Comparison of surface ozone simulation among selected regional models in MICS-Asia III – effects of chemistry and vertical transport for the causes of difference

In order to clarify the causes of variability among the model outputs for surface ozone in the Model Intercomparison Study Asia Phase III (MICS-Asia III), three regional models, CMAQ v.5.0.2, CMAQ v.4.7.1, and NAQPMS (abbreviated as NAQM in this paper), have been selected. Detailed analyses of monthly averaged diurnal variation have been performed for selected grids covering the metropolitan areas of Beijing and Tokyo and at a remote oceanic site, Oki. The chemical reaction mechanism, SAPRC99, used in the CMAQ models tended to give a higher net chemical ozone production than CBM-Z used in NAQM, agreeing with previous studies. Inclusion of the heterogeneous “renoxification” reaction of HNO3 (on soot surface)→ NO+NO2 only in NAQM would give a higher NO concentration resulting in a better agreement with observational data for NO and nighttime O3 mixing ratios. In addition to chemical processes, the difference in the vertical transport of O3 was found to affect the simulated results significantly. Particularly, the increase in downward O3 flux from the upper layer to the surface after dawn was found to be substantially different among the models. Larger early morning vertical transport of O3 simulated by CMAQ 5.0.2 is thought to be the reason for higher daytime O3 in July in this model. All three models overestimated the daytime ozone by ca. 20 ppbv at the remote site Oki in July, where in situ photochemical activity is minimal.


Introduction
In the Model Intercomparison Study Asia Phase III (MICS-Asia Phase III), one of the targets was to narrow down the difference in the model simulation results by using common key input parameters such as precursor emissions, meteorological fields, and boundary conditions to allow a more focused discussion on the causes of the difference among model outputs. In most of the past model intercomparison studies for chemical transport models (CTM) for air quality, such key parameters were not common to all the models, which made the discussion of the causes of the differences among the model outputs difficult, and the results often demonstrated that the ensemble mean of simulated mixing ratios agreed reasonably well with observations even though the disagreement among the models were often significantly large (for example, Han et al., 2008;Fiore et al., 2009).
In order to improve the situation of model intercomparison studies, participants of the MICS-Asia III studies agreed to use common emission data (Li, M. et al., 2017) Modeling System, which is abbreviated to NAQM in this paper hereafter for simplicity). It turned out, however, that even though these 12 models used the specified common key input components, large variabilities in the spatial distribution and absolute mixing ratios among the models were found for ozone (O 3 ) (Li, J. et al., in this special issue).
In the present study, three regional models, two WRF-CMAQ, v.5.02 and v.4.7.1, and WRF-NAQM were selected among the 12 above mentioned models to elucidate the causes of differences, and detailed comparisons were made for selected grids covering the metropolitan areas of Beijing and Tokyo, and at a remote oceanic site at Oki in April and July in 2010. We selected the two models of CMAQ because CMAQ models have been widely used to assess the air quality for ozone in Asia (e.g. Yamaji et al., 2008;Kurokawa et al., 2009;Fu et al., 2012), and the difference in the simulated results between different versions (v.5.02 and v.4.7.1) is of concern. Furthermore, we selected WRF-NAQM because this is one of the regional CTMs developed in Asia, giving substantially lower mixing ratios of surface ozone as compared to most of WRF-CMAQ including the selected two models (Li, J. et al., in this special issue). The metropolitan areas of the two megacities of Beijing and Tokyo have been selected for the comparison to test whether regional models can be applied to the mitigation policy of urban ozone pollution. Oki, an EANET (Acid Deposition Monitoring Network in East Asia) monitoring station located in the southern part of the Sea of Japan, was selected as a remote reference site located between the two megacities, as in situ photochemical production of O 3 is known to be minimal there (Jaffe et al., 1996).

Models
Basic features and the simulated whole domain of the regional models, CMAQ v.4.7 (Foley et al., 2010), v.5.0 (CMAS, 2011, and NAQM (Li, J. et al., 2016) used in this study, are given elsewhere in this issue (Li, J. et al., in this special issue). The employed horizontal resolution was 45 km for all the models, and the maximum height and number of vertical layers were 45 km and 40 layers, respectively, in the CMAQ models and 20 km and 20 layers, respectively, in the NAQM, so that the vertical resolution in the troposphere was about the same. The lowest layer for which the simulated data of ozone were extracted in this paper was 50 m from the ground. Model calculations by the CMAQ v.5.0.2, v.4.7.1, and NAQM were conducted at the University of Tennessee, USA, National Institute for Environmental Studies, Japan, and Institute of Atmospheric Physics, China, respectively. All the models used the common meteorological fields from WRF simulation and common emissions of MIX (0.25°× 0.25°) for 2010 (Li, M. et al., 2017) developed in the MICS-Asia III project. The initial and boundary conditions were supplied by global models, CHASER for CMAQ v.4.7.1 and NAQM, and GEOS-Chem for CMAQ v.5.0.2. It was agreed that either CHASER or GEOS-Chem may be used in the MICS-Asia III, since they were confirmed to give reasonably good agreement for the O 3 field in the Asian domain.
Other than these three key components (emissions, meteorological field, and boundary conditions), the three models employed different sub-models and parameters for e.g. the gas-phase and aerosol chemistry module, dry deposition parameters, boundary layer scheme, etc.

Comparison Domain and Observational Data
All the comparisons between the model simulations and the model using observational data were made for monthly averaged diurnal variations in the mixing ratios of O 3 and NO in April and July. April and July were chosen here because in situ photochemical build-up of O 3 in April is insignificant but the daytime maximum mixing ratio of O 3 is relatively high reflecting the well-known spring maximum of O 3 for the background in the northern hemisphere including East Asia (Monks, 2000;Pochanart et al., 2003), while in July, a much higher in situ photochemical build-up of O 3 is observed in urban areas in East Asia. Two representative megacities, Beijing and Tokyo, were selected as urban areas for the comparison. As a remote reference site, Oki, an EANET site situated between Beijing and Tokyo, was selected. The Oki site is located on a cliff of an island where the local emissions of NO x and VOCs are insignificant so that in situ production of O 3 is also minimal (Jaffe et al., 1996;Pochanart et al., 2002). Since the NO levels at Oki are too low to get any meaningful data using the conventional chemiluminescence NO x monitoring instrument, comparison with modeling results was made in this study only for O 3 at this site. All the calculations were conducted for the whole year of 2010 using the meteorological field and emission data for this year.    Other published observational data of diurnal variation of O 3 in Beijing in April are available by Xu et al. (2011) at four sites, two urban (Fengtai and Baolian), one suburban (Shunyi), and one rural (Shangdianzi) in summer (21 June-12 September) in 2007. Since the diurnal variation of the urban and suburban sites is consistent, the average of these three sites is plotted in Fig. 2 marked by a dashed line with triangles. No monthly average diurnal variation of O 3 is available for April in Beijing in the literature. Chen et al (2015) reported the monthly averaged daily maximum mixing ratio of O 3 to be ca. 60 ppbv at an urban site (Dongsi), and ca. 75 and ca. 65 ppbv at two suburban sites (Daxing and Shunyi, respectively) within the selected grids in this study. If we simply take the average of these three values, the daily maximum mixing ratio is ca.
65 ppbv (not shown in Fig. 2(a)). Only the IAP data are plotted for NO with solid lines in Figs.

2(c) and (d).
As can be seen in Figs parallel with the NAQM values for whole days in all cases, but the predicted mixing ratios by CMAQ 4.7.1 are by ca. 20 ppbv and ca. 40 ppbv higher than those predicted by NAQM in April and July, respectively, both in Beijing and in Tokyo. The O 3 mixing ratios predicted by CMAQ 5.0.2 have a peculiar seasonal characteristics, i.e., the mixing ratio is slightly higher but close to that predicted by NAQM within 10 ppbv both in Beijing and in Tokyo in April, whereas in July the daytime O 3 maximum predicted by CMAQ 5.0.2 is very close to that predicted by CMAQ 4.7.1, much higher than the value by NAQM. In Tokyo, the simulated mixing ratios of CMAQ 5.0.2 and NAQM are closer to the observations in April, and NAQM gives a closer matching with observations in July, while CMAQ 4.7.1 overestimates the values in both months as shown in Figs. 2 (b) and 3(b). A comparison with the observations will be discussed later including the uncertainty of the observational data in Beijing.
The observed mixing ratios of NO show a peak value at around 7 AM, a decrease during morning followed by a slow decay in the afternoon, and they start to build up during nighttime both in April and in July, in both Beijing and Tokyo. The peak values of the mixing ratios in the morning are ca. 13-14 and 6 ppbv in April and ca. 11 and 5-6 ppbv in July in Beijing and Tokyo, respectively. The minimum mixing ratios in the evening are ca. 1.7 and 1.4 ppbv in April and 2.3 and 1.3 ppbv in July in Beijing and Tokyo, respectively. Thus, it can be noted that the NO mixing ratios in Beijing are nearly the double of those in Tokyo.
The simulated mixing ratios of NO are generally in the order of NAQM > CMAQ 5.0.2 > CMAQ 4.7.1., but they vary considerably among the models. In April, CMAQ 5.0.2 gives morning peak values of 13-14ppbv in Beijing and ca. 5 ppbv in Tokyo, which agrees well with the observations. NAQM overpredicts the NO mixing ratio in April in Beijing but gives a reasonable agreement with the observations in Tokyo as shown in Figs 2(c) and 3(c). In contrast, CMAQ 4.7.1 gives a broad daytime peak of only ca. 2 ppbv in Beijing and ca. 1 ppbv in Tokyo in April, which is quite different from other models, and it underpredicts considerably the observational data. In July, only NAQM gives a morning peak mixing ratio of ca. 8 ppbv in Beijing and 5.5 ppbv in Tokyo, agreeing fairly well with the observations including diurnal variation (Figs. 2(d) and 3(d)). In contrast, both CMAQ 5.0.2 and 4.7.1 give morning peaks as low as 1-2 ppbv and nearly zero mixing ratio during nighttime, which are significantly lower than the observational values. The causes of the differences in the simulated results among the three models mentioned above must be due to either chemical or transport processes incorporated in the models. Here, possible causes of differences of those processes are discussed individually.

Comparison of Chemical Mechanism Sub-Modules
One of the differences in the three models in this study is the chemical reaction mechanism sub-module. CMAQ 5.0.2 and 4.7.1 incorporate SAPRC99 while NAQM employs CBM-Z. It has been well known that different photochemical mechanisms used in the regional chemical transport models produce different results in the prediction of O 3 . Jimenez et al. (2003) compared seven photochemical mechanisms including CBM-IV (Gery et al., 1989) and SAPRC99 using a box model. Comparisons of CBM-IV, CBM-V  and SAPRC99 incorporated into regional chemical transport models have been made by Faraji et al. (2007) and Luecken et al. (2008). The main differences among these mechanisms have been noted to be the lumping technique describing organic compounds into surrogate groups (Jimenez et al., 2003), the differences in the products of the reaction of aromatics with OH radical, and the overall branching ratio of radical generation and termination reactions (Faraji et al., 2007). The results of these studies gave a consistent picture that SAPRC99 gives higher concentrations of O 3 than CBM-IV both in the box model calculation and in regional model simulation over the United States. The O 3 concentration obtained by CBM-V is reported to be between the CBM-IV and SAPRC99 values . The reason for the higher concentration of O 3 by SAPRC99 has been deduced to be due to the more efficient peroxy radical production in the photochemical reaction scheme of SAPR99 compared to those of the CBM modules. with a process analysis package. The net chemical production was calculated in each grid and then the average was taken for all the grids. As revealed in Figure 5, the CMAQ models gave higher net ozone productions than the NAQM models did, which is consistent with the results of earlier studies, showing that the photochemical reaction scheme of SAPRC99 gives a higher O 3 production than do the CBM modules. The reaction scheme of CBM-Z is the revision of 3(a), (b) although the effect is marginal in the case of Beijing in April. The cause of the early morning peak of net O 3 production in NAQM might be due to the photolysis of higher HONO that is produced by the heterogeneous reaction of NO 2 . More quantitative sensitivity analyses should be performed to confirm these effects.
In April, the net chemical production of O 3 is, in general, negative in all the models for both Beijing and Tokyo, except for that in CMAQ 4.7.1 around midday and that in NAQM in early morning, showing slightly positive values. A tendency of higher net O 3 production is seen particularly for CMAQ 4.7.1, which may be the main cause of higher O 3 by this model both in Beijing and in Tokyo in April ( Fig. 2(a) and Fig. 3(a)). The daytime net O 3 production simulated by CMAQ 5.0.2 is similar to that simulated by CMAQ 4.7.1 in July but is substantially lower in April. Since the chemistry mechanism of SAPRC99 is used in both CMAQ versions, the difference may be related to the vertical transport of some relevant species.

Effects of heterogeneous "renoxification" reaction of HNO 3
have been well established (Burkholder et al., 2015 and earlier evaluations of the series) and more or less the same reaction rates are employed in both of SAPRC99 and CBM-Z. As for the heterogeneous processes affecting NO x , the reaction, is included in common in the heterogeneous inorganic chemistry sub-module, ISORROPIA and employed in the CMAQ and NAQM models.
It has been noted that the simulated gaseous HNO 3 concentration and HNO 3 /NO x ratio were found to be 2-10 times higher when using global and regional chemical transport models than the observational data during the PEM-West (Singh et al., 1996), TRACE-P (Talbot et al., 2003), and PEM-Tropics A and SONEX (Brunner et al., 2005) aircraft campaigns over the Pacific and Atlantic Ocean. The same result has also been reported by ground observations in the remote troposphere at Mauna Loa (Hauglustaine et al., 1996) and in the polluted boundary layer of Beijing-Tianjin-Hebei region .
Another concern regarding recent NO x chemistry has been focused on the high concentration of HONO in the urban atmosphere, which is thought to be produced by the heterogeneous reaction of NO 2 and H 2 O on the aerosol and ground surface (for example, Li, Y. et al., 2011;Gonçalves et al., 2012;Wong et al., 2013). Inclusion of the additional heterogeneous source of HONO not only affects the photochemical O 3 formation due to the increase of OH radicals but also increases HNO 3 due to the increase of the reaction, OH + NO 2 + M→ HNO 3 + M.  have shown that the inclusion of the heterogeneous formation of HONO gives more HNO 3 , which tends to give a larger overestimation of gaseous HNO 3 in the Beijing-Tianjin-Hebei region.
In order to solve the problem of overestimation of HNO 3, the heterogeneous reaction of HNO 3 on soot surface to reproduce NO and NO 2 has been proposed as "renoxification" process early by Lary et al. (1997) in the analysis of the above aircraft observation data. The heterogeneous reaction of HNO 3 on soot surface to produce NO/NO 2 has been confirmed experimentally in laboratory studies (Disselkamp et al., 2000;Muñoz and Rossi, 2002), although the product ratio and reaction mechanism are not well established yet. The steady state uptake coefficient γ ss of this reaction has been reported to be (4.6 ± 1,6) × 10 -3 for black soot using geometric surface area (Muñoz and Rossi, 2002).
Only NAQM among the three models studied here incorporates the following heterogeneous non-stoichiometric reactions on the surface of soot .
with γ HNO3 = 3.0 × 10 -3 for Reaction (5) and γ HONO = 1.0 × 10 -4 for Reaction (6). The "renoxification" by Reaction (5) (1)), which may also give a better agreement for O 3 with the observation, particularly during nighttime. However, no quantitative sensitivity analysis has been made in the present study, and it is highly recommended that verification of the importance of such heterogeneous renoxification reaction in model simulation be made against accurate measurements of gaseous HNO 3 together with other NO y in the polluted urban atmosphere.

Effects of Vertical Transport
Other than the difference in chemical reaction mechanisms, the difference in transport module could give rise to differences in the output of O 3 concentrations. In order to analyze the effects of transport, process analysis of horizontal and vertical transport of O 3 has been conducted. Since it has been found that there is not much difference in horizontal transport and surface deposition, and the chemical mechanisms of CMAQ 5.0.2 and CMAQ 4.7.1 are the same, the difference in model performance must be ascribed to the difference in vertical transport processes.  13 ppbv hr -1 ) are much higher than those in NAQM (< 5 ppbv hr -1 ). Thus, it is concluded that at least a part of much higher O 3 concentrations estimated by CMAQ 5.0.2 and 4.7.1 as compared to NAQM shown in Fig. 2(b) and Fig. 3(b) in Beijing and Tokyo in July can be ascribed to the higher downward O 3 flux estimated by the CMAQ models compared to NAQM.
A peculiar feature of vertical O 3 flux in the CMAQ 5.0.2 shown in Fig. 6 is the strong positive morning peak at around 7 and 6 am in Beijing in April and July, respectively, and also at 6-7 am in Tokyo in April. Here, it should be noted that the vertical transport was computed in the PPM scheme in CMAQ 5.0.2 instead of the direct extraction from WRF in CMAQ 4.7.1 as described in the method section. The PPM method has been known to introduce more downward flux of O 3 from higher layers to the surface layer. Another point to be noted is the delayed rise of vertical downward O 3 flux by nearly 2 hours in NAQM both in April and July in Beijing and Tokyo. Although this feature is not scrutinized in this study, it should be noted here that the vertical transport treatment affects significantly the simulated results of O 3 in regional chemical transport models.

Comparison of the Transport Process in CMAQ v. 5.0.2 and v. 4.7.1
As seen in Fig. 2  13 mixing ratio and hourly change shows a similar morning peak in April, the CMAQ 5.0.2 simulation doesn't give any such morning peak, which would correspond to a very low NO mixing ratio simulated by CMAQ 5.0.2 together with CMAQ 4.7.1 as seen in Fig. 2(d).
Although the phenomena could be caused by rapid oxidation of NO to NO 2 in summer, the reason is unknown at this stage.
It should be noted that after the large rise at 7-8 am, the hourly change of O 3 mixing ratio As for the observational data in Beijing, the daily maximum of O 3 mixing ratio in July is ca.
90 ppbv in Xu et al. (2011) and ca. 60 ppbv by the IAP data, while the nighttime minimums are both 10-20 ppbv consistently. Since the maximum O 3 mixing ratio in summer is expected to be higher in Beijing than in Tokyo (ca. 60 ppbv) due to higher NO (see Figs. 2(d) and 3(d)) and NO 2 levels by a factor of ∼2 (not shown), the higher observational data in Beijing than in Tokyo in Fig. 2(c) could be more representative for the average of the calculated grids in Beijing. Although there still is a large uncertainty in the monthly averaged observational data of O 3 in Beijing in 2010, a tendency of overestimation by CMAQ 5.0.2 and 4.7.1, and underestimation by NAQM in Beijing in July can be suggested.
In April in Beijing, Chen et al. (2015) reported the daily maximum mixing ratio of O 3 at ca.
65 ppbv in 2014-2015, which is substantially higher than the IAP data of ca. 40 ppbv in 2010.
An increase in surface ozone has been reported in Beijing at the rural sites of Shangdianzi during the period of 2004-2015 with regard to the maximum daily average 8 hr mixing ratios (MDA8) (Ma et al., 2016). Although the long-term increasing trend indicates an average rate of 1.13 ± 0.01 ppb yr -1 , no monthly data was reported, and the year-by-year variability is substantial. If we assume that the monthly averaged MDA8 values in April in 2010 are lower by 10 ppbv than those in 2015, the uncertainty of the daily maximum observational value in April in Beijing would be in the range of 40-55 ppbv. Thus, within the uncertainty range, a tendency of overestimation by CMAQ 4.7.1 and an underestimation by NAQM could be suggested.
As for the discussion of reproducibility of the model simulation, a comparison of three-year averaged values in more recent years after 2013, when routine monitoring data at considerably more sites within the targeted grids are available, would be highly desirable, particularly in Beijing.

Overestimation of O 3 at Oki, a remote oceanic site
At the remote site of Oki, an overestimation by ca. 20 ppb for daytime O 3 has been seen in July in Fig. 4 60 ppb and a summertime monthly minimum of 35-40 ppb (Pochanart et al., 1999(Pochanart et al., , 2002, which is consistent with the observational data shown in Fig. 4. The summertime minimum at Oki and other remote islands in this region are well established to be due to prevailing clean marine air (Pochanart et al., 2002;Yamaji et al., 2006).
Since the overestimation does not depend on the spatial resolution of the model, as noted above, and the daytime build-up of O 3 due to local photochemical activity is <10 ppbv in the observation and 5-15 ppbv in the simulation as shown in Fig. 4(b), the overestimation of O 3 concentration in July by all the three models cannot be ascribed to the direct influence of nearby terrestrial emissions of precursors in mainland Japan. The overestimation could be either due to a more frequent influence of terrestrial air masses by WRF compared to the real meteorology, or higher O 3 concentration in the oceanic air around this area affected by the influence of non-episodic terrestrial emissions including long-range transport. The reproduction of observed concentrations by models at Oki would be important for the analysis of air quality in Japan, since air masses passing through Oki provide a flowing-in background mixing ratio to mainland Japan.

Summary
In order to identify the causes of the substantial variability among the simulated modelling results for surface ozone in MICS-Asia III even though using the same emissions, meteorological field and boundary conditions, three regional models, namely CMAQ 5.0.2, 4.7.1 and NAQM were selected and a detailed comparison was made in the selected grids covering the metropolitan areas in Beijing and Tokyo and at the remote oceanic site of Oki. The analyses were made for the monthly averaged diurnal change of surface ozone in April and July in 2010.
The simulated O 3 concentration was the highest in CMAQ 4.7.1, followed by those in CMAQ 5.0.2 and NAQM both in Beijing and in Tokyo in April, while both CMAQ models gave much higher O 3 values than NAQM did in July. At Oki, the simulations for O 3 by all the three models agree well with each other and with the observation in April. In July, however, all the models overestimated daytime O 3 by ca. 20 ppb compared to the observation.
Three causes for the difference among model outputs have been identified and discussed.
(1) The chemistry mechanism sub-module, SAPRC99 used in the CMAQ was found to give higher net ozone production values than CBM-Z in NAQM, agreeing with previous studies.
(2) Higher NO concentrations have been predicted by NAQM than by CMAQ, possibly due to the inclusion of a heterogeneous "renoxification" reaction of HNO 3 (on soot surface) → NO + NO 2 , which gave a better agreement with observational concentration particularly for nighttime NO and O 3 .
(3) A vertical downward O 3 flux was found to affect substantially the diurnal pattern and mixing ratios of O 3 .