Analysis Of Landslide Susceptibility In Shaanxi Province Based On SOM-CNN

Geological disaster will bring immeasurable economic loss.The landslide's multiple occurrences,rapid development,and significant impact make it an important disaster in geological disasters.In all the geological disasters that occurred throughout the country over the years,the proportion of landslides is higher than 50%.Among the6181 geological disasters in 2019,there were 4220 landslides,with a proportion of up to 68.3%.From the perspective of the spatial distribution of geological hazards,Shaanxi has always been one of the most severely affected areas in China.Therefore,it is particularly important to understand the characteristics of landslides in Shaanxi Province,explore the distribution of landslides,analyze the susceptibility of landslides,and provide theoretical support for the monitoring and forecasting of landslides and effective prevention.Based on the data of landslide disasters in Shaanxi Province,the paper obtained as many landslide impact factors as possible according to the characteristics of Shaanxi Province.From the selection of evaluation unit to the comparative analysis of models,the process of constructing the landslide susceptibility analysis model in Shaanxi Province is elaborated in detail.A SOM-CNN model is constructed to partition the calculation results of landslide susceptibility in Shaanxi Province,and the distribution of each partition is briefly analyzed.1.Obtained raw data such as GDEMV2 data,MODIS remote sensing images and weather.Twenty three influencing factors were extracted from it,including seven topographic and geomorphic factors such as slope,aspect,and valley depth,three geological factors such as formation lithology,soil type,and fault,seven hydrological factors such as TWI,Melton coefficient,and flow path length.And factors related to NDVI,earthquakes,precipitation,human engineering activities.And by grouping the impact factors to count the number of landslide points,a brief analysis of different impact factors and the statistical law before the occurrence of landslide disasters.Pearson correlation coefficient was used to perform correlation analysis on the twenty three initially obtained impact factors,and then random forest was used for importance analysis to screen out fifteen relatively important and weakly correlated landslide impact factors.The highest of valley depth is 0.28,and the lowest of land use type is 0.01.2.Selecting two clustering algorithms of K-means and SOM neural network to partition the landslide susceptibility in the study area.The comparison shows that although the K-mean accuracy is higher,the partition is not reasonable.Therefore,the result after SOM clustering is selected as the basis for negative sample selection.Use:(1)randomly select negative samples from the extremely low susceptibility areas after SOM clustering,(2)randomly select negative samples from other areas except the extremely high susceptibility areas after SOM clustering,(3)randomly from areas that do not contain landslides Select negative samples.These three methods are used to obtain the sample set for model training and precision analysis.Finally,the method with higher accuracy and higher partition rationality is selected as the negative sample selection method in this study.3.On the basis of SOM network clustering,constructing a one-dimensional convolutional neural network as a model for landslide susceptibility analysis and mapping in Shaanxi Province,and compare it with the analysis results of support vector machine model and random forest model.The application accuracy of the three models is analyzed,and it is found through accuracy analysis that the three models are effective and reasonable for the analysis of landslide susceptibility in Shaanxi Province.However,comprehensively looking at the various indicators,the SOM-CNN model has higher prediction accuracy and zoning rationality than the other two comparison models.4.Analyzing the susceptibility of landslides in Shaanxi Province briefly.The susceptibility of landslides in southern Shaanxi has a tendency to decrease with the increase of DEM.In addition,the influence of faults on landslides in Shaanxi Province is also obvious.The high-prone areas in central Shaanxi basically coincide with the trend of faults.The high incidence areas in northern Shaanxi are more relevant to rivers and earthquakes.However,land use type,soil type and aspect have little effect on the overall landslide susceptibility in this area,which is consistent with the analysis of the importance of impact factors.