Urban PM2.5 Concentration Estimation Based on Remote Sensing and Land Use Regression Model: Evidence from the Bangkok Metropolitan Region

Abstract

Background and Objectives: Fine particulate matter (PM_2.5) pollution has been a persistent environmental issue in Thailand for over two decades. In 2023, Thailand ranked fifth in Southeast Asia among countries experiencing the worst air quality for over 30 consecutive days. Bangkok and its metropolitan area, as the country’s economic, political, social, and cultural hub, have undergone rapid urban expansion, intensifying cumulative pollution and continuously affecting residents’ quality of life. In 2022, only three months recorded good air quality. The highest PM_2.5 concentrations occurred during March and April, particularly in roadside areas, where levels began to pose adverse health impacts and showed an increasing trend. However, the limited number of air quality monitoring stations restricts comprehensive surveillance and accurate assessment of PM_2.5 concentrations. Aerosol Optical Depth (AOD) data obtained from the Himawari-8 satellite are widely recognized as a valuable alternative for estimating PM_2.5 levels in areas without ground-based monitoring due to their broad spatial coverage. Meteorological conditions and urban factors also influence PM_2.5 concentrations. Land Use Regression (LUR), extensively applied in previous studies, has demonstrated strong capability in explaining PM_2.5 factors. Therefore, this study aims to examine the relationships among PM_2.5 concentrations, AOD data, meteorological factors, and urban factors in the Bangkok Metropolitan Region. The LUR model is developed to estimate PM_2.5 levels in unmonitored areas and to generate monthly concentration maps covering the entire study area.

Methodology: Data were collected from 73 air quality monitoring stations operated by the Pollution Control Department and the Bangkok Metropolitan Administration to develop a PM_2.5 database. Meteorological factors—including wind speed, temperature, relative humidity, and air pressure—were compiled. In addition, Aerosol Optical Depth (AOD) data obtained from Himawari-8 in NetCDF format, with a spatial resolution of 5 × 5 square kilometers, were extracted at station locations. Subsequently, all datasets were organized into daily and monthly records. Urban factors comprised traffic speed, population density, and land use. For spatial analysis, buffer zones with radius of 200, 400, 600, 800, and 1,000 meters were mapped around each station. The buffer distance showing the highest correlation with PM_2.5 for each factor was then selected for model development. To investigate factor associations, Pearson’s correlation coefficient at a 0.01 significance level was applied to identify variables significantly associated with PM_2.5. These selected predictors were then incorporated into a Land Use Regression (LUR) model using stepwise selection to construct a multiple linear regression equation for estimating PM_2.5 concentrations. Model performance was subsequently evaluated using R² and Root Mean Square Error (RMSE), together with 10-fold and leave-one-out cross-validation (LOOCV). Finally, the resulting regression equation was used to produce monthly PM_2.5 maps for the study area.

Main Results: The results of PM_2.5 concentration estimation in the Bangkok Metropolitan Region area during 2019–2023 reveal pronounced spatial and temporal variability, with peak concentrations consistently occurring in the winter season each year. Analysis of the relationships between PM_2.5 and meteorological factors indicates that air pressure and relative humidity were the most influential factors. Relative humidity exhibited a significant negative correlation with PM_2.5 concentrations, whereas air pressure showed a significant positive correlation. Regarding urban factors, several land use categories were significantly associated with PM_2.5, reflecting the role of human activities in shaping pollution distribution patterns. Traffic speed and population density demonstrated weak negative correlations; Aerosol Optical Depth (AOD) data derived from Himawari-8 showed positive correlations with PM_2.5 at all monitoring stations. Although the strength of these correlations was moderate, their temporal consistency highlights the potential of AOD as a surrogate for ground-based monitoring data in areas with limited station coverage. The Land Use Regression (LUR) model yielded an R² value of 0.339 for the combined five-year dataset, with a Root Mean Square Error (RMSE) of 9.971. The highest R² value (0.456) was observed in 2020, demonstrating the model’s capacity to explain PM_2.5 variability in a complex urban environment. The model outputs were further applied to develop monthly PM_2.5 maps, providing a foundational dataset for area-based air quality management planning.

Conclusions: The research findings indicate that PM_2.5 concentrations exhibit pronounced seasonal variability, particularly during the winter period. Correlation analysis revealed that meteorological factors are significant determinants, exerting statistically significant influences on PM_2.5 levels. In contrast, urban factors play a comparatively lesser role in explaining pollution variability. Aerosol Optical Depth (AOD) data derived from the Himawari-8 demonstrated consistent correlations with PM_2.5 concentrations across all monitoring stations over multiple time periods. The Land Use Regression (LUR) model achieved a moderate level of predictive performance and was capable of generating monthly PM_2.5 concentration maps. These outputs can effectively support localized air quality monitoring and management strategies.