The Agglomeration of Food Services and Their Colocation with Surrounding Complementary Services in the Guangdong–Hong Kong–Macao Greater Bay Area

Abstract

This study explores the spatial distribution of food services and their colocation with surrounding complementary services. It investigates these issues within the Guangdong–Hong Kong–Macao Greater Bay Area (GBA), utilizing point-of-interest (POI) data, spatial kernel density, the HDBSCAN clustering algorithm, and colocation quotients. The findings are as follows: (1) this research reveals a significant spatial agglomeration of food services near the Pearl River, with notable food clusters across administrative boundaries; (2) Guangzhou, Shenzhen, Foshan, and Dongguan provide a significant quantity of food services, while Hong Kong and Macao feature the highest percentages of foreign cuisine; (3) the colocation between food services and surrounding services is concentrated along the Pearl River; (4) leisure, education, and residential services are key factors attracting the proximity of food services; (5) leisure, education, retail, and tourism services exhibit the strongest attractiveness to Chinese food, while residential and healthcare services are closely linked to the distribution of snacks, and transportation services attract snacks and beverages.

1. Introduction

In urban development, cities hold a prominent role as consumption centers [1], with service agglomeration brought about by the associated economic externalities [2,3,4,5,6,7]. Food services provide food to consumers and are an important part of service consumption. The spatial agglomeration of places providing food services, such as restaurants and bars, has received significant attention with the application of geo-coded data and big data. Research on the spatial distribution of urban food services and their interactive relationships with surrounding complementary services can provide a reference for optimizing the spatial planning of consumer services and help meet the diverse consumption needs of residents.

Existing empirical research has focused primarily on two aspects of food services that are relevant to the research topic: first, many scholars have analyzed the spatial distribution of food services in individual cities [8,9,10,11,12,13,14,15,16,17]; second, scholars have discussed factors influencing the location of food services, including population density, transportation, and so on [10,11,16,18,19,20,21,22,23,24,25,26]. The agglomeration of consumer services is often closely tied to the travel behavior of residents [27], including consumers’ travel methods, travel distance, travel purpose, travel cost, etc. Therefore, areas with convenient transportation also attract a large number of food services [8]. However, with changing consumer travel behaviors, such as increasing consumer mobility and the complexity of trip chains, we should rethink the spatial agglomeration of consumer amenities [28].

Building on prior research, this study aims to explore two key gaps that are closely linked to transitions in resident’s travel behaviors. First, while there is a growing body of work on the spatial agglomeration of consumer services [8,9,10,11,12,13,14,15,16,17], few studies have focused on the agglomeration of consumer services at the inter-city level. The rapid urbanization and formation of megaregions has been coupled with significant improvements in transportation, which have significantly increased consumer mobility. The improved transportation in megaregions enables residents to extend their activity scope beyond the administrative boundaries of individual cities [29,30,31].

Second, most studies focus on the agglomeration of similar services; the agglomeration of different types of services, especially complementary ones, has received less attention. Specifically, the influence of complementary services on food service agglomeration remains less explored, and the spatial concept of “surroundings” has been inadequately quantified, despite existing studies that examine the agglomeration of different service types [11,19,24,25,26]. This is particularly important when considering the trip chain [22,31,32,33], as consumers often visit multiple types of places in a single journey.

To address these gaps, we focus on the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) as a case study, using food services and other kinds of services POI data. We take the GBA as the studied region for three important reasons. First, the GBA is a prime example of a leading megaregion in China, characterized by increasing regional integration and transportation improvement. The GBA is also one of the world’s renowned megaregions. The research findings thus provide valuable references for studying other megaregions. Second, residents in the GBA show a strong preference for food services. According to China Statistical Yearbook, in 2023, Guangdong residents’ expenditure on food consumption, including food, tobacco, and alcohol, accounted for approximately 32.44% of total expenditure. This percentage (32.44%) was higher than other four provinces with high GDP, namely, Jiangsu (27.97%), Shandong (27.95%), Zhejiang (27.86%), and Henan (29.87%), and exceeded the national average in China (29.79%). It was also higher than that in other countries like the United States (14.2%) [34]. Food services are an essential component of service consumption in the GBA, making the study of food services in the GBA particularly representative. Third, due to the unique historical and political context, the GBA operates under three distinct statistical systems across Guangdong, Hong Kong, and Macao. The three distinct statistical systems have limited research using conventional data. However, geo-coded point-of-interest (POI) data provide an effective alternative for studying such diverse regions. In recent years, POI data have been used to study the GBA [35,36,37,38].

In this study, we use POI data for food services in the GBA in 2021 and employ spatial kernel density estimation, the HDBSCAN clustering algorithm, and colocation quotients to explore the spatial distribution patterns of food services and the colocation between food services and surrounding complementary services in the GBA.

The structure of this paper is arranged as follows: Section 2 presents the literature review and research questions; Section 3 presents the research area, the data, and our methodology; Section 4 discusses the spatial agglomeration of food services and the spatial attraction effect of surrounding complementary services on the agglomeration of food services; Section 5 is the discussion; and conclusions are drawn in Section 6.

2. Literature Review and Research Questions

When studying the agglomerations between service providers, we can start from two kinds of agglomerations: one is the agglomeration between the same services; another is the agglomeration between complementary services, which can be studied thought influencing factors. The following literature review examines the key findings on the spatial distribution and influencing factors of food service agglomerations, and also identifies critical gaps in existing research.

From a micro-level perspective, food services exhibit a mutual attraction that drives spatial agglomeration. In comparative or multipurpose consumption, the agglomeration of services can reduce consumer’s search and travel costs; thus, the agglomeration of food services can attract more consumers, and it allows service providers to share a larger market. Additionally, geographic proximity promotes spatial competition and knowledge spillovers, which can help service providers improve themselves. A study found that service providers may adopt differentiation strategies, such as variations in price, quality, and service types in response to intense spatial competition [6]. The improvements in consumer experience attained through providing various goods and services and lower prices [6,39] can also attract more consumers.

As for the spatial distribution of food services, many scholars have analyzed the spatial distribution of food services in individual cities, such as Shanghai, Beijing, Hangzhou, Nanjing, Hefei, Jakarta, Hamilton, Calabar, and London, and commonly used methods include kernel density estimation, Moran’s I index, nearest neighbour analysis, hotspot analysis, and so on [8,9,10,11,12,13,14,15,16,17]. From the microscopic mechanism, it can be inferred that catering clusters exist in reality, and the agglomeration of food services is found globally. Research on Jakarta reported a clustering trend among fast-food restaurants [11]. Another study used Hangzhou as a case study and discovered that 81.6% of KFCs had a McDonald’s nearby within 2500 m, and 68.5% of Shaxian Snacks had a Lanzhou Ramen restaurant within 400 m [16]. Research from the UK found that top-rated restaurants tend to be near other top-rated restaurants [17].

In terms of the influencing factors, scholars have discussed the impact on the location of food services of factors such as neighborhood sociodemographic characteristics, population density, economic development, transportation, walkable environments, and other related factors, and widely used methods include regression model, spatial autocorrelation model, colocation quotient, and so on [10,11,16,18,19,20,21,22,23,24,25,26]. Specifically, studies have explored the agglomeration between different types of services, including food services and other services. In theory, the colocation of complementary services brings several benefits. The agglomeration of complementary services can satisfy consumer’s various needs and allow them to save on search and travel costs, and it also can bring in more consumers, resulting in a larger market size. For example, large leisure facilities, such as city parks, museums, or large shopping malls, attract a significant number of consumers, nearby food service businesses can draw in customers from these popular facilities, and food services enhance the attractiveness of consumer spaces [40], thus attracting more consumers to visit these places. Some studies confirm the agglomeration between different types of services. In the case of Sweden, there are limited but noticeable strategic complementarities between retail services, consumer service branches, and restaurants [18]. Fast-food establishments tend to colocate with public facilities in Jakarta, including leisure/shopping, travel, education, religious, health, and work facilities [11]. In Beijing, the food services, retail, life services, accommodations, residences, and business sectors exhibit the most prevalent colocation patterns [24].

Despite significant progress in understanding food service agglomerations, key gaps remain: (1) inter-city interaction and administrative boundaries; (2) complementary services and spatial interactions.

First, few studies have focused on the agglomeration of consumer services at the inter-city level. In practice, the spatial development of Chinese cities is significantly influenced by the administrative boundaries. Due to decentralization, when economic activities exhibit positive externalities, local governments may strategically reduce public investments along jurisdictional borders [41], such as in transportation infrastructure, hospitals, schools, and other municipal facilities that are crucial for the agglomeration of food services [42]. Lagging infrastructure and economic development often appear in border areas, as well as environmental pollution issues. Additionally, certain services, such as food services, are inherently localized and non-tradable across regions. Because consumers tend to travel for a short time or distance to access these nearby services, the service cluster hinterlands for these services tend to remain localized. These two factors may contribute to the clustering of food service POIs within administrative boundaries.

However, consumer mobility has increased significantly. Private car ownership has increased rapidly in China. In 2023, there were 96 cities with car ownership exceeding 1,000,000. The supply of public transportation in megaregions, such as urban subways, inter-city subways, urban buses, inter-city buses, inter-city high-speed railways, and Mobility as a Service (MaaS), including Didi, is increasing. These changes in urban transportation have enhanced consumer mobility. With better consumer mobility, the travel distance can be extended, and travel costs can be reduced to a certain degree. With the integration of megaregions with increased consumer mobility, the hinterland of the services cluster is not limited to nearby local areas, and the scale of the service clusters may grow and extend across administrative boundaries.

Second, in general, compared with the agglomeration of similar services, the agglomeration of different types of services has received less attention. Specifically, there has been little exploration of how complementary services impact the agglomeration of food services, and there is less quantification of the spatial concept of “surroundings”, even though some studies presented above have focused on the agglomeration between different types of services [11,19,24,25,26]. The agglomeration of complementary services is critical to understanding the location of services when taking trip chains into consideration.

The colocation of complementary services caters to the complementary demands generated by residents’ trip chain modes. There is an increasing tendency for people to visit multiple stops along their trip chains [31], which start and end at home and include visits to one or more destinations, including restaurants and leisure places. For example, residents can go to the mall after work. Trip chain modes are evidenced by individual data from cities such as Shanghai, Beijing, and Nanjing [31,32,33]. The trip chain model shows differences between holidays and weekdays [32]. The complexity of the trip chain is often related to higher automobile dependency [31], which means that the improvement in consumer mobility increases the complexity of the trip chain. Notably, trip chains can contribute to consumption externalities and lead to industrial agglomeration [43].

To address these gaps, this study poses the following research questions. (1) What are the spatial distribution patterns of food services in the GBA? (2) Do the surrounding complementary services influence the agglomeration of the food services and colocate with the food services? (3) Where does the colocation pattern occur? (4) Does the influence of surrounding complementary services vary across different types of services?

3. Materials and Methods

3.1. Research Area

The research area is the Guangdong–Hong Kong–Macao Greater Bay Area (GBA), which includes 11 cities. Considering the significant differences in the administrative subdivisions of the various cities and urban planning within the GBA, the 11 cities are further divided into 65 districts (or counties, qu, xian in Chinese) (see Figure 1). Hong Kong is subdivided into three secondary districts: Hong Kong Island, Kowloon Peninsula, and New Territories. Macao is not further divided due to its small area. Dongguan and Zhongshan adopt a merging approach for their towns, and the newly defined areas are named on the basis of their geographic orientation (see Appendix A).

We use the GBA as the research area for three important reasons. First, it is a representative megaregion with increasing regional integration and transportation improvement in China and is a famous megaregion in the world. Second, the residents here have a preference for food services. Third, geo-coded POI data are particularly suitable for studying such diverse regions.

3.2. Research Data

POIs for food services in the GBA were obtained from AMAP in 2021 and categorized into five major types: Chinese food, foreign food, fast-food, snacks, and beverages. For surrounding complementary services, we selected seven service types: leisure, education, residence, retail, tourism, healthcare, and transportation. More details about the food services and surrounding complementary services are presented in

Table 1. Categories of services.

Categories		Subclasses	Quantity/10,000
Food services	Chinese food	Cantonese cuisine, hot pot, etc.	48.37
	Foreign food	Japanese cuisine, American cuisine, Indian cuisine, etc.	3.08
	Fast-food	Chinese fast-food restaurants, KFC, McDonald’s, etc.	19.20
	Snacks	Noodles, fried skewers, desserts, etc.	3.74
	Beverages	Milk tea shops, bar, coffee shops, etc.	7.58
Surrounding complementary services	Retail	Convenience stores, supermarkets, etc.	18.14
	Residence	Residential area	10.12
	Transportation	Bus stations, train stations, airports, subway stations, etc.	26.43
	Education	Middle schools, universities, etc.	0.95
	Leisure	Parks, cultural squares, exhibition hall, library, etc.	1.62
	Healthcare	Clinics, hospitals, etc.	6.06
	Tourism	Beaches, tourist attractions, etc.	1.04

.

3.3. Methodology and Analysis Flowcharts

The analysis process consists of three main components: (1) Data preprocessing. (2) Spatial agglomeration of food services in the GBA. We use spatial kernel density and the HDBSCAN clustering algorithm to analyze the POI data [10,15,16,44]. Spatial kernel density analysis allows for the spatial smoothing of discrete POI data to generate continuous distribution density maps and facilitates the examination of spatial distribution. Compared with the spatial kernel density analysis method, hierarchical density-based spatial clustering of applications with noise (HDBSCAN) is more suitable for datasets with smaller quantities. Thus, we employ the spatial kernel density method to analyze the spatial distribution of Chinese food, fast-food, and beverages, while we use the HDBSCAN method to analyze the spatial distribution of foreign food and snacks. (3) Analysis of the colocation pattern between food services and surrounding services. Colocation quotients (CLQ) measure the attraction between two different types of point features, allowing us to observe the spatial association between these two types of point features. This method has been applied in studies using geo-coded data [16,24,45]. We use the colocation quotients (CLQ) to test the attraction of surrounding complementary services to food services. A detailed technical route diagram is presented in Figure 2.

3.3.1. Spatial Kernel Density

Owing to the vast quantity of POI data for the food services within the GBA and the significant differences in the numbers among the five categories, we primarily employ the spatial kernel density method to analyze the spatial distribution of the top three categories of food services in terms of quantity: Chinese food, fast-food, and beverages.

The calculation formula for spatial kernel density is as follows:

$$f(x)={\displaystyle \frac{1}{nh}}{\sum }_{i=1}^{n}k\left({\displaystyle \frac{x-x_i}{h}}\right)$$

(1)

In this equation, (1) $f\left(x\right)$ represents the kernel density value at spatial location $x$; (2) n is the number of feature points located within a distance h from location $x$; (3) the kernel bandwidth h is larger than 0 and can be approximated by dividing the smaller value (in width or length) by 30; (4) $k(.)$ represents the kernel function, which is typically implemented using the Rosenblatt–Parzen kernel function; and (5) $x-x_i$ represents the distance from the feature point $x$ to the set of feature points {$x_1$, …, $x_n$}.

3.3.2. HDBSCAN Clustering Algorithm

HDBSCAN determines the mutual reachability distance on the basis of the minimum number of cluster points, constructing a minimum spanning tree and the hierarchical structure of clusters. The HDBSCAN algorithm primarily defines a mutual reachability distance to measure clusters of different densities. The formula for calculating the mutual reachability distance is as follows:

$$d_{mreach-k}(x,y)=max\left\{{core}_k\left(x\right),{core}_k\left(y\right),d(x,y)\right\}$$

(2)

In Equation (2), $d_{mreach-k}(x,y)$ denotes the mutual reachability distance between food service providers x and y. Here, $d(x,y)$ represents the distance between food service providers x and y, whereas ${core}_k\left(x\right)$ and ${core}_k\left(y\right)$ refer to the core distances of x and y on the basis of their $k$ nearest neighbors.

3.3.3. Colocation Quotients

The concept of CLQs was first proposed by Leslie and Kronenfeld [46]. We use CLQs to quantify the spatial association between food services and surrounding services.

CLQs consist of global colocation quotients (GCLQs) and local colocation quotients (LCLQs). In this study, GCLQs are used to evaluate whether surrounding services attract food services at a global scale. The specific formula for GCLQs is as follows:

$${GCLQ}_{A\to B}={\displaystyle \frac{N_{A\to B}/N_A}{N_B/(N-1)}}$$

(3)

In this formula, (1) N represents the total number of all types of points; (2) $N_A$ denotes the number of POIs of type A; (3) $N_B$ indicates the number of POIs of type B; and (4) $N_{A\to B}$ represents the number of type A points that have type B points as their nearest neighbors. ${GCLQ}_{A\to B}$ measures the extent to which type A points attract surrounding type B points.

LCLQs introduce the concept of geographic weighting into the GCLQ model. This approach not only aids in understanding the colocation of food services and surrounding services but also facilitates the display of spatial heterogeneity of colocation.

The specific formulas for LCLQ are as follows:

$${LCLQ}_{A_i\to B}={\displaystyle \frac{N_{A_i\to B}}{N_B/(N-1)}}$$

(4)

$$N_{A_i\to B}={\sum }_{j=1(j\ne i)}^N\left({\displaystyle \frac{w_{ij}{f}_{ij}}{\sum _{j=1(j\ne i)}^N{w}_{ij}}}\right)$$

(5)

$$w_{ij}=exp\left(-0.5\times {\displaystyle \frac{d_{ij}^2}{d_{ib}^2}}\right)$$

(6)

In these equations, (1) $N_{A_i\to B}$ represents the geographic weighted average of type B points within the bandwidth of point A_i; (2) ${LCLQ}_{A_i\to B}$ indicates the degree to which type B points are attracted by point A_i; (3) $f_{ij}$ is a binary variable (if j is a point of type B, then $f_{ij}$ = 1; otherwise, $f_{ij}$ = 0); (4) $w_{ij}$ denotes the geographic weight of point j relative to point I; (5) $d_{ij}$ represents the distance between points i and j; and (6) $d_{ib}$ is the bandwidth distance for point i.

Bandwidth is a critical parameter in the GCLQ and LCLQ, representing the number of nearest neighbor points considered in local calculations. In other words, bandwidth measures the extent of the “neighborhood” or “spatial proximity”. Referring to previous studies [47], we employ the adaptive bandwidth and explore the spatial association between food services and surrounding services in the GBA, with bandwidths defined by the first, fifth, tenth, fifteenth, and twentieth order in the GCLQ. As for the LCLQ, previous studies [48,49] suggest that varying bandwidths do not result in substantial changes to the spatial association pattern of features. Considering the limitations of this paper’s length, we explore the spatial association between food services and surrounding services at fifteenth-order bandwidth.

A GCLQ or LCLQ value greater than 1 suggests a strong spatial association between type A and type B points at the global scale or local scale. A GCLQ or LCLQ value of 1 indicates that both types are randomly distributed. Conversely, a GCLQ or LCLQ value less than 1 indicates weak spatial dependence, suggesting that the points are relatively independent of each other.

Given the GBA’s integration and improved transportation, we anticipate observing food services clusters along the Pearl River, with some catering clusters crossing administrative boundaries. Additionally, we expect a significant spatial agglomeration relationship between surrounding complementary services and food services along the Pearl River.

4. Results

4.1. Spatial Distribution of Food Services in the Guangdong–Hong Kong–Macao Greater Bay Area

4.1.1. Spatial Agglomeration of the POIs for Food Services

Overall, the food services in the GBA demonstrate a distinct monocentric structure, with sporadic food clusters in the peripheral districts. The main center of food services POIs is located predominantly along the Pearl River, with a notably greater quantity and density on the eastern bank than on the western bank. From the western bank to the eastern bank of the Pearl River, there is a clear agglomeration in northern Zhongshan, Foshan, Guangzhou, Dongguan, Shenzhen, and Hong Kong, forming a cohesive cluster. Additionally, the southern areas of Zhongshan, Zhuhai, and Macao form a continuous agglomeration region for food services (see ① in Figure 3).

At the urban scale, Guangzhou and Shenzhen emerge as the cities with the highest concentration of food services POIs, exhibiting a notable polycentric structure. Furthermore, as illustrated in Figure 3, although most clusters of food services are within administrative boundaries, certain clusters of food services have extended beyond the administrative boundaries. Notable examples of clusters include the Zhongshan–Zhuhai–Macao area (①), the Chancheng District–Nanhai District area (②), the Foshan (Nanhai District)–Guangzhou (Liwan District) area (③), the Dongguan–Shenzhen area (④), the Luohu District–Futian District area (⑤), the Tianhe District–Yuexiu District–Liwan District area (⑥), and the Kowloon Peninsula–Hong Kong Island area (⑦). In these regions, residents can dine out across administrative boundaries.

Given the significant differences in the number of POIs for various types of food services, spatial kernel density analysis is employed for Chinese food, fast-food, and beverages POIs, which are in large number, while the HDBSCAN method is applied to foreign foods and snacks, with fewer POIs.

Figure 4 shows the spatial agglomeration of Chinese food, fast-food, and beverages. Like the overall spatial agglomeration of the food services in the GBA, the high-density centers for Chinese food are concentrated primarily along the Pearl River. For fast-food, the kernel density values decline significantly, and the contiguous high-density regions are reduced compared with those of Chinese restaurants, indicating a lower spatial concentration. For beverages, the kernel density values are even lower than those of fast-food, revealing a further decrease in spatial concentration. Certain peripheral districts, especially districts in Zhaoqing, Jiangmen, and Huizhou, show no signs of agglomeration of beverages. Overall, the degree of agglomeration of Chinese cuisine is higher than that of fast-food and beverages, and the distribution range of Chinese cuisine is also larger than that of fast-food and beverages.

Figure 5 displays the spatial agglomeration of foreign food and snacks in the GBA in 2021. For foreign foods, the density clusters are located primarily along the Pearl River Delta. There are 14 clusters where the number of foreign food POIs exceeds 100, which are as follows: (1) three clusters in Hong Kong (Hong Kong Island, Kowloon, and New Territories), with the largest cluster located in the New Territories; (2) four clusters in Shenzhen (Futian, Nanshan, Baoan, and Longgang) are included, with Longgang’s cluster being the second largest in the GBA; (3) two clusters in Dongguan (Binhaiqu and Dongbuqu); (4) two clusters in Huizhou (Huicheng and Huidong); (5) three clusters in Guangzhou (Tianhe, Haizhu, and Huadu); and (6) one cluster in Zhongshan (Huojukaifaqu). For snacks, the clusters are much more extensive, with clusters found in many peripheral regions. There are six clusters where the number of snacks POIs exceeds 300: (1) Huizhou (Huicheng), with the largest cluster reaching 390 POIs; (2) Zhaoqing (Duanzhou); (3) Guangzhou (Huadu); (4) Foshan (Shunde); (5) Dongguan (Dongguanchengqu); and (6) Zhuhai (Xiangzhou). In summary, the quantity and distribution range of foreign food clusters are significantly less extensive than those of snacks. Foreign food is typically more expensive, requiring higher demand thresholds. As a result, it can only be sustained in densely populated and economically developed areas. In contrast, snack services cater to a broader consumer base with lower demand thresholds, leading to a wider and more dispersed distribution of snack clusters.

4.1.2. Regional Differences in Food Services

In terms of the quantity of food services, 29 districts have more than 10,000 food service POIs. Among these, Bao’an in Shenzhen has more than 40,000 POIs, making it the district with the highest number of food services. Other districts with more than 30,000 POIs include Baiyun (Guangzhou), Longgang (Shenzhen), Nanhai (Foshan), Shunde (Foshan), and Binhaiqu (Dongguan). In contrast, the districts with the fewest food service POIs are concentrated mainly in Zhaoqing, including the counties like Fengkai, Deqing, Dinghu, and Guangning, along with Dapeng and Yantian in Shenzhen.

In terms of the structural characteristics of food services, there are significant regional differences (see

Table 2. The percentage of different food services in districts.

Rank	Chinese Food		Foreign Food		Fast-Food		Snacks		Beverages
1	Longmen	70.72%	Hong Kong Island	31.01%	Guangming	29.93%	Yuexiu	7.37%	Macao	17.17%
2	Dapeng	70.14%	Kowloon	23.24%	Longhua	28.90%	Duanzhou	6.43%	Yuexiu	15.35%
3	Huidong	65.60%	New Territories	19.45%	Longgang	28.00%	Xinhui	6.40%	Liwan	14.19%
4	Gaoyao	65.56%	Macao	17.61%	Bao’an	27.66%	Pengjiang	6.36%	Futian	13.42%
5	Sihui	65.51%	Yuexiu	6.42%	Huangpu	27.23%	Chancheng	6.36%	Nanshan	12.81%
Average	59.71%		3.75%		22.21%		4.68%		9.64%

).

(1) For Chinese food, the average proportion of Chinese food in local food services across districts is 59.71%. In districts such as Longmen, Dapeng, Huidong, Guangyao, and Sihui, Chinese food accounts for more than 65% of local food services. These districts are located on the periphery of the GBA. The high proportion in Chinese food reflects a limited diversity in food service options in these districts, highlighting a disparity between central and peripheral regions.

(2) For foreign cuisine, Hong Kong and Macao, specifically Hong Kong Island, Kowloon, and New Territories, present a high percentage of foreign cuisine. Hong Kong and Macao are the most internationalized cities in the GBA, and they feature a diverse range of foreign dining options.

(3) For fast-food, the average proportion of fast-food in local food services across districts is 22.21%. The districts of Guangming, Longhua, Longgang, and Baoan, which are all located in Shenzhen, rank as the top four districts for fast-food. This indicates the high popularity of fast-food culture in Shenzhen.

(4) For snacks, the distribution is more dispersed across the GBA because the proportions in different districts are not very high.

(5) For beverages, the top districts are Macao, Yuexiu (Guangzhou), Liwan (Guangzhou), Futian (Shenzhen), and Nanshan (Shenzhen), reflecting a significant demand for beverage services in these regions.

In this section, the findings are as follows: (1) food services in the GBA exhibit large-scale agglomeration near the Pearl River, with smaller, scattered clusters in the peripheral regions; (2) some clusters of food services extend beyond the administrative boundaries; (3) Guangzhou, Shenzhen, Foshan, and Dongguan have a significant scale of food services, whereas Hong Kong and Macao feature the highest percentages of foreign cuisine.

4.2. Colocation Pattern Between Food Services and Surrounding Complementary Services

In this section, we delve into the spatial colocation patterns between food services and other types of service. The point-to-point analysis supports a more nuanced understanding of the spatial proximity and interactions between food services and various surrounding services. We utilize the colocation quotient (CLQ) methods, which include the global colocation quotient (GCLQ) and local colocation quotient (LCLQ).

4.2.1. Global Colocation Quotient

If the GCLQ is greater than 1, then surrounding services are significantly attracting food services at a global scale, and they tend to colocate.

Table 3. GCLQs ¹ of food services attracted by surrounding complementary services.

Surrounding Services	Food Services
	1st-Order	5th-Order	10th-Order	15th-Order	20th-Order
Retail	0.7096	0.7678	0.8014	0.8202	0.8328
Transportation	0.483	0.6616	0.7357	0.7745	0.7991
Tourism	0.5625	0.6761	0.7316	0.7629	0.7834
Education	0.604	0.743	0.8011	0.83	0.8483
Leisure	0.7636	0.8461	0.8864	0.9064	0.9183
Healthcare	0.5854	0.7439	0.8084	0.8403	0.8595
Residential	0.7008	0.7849	0.8203	0.8388	0.8512

¹ All the colocation quotients are significant at the confidence level of 0.01.

shows the results of the GCLQs in different bandwidths. The GCLQs are smaller than 1, and the analysis indicates no spatial colocation between food services and the seven types of surrounding services. These results can be attributed to the strong heterogeneity in colocation patterns. In some locations, food services and surrounding services are notably colocated, while in other locations, they are not.

4.2.2. Local Colocation Quotient

The LCLQ focuses on localized interactions and offers insights into specific areas where food services and surrounding complementary services colocate relatively closely. We next apply the LCLQ to explore the colocation between food services and surrounding complementary services at the local scale. If the LCLQ is greater than 1, then surrounding complementary services are significantly attracting food services, and they tend to colocate.

Figure 6 shows the results of the LCLQ. The LCLQs greater than 1 are represented by brown and yellow points, which are labeled “colocated”. Figure 6 shows that there exists colocation between food services and surrounding services, and most brown and yellow points are agglomerated along the Pearl River.

In Appendix B, we present the results of the LCLQ to show the attractiveness of the surrounding complementary services for different types of food services (Figure A1, Figure A2, Figure A3, Figure A4 and Figure A5).

To clearly show the colocation between food services and surrounding services, we calculate the proportion of LCLQs that are greater than 1 for each service. Put simply, we count the proportions of brown and yellow points in Figure 6, Figure A1, Figure A2, Figure A3, Figure A4 and Figure A5. The results are shown in

Table 4. The proportion (%) of colocation between food services and surrounding complementary services.

	All Food	Chinese Food	Foreign Food	Fast-Food	Snacks	Beverages
Leisure	47.86	42.76	25.33	34.68	34.09	32.83
Education	41.46	37.51	22.93	32.91	28.53	29.54
Residential	37.39	34.11	25.88	32.66	35.39	31.23
Retail	33.18	31.00	9.61	28.50	14.72	19.82
Tourism	33.18	30.33	14.98	24.28	20.21	20.16
Healthcare	30.85	27.06	24.84	26.70	32.51	28.28
Transportation	24.39	25.46	20.90	24.77	27.84	27.53

. The colocation analysis indicates that 47.86% of leisure services are significantly attracting food services nearby. Similarly, 41.46% of educational services, 37.39% of residential areas, 33.18% of retail and tourism sectors, 30.85% of healthcare services, and 24.39% of transportation services exhibit notable spatial attraction to food services. Leisure, education, and residential services are three important services that tend to colocate with food services, and transportation is the least important factor in attracting food service. Across different types of food services, leisure (42.76%), education (37.51%), retail (31.00%), and tourism (30.33%) services show the strongest attraction to Chinese food. Similarly, residential and healthcare services exhibit the highest attraction to snacks. Transportation services predominantly attract snacks and beverages.

In this section, the findings are as follows: (1) Although the global perspective does not reveal significant colocation between food services and surrounding complementary services, the local analysis reveals distinct patterns of colocation in specific regions. The colocation between food services and surrounding complementary services is concentrated along the Pearl River. (2) There are 47.86% leisure, 41.46% education, 37.39% residential, 33.18% retail and tourism, 30.85% healthcare, and 24.39% transportation services attracting food services nearby. Leisure, education, and residential services are recognized as significant contributors to the spatial proximity of food services. (3) Leisure, education, retail, and tourism services attract Chinese food the most, whereas residential and healthcare services primarily attract snacks, and transportation services are notably spatially associated with snacks and beverages.

4.3. Possible Explanations

In this section, we will discuss the possible reasons for the regional spatial distribution of food services POIs and the colocation between food services and surrounding complementary services from the following three perspectives.

(1) Population Agglomeration and Land Use Expansion.

The concentration of Food services in the Greater Bay Area (GBA) predominantly aligns with the built-up regions adjacent to the Pearl River [50]. Population agglomeration has played an important role in driving the expansion of built-up areas. Over time, the expansion of the built-up areas of the GBA has broken through administrative boundaries, forming a continuous urban landscape along the Pearl River Estuary. These interconnected built-up areas are densely populated and boast substantial market potential, thus attracting the concentration of food services.

Notably, service diversity exhibits significant spatial variation, with greater variety observed in core regions near the Pearl River. First, there is a high proportion of Chinese food in the peripheral regions, and this reflects a limited diversity in food service options in these districts. This phenomenon aligns with urban economics research, and large cities are likely to have more diverse types of services [1,7], including rare cuisines which are found only in the biggest, densest cities. Second, the colocation between food services and surrounding complementary services is also concentrated along the Pearl River Estuary. This echoes existing research which found that areas with high population and employment densities are associated with complex work trip chains and more non-work activity involvement [33]. Complementary services can only be sustained when the market size is large enough.

(2) Inter-city Transportation Connectivity.

The GBA’s extensive and efficient inter-city transportation infrastructure significantly enhances residents’ mobility. By the end of 2023, there was a road mileage reaching about 60,000 km, and the highway network density reached 9.7 km per 100 square kilometers. The railway operating mileage was more than 2700 km, and the urban rail operating mileage was 1373 km [51]. This improved transportation network fosters the development of cross-border food service clusters by enabling consumer traveling across municipal boundaries. For example, Guangzhou’s subway and bus lines extend from Liwan District in Guangzhou to Nanhai District in Foshan, contributing to the formation of the cross-border food service cluster (as shown in Figure 3 by ③). Similarly, there are also buses between Dongguan (Chang’an town) and the Shenzhen (Bao’an District). Close inter-city connectivity between Dongguan and Shenzhen facilitates the formation of another cross-border food service cluster (as shown in Figure 3 by ④).

(3) History and regional dietary cultures

The culinary landscape of the GBA is deeply rooted in its historical and cultural heritage, with a dominant preference for Chinese cuisine. Thus, the distribution range of Chinese food is larger than that of fast food and beverages (see Figure 4). However, variations in social and cultural characteristics across cities lead to differences in the spatial distribution of different types of catering. First, fast food and beverage establishments, including global chains like KFC and McDonald’s and local milk tea shops, are mainly concentrated in Guangzhou, Shenzhen, and Hong Kong. These cities have a high demand for quick-service dining and beverages due to their fast-paced lifestyles and large young consumers. Second, Hong Kong and Macao stand out for the international cuisines. Due to the legacy of their colonial history and cosmopolitan character, foreign culinary cultures have been deeply integrated in Hong Kong and Macao.

5. Discussion

In terms of our findings, this study reveals new findings, different from previous studies. First, some clusters of food services expand beyond administrative boundaries (see ① to ⑦ in Figure 3). The findings correspond to the regional integration process in the megaregion characterized by increasing regional integration and transportation improvement. Second, earlier studies have found that public services such as education, cultural facilities, and hospitals have no close relation with restaurant in Beijing [9,24]. However, in the GBA, leisure and education play crucial roles in the agglomeration of food services.

This study employs spatial kernel density estimation, the HDBSCAN clustering algorithm, and colocation quotients to address the research questions. However, several challenges remain for future exploration. Spatial kernel density analysis, a well-established method for analyzing the spatial patterns of POI data, offers the advantage of requiring minimal parameter adjustments by researchers. Spatial kernel density analysis is suitable for large datasets. However, it is less effective in identifying local details of spatial agglomeration or detecting irregular clusters. In contrast, the HDBSCAN clustering algorithm excels at uncovering local details and identifying irregular spatial clusters, particularly in smaller datasets. The HDBSCAN clustering algorithm offers advantages over traditional algorithms, such as DBSCAN and K-means, demonstrating greater robustness in handling varying densities of data. Nonetheless, its parameter selection is relatively subjective, necessitating a deeper understanding of the data by the researchers. The colocation quotient, used to examine the spatial agglomeration between two types of POI data, also presents challenges. Its parameter settings are similarly subjective, and the computational demands can be significant. Further research is needed to refine the spatial weight values for more robust analysis and to develop efficient methods for calculating colocation quotients.

This study has four limitations: first, although we examine different types of food services, the agglomeration of food services at different levels is unclear; second, we use POI data from only 2021 to study the agglomeration of food services, limiting the ability to analyze temporal changes; third, we do not consider the impact of the internet on the agglomeration of food services; fourth, we do not explore the mechanisms of the agglomeration of food services and their colocation with surrounding services in an empirical way.

Our findings provide implications for policy and urban planning. First, we find that some food clusters cross administrative boundaries. Considering that the agglomeration of food services attracts many consumers, the agglomeration of food services providers and consumers might create challenges in urban management. To mitigate potential negative externalities, such as traffic congestion, environmental pollution, and noise, joint governance and coordinated urban management across these areas are essential. In addition, the spatial planning of services should follow the trend of integration, and equal attention must also be paid to the needs of residents in border areas.

Second, leisure, education, residential, retail, tourist, healthcare, and transportation services usually colocate with food services. As shown in Table 4, leisure has the highest degree of colocation with food services (47.86%), followed by education (41.46%). However, the overall colocation between other services and food services remains below 40%, indicating that food service development often lags behind the expansion of transportation, tourism, and other urban services. This lag can hinder the ability of a given locality to meet the diverse consumption needs of residents. In future urban planning, more attention should be given to achieving adequate supporting services to foster balanced development. For instance, suburban and new town planning typically prioritizes industrial and residential developments [52], and soft infrastructure such as restaurants in new urban districts remains at a relative disadvantage [53]. A study of GBA revealed that newly developed urban areas suffer from a lack of public amenities and commercial facilities, exacerbating the disparity in development between central and peripheral urban regions [50]. Integrating food services and other supporting amenities into these plans is essential for fostering sustainable and inclusive urban growth.

6. Conclusions

This study explores the spatial distribution of food services and their colocation with surrounding complementary services in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) in 2021 via point of interest (POI) data and methods such as spatial kernel density estimation, the HDBSCAN clustering algorithm, and colocation quotients. Its findings are as follows: (1) the food services within the GBA exhibit a pronounced spatial concentration near the Pearl River, and some clusters of food services extend beyond administrative boundaries in the core areas of the GBA. (2) Leisure, education, and residential services are identified as important factors attracting the proximity of food services. Leisure, education, retail, and tourism services exhibit the strongest attraction to Chinese food, whereas residential and healthcare services show a closer association with snacks. Additionally, transportation services are particularly linked to snacks and beverages.

The potential contributions of this research are as follows: (1) we explore the agglomeration of urban services at the megaregion level, which helps to identify cross-boundary clusters in food services; (2) we confirm the colocation of food services with surrounding complementary services, contributing to the understanding of complementary agglomeration in urban settings.

This research points out several research directions for future research: (1) Future research may consider hierarchical heterogeneity in the spatial agglomeration of catering services. The agglomeration of catering services at different levels is another important dimension of consumption. For example, top restaurants tend to be located near other top restaurants [17]. Restaurants with higher prices are more clustered than restaurants with lower prices [3]. (2) Future research should incorporate data over a longer period to explore the dynamics of food service agglomeration. For example, during the COVID-19 pandemic, food services were destabilized [54,55]. With more data gathered over a greater time period, the dynamic of the agglomeration of food services can reflect the dynamic of urban resilience [54,55]. (3) This paper does not distinguish the differences between online and offline catering services. Online catering services are also important consumer amenities. Future studies can explore the differences in the agglomeration of online and offline food services and their colocation with other surrounding services. (4) Future research can use various data to verify and explain the mechanisms of the services’ agglomeration.