Using DMSP/OLS nighttime light data and K–means method to identify urban–rural fringe of megacities

Highlights

• Nightlights performed fine as an urban–rural diagnostic feature.

• Nightlight intensity and fluctuationwere combined to identify urban-rural fringe.

• K-means behaved better than mutation detection in detecting urban-rural fringe.

Abstract

Urban–rural fringe, which form a link between urban construction areas and rural hinterland, is the most sensitive area to urbanization. Its accurate identification is of great significance for the further study of urbanization related socio–economic and eco-environmental changes in the perspective of urban–rural contrast. Previous studies of urban–rural fringe identification had problems with narrow scope of application, low efficiency of identification, and the results were greatly influenced by subjective factors. Nighttime light, as an important product of human activities, can reflect the gradient changes of urban–rural landscapes, and can be used to identify urban–rural fringes. Therefore, a K–means–based approach was developed using Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) nighttime light data. Taking Beijing City as an example, in this study we delineated its urban–rural fringes. Our results indicate that a ring–shaped urban–rural fringe surrounds urban central areas, with an area of 3712 km², which is mainly located in new urban development zones. Inside the urban–rural fringe, lights fluctuated obviously, and the fluctuation index was up to 76.75. Meanwhile, the combination of nighttime light intensity and light fluctuation had better performance than that when they were considered separately in the identification of urban–rural fringes. Furthermore, the K–means algorithm based on nighttime light found more details related to urban–rural fringes when compared with the traditional mutation detection method. This study provided an approach to identifying urban–rural fringes accurately and objectively, which is conducive to the study of eco–environmental effects in the process of urbanization.

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.