Insights into airborne particulate matter: artificial intelligence-driven PM2.5 modelling in Hyderabad district, India

Abstract

Air pollution is one of the grave concerns of the modern era, claiming millions of lives and adversely impacting the economy. The primary objective of this study was to develop advanced forecast models for PM_2.5 levels in the Hyderabad district of India using artificial intelligence techniques. This study presents a novel approach to PM_2.5 modelling, leveraging the power of artificial intelligence (AI) and data-driven insights for Hyderabad District. Factor analysis was performed to check for correlations of PM_2.5 and aerosol optical depth (AOD) with various meteorological and pollutant variables, based on which it was observed that except temperature and solar radiation, all the variables showed considerable correlations with aerosols. The hybrid deep learning-based CNN – LSTM model was identified as the best-fit model for predicting PM_2.5 with an R² = 0.88, MSE = 68.93 (µg/m³)², RMSE = 8.30 µg/m³, and MAE = 6.45 µg/m³ as against the MLP – ARIMA and MLP models. A study on feature importance showed that AOD is a significant contributor to PM_2.5 prediction with a factor importance of 6.8%, ranking second only to meteorological factors. Wind direction and relative humidity exhibited factor importance values of 10.94% and 8.69%, respectively. The AI-driven PM_2.5 modelling approach offers a more comprehensive understanding of pollution patterns and their relationship with meteorological conditions and geographical characteristics. These results highlight the strong predictive power of the CNN – LSTM model and the significant influence of AOD and meteorological factors on PM_2.5 levels. These insights can inform policymakers, urban planners, and environmental agencies in formulating effective pollution control strategies and mitigation measures, leading to improved air quality and public health in the Hyderabad district and beyond.