Landslide Susceptibility Evaluation Method Based On Multi-source Data And Scale Segmentation

Landslide is a common geological disaster,which has the characteristics of high frequency,wide distribution and strong destruction,and has a great impact on human production and life.China is one of the countries greatly affected by landslides.There are huge casualties and property losses caused by landslides every year.The process of landslide susceptibility assessment is to use the characteristics of regional landslides to predict the future development trend of evaluation units in this region,and the prediction results can provide a scientific basis for landslide risk management,urban planning and landslide monitoring.The existing landslide susceptibility evaluation methods have a certain degree of reliability and practicability.To a certain extent,they solve the location and possibility of landslides.However,landslides are a natural phenomenon,and the environment and causes of landslides are changeable.Is a dynamic system.The subjective definition of humans and the complexity of the occurrence of landslides together lead to some problems in the evaluation of landslide susceptibility:(1)The process of multi-source data processing is unclear.(2)There is no analysis of the reasons for the influence of scale segmentation on the assessment of susceptibility.(3)There is no method to analyze the selection of non-landslide units.Based on this,firstly,the paper takes Shangyou County as the study area,on the basis of basic geological data,through the methods of on-site exploration and remote sensing interpretation,to investigate the geological disasters in Shangyou County.Secondly,according to the low-information area,select non-landslide units to construct an information-neural network/support vector machine coupling model.Based on the slope unit and terrain unit,the coupling model is verified and compared with information-neural network/support vector machine(ANN/SVM).Thirdly,the geographically weighted regression is used to segment the study area,and the coupling model is used to verify and evaluate each partition.Finally,the influence of the new coupling model and scale segmentation on the vulnerability assessment is analyzed,and the reasons for its influence are analyzed,and the better landslide vulnerability evaluation method is obtained by comparison.The main contents and results of this paper are as follows:(1)Based on the combination of basic geological data and field exploration data,the development characteristics of landslides are summarized.Different data formats,different coordinates,different file formats and different forms of multiple data are transformed into a unified framework of coordinates,formats and forms.Through data processing,16 landslide influencing factors are extracted,correlation analysis and weight calculation are carried out,and 14 weak correlation factors with large weights are selected to provide accurate data support for landslide vulnerability evaluation.(2)The information-ANN/SVM coupling model is established,and Shangyou County is taken as an example to evaluate the landslide vulnerability of the coupled model.Compared with the ANN/SVM landslide vulnerability evaluation accuracy,the results show that the evaluation accuracy of the coupling model is higher,and the accuracy is improved by 3% and10%.(3)It is proposed to use geographically weighted regression(GWR)to divide Shangyou County into regions and obtain local regions.Based on the slope unit and the terrain unit,the landslide susceptibility is evaluated on the local and regional scales.The analysis results show that after the scale segmentation,the evaluation of local scale is helpful to improve the accuracy of vulnerability evaluation.(4)Analyze the reasons for the influence of region segmentation and coupling model on the accuracy of susceptibility.The factor weight curve and weight bar chart of different regions are made by Origin software,and the chart shows that the weights of different regions are different,and the regional evaluation leads to the phenomenon of "average" of evaluation factor weights.It can be seen from the research results: regional susceptibility evaluation averages the influence of "spatial heterogeneity" to each grid in the region,and the influence weight of each evaluation factor on the local scale can better represent the influence degree of landslides in this region.During the local area evaluation process,it is found that the parameters of the model change with the change of training set and test set.This paper studies the influence of the changes of model parameters under different zones on the accuracy of model classification.Calculation results of Kappa coefficients by confusion matrices show: After region segmentation,the classification of the model is accurate,and the error rate is low,which makes the accuracy of vulnerability evaluation higher,verify the accuracy and significance of region segmentation.