Estimation of Ground-level Fine Particulate Matter (PM2.5) from Artificial Neural Networks Using Meteorological Parameters Over Major Capital Cities at Indo-Gangetic Plain (IGP)

Environmental Engineering

Fine particulate matter (PM2.5) has received a widespread attention all over the world because of its direct association with the degradation of air quality and the related human health effects. The regular monitoring of PM2.5 concentrations is critically necessary to analyze their aftermath effects but the limited number of ground based air quality monitoring stations acts as a barrier in the evaluation of space-time dynamics of air pollution. This study is conducted to estimate the ground-level concentrations of fine particulate matter (PM2.5) from statistical models developed using algorithms of artificial neural networks incorporating satellite measurements. A total of 12 ANN models were developed for three major cities of Indo-Gangetic Plain viz., Delhi, Lucknow and Patna using Moderate Resolution Imaging Spectroradiometer (MODIS) and INSAT-3D aerosol optical depth (AOD) to estimate the PM2.5 concentrations from Jan-Dec 2019. The ANN models were trained using the AOD, meteorological parameters and PM2.5 dataset of the previous two years (2017-2018) subdivided as train, validation and test groups in the ratio of 0.7, 0.15 and 0.15. High correlation values were obtained during the training of ANN models (R ≥ 0.7 for each group). The well trained models provided good correlation coefficients obtained from the regression analysis of estimated and observed values accompanied with low RMSE and absolute percentage error. The ANN model developed using MODIS Aqua AOD for Patna showed highest correlation value of 0.85 followed by Delhi (R = 0.70) and Lucknow (R = 0.66). Although the degree of correlation varies for different sites, this study demonstrates the potential of ANN in air quality monitoring.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.