中文摘要：

先前的研究已经证明从DMSP/OLS（Defense Meteorological Satellite Program/Operational Linescan System）稳定夜间灯光数据中可以提取城市建成区,并且以此估算了人口、GDP等其他各项指标,但对提取出的城市建成区空间分布模式研究较少。本文提出了一种改进的基于密度的聚类（Density-based Spatial Clustering of Qpplications with Boise,DBSCAN）算法,将对于点要素的聚类扩展到面要素,以面状对象的边缘作为界定包含关系的准则;确定了聚类参数,以要素间距离的突变点作为距离参数。按此方法对夜间灯光数据提取的中国城市对象按照密度分布的特点进行聚类,通过改变距离参数得到在不同尺度下中国城市的集聚形态。通过与实证资料的对比验证了该算法的有效性,为研究中国城市的空间分布及其演变提供了有力的研究方法。

英文摘要：

Understanding spatial distribution of urban clusters at regional and national scales is increasingly important for many fields especially urban planning.Previous Studies have demonstrated urban built-up areas can be derived from stable nighttime light satellite（DMSP-OLS）images.Population and economic variables（i.e.GDP）have been proved significant positive correlations with nocturnal light brightness.However,less studies focused on the spatial distribution of extracted urban built-up area.an improved DBSCAN algorithm is proposed to cluster the urban objects extracted from nighttime light image in different scales based on density,of which our urban spatial clusters are proved corresponding with urban agglomerations identified by statistical data.The traditional DBSCAN method is based on points which is not the same case with urban objects.The inclusion relation is refined,assuming that only if all the vertexes of each polygon are within the given distance,it is included in the area.Moreover,the parameters for the DBSCAN clustering model are determined by valleys of distances of every objects to classify urban spatial clusters.Besides,in a larger scale,the clustering results imply the different patterns of urban agglomerations on both sides of the Huhuanyong Line.