Comparison of Differently Parameterized SARIMA Models using CERES-Derived Aerosol Optical Depth over Indo-Gangetic Plain

The Indo-Gangetic Plain (IGP) is a globally recognized hotspot for high aerosol loading, necessitating precise modelling to understand its spatial and temporal dynamics. This study evaluates the performance of differently parameterized Seasonal Autoregressive Integrated Moving Average (SARIMA) models in forecasting the Aerosol Optical Depth (AOD) at 550 nm retrieved from the CERES (Clouds and the Earth's Radiant Energy System) satellite platform across eight  locations: Delhi, Dhaka, Jaipur, Kanpur, Karachi, Kolkata, Lahore, and Varanasi in the IGP. Using long-term AOD datasets from CERES during the period of 2005 to 2020, we tested various SARIMA configurations to capture seasonal trends and irregular variations specific to urban environments. The SARIMA configurations tested include configure_1: (1,0,1)(1,0,1)₁₂, configure_2: (1,1,1)(1,1,1)₁₂, configure_3: (2,0,1)(2,0,1)₁₂, and configure_4: (2,1,1)(2,1,1)₁₂ These configure models were compared with CERES-derived observations for AOD at the study sites for the next two years, that is, Jan, 2021 to Dec, 2022. Each configuration was assessed for data stationarity using the Augmented Dickey-Fuller (ADF) test and if not follows, then the differentiation method has been used to stationaries the series. The Model performance was evaluated using multiple statistical metrics, including normalized Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Root Mean Squared Error (RMSE), Mean Bias Error (MBE), and Mean Absolute Percentage Error (MAPE) for every configuration showed the low metric values. The result indicates high correlation coefficients, ranging from 0.54 to 0.91, and R-squared values, varying between 0.31 and 0.81 for all configurations that significantly determined the best-suited models for each location. Every modelled configuration has been checked with 95% and 99% confidence interval (with alpha=0.05 and 0.01, respectively) showing the p-value <0.001. These results emphasize the models' ability to replicate observed AOD patterns effectively. It reveal that parameter sensitivity plays a critical role in predictive accuracy, with optimal configurations varying across locations due to heterogeneity in aerosol sources and meteorological conditions. The present study underlines the importance of site-specific model tuning for reliable aerosol forecasting in densely populated and pollution-prone regions. These insights provide a foundation to enhance air quality prediction studies and address health, and climate impacts associated with aerosols in the IGP.