Evaluation of Machine Learning Models for Estimating PM2.5 Concentrations across Malaysia

Abstract

Southeast Asia (SEA) is a hotspot region for atmospheric pollution and haze conditions, due to extensive forest, agricultural and peat fires. This study aims to estimate the PM2.5 concentrations across Malaysia using machine-learning (ML) models like Random Forest (RF) and Support Vector Regression (SVR), based on satellite AOD (aerosol optical depth) observations, ground measured air pollutants (NO2, SO2, CO, O3) and meteorological parameters (air temperature, relative humidity, wind speed and direction). The estimated PM2.5 concentrations for a two-year period (2018–2019) are evaluated against measurements performed at 65 air-quality monitoring stations located at urban, industrial, suburban and rural sites. PM2.5 concentrations varied widely between the stations, with higher values (mean of 24.2 ± 21.6 µg m−3) at urban/industrial stations and lower (mean of 21.3 ± 18.4 µg m−3) at suburban/rural sites. Furthermore, pronounced seasonal variability in PM2.5 is recorded across Malaysia, with highest concentrations during the dry season (June–September). Seven models were developed for PM2.5 predictions, i.e., separately for urban/industrial and suburban/rural sites, for the four dominant seasons (dry, wet and two inter-monsoon), and an overall model, which displayed accuracies in the order of R2 = 0.46–0.76. The validation analysis reveals that the RF model (R2 = 0.53–0.76) exhibits slightly better performance than SVR, except for the overall model. This is the first study conducted in Malaysia for PM2.5 estimations at a national scale combining satellite aerosol retrievals with ground-based pollutants, meteorological factors and ML techniques. The satisfactory prediction of PM2.5 concentrations across Malaysia allows a continuous monitoring of the pollution levels at remote areas with absence of measurement networks.

Estimating ground-level PM2.5 using subset regression model and machine learning algorithms in Asian megacity, Dhaka, Bangladesh

“…Zaman et al ( 2017 ) revealed that PM 2.5 concentration was 11.61 μg/m 3 in Malaysia, while in Delhi the value was 18.99 μg/m 3 as explored by Saraswat et al ( 2017 ). Also, Zaman et al ( 2021 ) found higher levels of PM 2.5 concentration across Malaysia during the dry season, which was quite consistent with the present research work. Abaset et al ( 2004 ) found that during the dry season, levels of PM 2.5 were 230.3 μg/m 3 for urban/industrial areas and 226.3 μg/m 3 for suburban zones, with a mean of 31.26 μg/m 3 and 26.38 μg/m 3 , respectively.…”

“…A similar outcome was also obtained by Christakos et al ( 2018 ); Zhang et al ( 2021 ); and Lim et al ( 2019 ). The findings of Zaman et al ( 2021 ) are also consistent with the current result, with a cross-validation R 2 value of 0.84, a MAE of 10.07 μg/m 3 , and a lower RMSE value (12.1 μg/m 3 ). The findings of Shahriar et al ( 2020 ) represent that the R 2 values range between 0.66 and 0.78 and RMSE values range between 14.47 and 15.30 μg/m 3 .…”

“…Yang et al ( 2018 ) explored the correlation between PM 2.5 concentrations and the landscape pattern index. Zaman et al ( 2021 ) stated that there is no direct association between PM 2.5 and AOD, while Sinha et al ( 2015 ) exhibited that they have a strong correlation. Zaman et al ( 2021 ) investigated that there were only gasses (SO 2 , NO 2 , CO, NO) influencing PM 2.5 , but there was no correlation with meteorological factors.…”

Machine Learning Regression Models to Predict Particulate Matter (PM2.5)

“…RF validation worked somewhat better than SVR for spatial models, with statistical indicators of R 2 = 0.76, RMSE = 11.47 g for urban/industrial sites and R 2 = 0.64, RMSE = 10.76 for suburban/rural sites. The goal of this work is to predict PM 2.5 concentrations in Malaysia utilizing (ML) models derived from satellite AOD (aerosol optical depth) data, floor harmful emissions and weather parameters [5]. SVR calibration performed marginally better than RF calibration for the whole model, with R 2 = 0.69 and RMSE = 10.62 versus observed PM 2.5 concentrations.…”

“…Despite the fact that countless studies on PM 2.5 have been conducted in various aspects and areas, PM 2. 5 has not yet been introduced into the data science sector.…”