Hazard Assessment Of Landslide Based On Multi-source Heterogeneous Data Fusion Algorithm

Landslide poses a great threat to the safety of human life and property,and also causes serious damages to the ecological environment.Therefore,it is of great economic value and social significance to study effective landslide disaster evaluation landslide susceptibility assessment and landslide hazard assessment is to evaluate the geographical spatial distribution of potential landslide disasters by analyzing the relevant factors affecting landslide,to provide decision support for urban construction planning and landslide prevention.In this study,Shenzhen is taken as the research area,multi-source heterogeneous data and a variety of intelligent algorithms are used to evaluate the landslide susceptibility assessment,and the landslide assessment susceptibility evaluation results,historical landslide data,and rainfall data are used to realize the landslide hazard assessment based on rainfall-induced factors.Aiming at the multi-source heterogeneous data used in the study,a unified multisource heterogeneous data processing framework is established to extract five types of evaluation indexes,such as topography,landform,and human activities.According to the characteristics of urban geological hazards,the normalized light intensity index and population density are introduced to reflect the impact of human activities.To solve the problem of attribute redundancy,the pearson correlation coefficient and the variance inflation factor are used to perform collinearity analysis on the evaluation indexes,and the 28 evaluation indexes extracted are reduced to 18,to achieves attribute reduction.According to the correlation analysis of impact factors and landslides,the top five correlations between landslides are slope,land classification,VANUI,distance from buildings,and MNDWI.Aiming at the problem that it is difficult to select non-landslide points in landslide susceptibility and hazard assessment,a method is proposed to generate a landslide susceptibility map based on the statistical model of CF,and then select non-landslide points of low-risk areas.Compared with randomly generated landslide points,the accuracy can be improved by 11%.Aiming at the problem that it is difficult to select the best model in landslide susceptibility and hazard assessment,the paper introduces the ensemble learning algorithm Stacking to integrate the neural network,random forest,support vector machine,naive bayes and logical regression,and applies them to landslide susceptibility assessment and landslide susceptibility map generation,it is found that the integrated model Stacking has better performance than other single models,and the accuracy rate of the integrated model is 5% higher than other models.Using the landslide case of June 13,2008,and rainfall data onto the first six days in the study area,the calculation model of early rainfall and effective rainfall is established,and the landslide hazard assessment based on rainfall inducing factors is realized.It is found that the indicators based on the operational rainfall model to perform best,and the accuracy rate of other models can be increased by up to 5%.In view of different rainfall and sample conditions,it is found that different rainfall and sample number will have a greater impact on the prediction results,indicating that the influence on impact factors of landslide will change into the change of other conditions.