The applicability of artificial intelligence methods for the selection/elimination process to the stream networks in cartographic generalization

In this study, a new method was developed using Self Organizing Maps (SOM) method for the selection/elimination process in the generalization of stream networks and it was compared to Support Vector Machines (SVM) method. The most suitable attributes to be used as input to the SOM and SVM were explored. The attributes were weighted in accordance with the associations determined in chi-square independence test. The Radical Law was used in the determination of the number of features should be selected. An incremental approach was developed for the selection of the clusters in the SOM. In the case study, two different stream networks which have different drainage patterns (dendritic, trellis and rectangular) at 1:24,000-scale in the National Hydrography Dataset of United States Geological Survey were used in order to derive the stream networks at 1:100,000-scale. Derived stream networks are quite close to the original stream networks at 1:100,000-scale from both quantity and visual aspects. Stream density and drainage pattern were mostly maintained in each subbasin by SOM method. Continuous and semantically correct networks were obtained.