Identification And Stability Analysis Of Potential Landslide Based On Random Forest And Fractal Theory

As a natural geological disaster,landslide has the characteristics of wide distribution,strong destructiveness and frequent occurrence,which often brings huge economic losses to human beings and threatens people ’ s life safety.With the continuous development of society,the destruction of human activities on the natural environment is becoming more and more serious,resulting in a sharp increase in the frequency of landslide disasters in recent years.Early identification of potential landslides and analysis of their stability can provide a scientific basis for landslide disaster warning,prevention and risk assessment.In view of the shortcomings of existing potential landslide recognition methods,such as incomplete consideration of landslide influencing factors,poor recognition accuracy and insufficient stability analysis of recognition results,this paper studies the potential landslide recognition and stability analysis based on random forest and fractal theory.InSAR technology is used to obtain time series deformation,and various landslide influencing factors are extracted based on multi-source data.The random forest method is used to construct the potential landslide recognition model,and the time series fractal theory is introduced to analyze and evaluate the stability of potential landslides.In this paper,the identification and stability analysis of loess potential landslides in Ledu-Minhe area of Haidong City,Qinghai Province are studied.The specific work is as follows:(1)Based on the survey data and related research results,the landslide mechanism in the study area is analyzed,and the topography,geological conditions,human activities,meteorological and hydrological conditions,vegetation cover are determined as the main influencing factors of loess landslide in the study area.(2)Based on the Sentinel-1A data of the study area from January 2015 to December 2016,the short baseline interferometry(SBAS)method is used to extract the temporal deformation of the study area.The slope with an average deformation rate higher than 10 mm / year and a slope gradient greater than 10°is added to the landslide sample.At the same time,considering that deformation is the external representation of slope mechanical instability,this paper regards surface deformation as the influencing factor of landslide.(3)Based on the geological data,optical remote sensing images,DEM,meteorological data and other data in the study area,14 landslide influencing factors were extracted,including elevation,slope,aspect,curvature,valley,terrain humidity index TWI,faults,folds,lithology,vegetation cover,houses,farmland,roads,and mines.Based on the known landslides and large-scale variable slopes in the study area,the landslide samples are established,and the random forest algorithm is used to establish the potential landslide identification model in the study area.Finally,53 potential landslides are identified in the study area.According to the visual interpretation results of Google Earth images since 2017,30 of the 53 potential landslides identified have formed landslides in the later period,and the recognition accuracy rate is 56.6 %.(4)Based on the time-series deformation data of the study area obtained by SBAS,the time-series fractal method in fractal theory is used to analyze the stability of the landslide evolution process of Gaojiawan landslide occurred during the identification period,which verifies the effectiveness of the time-series fractal method for landslide stability analysis.On this basis,the time series fractal analysis of potential landslide samples in the identified study area was carried out,and the stability of these potential landslides was evaluated,which was verified by the visual interpretation results of Google Earth images.