Analysis of Daytime and Nighttime Ground Level Ozone Concentrations Using Boosted Regression Tree Technique

Abstract

This paper investigated the use of boosted regression trees (BRTs) to draw an inference about daytime and nighttime ozone formation in a coastal environment. Hourly ground-level ozone data for a full calendar year in 2010 were obtained from the Kemaman (CA 002) air quality monitoring station. A BRT model was developed using hourly ozone data as a response variable and nitric oxide (NO), Nitrogen Dioxide (NO2) and Nitrogen Dioxide (NOx) and meteorological parameters as explanatory variables. The ozone BRT algorithm model was constructed from multiple regression models, and the 'best iteration' of BRT model was performed by optimizing prediction performance. Sensitivity testing of the BRT model was conducted to determine the best parameters and good explanatory variables. Using the number of trees between 2,500-3,500, learning rate of 0.01, and interaction depth of 5 were found to be the best setting for developing the ozone boosting model. The performance of the O3 boosting models were assessed, and the fraction of predictions within two factor (FAC2), coefficient of determination (R2) and the index of agreement (IOA) of the model developed for day and nighttime are 0.93, 0.69 and 0.73 for daytime and 0.79, 0.55 and 0.69 for nighttime respectively. Results showed that the model developed was within the acceptable range and could be used to understand ozone formation and identify potential sources of ozone for estimating O3 concentrations during daytime and nighttime. Results indicated that the wind speed, wind direction, relative humidity, and temperature were the most dominant variables in terms of influencing ozone formation. Finally, empirical evidence of the production of a high ozone level by wind blowing from coastal areas towards the interior region, especially from industrial areas, was obtained.

Keywords

• stochastic
• algorithm
• ozone
• R software
• interactions