Meteorological variability and predictive forecasting of atmospheric particulate pollution

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Airborne particulate matter (PM) poses documented health risks and influences climate change.
• Understanding and forecasting PM variability is essential for mitigation and adaptation strategies.
• Atmospheric PM10 concentrations are influenced by meteorological conditions and contribute to global warming.

Purpose of the Study:

• To develop and validate predictive models for atmospheric PM10 concentrations in Brunei-Muara.
• To assess the impact of meteorological parameters and historical PM10 data on forecasting accuracy.
• To compare the performance of different machine learning models in predicting PM10.

Main Methods:

• Development and validation of PM10 predictive forecasting models using Multiple Linear Regression (MLR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN).
• Inclusion of time and meteorological parameters, alongside the previous day's PM10 concentration (PM10,t-1), as input variables.
• Evaluation of model performance using metrics such as Root Mean Square Error (RMSE) and R-squared (R²).

Main Results:

• Incorporating the previous day's PM10 concentration (PM10,t-1) significantly improved model predictive power by 57-92%.
• The MLR model with PM10,t-1 demonstrated the highest capability in capturing seasonal variability (RMSE = 1.549 μg/m³; R² = 0.984).
• RF and ANN models with PM10,t-1 provided accurate forecasts for the next 1, 2, and 3 days, respectively.

Conclusions:

• Predictive models incorporating meteorological data and historical PM10 concentrations can accurately forecast atmospheric PM10 levels.
• The inclusion of the previous day's PM10 concentration is a key factor in enhancing forecasting accuracy.
• Different modeling approaches (MLR, RF, ANN) offer varying strengths for short-term and medium-term PM10 prediction.