Tropical Ground-level Ozone Modeling in Urban Areas of Thailand

Abstract

This work applied multiple linear regression to model day time ozone (O3day) and daily maximum ozone (O3max) during 7:00-18:00 h by using its lagged day time ozone (O3lagday) and lagged daily maximum ozone (O3lagmax) in two urban areas (Bangkok and Samutprakarn) having different ozone precursor sources and urban topography, and in two seasons (wet and dry) having different meteorological variation by using a SAS® 9.2 software to analyze 16-year (1997-2012) data including 14,247,085 actual hourly measurements of O3, CO, NO2, and SO2 and meteorological variables such as temperature (T), wind speed (WS), relative humidity (RH) and solar radiation (SR). The results showed that a daily O3 average in Samutprakarn was 19.77 ± 11.30 ppb, higher than that in Bangkok at an average of 14.06 ± 8.74 ppb. In dry season, a daily O3 average (17.71 ± 10.08 ppb) was higher and more fluctuating than that in wet season (10.84 ± 6.24 ppb). O3day and O3max metrics showed the strongest correlation with O3lagday (r at 0.77 and 0.68 respectively) and O3lagmax (r at 0.68 and 0.66 respectively) following by RH, SR, CO, and NO2. The models with log-transformed O3 outcomes (lnO3day and lnO3max) and a lnO3lagday predictor provided better model R2 values. The lnO3day model had R2 ranging from 0.644 - 0.692 and was commonly predicted by lnO3lagday, CO7-18 and RHmin7-18. The lnO3max model had R2 ranging from 0.551 - 0.661 and was commonly predicted by lnO3lagday, COmin7-18 and RHmin7-18. The validation R2 values between observed O3 and predicted O3 using testing data ranged from 0.370 to 0.659. The predicted values trended to follow lagged O3 that was a dominant predictor. Model fitting could be improved in future if total VOCs data were available.

Keywords

• ground-level ozone
• ozone regression modeling
• wet and dry seasons
• meteorological effects