Date Published: May 24, 2018
Publisher: Public Library of Science
Author(s): Yao Yao, Dongsheng Chen, Le Chen, Huan Wang, Qingfeng Guan, Timothy Andrew Warner.
Urban extent data play an important role in urban management and urban studies, such as monitoring the process of urbanization and changes in the spatial configuration of urban areas. Traditional methods of extracting urban-extent information are primarily based on manual investigations and classifications using remote sensing images, and these methods have such problems as large costs in labor and time and low precision. This study proposes an improved, simplified and flexible method for extracting urban extents over multiple scales and the construction of spatiotemporal models using DMSP/OLS nighttime light (NTL) for practical situations. This method eliminates the regional temporal and spatial inconsistency of thresholding NTL in large-scale and multi-temporal scenes. Using this method, we have extracted the urban extents and calculated the corresponding areas on the county, municipal and provincial scales in China from 2000 to 2012. In addition, validation with the data of reference data shows that the overall accuracy (OA), Kappa and F1 Scores were 0.996, 0.793, and 0.782, respectively. We increased the spatial resolution of the urban extent to 500 m (approximately four times finer than the results of previous studies). Based on the urban extent dataset proposed above, we analyzed changes in urban extents over time and observed that urban sprawl has grown in all of the counties of China. We also identified three patterns of urban sprawl: Early Urban Growth, Constant Urban Growth and Recent Urban Growth. In addition, these trends of urban sprawl are consistent with the western, eastern and central cities of China, respectively, in terms of their spatial distribution, socioeconomic characteristics and historical background. Additionally, the urban extents display the spatial configurations of urban areas intuitively. The proposed urban extent dataset is available for download and can provide reference data and support for future studies of urbanization and urban planning.
Urban areas are dominated by the built environment, which includes all non-vegetation and human-constructed elements, such as roads, buildings, and runways. In this context, “dominated” implies coverage greater than 50% within a given landscape unit . Urban extents draw cities’ outlines, which contain urban areas, as well as man-made vegetation and bare soil in and around urban areas. In China, with the arrival of the Lewis turning point (the structural change from an excess supply of labor to one of labor shortages) and the new normal context, the pattern of urbanization that had driven large amounts of rural labor to urban areas has disappeared . Currently, the trend toward urban sprawl has been increasing. For example, hundreds of square kilometers of area have become occupied by urban development since 2000 [3, 4]. Urban extent data plays an important and basic role in the analysis of the processes and trends of urbanization.
We selected China as our study area, and the study period extends from 2000 to 2012. Since its reform and opening up, beginning in 1978, China has undergone rapid urbanization . The urban area increased from 7,438 km2 in 1981 to 32,520.7 km2 in 2005 in China. The change in urban sprawl is obvious . Thus, China is suitable for testing the temporal and spatial normalization model. The data on administrative boundaries in China came from the Database of Global Administrative Areas (GADM) (http://gadm.com), specifically version 2.8, which was released in 2015. The administrative boundaries stored in GADM are divided into four levels: Level 0, Level 1, Level 2, and Level 3, which address the national, provincial, municipal, and county administrative boundaries of China, respectively (Fig 1).
The workflow used in this study is illustrated in Fig 2. The purpose of our study is to extract an urban extent dataset covering China over multiple scales using an LULC image and time series of NTL data and to identify the different patterns of urban sprawl. In this study, we take four steps to address this problem. 1) First, we apply a spatial overlay analysis to exclude NTL data from non-city lights using the MOD44 W and Gas Flare masks. 2) Next, by comparing the NTL with the spatial correction reference data and extracting PIFs, we construct a temporal and spatial normalization model for NTL to eliminate the temporal and spatial inconsistency. In addition, we calculate the area of the urban extent on the provincial, municipal and county scales. 3) We then assess the accuracy of the results by performing validations of the urban extent dataset using the ancillary test data. 4) Finally, using the K-Means method, we cluster the extracted dataset to analyze the trends in urban sprawl in the temporal dimension.
To extract a time series of urban extents from DMSP/OLS NTL data using the thresholding method, two problems need to be solved: First, because of noise from non-urban light sources, the low spatial resolution of the OLS sensor, atmospheric effects and the accumulation of geolocation errors when mosaicking the NTL images, NTL data cannot be used directly to generate large-scale scenes of urban areas. Second, the lack of an on-board calibration system causes NTL images to not be comparable with each other. Therefore, given the temporal and spatial inconsistency of NTL data, they cannot be directly used to extract urban extents.