Artificial Intelligence Based Investigation for the Impact of High PM2.5 Concentration on Cloud Parameters over the Polluted Central IGP location, Kanpur

Abstract

This research focuses on using artificial neural network (ANN) models to assess daily surface PM2.5 concentrations by incorporating aerosol optical depth (AOD) and cloud parameters from the Moderate Resolution Imaging Spectroradiometer (MODIS), along with meteorological data, for the period from January 2017 to December 2021 over Kanpur. For this exercise, three ANN models were utilized: ANN1 (1 Layer, 14 Neurons), ANN2 (2 Layers, 14, 28 Neurons), and ANN3 (1 Layer, 14, 28, 14 Neurons). Statistical tests such as FAC2, MGE, NMB, MAPE, RMSE, R, and COE were conducted to validate the models. Initial results show that the ANN1 performed the best. The study also examined spatial and temporal changes to observe variations in PM2.5, AOD, and various cloud properties, including water vapor (WV), cloud effective radius (CER), cloud fraction (CF), cloud liquid water path (CLWP), cloud optical depth (COD), cloud top pressure (CTP), and cloud top temperature (CTT) on a seasonal and annual basis, as well as during high PM2.5 concentration conditions. During the study, the average daily PM2.5 was found to be approximately 100 ìg/m³ (ranging from 0.45 to 470.23 ìg/m³), while the average AOD was 0.79 (ranging from 0.09 to 3.55). High PM2.5 concentrations (three to five times higher than the NAAQS annual limit) significantly influenced crucial cloud microphysical properties. The research findings aid in estimating PM2.5 using satellite-retrieved AOD and meteorological data, providing insights into aerosol and cloud properties variability during high pollution events in the heavily polluted city of Kanpur, India.