集成土地利用数据和夜间灯光数据优化人口空间化模型

An Improved Population Spatialization Model by Combining Land Use Data and DMSP/OLS Data

人口统计数据空间化是解决统计数据与自然要素数据融合分析的有效途径。随着RS和GIS技术的发展,人口统计数据空间化方法推陈出新,其中,土地利用数据、夜间灯光数据是人口空间化研究中普遍利用的数据源,但各有优、缺点:土地利用数据中的城镇用地、农村居民点能准确表示人口分布的空间范围,却不能反映其内部的人口密度差异特征;夜间灯光数据的强度信息能体现人口分布的疏密程度,但其像元溢出问题显著夸大人口分布范围,像元过饱和现象也影响着人口数据空间化结果的精度。本研究以中国大陆沿海区域为例,尝试集成土地利用数据和夜间灯光数据优化人口空间化方法,设计了基于精度阈值和动态样本的渐进回归与分区建模的方法,获得了中国沿海2000、2005、2010年1 km分辨率人口空间化数据。结果表明,优化模型显著提高了研究区整体的精度,尤其适用于人口空间结构内部差异较为显著的区域。

Population spatialization could effectively remove the obstacles for data fusion between census data and geographic data. With the rapid development of RS and GIS techniques, the research methods of population spatialization have been updated greatly, and among them, both land use data and DMSP/OLS data are the most widely used data sources for population spatialization. However, both of them have advantages and disadvantages, in specific, the patches of cities and rural settlements in land use data indicate the area of population distribution accurately, but land use data lacks detailed features of population distribution, especially for that existing in the same type of land use; DMSP/OLS data manifests the spatial variations of population density, however it often overestimates the distribution area of population due to its ’overglow’ effect, and at the same time, the problem of pixel saturation in DMSP/OLS data also impairs the fitness of this data. In this paper,land use data and DMSP/OLS data are combined together to distinguish the night light data value of populated area, and the methodologies of population spatialization are improved by introducing the precision threshold and dynamic regionalization method. Census data in 2000, 2005 and 2010 for China’s coastal area are taken as examples to test the outcomes of the improved methodologies for population spatialization. The results show that：（1） due to the prominent spatial heterogeneity of population distribution in China’s coastal area, unitary model based on DMSP/OLS data for the whole study area exhibits very poor precision, therefore, land use maps are utilized to distinguish the populated and non- populated area based on DMSP/OLS images. Compared with the threshold method, land use maps more effectively removes the‘overglow’effect of DMSP/OLS data;（2）precision threshold of the regression model is adopted to dynamically divide the whole study area into several sub- regions, in specific, only counties that meet the testing accuracy defined by the precision threshold could retain in the regression model, otherwise, they should be regrouped into a new set of samples and be fitted by a new regression model. It is named as‘dynamic regionalization’method in this paper. The results show that this method further improves the overall accuracies of population spatialization data.