Ozone production during the field campaign RISFEX 2003 in the sea of Japan : analysis of sensitivity and behaviour based on an improved indicator

The ratio 8=kHC+OH[HC]/kNOx+OH[NOx] is used as an indicator for the sensitivity of ozone production (P (O3)) to HC and NOx in the field campaign RISFEX 2003 (RIShiri Fall EXperiment 2003) at Rishiri Island (45.07 N, 141.12 E, and 35 m a.s.l.) in the sea of Japan during September 2003. Four different sensitivity regimes are obtained based on the indicator. The sensitivity is found to show a distinctive pattern in each regime. In Regime I (8<1), P (O3) almost linearly increases with increasing HC and almost linearly decreases with increasing NOx. In Regime II (1<8<9± 5), there is a less-than-linear increase inP (O3) with HC and a less-than-linear decrease with NOx. P (O3) less-than-linearly increases with both HC and NOx in Regime III (9± 5<8<45± 7), and near linearly increases with NOx and is nearly constant with increasing HC in Regime IV (8>45± 7). During the campaign, 91 percent ofP (O3) data appear in Regime III and IV, indicating that NOx is a limiting factor of ozone production. Hence, it may be an efficient strategy to control NO x emission for ozone abatement at the site. Comparisons between the observed P (O3) and the ones modelled have represented general agreement. However, the model tends to underestimate P (O3) in Regime II, implying that an important source of peroxy radicals is possibly missed. In Regime IV, the modelled P (O3) is systematically larger than the measured one under a low j (O1D) condition, which may be caused by the over-estimated yields of peroxy radicals from the reactions of monoterpenes with ozone. A budget analysis indicates that sensitivity of P (O3) is declining with HC and enhancing with NO when the condition Correspondence to: B. Qi (b.qi@163.com) shifts from Regime II to Regime IV, which is also observed through the analysis of P (O3) sensitivity using8. Sensitivity studies forP (O3) are conducted to determine the effect of NOx and monoterpenes on ozone production and the conclusions are very consistent with those derived from the indicator. This study demonstrates that the ratio 8 could be a useful index to ascertain the sensitivity of P (O3) to HC and NOx in the clean marine boundary layer.


Introduction
Tropospheric ozone is a major constituent of air pollution which is detrimental to humans and vegetation (WMO, 1999).Ozone is also a greenhouse gas in the upper troposphere and has an important impact on the radiative balance of the atmosphere (Brasseur et al., 1998).The concentration of surface ozone in the Northern Hemisphere has increased by a factor of 2 and more in average since the preindustrial era (Bojkov et al., 1988).Previously, it is assumed that tropospheric ozone comes from the stratosphere.However, recent work has shown that a large fraction of the tropospheric ozone is due to in situ photochemical production in rural and remote environments (Chameides and Walker, 1973;Fishman andCrutzen, 1977, 1978;Liu et al., 1980Liu et al., , 1987;;Logan, 1985Logan, , 1998;;Monks et al., 2000).High tropospheric ozone has been a great problem in many parts of the world (Frank et al., 2001).To develop effective abatement strategies, it is necessary to investigate whether the ozone production is controlled by HC (hydrocarbons, which contains CH 4 and CO as well as anthropogenic and biogenic hydrocarbons), NO x or both.
Published by Copernicus Publications on behalf of the European Geosciences Union.Z. Q. Wang et al.: Ozone production during the field campaign RISFEX 2003 Essential for the ozone formation is the cycle of odd hydrogen (odd H) which can be defined as the sum of OH, HO 2 and RO 2 (Kleinman et al., 1986(Kleinman et al., , 1991)).During daytime, sources of hydroxyl radicals are the photolysis of ozone, of nitrous acid, of aldehydes and ozonolysis of alkenes which can be an important OH source in rural areas (Platt et al., 1986;Alicke et al., 2003;Paulson et al., 1996;Ariya et al., 2000).These formation paths lead to daytime OH levels in the range of several 10 6 cm −3 (Hein et al., 1997;Holland et al., 1998Holland et al., , 2003;;Andreas et al., 2003).OH radical reacts with inorganic and organic species to oxidize them, leading to HO 2 and a variety of organic peroxy radicals which can react with NO to convert it to NO 2 : The conversion of NO to NO 2 and the subsequent photolysis of NO 2 drive the ozone production: The reaction of OH with NO 2 provides a stable product and removes OH from the system: OH + NO 2 → HNO 3 (R7) The reactions of peroxy radicals with peroxy radicals also remove peroxy radicals from the system: During the odd hydrogen cycle, P (O 3 ) sensitivity greatly depends on the indirect competition between Reactions (R7) and (R8) as sinks for odd hydrogen radicals (Sillman, 1995(Sillman, , 1997)).As a result, the ratio between the products from Reactions (R8b) and (R7), the afternoon H 2 O 2 /HNO 3 , has been used as an indicator to ascertain ozone sensitivity (Sillman, 1995).However, the indicator is based on long-lived species and, thus, can only reflect the past sensitivity of the investigated air parcel (Frank et al., 2001).This makes it difficult to determine the local controlling factor of ozone formation, which is necessary for the development of abatement strategies.Therefore, we propose a new parameter as an indicator to determine the instantaneous sensitivity of ozone production, which is defined as: where k HC+OH and k NO x +OH are the combined rate coefficient for the reactions of OH radical with HC and NO x , respectively.
[HC], [NO] and [NO 2 ] are the concentrations of HC, NO and NO 2 , respectively.As (R8) rely on the level of peroxy radicals which is closely related to Reactions (R1) and (R2), the parameter reflects the relative size of two major sinks of odd hydrogen (Reactons R7 and R8) and, thus, is a valid indicator of P (O 3 ) sensitivity.Moreover, it is not based on photochemically produced long-lived species, but describes the instantaneous feature of an air parcel.So, is a more suitable tool for developing ozone abatement strategies.
The parameter is consistent with the indicator proposed by (Frank et al., 2001), which is defined as the ratio of the lifetimes of OH against the losses by reacting with VOCs and NO x .If the concentrations of HC except CH 4 and CO are used to calculate the value of , there exists a reciprocal relation between the values of and .The addition of CH 4 and CO in our work is because both of them can react with OH radical to produce peroxy radicals (CH 3 O 2 for CH 4 and HO 2 for CO), which are similar with other VOC species.
In this paper, we investigate as an improved indicator to determine the sensitivity of ozone production to HC and NO x and to analyse the behaviour of P (O 3 ) in different sensitive regions using the data from the RISFEX 2003 campaign.Furthermore, model studies based on Regional Atmospheric Chemistry Modelling (RACM) (Stockwell et al., 1997) are used to test the robustness of the indicator.

Experimental
The measurements were conducted at an observatory (Rishiri Island Observatory, RIO) (45.07 • N, 141.12 • E, 35 m a.s.l.) built on Rishiri Island, which is a round, dormant, volcanic island with a diameter of ca. 15 km and ca.20 km northwest of Hokkaido in the sea of Japan.In its center stands a 1721 m high mountain covered by a coniferous forest on its slope.RIO is situated on a foothill of the mountain ca.800 m away from the shore (Fig. 1).The population on the island is ca.7500 and the local pollution can be negligible (Tanimoto et al., 2000).A previous study at Rishiri Island has shown that the site receives the air masses usually from the clean Arctic, West Siberia and Pacific region as well as from polluted Japan and continental Asia (Tanimoto et al., 2000).
In the observatory, there are two 2-m high containers used for housing instruments and the inlets for trace gas and aerosol measurements.Peroxy radicals were measured by PERCA technique (Cantrell et al., 1982).The inlet of the PERCA instrument was mounted on the top of a container.NO and NO 2 were measured by a chemiluminescence instrument with a photolytic converter (CLD770AL and PLC760, Eco Physics).The detection limits for NO and NO 2 measurements were 22 and 45 pptv (S/N=2, 1-min measurement time), respectively.O 3 was measured by a .Geographical location of Rishiri island and Rishiri Island Observatory commercial UV absorption analyser (49C, Thermo).CO was measured with a non-dispersive infrared (NDIR) photometer instrument (48C, Thermo).Black carbon was measured with a commercial instrument based on absorption photometry (AE-21, Magee Scientific).Non-methane hydrocarbons (NMHCs) were measured by GC-FID and GC-MS followed by sampling the air into canisters.HCHO, CH 3 CHO, acetone, toluene and monoterpenes were measured by proton-transfer reaction-mass spectrometry (PTR-MS).Speciation and quantification of monoterpenes were made with two GC-based instruments.The solar actinic flux and photolysis frequency of O 3 to O( 1 D) (j (O 1 D)) and NO 2 to NO (j (NO 2 )) were measured by an actinic flux spectral radiometer (GmbH, Meteorologie Consult).Temperature, relative humidity, pressure, and wind direction and speed were recorded with conventional meteorological instruments.A further detailed description of the measurements and instruments has been given elsewhere (Qi et al., 2007).

Model
A time-dependent box model based on Regional Atmospheric Chemistry Modelling (RACM) (Stockwell et al., 1997) is developed to describe the remote marine boundary layer (MBL) chemistry and to determine the production rate of ozone.Kinetic rate constants are updated using the results in Sander et al. (2003).The model is also updated by incorporating more detailed monoterpenes chemistry (Kanaya et al., 2002a).The running of the model is constrained by measured stable chemical species, photolysis frequencies and meteorological parameters with the integrations conducting each day for 24 h starting at 00:00 and ending 24:00 JST (Japan standard time) at 10-min intervals given the measure-ment data available.If time resolution of the measurements is greater or less than once every 10 min, averaging or linear interpolation will be used to calculate the input data.Other unmeasured oxygenated hydrocarbon species are initially set to zero when each calculation is started and are allowed to accumulate with time integration.The calculations are performed for a 10-min-period with the integral time of ∼15 min to ensure that the concentrations of major radicals (HO 2 , OH and RO 2 ) have reached an approximate steady state.In daytime, the 1σ uncertainty of the model, based on the combined uncertainties in the kinetic rate coefficients and in the measured concentrations of species, is estimated using a Monte Carlo approach to be ca.± 30% (Carslaw et al., 1999).

Ozone production rate (P (O 3 )) during the campaign
The ozone production rate, P (O 3 ), can be approximately determined by the rate at which NO is oxidized to NO 2 by the reactions with peroxy radicals (Kleinman et al., 1995): where k is a combined rate coefficient for the oxidation of NO to NO 2 by all peroxy radicals.In this approach, the minor pathway of the higher organic peroxy radicals which lead to the formation of organic nitrates and the loss of NO 2 by reaction with OH are neglected, thus, Eq. (2) represents an upper limit for P (O 3 ) (Mihelcic et al., 2003).During RISFEX2003, the data coverage allows for the determination of P (O 3 ) during 18-21 September.The daytime weather during this period is typically clear in the morning hours (06:00-11:00 JST), and scattered clouds frequently appear overhead at noon and in the afternoon as shown in Fig. 2d.The daytime is defined as the interval of j (O 1 D)>1×10 −7 s −1 , corresponding to time from 06:00 to 18:00 JST.The local wind direction at the site is dominated by the south in daytime and shifts to the north in the night, exhibiting typical land-sea breeze.
We estimate the values of P (O 3 ) using the observed HO 2 and RO 2 concentration.The time series of observed P (O 3 ) together with pertinent chemical species and physical parameters in 10-min averages are shown in Fig. 2. We can see from Fig. 2a that P (O 3 ) varies greatly between these days, with midday values (10-min average) varying from 0.4 to 2.5 ppbv/hr.The scatter in P (O 3 ) is primarily caused by NO, since NO is more variable than RO x .The high P (O 3 ) occurring on 18 September is due to both high NO and peroxy radical levels.The daily mean P (O 3 ) for all data determined in this work is 0.93 ppbv/hr, similar to reported values obtained in MBL (Salisbury et al., 2002;Monks et al., 1998;Fleming et al., 2006).Figure 2b shows that RO x (the sum of HO 2 , RO 2 and OH radicals) signals increase quickly in the early  morning on clear-sky days, and reach a peak at ∼11:50 JST.In the afternoon, the radical signals decay consistently with the attenuation of UV radiation flux.In contrast, little variation is observed on the cloudy day (19 September).This suggests that the production of peroxy radicals is strongly driven by photochemistry, same as the previous observations in MBL (Burkert et al., 2001;Carpenter et al., 1997).The value of , calculated using the observed data of HC species and NO x concentrations, varies in the range of 3.5∼107.2,with no distinct diurnal variation (Fig. 2e).The averaged values and ranges of measured chemical species and meteorological parameters are listed in Table 1.

Sensitivity of P (O 3 ) to HC and NO x
Figure 3 shows the dependence of observed P (O 3 ) on .A modelled trend of P (O 3 ) with increasing is also shown in Fig. 3, which is similar to the variation of observed P (O 3 ).The running of the model is under the condition that the concentrations of all chemical species, except for NO and NO 2 , and physical parameters are constrained to those measured at 11:50 JST of 18 September.With the ratio of NO to NO 2 invariable, we change the concentrations of NO and NO 2 to obtain a series of different in the range of 0.01-1000.As Fig. 3 shows, the variation of P (O 3 ) can be divided into two different trends as changes.Under low conditions, P (O 3 ) increases as increases, indicating a positive correlation between P (O 3 ) and .When rises to a critical value, denoted as opt , P (O 3 ) reaches the maximum amount in the process of the continual increase of .At > opt , P (O 3 ) is found to be negatively correlated with .
To investigate the sensitivity of P (O 3 ) in different regions, we also calculate the relative sensitivity of P (O 3 ) to NO and HC, dlnP (O 3 )/dln[NO] and dlnP (O 3 )/dln[HC], which were found to be equal to: where k 1 and k 2 are rate constants for the reactions of HO 2 +HO 2 and HO 2 +RO 2 , respectively.L R represents all radical other radical reactions including HO+HO 2 and RO 2 +R O 2 , L N denotes all radical loss reactions between free radicals and NO or NO 2 including HO+NO 2 →HNO 3 and RO 2 +NO→ organic nitrate (Kleinman et al., 1997) 16) 58( 13) 52( 11) 55( 14) 37 94 studies (Milford et al., 1994;Frank et al., 2001), we notice that the definitions of the regimes are greatly different.Milford et al. (1994) distinguished only between two regimes, and considered that P (O 3 ) was NO x -limited when the value of dlnP (O 3 )/dln [NO] was larger than dlnP (O 3 )/dln[HC], otherwise it was HC-limited.From that classification, P (O 3 ) is HC-limited in Regime I and, therefore, ozone abatement must rely on the reduction of HC concentration.However, based on our former analysis, a rise of NO x can also reduce the production of ozone in this region.In the work of Frank et al. (2001), P (O 3 ) was distinguished between three regimes, and the definitions of Regime II and III were similar with Regime III and IV in our study, respectively.The difference is that they classify Regime I and II (in our work) into one region.Nevertheless, it is clear that the P (O 3 ) sensitivity for HC and NO x is completely different in Regime I and Regime II.
In this paper, the border between Regime I and II is fixed as =1 which indicates a comparable competition for OH between HC and NO x .The border between Regime II and III, denoted as opt , where the maximum P (O 3 ) occurs at the given HC level, is defined as the value of at which dlnP (O 3 )/dln[NO] equals to zero.Besides, the border between Regime III and IV is defined as the value of at which the ratio of dlnP (O 3 )/dln[HC] to dlnP (O 3 )/dln[NO] equals to 0.05.Based on the calculated sensitivity and four defined regimes, there are 24, 127 and 120 data points located in Regime II, III and IV, respectively, while none appears in Regime I during the campaign.It indicates that 91 percent of P (O 3 ) data is occurred in Regime III and IV, implying that NO x is a limiting factor for ozone production.Hence, the controlling of NO x emission may be an efficient strategy for ozone abatement at the site.During the period of 18-21 September, opt is ca.9 ± 5, which is consistent with the results in previous studies (Frank et al., 2001;Tonnesen et al., 2000;Kleinman et al., 2005).The indicator value of =9 is corresponding to a value of 0.20 for within the range of 0.1-0.3 reported by Frank et al. (2001) (the contributions of CO and CH 4 to L(HC) is ∼46% which is calculated by the observed data at 11:50 JST of 18 September).At the point of =9, the percentage of OH reacting with VOCs is about 83% which is in agreement with the value of 82%∼86% published by Tonnesen et al. (2000).From Kleinman et al. (2005), it was found that the maximum P (O 3 ) occurred when the ratio of HC reactivity to NO x concentrations was approximately equal to 1 s −1 ppbv −1 which was in the range of 0.9∼3.indicator value of 5∼15 found in our paper.The value of at the border between Regime III and IV is ca.45 ± 7, which is greatly lower than the value of ∼180 ( =0.01) proposed by Frank et al. (2001).It implies that the extent of Regime IV in our work is wider than their VOC-insensitive region.The indicator value of =45 is associated with ∼100 pptv NO which is in good agreement with the threshold NO concentration where HO x radicals begin at nearly constant with decreasing NO (Kanaya et al., 2002c), which can be considered as a distinctive characteristic of HC-insensitive region (Regime IV).
Figure 5 shows the dependence of calculated P (O 3 ) sensitivity and the parameter L N /Q upon for daytime 10-min averaged data.The indicator is shown to correlate with the relative sensitivity of P (O 3 ) in Fig. 5a, illustrating that it is a valid parameter to reflect P (O 3 ) sensitivity.As shown in Fig. 5b, is fully anti-correlated with the parameter L N /Q which is a successful indicator for determining the sensitivity of P (O 3 ) (Kleinman et al., 2005).Undoubtedly, the indicator is a more convenient parameter rather than L N /Q, as the value of could be easily measured or calculated (Frank et al., 2001).However, a constant L N /Q is always associated with a range of values, as represented in Fig. 5b.It implies that a problem with our indicator is that its correlation with P (O 3 ) sensitivity may shift under different conditions, which is similar with previous indicators.Therefore, it is necessary to test the robustness of the indicator.

Robustness study of
As discussed by Frank et al. (2001), the indicator can be used to find the instantaneous sensitivity regime of an air parcel, in contradistinction to other earlier proposed indicators based on long-lived species.More importantly, it is more robust than the indicators NO y and O 3 /NO z and of comparable robustness as the indicator H 2 O 2 /HNO 3 .As an improved parameter based on the indicator , naturally inherits the advantages of the original indicator, but it also has the same problem from very high percentages of large alkenes.Fortunately, the studied site is more free from the impact of human activities, thus, the emission of alkenes which are mainly emitted from diesel motors is greatly limited.Therefore, the improved indicator is suitable for ascertaining the sensitivity of P (O 3 ) on this island.Nonetheless, due to the addition of CH 4 and CO concentrations in determining the value of , it is necessary to retest the robustness of the indicator.
Firstly, the model calculations are used to investigate the effect of the atmospheric compositions of VOCs on the variation of opt .We separately raise the observed concentrations of each VOC species by a factor of 1.5 with holding other VOCs constant to calculate P (O 3 ) at different binned and then to determine the value of opt .The performance of the calculations is similar with that described in Sect.4.2.By comparing those opt values with the initial opt calculated from the observed data, we obtain the changes of    opt ) due to the increase of VOCs concentrations (Fig. 6). Figure 6a represents the percentage variations of opt relative to the initial value.As Fig. 6a shows, the increases of isoprene (ISO) and methacrolein (MACR) have a greater influence on opt than other VOCs, leading to a 1.7% increase and a 1.5% decrease of opt , respectively.This is partly attributed to their high fractional contributions to L(HC) (25% for ISO, while 12% for MACR) which are significantly higher than others (all are less than 5%).Considering that the fractional contributions of each VOCs to total reactivity are different from each other, we also calculate the changes of opt with one unit (ppbv) increase of VOCs (Fig. 6b).From Fig. 6b, it is shown that the increases of ISO, acetaldehyde (ALD), monoterpenes, toluene (TOL) and MACR lead to a great change of opt .Clearly, the increases of ISO and its oxidation product MACR show adverse effect on opt , which ensures a relatively constant value of opt under conditions with elevated ISO concentration on 18 September.However, TOL from anthropogenic can greatly reduce opt , illuminating that local opt is higher than it in more polluted environments.
Recent studies (Sillman, 1995;Thornton et al., 2002) have implied that the value of opt , at which the maximum amount of P (O 3 ) happens at a fixed level of HC, is closely related to the production rate of odd hydrogen radicals.Sillman (1995) found that the value of k OH HC/NO x at which the transition from NO x -sensitive to HC-sensitive conditions occurred, corresponding to opt in our work, was anti-correlate to Atmos.Chem. Phys., 10, 9579-9591, 2010 www.atmos-chem-phys.net/10/9579/2010/2002) also mentioned that the crossover between NO x -limited and NO x -saturated behaviour would shift to higher NO x levels, which implies a decrease of opt value when the production rate of radicals increases.Therefore, we use the model to test the robustness of our indicator to those factors which can improve the OH production rate, such as O 3 concentration, humidity, j (O 1 D) etc.
Figure 7 shows the dependence of P (O 3 ) on with different fixed O 3 concentration under high and low j (O 1 D) conditions.As shown in Fig. 7a, the variations of P (O 3 ) have the same tendency with increasing at each fixed concentration of O 3 .P (O 3 ) increases with increasing at < opt and then reaches the maximum value at opt .At > opt , the further increase of leads to a decrease of P (O 3 ).When the concentration of O 3 is constrained to 1 ppbv, the value of opt is ∼9.opt decreases to a value of 5 when O 3 concentration rises to 100 ppbv.It reveals a negative correlation between ozone concentrations and opt .Besides, the results from similar calculations under a low j (O 1 D) condition (Fig. 7b) are in qualitative agreement with those under a high j (O 1 D) condition.More importantly, it is found that the value of opt under the condition of 1 ppbv O 3 and high j 1 D) is nearly equal to it under the conditions of 100 ppbv O 3 and low j (O 1 D), which associates with the similar levels of OH radical, implying that the inherent factor of opt is a radical production rate rather than ozone concentration or photolysis rate.Figure 8 shows the dependence of P (O 3 ) on when the relative humidity (RH) is constrained to 10%, 20%, 50% and 90%, respectively.Under a high j (O 1 D) condition, as presented in Fig. 8a, the increase of RH associates with OH increase leading to a low opt .However, increasing RH under a low j (O 1 D) condition can not bring significant change in radical production and, thus, leads to a nearly constant opt (as shown in Fig. 8b).Results from Fig. 7 Fig. 8 provide reliable evidence of the value of opt anticorrelates with the radical production rate.From the definition of , we notice that the indicator value greatly relates to NO/NO 2 ratio when the sum concentration of NO x is fixed.Therefore, we also calculate opt at different NO/NO 2 ratio, as shown in Fig. 9.The value of opt is about 6 at low NO/NO 2 ratio (NO/NO 2 =0.01), and the corresponding maximum P (O 3 ) (P (O 3 ) max ) is very low due to the high NO 2 which can remove OH radical from the system.At high NO/NO 2 ratio (NO/NO 2 =100), the values of opt and P (O 3 ) max increase to 15 and 7.5 ppbv/hr, respectively.As represented in Fig. 9, it is clear that the decrease of NO/NO 2 ratio leads to opt decrease, which can be well-explained by relevant chemistry of NO x .In the fast photochemical cycling of NO x in daytime, the NO/NO 2 ratio is determined by the concentrations of O 3 and peroxy radicals as well as j (NO 2 ) (Cadle et al., 1952;Leighton et al., 1961;Crawford et al., 1996): where k 1 , k 2 and k 3 are the rate constant of the reaction of NO with O 3 , HO 2 and RO 2 , respectively.j (NO 2 ) is the photolysis rate of NO 2 .Equation ( 6) is proven to be suitable for remote locations (McFarland et al., 1978;Ritter et al., 1979;Fehsenfeld et al., 1983;Parrish et al., 1986;Trainer et al., 1987).Indubitably, an increase of NO/NO 2 ratio associates www.atmos-chem-phys.net/10/9579/2010/Atmos.Chem.Phys., 10, 9579-9591, 2010  with the decrease of peroxy radicals which indicates a lower opt , in accordance with our model results.As mentioned above, the value of opt is influenced by the local atmospheric composition of VOCs, ozone concentration, relative humidity, photolysis rate and NO/NO 2 ratio.However, opt varies within a range of 5∼15 in our case studies, indicating that is robust against those parameters and can be used as an effective indicator to investigate the sensitivity of ozone production to NO x and HC.In this paper, the variation of values at the border between Regime III and IV is not discussed, for that a reduction of NO x is always useful to decrease ozone formation in this region.

Behaviour of P (O 3 ) in different regimes
The ozone production rate is estimated additionally using the peroxy radicals derived from the model using Eq.(2).Basically, the modelled P (O 3 ) (P (O 3 ) mod ) track the diurnal and day-to-day variation of the observed P (O 3 )(P (O 3 ) obs ) well (Fig. 10).A detailed comparison between them is shown in Fig. 11.The P (O 3 ) obs /P (O 3 ) mod ratio is clearly higher than one unit when is lower than 10 (Regime II) under high j (O 1 D) condition (j (O 1 D)>10 −6 ), where the corresponding values of dlnP (O 3 )/dln[HC] are commonly higher than 0.8.It implies that the model tends to underestimate P (O 3 ) when P (O 3 ) is greatly sensitive to HC, which indicates that an important source of peroxy radicals may be missed, as men-   tioned by Qi et al. (2007).This behaviour reveals that the observed P (O 3 ) as well as peroxy radicals used to estimate the value of P (O 3 ) are significantly higher than the model results under high NO x conditions, which is consistent with several previous studies in different environments (Tan et al., 2001;Martinez et al., 2003;Ren et al., 2005).On the contrary, the underestimated trend becomes weaker with increasing in Regime IV where P (O 3 ) is insensitive to HC.In this region, the value of P (O 3 ) obs /P (O 3 ) mod is commonly lower than one unit under a low j (O 1 D) condition (j (O 1 D)<10 −6 ), implying that our model may overestimate the production of peroxy radicals which is not formed by photochemistry.
Figure 12 shows the calculated itemization of P (O 3 ) in different regimes.The average P (O 3 ) is 3.22 ppbv/hr in Regime II, 2.07 ppbv/hr in Regime III and 0.78 ppbv/hr in Regime IV, respectively, which illuminates a decrease of P (O 3 ) with increasing .In all regimes, P (O 3 ) is dominated by HO 2 +NO reaction and followed by MO 2 +NO reaction.These two reactions contribute ca.69% and 16% in Regime II, 67% and 17% in Regime III, 66% and 19% in Regime IV, respectively.Reactions of other peroxy radicals with NO contribute very little.As increases, the contribution ratio of HO 2 +NO reaction to the sum of RO 2 +NO  reactions decreases, this indicates a decrease of the efficiency of the RO 2 to HO 2 conversion via the reaction of RO 2 with NO.The percent contributions of all RO 2 +NO reactions show an increasing trend when shifts from Regime II to IV, except for an abnormally high contribution of TERP+NO in Regime II and ISOP+NO in Regime III, leading by the high OH reactivity towards monoterpenes (L(TER)) and isoprene (L(ISO)) in the corresponding regimes.The values of L(TER), NO concentration and the average P (O 3 ) formed by monoterpenes+NO reactions in Regime II are higher than these in Regime IV by a factor of 1.16, 14.25 and 5.76, implying the effect of NO on P (O 3 ) is weaker in Regime II than in Regime IV.For isoprene, the corresponding values of L(ISO), NO and the average P (O 3 ) in Regime III are higher than those in Regime IV by a factor of 1.36, 3.45 and 2.93, revealing that the effect of NO on P (O 3 ) in Regime III is weaker than in Regime IV.The discussion above reveals that the ozone production rate is more sensitive to NO in Regime IV than in Regime II and III.
As Fig. 12 shows, the 10-min average data of P (O 3 ) are categorized into three classes by j (O 1 D) (s −1 ) value: (1) j (O 1 D)<10 −6 s −1 (denoted as J1), (2) 10 −5 s −1 > j (O 1 D)>10 −6 s −1 (denoted as J2) and (3) j (O 1 D)>10 −5 s −1 (denoted as J3) in each regimes.In Regime II, the average value of P (O 3 ) is 0.28 ppbv/hr at J1, 0.74 ppbv/hr at J2 and 2.20 ppbv/hr at J3, respectively.As j (O 1 D) increases, P (O 3 ) in Regime II increases greatly, which is possibly because of the increasing concentration of peroxy radicals.This indicates that the concentration of peroxy radicals is an important factor in controlling ozone production in Regime II, thus, P (O 3 ) is sensitive to L(HC) as well as HC.In Regime III and Regime IV, the average value of P (O 3 ) is 0.21 and 0.06 ppbv/hr at J1 level, 0.53 and  0.20 ppbv/hr at J2 level and 1.33 and 0.51 ppbv/hr at J3 level, respectively.In these two regimes, the increase of P (O 3 ) (especially in Regime IV) caused by increasing j (O 1 D) is clearly less than in Regime II.The effect of peroxy radicals to control O 3 production becomes weaker with increasing , which implies a declining sensitivity of P (O 3 ) to HC.In the condition of J1 in Regime IV, the percentage contribution of monoterpenes+NO reactions to P (O 3 ) is clearly higher than in other conditions, which possibly associates with the elevated monoterpenes in this region.In such a case, the monoterpenes+O 3 reactions may be significant for the production of peroxy radicals.As these peroxy radicals cannot efficiently convert to HO 2 radical under low NO x condition, it leads to a low percentage contribution of HO 2 +NO reaction to ozone production (as shown in Fig. 12).Meanwhile, a low P (O 3 ) obs /P (O 3 ) mod ratio found in this region (Fig. 11) implies that the reactions of monoterpenes with ozone may over-predict the yield of peroxy radicals.The island is located in the sea of Japan and is more free from impact of human activities.However, the plants are abundant over the island, thus, the emission of BVOCs such as isoprene and monoterpenes can be significant (Tanimoto et al., 2000;Kanaya et al., 2002aKanaya et al., , 2002b)).In order to find the controlling factor for ozone production, it is necessary to analyse P (O 3 ) sensitivity to BVOCs and NO x .Firstly, sensitivity runs for P (O 3 ) are performed by changing NO x and monoterpenes concentrations.In the calculations, ozone

Fig. 3 .
Fig. 3. Dependence of P(O3) on the indicator Φ for daytime 10-min averaged data during 18-21 September, 2003.The color of the dots shows the values of j(O 1 D).The black curve shows the modeled trend of P(O3) with increasing Φ at 1150 JST of 18 September.

Fig. 3 .
Fig. 3. Dependence of P (O 3 ) on the indicator for daytime 10min averaged data during 18-21 September 2003.The colour of the dots shows the values of j (O 1 D).The black curve shows the modelled trend of P (O 3 ) with increasing at 11:50 JST of 18 September.

Fig. 5 .
Fig. 5. Dependence of (a) dlnP (O 3 )/dln[NO] (shown by red dot) and dlnP (O 3 )/dln[HC] (shown by open blue circle) and (b) L N /Q on for daytime 10-min average data during 18-21 September 2003.The ranges of two borders are also labelled in this figure. opt (

Fig. 6 .
Fig. 6.Detailed comparison of the changes of Φopt (ΔΦopt) due to the increase of VOCs concentrations at 1150 JST of 18 September.TER: monoterpenes.

Fig. 6 .
Fig. 6.Detailed comparison of the changes of opt ( opt ) due to the increase of VOCs concentrations at 11:50 JST of 18 September.TER: monoterpenes.
Fig. 7. Dependence of P(O3) on Φ when the concentration of O3 is constrained to 1 ppbv (shown by square), 5 ppbv (shown by upward pointing triangle), 10 ppbv (shown by downward pointing triangle), 50 ppbv (shown by asterisk) and 100 ppbv (shown by dot), respectively.The photolysis rates are constrained to those observed at (a) 1150JST and (b) 0600JST of 18 September.The NO/NO2 ratio and other conditions are constrained to those observed at 1150JST of 18 September.The color of the dots denotes the mixing ratio of OH radical.

Fig. 7 .
Fig. 7. Dependence of P (O 3 ) on when the concentration of O 3 is constrained to 1 ppbv (shown by square), 5 ppbv (shown by upward pointing triangle), 10 ppbv (shown by downward pointing triangle), 50 ppbv (shown by asterisk) and 100 ppbv (shown by dot), respectively.The photolysis rates are constrained to those observed at (a) 11:50 JST and (b) 06:00 JST of 18 September.The NO/NO 2 ratio and other conditions are constrained to those observed at 11:50 JST of 18 September.The colour of the dots denotes the mixing ratio of OH radical.

Fig. 8 .
Fig. 8. Dependence of P(O3) on Φ when the relative humidity is constrained to 10% (shown by square), 20% (shown by triangle), 50% (shown by asterisk) and 90% (shown by dot), respectively.The photolysis rates are constrained to those observed at (a) 1150JST and (b) 0600JST of 18 September.The NO/NO2 ratio and other conditions are constrained to those observed at 1150JST of 18 September.The color of the dots denotes the mixing ratio of OH radical.

Fig. 8 .
Fig. 8. Dependence of P (O 3 ) on when the relative humidity is constrained to 10% (shown by square), 20% (shown by triangle), 50% (shown by asterisk) and 90% (shown by dot), respectively.The photolysis rates are constrained to those observed at (a) 11:50 JST and (b) 06:00 JST of 18 September.The NO/NO 2 ratio and other conditions are constrained to those observed at 11:50 JST of 18 September.The colour of the dots denotes the mixing ratio of OH radical.

Table 1 .
Summary of observed P (O 3 ) and selected chemical species and parameters (1σ standard deviation given in brackets).
x .In Regime II, there is a less-than-linear increase in P (O 3 ) with HC and a less-than-linear decrease with NO x .P (O 3 ) less-than-linearly increases with both HC and NO x in Regime III, and near linearly increases with NO x and is nearly constant with increasing HC in Regime IV.Comparing the results with previous