Spatiotemporal Variability of Oxygen Isotope Anomaly in near Surface Air CO 2 over Urban , Semi-Urban and Ocean Areas in and around Taiwan

The most commonly used tracers to probe the atmospheric and biogeochemical cycles of CO2 are OCO, OCO, and OCO. Considering the number and diversity of sources and sinks affecting CO2, these tracers are not always sufficient to constrain the fluxes of CO2 between the atmosphere and biosphere/hydrosphere. In this context, OCO species was introduced but has rarely been used due to difficulties associated with its accurate measurement in natural samples. This tracer, expressed as an abundance anomaly in O, defined by ∆O = ln(1 + δO) – 0.516 × ln(1 + δO) can independently constrain the fluxes associated with the terrestrial processes. The advantage of utilizing ∆O over δO alone lies on the sensitivity of the former to the rates of biogeochemical processes involving multiple water reservoirs with spatial and temporal isotopic heterogeneities. To employ all the three oxygen isotopes for estimating fluxes of CO2, sources and processes affecting their partitioning have to be identified and quantified. Here, we measured ∆O values in near surface atmospheric CO2 from Taiwan in urban and semi-urban areas and over the South China Sea. Strong spatiotemporal variation was seen, with an average ∆O value of 0.332‰ and a mean variation of 0.043‰ (relative to VSMOW; 1-σ standard deviation for a total of 140 samples). The large variation reflects combinations of distinct air masses carrying CO2 from sources having different ∆O values: negative from combustion emissions, positive from the stratosphere, and a positive water-CO2 equilibration value from isotope exchange with leaf/soil/ocean waters. We observed that the variation of the semi-urban ∆O values is largely affected by local biogeochemistry and stratospheric intrusion with only minor influence from anthropogenic emissions. This is the first oxygen anomaly study for near surface CO2 covering diverse source characteristics and has enormous potential in air CO2 source identification and constraining the global carbon budget.


INTRODUCTION
About 90% of anthropogenic emissions of CO 2 are due to fossil fuel burning and cement manufacturing, while land use change is responsible for the rest (IPCC, 2007(IPCC, , 2013)).Not all the emitted CO 2 remains in the atmosphere.Land and oceans in total absorb more than half of the total emissions (Tans et al., 1990;Gurney et al., 2002;IPCC, 2007;Francey et al., 2013;IPCC, 2013).Due to sparse observations, the partitioning of the uptake between the land and ocean remains poorly constrained and highly debated (Gurney and Eckels, 2011;Wanninkhof et al., 2013).Current understanding of the sources and sinks of CO 2 is primarily based on 13 C (Cias et al., 1995; Francey et al., 1995; Yakir,  2011), 14 C (Levin et al., 2010), 18 O (Farquhar et al., 1993;Ciais et al., 1997;Cuntz et al., 2003b;Welp et al., 2011), clumped isotopes (Affek et al., 2007;Laskar and Liang, 2016;Laskar et al., 2016) and models that assimilate CO 2 observations (Peters et al., 2007;Gurney and Eckels, 2011;Wanninkhof et al., 2013). 13C is used to differentiate the uptake fluxes by land and oceans (Cias et al., 1995;Francey et al., 1995), and 18 O is useful for estimating terrestrial gross primary production (GPP) (Ciais et al., 1997;Farquhar et al., 1993). 17O in near surface air CO 2 has not yet been used widely, mainly because of the measurement difficulty due to its low abundance.Using a recently developed analytical method (Mahata et al., 2013), systematic and precise analysis of 17 O has become possible.In a typical chemical process, the oxygen isotope partitioning of CO 2 isotopologues is expected to follow a mass-dependent line: ln(1 + δ 17 O) = λ × ln(1 + δ 18 O). (1) The coefficient λ varies between 0.500 and 0.529 (Thiemens, 1999;Thiemens et al., 2014) and is taken to be 0.516 (λ o ) for atmospheric CO 2 .This value of λ o is chosen because it corresponds to the fractionation that occurs in transpiration (Landais et al., 2006) at an average relative humidity of 75% (more details about the choice of λ values are given in the Supplementary Material).With this selection of λ, the formulation of the partitioning of the tripleoxygen isotopes of CO 2 coming from exchanges with diverse waters (meteoric, vapor, soil, stem, and leaf) can be simplified since the triple isotope variations in these processes have similar λ values as λ o (Landais et al., 2006;Liang and Mahata, 2015).As a result, the oxygen isotope anomaly as defined below remains unchanged despite isotopic fractionation associated with oxygen transfer.However, it has been discovered that some atmospheric species follow a very different relation called mass independent fractionation, e.g., δ 17 O(O 3 ) ≈ δ 18 O(O 3 ) (relative to atmospheric O 2 ; see, e.g., Mauersberger et al., 2001;Liang et al., 2006;Thiemens, 2006;Barkan and Luz, 2012;Mahata et al., 2012) and δ 17 O(CO 2 ) ≈ 1.7 × δ 18 O(CO 2 ) in the stratosphere (where the δ-value is relative to tropospheric CO 2 ) (Thiemens et al., 1995;Lammerzahl et al., 2002;Liang et al., 2007;Liang et al., 2008;Wiegel et al., 2013).Oxygen isotope anomaly, expressed as ∆ 17 O, is used to quantify the deviation of the δ-values of a sample from the mass-dependent fractionation line and is defined by where the δ-values (and ∆ 17 O values), the isotopic composition of oxygen in CO 2 (or O 2 when referring to atmospheric O 2 ), are expressed in ‰ relative to V-SMOW, following the same definition as Liang and Mahata (2015).Note that we choose value of λ to be 0.516 in Eq. ( 2) but the selection of λ and the reference do not change the interpretation, as long as the same scales are used consistently.∆ 17 O in atmospheric O 2 has been widely used in geochemical modeling (Luz et al., 1999;Luz and Barkan, 2005).The value of ∆ 17 O in tropospheric O 2 is about -0.2‰ (using λ = 0.516; Thiemens et al., 1995;Luz and Barkan, 2005;Mahata et al., 2016) caused by combined effect of O 2 -O 3 -CO 2 photochemistry in the middle atmosphere and photosynthetic production in the biosphere (Juranek and Quay, 2013).The O 2 produced by the biosphere inherits the ∆ 17 O value of water (Luz et al., 1999).The signal produced by changes in biogeochemistry of oxygen and CO 2 has a time scale of the order of millennium and hence ∆ 17 O value of atmospheric O 2 can be used to study biospheric changes in the past thousands of years (Luz et al., 1999).The same can be done for CO 2 , if processes that control ∆ 17 O values in CO 2 are well identified.Hoag and co-authors (Hoag et al., 2005) investigated the contribution of stratospheric CO 2 to the troposphere (Yung et al., 1991;Thiemens et al., 1995;Barth and Zahn, 1997;Yung and Miller, 1997;Liang et al., 2007), and predicted a positive ∆ 17 O of ~0.15‰, above the mean value of emissions from the surface in the troposphere (set by exchange reactions with water bodies; Francey and Tans, 1987;Cuntz et al., 2003a, b;Hoag et al., 2005).The predicted anomaly (≈ 50 PgC year -1 × 1‰/ 400 PgC year -1 = 0.13‰) essentially represents a balance between the stratospheric input flux (~50 PgC year -1 ) with a ∆ 17 O value of ~1‰ (Boering et al., 2004;Liang et al., 2008) and the surface CO 2 oxygen isotope reset flux of ~400 PgC year -1 .The latter gives CO 2 a residence time of ~2 years in the atmosphere.In the present study our aim is to use ∆ 17 O values of atmospheric CO 2 to decipher its biogeochemical cycling in the atmosphere.

Air Sampling and Analytical Methods
Air samples were collected in pyrex flasks equipped with two high-vacuum stopcocks on two ends to facilitate flushing.For isotopic measurements, 1 L and for concentration measurements, 350 ml flasks were used.The flasks were pre-conditioned by passing dry and high purity nitrogen overnight.Two flasks of volumes 1 L and 350 mL, connected in series, were first flushed with the ambient air for 5 minutes at a flow rate of ~2 L min -1 and then compressed with air to ~2 atm pressure using oil free pump.Air moisture was removed during sampling by passing the air through a column of magnesium perchlorate that reduced the moisture content from the ambient value of 70-90% (in terms of relative humidity) to less than 1%, checked using a LI-COR analyzer (model 840A, LI-COR, USA).Air samples were dried during collection to minimize post sampling water-CO 2 exchange which could happen inside the flasks due to condensation of moisture on the wall of the flasks.CO 2 from air was extracted using cryogenic technique.Air was pumped through a series of five coiled traps with first two immersed in dryice-acetone slush (-77°C) for removing water and volatile organics and remaining three in liquid nitrogen (-196°C) for CO 2 collection.The pumping rate was maintained at 100 mL min -1 using a mass-flow controller during the sample extraction.The above process was checked by several control experiments to ensure that there was no escape of CO 2 and isotope fractionation during extraction and purification.
The CO 2 -O 2 oxygen isotope exchange method developed by Mahata et al. (2013) was used to measure the ∆ 17 O values of the tropospheric CO 2 .Briefly, the CO 2 samples (~30 µmole) were made to exchange with equal amount of O 2 (taken from a fixed flask) in presence of platinum in a reaction tube (made of quartz: 60 cm in length and 6.5 mm in diameter) heated by a cylindrical heater.The exchange reaction was carried out for 2 hrs at a temperature of 670°C.After exchange, the O 2 was separated from the CO 2 by cryogenic technique.This O 2 sample was collected in a ~3 mL high vacuum bottle containing a pellet of molecular sieve and analyzed for both 17 O and 18 O.As explained in Mahata et al. (2013) using the δ 18 O value of the CO 2 sample (measured initially) and the pre-exchange and post-exchange δ-values of O 2 , one can derive the ∆ 17 O value of the CO 2 (details are given in Mahata et al., 2013Mahata et al., , 2016)).Isotopic analyses were done using a FINNIGAN MAT 253 stable isotope ratio mass spectrometer in dual-inlet mode at Research Center for Environmental Changes, Academia Sinica, Taiwan.The analytical precision of the derived ∆ 17 O value of a sample of CO 2 was ~0.01‰ (1-σ standard deviation), and the accuracy was also ~0.01‰ (see Liang and Mahata, 2015;Mahata et al., 2016).The uncertainty in δ 13 C (V-PDB) and δ 18 O (V-SMOW) values (obtained in parallel with ∆ 17 O analysis) were 0.06‰ (e.g., see Liang and Mahata, 2015;Laskar and Liang, 2016).Details about the analytical methods, uncertainty and reproducibility of measurements are given in Supplementary Materials.Concentration of CO 2 was measured using a LI-COR infrared gas analyzer (model 840A, LI-COR, USA) at 4 Hz, running averaged with a 20s window, and the reproducibility was better than 1 ppmv.
CO 2 was collected between 2012 and 2014 (mainly from September 2013 through February 2014; see Table 1) at five places: (1) Academia Sinica campus (abbreviated AS; 121°36'51''E, 25°02'25''N; ~10 m above the ground level) in Taipei, Taiwan, in a place where C 3 plants were dominant, (2) the campus of National Taiwan University (NTU; 121°32'E, 25°01'N;~10 m above the ground level; ~10 km southwest of Academia Sinica) over a 60 × 60 m 2 grassland (C 4 plants, surrounded by dense C 3 plants and with intense vehicle emissions nearby), (3) South China Sea (SCS), (4) Daan forest park (C 3 tress and C 4 grass; CO 2 , however, was sampled over a piece of C 4 grassland; ~0.3 × 1 km 2 ) in central downtown, located ~1.5 km north of the NTU sampling site, and (5) Roosevelt road having heavy traffic during the sample collection and located at ~0.5 km west of the NTU station (see Fig. 1 for the geographic locations).These places will henceforth be called: AS, NTU, SCS, Daan and Roosevelt.Till date a total of 171 samples have been analyzed for the ∆ 17 O values in air CO 2 .The present study almost doubles the size of the published data of atmospheric ∆ 17 O (CO 2 ) measurements from Barkan and Luz (2012;n = 5) and Thiemens et al. (2014;n = 180).

RESULTS
Table 1 summarizes the results of the measured stable isotopic compositions including ∆ 17 O values of CO 2 (calculated for two values of λ 0.516 and 0.528 given for comparison) for AS, NTU, SCS, Daan, and Roosevelt.SCS had CO 2 concentration of 395.4 ± 7.3 ppmv, similar to the value reported at Mauna Loa measured by NOAA Earth System Research Laboratory (data available in NOAA website) at the time of sampling.Similarly, the δ 13 C value obtained for SCS was -8.43 ± 0.20‰ (same as that at Mauna Loa).Daan, though located in the center of the city, had second lowest concentration (401.4 ± 24.3 ppmv), next to SCS, among the five stations.One likely reason for getting a low value is that the Daan samples were collected in summer when photosynthetic uptake was high.The concentration was high at Roosevelt, as expected in a street with heavy traffic, where [CO 2 ] was 514.9 ± 41.7 ppmv.The anthropogenic CO 2 signature was also evident from its δ 13 C value (-12.22 ± 1.13‰), significantly lower than that of the other four locations.
Fig. 2 shows the time series of the concentration and isotope ratios of CO 2 at AS and NTU, the two places where most of the samples were collected.The data from these two stations were similar.Overall, the averaged CO 2 concentration ([CO 2 ]) was 417.1 ± 20.2 ppmv, δ 13 C value was -9.07 ± 0.70‰, δ 18 O value was 40.49 ± 0.80‰, and ∆ 17 O value was 0.332 ± 0.043‰.[CO 2 ] varied between ~350 and 475 ppmv, Table 1.Summary of isotopic data of CO 2 samples collected at various locations (see Fig. 1 for the sampling locations).Simple corrections in δ 13 C and δ 18 O values were made to remove the interference of N 2 O.Last two columns show the ∆ 17 O values (VSMOW) using λ = 0.516 and 0.528 respectively.0.528 value refers to the slope for vapor liquid fractionation in natural waters and is given for comparison.
The argument in favor of the localized processes is supported independently by the Keeling graphical approach and the magnitude of ∆ 17 O values presented below.The sources responsible for the changes of [CO 2 ] can be identified from the Keeling graphical analysis which is essentially a plot between the inverse of the concentration and a selected isotopic parameter (Pataki et al., 2003).The intercept of the carbon Keeling plot for AS data alone is -24.57± 0.77‰ (not shown) and is -23.88 ± 0.66‰ for AS and NTU combined data (Fig. 3(a)), similar to the values for C 3 -plant respiration and/or anthropogenic CO 2 with additional contribution from C 4 plant respiration (Bakwin et al., 1998;Newman et al., 2008).For comparison, the intercept from Roosevelt samples is -25.11 ± 3.41‰ (Fig. 3 Values of δ 13 C and δ 18 O are in ‰ relative to V-PDB and V-SMOW, respectively.The data refer to samples collected between September, 2013 and January, 2014.See Table 1 for details. for oceanic CO 2 (-13.48 ± 2.47‰) is consistent with a previous study carried out in a cruise from China to Antarctica (Chen et al., 2013).A detail analysis of the elevated δ 13 C intercept for the marine CO 2 is discussed elsewhere (Laskar and Liang, 2016).Keeling analysis for δ 18 O at Roosevelt gives an intercept of 24.41 ± 1.28‰ (Fig. 3(b)), close to tropospheric O 2 ; but no robust correlation is found for other sites probably due to multiple sources influencing δ 18 O values in those places.
The average ∆ 17 O value at Roosevelt is 0.248 ± 0.055‰, lower than that at AS (0.329 ± 0.044‰), NTU (0.337 ± 0.040‰), Daan (0.321 ± 0.028‰), and SCS (0.335 ± 0.034‰).The intercept in ∆ 17 O Keeling plot for CO 2 collected at Roosevelt is -0.258 ± 0.184‰ (Fig. 3(c)), suggesting combustion sources (Horvath et al., 2012).As ∆ 17 O is not a conserved quantity like δ 13 C and δ 18 O, the validity of Keeling approach in this case for a binary mixture of gases is questionable.However, it is found that the maximum possible uncertainty in the intercept could be 0.008‰ for a mixture of anthropogenic and background air CO 2 (Laskar et al., 2016).As this uncertainty is comparable to our analytical precision (0.01‰), we do not consider their difference to be significant and apply simple Keeling plot for source identification using ∆ 17 O also.The two downtown stations, NTU and Daan, do not show noticeable influence from anthropogenic CO 2 .The derived ∆ 17 O values of samples from AS, NTU, Daan, and SCS are all similar and possess similar variability.We ascribe the variability seen at AS, NTU and Daan to variations in local biogeochemistry and occasional stratospheric intrusions.The high variation at SCS, however, cannot be explained simply by local biogeochemistry, as it is not likely that oceans have CO 2 cycles as active as that of terrestrial surfaces.However, it is possible that the sampling locations at SCS are influenced by air from land sources (Kurita et al., 2015;Rangarajan et al., 2017).Further analysis, for example, in samples from remote Pacific Ocean (to minimize terrestrial alterations) is needed to resolve the processes responsible for the changes of ∆ 17 O values in this region.
The time series of ∆ 17 O values for AS and NTU is presented in the bottom panel of Fig. 2. The values vary from 0.216 to 0.427‰ with an average of 0.332 ± 0.043‰, a value that is similar to 0.31 ± 0.06‰ obtained in La Jolla (Thiemens et al., 2014) and 0.321 ± 0.007‰ in Jerusalem (Barkan and Luz, 2012); note that the values for the La Jolla and Jerusalem stations have been re-scaled using a λ value of 0.516 for comparison with our data.Fig. 4 presents all available ambient CO 2 data in three oxygen isotope plot.A linear fit to the combined data from AS and NTU gives a slope of 0.518 ± 0.004 (1-σ standard error) which is close to the slope 0.516 of terrestrial mass fractionation line we adopt here (shown by the solid line).Following Landais et al. (2006), this value is justified since the average relative humidity in Taipei is ~75% (Laskar et al., 2014).Moreover, the overall ratio of ln(1 + δ 17 O)/ln(1 + δ 18 O) in our samples is 0.524 ± 0.001, deviated only slightly from that of water-CO 2 equilibrium value of 0.523 (Hofmann et al., 2012;Barkan and Luz, 2012), suggesting that the near surface CO 2 in this region is primarily affected by local biogeochemistry with only minor signal from the stratosphere.A combined assessment of triple oxygen isotopes with eddy covariance flux data probably would help to identify and quantify the sources better.

Source Variation of ∆ 17 O Values
In addition to the known photochemical processes in the middle atmosphere (e.g., Liang et al., 2008), there are two processes at lower levels that can modify the ∆ 17 O values of CO 2 , namely, isotope exchange with water and combustion.The former can be estimated, assuming water-CO 2 equilibration Fig. 4. Three oxygen isotope plot of air CO 2 samples collected at various locations.Symbol notations are the same as those in Fig. 3.The terrestrial fractionation curve for λ = 0.516 is shown by the solid line.Data from La Jolla (Thiemens et al., 2014) are also shown (solid red) for comparison.
which follows λ = 0.523 (Barkan and Luz, 2012;Hofmann et al., 2012).Under the assumed value of λ here (λ o = 0.516), the ∆ 17 O value of CO 2 emitted from sources involving water-CO 2 equilibration like respiration is (0.523-0.516) × ln(α 18 O water-CO2 ) + ∆ 17 O water , where α 18 O water-CO2 and ∆ 17 O water are, respectively, the fractionation factor of CO 2 in equilibrium with water and the ∆ 17 O value of water.The α 18 O water-CO2 value is taken to be 1.041 at 25°C.The meteoric water in Taiwan has an average δ 18 O value of -4‰ in winter time (Peng et al., 2010).The ∆ 17 O value, however, has not been measured for rainwater.If we assume ∆ 17 O value to be the same as that of the global average (Luz and Barkan, 2010), ∆ 17 O water , in our definition, would be -0.015‰.The overall ∆ 17 O value from plant respiration at 25°C would then be 0.266‰ (= (0.523 -0.516) × ln(1.041)-0.015‰), which is similar to the equilibrated oceanic CO 2 value of 0.281‰ (assuming ocean water temperature of 25°C).
The CO 2 produced by combustion inherits the δ-values as well as the ∆ 17 O value of air O 2 , which is about -0.2‰ (Thiemens et al., 1995;Barkan and Luz, 2012;Horvath et al., 2012;Laskar and Liang, 2016).An idea for the source variation of ∆ 17 O values can be obtained from the Keeling plot of ∆ 17 O vs. 1/[CO 2 ] as well as a regression plot of ∆ 17 O vs. δ 13 C for the Roosevelt data: where ∆ 17 O and δ 13 C are in units of ‰ and [CO 2 ] in ppmv (see Figs. 3(c) and 3(d)).The intercept of -0.258 ± 0.184‰ (Eq.( 3)) is the ∆ 17 O value of O 2 source in combustion CO 2 .This value is close to that expected as the source of O 2 in fossil fuel combustion is the atmospheric O 2 with ∆ 17 O value of -0.2‰ as stated above.These regression equations (Eqs.( 3) and ( 4)) also state that a decrease of 0.01‰ in

Stratospheric Influence and Surface Processes
Fig. 5 shows how ∆ 17 O changes with the change in δ 13 C due to anthropogenic emissions for AS, NTU and Daan samples.Shaded region is the 1-σ envelop to the regression line obtained for the Roosevelt data presented in Fig. 3(d).CO 2 originating from anthropogenic sources would fall in the shaded region.A positive change in ∆ 17 O value can be explained by incursion of air from the stratosphere where the ∆ 17 O value of CO 2 could be more than 1‰.The negative values, on the other hand, reflect addition of CO 2 from fossil fuel burning and/or CO 2 which has undergone biogeochemical/ hydrospheric processing.We see that most of the points fall either well above the shaded region or below, and only a few samples lie in the shaded zone.This supports our inference from the Keeling plot method (Fig. 3) that in regions away from anthropogenic sources, local biogeochemistry controls the variation of CO 2 isotopologues.Daan, for example, located at the central downtown, does not show significant influence from car exhaust (shown by the triangles).Simple two-box mixing of background and anthropogenic CO 2 is found inadequate to explain the observed changes.For points falling above the shaded area, an addition of air having elevated ∆ 17 O values is required.Those points falling below the shaded region need biological/hydrospheric processing, in addition to anthropogenic contribution, thus providing an additional way for estimating CO 2 fluxes (and thus, the GPP) between the surface and atmosphere.An attempt to constrain either the strength of the stratospheric input or the surface reset rate is made below.(September 2013-February 2014), the diurnally averaged changing rate is 0.0007 ± 0.0158‰ hr -1 , suggesting that the region is highly variable in the strengths of carbon cycle/anthropogenic emission and stratospheric input, but the three sources are nearly in balance.The daytime photosynthetic uptake rate may be estimated from the samples having negative changes, because biological processes tend to produce CO 2 with a lower ∆ 17 O than that of the atmospheric value (Luz et al., 1999) in contrast to the positive change due to CO 2 from the stratosphere described above.If we assume tropospheric CO 2 in isotopic equilibrium with various waters (leaf water, soil water, and ocean water) and the air within the boundary layer is well mixed, the budget of ∆ 17 can be written as follows: where N is the column density of CO 2 in the boundary mixed layer, subscripts "a", "s", "l", "r", "o", and "st" of ∆ 17 represent the sampled air, the soil, the leaf, the respiration, the ocean water, and the stratosphere, respectively, and F the flux in and out of a reservoir such that the subscript "s" refers to soil invasion, "la" leaf-to-air, "r" respiration, "oa" oceanto-air, and "st" the air from the stratosphere.Typically the flux between air and leaf stomata can be described by where F A is the plant assimilation rate, c a the CO 2 mixing ratio in the air, and c c the mixing ratio in the chloroplasts.For C 3 plants (dominant in the study site, AS), we take c c /c a = 2/3 (Pearcy and Ehleringer, 1984).Then Eq. ( 5) reduces to: where we assume F A = F r and ∆ 17 s = ∆ 17 l = ∆ 17 r .The above equation also assumes that the boundary mixed layer matures in the late morning hours and no additional fluxes from the stratosphere/free troposphere can, thenceforth, enter the boundary layer and from the ocean to the study site.The value of 3 in the above equation is the same as used by Hoag et al. (2005) for C 3 plants assuming complete equilibration with water.The F A term in Eq. ( 7) is estimated assuming complete equilibration between water and CO 2 and is therefore a lower limit (Farquhar et al., 1993;Gilon and Yakir, 2001;Cousins et al., 2008).C 4 contribution is neglected in this estimate.
As mentioned, in our sampling region the average value of ∆ 17 O is 0.332‰ (Table 1) and the CO 2 surface concentration n is about 10 16 molecules cm -3 .The boundary mixed layer l PBL ranges from ~1 km at noon time to ~0.5 km in the middle of night (Chou et al., 2007) and N = n × l PBL = 10 21 molecules cm -2 , for l PBL = 1 km.If we take an averaged ∆ 17 a /t = -0.008‰hr -1 (Fig. 6) and ∆ 17 a = 0.332‰, F s + 3F A = 3 × 10 16 molecules cm -2 s -1 .Soil invasion (Wingate et al., 2009) is not well quantified and we assume F s = F A (enzyme catalyzed limit; see Wingate et al., 2009), then F A = 8 × 10 15 molecules cm -2 s -1 , a factor of ~5 higher than that estimated previously (Sims et al., 2005;Baldocchi, 2008;Beer et al., 2010).One explanation for the discrepancy is that we may have sampled different air masses from regions with different isotopic composition (Kurita et al., 2015;Rangarajan et al., 2017).So the observed temporal change may not reflect local biogeochemistry but spatial inhomogeneity.However, we note that similar variation is also seen in Daan, where in its surroundings car exhaust have distinctly different ∆ 17 O values, but no noticeable anthropogenic signal is observed (Fig. 5).This rules out the possibility that the variation originates from lateral transport.An alternate possibility concerns the validity of the assumption that molecules in the boundary layer are well mixed.It is known that in addition to the well-known boundary mixed layer, there exists a stagnant surface layer, roughly about 10-30 m height (Jacobson, 2005).Most of our current measurements were taken in this height range.Typical mixing time for establishing a well-mixed boundary layer is about one hour.So the surface layer ∆ 17 O signal from bottom-up transport to represent whole mixed layer has to be modified.Top-down transport of stratospheric signals to the surface layer, however, is less affected.As a result, the above estimate tends to overestimate the flux from the surface.Observationally, there are two ways to tackle the problem.One is to have a higher sampling rate to resolve eddy transport from the surface to the established mixed layer, i.e., at least 2-3 measurements per hour, the mixing time scale.The other is to sample air well above the surface layer.Rigorous error analysis for the numbers derived above cannot be done at this stage, because of poor realization and formulation of biological and dynamical processes occurring in the surface layer.
Similar calculation can be done for F st by assuming F la = F r = 10 15 molecules cm -2 s -1 (taken from direct flux measurements; Sims et al., 2005) and F oa = 0.The average increase of ∆ 17 a /t = 0.008‰ hr -1 suggests F st = 2 × 10 15 molecules cm -2 s -1 , a value that is not unreasonable if one compares the maximum flux in spring from a model simulation (Liang et al., 2008), echoing the conclusion arrived by Liang and Mahata (2015).The value could be higher if higher F la and F r are taken.

SUMMARY
Analysis of two oxygen isotope ratios (δ 17 O and δ 18 O) in atmospheric CO 2 has been carried out in a variety of ecosystems, covering urban, semi-urban, and oceanic areas in and around Taiwan region.In general, the ∆ 17 O value (representing isotope anomaly) in this region is highly variable, with mean value significantly lower in areas dominated by anthropogenic emissions.We show from measurements of samples from Daan forest park that local biogeochemistry dominates in determining the ∆ 17 O values.This is also reflected in the Keeling plot and shows insignificant contribution from air having depleted ∆ 17 O values from pollution sources.Detail analyses also show that anthropogenic processes are insufficient to explain the variability of ∆ 17 O values.
Overall, the ∆ 17 O value of atmospheric CO 2 from the Taiwan region is ~0.05‰ higher than that expected from surface emissions, assuming the surface emitted CO 2 is in isotopic equilibrium with local waters at 25°C (i.e., observed 0.332‰ versus calculated equilibrium values of 0.266-0.281‰).This excess is significantly smaller than that estimated (0.15‰) previously on a global scale (Hoag et al., 2005).There are two possible explanations.One is that the flux from the stratosphere is actually smaller than the one used by Hoag et al. (2005).The other explanation requires that the ∆ 17 O reset rate of CO 2 through processes occurring on the earth surface is faster than that estimated previously (e.g., see Hoag et al., 2005), and this agrees with the conclusion of a reduced CO 2 recycling time arrived by Welp et al. (2011).To fully utilize the ∆ 17 O analysis for understanding CO 2 biogeochemical cycle and to estimate and quantify missing sinks, more extensive measurements at various land surface types assisted by a regional model are needed.More specifically, to gain deeper insight into the variability in the ∆ 17 O values of CO 2 reported in this work requires that (i) CO 2 isotope measurements in the free troposphere (reachable around Taiwan, because in winter time, the surrounding mountain height is usually greater than the boundary mixed layer height), to provide a better estimate of the flux from the stratosphere, (ii) CO 2 isotope measurements in the boundary mixed layer, to reduce possible bias from sampling in the surface layer, and (iii) accurate determination of the boundary mixed layer height.These efforts would be able to provide a better understanding of the processes that affect the CO 2 isotopologues.

Fig. 1 .
Fig. 1.Locations of the sampling sites.Tips of the arrows in South China Sea (top left) show the sampling stations: OC1 and OC2 denote two ocean cruses conducted in June, 2013 and October, 2013, respectively (see Table 1 for dates and time of sampling).Sampling station at Academia Sinica is shown in the top right.The three stations in the downtown Taipei City, viz.Daan Park, Roosevelt Road and National Taiwan University (NTU) are shown at the bottom.

Fig. 2 .
Fig. 2. Time series of atmospheric CO 2 parameters ([CO 2 ], δ 13 C, δ 18 O, and ∆ 17 O values) at AS (black symbols) and NTU (red symbols).Values of δ 13 C and δ 18 O are in ‰ relative to V-PDB and V-SMOW, respectively.The data refer to samples collected between September, 2013 and January, 2014.SeeTable 1 for details.

Fig. 3 .
Fig. 3. Keeling plots of carbon (a) and oxygen (b, c) isotope ratios for air CO 2 samples collected at AS and NTU (black solid circles), SCS (grey solid circles), Daan (solid triangles), and Roosevelt (open triangles).Solid line is the linear regression obtained from the combined AS and NTU data and dotted line is from the Roosevelt Road data.To illustrate the contribution of anthropogenic emissions to ∆ 17 O values, scatter plot of ∆ 17 O-δ 13 C is also shown in (d).

∆
17 O value due to vehicle emission requires an addition of ~5 ppmv in [CO 2 ] or a decrease of 0.3‰ in δ 13 C.These empirical relations can be used to assess the alternation of ∆ 17 O values in urban areas by anthropogenic emissions.

Fig. 6
shows the rate of change of ∆ 17 O value plotted as a function of local time.Over the main sampling period

Fig. 5 .
Fig. 5. Scatter plot of the changes of ∆ 17 O and δ 13 C values, with 1-σ error bar overlaid (most of the points have error bars less than the symbol size), between two consecutive samples in a day; here the errors for individual ∆ 17 O and δ 13 C measurements are taken to be 0.01‰ and 0.06‰, respectively.Shaded regions (1-σ error bar) define the contribution from anthropogenic emissions, taken from Fig. 3(d).Solid blacks circles are for AS+NTU data and blue triangles for Daan.

Fig. 6 .
Fig. 6. ∆ 17 O changing rates, with 1-σ error bar overlaid, plotted against the local time.Dashed lines show the average rates for values greater and less than zero, respectively.Solid black circles represent AS and NTU data and blue triangles the Daan data.