This study was conducted in Guangzhou, China. Guangzhou is one of China’s five national central cities, the capital of Guangdong Province, and the main hub of the Pearl River Delta [15] . It was built 2228 years ago and was the first city in China to establish trade relations with the outside world, known as China’s “southern gateway” [15] [16] . Guangzhou is a modern city in China, serving as the political, economic, educational, cultural, and technological center of Guangdong Province and southern China [15] . The city covers an area of 7434 square kilometers and has a population of over 20 million with 10 million short-term residents [15] . Guangzhou is highly urbanized. The proportion of the urban population has continued to grow over the past seven years, exceeding 80%, far above the national average [16] . The location map of Guangzhou is shown in Figure 1 .

Guangzhou is China’s third largest medical center, with 5093 medical facilities (including 269 hospitals) and 6.5 beds per 1000 people. Health services are available in 15 minutes in urban and 30 minutes in rural areas. Eight national or regional medical centers in Guangzhou, including the First Affiliated Hospital of Guangzhou Medical University, are registered as the National Respiratory Medicine Center. The top 9 national hospitals in Guangdong are located in Guangzhou, among which 5 are specialized hospitals for treating COVID-19 patients.

Guangzhou is an international transportation hub with 15.3 million residents in southern China. On January 21, 2020, Guangzhou reported its first COVID-19 imported case. A total of 557 cases of COVID-19 symptoms were reported from January 21 to June 30, 2020. Since May 2, 2020, no local cases have been reported for more than 5 months. Guangzhou is one of the cities that experienced the early, mid-term, and recovery stages of the COVID-19 outbreak [17] .

The study has used COVID-19 testing sites located in Guangzhou City and sites’ locations and the information obtained from open Application Programming Interface (API) (platform of Gaode Maps ( https://www.amap.com/ ), which is one of the most extensive Web mapping, navigation, and Location-based services (LBS) platforms in China [18] . Population data for 11 districts of Guangzhou City were collected from the Guangzhou Statistics Bureau [19] .

Point pattern analysis using ArcGIS was calculated to examine the spatial dynamics of COVID-19 testing sites in Guangzhou City. Assessing spatial dynamics is essential for understanding how testing sites are varied. Both density-based and distance-based approaches were applied in this paper as follows;

Density-based approaches characterize the distribution pattern of points, and this paper used Kernel density to identify the spatial dynamics of COVID-19 testing sites in Guangzhou City. The kernel density method is an extension of the square method: like the square density, the kernel method calculates the local density of a subset of the study area, but its corresponding terms for the square density are different, with overlapping sub-regions providing movable subsets. The kernel density method generates a density grid with pixel sizes smaller than the kernel window. Assign the calculated density value of the Kernel window centered on each pixel. Equation (1) describes the Kernel Density estimation.

$\lambda \left(s\right)={\displaystyle {\sum }_{i=1}^n\frac{1}{\pi r^2}k\left(\frac{d_{is}}{r}\right)}$(1)

where $\lambda \left(s\right)$ is the density at location s, r is the search radius of the Kenel Density Estimation, k is the weight of a point i at distance d_is to location s. k is usually modeled as a function of the ratio between d_is and r. As a result, rather than choosing a uniform function that gives equal weight to all points within the radius r [20] .

The spread or distribution of something over a geographic area is measured via nearest-neighbor analysis. It offers a number that expresses how closely together or evenly spaced a group of points is. ANNA assesses the average distance between each point in the Guangzhou area and its nearest point. This paper calculated the nearest neighbor index based on the average distance between each COVID-19 testing site and its nearest neighbor site. The average distance from the nearest neighbor measures the rarity of points in the distribution.

The average nearest-neighbor ratio is determined as the difference between the observed and expected average distances, where the expected average distance is based on a fictitious random distribution with the same number of features covering the same total area [21] . ANNA can calculate using the Equation (2).

$\text{ANNA}={\bar{D}}_O/{\bar{D}}_E$(2)

where ${\bar{D}}_O$ is the observed mean distance between each feature and its nearest neighbor and ${\bar{D}}_E$ is the expected mean distance for the features given in a random pattern.

This paper used the word cloud plots to examine the characteristics of COVID-19 testing sites in Guangzhou City. A word cloud is a picture that visually displays word data. In other words, it is a group, or cluster, of words shown in various sizes. The more often and how essential a word is stated in a document, the larger and bolder it appears. Due to its ability to quickly reveal the most important or popular topics, word cloud visualization is particularly popular among presenters. Word clouds can be used as a summary, providing instant analysis & visualization of word data and feedback and providing insight into the most popular concepts or convey sentiment [22] . This paper used R’s word cloud package and the Rcolor Brewer package for colors. Additionally, the wordcloud2 package in R, which has a slightly different appearance and entertaining uses, can also be used to produce word clouds. This method has been used to investigate the attribute data described in the testing locations in Guangzhou City. It has imaged the keywords found in the attribute, and the size depicts the frequency of the word.

A network is a collection of interconnected components, such as edges (lines) and junctions (points), that show potential paths between two points. Utility networks and network datasets are the two types of networks that fall under the ArcGIS software. The ArcGIS Network Analyst extension to create and edit network datasets. Network analysis can be used to identify the quickest way to get from one selected location to another location [23] . This paper mainly used a location-allocation tool to assess the testing facilities from each centroid and assess the new location requirement to locate testing sites in each district with the demand.

Mathematically, the length of a line segment connecting any two points in Euclidean space is known as the Euclidean distance between them. The minimum distance between two pairs of points from the two objects is typically used to determine the distance between two objects that do not point. The distance between two points is known as the Euclidean distance in coordinate geometry. The length of the line segment separating the two places should be measured to determine the distance between them [24] . Euclidean distance can be calculated using Equation (3). Where “d” is the Euclidean distance, (x₁, y₁) is the coordinate of the first point, and (x₂, y₂) is the coordinate of the second point.

$d=\sqrt{{\left(x_2-x_1\right)}^2+{\left(y_2-y_1\right)}^2}$(3)

Euclidean allocation produces a grid where each empty cell is assigned to the source nearest to it. The nearest source is determined by Euclidean distance. Sources are all of the cells with a value, including zero. Empty cells are cells with no value. The value of the nearest target cell from the input picture, as determined by the Euclidean distance (i.e., straight-line distance), is assigned to grid cells in the output image using this tool [25] . This paper has used this to understand each testing site’s service area and estimate the areas for new testing locations in the study area.

Density mapping demonstrates where points or lines could be clustered in a specific region. These maps frequently use interpolation techniques to predict where a particular feature’s concentration (for example, population) could be throughout a given surface. Point density, line density, kernel density, and choropleth maps are examples of different density maps [26] . This study calculated the population density to understand the area with a higher population in Guangzhou City. Population density is the number of people in a given area of land. It primarily applies to people, yet it occasionally does to other living things. It is a crucial geographic phrase. The number of people living in an area per square kilometer, or another unit of land area, is called population density. Population density can be calculated using the following equation.

Population density = number of people ÷ the area they occupy(4)

The density map is presented using choropleth maps, and a choropleth map is a kind of statistical thematic map that uses color that correspond to an overall summary of a geographic characteristic within spatial enumeration units, such as population density. A simple way to see how a variable varies geographically or the degree of variability within a region is to use choropleth maps. The flow chart of the paper methodology is shown in Figure 2 .

The spatial distribution of COVID-19 testing sites and population density in Guangzhou City is shown in Figure 3(a) . There are 1183 total testing sites, and the results indicated that spatial disparities could be noticed. The highest testing sites are in Haizhu District, located in the city area with a 1,819,037 total population. The higher population in the Haizhu District lives in an area of 92.04 sq∙km. The population density of the Haizhu District is 19,763, and it is the third highest in Guangzhou City. The number of testing sites and population density of each district are shown in Table 1 . The highest population density (30,841) can be recognized in Yuexiu District, and the second highest is in Liwan District. There are 94 and 55 total testing sites, respectively. The lowest testing sites, 43, are located in the Huadu District, with a population density of 1699. Results revealed in this study have shown that testing facilities are not equally distributed throughout the study area.

In addition, inner cities of the Guangzhou Province, Yuexiu, Liwan, Tianhe, and Haizhu, have shown the highest aggregation of the COVID testing sites. The outer part of the inner cities (suburbs), mainly Luogang, Baiyun, Huangpu, and Panyu, have recorded fewer testing sites. However, some cluster areas can be recognized in the suburbs of Guangzhou. Outer city areas (primarily rural): Conghua, Huadu, Zengcheng, and Nansha have the lowest-density population and the lowest number of testing cities. Moreover, two prominent clusters can be recognized in the Conghua and Zengchend areas. Therefore, significant urban-rural disparities can be observed with the spatial distribution of COVID testing sites in the province, and testing sites have also been distributed according to the population density of the study area. The inner and outer suburbs have many COVID testing sites, and the outer part of the province has fewer health facilities than the Guangzhou Province.

The result of the ANNA is shown in Figure 3(b) . According to Figure 3(b) , testing locations are highly clustered. The nearest neighbor ratio is 0.36. Testing sites in Guangzhou City are concentrated in the city area, and the highest density can be recognized ( Figure 3(c) ). It has gradually decreased from the city area to the outer part of the city. Some higher-density clusters are located in every district in Guangzhou city. As the above results described, there are higher disparities in testing sites’ spatial allocation in Guangzhou City.

This paper presents characteristics of testing sites using word cloud plots in Figure 4 . As shown in Figure 3 , testing sites are mainly distributed around the urban areas. Guangzhou and Guangdong are the significant areas with the highest testing sites. They have focused on locating very close to the roads and cities. Guandong has repeated 936 times, and most sites are distributed in the city areas. Testing sites are in green areas, villages, parks, and riversides. However, there are few testing sites in these areas.

Hospitals and health care centers are the primary testing locations in Guangzhou, and health has been repeated 1174 times. Hospitals have shown 174 repeated times in the word cloud plots in the study area. Schools, stations, and university locations are included but not prominent. Conghua District, Zengcheng District, and Huadu District have shown lower population density than the other districts in Guangzhou City, 362 sq∙km, 908 sq∙km, and 1699 sq∙km, respectively. Nansha District has a low-density population (1280 sq∙km) compared to other districts. However, it consists of 106 testing sites around the city. Liwan has a 19,842 population density, and only 55 testing sites can be recognized. Baiyun District has the highest population of 3,742,991 and 171 testing sites. These testing sites are not distributed equally throughout the district. The second highest population density (2,658,397) is in Panyu District, and only 74 testing sites are located there.

Euclidean distance, allocation, and location-allocation methods are used to assess the accessibility, equality, and inequalities of testing site distribution in Guangzhou city. The Euclidean distance and allocation are shown in Figure 5(a) and Figure 5(b) . Good accessibility and equal facilities are distributed in the study area’s higher urban and southern parts. Figure 5(a) shows that the lowest distance between locations can be recognized in the Baiyun District, Tianhe District, Yuexin District, Liwan District, and Haizhu District. Testing site locations in the south part are also shown a similar pattern. However, the maximum distance between the points is nearly 10 km, and a higher distance can be recognized at the edge of the study area in the southern part.

Moreover, the north part of the study area presented a different pattern, and the highest distance of 18 km was recorded. Most of the testing sites in the North Part have been distributed monotonously along the main roads. However, some major town areas show a clustered pattern of testing site distribution. According to these results, the testing sites are not distributed equally throughout the study area.

The Euclidean allocation has shown the estimation of a service area for each testing location in the study area. Small service areas can be depicted in the urban cities in the study area, and it is less than 1 sq∙km from the testing site. However, this pattern is unequal throughout the study area, and the highest service area can be recognized, Conghua District, in the northern part of the study area. The highest service area is more than 200 sq∙km. Service areas with >100 sq∙km are located in Conghua, Huadu, and Zengcheng Districts in Guangzhou City. Furthermore, the upper part of the study area is also shown as a small service area with less than one sq∙km along the main roads, and other village areas have not shown good accessibility for the COVID-19 testing sites. Almost 21 testing sites have shown 50 to 100 sq∙km as their service area.

Figure 6 shows the distance from the districts’ centroids to each testing site in Guangzhou City. There are 255 testing sites away from the city, and >ten sq∙km distance can be shown. Around 30 locations are very close to the center locations, and 809 locations have shown a 1 - 10 sq∙km distance from the city centers. Very few testing centers are located in rural areas, and there is no high relative accessibility. The largest districts, namely Conghua District, Zengcheng District, and Huadu District, lack testing services. Testing sites are located on every side of the centers of Baiyuan District, Tianhe District, Yuexin District, Haizhu District, and Liwan District. However, other districts have not covered all the areas, and testing sites are insufficient to cover the whole area with reasonable accessibility. Sometimes, people must travel more than 10 km for a testing site in these districts. However, the testing location distribution of Panyu District is better than other districts. Inequalities in the spatial distribution and accessibility of testing sites can be identified.

Mainly, the inner suburbs of the Guangzhou Province have shown reasonable proximity and accessibility to COVID testing sites, which is significant for good accessibility. The people of this area can be reached at the testing sites traveling a short distance, and all sites are located within a minimum of 1 km area. Most of them have equal accessibility. However, this situation is recorded differently in the study area’s outer or rural part, and low accessibility can be recognized.