Study Of Land Distruction Susceptibility Mapping At County Scale In Loess Hilly And Gully Districts

As China's cultivated land area approaches the red line of 120 million hm2,land issues have become the focus of the whole society,and the scarce available land and fragile ecological environment due to the high loess have become China's key areas for land engineering and environmental protection research.The Loess Hilly and Gully Region is one of the typical sub-regions of the Loess Plateau.It faces complex land problems,especially the problem of slope damage caused by water erosion in the loess channel,and soil erosion and erosion.Geological disasters such as collapse,landslides,and mudslides caused by the action seriously threaten the scarce available land and restrict social and economic development.It can be seen that the research on the damaged land in the loess hilly and gully area is particularly important.In view of this,this paper selects the typical loess hilly and gully area---Zhidan County as the research area,fully collects regional geology,geomorphology,climate,hydrology,ecology and other data,and develops it based on remote sensing interpretation and ground investigation data.The zoning of disaster-prone lands in Zhidan County.The main research contents and results of the paper are as follows:1.The background of the regional geological environment of the study area is elaborated in detail.From the morphological characteristics of the damaged land and the geological environment in which it is located,the development and distribution rules of the damaged land in Zhidan County are comprehensively analyzed to extract the trenches for the study area.And the zoning modeling of disaster-prone lands provides the research basis.2.The actual starting points of 75 channels in the study area were interpreted based on remote sensing images.Based on this,a histogram statistical analysis method was used to determine the actual starting point of the channel.The diagnostic value of the sink flow on the loess canal was analyzed using the Euden index in the ROC curve.Finally,the optimal threshold of the sink flow for the loess canal extraction using the method of hydrological analysis was finally determined,thereby accurately extracting the loess ditch in the study area.The accuracy of the extraction results was evaluated.3.The topographic mapping unit of the study area was prepared based on the hydrological analysis module in the Arc GIS platform.The pixel mapping unit was selected as another basic mapping unit.Extract elevation,plane curvature,profile curvature,slope,surface roughness,annual average rainfall,formation lithology,channel buffer,water buffer,runoff intensity index,NDVI,land use type based on two mapping units,Residential buffer zone,road buffer zone,aspect and terrain humidity index to establish an evaluation index system.A quantitative evaluation index based on the box dimension was proposed.At the same time,the frequency density was used as a control.The relationship between the evaluation index and the damaged land was analyzed in detail using three information quantities,the proportion of the hierarchical grid,and the proportion of the damaged land.Association situation.4.In order to avoid mutual interference between the evaluation factors,the Pearson correlation coefficient and variance expansion factor between the induced factors are calculated to detect the multicollinearity problem of the disaster-induced land induced factors,and the factors causing the interference are excluded.Then,the information gain rate of each inducing factor is calculated,the factors with a contribution degree of 0 are excluded,and the screening of the inducing factor is completed.Finally,a modeling data set based on cell mapping unit and frequency density(data set 1),a modeling data set based on topographic mapping unit and frequency density(data set 2),a cell mapping unit based on cell mapping and box dimensions Modeling dataset(Dataset 3),modeling dataset based on terrain mapping units and box dimensions(Dataset 4).5.Index of entropy model(IOE),support vector machine model(SVM),kernel logistic regression model(KLR)and radial basis kernel function neural network model(RBFNN)were selected as the basis for modeling the susceptibility of disaster-damaged land in the study area.Using the verification sample sets in data sets 1,2,3,and 4,the 10-fold cross-validation method was used to optimize and adjust the(C,?)parameters of the radial basis kernel function in the machine learning model.At the same time,the guided clustering algorithm(Bagging)was selected to conduct integrated modeling of the three machine learning models,and the B-SVM model,B-KLR model and B-RBFNN model were constructed respectively.Finally,the susceptibility mapping of disaster-damaged land in the study area was completed.6.In order to achieve the comparison and verification of the comprehensive evaluation partition results and the model,a variety of statistical indicators are used to evaluate the classification results of the model,and the ROC curve and the SCAI seed verification method are used to comprehensively compare and analyze each classification model.The conclusion is as follows:In the study area,the topographic mapping unit is used as the basic mapping unit,and the input data set prepared using the box-dimensional quantization inducing factor is not prone to multicollinearity problems,and the diversity of data structure is maintained to the greatest extent.Because terrain mapping units can be combined with landforms,geological environments,and geological boundaries,each classification model performs better on topographic mapping units than pixel mapping units.The performance of the machine learning model is generally better than the binary statistical model,and the classification ability of the machine learning model after integration modeling is also due to the original model.Finally,it is concluded that the B-RBFNN model is more suitable for the susceptibility zoning of damaged land in the hilly and gully regions of the Loess Plateau.