| The landslide is a type of mountain deformation that occurs under the influence of external or internal forces,and is influenced by various triggering factors,including topography,geological environment,meteorological conditions,and human activities.Among these factors,meteorological conditions play a crucial role in triggering landslides.Currently,there are numerous studies that directly investigate landslides with meteorological conditions as the main research focus.However,there are relatively few studies that utilize atmospheric precipitation as an indirect environmental factor.In this study,remote sensing data and meteorological data were used to investigate the main influencing factors of landslides in Yunnan Province.Based on GNSS data,the temporal and spatial characteristics of rainfall and atmospheric precipitation were analyzed,as well as the correlation between rainfall,atmospheric precipitation,and landslides.Furthermore,a machine learning model was developed to predict the subsidence of landslides by considering multiple factors.The research content and analysis results are presented below..(1)Based on the analysis of the impacts on landslides caused by terrain,geological environment,climate conditions,and human factors.The characteristics of elevation,slope,aspect,lithology,water system,rainfall,precipitable water vapor,and road network were analyzed.The information value model was used to calculate the information value of each influencing factor,and the comprehensive information value of each region was obtained by raster overlay.The results indicated that the central and northeast regions of Yunnan Province were highly susceptible and moderately susceptible to landslides,where elevation,distance from rivers,rainfall,precipitable water vapor,and distance from roads had significant impacts on landslide occurrence.In the northwest region of Diqing Tibetan Autonomous Prefecture in western Yunnan,landslides were highly susceptible in the areas where different terrains intersected,and rainfall,precipitable water vapor,and distance from roads were the main factors leading to landslide occurrence.Given the unique distribution characteristics of landslide-prone areas in Diqing Tibetan Autonomous Prefecture,the study suggests that landslide prediction should consider displacement deformation,rainfall,and precipitable water vapor.(2)Based on GNSS and meteorological data,the characteristics of rainfall and precipitable water vapor changes in the study area were analyzed in detail at multiple spatiotemporal scales,including monthly and annual scales,using radial basis function interpolation,linear trend analysis,and curve trend analysis.The study showed that overall,the trend of rainfall and precipitable water vapor changes decreased from south to north,with the minimum change in rainfall being-0.52mm/30 d and the minimum change in precipitable water vapor being-0.14mm/30 d.The majority of the regions showed a decreasing trend annually.The correlation between landslides,rainfall,and precipitable water vapor was analyzed by combining deformation data in the region,using the Pearson correlation coefficient,gray correlation analysis,and combination weighting method.The study showed that the correlation between rainfall and precipitable water vapor can reach up to 0.79,the correlation between rainfall and landslides can reach up to 0.89,and the correlation between precipitable water vapor and landslides can reach up to 0.76,indicating that the occurrence of landslides is influenced by geographical location and meteorological conditions and has a strong correlation with rainfall and precipitable water vapor.(3)Based on the traditional machine learning model and the gradient boosting regression model,combined with the data of horizontal displacement,rainfall,and PWV,the prediction of the landslide settlement-time curve is conducted.The results show that the optimal gradient boosting regression model constructed can more accurately predict landslide settlement.By analyzing the test set data of the GZ01 monitoring point in August 2021,it is found that the relative error of settlement predicted based on the optimal gradient boosting regression model is within ±4mm.A further prediction of landslide settlement based on the GZ02 monitoring point shows that most of the residual results are distributed within ±5mm,indicating that this prediction model has high prediction accuracy and certain applicability to landslide settlement prediction. |
PDF Full Text Request