Modelling and forecasting of CERES-retrieved ultraviolet radiation and AOD using a seasonal-ARIMA model in urban regions of Indo-Gangetic Plain

Abstract

The Indo-Gangetic Plain (IGP), a densely populated and agriculturally productive region, experiences substantial emissions from industrial activities, vehicular traffic, and biomass burning. These emissions contribute to high aerosol loading and significant variability affecting the ultraviolet (UV) radiation, impacting air quality, climate, and public health. The present study employs Box-Jenkins, Seasonal Autoregressive Integrated Moving Average (SARIMA) model to analyze and forecast the ultraviolet radiation (UVA and UVB) and Aerosol Optical Depth (AOD) using data from the Clouds and the Earth's Radiant Energy System (CERES) over selected urban regions in the IGP. Time-series data from January 2005 to December 2020 (∼16 years) are utilized to capture seasonal patterns and long-term trends, enabling predictions for the next two years. Stationarity of UVA, UVB, and AOD data is tested using the Augmented Dickey-Fuller (ADF) test, followed by autocorrelation function (ACF) and partial autocorrelation function (PACF) analysis to determine the optimal model order. Model is fitted by estimating various parameters through maximum likelihood estimation (MLE) optimization techniques and found suitable with (1,1,1)(1,1,1)₁₂, considering a monthly seasonal factor for predicting UVA, UVB fluxes, and AOD. When SARIMA was used with exogenous variables for prediction of UVA, UVB and AOD it was found that different stations had different optimal configurations. The model validation involves error and comparative analyses. Error metrics, including mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE) in the predicted data, indicate low errors. Comparative analysis shows strong agreement between modelled and CERES data, with significant correlation metrics such as the R-squared coefficient, Pearson correlation coefficient (r), degree of agreement (DOA), and Taylor skill score (TSS). Model accuracy is further assessed using normalized Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), while model adequacy is evaluated through Ljung-Box and Jarque-Bera tests. Results indicate a strong seasonal dependence of UV radiation and AOD, with peaks during pre-monsoon and winter months, respectively. The SARIMA model effectively captures temporal variability and provides very good forecast for time series data valuable to researchers in air quality management and climate mitigation.