Global 3-D Simulations of the Triple Oxygen Isotope Signature Δ17O in Atmospheric CO2.

The triple oxygen isotope signature Δ17O in atmospheric CO2, also known as its "17O excess," has been proposed as a tracer for gross primary production (the gross uptake of CO2 by vegetation through photosynthesis). We present the first global 3-D model simulations for Δ17O in atmospheric CO2 together with a detailed model description and sensitivity analyses. In our 3-D model framework we include the stratospheric source of Δ17O in CO2 and the surface sinks from vegetation, soils, ocean, biomass burning, and fossil fuel combustion. The effect of oxidation of atmospheric CO on Δ17O in CO2 is also included in our model. We estimate that the global mean Δ17O (defined as Δ 17 O = ln ( δ 17 O + 1 ) - λ RL · ln ( δ 18 O + 1 ) with λ RL = 0.5229) of CO2 in the lowest 500 m of the atmosphere is 39.6 per meg, which is ∼20 per meg lower than estimates from existing box models. We compare our model results with a measured stratospheric Δ17O in CO2 profile from Sodankylä (Finland), which shows good agreement. In addition, we compare our model results with tropospheric measurements of Δ17O in CO2 from Göttingen (Germany) and Taipei (Taiwan), which shows some agreement but we also find substantial discrepancies that are subsequently discussed. Finally, we show model results for Zotino (Russia), Mauna Loa (United States), Manaus (Brazil), and South Pole, which we propose as possible locations for future measurements of Δ17O in tropospheric CO2 that can help to further increase our understanding of the global budget of Δ17O in atmospheric CO2. ©2019. The Authors.

Introduction

Oxygen has three naturally occurring stable isotopes 16O, 17O, and 18O of which 16O is by far the most abundant on Earth. For atmospheric CO2, the relative abundances of C16O16O, C17O16O, and C18O16O are 99.5%, 0.077%, and 0.41%, respectively (see, e.g., Eiler & Schauble, 2004). We can quantify the oxygen isotopic composition of a sample as where n refers to the rare oxygen isotope (i.e., n=17 or 18) and Vienna Standard Mean Ocean Water (VSMOW) is used as the reference standard and δ values are usually expressed in per mil (‰). The isotopic composition of oxygen‐containing molecules on Earth, like CO2 or H2O, is affected by processes such as diffusion, evaporation, and condensation. These processes depend on the mass of the molecules and therefore result in a mass‐dependent fractionation of the oxygen isotopes. As a consequence, the variations in δ 17O and δ 18O of oxygen‐containing substances on Earth are strongly correlated.

A deviation from the mass‐dependent fractionation can be expressed by the Δ17O signature (“triple oxygen isotope” or “17O excess”). In this study we consistently use the logarithmic definition for Δ17O (see Section S1 of the supporting information for an overview of alternative definitions that are commonly used) which is usually expressed in per mil (‰) or per meg (0.001‰), depending on the magnitude of the Δ17O signature, where λ RL is the reference line. We selected λ RL=0.5229, which is equal to the isotopic equilibration constant of CO2 and water (Barkan & Luz, 2012), since equilibration of CO2 with water is a key process in our study. As a consequence, the Δ17O signature of CO2 that equilibrates with a large amount of water will be reset to the Δ17O signature of the water reservoir. Relative to this selected reference line λ RL, other mass‐dependent processes (e.g., diffusion) result in a minor fractionation of oxygen isotopes (fractionation of Δ17O due to diffusion is described in section 2.3.1).

Stratospheric CO2 was shown to be anomalously enriched in oxygen isotopes with Δ17O≫ 0‰ in measurement campaigns performed with rockets (Thiemens et al., 1995a), aircraft (Boering et al., 2004; Thiemens et al., 1995b), balloons (Alexander et al., 2001; Kawagucci et al., 2008; Lämmerzahl et al., 2002; Mrozek et al., 2016), or using aircraft and balloons (Wiegel et al., 2013; Yeung et al., 2009). The anomalous isotopic composition of stratospheric CO2 has been linked to oxygen exchange with stratospheric O3, which has a positive Δ17O signature, by Yung et al. (1991). Photolysis of O3 produces the highly reactive radical O(1D) which can form the unstable CO3 * when colliding with CO2, which dissociates into CO2 and an oxygen radical

The oxygen atom that is removed by disintegration of CO3 * is random (except for the small fractionation of a few per mil favoring 18O remaining in the CO2 product; Mebel et al., 2004), such that there is an approximately two‐thirds probability that the reactions in equations (3) and (4) will result in the substitution of an oxygen atom in CO2 with an oxygen atom that was originally present in O3. This exchange of oxygen atoms from stratospheric O3 to CO2 is responsible for the transfer of the 17O anomaly (i.e., Δ17O≫0‰) from stratospheric O3 to stratospheric CO2.

In the upper troposphere, there is an influx of stratospheric CO2 with Δ17O≫0‰ (this stratospheric influence on Δ17O of tropospheric CO2 was recently observed by Laskar et al. (2019) in air samples from two aircraft flights). Following transport to the troposphere, the CO2 is mixed and can come into contact with liquid water in vegetation, soils, or oceans. When CO2 dissolves in liquid H2O, exchange of oxygen atoms occurs, such that the CO2 that is released back to the atmosphere has a signature of Δ17O≈0‰. The exchange between CO2 and H2O in vegetation is highly effective due to the presence of the enzyme carbonic anhydrase, whereas the exchange of oxygen isotopes between CO2 and cloud droplets is negligible due to the absence of carbonic anhydrase in the atmosphere (Francey & Tans, 1987). The resulting Δ17O signature in tropospheric CO2 reflects a dynamic balance of highly enriched stratospheric CO2 and equilibration that occurs in vegetation and other water reservoirs. Tropospheric measurements of Δ17O in CO2 have previously been performed in Jerusalem, Israel (Barkan & Luz, 2012); La Jolla, United States (Thiemens et al., 2014); Taipei, Taiwan (Liang & Mahata, 2015; Liang et al., 2017a, 2017b; Mahata et al., 2016a), Göttingen, Germany (Hofmann et al., 2017), and Palos Verdes, United States (Liang et al., 2017b).

Gross primary production (GPP; the gross uptake of CO2 by vegetation through photosynthesis) is a key process in the carbon cycle which is currently poorly constrained. Increasing our understanding of the terrestrial carbon cycle is essential for predicting future climate and atmospheric CO2 concentrations (Booth et al., 2012). An estimate of 120 PgC/year for global GPP was provided by Beer et al. (2010) by using machine learning techniques to extrapolate a database of eddy covariance measurements of CO2 to the global domain. An estimate of 150–175 PgC/year for global GPP was derived by Welp et al. (2011) based on the response of δ 18O in atmospheric CO2 after El Niño–Southern Oscillation events. The large spread in estimates of global GPP clearly indicates our current lack of understanding of the biospheric domain in the global carbon cycle.

Because the Δ17O signature of tropospheric CO2 strongly depends on the magnitude of the exchange of CO2 with liquid water in leaves, it is a potential tracer for GPP, as was first proposed by Hoag et al. (2005). Similarly, the δ 18O signature of tropospheric CO2 has been explored to constrain terrestrial carbon fluxes by Ciais et al. (1997a, 1997b), Peylin et al. (1997, 1999), and Cuntz et al. (2003a, 2003b). The main advantage of using Δ17O instead of δ 18O is that the signal is less affected by processes in the hydrological cycle (e.g., evaporation and condensation), since these are largely mass dependent (Hoag et al., 2005). Besides constraining gross terrestrial CO2 fluxes, other possible applications of Δ17O in atmospheric CO2 have been suggested, such as constraining stratospheric circulation and constraining the abundance and variability of O(1D) (e.g., Alexander et al., 2001).

The first two‐box model for Δ17O in tropospheric CO2 for the Northern and Southern Hemispheres was developed by Hoag et al. (2005). This conceptual box model takes into account the exchange fluxes of CO2 between the troposphere and the stratosphere, vegetation, and oceans. In addition, the supply of CO2 from fossil fuel combustion and land use change is incorporated in the box model. All these CO2 fluxes are associated with a reservoir‐specific Δ17O signature. The resulting Δ17O for tropospheric CO2 was calculated using a mass balance. Results from Hoag et al. (2005) can be converted into our reference frame, as defined in equation (2), assuming a global δ 18O signature of 41.5‰ (observations from Francey & Tans, 1987, show that the global mean δ 18O in CO2 is ∼0‰ PDB‐CO2, which can be converted using equation 5 from Brenninkmeijer et al. (1983) into 41.5‰ VSMOW), which yields Δ17O=0.066‰ for tropospheric CO2.

A more sophisticated global one‐box model was developed by Hofmann et al. (2017). This model takes into account that certain processes (e.g., diffusion of CO2 from the atmosphere into leaf stomata) can fractionate oxygen isotopes and influence the Δ17O signature of CO2. Another significant difference with the model from Hoag et al. (2005) is the soil invasion fluxes that are taken into account. Also, both models differ in the magnitude of the CO2 fluxes and the Δ17O reservoir signatures. Based on a Monte Carlo simulation where the uncertainty in the input variables is considered, Hofmann et al. (2017) predict Δ17O=0.061±0.033‰ for tropospheric CO2.

In recent years, there have been developments in the available measurement techniques for Δ17O in CO2. Mahata et al. (2013, 2016b) developed a measurement technique based on the equilibration between CO2 and O2 catalyzed by hot platinum, followed by measurement of the Δ17O signature of O2, from which the initial Δ17O signature of CO2 can be inferred with a precision of 8 per meg. Barkan and Luz (2012) developed a high‐precision measurement technique based on equilibration of CO2 and H2O, resulting in a precision of 5 per meg for Δ17O in CO2. Using laser‐based techniques, Stoltmann et al. (2017) were able to reach a precision for Δ17O in CO2 of better than 10 per meg. The quantum cascade laser developed by Aerodyne Research is also able to measure Δ17O in CO2 with high precision (McManus et al., 2015; Nelson et al., 2008). In addition, a recently developed ion fragment method allows to measure δ 17O and δ 18O directly on CO2 without the need of chemical conversion (Adnew et al., 2019). The recent developments in the measurement techniques for Δ17O in CO2 are essential for its application as tracer for the terrestrial carbon cycle.

Because of the recent advancements in measurement techniques for Δ17O in CO2, it is now possible to observe spatial and temporal gradients of Δ17O more accurately. To simulate the spatial and temporal variability of the Δ17O signal in atmospheric CO2, the available box models are not suitable and a 3‐D model framework is required. For this purpose, an oxygen isotope module for atmospheric CO2 was implemented in the atmospheric transport model TM5 (Huijnen et al., 2010; Krol et al., 2005). Results from an early version of our 3‐D model were compared with the Δ17O measurement series from Göttingen, Germany (Hofmann et al., 2017). A detailed description of our updated Δ17O model is given in section 2, and the changes in our current model with respect to the earlier version used by Hofmann et al. (2017) are summarized in section S2 of the supporting information. The model results are reported in section 3, followed by the discussion and conclusion in sections 4 and 5.

Methods

General Model Description

Our model framework for Δ17O in atmospheric CO2 is based on the atmospheric transport model TM5 (Krol et al., 2005), which is driven by ERA‐Interim meteorological fields (Dee et al., 2011) provided by the European Centre for Medium‐Range Weather Forecasts. TM5 uses a longitude‐latitude grid of 6° × 4°, 3° × 2°, or 1° × 1° resolution, depending on the chosen setup. Also, TM5 allows the use of two‐way nested zoom regions to simulate with a higher horizontal resolution for specific regions. For the vertical coordinate TM5 uses 25, 34, or 60 hybrid sigma‐pressure levels, such that the lowest model levels follow the surface elevation and the higher levels are (almost completely) isobaric. For this study, we performed simulations with the coarsest resolution (i.e., a horizontal resolution of 6° × 4° and 25 vertical levels with the highest model level at 47.8 Pa).

In our model we apply two‐way CO2 fluxes, exchanging between the stratosphere, biosphere, soil, ocean and the troposphere, and one‐way CO2 fluxes from fossil fuel combustion, biomass burning, and oxidation of CO into the troposphere, as illustrated in Figure 1. Modeling the gross two‐way exchange fluxes for some reservoirs is necessary to estimate the resulting Δ17O signature of tropospheric CO2. The CO2 fluxes in our model are time and space dependent and can originate from the stratosphere (described in section 2.2), the Earth surface (section 2.3) and are present within the troposphere itself in the case of oxidation of atmospheric CO (section 2.4). Also, the Δ17O signatures of the different reservoirs are indicated in Figure 1. The Δ17O signatures for stratospheric CO2, soil water, leaf water, and atmospheric CO are time and space dependent in our model. Note that for the exchange fluxes between the atmosphere and biosphere, kinetic fractionation affects the Δ17O signature (described in sections 2.3.1 and 2.3.2) and that the oxidation of CO by OH is not a mass‐dependent process, such that the Δ17O signature of atmospheric CO is not directly transferred to CO2 (described in more detail in section 2.4).

Figure 1

Conceptual overview of processes affecting the Δ17O signature of atmospheric CO2 in our model. The CO2 mass fluxes, indicated with symbol F, are given in units of PgC/year, and Δ17O signatures are given in ‰ as defined in equation (2) relative to a reference line λ RL = 0.5229. The reported values for CO2 mass fluxes are integrated over the global domain, averaged over the years 2012/2013 (as reported in Table S2 of the supporting information) and rounded to integer values. As a sign convention, the CO2 mass fluxes that tend to increase the tropospheric CO2 mass are expressed as positive numbers. The main source of Δ17O in tropospheric CO2 is exchange with the stratosphere (F SA and F AS), as described in section 2.2. The stratospheric signature in our model is time and space dependent, and the indicated value of 0.66‰ is the effective signature that is associated with stratosphere‐troposphere exchange (determined from the stratosphere‐troposphere CO2 mass flux and Δ17O isoflux as reported in Table S2 of the supporting information). The main sink for Δ17O in tropospheric CO2 is the exchange with leaves (F AL and F LA), which is associated with a large uncertainty. Also, the magnitude of the exchange fluxes between the soil and atmosphere (F ASI and F SIA) is uncertain. The implementation of the surface sources and sinks of CO2 is described in section 2.3. Note that the high signature is not directly transferred to CO2 because of fractionation of oxygen isotopes that occurs during the oxidation of CO, as described in section 2.4.

In our model framework we implemented CO2 and C17OO as independent tracers, while assuming a fixed atmospheric signature of δ 18O = 41.5‰ VSMOW. With the fixed δ 18O, we can translate the imposed boundary conditions (i.e., sources and sinks) of Δ17O into an equivalent boundary condition for the δ 17O signature, based on equation (2). Subsequently, the C17OO tracer mass can be determined from the local tracer mass of CO2 and δ 17O using equation (1). The C17OO tracer mass can then be transported in our atmospheric model. By again using δ 18O = 41.5‰ VSMOW, we can “translate” the simulated C17OO tracer mass back into Δ17O for analysis. By using a fixed δ 18O signature, we are able to simulate the transport of the Δ17O signature in CO2, without the need of explicitly modeling the variations in δ 18O that are strongly related to the water cycle (Ciais et al., 1997a, 1997b; Cuntz et al., 2003a, 2003b; Peylin et al., 1997, 1999). The consequence of this approach is that our model simulated δ 17O cannot be directly compared to δ 17O observations. Model output becomes meaningful after converting the simulated δ 17O fields using the fixed δ 18O signature into Δ17O fields. To convert isotopic signatures to isotope ratios, we use (Baertschi, 1976) and (Li et al., 1988). Note that more recent studies estimate the latter to be slightly higher, 386.7·10−6 and 382.7·10−6 according to Assonov and Brenninkmeijer (2003) and Kaiser (2008) respectively, but the effect on our simulated Δ17O is negligible.

We have defined several model parameters that can be set to user‐specified values. The motivation for this implementation is that many of the model parameters are uncertain (e.g., the magnitude of the soil invasion flux, as discussed in section 2.3.2), and this flexibility allows us to efficiently investigate the sensitivity to these model parameters. An overview of the most important model parameters and the available settings is given in Table 1. A more detailed explanation of the model parameters and available settings is given in the following sections. A summary of the model simulations that were conducted in this research is provided in Table 2.

Table 1

Overview of the Main Model Parameters and Available Settings for the 3‐D Δ17O Model

Reservoir	Section	Model parameter	Base setting	Alternative settings
Stratosphere	2.2	Δ17O–N2O fit	Least squares fit	Upper/lower 95% confidence limit fit
		[N2O] fit threshold	240 ppb level	Zero or positive value
		Relaxation time scale	0 hr (i.e., no relaxation)	Zero or positive value
Vegetation	2.3.1	Soil water Δ17O	Distributed from precipitation	Constant Δ^17O_soil
		Leaf water Δ17O	Dynamic from rel. humidity	Constant λ transp
Soil	2.3.2	Invasion flux magnitude	30 PgC/year globally	Zero or positive value
		Invasion flux distribution	Scaled from CO2 respiration flux	Scaled from H2 deposition velocity
Ocean	2.3.3	CO2 fluxes	Dynamically coupled to [CO2]	Calculated from predefined [CO2]
		C17OO fluxes	Dynamically coupled to [C17OO]	Calculated from predefined [C17OO]
Atmospheric CO	2.4	Setting	Not included	Included with nonzero ϵ CO+OH

Note. Note that the soil water signature is listed here under the vegetation reservoir, but it also affects the soil invasion fluxes. The model results with base settings are described in sections 3.1.1 and 3.1.2. The effect of some of the alternative settings on the model predictions is discussed in section 3.1.3.

Table 2

Overview of Performed Simulations for Sensitivity Analysis Including the Base Model Run

Name	Description
BASE	Base model run
ST_LOWER	95% confidence interval lower limit fit
ST_UPPER	95% confidence interval upper limit fit
SOIL_CONST	Δ17Osoil = −5 per meg
LEAF_CONST	λ transp = 0.5156
RESP_240	Respiration scaling; global magnitude 240 PgC/year
RESP_450	Respiration scaling; global magnitude 450 PgC/year
HYD_240	H2 deposition scaling; global magnitude 240 PgC/year
HYD_450	H2 deposition scaling; global magnitude 450 PgC/year
CO_ROCK	ϵ CO+OH from Röckmann et al. (1998a)
CO_FEIL	ϵ CO+OH from Feilberg et al. (2005)

Note. The resulting Δ17O signature of atmospheric CO2 and the Δ17O isofluxes for the base model run are discussed in sections 3.1.1 and 3.1.2. The results of the sensitivity analyses are given in section 3.1.3.

Stratospheric Source of Δ17O in CO2

N2O–Δ17O(CO2) Correlation

The production of isotopically anomalously enriched CO2 in the stratosphere has been linked to the exchange of oxygen atoms between O3 and CO2 via O(1D) as described in section 1 and shown in equations (3) and (4). Since the initial discovery of stratospheric CO2 with Δ17O≫0‰, a number of research groups were able to produce anomalously enriched CO2 from UV‐irradiated O2 or O3 and CO2 in controlled laboratory environments (Chakraborty & Bhattacharya, 2003; Johnston et al., 2000; Shaheen et al., 2007; Wen & Thiemens, 1993; Wiegel et al., 2013). Despite the knowledge gained through these studies, there are currently still many questions remaining regarding the dependence on temperature, pressure, photolysis wavelength, and concentrations of O2, O3, and CO2 in the stratosphere. Considering the uncertainties associated with explicitly modeling the production of Δ17O in CO2 based on the reactions in equations (3) and (4), we decided to impose Δ17O in stratospheric CO2 based on its observed correlation with N2O, which we expect to be a more robust approach.

The correlation between N2O and Δ17O in CO2 was first used by Luz et al. (1999) to estimate the stratospheric influx of Δ17O for CO2 and O2 into the troposphere. Boering et al. (2004) describe that atmospheric transport is the physical mechanism behind the N2O–Δ17O(CO2) correlation, as both N2O and Δ17O in CO2 are long‐lived tracers (the lifetime of N2O is approximately 120 years; Volk et al., 1997). The negative slope of the N2O–Δ17O(CO2) correlation is explained by the opposite effect of stratospheric photochemistry on N2O and Δ17O in CO2 (Δ17O in CO2 is produced from O(1D) originating from O3 photolysis, as described in section 1, and N2O is removed by photolysis and O(1D), as described in section 2.2.2).

Experimental data sets for stratospheric N2O and Δ17O in CO2 from Thiemens et al. (1995a), Boering et al. (2004), Kawagucci et al. (2008), and Wiegel et al. (2013) were examined to test the robustness of the N2O–Δ17O(CO2) correlation. The Δ17O values for these studies were recalculated from the reported δ 17O and δ 18O signatures using the definition of Δ17O as given in equation (2). The N2O mole fractions were detrended to account for the atmospheric growth rate of N2O and the difference in date of sample collection, according to the detrending procedure described in section 2.2.2. The reader is referred to the original works for details on experimental techniques and the associated uncertainties. Despite the difference in date and location of sample collection, there is a strong correlation between the N2O mole fraction and the Δ17O signature of CO2 that is linear for N2O in the range of 50 to 320 ppb as shown in Figure 2b. In the mesosphere the correlation between N2O and Δ17O in CO2 breaks down as discussed in detail by Mrozek (2017).

Figure 2

Overview of simulated and measured stratospheric N2O mole fraction and Δ17O signature in CO2. (a) Annual mean, zonal mean TM5 model predictions of detrended N2O mole fractions using a horizontal resolution of 6° × 4° and 25 vertical levels compared to detrended measurements of N2O mole fractions from Thiemens et al. (1995a), Boering et al. (2004), Kawagucci et al. (2008), and Wiegel et al. (2013) for stratospheric air in Northern Hemisphere. The background color indicates the value of the TM5 model prediction, and the color of the symbols indicates the measured value. (b) Δ17O signatures of stratospheric CO2 versus detrended N2O mole fraction, constructed from measurements by Thiemens et al. (1995a), Boering et al. (2004), Kawagucci et al. (2008), and Wiegel et al. (2013) and linear least squares fit with its corresponding 95% confidence interval. The error bars from Thiemens et al. (1995a) and Wiegel et al. (2013) are omitted from the figure to improve visibility.

We derived a linear fit for the detrended N2O mole fraction and Δ17O in CO2 using a least squares approach with equal weights assigned to each individual measurement (data for [N2O] < 50 ppb was excluded), based on the formulation where [N2O]dtd is the detrended N2O mole fraction. In addition to the least squares solution for the coefficients a and b in equation (5), we also constructed a 95% confidence interval, as shown in Figure 2b. The effect of the N2O–Δ17O fit on the resulting distribution of Δ17O in CO2 is tested by performing different simulations (BASE, ST_UPPER and ST_LOWER as defined in Table 2), the results of which are discussed in section 3.1.3.

In our model framework, the fit in equation (5) is implemented with a cutoff at 0‰, to prevent negative Δ17O values in the stratosphere. Also, a relaxation time can be specified in the model that determines the strength of the coupling between Δ17O for stratospheric CO2 and N2O mole fractions, such that where Δt is the model time step, τ relax is a user‐specified time scale, and and refer to Δ17O signature for the new and old time steps, respectively. In our model, we can apply the fit based on the vertical level (e.g., for cells with atmospheric pressure below 100 hPa) or depending on the local N2O mole fraction (e.g., for cells with N2O mole fractions below 280 ppb). The values used for these parameters in the base model run are summarized in Table 1.

N2O

We simulated N2O based on stratospheric sinks and optimized surface fluxes from Corazza et al. (2011) and Bergamaschi et al. (2015). The 2‐D surface fluxes have a time resolution of 1 month and a horizontal resolution of 6° × 4°. The 3‐D sink fields have the same time resolution and same horizontal resolution and consist of 25 vertical levels. The N2O surface fluxes are optimized for the years 2006 and 2007 by Corazza et al. (2011) and Bergamaschi et al. (2015), and we extrapolate the N2O sources for years outside of this range. The N2O sinks are climatological fields derived from the ECHAM5/MESSy1 model (Brühl et al., 2007). The sink fields distinguish between N2O loss caused by O(1D) (roughly 10% of total loss) and photolysis (roughly 90% of N2O loss) and have a strong seasonal cycle due to the changing orientation of the Earth with respect to the Sun. The sum of the yearly emissions is on average: ∼16 TgN/year, and the imbalance between the sources and sinks is ∼3.5 TgN/year, resulting in an increase of the N2O mass in our model. The global N2O emission and growth rate are in good agreement with results from Hirsch et al. (2006).

In this study, we are not interested in the atmospheric increase of the N2O mole fraction over time but its correlation with Δ17O in CO2. Assonov et al. (2013) have encountered the same issue and constructed a detrending method based on measured N2O at Mauna Loa. This detrending method assumes a constant growth rate for N2O mole fractions of α ref = 0.844±0.001 ppb/year, which is representative of tropospheric air but not suitable to the (upper) stratospheric air that we also consider in this study (e.g., upper stratospheric air samples from Thiemens et al. (1995a) with N2O mole fractions of less than 10 ppb). We modified the detrending method from Assonov et al. (2013) as described in section S3 of the supporting information to arrive at where X obs and X dtd refer to the observed and detrended mole fractions and where t obs and t ref are, respectively, the time of observation and the reference time (1 January 2007) on which the N2O mole fractions are projected. This detrending scheme is applied for (1) the validation of the N2O simulation against N2O observations, (2) the derivation of the N2O–Δ17O fit, and (3) the detrending of simulated stratospheric N2O before applying the correlation in TM5.

The modeled tropospheric N2O mole fraction is nearly constant (well mixed) at ∼320 ppb (for 1 January 2007), and the NH mole fraction is roughly 0.7–1 ppb higher than for the SH, which agrees well with the results from Hirsch et al. (2006). To test the uncertainty that is associated with our modeled N2O, we compare our model predictions for N2O with stratospheric measurements of N2O. Figure 2a shows a comparison of modeled zonal mean, yearly mean N2O with detrended experimental data from Thiemens et al. (1995a), Boering et al. (2004), Kawagucci et al. (2008), and Wiegel et al. (2013). For the measurements from Thiemens et al. (1995a), we assume that the latitude of measurements is equal to latitude of the launching site of the rocket. Our model prediction agrees well with the vertical profile from Kawagucci et al. (2008) at 39°N but overestimates the N2O mole fractions in the upper part of the vertical profile at 68°N. In Figure S1 of the supporting information we provide similar plots for each season.

Stratosphere‐Troposphere Exchange

The transport of air masses in our model, including stratosphere‐troposphere exchange (STE), is fully driven by the European Centre for Medium‐Range Weather Forecasts ERA‐Interim meteorological fields (Dee et al., 2011). Since STE is essential in this study, both for the transport of N2O and for CO2 with anomalous Δ17O, we aim to diagnose the magnitude and variability of STE. The diagnosed spatiotemporal variation of STE could help to explain variations in predicted Δ17O in the troposphere.

To diagnose the STE of CO2 in TM5, two artificial tracers were defined: CO2_trop and CO2_strat that have the same properties as the normal tracer CO2 but do not have any sources or sinks at the surface. For each time step, the tracer mass and tracer mass slopes of CO2_trop in tropospheric cells are copied from CO2, whereas the tracer mass and slopes of CO2_trop are set equal to zero for all stratospheric cells. The opposite procedure is performed each time step for the tracer CO2_strat after which all tracers in the model are transported. By diagnosing the tracer mass of CO2_trop that was transported into the stratosphere, we can determine for each time step a 2‐D field of the transport across the user‐defined tropopause. By combining the two gross exchange fluxes from CO2_trop and CO2_strat, we can calculate the net STE flux. This method allows the use of a static flat “tropopause” or a dynamic tropopause derived from the local temperature profile or the local N2O mole fraction. The transport of C17OO is tracked in a similar fashion, which allows for the calculation of the Δ17O stratospheric isoflux. Finally, we can determine the troposphere‐stratosphere flux F AS by integrating over the tropical region (30°S to 30°N) and the stratosphere‐troposphere flux F SA by integrating over the extratropical regions (outside the range 30°S to 30°N).

It is known that meteorological fields from data assimilation systems have the tendency to overestimate the Brewer‐Dobson circulation (Bregman et al., 2006; van Noije et al., 2004). The ERA‐Interim reanalysis performs better at simulating the Brewer‐Dobson circulation than its predecessor ERA‐40 (Monge‐Sanz et al., 2007), but upward transport is still too large compared to observations (Schoeberl et al., 2008). Also, the advection scheme for transport of tracer mass has an effect on the STE. Bönisch et al. (2008) showed that the “second‐order moments” scheme (Prather, 1986) is more accurate for stratospheric transport than the “slopes” scheme by Russell and Lerner (1981) that is used in our current model framework.

Given the importance of STE for our purposes and the difficulty of accurately modeling STE, we compared our diagnosed STE with data from Appenzeller et al. (1996) and Holton (1990). These studies were also used by Luz et al. (1999) to calculate the stratospheric source of Δ17O for tropospheric CO2 and O2 and in the box models by Hoag et al. (2005) and Hofmann et al. (2017). In order to determine the air mass flux crossing the tropopause, we switched off the CO2 sources and sinks at the surface and initialized the CO2 tracer with a constant mixing ratio throughout the entire domain. Using our method to track the STE of CO2 and the imposed constant CO2 mixing ratio, we inferred the air mass STE. The comparison of our derived STE and data from Appenzeller et al. (1996) and Holton (1990) is shown in Figure 3. It should be noted that the pressure levels for which the fluxes are given are not equal and also the years are different (as indicated in the legend). Still, some general conclusions about the STE in TM5 can be made. The magnitude of the STE from TM5 is for most months in between the estimates from Appenzeller et al. (1996) and Holton (1990) and the timing of the seasonality in STE agrees well. Despite the agreement, it should be noted that the range of reported values by Appenzeller et al. (1996) and Holton (1990) is large, and hence, considerable uncertainty is associated with our model derived STE. The implications of the large uncertainty in STE on the potential application of Δ17O in CO2 as tracer of GPP are further discussed in section 4.3.

Figure 3

Net air mass flux through ∼100‐hPa pressure levels from TM5 model simulation and from literature for two consecutive years. Mass fluxes from Appenzeller et al. (1996) for years 1992–1993 are given for the 118‐hPa surface. Mass fluxes from Holton (1990) are averaged over years 1958–1973; this averaged data are shown for the first years and are repeated for the second year. Monthly output was taken from our TM5 model simulation; the predicted mass flux is given for 123 hPa for years 2009 and 2010. (a) Fluxes for northern extratropical region (latitudes above 30°N). (b) Fluxes for southern extratropical region (latitudes below 30°S). (c) Fluxes for tropical region (latitudes between 30°N and 30°S). Note that for the tropical mass flux the vertical axis is shown on the right‐hand side of the figure and is reversed to facilitate easy visual comparison with the extratropical regions.

The mass fluxes from Appenzeller et al. (1996) are derived from the U.K. Meteorological Office data set (Swinbank & O'Neill, 1994) with a resolution of 3.75° longitude by 2.5° latitude and with a vertical resolution of ∼50 hPa in the lowermost stratosphere. We reproduced the STE graph by carefully extracting data points from the graphs in Appenzeller et al. (1996). STE mass fluxes by Holton (1990) are derived from climatological data of Oort (1983) specified on 5° latitude intervals and aggregated for the different seasons. Our TM5 simulation was performed with a horizontal resolution of 6° × 4° and for 25 vertical levels. The TM5 model uses hybrid sigma‐pressure levels; for the level at which the mass flux is diagnosed, the levels are almost completely isobaric.

Surface Sinks of Δ17O in CO2

Atmosphere‐Leaf Exchange

The atmosphere‐leaf exchange of CO2 is modeled using the Simple Biosphere/Carnegie‐Ames‐Stanford Approach (SiBCASA) model (Schaefer et al., 2008). To calculate photosynthesis, SiBCASA combines the C3 and C4 assimilation models (Collatz et al., 1992; Farquhar et al., 1980) with the Ball‐Berry‐Collatz stomatal conductance model (Collatz et al., 1991), from which the internal leaf CO2 concentration c can be calculated. SiBCASA is driven by ERA‐Interim meteorology with 3‐hourly time resolution and a spatial resolution of 1° × 1°. Furthermore, the spatial distribution of C3 and C4 vegetation is taken from Still et al. (2003) and SiBCASA uses a climatological mean seasonal leaf phenology based on satellite‐derived Normalized Difference Vegetation Index. SiBCASA results are first stored in full resolution in files that are subsequently read by our atmospheric transport model TM5.

The gross atmosphere‐leaf exchange fluxes can be derived from the ratio of leaf internal to atmospheric CO2 concentration c /c and the assimilation flux F A (which we obtain by scaling GPP with a factor 0.88, to take out the component that is released through autotrophic leaf respiration, similar to Ciais et al., 1997a), according to

We have used monthly averaged GPP‐weighted c /c ratios, similar to Ciais et al. (1997a, 1997b) and Peylin et al. (1997, 1999). Furthermore, our assimilation flux has 3‐hourly time resolution, whereas we assume that leaf respiration is a constant fraction of GPP. In future studies we recommend to include c /c and leaf respiration at the same temporal resolution as GPP, similar to the model by Cuntz et al. (2003a, 2003b) for δ 18O in CO2, as is also discussed in section 4.1.

In our model framework, we use the sign convention that positive fluxes increase the CO2 mass in the troposphere. The magnitude of global GPP in our model is taken from SiBCASA and is −133 PgC/year for 2011. This represents a larger uptake than the values of −100 and −120 PgC/year as used in the box models by Hoag et al. (2005) and Hofmann et al. (2017), respectively.

The average distribution of GPP‐weighted c /c for 2011 and the resulting gross atmosphere‐leaf flux F AL are shown in Figure 4. The presence of C4 vegetation in tropical Africa can be recognized clearly by the band of relatively low c /c ratios near the equator. Our c /c ratios are higher than what was used in the box models by Hofmann et al. (2017) (a fixed ratio of 0.7) and Hoag et al. (2005) (two thirds and one third for C3 and C4 vegetation, respectively, based on a study by Pearcy & Ehleringer, 1984). To prevent excessive atmosphere leaf fluxes in our model, we have imposed an upper limit such that c /c  ≤ 0.9 for all grid cells in the domain during all months of the simulation. Our global gross atmosphere‐leaf fluxes in Figure 4b exhibit a clear seasonal signal, peaking during the NH summer months. During the entire year, our atmosphere‐leaf flux is larger than the estimated −352 PgC/year from the box model by Hofmann et al. (2017), which can be explained by our higher c /c ratios and the larger magnitude of our assimilation flux F A.

Figure 4

Vegetation parameters as predicted by Simple Biosphere/Carnegie‐Ames‐Stanford Approach (SiBCASA). (a) Spatial distribution of gross primary production weighted c /c over the Earth surface averaged over the year 2011. (b) Temporal variation of global atmosphere‐leaf flux F AL as predicted by SiBCASA, partitioned over Northern Hemisphere (NH)/Southern Hemisphere (SH) and for C3/C4 vegetation.

A fraction of the CO2 that diffuses out of the leaf has equilibrated with leaf water inside the leaf. This can be expressed by dividing the gross leaf‐atmosphere flux F AL into an equilibrated and nonequilibrated part where refers to the fraction of a vegetation type and is the vegetation type‐specific equilibration constant. In our model we use and (Gillon & Yakir, 2000, 2001).

The isotopic signature associated with the gross atmosphere‐leaf exchange fluxes is determined by the signature of the source (atmospheric CO2 for F AL and F LAnoneq and leaf water for F LAeq) and kinetic fractionation during inflow and outflow of CO2 through the leaf stomata where and are the Δ17O signatures for atmospheric CO2 and for CO2 that has equilibrated with leaf water, α leaf=0.9926 is the fractionation factor for diffusion of C18OO relative to CO2 through leaf stomata (Farquhar et al., 1993), and λ kinetic=0.509 is the coefficient associated with kinetic fractionation of C17OO relative to C18OO (Young et al., 2002). A derivation and process‐based interpretation of equation (11) is given in section S4 of the supporting information. An alternative derivation for equations (11), (12), (13) is given in section S5 of the supporting information.

To calculate , we first need to determine the isotopic signature of soil water . We derive the δ 18O signature of soil water from the δ 18O signature of precipitation water, which we obtained from Bowen and Revenaugh (2003) through the portal http://www.waterisotopes.org. We use the yearly average precipitation water signatures, since the amplitude in the seasonal signal of soil water is weaker than for precipitation water and the phase of the seasonal signal can be shifted depending on the depth of the soil water in the soil layer (e.g., Affolter et al., 2015). Similar to Hofmann et al. (2017), we derive the Δ17O signature of soil water from its δ 18O signature by assuming that soil water falls on the Global Meteoric Water Line, that is, with λ GMWL = 0.528 and γ GMWL = 0.033‰ (Luz & Barkan, 2010). The resulting distribution of the has a maximum value near the equator and drops to its minimum close to the North Pole; see Figure 5b.

Figure 5

Δ17O signature of soil water and leaf water. (a) Annual mean distribution of for 2011. (b) Annual mean distribution of Δ17Osoil for 2011. (c) Temporal variation of for northern extratropical region (NET; latitudes above 30°N), tropical region (TROP; latitudes between 30°S and 30°N), and southern extratropical region (SET; latitudes below 30°S) during 2011.

The isotopic signature of leaf water (note that we use the same symbol for the Δ17O signature of CO2 that has equilibrated with leaf water, because for our selected reference level λRL these two signatures have the same value) is determined from the isotopic signature of soil water and the fractionation occurring due to the transpiration of water where α transp=1/0.9917 (West et al., 2008) is the fractionation factor of transpiration of H2 18O relative to H2 16O and λ transp is the exponent relating fractionation of H2 17O to transpiration of H2 18O where h is the relative humidity as was demonstrated by Landais et al. (2006). The resulting spatial distribution and temporal variation of is shown in Figure 5, where we used relative humidity data from ERA‐Interim. The isotopic signature attains its maximum in the African Sahara, where relative humidity is low, and has low values in the arctic region. The leaf signature for the northern and southern extratropical regions (NET and SET) exhibits a seasonal cycle of opposite phase with a peak‐to‐peak amplitude of ∼20 per meg. The in the tropical region has hardly any seasonality.

To test the effect of the soil water signature and the leaf water signature on Δ17O in CO2, we performed simulations with a spatially distributed and temporally and spatially distributed (BASE in Table 2) as well as a simulation with a constant soil water signature of −5 per meg (SOIL_CONST) and a simulation with a constant relative humidity of 0.8, which can be converted using equation (16) to λ transp=0.5156 (LEAF_CONST). The results of these simulations are given in section 3.1.3.

Respiration and Soil Invasion

The CO2 respiration flux is calculated in SiBCASA from multiple above and below ground carbon pools with different turnover rates, depending on temperature and moisture (Schaefer et al., 2008). The calculated respiration flux from SiBCASA is aggregated over a period of 1 month for each 1° × 1° grid cell. From the monthly respiration fluxes and the ERA‐Interim 2‐m temperature, the coefficient R 0 is determined (see equation (17) for its definition) and stored in a file. In our TM5 model, the CO2 respiration flux depends on temperature (and thus also on time) according to the following Q 10 relation (Potter et al., 1993) with Q 10=1.5 and T ref = 273.5 K. For T we used the 2‐m temperature from ERA‐Interim, which has a spatial resolution of 1° × 1° and a 3‐hourly time resolution, which allows us to simulate a diurnal cycle in the respiration flux. The coefficient R 0 is read from the SiBCASA output file and assures that the aggregated monthly respiration flux calculated according to equation (17) agrees with the monthly respiration flux for each cell from SiBCASA. The global respiration flux that we determine with SiBCASA for 2011 is 129 PgC/year (total respiration, including autotrophic leaf respiration).

The isotopic signature of respired CO2 (excluding the autotrophic leaf respired component, calculated similar to the net assimilation flux as described in section 2.3.1) is determined by equilibration with soil water, followed by kinetic fractionation due to diffusion through the soil column into the atmosphere where α soil=0.9928 is the kinetic fractionation factor of C18OO relative to CO2 for diffusion out of the soil column into the atmosphere (Miller et al., 1999).

The reported magnitudes of the global soil invasion flux cover a wide range: from 30 PgC/year (Stern et al., 2001) to 450 PgC/year (Wingate et al., 2009). The high soil invasion flux estimate is explained by the presence of the enzyme carbonic anhydrase in soils (Wingate et al., 2009). Similar to CO2, soil invasion fluxes of carbonyl sulfide (COS) are also affected by carbonic anhydrase (Ogée et al., 2016). The soil uptake of COS has been modeled by Launois et al. (2015) assuming that COS uptake scales linearly with v dep, the deposition velocity of molecular hydrogen to soils (based on the assumption that both processes are affected by similar soil microorganisms).

In this study, the global magnitude of the soil invasion flux is set to 30 PgC/year by default (normalized for years 2012–2013) but can be changed to any user‐specified value. Also, the spatial distribution of the soil invasion flux can be scaled with the biosphere CO2 respiration flux (i.e., F SIA∝F resp) or alternatively the hydrogen deposition velocity (i.e., F SIA∝v dep). See Table 1 for an overview of the available model settings for the soil invasion flux. To test the sensitivity of the Δ17O signature of atmospheric CO2 on the magnitude and spatial distribution of the soil invasion flux, we performed four additional simulations (RESP_240, RESP_450, HYD_240, and HYD_450 that are summarized in Table 2), for which the results are discussed in section 3.1.3.

The isotopic signature of CO2 that diffuses into soils (“ASI”) is determined from the local atmospheric Δ17O as predicted by our model. The Δ17O signature of CO2 released from the soil (“SIA”) is set equal to the signature of soil water described in section 2.3.1. Isotopic fractionation is not taken into account for the soil invasion fluxes, since the ingoing and outgoing fluxes have equal magnitude in our model (i.e., F SIA=−F ASI), and therefore, the kinetic fractionation effect on the atmospheric Δ17O budget cancels out.

Ocean Exchange

The exchange of CO2 between the atmosphere and the ocean is based on the relationship between wind speed and gas exchange over the ocean as reported by Wanninkhof (1992). The gas transfer coefficient k, in centimeter per hour, is calculated from where u is the wind speed in meter per second and Sc is the dimensionless Schmidt number. Note that the coefficient 0.31 in equation (19) is not dimensionless. Now, the two‐way CO2 exchange fluxes can be determined from where s is the solubility of CO2 in ocean water expressed in mol per cubic meter per atmosphere, is the partial pressure of CO2 in the atmosphere in unit μ atmosphere (≈0.1 Pa), and is the CO2 partial pressure difference between the ocean and the atmosphere in unit μ atmosphere. When we express k in meter per second, the CO2 fluxes have units of mol per squared meter per second. For cells that are covered with sea ice, the exchange fluxes are set to zero. The sea ice cover and wind speed data are taken from the ERA‐Interim data set (Dee et al., 2011), with a time resolution of 3 hr and a horizontal resolution of 1° × 1°. Data for solubility, CO2 partial pressure difference, and the Schmidt number are taken from Jacobson et al. (2007) with a horizontal resolution of 5° × 4° and a temporal resolution of 1 month.

The isotopic signature of ocean water is taken as ‰ (Luz & Barkan, 2010). Note that equilibration between CO2 and H2O does not result in a fractionation of our Δ17O signal, because we have taken the CO2–H2O equilibration constant as our reference line (i.e., ). We have neglected the kinetic fractionation effect for diffusion across the ocean‐atmosphere interface, since the associated fractionation factor for C18OO relative to CO2 is close to 1 (α ocean≈ 0.9992 according to Vogel et al., 1970) and the gross ocean fluxes largely cancel out (with a difference of ∼3 PgC/year on the global scale; see Figure 1).

In our model, the ocean sink for the CO2 and C17OO tracers can be determined from predefined constant CO2 and C17OO concentrations or dynamically coupled to the local concentrations of CO2 and C17OO above the ocean surface that the model calculates each time step (see Table 1 for an overview of the available model settings). For the results that we include in this paper, we always used the dynamic coupling between the ocean sink and the local atmospheric concentration.

Fossil Fuel Combustion and Biomass Burning

The CO2 fluxes from fossil combustion in our model are based on the Emissions Database for Global Atmospheric Research (EDGAR) version 4.2 from the Joint Research Centre of the European Union. The temporal resolution of this data set was improved by coupling to country and sector‐specific time profiles by the Institute for Energy Economics and the Rational Use of Energy from the University of Stuttgart. For our model we use the CO2 fluxes with a monthly time resolution and a horizontal resolution of 1° × 1°. We assign a signature of = −0.386‰ to the CO2 that is released by fossil fuel combustion, which is largely determined by the Δ17O signature of ambient O2 (Horváth et al., 2012). Laskar et al. (2016) reconstructed the same Δ17O signature for CO2 from car exhausts measured in a tunnel.

The CO2 released to the atmosphere by biomass burning is taken from the Global Fire Emissions Database version 4 (GFED4; Giglio et al., 2013). This data set is comprised by combining remotely sensed burned areas with modeled carbon pools from SiBCASA (van der Werf et al., 2010; van der Velde et al., 2014). The SiBCASA biomass burning emissions are available with a monthly time resolution and a spatial resolution of 1° × 1°. The isotopic signature of CO2 released by biomass burning is determined by the isotopic signature of ambient O2 and the wood intrinsic oxygen, resulting in an average signature of = −0.230‰ for released CO2 (Horváth et al., 2012).

Tropospheric Source of Δ17O in CO2

Tropospheric CO and Δ17O(CO) Budget

Most of the atmospheric CO2 originates from the Earth surface, where it is released directly in the form of CO2 through one of the processes as described in section 2.3. In addition, CO2 can be produced in the atmosphere through oxidation of atmospheric CO by the hydroxyl radical OH,

In this section we describe observed spatiotemporal patterns in Δ17O(CO), the processes driving Δ17O(CO) and the implications for the production of CO2 isotopologues. Subsequently, we describe in section 2.4.2 how the production of CO2 isotopologues from CO oxidation is implemented in our 3‐D atmospheric transport model.

Measurements have revealed a large positive Δ17O signature in atmospheric CO varying with season and location (measured at the per mil scale, similar to stratospheric CO2 shown in Figure 2). Huff and Thiemens (1998) report that Δ17O(CO) increases from a minimum of ∼0.3‰ during winter to a maximum of ∼2.7‰ during summer months in San Diego, California. Röckmann et al. (2002) measured a Δ17O(CO) winter minimum of ∼2‰ and summer maximum of ∼8‰ at high northern latitude stations in Alert, Canada, and Spitsbergen, Norway. At the tropical station Izaña, Tenerife, the seasonal cycle of Δ17O(CO) is much lower (∼1‰) but the annual average value is rather similar at about 5‰ (Röckmann et al., 1998a).

The most important source of the large Δ17O signature of CO is the oxidation of CO by OH (Röckmann et al., 1998a), which is not a mass‐dependent process (the rate coefficients for oxidation of C16O and C17O are approximately equal, whereas the rate coefficient for C18O is substantially higher than for C17O). This explains the observed seasonal cycle of Δ17O(CO), since OH levels are higher during the summer months than during winter months, which is more pronounced at higher latitudes. Besides this main oxidation sink with a global magnitude of ∼1 PgC/year (Holloway et al., 2000), CO is also taken up by soils at a global rate of 0.05–0.1 PgC/year (Sanhueza et al., 1998) which is a mass‐dependent process and thus not affecting Δ17O(CO).

Another contribution to the positive Δ17O in CO is the ozonolysis of nonmethane hydrocarbons (Röckmann et al., 1998b), but its effect on the Δ17O(CO) budget is less strong than the effect of the oxidation reaction. The main sources of CO (i.e., fossil fuel combustion, biomass burning and oxidation of atmospheric hydrocarbons) are considered to have a negligible contribution to the Δ17O(CO) budget (Brenninkmeijer et al., 1999).

The sources and sinks of CO and their isotopic composition are uncertain and characterized by strong spatial and temporal variability but allow us to describe the following implications for the production of Δ17O in CO2. As the OH levels increase after winter, the mass‐independent OH sink in equation (21) results in the production of CO2 with a negative Δ17O signature and the simultaneous increase in Δ17O of the remaining CO. Due to the increasing enrichment of the substrate C17O and depletion of the substrate C18O, the Δ17O isoflux from CO to CO2 will increase (i.e., become more positive or less negative) during the summer months. Since the sources of CO are largely mass dependent (i.e., with Δ17O(CO)≈0) and nearly all CO is removed through OH oxidation, we infer from mass conservation that the annual mean contribution of CO oxidation to the global budget of Δ17O in CO2 is minor (as will be confirmed in section 3.1.3.)

Production of CO2 Isotopologues

To simulate the production of Δ17O in CO2 from CO oxidation, we use climatological fields for C16O, C17O and C18O from Gromov (2013) with a global mean Δ17O(CO) signature of 5.0‰ and climatological OH fields from Spivakovsky et al. (2000). The OH fields are available for each month on a native TM5 resolution of 1° × 1° horizontally and 60 vertical sigma‐pressure levels. The climatological CO isotopologue fields are provided with a 5‐day time resolution on a T42 spectral resolution and a vertical grid of 19 hybrid sigma‐pressure levels and are regridded to match the temporal and spatial resolution of the OH fields.

We use a pressure‐dependent relation for the rate of oxidation of CO from DeMore et al. (1997) where p is the atmospheric pressure in the unit atmosphere and the unit of the rate coefficient k CO+OH is cubic centimeter per molecule per second. In our model this rate coefficient is based on climatological pressure fields derived from the orography of the Earth surface. The rate coefficients for the oxidation of the isotopologues C17O and C18O are determined with respect to the overall rate coefficient from for n = 17 or 18. The enrichment ϵ was measured in a controlled lab environment by Röckmann et al. (1998a) as ϵ 17=−0.21±1.30‰ and ϵ 18=−9.29±1.52‰ (for atmospheric pressure, according to Table 3.6 in Gromov, 2013). In a different lab study by Feilberg et al. (2002, 2005) enrichments of ϵ 17=0±4‰ and ϵ 18=−15±5‰ were found. To test the consequences of applying the different rate coefficients, we have performed simulations for both lab results (simulations CO_ROCK and CO_FEIL, as summarized in Table 2).

The oxygen in atmospheric OH likely does not have an anomalous Δ17O signature, since it equilibrates rapidly with atmospheric water vapor (Dubey et al., 1997; Lyons, 2001) and the Δ17O signature of water vapor is negligible compared to that of CO (Uemura et al., 2010). To calculate the production of CO2 isotopologues in our model, we assumed that Δ17O(OH) = 0‰, such that the temporal and spatial variation in the CO2 production fields is determined fully by that of the CO isotopologues, the OH concentration, and the rate coefficients in equations (22) and (23). To prevent interference with the stratospheric model described in section 2.2, we only apply the chemical production of Δ17O between the Earth surface and the 100‐hPa level.

From the derived C16OO, C17OO, and C18OO production fields, we calculated the associated Δ17O “flux” field. Subsequently, we scaled the C18OO fluxes such that the δ 18O fields for produced CO2 equal our assumed fixed value of 41.5‰ (see section 2.1). Finally, we scaled the C17OO flux fields to reobtain the Δ17O flux field. As mentioned in section 2.1, the motivation for using a fixed δ 18O for atmospheric CO2 is that this considerably simplifies the coupling with the hydrological cycle. This method implies that the simulated Δ17O signature is fully carried by the C17OO tracer in our atmospheric transport model.

Note that the contribution of mass‐independent CO2 through oxidation of atmospheric CO was not considered in the previous box models from Hoag et al. (2005) and Hofmann et al. (2017). Likewise, oxidation of CO is not included in our model runs with base settings (BASE), as summarized in Table 1. The resulting Δ17O in atmospheric CO2 for the simulations CO_ROCK and CO_FEIL (see Table 2) is presented and discussed in section 3.1.3.

Results

Global Model Simulations

Δ17O in Tropospheric CO2 for Base Model

In this section we show the results from the TM5 simulation with the base settings as summarized in Table 1 at a horizontal resolution of 6° × 4° and with 25 vertical levels. We started a simulation with an initial CO2 distribution from data assimilation system CarbonTracker (Peters et al., 2007, 2010; van der Laan‐Luijkx et al., 2017) and with Δ17O = 0 for each cell. After running the model for ∼10 years, we obtained a steady state (no further increase in the mean annual Δ17O signature) for the years 2012 and 2013 for which we show the results. We provide insight into the temporal and spatial patterns of modeled Δ17O in CO2 for the lowest ∼500 m of the atmosphere (lowest four model levels). The CO2 mass fluxes and corresponding Δ17O isofluxes between the different reservoirs are discussed in section 3.1.2.

In Figure 6, we show the temporal variation of monthly average Δ17O in CO2. The Hovmöller diagram in Figure 6a shows that the Northern Hemisphere experiences the largest seasonal variation and that the decrease in Δ17O occurs during the summer months for both hemispheres. Figure 6b shows the temporal variation of Δ17O in CO2 integrated over both hemispheres and for the global domain compared to box model predictions from Hoag et al. (2005) and Hofmann et al. (2017). Our 3‐D model predicts an average Δ17O signature of 39.6 per meg for CO2 in the lowest 500 m of the atmosphere, which is roughly 20 per meg lower than the prediction from the box model by Hofmann et al. (2017). This is an expected result since the exchange of CO2 with the biosphere, which represents the main sink of Δ17O, is higher in our model than for the box models. For the NH and SH we predict a mean Δ17O signature of 31.6 and 47.6 per meg and a seasonal cycle with a peak‐to‐peak amplitude of 17.7 and 5.1 per meg, respectively. The spatial and temporal patterns in simulated Δ17O confirm the potential of Δ17O as a tracer of GPP.

Figure 6

Monthly average of simulated Δ17O in CO2 for the lowest 500 m of the atmosphere using the TM5 model with base settings and a 6° × 4° horizontal resolution and 25 vertical levels. (a) Hovmöller diagram of Δ17O in CO2. (b) Time series of Δ17O in CO2 for TM5 integrated over NH, SH, and global domain, compared with predictions from box models by Hoag et al. (2005) and Hofmann et al. (2017).

The spatial distribution of Δ17O for the different seasons of 2013 is shown in Figure 7. Besides the North‐South gradient that was already visible in Figure 6, we can see that the Δ17O signature over oceans exceeds the Δ17O above land, which can be explained by the strong effect of the biosphere on atmospheric Δ17O. In addition, the tropical regions in South America and Africa have low Δ17O values during the entire year, with large zonal gradients, especially during December, January, and February and September, October, and November. Although the exchange of CO2 between the biosphere and atmosphere is highest for the tropical regions, the lowest Δ17O occurs in the NET. This is a direct consequence of the low Δ17O signatures of soil water and leaf water (see Figure 5c) in the NET compared to the tropics. Note also that fossil fuel combustion can have a strong effect on the local Δ17O signal, which explains the low Δ17O in CO2 simulated over parts of China.

Figure 7

Seasonal average distributions of simulated Δ17O in CO2 for lowest 500 m of atmosphere from the TM5 model with base settings using a 6° × 4° horizontal resolution and 25 vertical levels. (a) Seasonal average for December, January, and February (DJF) 2013. (b) Seasonal average for March, April, and May (MAM) 2013. (c) Seasonal average for June, July, and August (JJA) 2013. (d) Seasonal average for September, October, and November (SON) 2013.

CO2 Mass Fluxes and Δ17O Isofluxes for Base Model

To better understand the Δ17O budget, we analyzed the magnitudes and spatiotemporal variations of the simulated CO2 mass fluxes and Δ17O isofluxes. The definition of the Δ17O isoflux is where IF and F are, respectively, the Δ17O isoflux and CO2 mass flux from reservoir i to reservoir j. Furthermore, and are the signatures for the troposphere and for the source reservoir (which can also be the troposphere, e.g., for the isoflux from the atmosphere to the ocean IF AO). In this study we have used a reference level of  per meg, which is representative for the lowest ∼500 m of the atmosphere as described in section 3.1.1. The globally averaged yearly averaged CO2 mass fluxes and Δ17O isofluxes simulated by our TM5 model and the fluxes from the box models by Hofmann et al. (2017) and Hoag et al. (2005) are summarized in Table S2 of the supporting information.

In Figure 8 we show the global time series of the main biospheric and stratospheric Δ17O isofluxes from the model simulation with base settings for the years 2012–2013. For the global biospheric Δ17O isofluxes shown in Figure 8a, the atmosphere‐leaf isoflux IF AL has the largest seasonal variation with a peak‐to‐peak amplitude of ∼25‰ PgC/year. IF AL attains its peak (i.e., the most negative value) during the summer months in the Northern Hemisphere, similar to the seasonality in global carbon uptake by vegetation. The global equilibrated leaf‐atmosphere isoflux IF LAeq has a seasonal cycle with peak‐to‐peak amplitude of ∼10‰ PgC/year and is changing sign during the course of the year. The sign change in IF LAeq is related to the change in the isotopic signature of leaf water (see section 2.3.1) and the selected reference level . Finally, we see that global mean nonequilibrated leaf‐atmosphere isoflux IF LAnoneq is nearly constant during the year. Note that for all biospheric fluxes shown in Figure 8a the average value (and hence also the occurrence of sign changes for IF LAeq) is sensitive to the reference level .

Figure 8

Daily time series of main Δ17O isofluxes for TM5 simulation using base settings with 6° × 4° horizontal resolution and 25 vertical levels compared with independent global estimates. (a) Leaf exchange isofluxes from TM5 compared with predictions from the box model from Hofmann et al. (2017). (b) Net stratosphere‐troposphere Δ17O isoflux simulated with TM5 model compared with global estimates from Boering et al. (2004) and Kawagucci et al. (2008) based on observed N2O–Δ17O correlation and the stratospheric N2O loss rate.

The global net stratospheric Δ17O isoflux in Figure 8b has a mean value of ∼40‰ PgC/year, which agrees well with the estimates from Boering et al. (2004) and Kawagucci et al. (2008) that were derived from the observed N2O–Δ17O correlation and the estimated stratospheric N2O loss rate. Also, our global mean Δ17O stratospheric isoflux is close to the simulated flux by Liang et al. (2008). Our simulated stratospheric Δ17O isoflux has a seasonal cycle with a peak‐to‐peak amplitude of ∼40‰ PgC/year. On top of this, a relatively large day‐to‐day variability is associated with the stratospheric Δ17O isoflux. The average value of the stratospheric isoflux is not sensitive (compared to biospheric isofluxes) to small changes in the reference level, since whereas . During the Northern Hemispheric winter months, the global stratospheric influx of Δ17O is relatively high, while at the same time the biospheric sink of Δ17O is relatively weak, resulting in an increase of Δ17O in atmospheric CO2 on the global scale (which is visible in Figure 6). An overview of the temporal variation of all global CO2 mass fluxes and Δ17O isofluxes during the years 2012–2013 is given in Figures S2 and S3 of the supporting information.

The latitudinal distribution of the annual mean net CO2 mass fluxes and Δ17O isofluxes for 2012–2013 is shown in Figure 9 for different surface processes. Figure 9a clearly shows the dominance of fossil fuel combustion (“ff”) in the CO2 budget. In the warm tropics, the ocean is a source of CO2 to the atmosphere (F OA>F AO), whereas the ocean is a net sink of CO2 in the extratropics. Across all latitudes, vegetation exchange and biomass burning act as a net sink and source, respectively, and both processes peak in the tropical region. Soil invasion has no net contribution to the CO2 budget, since we assume that the uptake is equal to the release for each grid cell. The Δ17O isofluxes in Figure 9b are negative for all latitudinal bands for each surface process. The Δ17O isofluxes are dominated by the vegetation fluxes, although the contribution of fossil fuel combustion is significant in the Northern Hemisphere. Soil invasion Δ17O isofluxes are relatively small, for this simulation with base settings. More details for the contribution of different processes (e.g., the ingoing and outgoing leaf fluxes) as a function of latitude are presented in Figures S4 and S5 of the supporting information.

Figure 9

Net CO2 mass fluxes (a) and net Δ17O isofluxes (b) as function of latitude resulting from vegetation exchange (“veg”), soil invasion, ocean exchange, fossil fuel combustion (“ff”), and biomass burning (“bb”) for TM5 simulation with base settings, 6° × 4° horizontal resolution, and 25 vertical levels.

Model Sensitivity Analysis

Here we discuss the results of a sensitivity analysis for Δ17O in CO2. We have changed input values for the stratospheric N2O–Δ17O fit coefficients, the soil water and leaf water Δ17O signatures, the soil invasion fluxes, and the oxidation of atmospheric CO, as summarized in Table 2. In Table 3 we report the mean value and the peak‐to‐peak amplitude for Δ17O in CO2 for the lowest 500 m of the atmosphere for a selection of simulations with modified input settings. The peak‐to‐peak amplitude of global Δ17O was determined by fitting a sine function on the monthly values for global Δ17O for the years 2012 and 2013.

Table 3

Overview of the Mean Value and the Peak‐to‐Peak Amplitude of the Seasonal Cycle of Δ17O in CO2 for the Lowest 500 m of the Atmosphere for Different TM5 Model Simulations with Horizontal Resolution of 6° × 4° and with 25 Vertical Levels

	Mean Δ17O value (first column; per meg)
	Peak‐to‐peak Δ17O amplitude (second column, enclosed in parentheses; per meg)
Simulation	Global	NH	SH	Zotino	Mauna Loa	Manaus	South Pole
BASE	39.6 (6.5)	31.6 (17.7)	47.6 (5.1)	19.0 (36.1)	36.2 (19.5)	23.2 (2.9)	52.5 (7.4)
ST_LOWER	19.6 (5.4)	12.6 (14.4)	26.6 (3.9)	1.5 (31.4)	16.3 (15.5)	8.2 (2.9)	30.4 (5.5)
ST_UPPER	59.6 (7.7)	50.6 (21.1)	68.7 (6.3)	36.4 (40.9)	56.1 (23.5)	38.1 (2.9)	74.5 (9.2)
SOIL_CONST	40.5 (4.7)	34.7 (14.3)	46.3 (5.3)	27.8 (23.9)	38.7 (16.8)	18.5 (1.3)	51.1 (7.5)
LEAF_CONST	34.5 (6.7)	26.2 (17.8)	42.8 (4.9)	13.8 (36.1)	30.8 (19.7)	20.0 (2.3)	47.7 (7.1)
RESP_240	32.1 (6.4)	23.5 (17.4)	40.8 (4.9)	9.1 (35.3)	28.3 (19.2)	17.6 (3.0)	45.7 (7.2)
RESP_450	27.6 (6.3)	18.5 (17.1)	36.7 (4.8)	2.7 (34.5)	23.5 (18.9)	14.8 (3.0)	41.7 (7.0)
HYD_240	30.4 (6.6)	21.9 (17.7)	39.0 (4.9)	9.5 (35.6)	26.6 (19.3)	16.4 (3.1)	43.9 (7.1)
HYD_450	25.5 (6.6)	16.6 (17.5)	34.3 (4.7)	4.0 (35.5)	21.4 (19.0)	13.0 (3.1)	39.2 (6.9)
CO_ROCK	40.0 (6.5)	32.0 (17.6)	48.0 (5.1)	19.4 (36.0)	36.6 (19.4)	23.5 (2.9)	52.8 (7.4)
CO_FEIL	37.7 (6.4)	29.8 (17.5)	45.6 (5.1)	17.4 (35.5)	34.2 (19.3)	21.6 (3.1)	50.4 (7.3)

Note. The input settings for each simulation are summarized in Table 2. The global and hemispheric results are discussed in section 3.1.3, and the results for Zotino (60.80°N, 89.35°E), Mauna Loa (19.54°N, 155.58°W), Manaus (2.15°S, 59.00°W), and South Pole (90°S) are discussed in section 3.2.2. NH = Northern Hemisphere; SH = Southern Hemisphere.

According to Table 3, the change in the stratospheric N2O–Δ17O fit coefficients results in a change of roughly +20 and −20 per meg for the 95% upper (ST_UPPER) and lower limit (ST_LOWER) fits, respectively (see Figure 2b for the slope and offset of the fits) relative to the base model run (BASE). Clearly, the selected stratospheric fit is a key parameter for the resulting Δ17O in tropospheric CO2. Also, we see that the SH‐NH difference and the amplitude of global Δ17O increases when using the 95% upper limit confidence interval fit. As expected, the changes in these characteristics of the Δ17O distribution are reversed when using the 95% lower limit confidence interval fit. On annual basis, the effect of changing the stratospheric fit coefficients is smallest for the tropical forests in the Amazon and in Central Africa, as shown in Figure S6 of the supporting information, which is caused by the rapid exchange between the atmosphere and biosphere in these regions.

In the base model run, we use a spatial distribution for the soil water signature and spatial and temporal variation in the leaf water signature based on the local relative humidity, according to equations (15) and (16). We performed TM5 simulations with a constant soil water signature = −5 per meg (SOIL_CONST) and with a constant transpiration exponent λ transp = 0.5156 (LEAF_CONST; values that are also used in the box model from Hofmann et al., 2017). It should be noted that in this analysis we are changing not only the time and/or space dependency of and but also their global average value. In the base model run the global mean values are = −10.2 per meg and λ transp = 0.5160. Table 3 shows that changing to a constant  per meg has a small effect on global mean Δ17O in atmospheric CO2, whereas using a constant λ transp = 0.5160 results in a decrease of 5.1 per meg in global mean Δ17O. Finally, we see that changing the soil water signature to  per meg leads to decreases in both the North‐South difference and the amplitude of global Δ17O. In Figure S7 of the supporting information we show the annual mean difference of Δ17O for the TM5 simulations with modifications in the water signatures relative to the base model run.

The effect of a change in the global magnitude and the spatial distribution of the soil invasion flux can also be seen in Table 3. An increase from the base value of 30 to 240 PgC/year or even 450 PgC/year leads to a decrease in the global mean Δ17O signature of atmospheric CO2, where the magnitude of the Δ17O drop also depends on the spatial distribution of the soil invasion flux. For respiration scaling (BASE, RESP_240, and RESP_450) the soil invasion fluxes are mostly present in the tropical region, whereas for hydrogen scaling (HYD_240 and HYD_450) the soil invasion fluxes extend to higher latitudes, which have a lower signature and hence result in a lower Δ17O for atmospheric CO2. Also, we see in Table 3 that increasing the soil invasion fluxes leads to a small decrease in the amplitude of global and hemispheric Δ17O. In Figure S8 of the supporting information we show the global mean Δ17O distribution for changes in the soil invasion fluxes.

Finally, we show in Table 3 that incorporating the CO + OH reaction with the enrichment ϵ CO+OH from Röckmann et al. (1998a) (CO_ROCK) has a small positive effect on the resulting Δ17O of atmospheric CO2, whereas a larger negative effect was found for the fractionation factors from Feilberg et al. (2005) (CO_FEIL). Based on the enrichment coefficients given in section 2.4.2, we expect that more 18O‐enriched CO2 is produced in CO_FEIL than for CO_ROCK, which explains its lower resulting Δ17O signature in atmospheric CO2. Because the coefficients from Röckmann et al. (1998a) were also used to produce the CO isotopologue fields by Gromov (2013), we consider the results for CO_ROCK to be most representative. In Figure S9 of the supporting information we show the distribution of the annual mean anomalies for the calculated Δ17O relative to the base model run.

Local Model Simulations

Model‐Measurement Comparisons

To test the ability of our model to simulate Δ17O in atmospheric CO2, we compare our model results with a stratospheric profile measured above Sodankylä, Finland (Mrozek et al., 2016) and with tropospheric measurement series for Göttingen, Germany (Hofmann et al., 2017) and Taipei, Taiwan (Liang et al., 2017b). We selected these two data sets, because the measurement periods overlap (partially) with our model output for years 2010–2014. It should be noted that we are using a relatively coarse resolution for our model (a 6° × 4° horizontal resolution and 25 vertical levels) and that the model output are daily averages and therefore not fully representative for the observations.

In Figure 10, our TM5 model results are shown alongside the N2O mole fraction and Δ17O in CO2 profiles that were obtained from an AirCore with Stratospheric Air Sub‐sampler by Mrozek et al. (2016) above Sodankylä, Finland (67.35°N, 26.93°E) on 5 November 2014. Note that the N2O mole fractions that are reported by Mrozek et al. (2016) are not directly measured but inferred from measurements of CH4. The profile of N2O mole fractions from our simulation agrees reasonably well with the “measured” N2O profile. Contrary to the measured Δ17O in CO2 signatures, the simulated profile shows a monotonic increase with altitude. For the two observations at highest altitudes (at 24 and 39 hPa) we find that the simulated N2O is too low and that the simulated Δ17O in CO2 is too high, which suggests that the sampled air is younger than simulated in the transport model for these altitudes. The opposite is found for two of the three lowest observations (at 87 and 151 hPa) indicating that the sampled air was older than the simulated air. Note that the comparison of our model results with the data from Mrozek et al. (2016) is independent, since the experimental data from Mrozek et al. (2016) were not used as input for the N2O–Δ17O fit.

Figure 10

Comparison of vertical profiles measured over Sodankylä (67.35°N, 26.93°E; Mrozek et al., 2016) with TM5 model simulations with horizontal resolution of 6° × 4° and with 25 vertical levels. The ticks on the vertical axis coincide with the cell boundaries in the TM5 model with 25 vertical levels. (a) Stratospheric profile of N2O mole fraction. (b) Stratospheric profile of Δ17O in CO2 compared with TM5 least squares N2O–Δ17O fit simulation (BASE), 95% confidence upper limit fit simulation (ST_UPPER) and 95% confidence lower limit fit simulation (ST_LOWER).

In Hofmann et al. (2017), model predictions from an early version of our model (see section S2 of the supporting information for an overview of the differences with our current model) were compared with measurements of Δ17O in CO2 for Göttingen (51.56°N, 9.95°E) and Mt. Brocken (51.80°N, 10.62°E). We have repeated the analysis with our updated model and again find that there is a seasonal cycle in Δ17O that is driven by the biosphere. Also, we again find that our model does not show the significant drop in Δ17O that is reported based on observations (respectively a mean Δ17O of −12.8 and −108.2 per meg before and after 1 July 2011). This unexplained, large drop in the reported observations is discussed in more detail in section 6.2 of Hofmann et al. (2017). A comparison of the measured CO2 mole fraction and its Δ17O signature for Göttingen (51.56°N, 9.95°E) and Mt. Brocken (51.80°N, 10.62°E) with model predictions for the lowest level in TM5 (lowest ∼35 m) is given in Figure S10 of the supporting information.

We also compare our model predictions for Δ17O in tropospheric CO2 with measurement data obtained at the Academia Sinica campus (25.04°N, 121.61°E) and the National Taiwan University (25.01°N, 121.54°E) in Taipei, Taiwan, from Liang et al. (2017b). In Figure 11 we compare the measured and simulated CO2 mole fractions and the Δ17O signature. The uncertainty bar that is associated with the measured CO2 mole fractions is determined from the deviation between measurements taken at different times on the same day, showing the importance of local contributions and the development of the atmospheric boundary layer. The shading in Figure 11b indicates the spread related to the 95% confidence interval for the N2O–Δ17O(CO2) coefficients (slope and offset) that is used in the stratospheric module. The spread in model predictions for the different representations of the stratospheric source is substantial (∼40 per meg range) but cannot fully explain the model‐measurement discrepancy for this location. Compared to Göttingen, there is a smaller contribution of the biospheric fluxes since Taipei is surrounded by ocean. In addition, we expect a lower seasonality of the biosphere at the latitude of Taipei compared to Göttingen. Contrary to measurement series from Göttingen, our model predictions are lower (mean value of 31.1 per meg) than the Δ17O measurements from Taipei (mean value of 58.7 per meg).

Figure 11

Comparison of tropospheric measurements for the Academia Sinica campus (25.04°N, 121.61°E) and the National Taiwan University (25.01°N, 121.54°E) in Taipei, Taiwan, from Liang et al. (2017b) with daily model predictions for the lowest 35 m from TM5 with horizontal resolution of 6° × 4° and 25 vertical levels. (a) CO2 mixing ratios. (b) Δ17O in CO2. The shading indicates the spread in model estimates for the 95% confidence interval for the N2O–Δ17O fit for stratospheric CO2 (obtained from simulations ST_LOWER and ST_UPPER).

Future Measurements

The currently available measurement series for Δ17O of tropospheric CO2 have in common that the air was collected in the vicinity of the research groups that performed the measurements. Our objective here is to make use of our 3‐D model predictions to identify locations for which measurements of Δ17O in CO2 would be valuable for a better understanding of the global budget of Δ17O in CO2 and further model development. A global map of the peak‐to‐peak amplitude of simulated Δ17O in CO2 is shown in Figure S11a of the supporting information. We have selected four locations for which we describe the simulated patterns of Δ17O in CO2 in more detail.

Figure 12a shows the Δ17O signature for CO2 in the lowest 500 m of the atmosphere for a selection of locations. Zotino (60.80°N, 89.35°E) is the location of the Zotino Tall Tower Observatory (Heimann et al., 2014), where we expect a seasonal cycle of 36.1 per meg (see also Table 3), which is substantially larger than the measurement uncertainty of currently available measurement techniques (see also section 1). Also, the mean value of Δ17O at Zotino can be used to better constrain the magnitude of soil invasion fluxes (see Table 3). This site was also used in a study of the δ 18O in CO2 signal by Cuntz et al. (2002).

Figure 12

TM5 model predictions for Δ17O in atmospheric CO2 using base model settings with a horizontal resolution of 6° × 4° and with 25 vertical levels for selected locations. (a) Time series of Δ17O in CO2 for the lowest 500 m of the atmosphere for Zotino (60.80°N, 89.35°E), Mauna Loa (19.54°N, 155.58°W), Manaus (2.15°S, 59.00°W), and South Pole (90°S). (b) Longitudinal cross section through Manaus of Δ17O in CO2 for the lowest 500 m of the atmosphere in the dry season (defined here as months in the range July to October) and wet season; the vertical grid lines correspond to the longitudinal boundaries of the TM5 grid. (c) Vertical profile over Manaus of Δ17O in CO2 in the dry and wet seasons; the horizontal grid lines correspond to the vertical TM5 hybrid sigma‐pressure levels.

Mauna Loa (19.54°N, 155.58°W) and South Pole (90°S) are background stations that are famous for their long‐standing CO2 records that are operated by the National Oceanic and Atmospheric Administration and the Scripps Institution of Oceanography. The time series of Δ17O for CO2 in the lowest 500 m of the atmosphere (above the local surface) for Mauna Loa and South Pole in Figure 12a exhibit a seasonal cycle in antiphase with each other. Also, South Pole is an interesting location because we expect a high annual mean Δ17O signature. Dry air samples from the South Pole (British Antarctic Survey station) were collected in 2017 and are currently being analyzed for their Δ17O in CO2 signatures by the Centre for Isotope Research in Groningen, the Netherlands.

Also, we included model predictions of Δ17O in CO2 for Manaus (2.15°S, 59.00°W), the location of the Amazon Tall Tower Observatory (Andreae et al., 2015). Although the annual variation of Δ17O in CO2 is small in the lowest 500 m of the atmosphere for Manaus, there is a relatively strong gradient for Δ17O in the longitudinal direction across Manaus (Figure 12b) and a strong vertical gradient above Manaus (Figure 12c). Measurements in and around the Amazon region that are ongoing since February 2018 and analyzed at the LaGEE lab in Brazil could show whether these predicted features in the Δ17O distribution can be observed. The zonal mean annual mean vertical profile for Δ17O in CO2 as a function of latitude can be seen in Figure S11b of the supporting information.

Discussion

Possible Improvements of Model for Δ17O in CO2

In this section we discuss some model features that could be added to or improved with respect to our current 3‐D model for Δ17O in CO2. In our current model we represent the stratospheric source of Δ17O by simulating N2O and converting stratospheric N2O mole fractions into Δ17O signatures based on their observed correlation as described in section 2.2.1. Although we feel that this is a robust and straightforward approach, we generally prefer to simulate the actual physical processes. As more details of the production process are unfolded by the scientific community, we foresee that it becomes more feasible to implement an explicit description of the production of Δ17O in CO2 in future model versions.

To calculate the atmosphere leaf fluxes F AL and F LA, we use GPP from SiBCASA at a 3‐hourly temporal resolution and GPP‐weighted c /c values from SiBCASA at a monthly temporal resolution as described in section 2.3.1. Also, we assume in our current model that leaf respiration is a constant fraction of 12% of GPP, similar to Ciais et al. (1997a). In future studies we intend to use c /c values and leaf respiration from SiBCASA at a 3‐hourly temporal resolution to be fully consistent with the temporal resolution of GPP. In the comprehensive δ 18O model from Cuntz et al. (2003a, 2003b) these components are also simulated at the same temporal resolution.

For some input fields we use year‐specific data, such as the meteorological data ERA‐Interim (Dee et al., 2011) that drives the atmospheric transport in TM5. Also, the vegetation‐atmosphere fluxes from the SiBCASA model are calculated using the ERA‐Interim meteorology. For other input fields, we resort to annually repeating fields, such as for the CO isotopologue fields (Gromov, 2013) and the OH fields (Spivakovsky et al., 2000). In general, we preferably use year‐specific input data to capture interannual variability of the different processes. Especially for CO oxidation, we expect some interannual variability due to the irregular occurrence of wildfires (which is major source of CO; Holloway et al., 2000) that we are now not able to simulate.

Another possible improvement is the resolution of the transport model for the performed simulations, which is relatively coarse (a horizontal resolution of 6° × 4° and a vertical resolution of 25 layers). A finer horizontal resolution could lead to better agreement with local surface measurements, and a finer vertical resolution could be more representative for the STE, which is of importance to the Δ17O in CO2 budget and its ability to be used as tracer of GPP, as discussed in section 4.3. For follow‐up studies focusing on specific regions, we intend to use finer spatial resolutions.

Finally, a valuable extension of this model would be to implement a “tracer tagging” method that allows to disentangle the contributions of different processes (e.g., biosphere exchange or fossil fuel combustion) on the resulting Δ17O signature of CO2. This would allow to effectively attribute the seasonal patters, interannual variability, or local disturbances that appear in the simulated Δ17O signature to these processes. Such a tracer tagging technique was also used in the δ 18O studies from Ciais et al. (1997a, 1997b), Peylin et al. (1997, 1999), and Cuntz et al. (2003a, 2003b) to quantify the contribution of different processes to the simulated δ 18O signature for atmospheric CO2.

Required Measurements of Δ17O in CO2

In this section we discuss issues related to the measurements of Δ17O in CO2. For δ 18O in CO2 there is a vast network of well characterized measurement stations operated by the National Oceanic and Atmospheric Administration (NOAA) and collaborating organizations that measure δ 18O in CO2 at a regular basis in addition to other atmospheric compounds and meteorological variables. These flasks are typically already collected with dried air, and with new measurement techniques for Δ17O in CO2 the air in these flasks is sufficient for a high‐precision (± 20 per meg) analysis. The opportunity to start a global characterization of actual signatures followed by a monitoring effort across a subset of most interesting sites thus could be seized. In section 3.2.2, we describe in more detail four locations where measurements have, or could be, started using existing resources.

Besides these observations of Δ17O on the global scale that can help to understand the budget of Δ17O in CO2, there is also a need to measure the individual processes that affect Δ17O in CO2. The value of experiments that unravel the remaining uncertainties about the stratospheric production of Δ17O was already mentioned in section 4.1. Also, controlled laboratory measurements on the effect of plant assimilation on the Δ17O signature of atmospheric CO2 could be valuable to test the assumptions used in our current model that are for a large part based on earlier works on δ 18O in CO2 (e.g., Gillon & Yakir, 2000, 2001). Furthermore, field scale studies can help to quantify the effect of these leaf‐scale processes and entrainment on Δ17O in the atmospheric boundary layer (as done for δ13C and δ18O by Vilà‐Guerau de Arellano et al., 2019)

Potential of Δ17O in CO2 as Tracer of GPP

In this final discussion section we reflect on the potential of Δ17O in CO2 to function as a tracer of GPP. One of the main requirements for its use as tracer of GPP is that the stratospheric influx of Δ17O in CO2 can be quantified accurately. However, as described in section 2.2.3, estimates for the STE vary considerably. Combination with other tracers (e.g., 7Be, as described by Dutkiewicz & Husain, 1985) might be necessary to reduce the uncertainty in STE.

One of the key variables in the budget of Δ17O in CO2 is the c /c ratio that relates the gross exchange fluxes between atmosphere and leaf to GPP as described in section 2.3.1. Cuntz (2011) pointed out in a commentary about the GPP estimate by Welp et al. (2011) that uncertainty in the c /c ratio (or the percentage of the CO2 that diffuses into a leaf that is fixed) can have significant effects on the inferred GPP. This exemplifies the necessity to better constrain c /c , which might be achieved with δ 13C observations (Peters et al., 2018).

Similarly, the large uncertainty in the magnitude of soil invasion fluxes that was reported by Wingate et al. (2009) has implications for the potential use of Δ17O in CO2 as a tracer of GPP. If the soil invasion fluxes are underestimated, this could lead to overestimating GPP since these processes have a similar effect on Δ17O in CO2. The ongoing research on carbonic anhydrase in soils from the COS community might also lead to better quantification of the CO2 soil invasion fluxes and as such benefit the use of Δ17O in CO2 as tracer of GPP.

Finally, we address the effect of the hydrological cycle on the budget of Δ17O in atmospheric CO2. The main reason to explore the use of Δ17O as tracer for GPP instead of δ 18O was that Δ17O is hardly sensitive to the hydrological cycle which greatly simplifies its interpretation and modeling according to Hoag et al. (2005). Still, we have put much effort in calculating the Δ17O isotopic composition of different water reservoirs (e.g., soil water and leaf water, as discussed in section 2.3.1) and we find that changing these values can have a significant effect at high northern latitudes, as described in section 3.1.3. Also, a recent study by Tian et al. (2018) shows that Δ17O of precipitation collected at Indianapolis (Indiana, USA), can vary considerably within months. As such, the use of Δ17O in CO2 could be more involved than originally envisioned by Hoag et al. (2005) depending on the specifics of the application.

Conclusions

We developed a 3‐D model framework for Δ17O (defined as , with λ RL = 0.5229) in atmospheric CO2, using the terrestrial biosphere model SiBCASA and atmospheric transport model TM5. In our model framework, the stratospheric source of Δ17O in CO2 is based on the observed N2O–Δ17O correlation using available stratospheric data. We included the CO2 exchange fluxes from biosphere, oceans, and soils with the atmosphere. Also, we added the release of CO2 to the atmosphere from fossil fuel combustion and biomass burning and the production of CO2 through the oxidation of atmospheric CO.

Our 3‐D model (with base model settings) predicts an average Δ17O signature of 39.6 per meg for CO2 in the lowest 500 m of the atmosphere, which is roughly 20 per meg lower than the prediction from the box model by Hofmann et al. (2017). This difference can be attributed mostly to the larger biosphere‐atmosphere exchange in the 3‐D model (global mean F AL=−514.5 PgC/year for 2012/2013) compared to the box model (F AL=−352 PgC/year) by Hofmann et al. (2017). For the NH and SH we predict a mean Δ17O signature of 31.6 and 47.6 per meg, respectively. In addition, the Δ17O signature exhibits a seasonal cycle with a peak‐to‐peak amplitude of 17.7 for the NH and 5.1 per meg for the SH, showing the largest drop in Δ17O during the respective summer months for both hemispheres.

We showed that Δ17O model predictions are sensitive to changes in the coefficients describing the N2O–Δ17O correlation for stratospheric CO2. Also, the magnitude and spatial distribution of the soil invasion fluxes have a significant effect on Δ17O in atmospheric CO2. Furthermore, it was found that using a spatially explicit soil water signature and time‐ and space‐dependent leaf water signature has a limited effect on the resulting Δ17O in atmospheric CO2 and that the oxidation of CO has a minor effect on Δ17O in atmospheric CO2.

We compared our model predictions with a stratospheric profile of Δ17O in CO2 measured above Sodankylä, Finland (Mrozek et al., 2016), which showed good agreement indicating that our 3‐D model is able to simulate these large‐scale features of Δ17O in atmospheric CO2. Comparisons of model predictions with currently available tropospheric measurements of Δ17O in CO2 remain inconclusive due to the unexpected interannual variability for measurements from Göttingen, Germany (Hofmann et al., 2017) and the influence of local disturbances that cannot be resolved in our global model for Taipei, Taiwan (Liang & Mahata, 2015).

We identified Zotino, Russia (60.80°N, 89.35°E) as a suitable location to detect a large seasonal cycle of Δ17O in CO2 of 36.1 per meg, which is substantially larger than the uncertainty of several recently developed measurement techniques for Δ17O in CO2. Mauna Loa, USA (19.54°N, 155.58°W) and South Pole (90°S) are suitable background locations for which we predict a mean Δ17O in CO2 of 36.2 and 52.5 per meg respectively. For Manaus, Brazil (2.15°S, 59.00°W) we predict a small seasonal cycle in Δ17O in CO2 of 2.9 per meg but a strong vertical and longitudinal gradient. Measurements at the suggested locations or at comparable sites could help to further increase our understanding of the global Δ17O budget for tropospheric CO2.

Supporting Information S1

Click here for additional data file.