The Quantification And Geomorphological Classification Of Ridgeline Based On Graph Theory

The geomorphological types in China are complex and affect the production and life of human beings,while the geomorphological classification plays an important role in the scale and layout of human construction activities.At present,most researches on geomorphic classification are based on Digital Elevation Model(DEM)to extract its topographic indexes,and to determine its geomorphic types through combination quantification.The topographic feature lines directly express the topographic skeleton and depict the landform.How to classify landforms based on topographic feature lines is worth further exploration.According to the principle of graph theory,terrain feature lines can be regarded as networks or graphs.Can quantitative indicators be designed on the basis of graph or network to describe the difference of terrain and then implement terrain classification? If progress can be made,it may become a new Angle and new direction of geomorphic classification research.Based on the typicality,representativeness and scientificity of the sample areas,this paper chooses Shaanxi and Gansu provinces as the study areas.The core area of the study is located in Shaanxi Province,mainly concentrated on the Loess Plateau and the Qinling-Daba Mountains.From north to south,the landform types include plateau,hill,plain,basin and mountain,etc.The sample area is rich in landform types,which is suitable for the study of landform classification.In order to compare the geomorphic stability between small watershed boundaries and large area samples,small watershed samples of different geomorphic types and large area samples in the core region were used to determine the characteristic scale of analysis using multi-scale spatial autocorrelation,and ridge lines were extracted.Based on the principle of graph theory,topological parameters were developed to quantitatively describe the ridge features.Cluster analysis is used to explore the quantification features of ridge lines based on graph theory and its application in geomorphological classification.The main contents of the study and the main conclusions are as follows:1.The positive topography and its contour line are similar to the ridge line to some extent,and the analysis scale of positive topography is very important to the ridge line.Firstly,the positive topography regions of different landform types were calculated,and then transformed into vector-based positive topography.The incremental spatial autocorrelation analysis was carried out on the positive topography,and the characteristic scale of the study area was determined according to the significance index.The results show that the best scale is generally 40×40 in the Qinling-Daba Mountain region,and 33×33 in the Loess Plateau region.This indicates that the incremental spatial autocorrelation analysis is helpful to determine the optimal analysis window for ridge line extraction2.According to the principle of graph theory,the ridge line is regarded as a tree graph,which is quantitatively analyzed to study and determine the appropriate graph theory parameters;A series of parameters including entropy,edge adjacency index,height difference adjacency index,number of graphs,density of graphs,Randic X,point-line connectivity,graph complexity,and elevation mean,standard deviation and range of the largest subgraph are constructed.Based on principal component analysis,the selected graph theory parameters are processed and the main parameters are retained.Clustering analysis was carried out by using the method of system clustering and K-means clustering,and the similarities and differences of the two were compared.The results showed that:In terms of classification parameters,in small watershed geomorphic sample area,the same method,the classification effect of 11 parameters and 7 parameters can achieve the same effect in some categories.As for the landforms of large sample areas,the results of the four parameters and the 11 parameters are consistent when the landforms are divided into two categories,and the loess plateau landforms and the Qinling-Daba mountain landforms can be well identified.However,the 11 parameters can further identify the beam-shaped low mountain landform,while the 7 parameters cannot.Comparatively,the accuracy of 7-parameter classification is not as good as11-parameter classification.In a large sample area,the complexity of landforms is higher,and there may be more than one landform,which requires more comprehensive quantitative parameters to summarize.However,although the parameters determined by principal component analysis summarize the vast majority of the information of classification parameters,they still cannot replace all of them.As far as clustering methods are concerned,the results are different among different methods.The first classification method of systematic clustering can effectively identify four landform types: loess tableland,loess hillock,middle mountain.K-means clustering can identify three landform types,namely,middle-alpine landform,alpine landform and loess tableland landform.This shows that the classification method of hierarchical clustering is better than K-means clustering.In terms of the classification area,the landforms of the Loess Plateau and the Qinling-Daba Mountains can be recognized by several classification methods in the large sample area,but the landforms of small watershed cannot be effectively distinguished.It is suggested that the large-scale geomorphic sample area can well identify the macroscale geomorphic types such as the Loess Plateau and the Qinling-Daba Mountains,which provides a direction for exploring the macroscale geomorphic classification3.When the landform classification and prediction were divided into two categories(Loess Plateau landform and Qinling-Daba mountain landform)by random forest,the classification accuracy reached 95.4%.The accuracy of classification is 80%when it is divided into four categories(Qinling-Daba mountain,Loess ridge,Loess hilllock,Loess tableland).