Finding Optimal Stations Using Euclidean Distance and Adjustable Surrounding Sphere

Air quality monitoring network (AQMN) plays an important role in air pollution management. However, setting up an initial network in a city often lacks necessary information such as historical pollution and geographical data, which makes it challenging to establish an effective network. Meanwhile, cities with an existing one do not adequately represent spatial coverage of air pollution issues or face rapid urbanization where additional stations are needed. To resolve the two cases, we propose four methods for finding stations and constructing a network using Euclidean distance and the k-nearest neighbor algorithm, consisting of Euclidean Distance (ED), Fixed Surrounding Sphere (FSS), Euclidean Distance + Fixed Surrounding Sphere (ED + FSS), and Euclidean Distance + Adjustable Surrounding Sphere (ED + ASS). We introduce and apply a coverage percentage and weighted coverage degree for evaluating the results from our proposed methods. Our experiment result shows that ED + ASS is better than other methods for finding stations to enhance spatial coverage. In the case of setting up the initial networks, coverage percentages are improved up to 22%, 37%, and 56% compared with the existing network, and adding a station in the existing one improved up by 34%, 130%, and 39%, in Sejong, Bonn, and Bangkok cities, respectively. Our method depicts acceptable results and will be implemented as a guide for establishing a new network and can be a tool for improving spatial coverage of the existing network for future expansions in air monitoring.


Introduction
Air quality monitoring networks (AQMN) are established as tools that determine policies and strategies for achieving air quality standards. A plan for designing an AQMN depends on objectives such as urban planning, environmental policies, and budget. Generally, designing an AQMN is done by environmental authorities or governmental organizations based on empirical judgments. An expert group assesses various criteria to make their decisions, such as budget and population thresholds. These play a significant role in determining the required number and location of monitoring stations. For example, Thailand and South Korea set up air monitoring stations in government office areas to reduce the cost of installation, maintenance, and the safety of the devices [1,2]. Guidelines of the USA and Australia suggest installing new air quality monitoring stations based on population size [3,4]. In such cases, monitoring stations are often considered in an ad hoc fashion.
The critical issues in AQMN designing are separated into a setting up network for allocating optimum stations and optimizing the existing network to better reflect air quality in the area. There are different methods to design a new network, all of which must

Proposed Methods
In this study, finding optimal stations' main target goal is achieving the maximum spatial coverage while still preserving appropriate overlapping areas. The maximum spatial coverage designed can be realized through the optimal placement of the stations in the city. Moreover, maintaining a relevant overlapped area can enhance effectiveness for more reliable and accurate data collection. Accordingly, in this section, we proposed finding the stations based on the Euclidean Distance and the k-nearest neighbor's algorithm (k-NN). The proposed methods framework consists of four main methods and two evaluation criteria, as shown in Figure 2. Bangkok is the capital city of Thailand. The city is located in almost the middle of Thailand and occupies 1568.7 km 2 . Because this city has the highest population density in Thailand, Bangkok city has an air pollution problem from traffic and energy consumption. There are 12 air quality monitoring stations from the pollution control department, which

Proposed Methods
In this study, finding optimal stations' main target goal is achieving the maximu spatial coverage while still preserving appropriate overlapping areas. The maximum sp tial coverage designed can be realized through the optimal placement of the stations the city. Moreover, maintaining a relevant overlapped area can enhance effectiveness more reliable and accurate data collection. Accordingly, in this section, we proposed fin ing the stations based on the Euclidean Distance and the k-nearest neighbor's algorith (k-NN). The proposed methods framework consists of four main methods and two ev uation criteria, as shown in Figure 2. In the proposed framework, our input consists of a map and parameters. The map of the study area is divided into a square of continue grid. A centroid of each square grid is a point that consists of latitude-longitude coordinates. For simple understanding throughout this paper, we use the term "location" (L i ), i = 1 . . . N to represent the centroid of a grid. The parameters consisted of (i) a grid index at the center of the map, (ii) a specified number of stations, and (iii) the diameter length of the surrounding sphere of each station as described in Section 6.1. Next, we give stations in area A, which consist of η stations. The value S = {S j , . . . , S η } is a list of stations. Each station S j , j = 1 . . . η is at a location in our study area. All of the input will use to calculate by our proposed methods. M1: Euclidean Distance (ED) and M2: Fixed Surrounding Sphere are the initial method for finding the next stations. The M3 method is a combination of previous methods. The M4 is an upgraded version of M3. Our evaluation criteria, coverage percentage (COV), and weighted coverage degree (WCD) are used to evaluate the results from the methods then return the optimal network. The comprehensive methods are described in the following sections.

Euclidean Distance (ED)
In this study, we apply the Euclidean Distance function to calculate distance from location (L i ) to three nearest neighbor stations {NS 1 , NS 2 , NS 3 }, as illustrated in Figure 3. We use k-NN (k = 3) because the nearby stations can exchange data reliability with existing stations. Let NS 1 , NS 2 , and NS 3 be members of the set of nearest neighboring stations of L i , where L i is a position to calculate ED i . We calculate the Euclidean Distance (ED i ) at any L i as:

Fixed Surrounding Sphere (FSS)
The fixed surrounding sphere (FSS) method was inspired by the original sphere o influence (SOI) [12]. We applied such an idea for determining the area surrounding eac station without using pollution concentration data. The constant value is identified to diameter length of FSS with reference to city air quality monitoring network designing i Seoul city [2]. In Seoul's existing air monitoring network, the 25 stations are located ap proximately 5 km away from each other. They are not located close to the road, high emis sion concentration sources, or high-density population areas. On the other hand, thes stations are distributed throughout the city. Consequently, we predefined a fixed diame ter (dms) of FSS to divide the covered and non-covered areas under stations. As shown i Figure 4, the triangle symbols represent stations. The surrounding covered area of th stations is illustrated as a circle shape in Figure 4a and a pie shape in Figure 4b by deter mining a diameter length. The areas outside represent the non-covered areas. The shape depend on the position of stations on the map. However, such areas are defined as cov ered areas, although the shapes of the areas are different. ED i denotes Euclidean Distance at L i , where i is an index of the location, and each parenthesized value is the distance from L i to NS 1 , NS 2 , and NS 3 , respectively. If k < 3, then we adapt the equation by using only existing stations. For example, if k = 2, then D L i ,NS 3 is equal to zero. Thus, ED i is calculated using only D L i ,NS 1 and D L i ,NS 2 .
In order to calculate and find the next station, the following definitions are used. Let S denote a list of stations in a network and let L t be a candidate station to be evaluated for the possibility of it being the next station. Thus, the current station network is S + L t . Here, we calculate ED i according to Equation (1). Subsequently, we sum ED i values as: TED Lt denotes the total of ED i , where L t is a candidate for the next station, and N is the number of locations. The candidate with the lowest TED Lt will be defined as the additional station.

Fixed Surrounding Sphere (FSS)
The fixed surrounding sphere (FSS) method was inspired by the original sphere of influence (SOI) [12]. We applied such an idea for determining the area surrounding each station without using pollution concentration data. The constant value is identified to a diameter length of FSS with reference to city air quality monitoring network designing in Seoul city [2]. In Seoul's existing air monitoring network, the 25 stations are located approximately 5 km away from each other. They are not located close to the road, high emission concentration sources, or high-density population areas. On the other hand, these stations are distributed throughout the city. Consequently, we predefined a fixed diameter (dm s ) of FSS to divide the covered and non-covered areas under stations. As shown in Figure 4, the triangle symbols represent stations. The surrounding covered area of the stations is illustrated as a circle shape in Figure 4a and a pie shape in Figure 4b by determining a diameter length. The areas outside represent the non-covered areas. The shapes depend on the position of stations on the map. However, such areas are defined as covered areas, although the shapes of the areas are different.

Fixed Surrounding Sphere (FSS)
The fixed surrounding sphere (FSS) method was inspired by the original sphere influence (SOI) [12]. We applied such an idea for determining the area surrounding ea station without using pollution concentration data. The constant value is identified to diameter length of FSS with reference to city air quality monitoring network designing Seoul city [2]. In Seoul's existing air monitoring network, the 25 stations are located a proximately 5 km away from each other. They are not located close to the road, high em sion concentration sources, or high-density population areas. On the other hand, the stations are distributed throughout the city. Consequently, we predefined a fixed diam ter (dms) of FSS to divide the covered and non-covered areas under stations. As shown Figure 4, the triangle symbols represent stations. The surrounding covered area of t stations is illustrated as a circle shape in Figure 4a and a pie shape in Figure 4b by dete mining a diameter length. The areas outside represent the non-covered areas. The shap depend on the position of stations on the map. However, such areas are defined as co ered areas, although the shapes of the areas are different.  The procedure to classify the covered and non-covered areas under a station are described in this subsection. Let station S j be a location of station and the diameter of S j be equal to dm s . The location (L i ) is covered (monitored) by a station S j when the distance from L i to S j is less than dm s 2 and also such L i will be members of all covered areas for A sj . Equation (3) represents the probability p(L i ,S j ) that a location is covered by station S j .
i is the index of location from 0 to N − 1, where N represents the number of locations in area A, and j provides an index of stations in S.
The Algorithm 1 Fixed Surrounding Sphere (FSS) can be described as follows: Equation (3) represents the probability p(Li,Sj) that a location is covered by station p , = 1: "covered area", ( , ) ≤ 2 0: "non-covered area", ℎ ∈ i is the index of location from 0 to N − 1, where N represents the number of lo in area A, and j provides an index of stations in S.
The Algorithm 1 Fixed Surrounding Sphere (FSS) can be described as follow

10:
The Lt with the smallest value of #poor will be chosen as the next location of station.

11:
Lt is permanently appended to the station list S. This method combines the concepts Euclidean Distance (ED) and Fixed Surro Sphere (FSS). However, the difference between this method and the previous on location of the first station. For ED and FSS, the first stations are set at the cente map, while for ED + FSS, all map locations are tried as a first station. The outpu method is a multi-list of stations in which the first stations are different. Conseq further evaluation criteria have been introduced to evaluate the best network with imum of spatial coverage.
We divide ED + FSS into two processes: (i) finding the next stations and (ii) eva a maximum coverage percentage while still preserving an appropriate overlapp for the network. The first process, finding the next stations, is described with Algo Euclidean Distance + Fixed Surrounding Sphere (ED + FSS) as follows:

Euclidean Distance + Fixed Surrounding Sphere (ED + FSS)
This method combines the concepts Euclidean Distance (ED) and Fixed Surrounding Sphere (FSS). However, the difference between this method and the previous one is the location of the first station. For ED and FSS, the first stations are set at the center of the map, while for ED + FSS, all map locations are tried as a first station. The output of this method is a multi-list of stations in which the first stations are different. Consequently, further evaluation criteria have been introduced to evaluate the best network with a maximum of spatial coverage.
We divide ED + FSS into two processes: (i) finding the next stations and (ii) evaluating a maximum coverage percentage while still preserving an appropriate overlapped area for the network. The first process, finding the next stations, is described with Algorithm 2 Euclidean Distance + Fixed Surrounding Sphere (ED + FSS) as follows: The Lt with the smallest value of TEDLt will be chosen as the next location of station.

12:
Lt is permanently appended to the station list S. After we obtain a multi-list of stations, the next process is to evaluate the maximu coverage percentages of all Sx. The ED + FSS will use two criteria, Coverage Percenta (COV) and Weighted Coverage Degree (WCD), for selecting the best network. The scriptions of COV and WCD criteria are outlined in Section 5.

Euclidean Distance + Adjustable Surrounding Sphere (ED + ASS)
This method is an upgrade from ED + FSS. We change from using a fixed diame to an adjustable diameter, which depends on a station's covered area proportion. O proposed method considers economic benefits based on deployment costs and station cation with the highest spatial coverage for a specified number of economically feasi stations. The covered area proportion can be adjusted as needed. For example, if the vironmental authority has a budget limit for establishing monitoring stations, an arbitr ratio can be predefined as 50% or 70%. On the other hand, if they require high spa resolution monitoring or an unlimited budget, the ratio can be predefined as 10% or 30 As a warmup to ED + ASS, we select a station at the center of the map (Scenter) calculating the length of the diameter. Next, we calculate EDi from all locations (Li) map to the Scenter with Equation (1). We pass all the EDi values into a list. Subsequently, sort the obtained list in ascending order, which means that locations closer to the stat with lower EDi values will be at the beginning of the list. The cut off position correspon to an arbitrary ratio, as mentioned above. We calculate the cut off position value w Equation (4).
ratio is predefined covered area proportion of the station, length (list) is equal N where represents the total number of locations in area.
After we obtain a multi-list of stations, the next process is to evaluate the maximum coverage percentages of all S x . The ED + FSS will use two criteria, Coverage Percentage (COV) and Weighted Coverage Degree (WCD), for selecting the best network. The descriptions of COV and WCD criteria are outlined in Section 5.

Euclidean Distance + Adjustable Surrounding Sphere (ED + ASS)
This method is an upgrade from ED + FSS. We change from using a fixed diameter to an adjustable diameter, which depends on a station's covered area proportion. Our proposed method considers economic benefits based on deployment costs and station location with the highest spatial coverage for a specified number of economically feasible stations. The covered area proportion can be adjusted as needed. For example, if the environmental authority has a budget limit for establishing monitoring stations, an arbitrary ratio can be predefined as 50% or 70%. On the other hand, if they require high spatial resolution monitoring or an unlimited budget, the ratio can be predefined as 10% or 30%.
As a warmup to ED + ASS, we select a station at the center of the map (S center ) for calculating the length of the diameter. Next, we calculate ED i from all locations (L i ) on map to the S center with Equation (1). We pass all the ED i values into a list. Subsequently, we sort the obtained list in ascending order, which means that locations closer to the station with lower ED i values will be at the beginning of the list. The cut off position corresponds to an arbitrary ratio, as mentioned above. We calculate the cut off position value with Equation (4).
Cut o f f position = round(ratio × length (list)) (4) ratio is predefined covered area proportion of the station, length (list) is equal N where N represents the total number of locations in area.
In the sorted list, we access the index of the list at the cut off position and get the value of that ED i value. The value of ED i is multiplied by two and defined as a diameter length of the ED + ASS method. Once, we have obtained the diameter length, and then we can continue the procedure of establishing the station network and evaluating the global maximum coverage of the station network in the same way as we did with ED + FSS.

Evaluation Criteria
In ED + FSS and ED + ASS, there is a multi-list of station networks with different first locations that must be processed to evaluate the maximum coverage percentage. Consequently, in order to find the best station network, the evaluation criteria are designed and described next.

Coverage Percentage (COV)
The covered area of stations (A S ) can be determined using Equation (3). Given the list of stations S located in a study area, we can assess the k-coverage when a location is covered by at least k different stations. The parameter k is called Coverage Degree. It means at least k stations cover each location in the study area. There are previous studies of wireless network coverage that have discussed the required k value of a network. Such studies explain a proper value of k that depends on the application. For example, an application requires k = 1 in a monitoring environment in which fault tolerance is not important. Meanwhile, k > 1 should be used when stronger monitoring is required, such as in an industrial or dangerous chemical region. Furthermore, in cases requiring fault tolerance, k ≥ 3 is required. Therefore, it is clear that the station networks with higher k-coverage are more reliable [32,33].
Suppose that there are four stations. The circle shapes represent covered areas of a station and × symbols represent L i for coverage degree assessment. If × symbols are within the covered areas of one station, then we define such L i as a C1. If × symbol lies within the covered areas of two, three, and four stations, then it is denoted by C2, C3, and C4 as shown in Figure 5a-d, respectively. The area outside the circle is defined as a C0, which means it is a non-covered area. length of the ED + ASS method. Once, we have obtained the diameter length, and then we can continue the procedure of establishing the station network and evaluating the global maximum coverage of the station network in the same way as we did with ED + FSS.

Evaluation Criteria
In ED + FSS and ED + ASS, there is a multi-list of station networks with different first locations that must be processed to evaluate the maximum coverage percentage. Consequently, in order to find the best station network, the evaluation criteria are designed and described next.

Coverage Percentage (COV)
The covered area of stations (AS) can be determined using Equation (3). Given the list of stations S located in a study area, we can assess the k-coverage when a location is covered by at least k different stations. The parameter k is called Coverage Degree. It means at least k stations cover each location in the study area. There are previous studies of wireless network coverage that have discussed the required k value of a network. Such studies explain a proper value of k that depends on the application. For example, an application requires k = 1 in a monitoring environment in which fault tolerance is not important. Meanwhile, k > 1 should be used when stronger monitoring is required, such as in an industrial or dangerous chemical region. Furthermore, in cases requiring fault tolerance, k ≥ 3 is required. Therefore, it is clear that the station networks with higher k-coverage are more reliable [32,33].
Suppose that there are four stations. The circle shapes represent covered areas of a station and × symbols represent Li for coverage degree assessment. If × symbols are within the covered areas of one station, then we define such Li as a C1. If × symbol lies within the covered areas of two, three, and four stations, then it is denoted by C2, C3, and C4 as shown in Figure 5a, 5b, 5c, and 5d, respectively. The area outside the circle is defined as a C0, which means it is a non-covered area. We calculate the coverage percentage using the count Li in each of the coverage degrees. The coverage percentage (COV) is used to evaluate results in our proposed methods and can be calculated with Equation (5).
Ci denotes the summation of Li in coverage degree, i is k-coverage number, k indicates a number of stations in the network, and N is total number of locations in study area A.

Weighted Coverage Degree (WCD)
Weighted coverage degree (WCD) is an additional criterion. It is used whenever COV cannot give a unique answer to the best network. The WCD corresponds to number of We calculate the coverage percentage using the count L i in each of the coverage degrees. The coverage percentage (COV) is used to evaluate results in our proposed methods and can be calculated with Equation (5).
C i denotes the summation of L i in coverage degree, i is k-coverage number, k indicates a number of stations in the network, and N is total number of locations in study area A.

Weighted Coverage Degree (WCD)
Weighted coverage degree (WCD) is an additional criterion. It is used whenever COV cannot give a unique answer to the best network. The WCD corresponds to number of stations and k-coverage. For example, if the study area has four stations, here the k-coverage k = 4 (coverage degree: C0, C1, C2, C3, C4) and the weight has five values of coverage degree. We calculate the weight value by employing the Divide and Conquer concept. The ∑W i is a value equal to one. The W k is calculated from ∑W i divided by two. The next weight value at W k-i will decrease from the previous by half, which means W k-i = W k divided by two. The weight calculation is done continuously until the last weight at W 0 is set equal to W 1 . Table 1 is an example of weight value generation when the coverage degree is equal to 4. The WCD can be calculated by coverage degree weighting with Equation (6).
. . , k}, k indicates a number of stations in the network, and the weights are associated to Coverage Degree (C i ).

Results and Discussion
This section explains the parameters used in the experiment and compares each method's pros and cons. In scenario 1, setting up a network by considering four cases as following, (i) spatial coverage, (ii) performance of coverage percentage versus the number of the stations added incrementally, (iii) coverage percentage versus a specified number of stations, and (iv) flexibility to apply our methods to different cities. In scenario 2, finding an additional station to improve the current network is evaluated.

Experimental Parameter Settings
The parameter settings are shown in Table 2. The area size (A) is the number of locations in study areas. The centers of the maps are located at indices 1056, 3516, and 218 for Sejong, Bangkok, and Bonn. The specified number of stations is equal to the existing stations in the cities. The diameter length of the M3: ED + FSS is a fixed value of ten kilometers and the M4: ED + ASS is predetermined. We consider the proportion of covered area from reasonable based on the number of existing stations, as shown in Figure 6. Finally, we determined that 30% is a proper value for our experiment.   Table 3 shows M1: ED, M2: FSS and compares them with the existing stations in Sejong city. In Table 3, triangle symbols depict existing stations, and circles represent the spatial coverage. Let us consider a result in M1: ED, where we defined the first station at the center and used four stations as input parameters. The output from M1: ED is shown in Table 3 (b). Three stations are located near each other, but one station is located far away from its neighbors. We found significant inefficient spatial coverage in that case because almost all covered areas overlap. The COV of M1: ED is 57%, which is a 16% decrease in the current value, and WCD shows a value of 12.58, which increased 48% when compared with the current value. Next, consider the result in M2: FSS; we used the input parameters as same as M1: ED except adding a fixed diameter of ten kilometers to be used in the calculation. The result shows that most stations are located apart from the first station at the center, as shown in Table 3 (c). As a result, M2: FSS achieves the best coverage percentage up to 91% and an increase of 34% compared to the current value. On the other hand, the WCD value shows 7.44, which decreased by 13%.  The M1: ED shows large overlapping areas which make a strengthened area with neighboring stations. The area of overlap can make data more reliable in case of data verification between neighboring stations and also produce a network which is fault tolerant.  Table 3 shows M1: ED, M2: FSS and compares them with the existing stations in Sejong city. In Table 3, triangle symbols depict existing stations, and circles represent the spatial coverage. Let us consider a result in M1: ED, where we defined the first station at the center and used four stations as input parameters. The output from M1: ED is shown in Table 3 (b). Three stations are located near each other, but one station is located far away from its neighbors. We found significant inefficient spatial coverage in that case because almost all covered areas overlap. The COV of M1: ED is 57%, which is a 16% decrease in the current value, and WCD shows a value of 12.58, which increased 48% when compared with the current value. Next, consider the result in M2: FSS; we used the input parameters as same as M1: ED except adding a fixed diameter of ten kilometers to be used in the calculation. The result shows that most stations are located apart from the first station at the center, as shown in Table 3 (c). As a result, M2: FSS achieves the best coverage percentage up to 91% and an increase of 34% compared to the current value. On the other hand, the WCD value shows 7.44, which decreased by 13%.  Table 3 shows M1: ED, M2: FSS and compares them with the existing stations in Sejong city. In Table 3, triangle symbols depict existing stations, and circles represent the spatial coverage. Let us consider a result in M1: ED, where we defined the first station at the center and used four stations as input parameters. The output from M1: ED is shown in Table 3 (b). Three stations are located near each other, but one station is located far away from its neighbors. We found significant inefficient spatial coverage in that case because almost all covered areas overlap. The COV of M1: ED is 57%, which is a 16% decrease in the current value, and WCD shows a value of 12.58, which increased 48% when compared with the current value. Next, consider the result in M2: FSS; we used the input parameters as same as M1: ED except adding a fixed diameter of ten kilometers to be used in the calculation. The result shows that most stations are located apart from the first station at the center, as shown in Table 3 (c). As a result, M2: FSS achieves the best coverage percentage up to 91% and an increase of 34% compared to the current value. On the other hand, the WCD value shows 7.44, which decreased by 13%. The M1: ED shows large overlapping areas which make a strengthened area with neighboring stations. The area of overlap can make data more reliable in case of data verification between neighboring stations and also produce a network which is fault tolerant.  Table 3 shows M1: ED, M2: FSS and compares them with the existing stations in Sejong city. In Table 3, triangle symbols depict existing stations, and circles represent the spatial coverage. Let us consider a result in M1: ED, where we defined the first station at the center and used four stations as input parameters. The output from M1: ED is shown in Table 3 (b). Three stations are located near each other, but one station is located far away from its neighbors. We found significant inefficient spatial coverage in that case because almost all covered areas overlap. The COV of M1: ED is 57%, which is a 16% decrease in the current value, and WCD shows a value of 12.58, which increased 48% when compared with the current value. Next, consider the result in M2: FSS; we used the input parameters as same as M1: ED except adding a fixed diameter of ten kilometers to be used in the calculation. The result shows that most stations are located apart from the first station at the center, as shown in Table 3 (c). As a result, M2: FSS achieves the best coverage percentage up to 91% and an increase of 34% compared to the current value. On the other hand, the WCD value shows 7.44, which decreased by 13%. The M1: ED shows large overlapping areas which make a strengthened area with neighboring stations. The area of overlap can make data more reliable in case of data verification between neighboring stations and also produce a network which is fault tolerant.  Table 3 shows M1: ED, M2: FSS and compares them with the existing stations in Sejong city. In Table 3, triangle symbols depict existing stations, and circles represent the spatial coverage. Let us consider a result in M1: ED, where we defined the first station at the center and used four stations as input parameters. The output from M1: ED is shown in Table 3 (b). Three stations are located near each other, but one station is located far away from its neighbors. We found significant inefficient spatial coverage in that case because almost all covered areas overlap. The COV of M1: ED is 57%, which is a 16% decrease in the current value, and WCD shows a value of 12.58, which increased 48% when compared with the current value. Next, consider the result in M2: FSS; we used the input parameters as same as M1: ED except adding a fixed diameter of ten kilometers to be used in the calculation. The result shows that most stations are located apart from the first station at the center, as shown in Table 3 (c). As a result, M2: FSS achieves the best coverage percentage up to 91% and an increase of 34% compared to the current value. On the other hand, the WCD value shows 7.44, which decreased by 13%. The M1: ED shows large overlapping areas which make a strengthened area with neighboring stations. The area of overlap can make data more reliable in case of data verification between neighboring stations and also produce a network which is fault tolerant.
(c) The M1: ED shows large overlapping areas which make a strengthened area with neighboring stations. The area of overlap can make data more reliable in case of data verification between neighboring stations and also produce a network which is fault tolerant. We denominate such overlap areas as confidence areas because they can enhance data reliability. However, this method shows spatial coverage weaknesses. In contrast, M2: FSS shows achieving good spatial coverage which can cover an area up to 91% in the case of four stations in Sejong city. For the spatial coverage assessment, it is possible to conclude that M2: FSS is better than M1: ED.

Performance of Coverage Percentage Versus the Number of Stations Added Incrementally
Our previous result when establishing four stations using M2: FSS shows a high coverage percentage of about 91%, which is better than M1: ED. In this case, we compared the performance of M2: FSS and M3: ED + FSS in terms of coverage percentage versus the number of stations added incrementally. The specified number of stations is seven. In Figure 7, the plot graph shows a comparison coverage percentage (COV) as the number of stations increases. The dashed line with square symbols indicates M2: FSS that shows coverage percentage increasing sharply and degrading from stations numbered 4 to 7. In contrast, the dotted line with triangle symbols indicates M3: ED + FSS and offers a relatively stable increasing trend from stations numbered 1 to 7. We conclude that M3: ED + FSS shows a coverage increasing trend better than that of M2: FSS. We denominate such overlap areas as confidence areas because they can enhance dat reliability. However, this method shows spatial coverage weaknesses. In contrast, M2: FS shows achieving good spatial coverage which can cover an area up to 91% in the case o four stations in Sejong city. For the spatial coverage assessment, it is possible to conclud that M2: FSS is better than M1: ED.  Next, we investigated the additional stations of M2: FSS. As shown in Table 4 (a light color of triangle symbols indicated the three additional stations which are locate close to the border.  Next, we investigated the additional stations of M2: FSS. As shown in Table 4 (a), light color of triangle symbols indicated the three additional stations which are located close to the border.
Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.  Next, we investigated the additional stations of M2: FSS. As shown in Table 4 (a), light color of triangle symbols indicated the three additional stations which are located close to the border.  Next, we investigated the additional stations of M2: FSS. As shown in Table 4 (a), light color of triangle symbols indicated the three additional stations which are located close to the border. In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. The M4: ED + ASS shows good spatial coverage and balance of stations based on a specified number of stations. For this section, we compared our method's capability to apply to different sizes of cities and evaluated balancing and distribution of the stations in cities. We used M3: ED + FSS and M4: ED + ASS in Bangkok, Sejong, and Bonn. The cities in that list are in descending order by size. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. The M4: ED + ASS shows good spatial coverage and balance of stations based on a specified number of stations. For this section, we compared our method's capability to apply to different sizes of cities and evaluated balancing and distribution of the stations in cities. We used M3: ED + FSS and M4: ED + ASS in Bangkok, Sejong, and Bonn. The cities in that list are in descending order by size. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. The M4: ED + ASS shows good spatial coverage and balance of stations based on a specified number of stations. For this section, we compared our method's capability to apply to different sizes of cities and evaluated balancing and distribution of the stations in cities. We used M3: ED + FSS and M4: ED + ASS in Bangkok, Sejong, and Bonn. The cities in that list are in descending order by size. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. Although the M2: FSS achieves excellent spatial coverage (98%) nonetheless, the additional stations cause loss of area coverage, which shows the transparent area inside circles. On the other hand, all seven stations from the M3: ED + FSS method are located inside the city. These stations provide both sufficient spatial coverage (94%) and right overlapping area, as clearly shown in Table 4 (b). According to the result, we can assert a performance of M3: ED + FSS in achieving spatial coverage without loss of area coverage while still preserving the overlapped areas. The nearby stations can enhance strength to neighboring stations, which makes the network more robust and its data more reliable.

Coverage Percentage Versus a Specified Number of Stations
In this experimental case, we compared the performance of M3: ED + FSS and M4: ED + ASS concerning coverage percentage versus a specified number of stations. The location of station results from two methods is shown in Table 5. This part of our experiment aims to consider the flexibility of finding stations when we predefine the number of stations.
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations. The M4: ED + ASS shows good spatial coverage and balance of stations based on a specified number of stations. For this section, we compared our method's capability to apply to different sizes of cities and evaluated balancing and distribution of the stations in cities. We used M3: ED + FSS and M4: ED + ASS in Bangkok, Sejong, and Bonn. The cities in that list are in descending order by size.

COV: 99%
We will evaluate the coverage percentage and distribution of the stations in the study area. The results based on M3: ED + FSS show that the stations in similar aligned positions seem like a straight line in all four cases. On the other hand, results for M4: ED + ASS show a better balance of stations. The stations are readjusted whenever the number of stations changes. Furthermore, in four cases of a specified number of stations, the M4: ED + ASS has a higher coverage percentage and is better than M3: ED + FSS. Clearly, M4: ED + ASS is better than M3: ED + FSS to better cover the percentage and balance of stations.

Flexibility to Apply Our Methods for Different Cities
The M4: ED + ASS shows good spatial coverage and balance of stations based on a specified number of stations. For this section, we compared our method's capability to apply to different sizes of cities and evaluated balancing and distribution of the stations in cities. We used M3: ED + FSS and M4: ED + ASS in Bangkok, Sejong, and Bonn. The cities in that list are in descending order by size. Table 6 (bkk-1), (sj-1), and (bo-1) shows the existing stations and current coverage percentages. There are twelve stations in Bangkok, four stations in Sejong, and one station in Bonn. The result of M3: ED + FSS in Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.    Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-1), (sj-1), and (bo-1) shows the existing stations and current coverage percentages. There are twelve stations in Bangkok, four stations in Sejong, and one station in Bonn. The result of M3: ED + FSS in Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.   Table 6 (bkk-1), (sj-1), and (bo-1) shows the existing stations and current coverage percentages. There are twelve stations in Bangkok, four stations in Sejong, and one station in Bonn. The result of M3: ED + FSS in Table 6 (bkk-2) shows an imbalance of stations. On the other hand, the stations from M4: ED + ASS in Table 6 (bkk-3) are evenly distributed over the city. One benefit of balancing locations and overlapping areas is that the network can better support the future urban expansion and provide better air pollution monitoring. Although in Table 6, (sj-3) and (bo-3) cannot clearly show the balancing of stations when compared with Table 6 (sj-2) and (bo-2), the COV values of M4: ED + ASS are higher than M3: ED + FSS and show significantly increased rates of 56%, 22%, and 37%, in Bangkok, Sejong, and Bonn, respectively. These results allow us to conclude that our M4: ED + ASS is more flexible than M3: ED + FSS for designing new station networks in any city size.  For this section, we applied M4: ED + ASS to add one station into the network. Then, we evaluated the results with our proposed criteria: COV and WCD. In Table 7, triangle (bo-3) COV: 37% (+37%)

Assessment of Scenario 2: Finding Additional Stations to Improve the Current Network
For this section, we applied M4: ED + ASS to add one station into the network. Then, we evaluated the results with our proposed criteria: COV and WCD. In Table 7, triangle symbols indicate the existing stations, and star symbol indicates the additional station. The additional stations show the performance of our proposed method and evaluation criteria. They can improve coverage percentages by 39%, 34%, and 130% in Bangkok, Sejong, and Bonn.  COV: 89% (+39%) COV: 91% (+34%) COV: 62% (+ One of the key limitations of our study is the appropriate ratio of satisfaction versus confidence area. As shown in Figure 8, the dashed line pares coverage percentage and the number of stations. The 6th station coverage percentage of 90%. Increasing the number of stations from 7 u significantly increase coverage. Thus in future work, a change of strateg achieving the spatial coverage to increase the overlapped regions.

Conclusions
Increasing spatial coverage of AQMN gives effective environmen Especially in setting up a network in cities that lack historical pollution tions in cities with an existing network that face urbanization can make design adequate spatial coverage and effective AQMN. Our experiment the ability of the proposed method to provide maximum spatial coverage number of stations. The results showed that ED + ASS achieved effectiv hance spatial coverage and proper confidence area. In setting up a netw such a method suggested a balance of station locations and achieved m coverage. The stations are neither located close to the border, nor are th However, they are evenly distributed over the city. These results indicat planner considers economic benefits and the investment costs for constru the proposed method can give an excellent. In adding stations in cities, th  One of the key limitations of our study is the appropriate ratio of spatial co satisfaction versus confidence area. As shown in Figure 8, the dashed line plot grap pares coverage percentage and the number of stations. The 6th stations show the coverage percentage of 90%. Increasing the number of stations from 7 up to 12 do significantly increase coverage. Thus in future work, a change of strategy is neede achieving the spatial coverage to increase the overlapped regions.

Conclusions
Increasing spatial coverage of AQMN gives effective environmental manage Especially in setting up a network in cities that lack historical pollution data, addi tions in cities with an existing network that face urbanization can make it challeng design adequate spatial coverage and effective AQMN. Our experimental results p the ability of the proposed method to provide maximum spatial coverage based on a number of stations. The results showed that ED + ASS achieved effectiveness both hance spatial coverage and proper confidence area. In setting up a network in the such a method suggested a balance of station locations and achieved maximum coverage. The stations are neither located close to the border, nor are they close tog However, they are evenly distributed over the city. These results indicated that if t planner considers economic benefits and the investment costs for constructing a ne the proposed method can give an excellent. In adding stations in cities, the results sh  One of the key limitations of our study is the appropriate ratio of spatial coverage satisfaction versus confidence area. As shown in Figure 8, the dashed line plot graph compares coverage percentage and the number of stations. The 6th stations show the most coverage percentage of 90%. Increasing the number of stations from 7 up to 12 does not significantly increase coverage. Thus in future work, a change of strategy is needed after achieving the spatial coverage to increase the overlapped regions.

Conclusions
Increasing spatial coverage of AQMN gives effective environmental management. Especially in setting up a network in cities that lack historical pollution data, adding stations in cities with an existing network that face urbanization can make it challenging to design adequate spatial coverage and effective AQMN. Our experimental results proved the ability of the proposed method to provide maximum spatial coverage based on a given number of stations. The results showed that ED + ASS achieved effectiveness both in enhance spatial coverage and proper confidence area. In setting up a network in the cities, such a method suggested a balance of station locations and achieved maximum spatial coverage. The stations are neither located close to the border, nor are they close together. However, they are evenly distributed over the city. These results indicated that if the city planner considers economic benefits and the investment costs for constructing a network, the proposed method can give an excellent. In adding stations in cities, the results showed One of the key limitations of our study is the appropriate ratio of spatial coverage satisfaction versus confidence area. As shown in Figure 8, the dashed line plot graph compares coverage percentage and the number of stations. The 6th stations show the most coverage percentage of 90%. Increasing the number of stations from 7 up to 12 does not significantly increase coverage. Thus in future work, a change of strategy is needed after achieving the spatial coverage to increase the overlapped regions. criteria. They can improve coverage percentages by 39%, 34%, and 130% in Bangkok, Sejong, and Bonn. One of the key limitations of our study is the appropriate ratio of spatial coverage satisfaction versus confidence area. As shown in Figure 8, the dashed line plot graph compares coverage percentage and the number of stations. The 6th stations show the most coverage percentage of 90%. Increasing the number of stations from 7 up to 12 does not significantly increase coverage. Thus in future work, a change of strategy is needed after achieving the spatial coverage to increase the overlapped regions.

Conclusions
Increasing spatial coverage of AQMN gives effective environmental management. Especially in setting up a network in cities that lack historical pollution data, adding stations in cities with an existing network that face urbanization can make it challenging to design adequate spatial coverage and effective AQMN. Our experimental results proved the ability of the proposed method to provide maximum spatial coverage based on a given number of stations. The results showed that ED + ASS achieved effectiveness both in enhance spatial coverage and proper confidence area. In setting up a network in the cities, such a method suggested a balance of station locations and achieved maximum spatial coverage. The stations are neither located close to the border, nor are they close together. However, they are evenly distributed over the city. These results indicated that if the city planner considers economic benefits and the investment costs for constructing a network,

Conclusions
Increasing spatial coverage of AQMN gives effective environmental management. Especially in setting up a network in cities that lack historical pollution data, adding stations in cities with an existing network that face urbanization can make it challenging to design adequate spatial coverage and effective AQMN. Our experimental results proved the ability of the proposed method to provide maximum spatial coverage based on a given number of stations. The results showed that ED + ASS achieved effectiveness both in enhance spatial coverage and proper confidence area. In setting up a network in the cities, such a method suggested a balance of station locations and achieved maximum spatial coverage. The stations are neither located close to the border, nor are they close together. However, they are evenly distributed over the city. These results indicated that if the city planner considers economic benefits and the investment costs for constructing a network, the proposed method can give an excellent. In adding stations in cities, the results showed an optimal for the next station and enhanced the spatial coverage. Our proposed method showed effectiveness for expanding the existing networks as well as setting up AQMN.
In future work, we will consider the ratio of spatial coverage satisfaction versus confidence area as we mentioned in our experiment. We will modify our methods by changing strategies after achieving the spatial coverage to increase the overlapped regions of confidence areas. Moreover, we will integrate the historical pollution information, the city characteristics data, and land-use to find stations for better air pollution monitoring.