Study On The Evaluation Methodology Of Landslide Susceptibility Based On Multi-scale Analysis

Landslides are large adverse natural geological hazards.Due to its unique location,C hina has a variety of landform types,of which mountains are the largest type of landmark.And,being affected by tectonic movements,landslides and geological disasters occur frequently.Landslide susceptibility evaluation refers to the quantitative evaluation and prediction of the possible locations and probabilities of landslides.It is the focus of current landslide research to leverage the prediction results to prevent and control landslides.Methods for assessing landslide susceptibility can be divided into two categories.One is a deterministic approach to predicting the probability of landslides in a particular area by combining rock and soil mechanics models with basic spatial geologic information.The other is to use computer and mathematical statistical methods to study the characteristics of the geological environment at the location of the landslide,and to obtain similar or similar high-risk areas of the geological environment in the study area,so as to achieve an uncertain assessment and prediction method.Since the late 1990 s,with the "3S"(GPS-Global Positioning System;RS-Remote Sensing;GIS-Geographical Information System)in regional disaster research has provided more efficient and accurate means of obtaining relevant data and information such as landslide deformation characteristics,landslide distribution and landslide influence factors,which has greatly promoted the progress of non-deterministic methods of landslide susceptibility assess ment.In addition,the issue of scale in landslide susceptibility assessment ha s begun to enter the scope of research by domestic and international researchers.In this regard,researchers often use scale as an initial condition for landslide susceptibility assessment,and select evaluation models and evaluation factors based on scale,without noticing that different evaluation results will be obtained with different scales.Excessive association of scale issues with the study scope,evaluation units or spatial resolution of the data is not done in the form of independent scientific questions.The internal relationship between the different scales and the evaluation results of landslide susceptibility is omitted,and the relationship between the different scales and the evaluation factors of landslide susceptibility and the parameters of the evaluation model is not explored,so it is impossible to fundamentally explore how scale issues affect the evaluation results of landslide susceptibility.To address the above issues,landslide susceptibility assessment was carried out in Lixian County.Based on related theoretical and technical approaches from computer science,surveying and mapping,geology,statistical analysis and other disciplines,landslide susceptibility assessment methods based on multiscale analysis have been studied.The landslide susceptibility evaluation is done by combining two evaluation models,global and local scale,CNN model and PSO-CNN coupled model.By comparing the results of the four evaluations,we quantitatively analyze the internal relationship between the evaluation results,the evaluation factors and the evaluation model parameters for landslide susceptibility at different scales,and conclude the paper as follows.(1)Multi-source data were collected and collated based on Arc GIS platform,and 14 landslide evaluation factors were extracted by analyzing the characteristics and attributes of various data and combining with the relevant data of Lixian County.A coupled model of PSO-CNN is constructed and the optimal solution of CNN model parameters is searched by PSO algorithm to improve the accuracy and efficiency of model training(2)The multi-scale analysis method is applied to the whole process of landslide susceptibility evaluation and the idea of geographical weighted regression is used to solve the problem of local segmentation.The evaluation scale is divided into a global scale and a local scale,and we discuss the differences in the evaluation results under the two scales.(3)A variety of methods were selected and used to quantify the differences between the global scale and local scale landslide susceptibility evaluation results,including specific category accuracy analysis,overall prediction accuracy analysis,ROC curve analysis.(4)The influence degree of global and local landslide susceptibility evaluation factors on landslide and the optimal solution of landslide susceptibility evaluation model parameters at different scales were analyzed.It is shown that the importance of the landslide evaluation factor at the global scale is close to the average value of each local landslide evaluation factor.The number of samples at the local scale is small compared to the global scale and they share some of the same features.Therefore,parameter optimization of the model evaluated based on sample training can achieve better results.