Simulating the integrated summertime Delta(CO2)-C-14 signature from anthropogenic emissions over Western Europe

. Radiocarbon dioxide ( 14 CO 2 , reported in 1 14 CO 2 ) can be used to determine the fossil fuel CO 2 addition to the atmosphere, since fossil fuel CO 2 no longer contains any 14 C. After the release of CO 2 at the source, atmospheric transport causes dilution of strong local signals into the background and detectable gradients of 1 14 CO 2 only remain in areas with high fossil fuel emissions. This fossil fuel signal can moreover be partially masked by the enriching effect that anthropogenic emissions of 14 CO 2 from the nuclear industry have on the atmospheric 1 14 CO 2 signature. In this paper, we investigate the regional gradients in 14 CO 2 over the European continent and quantify the effect of the emissions from nuclear industry. We simulate the emissions and transport of fossil fuel CO 2 and nuclear 14 CO 2 for Western Europe using the Weather Research and Forecast model (WRF-Chem) for a period covering 6 summer months in 2008. We evaluate the expected CO 2 gradients and the resulting 1 14 CO 2 in simulated integrated air samples over this period, as well as in simulated plant samples. Weﬁnd that the average gradients of fossil fuel CO 2 in the lower 1200 m of the atmosphere are close to 15 ppm at a 12 km × 12 km horizontal resolution. The nuclear inﬂu-ence on 1 14 CO 2 signatures varies considerably over the domain and for large areas in France and the UK it can range from 20 to more than 500 % of the inﬂuence of fossil fuel emissions. Our simulations suggest that the resulting gradients in 1 14 CO 2 are well captured in plant samples, but due to their time-varying uptake of CO 2 , their signature can be different with over 3 ‰ from the atmospheric samples in some regions. We conclude that the framework presented will be well-suited for the interpretation of actual air and plant 14 CO 2 samples


Introduction
The magnitude of anthropogenic fossil fuel CO 2 emissions is relatively well known on the global scale (Raupach et al., 2007;Friedlingstein et al., 2010) as bottom-up inventories constrain the sum of all emissions to within 6-10 % uncertainty (Marland and Rotty, 1984;Turnbull et al., 2006;Marland, 2008).But it is widely acknowledged that confidence in the estimated magnitude of these emissions reduces quickly when we consider the regional and national scale (Olivier and Peters, 2002;Gurney et al., 2009;Francey et al., 2013).At length scales of 150 km and smaller, bottom-up emission maps can differ up to 50 % (Ciais et al., 2010).This is partly a disaggregation problem that arises when nationally reported data on economic activity, energy use, and fuel trade statistics must be attributed to smaller geographic areas and more diverse processes.At the same time, there is a challenge to aggregate available bottom-up information on the level of individual roads, or power plants, or industrial complexes to a larger scale consistently.In between these two lies an important opportunity for atmospheric monitoring, as it can independently verify the reported emission magnitudes at the intermediate scales, uniquely constrained by the integrating capacity of atmospheric transport.
Published by Copernicus Publications on behalf of the European Geosciences Union.
Several atmospheric monitoring strategies for fossil fuel emissions have been applied in recent years.Most of these use spatiotemporal variations in CO 2 mole fractions (Koffi et al., 2013), often augmented with various other energy related gases such as CO (Levin and Karstens, 2007), NO x (Lopez et al., 2013), or SF 6 (Turnbull et al., 2006).An advantage of these other gases is that they can be measured continuously and relatively cheaply with commercially available analyzers, of which many have already been deployed.However, one of the disadvantages lies in attribution, as each process induces its own typical ratio of these gases to the atmosphere.An example is the much higher CO / CO 2 ratio produced by traffic emissions than by power plants.Another disadvantage is that not all of these trace gases are direct proxies for fossil fuel CO 2 release as some have totally independent, but co-located sources with the sources of anthropogenic CO 2 emissions.This is in large contrast with the one tracer that is generally considered the "gold standard" for fossil fuel related CO 2 detection: radiocarbon dioxide or 14 CO 2 (Kuc et al., 2003;Levin et al., 2003Levin et al., , 2008;;Levin and Karstens, 2007;Levin and Rödenbeck, 2008;Turnbull et al., 2006;Djuricin et al., 2010;Miller et al., 2012), reported usually as 14 CO 2 (Stuiver and Polach, 1977;Mook and van der Plicht, 1999).
Radiocarbon derives its strength for fossil fuel monitoring from the absence of any 14 C in carbon that is much older than the typical half-life time of the radiocarbon −5700±30 years (Roberts and Southon, 2007).This typically applies only to carbon in fossil reservoirs, as other carbon reservoirs are continuously supplied with fresh 14 C from exchange with the atmosphere where 14 CO 2 is produced in the stratosphere and upper troposphere (Libby, 1946;Anderson et al., 1947).In the natural carbon balance this 14 C would cycle through the atmospheric, biospheric, and oceanic reservoir until it decays.But very large anthropogenic disturbances on this natural cycle come specifically from (a) large scale burning of very old and 14 C depleted carbon from fossil reservoirs, the "Suess effect" (Suess, 1955;Levin et al., 1980), and (b) production of highly enriched 14 C in CO 2 such as from nuclear bomb tests (Nydal, 1968), or some methods of nuclear power production (McCartney et al., 1988a, b).Samples of 14 CO 2 taken from the atmosphere, but also from the oceans and biosphere that exchange with it, consistently show their dominant influence on the 14 CO 2 budget of the past decades (e.g.: Levin et al., 1989Levin et al., , 2010;;Meijer et al., 1996;Nydal and Gislefoss, 1996;Levin and Hesshaimer, 2000;Randerson et al., 2002;Naegler and Levin, 2006;Graven et al., 2012a, b).
Monitoring of atmospheric 14 CO 2 is done through several methods.One commonly applied approach is by absorption of gaseous CO 2 into a sodium hydroxide solution from which the carbon content is extracted for 14 C / C analysis either by radioactive decay counters, or converted into a graphite target for analysis by accelerator mass spectrometry.The air flowing into the solution typically integrates the absorbed CO 2 with sampling time of days, weeks, or even longer periods.
While there is a new technique, which uses integrated flask sampling (Turnbull et al., 2012), the other method generally used is to collect an air sample in a flask, which is filled within less than a minute and thus representative of a much smaller atmospheric time-window.Compared to these, at the other end of the time spectrum is the use of plants to sample 14 C / C ratios in the atmosphere through their photosynthetic fixation of atmospheric CO 2 .Depending on the species these integrate over sampling windows of a full growing season (annual crops, fruits - Shibata et al., 2005;Hsueh et al., 2007;Palstra et al., 2008;Riley et al., 2008;Wang et al., 2013) or longer (trees, tree-rings -Suess, 1955;Stuiver and Quay, 1981;Wang et al., 2012).
An effective monitoring strategy for fossil fuel emissions is likely to take advantage of all methods available to collect 14 C samples, and combine these with high resolution monitoring of related gases (e.g.CO, SF 6 ).Levin andKarstens (2007), van der Laan et al. (2010) and Vogel et al. (2010) already demonstrated the viability of a monitoring method in which observed CO / CO 2 ratios are periodically calibrated with 14 CO 2 to estimate fossil fuel emissions at high temporal resolutions.More recently, this strategy was also employed by Lopez et al. (2013), where additionally the CO 2 / NO x ratios were used to estimate fossil fuel derived CO 2 from continuous CO and NO x observations in Paris.Turnbull et al. (2011) showed for the city of Sacramento, that using a combination of 14 CO 2 and CO observations can reveal structural detail in CO 2 from fossil fuel and biospheric sources that cannot be obtained by CO 2 measurements alone.Van der Laan et al. (2010) and recently Vogel et al. (2013) showed that the agreement between modeled fossil fuel CO 2 estimates and observations of 14 C-corrected CO can be further improved by including 222 Rn as a tracer for the vertical mixing.Finally, Hsueh et al. (2007) and Riley et al. (2008) used 14 C / C ratios in corn leaves and C3 grasses to reveal fossil fuel emission patterns on city, state, and national scales.Given so many different methods to use 14 C in monitoring strategies, its increasing accuracy, reduction in required sample size, and decreasing costs, it is likely that this tracer will play a more important role in the future of the carbon observing network.
The quantitative estimation of fossil fuel emissions from all of the 14 C-based monitoring strategies above requires different methods and emphasizes different terms in the 14 CO 2 budget.For example, interpretation of 14 C in air samples from aircraft requires detailed dispersion modeling of surface emissions into a highly dynamic atmosphere, while interpretation of monthly integrated air samples from tall towers requires the inclusion of the re-emergence of old 14 C signals after longer turn-over in the oceans and biosphere.In a recent publication (Bozhinova et al., 2013), we showed that the interpretation of growing season integrated plant samples additionally requires simulation of location and weather dependent photosynthetic uptake and plant development patterns.A successful 14 C monitoring strategy will thus depend strongly on our ability to capture these diverse processes on diverse scales.
In this work, we present a newly-built framework designed to interpret 14 CO 2 from different types of samples and from different monitoring strategies.The framework includes atmospheric transport of surface emissions of total CO 2 and 14 CO 2 on hourly scales on a model grid of a few kilometers, but integrates signals up to seasonal time scales and even down into the leaves of growing crops (maize and wheat).Both regional transport and plant growth are based on meteorological drivers that are kept consistent with large-scale weather reanalyses.In addition to fossil fuel signals in the atmosphere and in plants, we simulate the spread of nuclear derived 14 C release from major reprocessing plants and from operational nuclear power production plants across Europe based on work of Graven and Gruber (2011).We applied our framework to the European domain for the summer of 2008.After explaining the components of the framework (Sect.2) we will demonstrate its application (Sect.3.1), assess the fossil and nuclear derived 14 C gradients across Europe (Sect.3.2), and simulate the signal that will be recorded into annual crops growing across the domain (Sect.3.3).We will evaluate its potential benefits compared to simpler but less realistic fossil fuel estimation methods from integrated samples alone (Sect.3.4).We will conclude with a discussion (Sect.4) of the application of this framework to actual measurements and recommendations for future studies.

Methods
2.1 The regional atmospheric CO 2 and 14 CO 2 budget The regional CO 2 mole fractions and 14 CO 2 signature of the atmosphere observed at a particular location are described in Eqs. ( 1) and (2), following the methodology used by Levin et al. (2003), Turnbull et al. (2006), Hsueh et al. (2007), Palstra et al. (2008) and described thoroughly in Turnbull et al. (2009b).Here the x and CO 2x (or 14 CO 2x ) indicate the 14 CO 2 signature of CO 2 (or 14 CO 2 ) mole fractions of particular origin, expressed in the index as follows: obs -observed at location, bg -background, fffossil fuels, p -photosynthetic uptake, r -ecosystem respiration, o -ocean, n -nuclear and s -stratospheric.
Several of the terms in both equations can be omitted or transformed in our study, as described next.
We set p = bg similar to the approach in Turnbull et al. (2006) as the calculation of 14 CO 2 accounts for changes in the signature of the photosynthesized CO 2 flux due to fractionation.The atmosphere-ocean exchange in the northern Atlantic makes the region generally a sink of carbon (Watson et al., 2009), but we assume that its transport to our domain is uniform and captured by the inflow of background air and thus also carries the signature bg .For the ecosystem respiration and ocean exchange the terms r and o can be also written as bg + dis bio and bg + dis ocean , where the disequilibrium terms ( dis ) describe the difference between the signature of the carbon in the particular reservoir and the current atmospheric background.These differences arise from the past enrichment of the atmosphere with 14 CO 2 from the atmospheric nuclear bomb tests since the 1960s.In the following decades this enrichment was incorporated into the different carbon reservoirs (Levin and Kromer, 1997;Levin and Hesshaimer, 2000) and currently these terms are of dominant importance only in particular regions of the globe.For our domain both terms are considered of much smaller influence than the dominant effect of the fossil fuels and are consequently omitted (Levin and Karstens, 2007;Hsueh et al., 2007;Palstra et al., 2008;Turnbull et al., 2009b;Naegler and Levin, 2009a, b;Levin et al., 2010).Because we currently do not correct for this, the omission of the biospheric disequilibrium in the region and period of our study will likely result in a small bias in our results, as our atmosphere will be less enriched during the period of peak biospheric activity.For the northern hemisphere Turnbull et al. (2006) estimates an overestimation of fossil fuel CO 2 by 0.2-0.5 ppm or up to 1.3 ‰ enrichment in 14 CO 2 due to this lack of disequilibrium influence, while Levin et al. (2008) evaluates this influence on the observational sites in Germany to be within 0.2 ppm or about 0.5 ‰ enrichment.The intrusion of 14 CO 2 -enriched stratospheric air can be of importance for observations in the upper troposphere or higher, however in our case this term can be considered as part of the background, as the stratospheric 14 CO 2 is already well mixed by the time it reaches the lower troposphere.
Most studies ignore the effects of anthropogenic nuclear production of 14 CO 2 on the atmospheric 14 CO 2 since on the global scale this production averages to the smallest contribution, compared to the other terms (Turnbull et al., 2009a) and few try to quantify and correct for it in observations taken nearby nuclear power plants (Levin et al., 2003).However, Graven and Gruber (2011) showed that the regional influence of a dense nuclear power plant network cannot be ignored.They estimated the potential bias in the recalculation of fossil fuel CO 2 due to nuclear power plant production is on average between 0.5 and 1 ppm for Europe, but the horizontal resolution of their transport model (1.8 Sellafield, United Kingdom), which generally have higher than average emissions of 14 CO 2 (McCartney et al., 1988a).
Particularly the site of La Hague is estimated to be the largest current point-source of 14 CO 2 emissions in the world, in recent years accounting for more than 10 % of the global budget of nuclear produced 14 CO 2 (Graven and Gruber, 2011).The magnitude of this source and its spatial location close to the major fossil fuel emitters in Europe pose a challenge in estimating the uncertainty with which the method of recalculating fossil fuel CO 2 can be applied in the region.All these considerations allow us to simplify Eqs. ( 1) and (2) to Eqs. ( 3) and (4). (3) obs CO 2obs = bg (CO 2bg + CO 2p + CO 2r ) The instantaneous 14 CO 2 signature of the atmosphere is calculated using Eq. ( 4), using the specific signatures for various sources of CO 2 (various terms) as listed below: 1. Fossil fuels are entirely devoid of 14 CO 2 and their ff = −1000 ‰.
2. The nuclear emissions are of pure 14 CO 2 and in this formulation n is the 14 CO 2 signature that a pure 14 CO 2 sample would have.We calculate it using the activity of pure 14 CO 2 sample in the formulation of 14 CO 2 as follows: where N a = 6.022 × 10 23 mol −1 is the Avogadro constant, λ = 3.8534 × 10 −12 Bq is the decay rate of 14 C and m14 C = 14.0 g mol −1 is the molar mass of the isotope.In a sample of a pure 14 CO 2 there is no fractionation and the calculation of 14 CO 2 (Stuiver and Polach, 1977;Mook and van der Plicht, 1999) can be simplified to the ratio between the activity of the sample and activity of the referenced standard A ABS = 0.226 Bq g C −1 (Mook and van der Plicht, 1999): The resulting n ≈ 0.7 × 10 15 [‰] is much higher than any of the other signatures, but this is balanced by the concentrations of the 14 CO 2 , which are only a very small fraction (∼ 10 −12 ) of the observed CO 2 concentrations.
3. Finally, we use bg from monthly observed 14 CO 2 at the high alpine station Jungfraujoch (3580 m a.s.l., Switzerland) (Levin et al., 2010), which is considered representative for European 14 CO 2 background.These are shown in red on Fig. 3a.
We note that the choice of background can be crucial for the estimation of obs and consequently for the recalculation of CO 2ff .Local influences captured in the background might modify the seasonality of the derived obs and result in biases when applied to observations from other locations.These influences include local fossil fuel or nuclear signals, biospheric enrichment or modified vertical mixing during parts of the year (Turnbull et al., 2009b).
The transport and resulting spatiotemporal gradients in total CO 2 and 14 CO 2 over Europe are simulated with WRF-CHEM model, described next.

WRF-CHEM
For our simulation with WRF-Chem (version 3.2.1)(Skamarock et al., 2008) we use meteorological fields from the National Centers for Environmental Prediction Final (FNL) Operational Global Analyses (NCEP, US National Centers for Environmental Prediction, 2011) at 1 • ×1 • for lateral meteorological boundary conditions, which are updated every 6 h.We model the atmospheric transport and weather for the period between April and September 2008 including.We use three domains with horizontal resolution of 36, 12 and 4 km and, respectively, 60×62, 109×100 and 91×109 grid points, centered over Western Europe and the Netherlands, as shown in Fig. 1.Our vertical resolution includes 27 pressure levels, 18 of which are in the lower 2 km of the troposphere, and the time step used is 180 s in the outer domain.Important physics schemes used are the Mellor-Yamada Nakanishi and Niino (MYNN2.5)boundary layer scheme (Nakanishi and Niino, 2006), the Rapid Radiation Transfer Model (RRTM) as our longwave radiation scheme (Mlawer et al., 1997), and the Dudhia shortwave radiation scheme (Dudhia, 1989).We use the Unified Noah Land-Surface Model (Ek et al., 2003) as our surface physics scheme and additionally use timevarying surface conditions, which we update every 6 h.
We use separate passive tracers for the different CO 2 terms in Eq. ( 4).We prescribe our initial and lateral boundary conditions for the background CO 2 , while the biospheric uptake, respiration, fossil fuel CO 2 and nuclear 14 CO 2 are implemented with surface fluxes only, which are prescribed and provided to the model every hour.Once CO 2 leaves our outer domain it will not re-enter it again.This setup reflects our interest in the recent influence of the biosphere and anthropogenic emissions.For this reason we will avoid using direct results from the outer domain, and instead use only the nested domains, where boundary conditions for all tracers are provided through their respective parent domain.
The background (CO 2bg ) initial and boundary conditions are implemented using 3-D mole fraction output from Car-bonTracker (Peters et al., 2010) for 2008 at 1 • × 1 • resolution and interpolated vertically from 34 to 27 levels using the pressure fields.The CO 2 lateral boundary conditions are added to the standard meteorological boundary conditions and also updated every 6 h.Our biospheric fluxes (CO 2r and CO 2p ) are generated using the SiBCASA model (Schaefer et al., 2008;van der Velde et al., 2014), which used meteorological fields from the European Centre for Medium-Range Weather Forecasts (ECMWF).It provides us with monthly averaged gross photosynthetic production (GPP) and terrestrial ecosystem respiration (TER) at 1 • × 1 • resolution.Due to the coarse resolution of the SiBCASA model, we find land-use categories in the higher resolution map of WRF that are not in the natural land-use map of SiBCASA.To address this issue, we ran 9 simulations with SiBCASA prescribing a single vegetation category, alternating through all the vegetation categories to produce biospheric fluxes for the different land-use categories within the resolution of WRF.For temporal interpolation of the monthly fluxes, we scale the GPP and TER with the instantaneous WRF meteorological variables (temperature at 2 m and shortwave solar radiation) following the method described in Olsen and Randerson (2004).

Atmos
Anthropogenic (fossil fuel) CO 2 emissions (CO 2ff ) are from the Institute for Energy Economics and the Rational Use of Energy (IER, Stuttgart, Pregger et al., 2007) at a horizontal resolution of 5 (geographical) minutes over Europe in the form of annual emissions at the location and temporal profiles to add variability during different months, weekdays and hours during the day.These are then aggregated to every WRF domain horizontal resolution and updated every hour for the duration of our simulation.The emissions are introduced only at the lowest (surface) level of the model.Anthropogenic (nuclear) 14 CO 2 emissions ( 14 CO 2n ) are obtained by applying the method described in Graven and Gruber (2011) for the year of 2008.We used information from the International Atomic Energy Agency Power Reactor Information System (IAEA PRIS, available online at http://www.iaea.org/pris)for the energy production of the nuclear reactors in our domain and reported 14 CO 2 discharges for the spent fuel reprocessing sites (van der Stricht and Janssens, 2010).The data is available only on annual scale and once converted from energy production to emissions of 14 CO 2 , these are scaled down to hourly emissions, assuming continuous and constant emission during the year.This is likely true when the nuclear reactors are operating, however, in reality regular maintenance and temporary shutdowns of individual reactors would result in periods of weeks and sometimes months of lower energy production and subsequently lower 14 CO 2 discharge.We will further comment on these assumptions in our Discussion (Sect.4).

Integrated 14 CO 2 air and plant samples
Integrated 14 CO 2 samples ( absorption ), where the sampling rate is usually constant (e.g. in various CO 2 absorption setups), are represented with the concentration-weighed timeaverage 14 CO 2 signature for the period and height of sampling, as seen in Eq. ( 7).When actual sampling is restricted to specific wind conditions or times-of-day, we include this in our model sampling scheme as well.
Plant samples ( plant ) integrate the atmospheric 14 CO 2 signature with CO 2 assimilation rate which varies depending on various meteorological and phenological factors.Photosynthetic uptake and the allocation of the assimilated CO 2 in the different plant parts strongly depend on the weather conditions and plant development.To simulate such samples we use WRF meteorological fields in the crop growth model SUCROS2 (van Laar et al., 1997) and use the modeled daily growth increment as a weighting function (averaging kernel) on the daytime atmospheric 14 CO 2 signatures (Bozhinova et al., 2013).For each location we use the same sowing date and the model simulates the crop development until it reaches flowering, when we calculate plant .More explicitly these integrated sample signatures are calculated as follows: where X t is the growth increment at time t, which in the case of SUCROS2 simulation is the dry matter weight increment at day t.  3 Results

Model evaluation -how realistic are our CO 2 and 14 CO 2 simulations?
The meteorological conditions for 2008 that were simulated by WRF and used for the plant growth simulation in SUCROS2 were previously assessed in Bozhinova et al. (2013).Here we assess the model performance compared to observed CO 2 fluxes, CO 2 mole fractions, and boundary layer heights.Figure 2 shows this comparison at the observational tower of Cabauw, the Netherlands (data available at http://www.cesar-observatory.nl).The simulated net CO 2 flux (NEE) compares well to observations with a rootmean squared deviation (RMSD) of 0.26 mg CO 2 m −1 s −1 and correlation coefficient (r) for the entire period of 0.70, which is even higher in clear days.Overestimates of NEE occur during cloudy conditions, which are notoriously difficult to represent in many mesoscale models.The CO 2 mole fractions compare well to observations (Vermeulen et al., 2011) and overall model performance is similar to other studies for the region (Tolk et al., 2009;Meesters et al., 2012).Similar to Steeneveld et al. (2008), Tolk et al. (2009), Ahmadov et al. (2009) the night-time stable boundary layer poses a challenge to the model.Note that the skill at modeling the boundary layer height can be of a particular importance for the correct simulation of the CO 2 budget, as it controls the diurnal evolution of the CO 2 mole fractions (Vilà-Guerau de Arellano et al., 2004;Pino et al., 2012).Thus, we have included this comparison in the last panel of Fig. 2.More detailed statistics for this and other stations and observations are listed in Table 1.We show the mean difference between the predicted and observed time series, with the according RMSD, and calculated correlation coefficient and coefficient of determination (Willmott, 1982) for each location.While in Table 1 we show the statistics for the daily time-series, we also evaluated their hourly and daytimeonly counterparts and the differences between each.Overall, our comparison shows that although the model overestimates the night-time CO 2 concentrations, it captures the observed daytime CO 2 mole fractions features and their variability on scales of hours to days satisfactorily over the full period simulated for Cabauw.
We next analyze the results for the 14 CO 2 signature corresponding to these CO 2 mole fractions to evaluate our skill at modeling the large scale 14 CO 2 over Europe.1.For the high-altitude locations of Jungfraujoch and Schauinsland the model topography differed significantly from the altitude of the observational site.Similar to the procedure Table 1.The observational sites with data used in this study and statistics for the daily concentrations of CO 2 and CO 2ff estimated from CO observations, hourly flux CO 2 and monthly integrated 14 CO 2 observations as compared with modeled results.Here Pi-Oi represents the mean model-data difference and σ Pi-Oi is the spread of this difference.Both expressions carry the units described in the header of each section.r and d are respectively the Pearson's coefficient of correlation and the coefficient of determination.n is the number of members used in the statistical analysis.

Site
Latitude [ described in Turnbull et al. (2009b) we sampled a model layer in the free troposphere instead of at the modeled surface to better represent the observations.At all other sites we sample the pressure-weighted signature of the boundary layer, applying a minimum boundary layer height of 350 m during the night to avoid sampling too low surface signatures in a too stable nighttime boundary layer.The comparison shows we capture reasonably well the seasonal cycle for most sites, however the model generally underestimates the 14 CO 2 .This is partly caused by the omitted biospheric disequilibrium term, which accounts on average for up to 1.5 ‰ at these latitudes.Additional bias could be introduced through our choice of background site.In their study, Turnbull et al. (2009b) showed that the signature of free tropospheric air in the northern-hemispheric mid-latitudes can vary within 3 ‰ and additionally the signatures at mountain background sites (as Jungfraujoch) are slightly influenced by local fossil fuel emissions.
In the lowest left panel of Fig. 3 we show the comparison for Heidelberg, where observations are collected as weekly night-time (between 19:00 and 07:00 local time) integrated samples.On higher temporal resolution our model estimates reproduce the temporal variations of the observations well.Still, the already discussed underestimation in 14 CO 2 is also present at this site, which is located near a large urban area with considerable fossil fuel emissions.During the period from May to August, this underestimation is on average 5 ‰ in the model (∼ 1.8 ppm of fossil fuel CO 2 ).In the lowest right panel of Fig. 3 we show the comparison between the observed and modeled signatures at Lutjewad for the wind-specific measurements at this site in addition to the observed monthly samples that were continuously integrated.The monthly 14 CO 2 observations for 2008 from this location show atypical seasonality with a lack of the expected summer maximum, and 10 to 20 ‰ lower 14 CO 2 than the observations in Jungfraujoch and Schauinsland in that year.Although this suggests a large fossil fuel CO 2 signal for 2008, we could not find further evidence of this in the rest of the Lutjewad observational record (CO, CO 2 ), nor in the selected southerly wind sector data (van der Laan  et al., 2010), which our model matches rather well.Since the measurements themselves seem valid, this feature in the continuous monthly Lutjewad 14 CO 2 data remains unexplained.We will however take a closer look at the temporal variability of the different 14 CO 2 components and the general model performance at Lutjewad for the more accurately simulated southerly wind sector.
Figure 4 shows the 6-month hourly comparison of simulated and observed CO 2 and fossil derived CO 2 for Lutjewad.The latter is derived from 14 C-corrected high-resolution CO observations (van der Laan et al., 2010).Statistics for the comparison are also shown in Table 1.The fossil fuel signal dominates over any variability in the background, clearly defining periods with enhanced transport of fossil fuel CO 2 to the location (late April, start of May, start of July, start of August) as compared to less polluted air transported from the North Sea (mid-May, mid-June).The larger mismatch in particular periods (second half of April, start of May) can be attributed to the specific way the CO observations are calibrated using the 3-year fit of the 14 C-CO ratio at the site.While this would ensure that on an annual scale the actual 14 C-CO relation is reached, on the bi-weekly scale of the 14 C observations this sometimes results underestimation of the 14 C-CO ratio compared to the observed values and consequently overestimation of the estimated fossil fuel CO 2 .For more information, see van der Laan et al. ( 2010).
In the last panels we see this influence on the resulting 14 CO 2 signature and especially its high temporal variability that is not captured in the typically integrated monthly samples.Note that even though station Lutjewad is far away from nuclear emission sources, the signal from nuclear activity (shown in the last panel) can sometimes be of the same order of magnitude as the fossil fuel signal.This shows that it is important to evaluate the nuclear influence at every measurement site using a model like presented here, as it will contribute to the uncertainty in the recalculation of fossil fuel CO 2 .

Fossil fuel vs. nuclear emissions influence on 14 CO 2
The lowest 14 CO 2 values in the domain are modeled in the regions with high fossil fuel emission in Germany (the Ruhrgebiet), and the highest 14 CO 2 is near the large emitting sites in western France and UK.This pattern can be clearly seen in Fig. 5a-c where results averaged over the lower 1200 m of the atmosphere over the full 6 months are shown.Note that the nuclear enrichment reaches much higher amplitude than the opposite effect by the fossil CO 2 , but its influence on the atmospheric 14 CO 2 is usually restricted to the vicinity of the average nuclear power plant reactors.The influence is more pronounced in the western part of our domain, where it captures the influence from the spent fuel reprocessing plant in La Hague (France) and several newer generation nuclear reactors in the UK.Even then, the influence of the nuclear enrichment averaged over 6 months is typically about 1 to 6 ‰ in areas that are not in direct vicinity of the sources.As a comparison, the fossil fuel influence in our domain on the same temporal and spatial scale is mostly between −3 and −15 ‰ outside the very polluted area of the Ruhrgebiet, Germany.
As the nuclear enrichment will (partially) mask the effect of fossil fuel CO 2 on the atmospheric 14 CO 2 , we show in Fig. 5d the average 6-month ratio of the influences due to nuclear and fossil fuel sources in our domain.Again, in most of the eastern and central parts of our domain the nuclear influence is less than 10 % the fossil fuel influence.This differs from the western part of our domain, where the ratio varies between 3 times smaller to about the same magnitude as the fossil fuel contribution and even to a more than 5 times larger influence in the area around the nuclear sources.The area affected depends on the strength of the source, and in our case the influence of most water-cooled reactors rarely exceeds the grid cell of the source, while for the gas-cooled reactors the influence can be seen up to 50 km distance.These findings are consistent with Graven and Gruber (2011).The magnitude of the enrichment and size of the area influenced are both highly variable and strongly dependent on the atmospheric transport.As a result, in months with dominant easterly winds the nuclear enrichment has a minimum effect in our domain, as most of the nuclear emissions are transported towards the Atlantic ocean and out of our area of interest.However, in months with dominant westerly winds, which is the prevailing wind direction, the nuclear 14 CO 2 spreads widely over the domain.Graven and Gruber (2011)  70 % confidence interval of the emission factors ("low estimate" and "high estimate" runs).In Fig. 6 we show these results as the absolute difference when compared to the mean run.While our largest source of nuclear emissions -located in La Hague, France, has directly reported emissions of 14 CO 2 and is thus not subject to uncertainty in the emission factors, considerably higher or lower 14 CO 2 signatures could be associated with the nuclear estimates in the United Kingdom, southern Germany and central France.
For sites located in northern and central France, southern Germany and the UK the nuclear enrichment means that corrections are needed that account for the nuclear influence in the observed 14 CO 2 before estimating the fossil fuel influence.As an illustration, we show in Fig. 7 the influence of the different anthropogenic emissions for three locations with different characteristics in our domain: Cambridge (UK), Cabauw (the Netherlands) and Kosetice (Czech Republic).The locations were chosen to be in rural or agricultural areas, without large local CO 2 emissions.As seen in Fig. 7, the western part of our domain (represented by Cambridge) has an equal influence from fossil fuel and nuclear emissions; the center (represented by Cabauw) experiences some events with relatively high nuclear emissions influence, but is influenced mostly by the very high fossil fuel emissions in this region (on average about 3 times higher than in Cambridge).In the east (represented by Kosetice) there is no significant signal of influence of nuclear emissions, but the influence of fossil fuel emissions is also considerably lower.

14 CO 2 plant vs. atmospheric samples
In our previous work (Bozhinova et al., 2013) we described a method to model the 14 CO 2 in plant samples as the first step in quantifying the differences between such samples and integrated atmospheric samples.Here we build on this work by calculating the plant signature resulting from uptake of spatially and temporally variable atmospheric 14 CO 2 .The results for modeled samples from maize leaves at flowering, are shown in Fig. 8. Clearly, spatial gradients in 14 CO 2 in plants are sizeable compared to the measurement precision of approximately 2 ‰.The regions with high influence from anthropogenic emissions from Fig. 5, namely the Ruhrgebiet in western Germany and the Benelux are also visible in the modeled plant signature, and so are some hot spots around larger european cities, like Frankfurt, Paris, London and others.It is important to point out that in addition to fossil fuel and nuclear gradients, plants develop at a different rate in different parts of the domain, and even the different parts of a plant (roots, stems, leaves, fruits) grow during different time periods.The plant-sampled 14 CO 2 includes the effect of the covariance between the atmospheric 14 CO 2 variability and the variability in the assimilation of CO 2 in the plant during growth, which is absent in traditional integrated samples where the absorption of CO 2 is based on constant flow rate through an alkaline solution and thus only varies with the CO 2 concentration present in the flow (Hsueh et al., 2007).In Fig. 9 (left) we show this effect of the plant growth on the resulting plant 14 CO 2 signature when comparing the resulting plant signature with the daytime atmospheric average we provide to our crop model.We should stress, that this is the magnitude of the error one should expect if the plantsampled 14 CO 2 is assumed equal to the atmospheric mean 14 CO 2 for the growing period of the plant.For many parts of Europe in our simulated period this error is approaching the measurement precision of the 14 CO 2 analysis (of approximately ±2 ‰).In the region located between the areas with high fossil fuel and large nuclear emitters, however, the magnitude of the error can be several times larger.This is likely due to the absorption of some very high signature values in periods when the wind direction is directly from the nuclear source.Actual plant samples, taken during different period than the one investigated here (namely 2010-2012), will be used to further investigate these signatures in a follow-up publication.
We also evaluated the bias that would be introduced if the nuclear influence is not included in the modeling of the plant samples.We show this on Fig. 9 (right) as the difference between the plant signatures when the nuclear influence is included or excluded from the simulation.For the continental part of our domain this bias mostly stays within 0-4 ‰, while in the United Kingdom it ranges from 2-8 ‰ and higher.This suggests that also when interpreting plant samples, the ability  to correctly account for nuclear influences such as through a modeling system could be important.

Direct estimation of the fossil fuel CO 2 emissions
While the entire emission map of Europe might be difficult to verify, most of the fossil fuel CO 2 emissions are produced at only a number of locations.For instance, 10 % of all emissions in our domain come from only 30 grid cells and more than half of these are located in densely populated cities or urban conglomerations.This might provide an opportunity for a better fossil fuel estimate of the highest emitting regions in Europe even when only selected locations are visited in a plant sampling campaign.One could for instance assume that the 14 CO 2 signatures in plants in these high-emission areas directly reflect the local anthropogenic sources, and a straightforward determination of their 14 CO 2 signature would suffice to estimate emissions using a simple box-model approach.We show in the following analysis that this simplification can lead to large errors though, and a more complete modeling framework like ours is needed for a proper interpretation of 14 CO 2 .
In our modeling framework, we know the exact emissions we prescribe in each grid box as well as the resulting atmospheric 14  emissions over a 60 km × 60 km area around 25 large European cities, mix them through a 500 m deep boundary layer (typical 24 h average for our domain), and assume the air to have a residence time of 3.3 h (corresponding to a typical wind speed of 5 ms −1 through a 60 km domain), we can make a simple estimate of the resulting 14 CO 2 signature relative to the background from Jungfraujoch.This box-model estimate is shown in Fig. 10 as the continuous straight line, in which the downward slope with increasing emissions is controlled mostly by the assumed residence time and the prescribed boundary layer height.
If we compare this linear relationship with the simulated 14 CO 2 signatures over these cities simulated with the full model developed in this paper (including its detailed horizontal advection, vertical mixing, and nuclear influence), one can see the large variability and substantial bias one would incur using the simple box-model approach.Up to 8 ‰ differences from this line would be found for Paris and Cologne, while the nuclear influence would lift Birmingham plant samples back toward the Jungfraujoch background 14 CO 2 despite its emissions being similar to Berlin.Even if the full model-derived slope of approximately −4.85 ‰ per 10 000 mol km −2 h −1 could be reproduced with the boxmodel, the coefficient of determination (R 2 ) would be just over 0.7, meaning that close to 30 % of the spatial variance in emissions across Western Europe will not be captured in the simple approach.We therefore caution strongly against a simplified quantitative interpretation of 14 CO 2 signatures, both in plants and in the atmosphere.While the left figure shows the error that should be expected if the plant growth is not taken into account and the plant signature is assumed to be equal to the atmospheric average, the right one shows the error that will be introduced if nuclear emissions of 14 CO 2 are not accounted for in the model simulation.
With a typical 14 CO 2 single measurement precision of about ±2 ‰ and the full model-derived slope given above, we can tentatively estimate that even a perfect modeling framework will have a remaining uncertainty of 4000 mol km −2 h −1 for area-average emissions in these top-25 emitters over Europe.This is quite substantial (20-50 %) for most of them, with the possible exception of the cities in the German Ruhr area (5-15 %).We therefore see an important role for a monitoring program of 14 CO 2 signatures in which emissions from all major sources are captured in multiple samples from multiple locations to minimize dependence on single observations and single atmospheric transport conditions.A modeling framework that can capture the specific characteristics of the regional atmospheric transport, fossil fuel emissions, and nuclear contributions like the one presented here would bring added value to the interpretation of such data.

Discussion
Our modeling results show that over a significant part of our domain, the nuclear influence on the atmospheric 14 CO 2 signature will be more than 10 % (ratio = 0.1 on Fig. 5d) of the estimated fossil fuel influence, introducing considerable uncertainty to the method of using 14 CO 2 to calculate the fossil fuel CO 2 addition to the regional atmosphere.The strongest gradients of 14 CO 2 in Western Europe are found in the relatively polluted region in western Germany and the Netherlands due to the high population density and large industry sector there, and hence high CO 2 emissions.As was shown for California by Riley et al. (2008), more detailed 14 CO 2 observations in this region can possibly prove useful in lowering the uncertainty of the regional fossil fuel emission estimates.Furthermore, the high fossil-to-nuclear ratio ensures that uncertainties arising from nuclear emissions will be at their minimum.This result relies partly on the underlying emission maps for the anthropogenic (fossil fuel) CO 2 and (nuclear) 14 CO 2 emissions.We should consider various factors that are uncertain or unknown at this point for these emissions (Peylin et al., 2011;Graven and Gruber, 2011) -such as temporal characters, vertical resolution and even small irregularities in the spatial allocation of the emission sources.All our anthropogenic emissions are currently introduced in the lowest (surface) layer of our model, but according to the emission database used (IER, Stuttgart), most of the industrial emission stacks are located on average at 100 to 300 m height.Using this information in our model will likely result in the emitted CO 2 being transported away faster, and result in less local enrichment.This is also true for our nuclear emissions sources, but information on their vertical emission heights is more difficult to find.
For the fossil fuel CO 2 emissions we apply temporal profiles that disaggregate monthly, weekly and diurnal signals from the provided annual emissions.For the nuclear emissions such profiles are unknown and information on their temporal heterogeneity is not publicly available.In this study we consider these emissions as continuous and constant throughout the year.This is a relatively safe assumption for the emissions from nuclear power plants as their 14 CO 2 is a by-product of the normal operation of the reactor.Temporary shutdowns for scheduled maintenance that covers periods of weeks and sometimes months would invalidate this assumed emission pattern.Continuous constant emissions are not likely for reprocessing sites, where the emissions will depend on the type and amount of fuel being reprocessed.
Additionally, there is uncertainty if these emissions are released continuously or in a few large venting events, where the venting procedures are moreover likely to be reactor-type dependent.Currently, we lack the information to account for such complications.
When using flask samples for 14 CO 2 measurement nuclear enrichment can relatively easily be recognized.However, in integrated air and plant samples this signal will be averaged over the total sampling period.Depending on weather variability, local fossil fuel CO 2 addition and the proximity to the nuclear sources, the enrichment in 14 CO 2 can often be within the measurement precision (of approximately ±2 ‰) as we have shown.Thus, integrated samples likely have too low time resolution and sensitivity to attribute nuclear emissions, and areas where this influence is high would profit from flask sampling of 14 CO 2 in addition to integrated plant sampling.Because plant samples can be used only as complementary observations during particular seasons and depending on the species sampled a dual monitoring approach with flasks and integrated samples seems best.Based on our results, a better characterization of the temporal structure of the nuclear emissions is a prerequisite for any 14 CO 2 -based monitoring effort in Europe.
Our study is subject to known uncertainties in atmospheric transport of mesoscale models.An inaccurate simulation of wind speed and direction (Lin and Gerbig, 2005;Gerbig et al., 2008;Ahmadov et al., 2009) or boundary layer height development (Vilà-Guerau de Arellano et al., 2004;Steeneveld et al., 2008;Pino et al., 2012) will affect the transport of emission plumes and resulting mole fractions.Resolving more meso-scale circulations, and improved representation of topography can be particularly advantageous, as they can cause large gradients in CO 2 (de Wekker et al., 2005;van der Molen and Dolman, 2007).While WRF-Chem is used for a variety of atmospheric transport studies (among others: Tie et al., 2009;de Foy et al., 2011;Lee et al., 2011;Stuefer et al., 2013), more general air quality studies have shown that an ensemble of models can forecast air pollution situations more accurately than a single model (Galmarini et al., 2004(Galmarini et al., , 2013)).While in our research we focused on the transport of CO 2 and 14 CO 2 , other chemically active tracers (e.g.CO, NO x ) that are regularly measured and connected with anthropogenic emissions could be used too.Including 222 Rn as an additional tracer can help lowering the uncertainty associated with the vertical mixing in the model and provide correction factors to be applied to the other passive tracers, as shown in van der Laan et al. ( 2010), Vogel et al. (2013).
Considering future uses of 14 CO 2 observations as additional constraint on the carbon cycle, we should note that atmospheric inversions currently typically use only afternoon observational data.In that case, plant-sampled 14 CO 2 observations may provide a better representation of the afternoon atmospheric 14 CO 2 signals than conventional integrated samples that also absorb CO 2 during the night.
www.atmos-chem-phys.net/14/7273/2014/Atmos.Chem.Phys., 14, 7273-7290, 2014 However, the use of plant samples is typically limited to the summertime, which is a period with lower anthropogenic CO 2 emissions, more vertical mixing and larger biospheric fluxes.This will correspond to larger uncertainty in the recalculation of the fossil fuel CO 2 emissions compared to wintertime.
We explored the possibility that a relatively simple boxmodel can be used to calculate the emissions directly from 14 CO 2 observations, and showed its inability to capture the variability in 14 CO 2 signals across 25 European cities.Using such a simple box model has high inherent uncertainty for the reconstructed emissions, a portion of which is a direct consequence of the 14 CO 2 measurement precision.
Our results suggest that a combination of the available sampling methods should be used when planning a 14 CO 2 observational network for fossil fuel emissions estimates.Integrated air and plant samples alone can provide a longer period observations at a lower cost, but are less useful for evaluation of large nuclear influences in shorter periods.Flask samples are much better suited for this, however their continuous analysis is too costly.A possible compromise could be to obtain flask samples for a limited period alongside integrated samples for new sampling locations.This would already provide information about the possible nuclear enrichment and the wind directions from which it usually occurs.Additionally, while integrated air samples are the current standard for quasi-continuous observations of 14 CO 2 , plant samples can be obtained at a much higher spatial resolution without additional infrastructure investment.Their use is however constrained to the sunlit part of the day and generally the summer season, and the exact time and locations where the chosen crop grows.

Conclusions
In this work, we demonstrated the ability of our modeling framework to simulate the atmospheric transport of CO 2 and consequently the atmospheric 14 CO 2 signature in integrated air and plant samples in Western Europe.Based on our results we reach the following conclusions.
1. Simulated spatial gradients of 14 CO 2 are of measurable size and the 6-month average CO 2ff concentrations in the lower 1 km of the atmosphere across Western Europe are between 1 to 18 ppm.
2. Enrichment by 14 CO 2 from nuclear sources can partly mask the Suess effect close to nuclear emissions, particularly in large parts of UK and northwestern France.This is consistent with previous studies (Graven and Gruber, 2011) and we show that in these regions the strength of the nuclear influence can exceed the influence from fossil fuel emissions.
3. The simulated plant 14 CO 2 signatures show spatial gradients consistent with the simulated atmospheric gradients.Plant growth variability induces differences between the simulated plant and the daytime atmospheric mean for the period of growth, of a magnitude that is mostly within the measurement precision of ±2 ‰, but can be up to ±7 ‰ in some areas.
4. Integrated 14 CO 2 samples from areas outside the immediate enrichment area of nuclear emission sources are not sensitive to occasional advection of enriched air due to their long absorption period.However, to properly account for the nuclear enrichment term on smaller time scales, improvements in temporal profiles of nuclear emissions are needed.
5. New 14 CO 2 sampling strategies should take advantage of different sampling methods, as their combined use will provide a more comprehensive picture of the atmospheric 14 CO 2 temporal and spatial distribution.

Figure 1 .
Figure 1.The location of modeled domains.The respective horizontal resolutions are according to the color of the domain boundaries: red -36 km × 36 km; blue -12 km × 12 km; green -4 km × 4 km.The scatter markers indicate the locations of various observational sites used in this study.

Figure 2 .
Figure 2. Comparison between modeled and observed CO 2 fluxes, concentrations and boundary layer height for the location of Cabauw for one month in the simulated season.Performance is usually better on clear days as compared to cloudy ones, as indicated in the graph with the gray background.
Figure 3 shows the comparison between integrated (monthly, bi-weekly or weekly) samples and their modeled counterparts for six measurement sites -Jungfraujoch, Switzerland, Heidelberg and Schauinsland, Germany (Institut für Umweltphysik, University of Heidelberg, Germany, Levin et al., 2013), Prague-Bulovka and Kosetice, Czech Republic (Academy of Sciences of the Czech Republic, Svetlik et al., 2010) and Lutjewad, the Netherlands (Centre for Isotope Research, University of Groningen, The Netherlands, unpublished data for the monthly integrated samples, south sector data was previously used in van der Laan et al., 2010).Complementary statistics are included in Table

Figure 3 .
Figure 3.Comparison between observed and modeled atmospheric 14 CO 2 integrated samples for six observational sites.Red circles in the Jungfraujoch graph show the monthly fit used as the signature of the background CO 2 ( bg ) in our calculations.Observations are monthly continuously integrated samples at Jungfraujoch, Schauinsland, Kosetice, and Prague.At Heidelberg the weekly samples integrate only during the night-time.At Lutjewad the bi-weekly samples only integrate during periods of southerly winds, and the monthly integral over all sectors (discussed in the main text) is shown in red.

Figure 4 .
Figure 4. 6 months of hourly results for Lutjewad at 60 m height.Comparison between observed and modeled (a) CO 2 concentrations, (b) CO 2ff concentrations (c) atmospheric 14 CO 2 and (d) the contribution of different compounds for the resulting 14 CO 2 .The transport of air that is enriched in fossil fuel CO 2 is directly connected to the variations in the 14 CO 2 signal at the location, but these are not captured by current observations due to their low temporal resolution.

Figure 5 .
Figure5.Spatial distribution for the 6-month averaged (a) fossil fuel CO 2 emissions influence, (b) nuclear 14 CO 2 emissions influence, (c) resulting 14 CO 2 signature in the atmosphere and (d) the ratio between the nuclear and fossil fuel influences on the atmospheric signature, all averaged over the lower 1200 m of the atmosphere.While the largest influence over Europe is from fossil fuel CO 2 , the effect of the nuclear emissions of 14 CO 2 can be of comparable magnitude for large areas in France and UK.

Figure 6 .
Figure 6.Spatial distribution for the uncertainty in the nuclear CO 2 influence simulated for August and September, due to the uncertainty in the emission factors associated with different reactor types.(Left) The nuclear influence modeled with the central estimate of the reactor emission factors; (middle) the absolute difference between the lower estimate and central estimate; (right) the absolute difference between the higher estimate and the central estimate.Low and high estimates refer to the 70 % confidence interval for the emission factors.

Figure 7 .
Figure 7. Time series for the relative importance of nuclear vs. fossil fuel influence on the resulting atmospheric 14 CO 2 for three locations in our domain -near Cambridge (UK), Cabauw (the Netherlands) and Kosetice (Czech Republic).

Figure 8 .
Figure 8. Modeled absolute 14 CO 2 signature of maize leaves at flowering.Both the highly industrialized areas in Germany, where the atmospheric 14 CO 2 is lower than the background, and the enriched areas near the big nuclear sources in France and UK are visible also in the plants.Even on this resolution we see in the plant signature the hot spots around Paris, London, Frankfurt, and many other big cities.

Figure 9 .
Figure 9. Difference between 14 CO 2 modeled in plants and the daytime atmospheric average (left) and between modeled plants with and without taking the nuclear influence into account (right).While the left figure shows the error that should be expected if the plant growth is not taken into account and the plant signature is assumed to be equal to the atmospheric average, the right one shows the error that will be introduced if nuclear emissions of 14 CO 2 are not accounted for in the model simulation.

Atmos. Chem. Phys., 14, 7273-7290, 2014 www
Comparison between the results of the simple box model (see main text) and the modeled maize leaves 14 CO 2 signature at city center and fossil fuel CO 2 emissions averaged for 5×5 grid around the city center on 12 km horizontal resolution. .atmos-chem-phys.net/14/7273/2014/