ORIGINAL RESEARCH
Performance of Multivariate Time Series on Forecasting the Tropospheric Ozone (O3)
 
More details
Hide details
1
Cluster of Water and Environmental Engineering, Faculty of Civil Engineering and Built Environment, Universiti Tun Hussein Onn Malaysia, 86400 Batu Pahat, Johor
2
Mathematics Division, School of Distance Education, Universiti Sains Malaysia, 11800 Penang
CORRESPONDING AUTHOR
Hazrul Abdul Hamid   

School of Distance Education, Universiti Sains Malaysia, 11800, Gelugor, Malaysia
Submission date: 2021-01-14
Final revision date: 2021-03-19
Acceptance date: 2021-03-25
Online publication date: 2021-09-10
 
 
KEYWORDS
TOPICS
ABSTRACT
The tropospheric ozone (O3) is known as a hazardous ambient air pollutant that adversely affects human health. Recently, many studies have been devoted to assess and monitor the O3 concentration due to its impact on society health. This study was carried out to develop a model to forecast O3 concentration. A comparison between univariate and multivariate time series to examine the most appropriate model was made. The air quality data used in this research was collected from three (3) stations namely Perai, Penang (industrial), Alor Setar, Kedah (urban) and Jerantut, Pahang (background). The selection of background station allows for comparisons to be made with stations closer to anthropogenic emissions. Based on Akaike Information Criterion (AIC), the appropriate multivariate time series to develop a forecasting model for Perai, Alor Setar and Jerantut monitoring stations were vector autoregressive - VAR(3), vector autoregressive - VAR(2) and vector moving average - VMA(2), respectively. The lowest root mean square error (RMSE) is 0.0053 which is for the multivariate time series model in Perai while for normalized absolute error (NAE) and mean absolute error (MAE), the lowest is in Jerantut with 0.0850 and 0.0013 respectively. Validation of the models using three error measures shows that the multivariate time series model performed better compared to the univariate time series model.
eISSN:2083-5906
ISSN:1230-1485