Landslide Risk Analysis Based On Multi-source Data

The characteristics of Guizhou karst area are obvious,the topography is complex,the rainstorm is frequent,which leads to the frequent occurrence of landslide disaster,which poses a serious threat to the safety of local people's lives and property.Therefore,it is of great practical significance to carry out regional landslide disaster risk assessment for disaster prevention and mitigation.In this paper,Panzhou City,which is located in the Guizhou Province,is taken as the research area.Combined with the geological hazards and computer-related theories.Adopting GIS and remote sensing as the main technical means,and using machine learning and statistical analysis as the main theoretical methods,to achieve the purpose of regional landslide risk analysis.This paper produced landslide susceptibility maps using random forest and logistic regression method based on the high-quality evaluation index system,and carried out the accuracy assessment of the susceptibility maps by the field validation and model quality valuation.Then,landslide hazard map was calculated by the landslide happening spatial location,temporal frequency and landslide magnitude.Subsequently,based on qualitative analysis methods,the assessing and analysis model of regional risk element are established.Finally,through the analysis of rainfall,land use and NDVI change,this paper forecasts the landslide hazard and risk change and trend under the condition of factors change.The main results are as follows:1.Analysis of regional landslide susceptibility.By correlation test and rough set methods,the paper establishes a high-quality evaluation index system,which includes altitude,lithology,distance from streams,rainfall,land use,distance from roads,distance from residential area and NDVI.The results of landslide susceptibility maps show that high susceptibility prone are mainly distributed in the areas with low vegetation coverage,intensity human social activities and density rivers.The accuracy assessment of the susceptibility maps were carried out by the area under curvature approach and field validation,the results show that the prediction results of random forest model not only has higher prediction accuracy,but also are matched with the actual situation than logistic regression model,moreover,the random forest model obtained the AUC value is 0.82 by the validation data.2.Risk analysis of regional landslide disaster.Firstly,the temporal frequency and magnitude of landslide are analyzed according to the landslide record.The analysis results show that the probability of recurrence of landslide with 4 to 5 magnitude is 0.52.Then,the landslide hazard map is obtained by the comprehensive calculation of the landslide susceptibility,temporal frequency and magnitude,the result shows that the landslide hazard is mainly affected by the landslide susceptibility results;the areas with high landslide hazard are mainly distributed in the Triassic and Permian strata;the areas with high landslide hazard have strong human activities and obvious distribution along both banks of the river.3.Landslide risk assessment for Panzhou city was carried out by the qualitative analysis method.The economic high risk areas mainly concentrate in areas of the large economic value: the urban areas,township streets and traffic construction land etc.The life high risk areas mainly distribute in the high landslide hazard areas and the densely populated areas.4.The change situation and trend of landslide risk and hazard in the study area were predicted under the variations of rainfall,land use and NDVI,and the population density of each towns.The results show that rainfall and NDVI index have increased.Construction land and waters land have increased,but cultivated land and forest land have decreased.Landslide hazard were in an increasing trend on the whole,but part of the areas has changed.Both of the economic and life risks were in an increasing trend.