References

Stratospheric and upper tropospheric air samples were collected during 1994-2004 over Sanriku, Japan and in 1997 over Kiruna, Sweden. Using these archived air sam- ples, we determined the triple oxygen-isotope composition of stratospheric CO2 and the N2O mixing ratio. The maxi- mum 1 17 OCO2 value of +12.2‰, resembling that observed previously in the mesosphere at 60 km height, was found in the middle stratosphere over Kiruna at 25.6 km height, suggesting that upper stratospheric and mesospheric air de- scended to the middle stratosphere through strong downward advection. A least-squares regression analysis of our ob- servations on a 18 OCO2 - 17 OCO2 plot (r 2 >0.95) shows a slope of 1.63±0.10, which is similar to the reported value of 1.71±0.06, thereby confirming the linearity of three iso- tope correlation with the slope of 1.6-1.7 in the mid-latitude lower and middle stratosphere. The slope decrease with in- creasing altitude and a curvy trend in three-isotope corre- lation reported from previous studies were not statistically significant. Using negative linear correlations of 1 17 OCO2 and 18 OCO2 with the N2O mixing ratio, we quantified triple oxygen-isotope fluxes of CO 2 to the troposphere as +48‰ GtC/yr (1 17 OCO2 ) and +38‰ GtC/yr ( 18 OCO2 ) with 30% uncertainty. Comparing recent model results and ob- servations, underestimation of the three isotope slope and the maximum 1 17 OCO2 value in the model were clarified, sug- gesting a smaller O2 photolysis contribution than that of the model. Simultaneous observations of 18 OCO2 , 17 OCO2 , and N2O mixing ratios can elucidate triple oxygen isotopes in CO2 and clarify complex interactions among physical, chem- ical, and photochemical processes occurring in the middle atmosphere.

Tropospheric CO 2 has an almost constant δ 18 O value (∼+41‰) with a small 17 O anomaly which results from rapid oxygen isotope exchange between tropospheric CO 2 and surface water such as leaf water and seawater (Thiemens et al., 1991;Ciais et al., 1997;Hoag et al., 2005).In contrast, anomalously high δ 18 O values and a positive 17 O anomaly have been observed in stratospheric CO 2 over Texas and New Mexico (Thiemens et al., 1991), in southern high-latitude regions (Thiemens et al., 1995a), over Sanriku in Japan (Gamo et al., 1989;, 1995;Aoki et al., 2003;Kawagucci et al., 2005), over Kiruna in Sweden (Alexander et al., 2001;Lämmerzahl et al., 2002), Aire-sur-l'Adour in France (Lämmerzahl et al., 2002), in northern high-latitudinal regions (Boering et al., 2004), and also in the upper stratosphere and lower mesosphere over New Mexico (Thiemens et al., 1995b).The 17 O anomaly observed for middle atmospheric (stratospheric and mesospheric) CO 2 is generally thought to be derived from a S. Kawagucci et al.: Oxygen isotope anomaly in stratospheric CO 2 series of chemical reactions of which O 3 formation is the principal source of extra heavy δ 18 O and 17 O (Mauersberger, 1987;Yung et al., 1991;Lacoursiere et al., 1999;Perri et al., 2003): In addition to those field observations, several laboratory experiments and model calculations have been conducted to elucidate detailed mechanisms of mass-independent oxygen isotope fractionation of CO 2 , O 3 , and O 2 in the middle atmosphere (Wen and Thiemens, 1993;Johnston et al., 2000;Chakraborty and Bhattacharya, 2003;Liang et al., 2007;Shaheen et al., 2007).In addition, the flux of oxygen anomaly in CO 2 from the stratosphere to the troposphere is an important parameter for the tropospheric CO 2 budget, which is useful to investigate carbon circulation between the atmosphere and the biosphere (Cuntz et al., 2003;Boering et al., 2004;Hoag et al., 2005).
Although several observational studies have revealed anomalous oxygen isotopic compositions in stratospheric CO 2 , few samples or parameters have been obtained for each observation, rendering further detailed quantitative analyses difficult.For example, some reports have presented results from fewer than 10 stratospheric air samples (Thiemens et al., 1991(Thiemens et al., , 1995a;;Alexander et al., 2001).In more extensive studies, interesting correlations have been reported for the lower stratosphere of δ 18 O CO 2 -δ 17 O CO 2 (Lämmerzahl et al., 2002), δ 18 O CO 2 -N 2 O mixing ratio (Aoki et al., 2003), and 17 O CO 2 -N 2 O mixing ratio (Boering et al., 2004).However, no report has described a comprehensive dataset to elucidate these unique relations in the same observation, which hinders further comparison of direct data to modeling studies.
The correlation plot between δ 18 O and δ 17 O is useful to discuss mass-independent isotope fractionation processes because vertical deviation from a terrestrial fractionation line (TFL: slope=0.516)corresponds to the 17 O anomaly.Several previous studies of the triple oxygen isotope composition of middle atmospheric CO 2 have revealed a linear correlation between δ 18 O and δ 17 O.However, slopes of the linear correlation lines and the degree of the mutual relation differ among studies.For example, Lämmerzahl et al. (2002) reported a close linear relation with a slope of 1.71±0.06(2σ ) in the lower/middle stratosphere air over Aire-sur-l'Adour (44 • N) and Kiruna (68 • N); in contrast, Thiemens et al. (1995b) reported a slope of 1.18±0.17for the upper stratosphere/lower mesospheric air over New Mexico (32 • N).This difference in slope was interpreted as a gradual decrease in slope occurring concomitantly with increasing altitude (Kawagucci et al., 2005;Liang et al., 2007), although such a change in slopes has not been observed to date.
As described in this paper, we report triple oxygen isotope compositions of CO 2 and its relation with the N 2 O mixing ratio of the middle-latitude lower/middle stratospheric CO 2 samples collected during 11 years' air sampling over Sanriku, Japan and Kiruna, Sweden.The long-term stratospheric air sampling program over Sanriku was started in 1985 (Gamo et al., 1989).It continued during 1994-2004, while additional stratospheric air sampling was carried out over Kiruna in February 1997 within the polar vortex.Oxygen isotope fluxes in CO 2 from the stratosphere to the troposphere and triple oxygen isotope fractionation processes in the middle atmosphere are discussed using the comprehensive dataset of triple oxygen isotope composition of stratospheric CO 2 and the mixing ratio of N 2 O.

Sampling and sample storage
Stratospheric whole air samples were collected during six balloon flights over Sanriku, Japan (39 • N, Table 1) during 1994-2004 using a balloon-borne cryogenic sampler developed by the Japan Aerospace Exploration Agency (JAXA).Using the same sampling system, additional vertical sampling was carried out at Kiruna, Sweden (68 • N) on 22 February 1997 (Table 1).Details of our air sampler and sampling procedure have been reported elsewhere (Nakazawa et al., 1995;Honda et al., 1996;Aoki et al., 2003;Honda, 2001).Herein, we describe them briefly.The air sampling system can collect multiple whole-air samples into pre-evacuated 760 cm 3 stainless steel sample bottles.Chemical and isotopic compositions in the sampled air are only slightly affected by the collection process (Honda et al., 1996;Honda, 2001).Preliminary experimental tests showed that a possible change of δ 18 O CO 2 caused by isotope exchange between CO 2 and water on the inner walls of the sample bottle is comparable to our analytical precision of 0.05‰ (Honda, 2001;Gamo et al., 1995;Aoki et al., 2003).For 17 O CO 2 , such preliminary experimental tests were not carried out, but we consider that this sampling process also provides results within our analytical precision of 0.5‰, as discussed previously (Boering et al., 2004).
In the laboratory, each whole air sample in the stainless steel bottle of the sampler was divided into several aliquots.From each aliquot, CO 2 was extracted and separated from water vapor and other components using cryogenic trapping, which is commonly used in CO 2 isotope analysis.The separated CO 2 gas was then sealed into a glass ampoule that had been combusted at 400 • C and kept in a desiccator to avoid contamination by water and organic matter.These CO 2 samples had been stored for more than a decade before current isotope analysis.Although the storage period is quite long, the change in δ 18 O CO 2 is expected to be negligible because very little water remains in the ampoules.After 2001, we changed the sample storage method.Aliquots of the whole-air samples were transferred directly into preevacuated 1000 cm 3 stainless steel gas canisters at a positive pressure of 2 atm without CO 2 separation.They were stored for more than three years before analyses.It is noteworthy that the quality of our whole-air sampling and handling procedures was verified by concurrent analyses of multiple compositions, such as concentrations of CO 2 , N 2 O, and CH 4 , and δ 13 C CO 2 .The sample air was regarded as defective and eliminated from this study when an apparent anomaly was observed in any chemical or isotope signature.

Oxygen isotope analysis using CF-IRMS
Both δ 18 O CO 2 and δ 17 O CO 2 values of the stratospheric CO 2 were analyzed simultaneously using a new continuous flowisotope ratio mass spectrometry (CF-IRMS) technique.Details of the analytical method and δ 17 O calculation were reported in Kawagucci et al. (2005).Herein, we describe them briefly.The CO 2 and other molecules in a sample gas are separated using gas chromatography with helium as the carrier gas.After separation, CO 2 is split and analyzed on-line using two procedures: 1) direct introduction into the IRMS (Finnigan MAT252) for isotope analysis, and 2) eliminating the 17 O anomaly in the CO 2 by exchanging the oxygen atoms with high-temperature CuO, with subsequent isotope analysis using IRMS.This CF-IRMS measurement yield two isotope values for "raw" CO 2 and "exchanged" CO 2 and results in four isotope ratios of 45/44 raw , 46/44 raw , 45/44 ex , and 46/44 ex .The δ 17 O value is calculable accurately using these measured isotope ratios and several constants of λ=0.516 (Santrock et al., 1985), K=0.0099235 (Santrock et al., 1985), 13 R VPDB =0.0112372 (Craig, 1957), and 18 R VSMOW =0.0020052 (Baertschi, 1976).The total analytical errors of this method in determining δ 18 O CO 2 , δ 17 O CO 2 , and 17 O CO 2 values are estimated respectively as, at most, ±0.05‰, ±0.5‰, and ±0.5‰.The N 2 O mixing ratio was also determined using a gas chromatographic technique with analytical precision of 2 ppbv using another aliquot from the same stratospheric air sample (Aoki et al., 2003).
The measurements over Sanriku revealed a nearly linear altitude dependence of 17 O CO 2 .Some deviations from the linearity might reflect dynamic variation in the stratosphere.The vertical distribution of 17 O CO 2 over Kiruna exhibited a different profile: 17 O anomalies higher than 20 km height increased more quickly with increasing altitude than those lower than 20 km height, which agrees well with the results of previous studies of the stratosphere over Kiruna (Alexander et al., 2001;Lämmerzahl et al., 2002).In particular, large 17 O CO 2 values (>+5‰) even in the lower/middle stratosphere (20-26 km), observed in this study and by Lammerzahl et al. (2002), are noteworthy.The characteristic vertical profile of 17 O CO 2 over Kiruna resembles the typical one observed for long-lived tracers, such as N 2 O and CH 4 mixing ratios within a polar vortex: Upper stratospheric and mesospheric air descends to the lower/middle stratosphere through strong downward advection (Waugh and Hall, 2002), resulting in upper stratospheric and mesospheric characteristics observed even in the lower/middle stratosphere.Indeed, 17 O CO 2 at 25.6 km height in our observation over Kiruna reached +12.2‰ (Table 1, Fig. 1), which is significantly greater than those over Sanriku at a similar height and which is as large as the highest value previously observed at 60 km height over New Mexico (Thiemens et al., 1995b).That result suggests the probable transport of a mesospheric air mass into the middle stratosphere over the Kiruna area.

Triple oxygen isotopic compositions
The triple oxygen isotopic compositions (δ 18 O CO 2 and δ 17 O CO 2 ) determined in this study are depicted in Fig. 2.
We conclude that the slope on the triple oxygen isotope plot within mid-latitude lower/middle stratospheric CO 2 (39-68 • N; <36 km) is always constant for the observation term.In contrast, Thiemens et al. (1995b) reported a slope of 1.18±0.17(r 2 >0.96; n=12) in their observation of both δ 18 O CO 2 and δ 17 O CO 2 values in upper stratospheric/lower mesospheric CO 2 , which is much lower than observations of the lower/middle stratosphere reported herein and by Lammerzahl et al. (2002).Furthermore, ANCOVA tests demonstrate significant differences in the regression between sults reported by Thiemens et al. (1995b)  The observed difference in the slopes implies that δ 18 O CO 2 and δ 17 O CO 2 values might not have a simple linear relation, but rather an altitude-dependent difference by which the slopes become smaller as altitude increases (Kawagucci et al., 2005).Liang et al. (2007) also reported a gradual slope decline with increasing altitude using a theoretical model that incorporated a considerable impact of O 2 photolysis to O( 1 D) in the mesosphere.Kawagucci et al. (2005) reported that the slopes in the δ 18 O CO 2 and δ 17 O CO 2 correlation line seem to be different for altitudes greater than and less than 25 km over Sanriku, although the difference was not statistically significant.Using the current dataset over Sanriku, which includes more data than reported in Kawagucci et al. (2005), the separate linear fit for the lower division (<25 km) yields a slightly larger slope of 1.60±0.13(r 2 >0.88; n=22) than that for the higher division (>25 km) of 1.46±0.11(r 2 >0.90; n=21).However, ANCOVA tests for these slopes reveal no significant difference at a 95% confidence level (F (1,39) =0.71; p<0.05).In addition, a leastsquares quadratic fit for our stratospheric dataset showed a correlation coefficient of r 2 >0.95, which is the same as that for a linear fit (r 2 >0.95), meaning that it is not possible to identify the gradual slope decline in the lower/middle stratosphere from our dataset.For those reasons, it is difficult to prove the gradual slope decline in the stratosphere, even from our dataset.Further observation of the triple oxygen isotopes in the upper stratosphere/lower mesosphere is necessary to prove the gradual decline in the slope.2004).The terrestrial fractionation line (TFL: slope=0.516)and also an isoclinic line for 17 O CO 2 =+12.2‰ that marks the maximum value of the atmospheric observations are shown respectively as a bold solid line and a dashed line.Thin solid and dotted lines respectively represent least square linear fits for our dataset and the dataset described by Thiemens et al. (1995b).A gray solid curve represents the model simulation result (from Fig. 5 in Liang et al., 2007) for comparison.

Correlation between N 2 O mixing ratio and oxygen isotopes in CO 2
The physical processes in the stratosphere that are responsible for the characteristic spatial distribution of 17 O CO 2 in both Sanriku and Kiruna (Fig. 1) can be studied by plotting 17 O CO 2 against the N 2 O mixing ratio (Boering et al., 2004).The present data, except for the five tropospheric samples and the sample showing the largest 17 O anomaly of +12.2‰ over Kiruna (Table 1), exhibited a linear negative correlation between 17 O CO 2 and N 2 O mixing ratio (r 2 >0.90) (Fig. 3a), with some deviations from the fit line which are larger than the analytical errors.The linear correlation suggests that the relative reaction rates between the heavy oxygen isotope accumulation in CO 2 (R1-R2) and photochemical N 2 O destruction are almost uniform in the lower and middle stratosphere.On the other hand, the sample showing the largest 17 O anomaly of +12.2‰ in the middle stratosphere over Kiruna (Table 1, Fig. 1) deviates from the general linear correlation of N 2 O-17 O CO 2 (Fig. 3a).It overlaps with the values from the mesosphere (Thiemens et al., 1995b), suggesting its mesosphere origin, as discussed above.In addition to 17 O CO 2 , δ 18 O CO 2 showed a negative linear correlation with the N 2 O mixing ratio (Fig. 3b), as presented in Aoki et al. (2003).Deviations comparable to those in the N 2 O-17 O CO 2 correlation were also observed in the N 2 O-δ 18 O CO 2 correlation, although the degree of the deviations was much smaller than for those described in Boering et al. (2004) (Fig. 3).The observed deviations from the linear correlations in the plots of N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 could reflect either some changes in the major natural (photo)chemical processes in the stratosphere, in which the relative destruction rate of N 2 O against the heavy oxygen isotope accumulation rate in CO 2 (R1-R2) have varied, or some additional isotope fractionation processes for CO 2 other than the Reactions (R1)-(R2).The N 2 O destruction rate is almost entirely (∼90%) governed by UV radiation, although the accumulation rate of heavy oxygen isotopes in CO 2 (R2) is controlled by the O( 1 D) concentration governed by both the O 3 concentration and its photolysis rate (R1a).Therefore, independent changes in the reaction rates in response to variations in stratospheric chemical conditions, such as abundances of O 3 and O( 1 D) that would control the accumulation rate with no relation to N 2 O destruction, for example, might cause deviation from linear N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 corre- lation.However, isotope fractionation processes other than the (R1)-(R2) reactions, most of which fractionate under the mass-dependent relation that is usual for almost all physical and chemical processes (e.g.Thiemens et al., 1999), cause changes only in the N 2 O-δ 18 O CO 2 correlation.Consequently, a comparison of the deviations between the N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 correlations is useful to examine potential additional isotope fractionation processes.For this purpose, we introduce new parameters, VDF-17 O and VDF-δ 18 O, which correspond to the vertical deviation of each datum from the fit lines of either N 2 O-17 O CO 2 or N 2 O-δ 18 O CO 2 plot (Fig. 3) using the following equations: Calculated VDF values for stratospheric samples are within ±2‰, as presented in Fig. 4, although the VDF values for the sample showing the largest 17 O anomaly of +12.2‰ are out of that range (>+5‰), indicating a distinguishable origin as discussed above.By plotting VDF-17 O against VDF-δ 18 O, we can distinguish whether those deviations are explainable according to changes in the relative destruction rate of N 2 O or not.That is to say, if the relative reaction rate change were the only process responsible for the deviations, the correlation between the VDF-δ 18 O and VDF-17 O values would be expected to be linear with a slope of +1.1-1.2 (Fig. 4), as calculated from the constant fractionation slope of +1.6-1.7 between δ 18 O CO 2 and δ 17 O CO 2 (Fig. 2) and the 17 O definition of 17 O=δ 17 O-0.516×δ 18O.Although the observed VDF-17 O values exhibit a correlation with the VDF-δ 18 O values, showing a positive slope of +0.6 (Fig. 4), the linearity is not good (r 2 >0.31).Therefore, we conclude that, aside from changes in the relative reaction rate in the stratosphere that produce the slope of +1.1-1.2, some additional isotope fractionation processes having different slopes on the plot between the VDF values, such as the slope of 0 (mass-dependent fractionation processes), were involved in the observed isotopic compositions of CO 2 as well.
One such additional isotope fractionation process is, unfortunately, sample alteration during sampling and storage despite careful handling.For example, oxygen isotope exchange with stratospheric H 2 O, which includes highly 18 Odepleted δ 18 O from CO 2 (Johnson et al., 2001), that takes place in canisters reduces δ 18 O CO 2 values, with little accompanying 17 O CO 2 variation (Boering et al., 2004).In this case, the maximum extent of δ 18 O change can be estimated as 1.5‰ from Fig. 4.That is to say, we can obtain linear correlation of the slope of 1.1-1.2 on the plot between VDFδ 18 O and VDF-17 O, by adding, at most, 1.5‰ correction on VDF-δ 18 O.Therefore, we adopt the value for the maximum extent of errors in δ 18 O CO 2 for subsequent discussions.

Net oxygen isotope flux to the troposphere
The linear correlations of 17 O CO 2 and δ 18 O CO 2 with the N 2 O mixing ratio enable us to quantify net isotope fluxes of 17 O and δ 18 O in CO 2 from the stratosphere to the troposphere.A slope of the fitting line between two longlived tracers is known to be equal to the ratio of their net vertical fluxes (Plumb and Ko, 1992); furthermore, the net vertical flux of N 2 O is estimated as the global N 2 O loss rate of 13 MtN/yr with uncertainty of ±25% (Prather and Ehhalt, 2001).Using the dataset in the stratosphere (excluding five tropospheric and one mesosphere-derived samples)(Table 1), least-squares linear fitting is applied to correlation plots of N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 .The fittings yielded slopes of −0.0234±0.0022(‰/ppbv, r 2 >0.90) and −0.0185±0.0022(‰/ppbv, r 2 >0.85), respectively, for N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 .The best fit lines are shown as solid lines with the formulas presented in Fig. 3a  and b.Assuming that the CO 2 mixing ratio in the stratosphere during our observation period is constant (370 ppmv), the resulting net isotope fluxes of 17 O CO 2 and δ 18 O CO 2 are estimated as +48‰ GtC/yr and +38‰ GtC/yr, respectively.We estimate that the calculated isotope fluxes might include ∼30% uncertainty because of the uncertainty in the estimated N 2 O loss rate used here, uncertainty in the least-squares linear fitting, and the assumption of a constant CO 2 mixing ratio.The isotope flux of 17 O CO 2 estimated in this study resembles that reported by Boering et al. (2004).

Comparison of current observation and model calculation
Liang et al. (2007) first investigated the considerable contribution of O 2 photolysis in the upper mesosphere to the oxygen isotope composition of middle atmospheric CO 2 .3a) (see Liang et al., 2007).
On the δ 18 O CO 2 -δ 17 O CO 2 plot (Fig. 2), the slope of the solid curve shows good fit for the lower stratosphere, but it declines gradually and underestimates 17 O CO 2 concomitantly with increasing altitude (increasing δ 18 O CO 2 ) in the model (Fig. 2) in comparison to observations from various latitudes and altitudes (Thiemens et al., 1995b;Lammerzahl et al., 2002;this study).In particular, the maximum 17 O CO 2 value in the simulation result, which corresponds to ca. 55 km height, does not reach that in the observation results (isoclinic line in Fig. 2).Furthermore, on the N 2 O-17 O CO 2 plot, the change of the eddy diffusion constant from the original value (solid curve) to the increased value (dashed curve) results in an excellent fit between the model and observation at a range of [N 2 O]>∼50 ppbv, corresponding to the lower/middle stratosphere, although the discrepancy in the maximum 17 O CO 2 values between the model and observation is more significant.The cause of the underestimation of the maximum 17 O CO 2 in the model will be discussed later.In addition to that disagreement, at a range of [N 2 O]<∼50 ppbv, the observation result over Sanriku demonstrates smaller 17 O anomalies in CO 2 compared to the model simulation result.The wide ranges of the observed 17 O CO 2 values at the range of [N 2 O]<∼50 ppbv would be informative for middle/upper stratospheric chemistry and physics, although only a fragmentary dataset of chemical and isotopic compositions in the middle/upper stratosphere is available for comparison to those in the lower/middle stratosphere ([N 2 O]>ca.50 ppbv).Additional intensive observation throughout the whole stratosphere will elucidate the complex chemistry and air transport in the upper stratosphere.
Oxygen isotope fractionation in middle atmospheric CO 2 occurs through isotope exchange between CO 2 and O( 1 D) produced by photolysis of O 3 and O 2 (Yung et al., 1991;Liang et al., 2007).The possibility of oxygen isotope fractionation from the CO 2 photodissociation reaction has also been proposed (Bhattacharya et al., 2000), although the actual contribution of the isotope fractionation was considered to be negligible throughout the middle atmosphere (Liang et al., 2007).Oxygen isotope fractionation in the series of reactions of CO 2 with O( 1 D) through photolysis of O 3 (R1a-R2), which dominates in the stratosphere, engenders a steep slope of ∼1.7, as observed in the lower/middle stratospheric CO 2 (Fig. 2).This engenders the strengthening 17 O anomaly with increasing altitude.On the other hand, Liang et al. (2007) pointed out that photolysis of O 2 might dominate O( 1 D) production in the upper mesosphere and that the three isotope slopes of the Reactions (R1b)-(R2) in this region might be as low as 0.3.Consequently, 17 O CO 2 will decrease again with increasing altitude if this pattern pertains in the upper mesosphere (Figs. 2 and 3a).Therefore, we conclude herein that the underestimation of maximum 17 O CO 2 value (Figs. 2 and 3a) in the model simulation implies some problems related to the treatment of O 2 photolysis (R1b-R2) in the model.A possible reason is overestimation of the contribution of O 2 photolysis in the model.In this case, changing the parameters in the model would yield a better fit of the simulated with the observed data without changing the underlying theory of the model.

Concluding remarks
Decade-long observations of triple oxygen isotope composition in stratospheric CO 2 over Sanriku and Kiruna within the polar vortex revealed the following: 1. Observed δ 18 O CO 2 and δ 17 O CO 2 showed a good linear correlation (r 2 >0.95) with a slope of 1.63±0.10(2σ ), which is consistent with that observed in the middlelatitude lower/middle stratosphere (Lammerzahl et al., 2002).However, it differs clearly from that observed in the upper stratosphere/lower mesosphere (Thiemens et al., 1995b).The reason for the difference in the slopes remains unknown.
2. Both N 2 O-17 O CO 2 and N 2 O-δ 18 O CO 2 plots showed a negative linear correlation, as previously reported respectively by Boering et al. (2004) and Aoki et al. (2003).Based on those linear relations, CO 2 oxygen isotope fluxes from the stratosphere to the troposphere are estimated as +48‰ GtC/yr ( 17 O CO 2 ) and +38‰ GtC/yr (δ 18 O CO 2 ) with ∼30% uncertainties.Results of the current study show that long-term observations of δ 18 O CO 2 , δ 17 O CO 2 , and N 2 O mixing ratios are informative when combined with a model study that particularly addresses correlations of N 2 O-17 O CO 2 and δ 18 O CO 2δ 17 O CO 2 .These results can elucidate triple oxygen isotope fractionations in CO 2 and help to untangle complex interactions among physical, chemical, and photochemical processes in the middle atmosphere. Fig.1

Fig. 2 .
Fig. 2. δ 18 O CO 2 -δ 17 O CO 2 plot.Symbols are similar to those used for Fig. 1.Filled squares (BO04) represent datasets obtained fromBoering et al. (2004).The terrestrial fractionation line (TFL: slope=0.516)and also an isoclinic line for 17 O CO 2 =+12.2‰ that marks the maximum value of the atmospheric observations are shown respectively as a bold solid line and a dashed line.Thin solid and dotted lines respectively represent least square linear fits for our dataset and the dataset described byThiemens et al. (1995b).A gray solid curve represents the model simulation result (from Fig.5inLiang et al., 2007) for comparison.

Fig. 3 .
Fig. 3. N 2 O mixing ratio and (a) 17 O CO 2 or (b) δ 18 O CO 2 plots.Symbols are the same as those shown in Figs. 1 and 2. Solid bold lines represent linear square regression lines for our stratospheric observation, excluding a sample showing 17 O CO 2 =+12.2‰ and δ 18 O CO 2 =+52.49‰,used for isotope flux estimations.Solid and dash curves in Fig. 3a represent model simulation results referred from Liang et al. (2007).

Fig. 4 .
Fig.4 3. A comparison between a recent model study and ob-servations revealed underestimation of the maximum 17 O CO 2 value in the model.One reason might be overestimation of O 2 photolysis considered in the model.

Table 1 .
Measured oxygen isotopic compositions of CO 2 are shown with the sampling location, date, altitude, and N 2 O mixing ratio.A sample denoted with # is regarded as mesosphere-derived air.Five samples marked with * are regarded as tropospheric air.Errors of δ 18 O, δ 17 O, and 17 O are estimated respectively as at most, 0.05‰, 0.5‰, and 0.5‰.
S.Kawagucci et al.:Oxygen isotope anomaly in stratospheric CO 2 portrays altitude profiles of 17 O CO 2 for each balloon launch.The 17 O anomalies reached maximum values of +7.6‰ and +12.2‰, respectively, over Sanriku and Kiruna.For comparison, similar altitude profiles reported in the previous studies are presented in that figure as well They simulated correlation of N 2 O-17 O CO 2 and δ 18 O CO 2δ 17 O CO 2 using a 1D vertical model.That innovative model, however, has not been tested yet, especially in terms of its adequacy for assessing effects of O 2 photolysis on CO 2 .Herein, we verify the theory in Liang's model to compare the simulation results with the observation results.The simulation results are included in the current correlation plots of δ 18 O CO 2 -δ 17 O CO 2 and N 2 O-17 O CO 2 portrayed in Figs. 2 and 3a, for which two different eddy diffusion constants were used for comparison (original values: solid curves in Figs. 2 and 3a; increased by 30% for altitudes of less than 40 km: dashed curve only in Fig.