1. Introduction
Ambient air pollution has been related to increased levels of mortality and morbidity in megacities [
1]. It is closely related to people’s daily lives. The relatively high growth of socio-economic, industrial, and urbanization activities in urban and suburban areas has great potential for increasing energy consumption, which is one of the factors of air pollution. There are several types of outdoor air pollution, such as Cox, NOx, Sox, Ox, and one of the pollutants that is quite dangerous for human health is PM
2.5. PM
2.5, usually called fine particulate matter, is an air pollutant consisting of a mixture of solid and liquid particles. PM
2.5 has a diameter of less than 2.5
, and thus, it is often referred to as smooth PM, which also consists of ultrafine particles with a diameter of less than 0.1
[
2]. Epidemiological research associated with the study of PM
2.5 exposure to the incidence of lung cancer, respiratory problems, and early death has shown a positive association [
3,
4]. The results show that it harms human health.
Several studies have shown that the concentrations of PM (particulate matter) can be affected by geographical features such as land use information, meteorology, and satellite data, including temporal variation parameters [
5]. Methods for evaluating local urban variability to fine particulate matter were needed for these studies. Various models have been established around the world to explore the statistical correlations between ground monitoring stations of PM
2.5 and the variables derived from geographical data information due to develops in the performance of GIS (Geographical Information System) technology.
Research is being conducted on the impact of geographical parameters on improvement air quality has been carried out. However, it is a big challenge to access PM2.5 data, particularly in developing countries. During the last period, some methods have been established to tackle challenges related to air pollution, such as interpolation using kriging and IDW (Inverse Distance Weighting) [
6,
7], and LUR (Land Use Regression) [
8,
9,
10]. The interpolation of pollutant concentrations is based on the monitoring sites with densely clustered stations, whereas it is difficult to monitor locations with few stations. LUR models have proven to be relevant for these approaches in recent years. The method is to develop statistical regression models based on GIS platforms. These can be used to estimate air pollutant levels in a particular site by establishing a statistical correlation between pollutant observations and potential prediction variables [
11,
12,
13].
In major cities, the types of land use can affect the level of PM through urban area development [
14]. The land changes of forest, grasslands, agriculture in residential areas, industrial sites, and commercial centers frequently lead to increased emission levels. Air pollution concentration is related to changes in meteorological conditions. The development of prediction method starts by analyzing periods of serious atmospheric pollution, which are correlated with meteorological conditions monitored during those periods. These elements are considered as predictors [
15]. Meteorological conditions such as wind speed, wind direction, and rainfall have an impact on particulate matter exposure [
16]. Furthermore, satellite data using remote sensing, such as identification of greenness information, are efficient for measuring a wide range with multitemporal variations and easily available. The Normalized Difference Vegetation Index (NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) sensors have been commonly used in greenness inventory management, which is a satellite-based method that has established a strong dynamic range and responsiveness for recording and calculating spatial-temporal variations in vegetation density [
17,
18]. NDVI has been used to determine green space area, which has a negative correlation with particulate matter [
19].
Indonesia, with an area 1,922,570 km² (land only), has five ground stations monitoring PM
2.5 concentration, one of the station locations is in DKI Jakarta [
20]. In contrast, Taiwan, with an area of 36,000 km², has 78 ground stations to monitor air quality concentrations. For controlling and preventing air pollution from exceeding air quality standards, monitoring mass concentrations and multi-element identification of PM
2.5 can be used for particle characterization and estimation of pollutant sources. In general, air quality monitoring, especially for PM
2.5 is performed in various major cities in each country.
This study compares the air quality DKI Jakarta, in Indonesia and the Taipei Metropolis, in Taiwan using LUR model development for land use, meteorology, and greenness related to PM2.5. All the parameters have multi-temporal variability in order to achieve better performance in the estimation of PM2.5. The result acquired from this research will be particularly useful when developing LUR models in each country, in epidemiological studies and environmental health research in each country.
4. Discussion
This study implemented land use, meteorology and greenness data and long-term PM
2.5 monitoring data, from DKI Jakarta in Indonesia and Taipei Metropolis in Taiwan. The study covered the capital cities of the two countries. The variables were concerned with multi-temporal variability data. The LUR model has been suitable method for estimating the concentration of pollutants in several areas, especially for PM
2.5 [
9,
11,
12]. The model shows good predictive ability (R
2 = 56% for DKI Jakarta, and 84% for Taipei Metropolis), with spatial resolution of 50 × 50 meters. LUR can be used to determine the relations between PM
2.5 and various variables. The results were derived from the proposed two model development approaches. The model can be validated using 10-fold cross-validation which shows the model robustness. It can also be shown by the R
2 value of cross validation (R
2 = 56% for DKI Jakarta, and 84% for Taipei Metropolis). The results spatial-temporal maps can be obtained using the LUR model in each country.
A limitation of our study was the lack of traffic-related and satellite data (such as, AOD (Aerosol Optical Depth) from MODIS sensor) to determine the level of PM2.5 concentrations. Land use information in the two countries is based on different information standards, which could affect the number of variables used in selecting correlations. However, AOD (MCD19A2 products) from the MODIS sensor did not show pixel values, due to the cloud density/null value in the study area. It was difficult to obtain more specific GIS data for these sources.
However, this study showed that DKI Jakarta has a lower R2 value rather than Taiwan. This could be because to the total numbers of monitoring stations are not balanced between these regions. The small number of monitoring stations in DKI Jakarta could affect LUR model development due to the number of predictors factored, which might have influenced the performance of LUR model establishment. The varying distribution of station monitoring networks across countries could contribute to variations in PM2.5 level.
The proportion of residential areas, major roads, railways, airports, and quarrying sites has important effects across land use characteristics in the model of PM2.5. However, according to the land use data, residential areas showed a significant positive correlation with PM
2.5 concentration in DKI Jakarta. Studies of PM, the correlation have demonstrated high correlation between PM
2.5 level and residential areas, especially in high-density residential areas [
29,
30].
Several studies have demonstrated that meteorological factor (such as humidity, temperature, wind speed and wind direction) highly correlate with air pollution, espacially PM
2.5 concentrations [
31,
32,
33]. Statistical modelling of PM
2.5 in previous studies consistently showed higher concentrations in spring periods compared to fall periods [
34]. Coefficient estimation from the LUR model showed that PM
2.5 has a positive correlation with spring periods and a negative correlation with fall periods.
NDVI was obtained from satellite data using MODIS sensors to calculate greenness in the research area. However, NDVI was not differentiated among greenness types (such as planted area, green spaces, public parks, etc.). The difference of greenness type might be relevant to air pollutants. Previous studies have shown that greenness can affect air pollution as well [
35].The annual averages of PM
2.5 in Taipei Metropolis and DKI Jakarta exceed the national policy threshold standard in Indonesia and Taiwan, which is 15
. Meanwhile, the concentration of PM
2.5 both countries also exceed the threshold standard of the WHO. Accurate and frequent ground monitoring stations of air quality concentrations are necessary for the capital cities in each country. They can be used to assess air quality, identify the most relevant potential sources, strengthen management to control air quality, and provide advice to policymakers, especially to the governments. By a adopting remote sensing and GIS techniques, a comprehensive strategy can be used to develop an air quality monitoring network, which allows for controlling air quality for the protection of human health [
36].
5. Conclusions
This study analyzed recent trends of PM2.5 concentrations from 2016–2018 in two countries in Asia. LUR models were used to predict of PM2.5 level. The assessment of PM2.5 based on spatial-temporal was strongly influenced by land use, meteorological conditions, and MODIS NDVI. PM2.5 pollution in the Taipei Metropolis region, has a positive correlation with PM10, SO2, NO2, spring conditions, main roads, railways, airports, mining areas, and has a negative correlation with UV, rainfall, airports within close distances. PM2.5 pollution in DKI Jakarta, was strongly influenced by humidity, NDVI, temperature and residential areas. PM2.5 pollution in DKI Jakarta, has a positive correlation with residential areas, temperatures and has a negative correlation with NDVI, humidity. The R2 values of the resulting model were 0.84 and 0.56 for Taipei Metropolis and DKI Jakarta, respectively. Meanwhile, the 10-cross validation result shows R2 values of 0.83 and 0.61 for Taipei Metropolis and DKI Jakarta, respectively.