Uncertainty And Optimization Of DEM Generalization

In the study of digital terrain analysis, many methods are applied to extract the important information from digital terrain model. The information includes the basic topographical parameters(such as slope, aspect and curvature), the hydrological data(such as flow direction, flow accumulation and watershed), and the terrain features(such as valley line, ridge line, peak and saddle). The existence of effect of scale in spatial data makes the different scales have different terrain presentation. However, for many study regions, only one single or a small amount of data models(DEM, Digital Elevation Model) of the specific target scales can be used. The lack of data makes it hard to do terrain analysis in more abundant scales. In order to solve this problem, the most important thing is to construct the multi-scale terrain models, and DEM generalization can do this task well.The DEM generalization exists uncertainty. This uncertainty is caused by two aspects, the first one is the elevation error of original terrain model(such as original DEM), and the second one is the model of generalization. For investigating the influence of the uncertainty caused by original terrain model error, one elevation error field is generated based on two terrain models of the same region, which have different elevation accuracy, and the sequential Gaussian simulation is used to generate a series of error field simulations, these simulations are added to the original terrain in order to generate the simulations of the original terrain. For investigating the influence of the uncertainty caused by model of generalization, three feature-based models are selected to generalize the original terrain, and the results are analyzed. The methods of analysis for the generalized results include two parts, namely, qualitative analysis and quantitative comparison. For qualitative analysis, the contours of the original terrain and the generalized terrain are extracted and overlaid. And for quantitative comparison, the elevation accuracy and the degree of skeleton maintenance of generalized terrain are calculated. The results show that the uncertainty is small when the model of generalization can utilize more original features.The uncertainty of terrain generalization will propagate to the generalized terrain, which makes the real generalized terrain of the specific scale unknown. The traditional models can only obtain single result for the specific target scale, while it is not rigorous to calculate the generalized result, which has uncertainty, with the certain solution. So in the last part of our research, one new DEM generalization method, based on spatial simulated annealing, is proposed. For this new method, one composition of points are determined to reconstruct an initial generalized surface, and the initial surface will be optimized based on a fixed objective function, which acts as the quality criterion of generalized terrain, the final result is obtained when the quality criterion cannot be improved anymore. This characteristic makes the new method have the ability of searching the optimal generalized terrain surface. It means when the quality criterion fixed, the generalized result which is the best terrain representation under this criterion, will be found. The results show that the new method can accomplish the generalization optimization in different kinds of terrain surfaces finely and consistently.