Influence of Land Use and Meteorological Factors on PM_2.5 and PM₁₀ Concentrations in Bangkok, Thailand

Abstract

Particulate matter (PM) is regarded a major problem worldwide because of the harm it causes to human health. Concentrations of PM with particle diameter less than 2.5 µm (PM_2.5) and with particle diameter less than 10 µm (PM₁₀) are based on various emission sources as well as meteorological factors. In Bangkok, where the PM_2.5 and PM₁₀ monitoring stations are few, the ability to estimate concentrations at any location based on its environment will benefit healthcare policymakers. This research aimed to study the influence of land use, traffic load, and meteorological factors on the PM_2.5 and PM₁₀ concentrations in Bangkok using a land-use regression (LUR) approach. The backward stepwise selection method was applied to select the significant variables to be included in the resultant models. Results showed that the adjusted coefficient of determination of the PM_2.5 and PM₁₀ LUR models were 0.58 and 0.57, respectively, which are in the same range as reported in the previous studies. The meteorological variables included in both models were rainfall and air pressure; wind speed contributed to only the PM_2.5 LUR model. Further, the land-use types selected in the PM_2.5 LUR model were industrial and transportation areas. The PM₁₀ LUR model included residential, commercial, industrial, and agricultural areas. Traffic load was excluded from both models. The root mean squared error obtained by 10-fold cross validation was 9.77 and 16.95 for the PM_2.5 and PM₁₀ LUR models, respectively.

1. Introduction

Degradation of air quality poses a significant threat to both human health and the environment in Southeast Asia. Air pollution caused by fine and coarse particulate matter (with particle diameter less than 2.5 and 10 µm: PM_2.5 and PM₁₀, respectively) is an important issue severely affects human health and causes the risk of premature death. Exposure to air pollution is known to be associated with the risk of stroke, heart disease, chronic obstructive pulmonary disease, lung cancer, and acute respiratory infections [1,2]. The amount of particulate matter (PM) in the air has influenced hospital admissions for respiratory and cardiovascular diseases in Bangkok [3,4].

Air pollution impact assessment requires considerable data, including data on air quality, pollution sources, land use, demography, and meteorological data. Previous studies on air quality in Bangkok mainly used data from the monitoring stations of the Pollution Control Department (PCD) [3,4,5,6,7,8,9]. Although these stations provide high-quality, high-frequency air quality data, the spatial resolution is limited. Bangkok Metropolitan Authority has 50 air quality monitoring stations, but some may not be operational throughout the year due to stops for maintenance.

A land-use regression (LUR) model examines the spatial variation of pollutants within urban areas and the pollutant concentrations detected in the area [10,11,12,13,14,15]. The LUR model forecasts air pollutant concentrations in areas where monitoring stations are unavailable. These models use multiple linear regression analysis to explore the relationship between observed air pollutants and spatial factors such as traffic conditions, population, local pollution sources, land use, land cover, elevation, and distance of monitors to the sea [15,16,17,18]. Such models were developed for many cities. In previous studies on LUR models, the focus was on nitrogen dioxide [8,10,12,19,20,21,22,23,24], which is one of the major air pollutants originating from road traffic and fossil fuel combustion in metropolitan areas. However, LUR modeling for predicting PM_2.5 and PM₁₀ concentrations has become crucial as PM appears to be important air pollutants in many metropolitan areas, especially in Asia. China has accumulated a substantial body of research on the application of LUR models to investigate the spatial distribution of PM 2.5 concentrations using limited quantities of data from fixed monitoring stations in several cities such as Liaoning [15], Shanghai [16], Beijing [25], Xi’an [26], and Guangzhou [27].

A few studies used the LUR model to evaluate air pollution in Thailand. Cheewinsiriwat [8] evaluated the NO₂ concentration in Inner Bangkok using an LUR model and data from the 12 monitoring stations of the Pollution Control Department. The predictor variables included traffic volume, land use, road areas, and meteorological data. The LUR model demonstrated variation in NO₂. In addition, humidity, temperature, wind speed, rainfall, residential land use, and industrial land use influenced NO₂ in the study area. However, because of insufficient and inaccurate data, the traffic volume data were excluded from the multiple linear regression analysis processes. In another recent study, the LUR approach was employed by Chalermpong et al. [9] to investigate the relation between daily PM_2.5 concentration and other predictive factors using data from 2019. The findings revealed remarkable seasonal influences on PM_2.5 and considerable impacts of open biomass burning and meteorological variables. However, time-invariant factors such as traffic, transportation, and land use exhibited influences according to the LUR models. Further research based on more accurate traffic data is required to determine the association between traffic intensity and PM_2.5 concentrations. Although Chalermpong et al. [9] established a relationship between PM_2.5 concentrations and associated factors, there is still a lack of understanding regarding PM₁₀ concentrations, which may also harm human health. While Chalermpong et al. [9] used annual average daily traffic, this study used the daily traffic count derived from a model developed by Sukitpaneenit and Stettler [28].

The aim of this research was to analyze the relationship between the land use and meteorological factors and PM_2.5 and PM₁₀ concentrations in Bangkok based on an LUR approach. In this study, the backward method was used for model fitting and the leave-one cross validation (LOOCV) and 10-fold CV method were used for evaluating model predictions. The results of this study may contribute toward a better understanding of exposure to PM_2.5 and PM₁₀ in urban environments. The findings may allow us to identify locations where public health is at risk and address pollution issues in urban areas.

2. Materials and Methods

Meteorological, land use, and traffic load data were used in this study as the predictor variables to predict the response variables, namely, PM_2.5 and PM₁₀ concentrations. The predictors were statistically tested. If any of the predictors did not exhibit collinearity with each other and if they were significantly correlated to PM_2.5 or PM₁₀, they were selected as input variables for the LUR model. This model then again selected a number of the variables that yielded the best coefficient of determination R². Then, two equations—one each for PM_2.5 and PM₁₀—were derived from the models. The models were also validated using LOOCV techniques with 10-fold CV using the R program.

2.1. Dataset and Data Preparation

2.1.1. PM_2.5 and PM₁₀

The PM_2.5 and PM₁₀ concentrations and meteorological data in 2019 were measured at the monitoring stations. Bangkok Metropolitan Authority has 50 weather monitoring stations in Bangkok. In 2019, every station measured meteorological data such as rainfall, wind speed, temperature, relative humidity, and air pressure; some stations measured PM_2.5 concentrations, others measured PM₁₀ concentrations, and some measured both types of PM. The daily average concentrations of PM_2.5 (18 stations) and PM₁₀ (19 stations) in June and December 2019 were used in the model to represent the low-PM season in June and high-PM season in December (Figure 1). The PM_2.5 concentrations averaged daily per hour were 6–37 µg/m³ (average: 18 µg/m³) in June 2019, and 15–82 µg/m³ (average: 42 µg/m³) in December 2019. The PM₁₀ concentrations averaged daily per hour were 11–87 µg/m³ (average: 38 µg/m³), and 31–131 µg/m³ (average: 75 µg/m³) in December 2019. The concentrations in December were about twice those in June.

2.1.2. Meteorological Data

As the daily average rainfall, wind speed, temperature, relative humidity, and air pressure were measured at the same stations as the PM_2.5 and PM₁₀ data, the measured values could be used directly for LUR modeling.

Rainfall

In June 2019, rainfall occurred almost every day. The daily accumulated rainfall was mostly less than 40 mm, but the highest was about 100 mm. In contrast, there was only 1 rainy day, with very little rainfall, in December 2019.

Wind Speed

Wind speed was up to 2 m/s throughout June. In December, the speed reached up to 2 m/s only in the first week and reduced to 1 m/s in the rest of the month.

Temperature

In Bangkok, June and December correspond to the rainy and winter seasons, respectively. The temperatures in June and December 2019 were in the ranges of 27–34 °C and 21–31 °C, respectively. In fact, in the first week of December, the temperature was below 26 °C.

Relative Humidity

The relative humidity was about 60–90% in June 2019, and it decreased to 40–80% in December 2019. The maximum value in the first half of December was less than 65%.

Air Pressure

The air pressure was in the range of 1004–1010 hPa in June 2019, and it rose to 1011–1019 hPa in December 2019. The highest air pressure of 1015–1019 hPa in the study period was recorded in the first week of December.

2.1.3. Land Use

The 2019 land use data of Bangkok were provided by the Land Development Department, Ministry of Agriculture and Cooperatives. The land-use data were categorized into eight types: residential, commercial, industrial, government, transportation, open space, agriculture, and water bodies, as shown in Figure 2. The three most common land-use types in Bangkok are residential (31.60%), commercial (24.18%), and agricultural (23.68%).

The activities in the area surrounding monitoring stations affect the PM_2.5 and PM₁₀ concentrations. Hence, the areal sizes of all the land use types around the stations were calculated and used as inputs to the models. Buffer zones of radii 200, 400, 600, 800, and 1000 m from the monitoring stations were created. The land use areas intersecting each buffer zone were calculated and summarized by the land use type. The appropriate radial extent of each land use type from the stations may vary according to the different activities that affect PM_2.5 and PM₁₀ concentrations.

To acquire the appropriate radial extents of each land use type, correlations of the areas with 200, 400, 600, 800, and 1000 m radii with PM_2.5 and with PM₁₀ concentrations were tested. For each land-use type, the radius with the highest correlation, with a significance level of 0.05, was selected as the appropriate radius. The areal extent of the selected radius of each land-use type was input to the PM_2.5 and PM₁₀ LUR models.

2.1.4. Traffic Load

The traffic load was derived from a model developed by Sukitpaneenit and Stettler [28]. This model estimates the daily traffic count (vehicles/hour) of all vehicle types based on the relationship between traffic flow, traffic speed, and traffic density using data from the Global Positioning System receivers installed in 3000 taxis. The daily traffic count was translated into the traffic load of each road segment by multiplying the count with the length of each road segment (Equation (1)).

Traffic Load = Daily Traffic Count × Road Segment Length

(1)

Traffic load affects the PM_2.5 and PM₁₀ concentrations in the same way as land use. Therefore, the traffic load values within radii of 200, 400, 600, 800, and 1000 m from each station were tested to determine the appropriate radius with the highest correlation, with a significance level of 0.05, to the PM_2.5 and PM₁₀ concentrations. The traffic load values of the selected radius for PM_2.5 and PM₁₀ were input to the PM_2.5 and PM₁₀ LUR models, respectively.

2.2. LUR Model

As mentioned earlier, two LUR models were built, one each for PM_2.5 and PM₁₀. After selecting the predictor variables that were noncollinear with others and that were significant at the 0.05 level for PM_2.5 or PM₁₀ concentrations, the variables of each model were input into a multiple linear regression using the backward method in R program. Initially, all predictor variables were entered into a regression analysis; thereafter, one variable with the maximum p-value at a time was removed until all the variables remaining in the model were statistically significant. These remaining predictor variables and their coefficient values were used in the PM_2.5 and PM₁₀ LUR equations.

The two equations were then applied to estimate the PM_2.5 and PM₁₀ concentrations at any unmeasured location. The R² values from each model indicate the percentage of variation in PM concentrations that can be explained by the predictors in the model.

2.3. Model Validation

K-fold CV in the R program was used to validate the stability of the obtained LUR models. Since all data were used to create the models, the data were sliced into 10 portions; among them, 9 portions were used as a training set and the remaining portion as a testing or validating set. The process was repeated 10 times, iterating through the training and testing sets, in the so-called 10-fold CV method, to obtain the mean squared error (MSE) of the 10-fold CV. The root mean squared error (RMSE) was then computed to determine the error (+/− µg/m³) in the model-estimated PM_2.5 and PM₁₀ concentrations.

2.4. Estimated PM_2.5 and PM₁₀ Concentration Map

The equations derived from the PM_2.5 and PM₁₀ LUR models were applied to a 200 × 200 m grid covering Bangkok to reproduce estimated PM_2.5 and PM₁₀ concentration maps. Because the data for June and December 2019 were used in this research, the middle of each month (June 15 and December 15) was selected to generate the estimated PM_2.5 and PM₁₀ concentrations maps.

A vector grid of resolution 200 × 200 m was created to store the values of the variables in each LUR model. For meteorological data, the known values of the meteorological variables were measured only at the monitoring stations. Then, interpolation using the inverse distance weighted method was applied to acquire the values of grid cells at any unmeasured location. For land-use and traffic-load data, the areal extents of each required land-use type and the traffic load around the monitoring stations were obtained by buffering the grid cell centers with the selected radius and then calculating the area of each land-use type with the buffer zone.

When all the required values of the variables in the PM_2.5 and PM₁₀ LUR models were stored in each grid cell, the equation of each model was applied to estimate the PM_2.5 and PM₁₀ concentrations in Bangkok on 15 June and 15 December 2019.

3. Results

3.1. PM_2.5

3.1.1. Variables Selection

The correlations of all the five meteorological variables with PM_2.5 concentrations were statistically significant at the 0.01 level of significance (99% confidence; see

Table 1. Correlations between the meteorological variables and PM_2.5 concentrations.

	Rainfall	Wind Speed	Temperature	Humidity	Air Pressure
PM2.5 Concentration	−0.218 **	−0.295 **	−0.475 **	−0.557 **	0.711 **

** Correlation is significant at the 0.01 level (2-tailed).

). However, air pressure exhibited collinearly with temperature and relative humidity; therefore, only rainfall, wind speed, and air pressure were entered into the PM_2.5 LUR model. Air pressure showed a high positive correlation with PM_2.5 concentrations, whereas rainfall and wind speed showed low negative correlations with PM_2.5 concentrations.

Six among the eight land-use types were statistically significant at the 0.05 level (95% confidence) and were selected as inputs to the PM_2.5 LUR model. These six types were commercial, industrial, government, transportation, open space, and water bodies (

Table 2. Correlations between the land use types and PM_2.5 concentrations and the selected radius from the monitoring stations.

Land Use Type	Selected Radius from the Monitoring Stations	Correlation
Residential	N/A	Not significant at 0.05
Commercial	200	0.070 *
Industrial	400	0.096 **
Government	800	−0.078 *
Transportation	200	−0.065 *
Open Space	400	−0.093 **
Agriculture	N/A	Not significant at 0.05
Water Bodies	800	0.110 **

** Correlationis significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

). Residential and agricultural land-use types were excluded as they were not correlated with statistically significantly with PM_2.5 concentrations. The data in Table 2 show that the selected radius for commercial and road areas were 200 m, for industrial and open space were 400 m, and for government and water bodies were 800 m. Commercial, industrial, and water bodies showed positive correlations, whereas government, open space, and road areas showed negative correlations. However, all the selected land use types showed very low correlation with PM_2.5 concentrations.

Traffic load was not statistically significant correlated with the PM_2.5 concentrations at any radius (

Table 3. Correlations between traffic loads with different radii from the monitoring stations and PM_2.5 concentrations.

	Radius from the Monitoring Stations
	200	400	600	800	1000
PM2.5 Concentration	−0.049	−0.055	−0.016	−0.016	0.023

). Therefore, this variable was excluded from the PM_2.5 LUR model.

3.1.2. PM_2.5 LUR Model

After variable selection, the PM_2.5 LUR model consisted of three meteorological variables (rainfall, wind speed, and air pressure) and six land-use variables (commercial, industrial, governmental, road area, open space, green area, and water bodies). After applying backward stepwise regression in R program, only five variables were left in the PM_2.5 LUR model, as shown in Equation (2).

$$\begin{array}{l}{\mathrm{PM}}_{2.5}=-2949.0665-0.0948\mathrm{Rainfall}-13.4344\mathrm{WindSpeed}+\\ 2.9583\mathrm{AirPressure}+0.0002468\mathrm{Ind}400-0.0001721\mathrm{Trans}200\end{array}$$

(2)

The adjusted R² of the model was 0.58; in other words, the model could explain 58% of the variations in PM_2.5 concentration. Equation (2) indicates that rainfall, wind speed, and transportation area were negatively correlated with the PM_2.5 concentrations, whereas air pressure and industrial areas were positively correlated.

3.1.3. Validation

The results obtained by using 10-fold CV on the PM_2.5 LUR model in the R program showed that the MSE was 95.5146 (RMSE = 9.77). Therefore, the model estimated that PM_2.5 concentrations can have errors of about ±9.77 µg/m³.

3.2. PM₁₀

3.2.1. Variable Selection

Only four meteorological variables, namely, rainfall, temperature, relative humidity, and air pressure, were statistically significant at the 0.01 level (99% confidence) as shown in

Table 4. Correlations between the meteorological variables and PM₁₀ concentrations.

	Rainfall	Wind Speed	Temperature	Humidity	Air Pressure
PM10 Concentration	−0.221 **	−0.032	−0.469 **	−0.563 **	0.682 **

** Correlation is significant at the 0.01 level (2-tailed).

. Similar to the case of the PM_2.5 LUR model, air pressure exhibited collinearly with temperature and humidity; therefore, only rainfall and air pressure were entered into the PM₁₀ LUR model. Rainfall was negatively correlated with the PM₁₀ concentrations, whereas air pressure was positively correlated.

All land-use types were statistically correlated with the PM₁₀ concentrations at a significance level of 0.05 (95% confidence), see

Table 5. Correlations between the land-use types and PM₁₀ concentrations and selected radius from the monitoring stations.

Land Use Type	Selected Radius from the Monitoring Stations	Correlation
Residential	200	0.115 **
Commercial	200	−0.119 **
Industrial	600	0.163 **
Government	400	0.102 **
Transportation	200	0.073 *
Open Space	800	0.117 **
Agriculture	200	−0.158 **
Water Bodies	400	0.080 **

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

. Therefore, all of them were used to create the PM₁₀ LUR model. The residential, commercial, agricultural, and transportation areas were selected with a radius of 200 m from the monitoring stations. Government area and water bodies were selected with a radius of 400 m. Industrial area and open space were selected with radii of 600 and 800 m, respectively. Only commercial area and agriculture were negatively correlated, whereas the others were positively correlated with PM₁₀ concentrations.

Traffic load within a radius of 400 m from the monitoring stations was the only buffer distance for which correlation with PM₁₀ concentration was significant at the 0.05 level (

Table 6. Correlations between traffic loads within different radii from the monitoring stations and PM₁₀ concentrations.

	Radius from the Monitoring Stations
	200	400	600	800	1000
PM10 Concentration	−0.040	−0.065 *	0.053	−0.020	0.059

* Correlation is significant at the 0.05 level (2-tailed).

). Therefore, it was included in the PM₁₀ LUR model.

3.2.2. PM₁₀ LUR Model

The PM₁₀ LUR model was composed of two meteorological variables (rainfall and air pressure), eight land-use variables (residential, commercial, industrial, government, transportation, open space, agriculture, and water bodies), and traffic load. Then, backward stepwise regression analysis was used to fit a model (Equation (3)) with adjusted R² at 0.57. That is, this model could explain 57% of the variations in PM₁₀ concentration.

$$\begin{array}{l}{\mathrm{PM}}_{10}=-5050.5599-0.1309\mathrm{Rainfall}+5.0845\mathrm{AirPressure}-\\ 0.00022\mathrm{Res}200-0.0002932\mathrm{Com}200+0.0001149\mathrm{Ind}600-\\ 0.0005785\mathrm{Agric}200\end{array}$$

(3)

The final PM₁₀ LUR model had two meteorological variables and four land-use types. Government area, transportation area, open space, and water bodies were excluded from the model. The traffic load was also excluded. According to Equation (3), air pressure and industrial area were positively correlated with the PM₁₀ concentrations, while the rest were negatively correlated.

3.2.3. Validation

Through 10-fold CV on the PM₁₀ LUR model in R program, the MSE was obtained as 287.4693 (RMSE = 16.95). Therefore, the model-estimated PM₁₀ concentrations can have errors of about ±16.95 µg/m³.

3.3. Estimated PM_2.5 and PM₁₀ Concentration Maps

The equations of the PM_2.5 and PM₁₀ LUR models were applied to estimate PM_2.5 and PM₁₀ concentrations on 15 June and 15 December, 2019 for the entirety of Bangkok. The estimated PM maps are shown in Figure 3. The concentrations were classified according to the standard classification of the Pollution Control Department. The maps show that the areas with high levels of PM that may affect human health were sparsely distributed.

With regard to PM_2.5, most areas in Bangkok in June 2019 were classified as good and very good to human health, while in December 2019, most areas were classified as moderately harmful, which is harmful for vulnerable groups of people. There were more areas with high and very high levels of PM, which affect human health, in December than in June.

With regard to PM₁₀, similar to PM_2.5, most areas were very good for human health in June. In December, most of the areas were changed to good for human health or moderately harmful. Few spots had high and very high levels of PM₁₀ concentrations.

4. Discussion

Based on previous studies, the factors used in the LUR models were varied to suit the characteristics of each study area. These factors are commonly related to transportation (such as road area, traffic load, and road length), land use (such as residential, commercial, industrial, agriculture, number of populations, and number of households), topography (such as distance from sea and elevation), and meteorological factors (such as rainfall, wind speed, temperature, and humidity) [9,15,17,26,27,28,29,30]. In this study, similar factors available within the study area, including traffic load, area of each land-use type, and meteorological data, were used. Compared with the previous research, the models built in this study have R² values of 0.57 and 0.58 for PM_2.5 and PM₁₀ concentrations, respectively, indicating that the model can moderately explain the variations in the concentrations. The R² values for PM_2.5 of the previous research were in the ranges of 0.46 and 0.53 for the study areas of Bangkok [9] and Taichung, respectively, and 0.71 and 0.88 for the study areas in Xi-an and Shanghai, respectively [16,17,26,27].

The PM concentration levels are affected by seasons as the correlations between PM and meteorological factors were statistically higher than those between PM and land-use and traffic-load factors; this finding is similar to that of the study of Chalermpong et al. [9]. Although rainfall and air pressure influence both PM_2.5 and PM₁₀ concentrations, a wind speed of about 1–2 m/s in Bangkok significantly affected only PM_2.5 concentrations. The different influence of wind speed on PM_2.5 and PM₁₀ were also mentioned in Pengchan et al. [31], i.e., that stronger wind could blow fine PM far away and, simultaneously, re-suspend coarse PM from the ground up in the air. Moreover, Liu et al. [32] also reported the different effects of wind speed on the concentrations of PM_2.5 and PM_10. They revealed that wind speed greater than 2 m/s decreased the concentration of PM_2.5 and PM₁₀, and wind speed greater than 4 m/s increased the concentration of PM₁₀. As wind speed in Bangkok is lower than in the study area of Liu et al. [32], the finding from this study reveals that wind speed of 1–2 m/s caused PM_2.5 to increase but not strong enough on PM₁₀ to see a pattern or relationship.

This study did not include wind direction in the analysis, although some previous studies mentioned that long-range transport affect the concentrations of PM along the wind pathway [33,34,35]. Li et al. [34] introduced the semicircular-buffer-based LUR model to incorporate wind direction in the LUR model and reported that including wind direction in their LUR model yielded better results than traditional LUR. Furthermore, Wimolwattanapun et al. [35] found that wind speed and direction helped in identifying the PM sources and each source was influenced by the wind in the same direction at their study sites in Bangkok and Pathumthani. Therefore, in our further study, wind direction should be included in the LUR modeling to cover the long-range transport issue.

Shi et al. [15] stated that main road length and distance to road affected PM concentrations; however, in this study, a road area within 200 m from the measurement point affected only PM_2.5 concentrations and showed a negative relationship with the PM. Traffic load was also excluded from the models, suggesting that transportation-related factors may not be the main contribution to the PM concentrations in Bangkok. Likewise, Han et al. [26] also mentioned that the emission from traffic had less contribution to PM concentrations than the contribution from industries.

Land-use types are expected to influence the PM concentrations as it is known that some land-use types such as industrial areas are a source of pollution while some others such as agricultural area are not. This study revealed that the correlations between each land-use type and the PM concentrations were quite low. In addition, only some land-use types were included in the models. Industrial areas within 400–600 m comprised the main land-use type that contributed to both PM_2.5 and PM₁₀ concentrations. In contrast, residential, commercial, and agriculture areas within 200 m affected only the PM₁₀ concentrations.

The RMSE values obtained by CV in this study are comparable to those reported by Han et al. [26]. In this study, the RMSEs were ±9.77 for the PM_2.5 model and ±16.95 for the PM₁₀ model, while those reported by Han et al. [26] were ±8.35 for the PM_2.5 model and ±14.84 for the PM₁₀ model.

The reproduced maps of the estimated PM_2.5 and PM₁₀ concentrations in Bangkok showed that the model can help identify locations that require measures to safeguard human health even in the absence of monitoring stations there. The maps also revealed that Bangkok was more affected by PM_2.5 than by PM₁₀ as in December, most areas were regarded to have moderately harmful PM_2.5 levels but the PM₁₀ level was considered to be within the limit of good for human health.

5. Conclusions

To summarize, an LUR approach was adopted in this study to estimate the PM_2.5 and PM₁₀ concentrations in Bangkok. Researches conducted across the world based on this approach have revealed that the models comprise different factors and different coefficients of such factors depending on the study areas. This study revealed that meteorological factors are strongly related to PM concentrations, and that the industrial area is a source of PM_2.5 and PM₁₀ pollutants. The models yielded R² values of about 0.57–0.58, which is within the range reported in previous research. In further study, using daily PM and meteorological data of the whole year can help improve the models as the present results indicated that meteorological data play an important role in determining the PM concentrations in Bangkok.