Doctoral Dissertation Research: An Investigation of the Scale Effects of DEM-based Fuzzy k-means Landform Classifications

本项目主要研究模糊k-means地形分类中空间尺度的三个相互关联的组成部分：分类的空间分辨率、属性计算的空间分辨率以及邻域窗口大小对计算属性的影响。这些尺度分量在利用该分类方法描述基于地形属性和生物物理属性之间存在的遗传联系的连续生物物理模式时具有潜在的重要影响。该项目包括三个实验。第一个实验比较了在10到200米的10种分类分辨率下进行的10种分类。第二个实验将不同的属性分辨率（10 ~ 200 m）合并为统一的分类分辨率（10 m），然后比较这些结果。第三个实验比较了使用不同邻域窗口大小计算地形属性的分类效果。该项目将使用三种方法来比较模糊k均值分类结果。首先根据中心属性的相对差异计算两类中心的属性距离；第二部分分析了去模糊化后两类之间的硬类重叠程度；第三步检验由一个隶属度分布减去另一个隶属度分布得到的剩余隶属度曲面。实验结果表明，模糊k均值地形分类对所选空间尺度的敏感性；分类如何在不同空间尺度上定义不同的景观格局；被测地形属性如何对尺度变化做出不同的响应；研究区是否存在尺度阈值或自然边界。该项目的科学价值在于，模糊k均值分类可以潜在地使用广泛可用的数字高程模型（DEM）来描述环境特征的地理变化，而这些特征通常极难测量。对空间尺度问题的关注是建模过程的核心，因为地形表面和环境在不同的空间尺度范围内变化，而环境特征的描绘依赖于适当的空间尺度选择。研究结果也将有助于提高我们对地理学、地理信息科学（GIS）和环境建模中尺度问题的认识。作为博士论文研究改进奖，该奖项还将为计划在未来独立从事这些和类似研究问题的年轻学者提供支持。

BCS-0425273John P WilsonYongxin DengUniversity of Southern California This project focuses on three connected components of spatial scales in fuzzy k-means landform classifications: the spatial resolution of the classification, the spatial resolution of the attribute calculation, and the effects of size of neighborhood window on calculated attributes. These scale components have potentially important impacts when using this classification method to delineate continuous biophysical patterns based on the genetic linkages existing between terrain attributes and biophysical properties. The project includes three experiments. The first experiment compares 10 classifications performed at 10 classification resolutions ranging from 10 to 200 m. The second experiment incorporates different attribute resolutions (ranging from 10 to 200 m) into a uniform classification resolution (10 m), and then compares these results. The third experiment compares the classification effects of using different sizes of neighborhood windows to calculate terrain attributes. The project will use three methods to compare the fuzzy k-means classification results. The first calculates the attribute distance of two class centers based on the relative differences of center attributes; the second analyzes the extent of hard class overlaps between two classifications after they are defuzzified; and the third examines the residual membership surfaces obtained by subtracting one membership distribution from another. The experimental results will show how sensitive fuzzy k-means landform classifications are to the selected spatial scales; how the classifications may define different landscape patterns at different spatial scales; how the tested terrain attributes may respond differently to scale changes; and whether or not there are scale thresholds or natural boundaries present in the study area. The scientific value of this project lies in the fact that fuzzy k-means classification can potentially use widely available digital elevation models (DEM) to depict the geographic variations of environmental characteristics, which are often extremely difficult to measure. The focus on spatial scale issues is central in this modeling process because both the terrain surface and the environment vary at a diverse range of spatial scales, and the delineation of environmental characteristics relies on appropriate selection of spatial scales. The research results will also help to advance our knowledge of scale issues in geography, Geographic Information Science (GIS), and environmental modeling. As a Doctoral Dissertation Research Improvement award, this award will also provide support to a young scholar that plans to pursue these and similar research questions independently in the future.