Using DMSP/OLS nighttime light data and K–means method to identify urban–rural fringe of megacities
Introduction
With the expansion of urbanized areas, urbanism factors are increasing in the suburbs of large cities; in other terms, at the periphery of urban areas, rurality factors are declining. Although these periphery areas have not yet reached the urban category in terms of urbanization level and quality, it is no longer suitable to describe these areas to be rural. In this case, the urban–rural dichotomy is unable to fully reflect the characteristics of the regional structure of cities, and thus the concept of the urban–rural fringe has emerged. As a dynamic, unified, transitional region, the urban–rural fringe has become a focus of environmental protection. They are fragile urban management zones, and the social, economic and environmental conflicts they stimulate and expose are becoming more and more obvious (Bauer & Swallow, 2013; Kolbe, Miller, Cameron, & Culley, 2016; Yan, Chen, & Xia, 2018). Many scholars have begun to pay attention to the ecological and socio–economic changes in urban–rural fringes (Cheng, Wang, Wang, & Lin, 2018; Koch et al., 2019; Wei, Na, Cheng, Wu, & Liu, 2020). In order to study the social structure, landscape pattern and the effect on urban ecological environments in urban–rural fringes, it is necessary to identify urban–rural fringes quantitatively, which is of great significance for cities to develop more sustainably.
The identification of urban–rural fringes was originally defined by qualitative methods, i.e. by empirical values to delineate the scope of urban–rural fringes, such as the distance from built–up areas or central city (John & John, 1965; Sylla, Lasota, & Szewranski, 2019). With the improvement of basic urban data, identification methods have transformed into both qualitative and quantitative. For example, the ratio of non–agricultural populations and agricultural populations, economic structure, population density, and housing completion have been applied to delineate the scope of urban–rural fringes (Gober & Burns, 2002; Sharp & Clark, 2008; Surhone, Timpledon, & Marseken, 2010), or to build comprehensive indicators (Garreau, 1992). With the development of RS and GIS technologies, some scholars have begun to use threshold method (Imhoff, Zhang, Wolfe, & Bounoua, 2010; Sharp & Clark, 2008), mutation detection (Peng et al., 2016b, 2018), spatial clustering (Gao et al., 2014) and other quantitative methods to delineate the boundary of urban–rural fringes.
Although these methods break the limitations of administrative boundaries in qualitative or semi–qualitative methods, there are still some problems. For example, single or multiple indicators lead to excessive data processing workloads and cumbersome processing methods (Cao, Liu, Liu, & Miao, 2012). Threshold values are obtained by repeated experiments (Zhou et al., 2014), which affect the identification efficiency and the longitudinal comparability of the identification results. The mutation detection method has certain requirements on urban morphology, and the boundaries of urban–rural fringes are determined subjectively and connected manually (Yang, Ma, Tan, & Li, 2017). Therefore, the identification results based on the mutation detection method are greatly influenced by subjective factors, and it is difficult for application and popularization of this identification method (Peng et al., 2018).
As a classical clustering algorithm, the K–means algorithm has a fast and efficient performance when processing large datasets (Delmelle, 2015; Rokach & Maimon, 2005). At the same time, this unsupervised classification algorithm can directly identify regional types at the pixel scale, and it has no special requirements for urban morphology. That is to say, this method has a wider scope of application and less uncertainties in the identification results than other methods, and it can also solve the problem of insufficient objectivity reported in previous studies. Of course, the classification results of K–means are also affected by the diagnostic features. The obvious diversities in land use, population structure and economic activities between urban and rural areas can be reflected in the landscape (Hu, Tong, Frazier, & Liu, 2015). Many studies have explored the impact of urbanization on urban–rural gradient from the perspective of landscape (Díaz–Palacios–Sisternes, Ayuga, & García, 2014; Vizzari & Sigura, 2015), confirming that urban and rural areas can be distinguished by some landscape features (Haregeweyn, Fikadu, Tsunekawa, Tsubo, & Meshesha, 2012; Thorn, Thornton, & Helfgott, 2015), and nighttime light is one of them.
Nighttime light data have benefitted from the development of remote–sensing technology. At night, sensors have the ability to detect weak near–infrared radiation, such as city lights, even from smaller–scale residential areas, traffic flow, etc., which makes urban areas distinctly different from dark rural backgrounds. Therefore, nighttime light data have been applied to detect urban spatial patterns, urbanization processes, human activities and their impact on the ecological environment and other fields (Bennett & Smith, 2017; Chang, Wang, Zhang, Niu, & Wang, 2020). Nighttime light data have the ability to depict the regional spatial features of transitional urban–rural areas, and thus can be used as a diagnostic feature to distinguish between urban and rural areas. As such, they not only break the limitations of administrative boundaries and statistical calibers, but also overcome the problem of poor data continuity.
Beijing City has witnessed rapid urban development since the reform and opening up, and the urban suburbanization and suburb urbanization of Beijing City have been in the integration process (Li, Zhou, & Ouyang, 2013; Peng et al., 2018; Wang, Ma, & Zhao, 2014), resulting in the continuous reconstruction of urban spatial patterns. It is very important to delineate the urban–rural fringe of Beijing City for defining the urban spatial pattern, and for further studying the natural, social and economic changes of urban–rural fringes. Therefore, in this study we take Beijing City as an example, and use the K–means algorithm to identify urban–rural fringes based on DMSP/OLS nighttime light data, considering both light intensity and light fluctuation. The research aims were: (1) to analyze the spatial pattern of nighttime light intensity and light fluctuation in Beijing City, especially focusing on the gradient change between urban area, urban–rural fringe, and rural area; (2) to identify urban–rural fringe by K–means algorithm and recognize the urban structure of Beijing City; (3) to examine the performance of combining nighttime light intensity and light fluctuation; and (4) to explore the efficiency of K–means algorithm by comparing the results of proposed method with traditional mutation detection methods.
Section snippets
Study area and data sources
Beijing City, with a total area of 16,410 km2, is located north of the North China Plain (39°28′–41°05′N, 115°25′–117°30′E), which extends about 160 km from east to west, and about 176 km from north to south. Surrounded by mountains in the north, west and northeast, the terrain of Beijing City drops gradually from northwest to southeast. As the typical semi–humid continental monsoon climate in warm temperate, it is rainy and hot in summer, dry and cold in winter, and it has short spring and
Spatial pattern of nighttime light intensity and light fluctuation
From the city center to the surrounding areas, it can be seen that the nighttime light intensity of Beijing City shows a downward trend (Fig. 2(a)). In the midland of Beijing City, there are a series of high–value areas with a maximum DN of 103.676, which are mainly located within the 5th Ring Road that covers the core functional zone and urban function extended zone, which are the main urban areas of Beijing City. Outside the city center, between the 5th and 6th Ring Road and outer area near
Performance of combining nighttime light intensity and light fluctuation
The combination of multiple dimensions often results in more information than when they are used separately (Lu, Tian, Zhou, & Ge, 2015; Wang, Wan, Guo, Hu, & Zhou, 2017). In this study, we used two indicators of nighttime light intensity and light fluctuation, which were both derived from nighttime light data, but they reflect different characteristics of nighttime lights. We analyzed the probability density distribution of DN and FI of the three different regional types (urban areas, rural
Conclusions
Urban–rural fringes, as a link between urban construction areas and rural hinterland, are transitional regions with complex and rapid changes in land use types and socio–economic activities. In this study, an urban–rural fringe identification method based on K–means algorithm was proposed, which combined the intensity and fluctuation of nighttime light. The intensity and fluctuation of nighttime light were proved to effectively reflect the gradient changes of the urban–rural landscape via a
CRediT authorship contribution statement
Zhao Feng: Software, Writing - original draft, Visualization. Jian Peng: Supervision, Writing - review & editing. Jiansheng Wu: Conceptualization, Methodology, Software, Validation.
Acknowledgment
This work was supported by the National Natural Science Foundation of China [No. 41671180].
References (41)
- et al.
Conserving metapopulations in human–altered landscapes at the urban–rural fringe
Ecological Economics
(2013) - et al.
Advances in using multitemporal night–time lights satellite imagery to detect, estimate, and monitor socioeconomic dynamics
Remote Sensing of Environment
(2017) - et al.
A study of the impacts of urban expansion on vegetation primary productivity levels in the Jing–Jin–Ji region, based on nighttime light data
Journal of Cleaner Production
(2020) - et al.
Mix leading to success? Exploring the innovative development model in peri–urban China
Habitat International
(2018) Five decades of neighborhood classifications and their transitions: A comparison of four us cities, 1970–2010
Applied Geography
(2015)- et al.
A method for detecting and describing land use transformations: An examination of madrid's southern urban–rural gradient between 1990 and 2006
Cities
(2014) - et al.
Clustering urban multifunctional landscapes using the self–organizing feature map neural network model
Journal of Urban Planning and Development
(2014) - et al.
The dynamics of urban expansion and its impacts on land use/land cover change and small–scale farmers living near the urban–rural fringe: A case study of bahir dar, Ethiopia
Landscape and Urban Planning
(2012) - et al.
Urban boundary extraction and sprawl analysis using landsat images: A case study in wuhan, China
Habitat International
(2015) - et al.
Remote sensing of the urban heat island effect across biomes in the continental USA
Remote Sensing of Environment
(2010)