Air quality analysis and modelling of particulate matter (PM2.5 and PM10) of Ghaziabad city in India using Artificial Intelligence techniques

Abstract

Air pollution affects 91% of the global population, causing approximately 4.2 million deaths annually, according to the World Health Organization. This study presents a comprehensive analysis of spatiotemporal air quality patterns in Ghaziabad, focusing on seasonal variations, aerosol characteristics, correlation analysis, machine learning-based modelling, sensitivity analysis, and short-term prediction of PM${}_{\mathrm{2.5}}$ and PM₁₀ concentrations using data from four monitoring stations (MS1, MS2, MS3, MS4). Alarming levels of PM₁₀ and PM${}_{\mathrm{2.5}}$, frequently exceeding permissible standards, were observed, particularly at MS2, where industrial activities led to an 81.29% exceedance rate for PM₁₀ with a maximum concentration increase of 447.23%. PM${}_{\mathrm{2.5}}$ concentrations at MS2 reached $360.93\mu g$/m³, representing a 501.55% increase. Meteorological circumstances, particularly during winter, significantly increased pollution levels. SO₂ and ozone concentrations adhered to CPCB (Central Pollution Control Board) guidelines; nonetheless, winter months experienced a significant increase in overall pollutant levels. Positive correlations were identified between PM${}_{\mathrm{2.5}}$ and PM₁₀ with NO₂ (r $=$ 0.54, r $=$ 0.51), CO (r $=$ 0.51, r $=$ 0.45), and SO₂ (r $=$ 0.18, r $=$ 0.34), while negative correlations were noted with ozone (r $=$ −0.02, r $=$ −0.18), wind speed (r $=$ −0.17, r $=$ −0.20), and relative humidity (r $=$ −0.08, r $=$ −0.37). Solar radiation also showed a negative correlation (r $=$ −0.32, r $=$ −0.13). The study optimized predictive models for air quality forecasting using historical data. The XGBoost model outperformed others in predicting PM${}_{\mathrm{2.5}}$ and PM₁₀ concentrations, achieving the lowest Mean Absolute Error (MAE) and highest R² values (PM${}_{\mathrm{2.5}}$: MAE $13.24\mu g$/m³, R² 0.8960 and PM₁₀: MAE $27.46\mu g$/m³, R² 0.8397). Sensitivity analysis identified PM₁₀ concentration as the most influential predictor of PM${}_{\mathrm{2.5}}$ levels, contributing approximately 63.56% to the model’s predictive power, followed by solar radiation (9.74%) and relative humidity (8.30%). The model accurately forecasted air quality for 2023, demonstrating high reliability (PM${}_{\mathrm{2.5}}$ for 2023: MAE $14.64\mu g$/m³, R² 0.8850, and PM₁₀ for 2023: MAE $27.66\mu g$/m³, R² 0.8234). These robust short-term forecasts are essential for public health planning and environmental management, enabling proactive measures to mitigate pollution levels and safeguard public health. Reliable predictions facilitate targeted actions, supporting policy decisions to reduce air pollution and its adverse effects on the population.