The influence of 14 CO 2 releases from regional nuclear facilities at the Heidelberg 14 CO 2 sampling site ( 1986 – 2014 )

Abstract. Atmospheric Δ14CO2 measurements are a
well-established tool to estimate the regional fossil-fuel-derived
CO2 component. However, emissions from nuclear facilities can
significantly alter the regional Δ14CO2 level. In order to
accurately quantify the signal originating from fossil CO2 emissions, a
correction term for anthropogenic 14CO2 sources has to be
determined. In this study, the HYSPLIT atmospheric dispersion model has been
applied to calculate this correction for the long-term Δ14CO2 monitoring site in Heidelberg. Wind fields with a spatial
resolution of 2.5∘ × 2.5∘,
1∘ × 1∘, and 0.5∘ × 0.5∘ show systematic deviations, with coarser
resolved wind fields leading to higher mean values for the correction. The
finally applied mean Δ14CO2 correction for the period from
1986–2014 is 2.3 ‰ with a standard deviation of 2.1 ‰
and maximum values up to 15.2 ‰.
These results are based on the 0.5∘ × 0.5∘ wind field
simulations in years when these fields were available (2009, 2011–2014), and
for the other years they are based on 2.5∘ × 2.5∘ wind
field simulations, corrected with a factor of 0.43. After operations at the
Philippsburg boiling water reactor ceased in 2011, the monthly nuclear
correction terms decreased to less than 2 ‰, with a
mean value of 0.44 ± 0.32 ‰ from 2012 to 2014.



Introduction 10
Evaluation of the perturbation of atmospheric 14 C by nuclear bomb tests in the middle of the last century has given very useful insight into carbon cycle dynamics (e.g. Levin and Hesshaimer, 2000). Today this artificial spike has almost equilibrated with the fast exchanging carbon reservoirs, and the currently observed global  14 CO2 trend is almost exclusively due to the ongoing input of 14 C-free fossil fuel CO2 into the atmosphere (Naegler and Levin, 2009;Levin et al., 2010;Graven, 2016). This longterm trend can potentially be used to estimate the global input of fossil fuel CO2 into the atmosphere. However, the uncertainty 15 of this estimate is still large (ca. 30%, Levin et al., 2010) due to the uncertainty of the large 14 CO2 disequilibrium fluxes from biosphere and ocean, as well as artificial 14 C sources. On the continental scale, however, atmospheric  14 CO2 measurements provide a powerful and the only direct and quantitative tool for estimating the regional fossil fuel component. 14 CO2 measurements at a polluted station allow separating fossil fuel-derived regional CO2 enhancements relative to a clean reference level from those originating from biospheric fluxes if also the 14 CO2 level at the reference site is known (Levin et al., 2003;20 Turnbull et al., 2009). However, on that local to regional scale (several 10 km) 14 CO2 emissions from nuclear facilities, such as boiling water reactors, can significantly contaminate atmospheric 14 CO2. The 14 C signals from such point sources are well detectable in their immediate neighborhood in atmospheric CO2 (and CH4, e.g. Levin et al., 1992) but also in plant samples (Levin et al., 1988). 14 CO2 "plumes" from point sources normally quickly disperse at distances of some tens of kilometers. But if a sampling station is located in the catchment of such 14 CO2 point sources, special care is required to accurately quantify the 25 14 CO2 contamination and correct for it to estimate reliable fossil fuel CO2 values (e.g. Levin et al., 2003).
Here we present results from HYSPLIT dispersion modelling (Draxler and Hess, 1998) of 14 CO2 emissions from five nuclear installations in the < 60 km neighborhood of our long-term 14 CO2 monitoring site in Heidelberg. We apply the HYSPLIT model for the period of 1986-2014 with available wind fields of 2.5° x 2.5°, 1° x 1° and 0.5° x 0.5° resolution. Using reported 30 14 CO2 emission rates, these model estimates for the Heidelberg sampling site allow us to correct for the local 14 CO2 contaminations from nuclear facilities (Kuderer, 2016). Our model results, however, turned out to strongly depend on the Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-84 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 1 February 2018 c Author(s) 2018. CC BY 4.0 License. resolution of the wind field used for the calculation. We discuss this important finding and present the currently most reliable corrections of our long-term 14 CO2 measurements.

Site description
The Heidelberg 14 CO2 sampling site is located on the University campus in the outskirts of Heidelberg, a medium size city in Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-84 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 1 February 2018 c Author(s) 2018. CC BY 4.0 License.
Five nuclear installations with reported 14 CO2 emissions are found at distances between 25 km and 55 km to the Heidelberg station. Figure 1 shows their locations; details of reactor type, installed electrical output, period of operation, distance from the Heidelberg station and mean reported 14 CO2 emission during their operation up to 2014 are listed in Tab. 1. As the prevailing 5 winds in the Upper Rhine valley are from south-west, Philippsburg (KKP I & II) is the most important source of potential 14 CO2 contamination in Heidelberg. Philippsburg I is the only boiling water reactor (BWR) with its major 14 C emissions being 14 CO2, whereas pressurized water reactors (PWR) emit 14 C mainly as 14 CH4. All the other nuclear installations except for Neckarwestheim II (GKN II) emit less than 15 % of Philippsburg I. Neckarwestheim is, however, located to the southeast of Heidelberg in the Neckar valley at a distance of 55km, so that its relative contribution to the total 14 CO2 contamination is only 10 less than 10 % (see Table 2).

14 CO2 sampling and analysis
Two-and, for limited periods, also one-week integrated large volume samples of atmospheric CO2 were collected from the roof of the Institute's buildings by quantitative chemical absorption in basic sodium hydroxide (NaOH) solution, as described by Levin et al. (1980). Except for the first few years, samples were collected only during night (from 19:00 to 7:00 Central 15 European Winter Time), in order to avoid CO2 contamination from local traffic. Moving the Institute to a new building in the year 2000 required parallel CO2 sampling at both, the old and the new sampling locations on the Heidelberg University campus, in order to quantify possible differences and then allow combining the data sets from the two locations about 500 m apart. As the new building is located closer to the Heidelberg city center, slightly lower  Levin et al. (2008)). 14 CO2 samples were processed in the Heidelberg 14 C laboratory by acidification of the NaOH solution in a vacuum system. 25 The extracted CO2 was subsequently purified over charcoal. The 14 C/C ratio was then measured by low level counting (Kromer and Münnich, 1992). All results are presented here as 13 C-corrected  14 C deviations from the international reference standard (Oxalic acid) in permil. They are corrected for decay to the date of CO2 sampling (Stuiver and Polach, 1977). Note that Stuiver and Polach (1977) refer to this 14 C notation as not  14 C, however in order to be consistent with other atmospheric radiocarbon literature we stick to using  14 C instead of  Precision of  14 C values was of order 4-5 ‰ in the 1980s and 1990s, of 3-4 ‰ 30 in the 2000s and of 2-3 ‰ thereafter.

Reported 14 CO2 emissions from nuclear facilities in the surroundings of Heidelberg
According to the German Atomic Energy Act (Strahlenschutzverordnung, 2001), emissions of radioactive substances from nuclear facilities with the exhaust air must be monitored and reported quarterly to regional and federal authorities. The Bundesamt für Strahlenschutz (BfS, German Federal Office for Radiation Protection), releases yearly reports on radioactive emissions from all German reactors and research facilities; here the 14 CO2 emissions are reported separately from other 5 radioactive substances. These BfS reports are available for the years 1986 -2014. For Philippsburg I and II higher resolution, i.e. monthly emission data are available (KKP pers. comm.); these monthly data were used in this work to estimate the 14 CO2 contamination in Heidelberg.  (lower panel) shows the distribution of monthly emissions from Philippsburg I and II for the years 1986 -2012. Note the huge variability of monthly emissions, which can differ from month to month by more than a factor of two. Graven and Gruber (2011) estimated mean emission factors of 0.06 TBq 14 CO2 GWa -1 for PWRs and 0.51 TBq 14 CO2 GWa -1 for BWRs. From our 5 emission data and corresponding power production reports, we do see, however, large differences from these emission factors and for PWRs no correlation at all, as displayed in Fig. 3. Moreover, keeping in mind the huge month-to-month variability of 14 CO2 emissions from Philippsburg KKP I & II (Fig. 2, lower panel), underlines the necessity of reliable high-resolution 14 CO2 emission data from nuclear installations, if accurate corrections shall be applied to atmospheric 14 CO2 observations for fossil fuel CO2 estimates. 10

The HYSPLIT model
The Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) from NOAA offers a variety of services ranging from computing simple air parcel trajectories up to complex dispersion simulations (Draxler and Hess 1998). During the simulations, virtual particles are emitted at the source location and advected to the new particle position, described by the 5 position vector P, using the input wind velocity vector field V: ( 1 ) The advection equation is solved with a dynamic time step ∆t, demanding that the advective displacement is smaller than the size of a grid cell (Draxler, 1999 where the turbulent velocity components U', W' are estimated from the standard deviations σ of the horizontal or respective 15 vertical velocity components (Fay et al., 1995). For more details, see Stein et al. (2015) and references therein. After the advective and dispersive displacement, the HYSPLIT model computes the particle concentration in every grid cell, which gives a dilution factor f (see Eq. 3), describing how much the point source emissions are diluted over the respective grid. This dilution factor is strongly depending on the prevailing meteorological conditions.

Wind fields 20
Previous studies have shown that HYSPLIT calculations are sensitive to the meteorological input data (e.g., Cabello et al., 2008;Lin et al., 2015). Here we used three different wind velocity fields that have a horizontal resolution of 2.5° x 2.5°, 1° x 1° and 0.5° x 0.5°. The GDAS (Global Data Assimilation System) assimilates meteorological observations in numerical weather prediction models and archives the results. The one degree fields GDAS1 are available since 2005 and the half degree fields GDAS0p5 since 2008. GDAS1 and GDAS0p5 differ besides the horizontal also in the vertical resolution (Lin et al., 25 2015). The NCEP/NCAR (National Centre for Environmental Prediction/National Centre for Atmospheric Research) reanalysis provides atmospheric analyses with a spatial resolution of 2.5° x 2.5°, using historical data from 1948 onwards. All three wind fields are readily available at ftp://arlftp.arlhq.noaa.gov/pub/archives/.

Estimation of  14 Cnuclear
The 14 C signal at the sampling site  14 Cnuclear originating from 14 CO2 emissions from each nuclear facility is calculated by scaling the meteorological dilution factor f (s m -3 ) at the measurement station obtained from the HYSPLIT simulation with the time-varying emission strength Q (Bq s -1 ) of the source. This specific 14 C activity is converted (according to its definition from Stuiver and Polach (1977)) into  14 Cnuclear in ‰ according to Eq. 3 5 with the molar volume at standard atmospheric temperature and pressure (STP) Vm = 24.465 mole m -³, molar mass of carbon MC = 12 g mole -1 , mole fraction of CO2, ΧCO2, and specific activity of the 14 C standard a = 0.238 Bq gC -1 . (corresponding to the long-term average emission from this facility). The different symbols distinguish the results when using 5 the three different wind fields, i.e. with resolution of 2.5° x 2.5° (black diamonds), of 1° x 1° (blue triangles) and of the highest resolution of 0.5° x 0.5° (red circles). The two-week integrated  14 Cnuclear signals vary between 0‰ and 16 ‰ for the coarse resolution wind field, and show on average lower signals when using the higher resolved wind fields. There are, however, also situations when we obtain lower contamination signals with the coarse resolution wind field than with the higher resolved fields. The 1° x 1° wind field also yields, on average, slightly higher  14 Cnuclear signals from Philippsburg than the highest 10 resolution 0.5° x 0.5° wind field, but the differences between those two are often only marginal. Looking at the contributions from the Neckarwestheim reactors (GKN I & II) (Figure 4 lower panel), we also estimate the largest  14 Cnuclear signals with the low-resolution wind field, while the highest resolution wind field yields the smallest signals. The mean ratio between the contamination signals estimated with the highest resolution wind field and those estimated with the 2.5° x 2.5° resolution field is 0.43. We consider the results from the higher-resolution wind fields more reliable to calculate  14 Cnuclear than those with the 15 coarse resolution field (see discussion below). We further conclude that the contributions from Neckarwestheim 14 CO2 emissions on the Heidelberg  14 CO2 signal are, on average, about one order of magnitude smaller than those from Philippsburg and, thus, with an average  14 Cnuclear of less than 0.2 ‰, almost negligible.

Estimation of  14 Cnuclear in Heidelberg from all five nuclear installations
Owing to its source strength and proximity to Heidelberg, Philippsburg is the dominant contributor to the nuclear 20 contamination at our sampling site. Therefore, and considering the high month-to-month variability of emissions (Fig. 2, lower panel), it is important to use monthly resolved emission data to estimate the  14 Cnuclear signals originating from KKP I & II.
The other four nuclear installations are secondary contributors permitting the use of annual average 14 CO2 emission rates in absence of higher temporally resolved emission data. For each source location, the HYSPLIT model was run for every calendar day separately covering the period 1986 -2014. Due to the small distance between 14 C sources and the measurement station, 25 simulations were limited to 48h, where each run consisted of a 24-hour period, when particles were emitted with a constant rate, followed by 24 hours of sole propagation of the particles. Thus, for each day the simulated nuclear 14 C activity included the actual emissions of this day arriving at the sampling site and the propagated emissions from the day before. This could potentially lead to loss of particles, which arrive at the measurement site more than 24-48 hours after the release, but for an extended reference period no such effect has been observed. Typical travel times from the nuclear power plants to Heidelberg 30 are in the order of 6-12 hours.
Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-84 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 1 February 2018 c Author(s) 2018. CC BY 4.0 License.   In the years before the KKP I shutdown, about 1 % of all corrections were above 10 ‰ 5 and less than 2 % above 5 ‰. The mean correction was 2.3 ‰ with a standard deviation of 2.1 ‰. After the shutdown of the BWR KKP I, the largest 14 CO2 source before 2011,  14 Cnuclear decreased to less than 2 ‰, with a mean value of (0.44 ± 0.32) ‰ from 2012 to 2014. It is therefore feasible to apply only an average correction of this size to the Heidelberg measurements of all subsequent years.

Uncertainty of estimated  14 Cnuclear 10
The uncertainty of our  14 Cnuclear estimates originates from uncertainties in emission data and uncertainties in the HYSPLIT model transport. From comparison of results based on the differently resolved wind fields (Fig. 4), we find the largest deviations between the 2.5° x 2.5° and the 1° x 1° fields while the average differences between the two finer resolved wind fields are of order 30 %, they can, however, be as large as a factor of two for individual two-week periods. The uncertainty of the measured monthly emission data is probably less than 10-20 % and thus small if compared to the uncertainty of the model 15 transport (although sub-monthly variability in the emissions may also contribute to the uncertainty of the  14 Cnuclear estimates).
For the contributions from nuclear installations where only annual average emission data were available to us, the uncertainty of emissions is estimated to 30 %. As the contribution from all four installations except Philippsburg contribute on average only 12 % (Tab. 2) this uncertainty is small compared to the transport uncertainty of the contributions from Philippsburg. We, therefore, estimate the typical uncertainty of individual total  14 Cnuclear signals to less than 35 %. 20

Discussion and Conclusions
Our HYSPLIT estimates of 14 CO2 contaminations from nuclear facilities in the catchment area of Heidelberg showed large differences when using wind fields of different resolution. The calculated mean contamination was approximately twice as large when using the coarse resolution 2.5° x 2.5° wind field compared to the two higher resolution fields. Previous studies have shown, that meteorological coarse grid re-analyses can be well suited to capture synoptic-scale dynamical processes, but 25 biases in surface wind speeds may be introduced as re-analysis data are not well adapted to reproduce transient strong wind events occurring at the mesoscale and generating a large sub-grid scale variability (Largeron et al., 2015). These can arise in HYSPLIT trajectory calculations, which are the basis for concentration simulations, when the air mass passes through areas with complicated topography and meteorological patterns that are on a smaller scale than the data resolution (Su et al., 2015).
Another and possibly more important factor is that atmospheric dispersion is included in the model by using the standard 30 deviation of the interpolated velocity field. Linearly interpolating the coarse wind field to the internal HYSPLIT grid (here Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-84 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 1 February 2018 c Author(s) 2018. CC BY 4.0 License. 0.05° x 0.05°) leads to a less variable velocity field compared to initially starting with a fine grid. This generates more distinct plume shapes in coarse grid simulations (Kuderer, 2016). Therefore, using the coarse wind field may underestimate the effect of atmospheric dispersion, leading to high values when the plume directly passes the measurement point. We expect this to occur frequently in the case of the Philippsburg 14 CO2 plume, where the source lies in the main wind direction at rather short distance from the measurement point. In the case of Neckarwestheim, this explanation does not hold. However, also here we 5 consider the results obtained with the finest resolution wind field as more accurate. GKN lies in the hilly Neckar valley with a complex topography, which is probably better represented by the finer resolution wind fields. Overall, we expect the HYSPLIT estimates that are based on higher resolution wind fields to provide more realistic results, in particular as the topography around Heidelberg is not flat. We therefore correct the HYSPLIT results obtained with the 2.5° x 2.5° wind fields for the earlier years when high-resolution wind fields (0.5° x 0.5°) are not available (see above). 10 In an earlier study by Levin et al. (2003), KKP I & II were considered as the sole sources for the nuclear contamination at the Heidelberg sampling site. A Gaussian plume model (Turner, 1970) with a constant mean dispersion factor had been applied there to calculate  14 Cnuclear as a first approximation, but using the same monthly 14 CO2 emissions as in the present study. The mean nuclear signal estimated by Levin et al. (2003) was  14 Cnuclear = (4.8 ± 2.0) ‰ ranging from 0.2 ‰ to 10 ‰ for monthly 15 mean values. This earlier estimate of 14 CO2 contamination is approximately twice the value obtained with the HYSPLIT model and the high-resolution wind fields. Graven and Gruber (2011) used the TM3 model with a spatial resolution of 1.8° x 1.8° and estimated for 2005 a total  14 Cnuclear of 2.1 (1.1 -3.7) ‰ for the Heidelberg grid cell. Their estimate is in agreement with our results for that year ((2.1 ± 1.6) ‰) obtained with the high-resolution wind field. As in the present study, Graven and Gruber (2011) also included 14 C contributions from other nuclear installations in their estimates. However, their assumed 20 emissions from the Philippsburg I reactor were estimated with the average emission factor for BWR, which is about 20 % smaller than the measured value for 2005 used in our estimate. They also mention that their Eulerian model may have underestimated the true contamination due to its coarse resolution, which would dilute point source emissions over a large grid in an Eularian approach.

25
These comparisons with earlier studies indicate that more work and higher resolution models and wind fields are needed to reduce the uncertainty of the 14 CO2 contamination estimates from nuclear installations at measurement sites where  14 CO2 observations shall be used to precisely determine the regional fossil fuel CO2 component. Currently, we have to take into account a model transport uncertainty of about 1-2 ‰ in the estimated  14 Cnuclear contamination, if the measurement site is located closer than about 30 km downwind from a nuclear facility, which has a 14 CO2 emission rate of about 0.5 TBq yr -1 30 similar to the Philippsburg I boiling water reactor with 1 MWe power production. Other reactor types, such as the Canadian CANDU reactors may have significantly larger emission rates (Graven and Gruber, 2011;Vogel et al., 2013); the uncertainty of corresponding  14 Cnuclear estimates in their close neighborhood may then be considerably larger.
The limited accuracy and temporal resolution of 14 CO2 emission rates from nuclear installations cause additional uncertainty on the  14 Cnuclear estimates, as generally only annual mean emissions are reported. Graven and Gruber (2011) assume that 14 CO2 emissions are proportional to the annual power production. However, the present study on the influence from German reactors on the Heidelberg measurement site does not fully support this finding. Figure 4 does not show significant correlations 5 between annual 14 CO2 emissions and corresponding electricity supply. Therefore, assuming emission factors as suggested by Graven and Gruber (2011) will add considerable uncertainty to the  14 Cnuclear estimates, which may be as large as the uncertainties estimated here for model transport error.
Overall, we conclude that careful investigation of potential 14 CO2 emissions in the catchment of sampling sites is required 10 when using 14 CO2 observations for fossil fuel CO2 estimates. The differences of our modelling results, when based on differently resolved wind fields, together with the findings from earlier studies suggest that current  14 Cnuclear estimates may be wrong by a factor of two. Therefore, careful investigations with high-resolution models must be performed at all stations where 14 C-based fossil fuel CO2 measurements are conducted. We plan such studies for the European ICOS atmospheric station network (https://www.icos-ri.eu/icos-stations-network). The basis must be high-resolution 14 CO2 emissions data from nuclear 15 facilities, which need to be made available for these investigations, if contamination estimates shall be accurate.

Data availability
Data will be available from the Heidelberg University data depository under https://heidata.uni-heidelberg.de/dataverse/carbon