The Center of City Function in Guiyang, China: An Evaluation with Emerging Data

2.1 Application of emerging data in cities

Traditional research data, including statistical yearbook and census data, have shown great inadaptability in current urban research. In recent years, analyzing the spatial structure of cities with urban big data has provided a new research paradigm and perspective for urban research. The current widely used urban data includes rail transit card data, mobile signaling data, network review data, thermal map data, etc. [8-11], A large number of Chinese scholars, including Long and Zhen [12, 13], took the advantages of these new data to explore the urban space, and summed up plenty of new theoretical and methodological models. However, in China, there are still problems for these urban big data in obtaining difficulty, less acquisitive methods and confidentiality protection, which plague the researchers. Withal, POI data based on open-source mapping provides an accurate and effective alternative. First, the POI data is a virtual abstract representation of the real geographic entity in space, which equipped with spatial attribute information. Secondly, it is large in quantity and easy to obtain, which makes it gradually become one of the important data in urban research [14]. Based on the POI data, scholars have concentrated on the exploration of the formation of urban space [15], the identification of urban functional areas [16], etc., and because of its good research performance, it has begun to be widely used in urban research [17].

2.2 Central identification and temporal evolution method

It is noteworthy that different methods are applicable to different types of data, and what this study emphasizes is the central population, the activities taking place in the center, and the physical properties of the center itself.

(1) In 1991, Giuliano developed a central identification method based on the minimum cut-off point of the employment density of regional personnel [18], which was then subsequently widely used [19]. This method mainly determines a group of continuous plots as a center, whose minimum employment density is D and the minimum total employment number is E, and achieves the purpose by judging the cut-off point of the two indicators, but the set threshold value will produce obvious differences to the research results. Different from the method that requires uniform setting of minimum cut-off value, Agarwal et al. proposed a density gradient value based on estimation [20] to determine the center in 2012, including determining the sub-center by local peaks [21]. However, the use of this method requires more detailed local knowledge to ensure the accuracy of peak selection, thus, it can also become the method with least efficiency and accuracy when applied in more cities. Li et al. identified 658 Chinese cities based on the population distribution reflected in Baidu heat map [22], and through detailed grid population data, the diversity of the centers of 318 cities in China was further identified [23]. These results allowed the center identification program to be applied to numerous cities and regions without requiring much local knowledge. Based on the existing urban center identification methods, the exploration of cities is still subjective and inaccurate [24].

(2) The trend of urban spatial structure and the correlation of urban geographical space can be an enormous challenge which derives from a fact that static or aggregated measurement cannot reveal the real impact of clustering [25]. Therefore, it is necessary to provide a method to analyze its dynamic evolution and correlation in a short time [26]. At present, most researches adopt the correlation methods of spatial statistical analysis, which mainly tend to reveal the correlation in geographical space location. Therefore, most researchers begin to try global and local statistics to describe its change trend [27]. For instance, Malleson explored the changing trend of crime rate in a space through visualization of getis-ord Gi* and charts, which effectively reduced the risk of criminal events [28]. And Mazzulla tested the travel space with getis-ord General G, and described its growth trend in detail through the expression of temporal atlas [29]. In a word, spatial statistical analysis can all serve well in promoting the trend assessment of elements in geographical space [30, 31].

There are two main objectives in this study: The first is to determine the city function center of Guiyang, China, and the second is to explore the changing trend of the function center.

The first objective identifies the city center by the point density peak of the POI. The point density in the city is first calculated to generate a continuous density surface in the area, which is then divided into 10 categories with natural breaks (Jenks), and on each surface according to the standard of no more than 20% density difference to determine the city function center.

The second objective make use of the Getis-Ord General G and Getis-Ord Gi* to explore global and local geographic features, structural patterns, and spatial clustering in Guiyang.

Getis-Ord General G:

$G(d)=\sum \Sigma w_{i j}(d) x_{i} x_{i} / \sum \Sigma x_{i} x_{i}$        (1)

where, d is the distance to the urban center of Guiyang, China; $w_{i j}(d)$ is the spatial weight within the study area; $x_{i}$ and $x_{j}$ are the index of urban land extension intensity respectively. Under the assumption of space non-agglomeration, the expected value of $G(d)$ is $E(G)$:

$E(G)=W /[n(n-1)] \quad w=\sum \Sigma w_{i j}(d)$          (2)

Under the condition of normal distribution, the statistical test value of $G(d)$ is $Z(G)$:

$Z(G)=[G-E(G)] / \sqrt{\operatorname{Var}(G)} \quad\left[E(G)=W_{n}(n-1)\right]$          (3)

When $G(d)$ is higher than $E(G)$ and $Z(G)$ value is significant, high-value clusters appear in the study area. When $G(d)$ approaches $E(G)$, the variables in the research area show a random distribution.

The expression for Getis-Ord Gi* is:

$Z\left(G_{i}^{*}\right)=\sum_{j}^{n} w_{i j}(d) x_{j} / \sum_{j}^{n} x_{j}$         (4)

Through POI, this study identified the city function center and its change trend of Guiyang, China. According to the research, first of all, functional centers show the functional diversity of city centers, and the previous methods and data are not enough for accurate identification and judgment. Through emerging data represented by POI, most city functions are expressed objectively, and the identified city function centers also have high credibility. Secondly, it can avoid the errors brought by previous methods, including the subjective arbitrary setting of the regional threshold and the scale effect in the spatial scale. The practical case of Guiyang reveals the value of this study, which can be used to conduct further research on a large number of cities, thus providing reliable and effective practical experience for solving the research on the spatial structure characteristics of Chinese cities. Finally, the administrative boundary is not used as the research area as usual in Chinese cities, but the actual functional area of the city is replaced, which makes the analysis results more objective and to some extent explains the differences between Chinese cities and those in European and American before.

Of course, there are still some shortcomings need to be improved in this study. First, the study only considers the area and density of each city function center, and these methods are not substitutable. For example, land use classification can be verified through remote sensing images. Besides, the number of POI in city function centers is not homogeneous in each city function category. In this study, the number of POI of life function occupied the largest proportion, while the proportion of POI of leisure and business was less than 10%. However, they were all treated equally without considering whether each type of POI should be weighted and averaged according to the proportion of quantity. These problems need to be further explored in the following research.

The emerging data headed by POI provides a brand-new research perspective and paradigm for the study of urban spatial structure. Both the ideas and methods of urban planning have been greatly changed, and the new innovation is undoubtedly facing great challenges. This study attempts to make use of new data to innovate research, so it focuses on the morphological dimension of urban internal scale and urban spatial structure. However, compared with the traditional research, such kind is still less, which needs to be further deepened by continuous theoretical supplement. Finally, when considering the regional area and density ratio of POI, it only limits the regional area preliminarily. Based on the focus of this study is morphological change, further studies are needed to solve the problem of regional area ratio.