Multiresolution Land Surface Segmentation And Generalization Of Dem: Tentative Searching For The Optimal Settings Of Detecting Elementary Forms

Before applying a land surface segmentation algorithm, generalizing the input digital elevation model (DEM) can help optimize the multiresolution land surface segmentation. We tested an indicator of optimal generalization of DEM (the index K-0 (G(nn))) and the setting of the scale parameter (SP) in the multiresolution segmentation by geographical object-based image analysis (GEOBIA). The concept of elementary forms was used as a theoretical base. We tested the utility of the estimation of scale parameter (ESP2 tool) for the estimation of SP. The DEM was generalized by approximating land surface with a fourth-order polynomial using the least-squares method for various sizes of a moving window. Altitude, slope, aspect, normal slope line (profile) curvature, and contour line (plan) curvature were used as the input variables of segmentation. The results from the three areas of interest differing in morphology show a clear dependence of local variance (LV) on the average segmentation area (ASA) and generalization level. The relationship between LV and ASA can be effectively used to determine the most suitable SP, however, the ESP2 tool is not ideal to define the LV-ASA relationship. The relationship better defines the nested hierarchy of land surface in case the DEM is generalized for the optimal generalization level (i.e., windows size). The optimum DEM generalization for detecting elementary forms of a particular geomorphological hierarchical level was achieved by using the moving window size for which the index K-0 (G(nn)) reaches the local maximum.