Forecasting air pollutant concentrations in Colombo, Sri Lanka: A time series analysis of major air pollutant parameters

Abstract

Sri Lanka’s commercial capital is Colombo and virtually all significant lines of financial and business activities take place in Colombo or within its periphery. However, the daily developing city of Colombo is incurring the harm of atmospheric pollution, which is one of the most dangerous disasters due to urbanization. As regards the issues of air pollution, people have diseases that affect their respiratory tracts. According to the World Health Organization, concerns have indicated that air pollution is considered one of the most lethal pollutants globally. The poisonous air particles could cause human deaths. Hence, the quantity of pollutants in the air and the conditions that affect air quality need to be analyzed. Therefore, the major aim of this research is to estimate the values of the major air pollutant parameters in the Colombo district by building predictive models for them. As for this, the historical weekly air pollutant parameters of the Colombo district were collected from the National Building Research Organization (NBRO) for the period of April 2020 to September 2023 with an aim to quantify as well as to understand the typical patterns of the air pollutants concentration in that region. Major air pollutant parameters such as PM2.5, PM10, NO2, and SO2 were considered in this study. Then, univariate time series models were fitted for the weekly data related to the air pollutants in the short time duration, and the accuracy of the models were assessed using RMSE, MAPE, and MAE values. For each parameter, ten candidate models were created separately, and the model with the lowest AIC value and all significant coefficients was selected as the best model. Also, the diagnostic tests recommended that the residuals of all models were normally distributed, exhibited no heteroscedasticity and no autocorrelation of residuals. Indicating that these models can be used in future predictions. Here, ARIMA(2,1,0), ARIMA(2,1,0), ARIMA(2,1,2), and ARIMA(3,1,1) models were discovered, for PM2.5, PM10, NO2, and SO2, respectively. The MAPE values are 15.434, 16.374, 21.130 and 17.902, respectively. The predictive modes suggest that these air pollutants have increased and decreased over time during our testing period. When analyzing the univariate time series model of various air pollution components, it was noted that the forecasted measurement values were slightly higher. The main reason is that other factors such as different air pollution parameters and meteorological factors from past data were not considered. Therefore, the accuracy of future predictions may be compromised if past data on these additional factors are not incorporated into the modeling process. Therefore, as future works, further analysis using multivariate models has been used to determine the relationship between meteorological factors and air quality parameters.