the secondary identification to complete the tertiary identification and identify seven potential landslides.Compared with large-scale remote sensing interpretation,the tertiary identification step can save interpretation costs and reduce the workload.The seven potential landslides identified included the residual body of the Xinmo landslide,which is still undergoing deformation.The deformation of the Quaternary sediment deformation body in Tuanjie Village has been monitored by relevant units,which directly verifies the reliability of the identification results.Subsequently,a fourth-level identification was completed through on-site verification,and it was finally determined that the Luoduizhai potential landslide,the Tuanjiecun potential landslide(the confirmed deformation body),and the Liuducun potential landslide were large-scale potential landslides,while the others were not.Thus,the large-scale potential landslide chain identification method based on an existing landslide database proposed in this paper successfully achieved the early identification of large-scale potential landslides and its reliability was proved.5.The analysis of deformation evolution trends for typical potential landslides was performed.This paper introduced a neural network algorithm based on time series data,denoted as LSTM(Long Short Term Memory).This algorithm combines the cumulative deformation values of potential landslides analyzed by SBAS-In SAR and the deformation points around them with factors such as the high-precision slope,aspect,elevation,lithology,and distance from roads,to conduct multidimensional learning and predict the future deformation trend of potential landslides.Compared with traditional single-dimensional neural network learning,the results of multidimensional learning analysis have a higher accuracy and also partially solve the problem of predicting potential landslide deformation in the absence of actual deformation monitoring data.The results reveal the potential landslide in Luoduzhai to be a push-type landslide,with a sinking deformation area yet relatively stable surroundings.The potential landslide in Tuanjie Village is determined as a pull-type landslide,and the deformation area shows a downward trend while the surrounding area exhibits an upward trend within the next three years.The deformation range of both potential landslides is not observed to significantly increase. |
PDF Full Text Request