基于栅格数字高程模型提取特征地貌技术研究

AUTOMATED EXTRACTION OF THE CHARACTERISTICS OF TOPOGRAPHY FROM GRID DIGITAL ELEVATION DATA

本文对近年来基于栅格数字高程模型（ＤＥＭ）提取特征地貌技术进行了详细的研究，认为该技术的关键在于两个方面：一是如何定义地貌形态结构，二是提取算法的设计。本文提出了基于地貌学角度来定义地貌形态结构的方法，利用有限个数的形态要素的空间组合和对比分析来获取特征地貌，并可以对各种特征地貌形态进行符合物理意义的改进。

Automated extraction of the characteristics of topography from digital elevation models (DEM) is an important application technique used for qualitative analysis of remote sensing observations of the earth system, for positional analysis in GIS database construction, and for complex analysis of multi dimensional information and polymorphic information. Based on our study of the extraction techniques using DEM in recent years, we find that the key to their successful application lies in two factors. The first is how to adequately define topographic structure morphologically or hydrologically, and the second is how to design algorithms for the defined topographic structure. The definitions of and the relationships among geomorphic entities must be concerned with the most significant attributes of the landscape. Different definitions of a morphostructure will lead to different algorithm designs and different analytical results. Hydrologically based method employing many composites of thresholds to extract surface features may yield obscure and inaccurate results. On the other hand, morphologically based method using spatial composites and contrast analysis of finite morphologic elements can produce accurate results. In the last 20 years, two approaches to the identification of topographic features from DEM have emerged, one seeking to classify individual pixels or connected pixel regions on the basis of local surface characteristics, and the other attempting to accumulate more global surface information to build a more consistent structure of the surface features. The typical local method of pixel classification includes a hill, depression, cape, ravine, regular slope or saddle. These elements are included by inspecting the eight connected pixels adjacent to a cell and by computing the central pixel. An alternative to such a local method is to base the stream, divide, and basin extraction directly on the synthesized global information through pixel integration into larger structures. In this paper, we develop an algorithm based on global information. For experimentation, we constructed GIS databases and studied the ravine district in the loess hills of the Wangjiagou drainage area in Lishi, Shanxi Province. The primitive drainage catchments are generally characterized as having uniform aspect and elevation, arranged as leaves on the stream channel tree graph. Landform features extracted from the algorithm match the ground truth quite well. Topological operations on the channel network may be extended to other drainage catchments by building attribute files parallel to the topologic stream graph code.