Observations of radiocarbon in CO2at La Jolla, California, USA 1992–2007: Analysis of the long-term trend


  • Heather D. Graven,

    1. Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California, USA
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    Search for more papers by this author
  • Thomas P. Guilderson,

    1. Center for Accelerator Mass Spectrometry, Lawrence Livermore National Laboratory, Livermore, California, USA
    2. Department of Ocean Sciences, University of California, Santa Cruz, California, USA
    Search for more papers by this author
  • Ralph F. Keeling

    1. Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California, USA
    Search for more papers by this author


[1] High precision measurements of Δ14C were performed on CO2 sampled at La Jolla, California, USA over 1992–2007. A decreasing trend in Δ14C was observed, which averaged −5.5 ‰ yr−1 yet showed significant interannual variability. Contributions to the trend in global tropospheric Δ14C by exchanges with the ocean, terrestrial biosphere and stratosphere, by natural and anthropogenic 14C production and by 14C-free fossil fuel CO2 emissions were estimated using simple models. Dilution by fossil fuel emissions made the strongest contribution to the Δ14C trend while oceanic 14C uptake showed the most significant change between 1992 and 2007, weakening by 70%. Relatively steady positive influences from the stratosphere, terrestrial biosphere and 14C production moderated the decreasing trend. The most prominent excursion from the average trend occurred when Δ14C decreased rapidly in 2000. The rapid decline in Δ14C was concurrent with a rapid decline in atmospheric O2, suggesting a possible cause may be the anomalous ventilation of deep 14C-poor water in the North Pacific Ocean. We additionally find the presence of a 28-month period of oscillation in the Δ14C record at La Jolla.

1. Introduction

[2] Long term atmospheric measurements of radiocarbon, 14C, in CO2 began in 1954 at Wellington, New Zealand [Rafter, 1955]. Since then, observations from New Zealand [Rafter and Fergusson, 1957; Manning et al.,1990; Currie et al., 2009], Norway [Nydal and Lövseth, 1965, 1983], central Europe [Levin et al., 1985; Levin and Kromer, 2004] and other sites have recorded large changes in the 14C/C ratio of CO2. In the 1950s and 1960s, testing of nuclear weapons added an excess of 14C atoms that approximately doubled the atmospheric inventory of 14C. As the testing ceased and bomb-derived14C entered the oceanic and terrestrial carbon reservoirs, observations of 14C/C in CO2revealed a quasi-exponential decline that enabled investigation of mixing rates between different parts of the atmosphere and exchange rates between the atmosphere and the ocean and terrestrial ecosystems [e.g.,Rafter and Fergusson, 1957; Lal and Rama, 1966; Goudriaan, 1992; Hesshaimer et al., 1994; Trumbore, 2000; Naegler et al., 2006].

[3] Presently, 14C exchanges between the atmosphere and the ocean and terrestrial biosphere are redistributing the bomb-derived excess14C from short term to longer term reservoirs. 14C exchanges are also responding to the dilution of atmospheric 14C by fossil fuel-derived CO2 which has no 14C because of radioactive decay [Suess, 1955; Keeling, 1979; Tans et al., 1979; Stuiver and Quay, 1981]. Since the response of land and ocean carbon reservoirs to these perturbations in 14C/C is governed by the same exchange processes that determine anthropogenic CO2 uptake and storage, continued observation and understanding of 14C dynamics can provide constraints on terrestrial and oceanic carbon cycling and the potential magnitude and sustainability of CO2 sinks [Levin and Hesshaimer, 2000; Randerson et al., 2002].

[4] 14C/C ratios can also be used to identify local additions of CO2 from fossil fuel emissions by observation of 14C dilution in comparison to background air [e.g., Tans et al., 1979; Levin et al., 1989; Meijer et al., 1996; Turnbull et al., 2006; Levin and Rödenbeck, 2008]. Quantification of fossil fuel-derived CO2 can be useful for resolving budgets of CO2 contributions from industrial versus biospheric or oceanic sources [e.g., Turnbull et al., 2006; Graven et al., 2009], for detecting temporal or spatial patterns in fluxes of different types [e.g., Hsueh et al., 2007; Turnbull et al., 2009a], or for estimating fossil fuel emissions within a catchment area [e.g., Levin et al., 2003; van der Laan et al., 2010; Turnbull et al., 2011]. The use of atmospheric observations to estimate fossil fuel emissions promises to become important as emissions of CO2and other greenhouse gases are more heavily regulated and require independent verification of economic data-based inventories [Nisbet, 2005; Marquis and Tans, 2008]. Measurements that resolve variability in background air are essential to this technique since uncertainty in background air composition limits the precision of observation-based estimates of fossil fuel-derived CO2 [Graven et al., 2009; Turnbull et al., 2009b].

[5] Measurements of 14C/C ratios are typically referenced to the Modern Standard and reported as Δ14C, which includes a correction for radioactive decay between sampling and analysis and a correction for mass-dependent fractionation using measurements of13C/12C in the sample [Stuiver and Polach, 1977]. Use of Δ14C notation eliminates the effect of fractionating processes such as photosynthetic assimilation or oceanic CO2 uptake on the 14C/C ratio in CO2. Δ14C in CO2 is therefore sensitive to natural and anthropogenic 14C production as well as to exchanges of carbon with reservoirs that have a different Δ14C signature.

[6] Here we present Δ14C measurements in CO2 samples from La Jolla, California, USA collected at roughly monthly intervals between 1992 and 2007. In this paper, we focus on analyzing the Δ14C trend over the 16-year record. We compare the observed trend to simple models of the contributions to the global tropospheric trend in Δ14C between 1992 and 2007 from the release of fossil fuel CO2, natural and anthropogenic production of 14C and carbon exchanges with the stratosphere, ocean, and biosphere. We then evaluate interannual variability in the trend of Δ14C. In the accompanying paper, we present Δ14C observations from 6 other global sites for 2- to 9-year periods ending in 2007 and examine spatial gradients and seasonal cycles [Graven et al., 2012].

2. Methods

[7] Atmospheric flask sampling at La Jolla, California is conducted at the Scripps Pier (32.87°N, 117.25°W) when meteorological conditions are favorable for collecting clean, marine air and avoiding local contamination. Such conditions are met when strong, stable winds originating from the southwesterly sector (offshore) are present and a low, stable CO2 concentration is identified with a continuous CO2 analyzer.

[8] The representativeness of air collected at La Jolla under clean air conditions can be assessed by comparing observed CO2 concentrations with other background sites. Observed annual mean CO2 gradients are less than ±0.2 ppm between La Jolla and the Pacific Ocean Station at 30°N, 122.85°W [Conway and Tans, 2004] and less than ±0.5 ppm between La Jolla and other Scripps CO2 stations in closest proximity (Kumukahi, Hawaii and Point Barrow, Alaska) [Keeling and Piper, 2001; Keeling et al., 2011]. The CO2 concentration in the air collected at La Jolla is slightly higher than Kumukahi and slightly lower than Point Barrow, in accordance with the observed meridional CO2 gradient in background air [Keeling and Piper, 2001; Masarie and Tans, 1995; Keeling et al., 2011]. The seasonal cycle of Δ14C at La Jolla, presented in the accompanying paper, also supports the representativeness of clean, marine air sampled under these conditions. If local contamination strongly contributed to the seasonal cycle of Δ14C at La Jolla, the minimum in Δ14C would be expected to occur in the fall months when polluted continental air is transported offshore most frequently [Conil and Hall, 2006; Riley et al., 2008]. In fact, the maximum Δ14C occurs in the fall months at La Jolla, consistent with other Northern Hemisphere observations of background air [Graven et al., 2012; Levin and Kromer, 2004; Turnbull et al., 2007].

[9] Evacuated 5 L spherical glass flasks are sampled by opening a single ground taper joint stopcock sealed with Apiezon® grease and filling with whole air. At La Jolla, six flasks are sampled concurrently, while at other Scripps CO2 stations, 2–3 flasks are sampled concurrently. Flask air is dried and the CO2concentration is measured by non-dispersive infrared gas analysis [Keeling et al., 2002] at the Scripps laboratory. CO2 is extracted from the remaining flask air by passing 2–4 L of air through a spiral quartz trap immersed in liquid nitrogen, then the CO2 sample is sealed and stored in a Pyrex® tube. The same flask handling, storage and extraction procedures are used for analysis of δ13C in CO2 at Scripps with measurement precision of <0.03 ‰ [Guenther et al., 2001]. An archive of such CO2 samples dating back to July 1992 from La Jolla was available for analysis. Samples collected between July 1992 and December 2007 were analyzed for 14C in CO2, including 79 sample dates for which two or more replicate samples were measured. These CO2 samples were analyzed together with CO2 samples from six other sites [Graven et al., 2012].

[10] All CO2 samples were converted to graphite and analyzed by accelerator mass spectrometry (AMS) at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Laboratory (LLNL) between 2003 and 2009 [Graven et al., 2007; Graven, 2008]. We utilize the Δ14C notation implicitly as a geochemical sample with known age and δ13C correction (equivalent to Δ in work by Stuiver and Polach [1977]). Ratios of 14C/C are corrected for decay between sampling and analysis dates and for mass dependent fractionation using δ13C. The δ13C correction uses δ13C measured in concurrently sampled CO2 [Guenther et al., 2001] to normalize atmospheric samples to the −25 ‰ reference. Unlike some smaller AMS systems which cause significant fractionation during ionization, there is no evidence for fractionation within the LLNL AMS ion source [Proctor et al., 1990]. This is shown by nearly constant 14C/13C ratios measured throughout the analysis of individual samples [Fallon et al., 2007]. Slight drifts in 14C/13C ratios can be attributed to stripping efficiency and are successfully canceled by normalization with reference materials. Therefore, in-lineδ13C correction is not required or performed at LLNL.

[11] Individual measurement uncertainty in Δ14C is ±1.7 ‰ for most samples, where uncertainty is determined by the reproducibility of Δ14C in CO2 extracted from whole air reference cylinders [Graven et al., 2007; Graven, 2008]. Measurements conducted prior to 2006 have uncertainties between ±1.7 and ±3.3 ‰ [Graven, 2008], since sample handling and data processing were not yet optimized for CO2 samples. However, drawing on a large archive of existing samples allowed the samples to be selected randomly for analysis between 2003 and 2009. Samples from different years and different sites were included in each measurement batch [Graven, 2008], so that differences in uncertainty between measurement batches can be expected to have little impact on the features contained in the atmospheric records.

3. Δ14C Observations

[12] Measurements of Δ14C are shown in Figure 1 and listed in Appendix A. These data are also available at the Scripps CO2 Program Web site: http://scrippsco2.ucsd.edu/.

Figure 1.

Δ14C measured in CO2 sampled at La Jolla with a cubic smoothing spline. Replicate measurements have been averaged. Error bars show measurement uncertainty or the standard deviation in Δ14C of replicate samples, whichever is larger.

[13] Figure 2a shows seasonally adjusted observations of Δ14C at La Jolla to emphasize variations in the data that occur at timescales longer than one year. Seasonal cycles are presented and discussed in the accompanying paper [Graven et al., 2012]. The seasonal cycles were removed by first detrending the data with a cubic smoothing spline with cutoff period of 24 months [Enting, 1987]. Second, a loose cubic smoothing spline was fit to the detrended data (cutoff period of 4 months) and finally, the loose spline was subtracted from the original observations.

Figure 2.

(a) Seasonally adjusted Δ14C at La Jolla with a cubic smoothing spline. (b) The derivative of the spline curve from Figure 2a is shown as the solid line. Also shown in Figure 2b are lines representing the derivative of a linear fit (dashed, −5.5 ‰ yr−1) and an exponential fit (dash-dotted).

[14] Δ14C at La Jolla decreased by nearly 100 ‰ between 1992 and 2007. A linear least squares fit to all measurements results in a slope of −5.5 ± 0.1 ‰ yr−1, where 0.1 is the 1-σ uncertainty [Cantrell, 2008]. The trend was slightly weaker during the second half of the observation period; a linear fit to observations between mid-2001 and the end of 2007 yields a slope of −5.0 ± 0.2 ‰ yr−1 [Graven et al., 2012] while a fit to observations between mid-1992 to mid-2001 yields −5.7 ± 0.1 ‰ yr−1. We have also fit an exponential trend to the observed Δ14C; the derivative of the exponential and linear fits are shown in Figure 2b. Compared to the constant trend of −5.5 ‰ yr−1 computed by the linear fit, the exponential derivative slows from −8 ‰ yr−1 in 1992 to −3 ‰ yr−1 in 2007.

[15] Figure 2b also shows the derivative of the seasonally adjusted Δ14C observations. The growth rate of Δ14C at La Jolla showed large variability over 1992–2007. The strongest excursion from the linear or exponential trend was an especially rapid decrease in Δ14C in 2000. Local minima in the trend of Δ14C are also observed at roughly 2 year intervals, suggesting a short term periodicity exists in Δ14C at La Jolla. Features of variability in the trend are not likely to be artifacts of the spline fitting technique, since trends calculated using annual means (Section 5.2) and low-pass filtering (Section 5.3) show similar features.

4. Global Trend in Tropospheric Δ14C

4.1. Description of Box Model Formulation

[16] In this section, we will describe the box model formulation used to simulate influences on the trend in Δ14C from the emission of fossil fuel-derived14C-free CO2, natural and anthropogenic 14C production, and 14C and carbon exchanges between the troposphere and the ocean, the land biosphere, and the stratosphere. In Section 4.2, we will present the results of the model and in Section 5.1, we will discuss the results, the limitations of the simple models used here, and a comparison with similar calculations performed by Levin et al. [2010].

[17] The influence on tropospheric Δ14C from exchange with a carbon reservoir is primarily determined by the Δ14C disequilibrium. In order to formulate the most precise estimate of these influences, we used the observed tropospheric Δ14C, δ13C and CO2 to calculate the evolution of Δ14C and carbon and isotopic fluxes in separate, uncoupled forward models of the ocean, biosphere and stratosphere. We then synthesize the separate model contributions and evaluate the correspondence between the simulated and observed Δ14C trend. While individual components are forced by observed atmospheric changes, they are not constrained to add up to the observed overall Δ14C change in the atmosphere. A comparison of the sum of the components with observations thus provides an important consistency check. Estimates of global averages were formulated by weighted averages of clean-air station data including results fromNeftel et al. [1994] and Keeling and Whorf [2005] for atmospheric CO2 concentrations, from Friedli et al. [1986] and Keeling et al. [2005] for δ13C, and from this work and that of Stuiver et al. [1998], Levin and Kromer [2004], Levin et al. [2007] and Graven et al. [2012] for Δ14C.

[18] The box model setup is depicted in Figure 3 and summarized in Table 1. The biospheric and oceanic components were initialized with simulations of constant preindustrial atmospheric composition lasting 30,000 years to achieve steady state. Then, simulations of the biospheric and oceanic components were conducted using records of atmospheric composition beginning in 1511 for Δ14C [Stuiver et al., 1998] and 1720 and 1744 for CO2 concentrations and δ13C [Neftel et al., 1994; Friedli et al., 1986], respectively. The stratospheric component was simulated beginning in 1900. We present model results for the period of observation at La Jolla: 1992 through the end of 2007.

Figure 3.

Schematic of the box model setup and parameter values used to estimate contributions to the global tropospheric Δ14C trend. Colored lines indicate carbon and isotopic fluxes that correspond to the individual contributions plotted in Figure 4; dashed lines indicate fluxes to stratospheric reservoirs that affect the troposphere indirectly.

[19] The decrease in Δ14C caused by fossil fuel emissions was estimated by mixing the global annual CO2 emissions from inventories of economic data [Marland et al., 2008; Canadell et al., 2007] into the entire troposphere (78% of the atmosphere). We included a 10% uncertainty in reported emissions, slightly larger than estimates of 8% by Andres et al. [1996] and 5% by Canadell et al. [2007].

[20] Air-sea exchange was estimated by several simulations of a 43-box diffusion model including CO2 and 14C and 13C isotopes [Oeschger et al., 1975], using a piston velocity of 14.8 to 18.1 cm hr−1 and an eddy diffusion coefficient of 3000 to 6000 m2 yr−1. The specification of piston velocity uses results from previous studies that constrained the globally averaged piston velocity with observations of the oceanic inventory of bomb-derived14C [Sweeney et al., 2007; Naegler et al., 2006] and oceanic Δ14C and δ13C distributions [Krakauer et al., 2006]. We selected three values across the range in piston velocity (14.8, 16.3 and 18.1 cm hr−1). Then we selected eddy diffusion coefficients that allowed the simulated average oceanic depth profile of natural 14C [Oeschger et al., 1975] and the simulated oceanic inventory of anthropogenic CO2 [Sabine et al., 2004] to roughly match observations, in addition to allowing the simulated total bomb-derived14C in our modeled carbon system (including the atmosphere and terrestrial biosphere, see below) to be consistent with Naegler and Levin [2006]. We used four values for the eddy diffusion coefficient (3000, 4000, 5000 and 6000 m2 yr−1), which are specific to our model setup. One combination of piston velocity and eddy diffusion coefficient (18.1 cm hr−1 and 6000 m2 yr−1) did not fit our constraints, so this pair of parameter values was excluded.

[21] Terrestrial ecosystem exchange was estimated by several simulations of a one-box model of the biosphere. The biosphere was assigned a total preindustrial amount of overturning biomass of 470 to 1370 Pg C and a CO2 fertilization factor [Keeling et al., 1989] of 0 to 0.4, with global net primary production (NPP) of 24 to 42 Pg C yr−1and average ecosystem residence time of 16 to 35 years between 1992–2007. Sets of parameter values were chosen to match a total bomb-derived14C excess of 615 ⋅ 1026 atoms [Naegler and Levin, 2006], including the atmospheric inventory and the ocean inventory from simulations of our one-dimensional model (see above). We allowed a range in total bomb-derived14C of ±35 ⋅ 1026 atoms, twice as large as the uncertainty reported by Naegler and Levin [2006]. We neglect the fraction of NPP for which the assimilated carbon is returned to the atmosphere within 1–3 years, one-third or more of NPP [Randerson et al., 2006], since the respiration of such young carbon does not substantially affect tropospheric Δ14C over 1992–2007 (less than 0.4 ‰ yr−1). Allowing for the fraction of NPP and biomass that are neglected by this assumption, the total biospheric mass and NPP in our simple model (470–1370 Pg C and 24–42 Pg C yr−1) are similar to current dynamic global vegetation models [Cramer et al., 2001; Friedlingstein et al., 2006]. Additional annual carbon, 13C and 14C fluxes to or from the biosphere were assigned based on the land use (LU) flux of Houghton [2008] and the residual fluxes from a single deconvolution between the fossil fuel and land use sources, the observed atmospheric CO2 growth and the box diffusion model representation of oceanic CO2 uptake [Siegenthaler and Oeschger, 1987].

[22] Cosmogenic production was simulated to occur at an average rate of 2.16 ⋅ 1026 atoms yr−1, which is 35–40% lower than the rate estimated by Lal [1992] and Masarik and Beer [2009]. The total production rate was reduced from Lal [1992] and Masarik and Beer [2009]in order to match the pre-bomb global14C inventory simulated by our oceanic and biospheric models. Observations of 14CO [Manning et al., 2005] support a higher cosmogenic production rate, similar to that predicted by Lal [1992] and Masarik and Beer [2009]. However, as in work by Levin et al. [2010], the use of a smaller average production rate was necessary to achieve a steady state in our modeled carbon system. Modulation of 14C production by the sunspot cycle was included according to observations of the cosmic neutron flux at Climax, Colorado, USA (available at http://ulysses.sr.unh.edu/NeutronMonitor/neutron_mon.html), using relationships between the cosmic neutron flux and the solar modulation parameter from Lal [1992], Masarik and Beer [1999] and Lowe and Allan [2002]. Simulated production varied by ±15% [Lal, 1992; Masarik and Beer, 1999] over the sunspot cycle and an uncertainty of ±10% in the total production rate was included. One-half to two-thirds of cosmogenic14C production was prescribed to occur in the stratosphere, with the rest occurring in the troposphere [O'Brien, 1979; Jöckel et al., 1999].

[23] Production of 14C by nuclear power plants was calculated using energy statistics and emission factors of 14C release per unit electrical power generation for 6 different reactor types [United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR), 2000; Graven and Gruber, 2011]. Electrical output from each type of reactor was gathered from the International Atomic Energy Agency's Power Reactor Information System (available at http://www.iaea.org/programmes/a2/). We also accounted for 14C released by 4 spent nuclear fuel reprocessing facilities where 14C emission data was available [UNSCEAR, 2000; Schneider and Marignac, 2008; Nakada et al., 2008; UK Environmental Agency, 2008; Anzai et al., 2008], which amounted to 7–16% of the release from nuclear power plants. Total 14C production from the nuclear energy industry was 0.28 ⋅ 1026 atoms in 1992, which was 13% of the rate of cosmogenic production. Anthropogenic 14C production increased to 0.36 ⋅ 1026 atoms in 2000 then remained largely constant until 2007. Our calculation includes 14C released from nuclear power plants as methane since these releases will eventually oxidize to form 14CO2. This assumes that the nuclear-derived atmospheric14CH4 inventory did not change substantially, which is reasonable considering the increase in production was modest over 1992–2000 (30%) and steady thereafter. We assume an uncertainty of ±50% in the total production of 14C by the nuclear energy industry.

[24] Finally, the trend in tropospheric Δ14C caused by stratosphere-troposphere transport was estimated by a two-box model of the stratosphere with a residence time in the lower box of 1 to 1.5 years and a residence time in the upper box of 5 years, with 28% of the stratospheric mass in the upper box, similar toRanderson et al. [2002]. Cosmogenic production of 14C in the stratosphere was simulated to occur either mainly in the lower stratosphere (75%) or to be distributed equally between the stratospheric boxes.

4.2. Modeled Influences on Tropospheric Δ14C Trend

[25] Fossil fuel CO2 emissions increased from 6.1 to 8.5 Pg C yr−1 between 1992 and 2007, ranging between −0.6 to 5.3% growth each year [Marland et al., 2008; Canadell et al., 2007]. The resulting trend in tropospheric Δ14C was −11.4 ± 1.1 ‰ yr−1 in 1992 and decreased to −13.2 ± 1.3 ‰ yr−1 in 2007 (Figure 4a and Table 1). The Δ14C trend caused by fossil fuel emissions changed by a smaller fraction than the emissions themselves because rising CO2 concentrations and declining Δ14C in CO2 are reducing the sensitivity of atmospheric Δ14C to fossil fuel-derived CO2 [Graven et al., 2012; Levin et al., 2010].

Figure 4.

(a) Contributions to the global Δ14C trend by stratospheric exchange (red), biospheric exchange (green), 14C production in the troposphere (yellow), oceanic exchange (blue) and fossil fuel combustion (black). The filled areas reflect the uncertainty or range of plausible values included in the models for each process and the lines show the middle of the range. (b) The sum of the modeled components is shown as the gray filled area, where the area encompasses the sum of the trends in Figure 4a plus and minus a quadrature sum of the range/uncertainty for each independent process. The derivative of the seasonally adjusted observations at La Jolla is shown by the black line in Figure 4b, repeated from Figure 2b.

[26] Air-sea fluxes contributed to a decreasing trend in Δ14C and showed the largest change over 1992–2007, from −8.7 ± 2.0 ‰ yr−1 in 1992 to −2.4 ± 1.6 ‰ yr−1 at the end of 2007 (Figure 4a and Table 1). The disequilibrium between mixed layer and tropospheric Δ14C as predicted by the box diffusion model shrank from −69 ± 22 ‰ in 1992 to −21 ± 15 ‰ at the end of 2007. In comparison, the simulated disequilibrium of the preindustrial state was −58 ± 7 ‰, which matches observations through our tuning of the model (Section 4.1). The air-sea flux may be considered to consist of a steady state component that roughly balances cosmogenic production and an anthropogenic component that responds to Δ14C perturbations from nuclear sources of 14C and from 14C dilution by fossil fuel emissions. The air-sea disequilibrium is now weaker than the preindustrial air-sea disequilibrium in our box model simulations, suggesting that the total oceanic uptake of14C may have become smaller than cosmogenic production in recent years. This indicates that the anthropogenic 14C flux reversed sign. While net removal of 14C from the atmosphere continued through 2007, dilution by fossil fuel CO2may have become a stronger influence on the anthropogenic air-sea14C flux than bomb-derived excess14C.

Table 1. Simulated Contributions to the Tropospheric Δ14C Trend (‰ yr−1) From This Work and From Levin et al. [2010] in 1992 and the End of 2007a
ComponentThis WorkLevin et al. [2010]
  • a

    Uncertainties are omitted here for clarity but described in Sections 4.2 and 5.1 and shown in Figure 4 for this work.

Fossil fuel−11.4−13.2Marland et al. [2008]−11.7−13.5Marland et al. [2007]
Ocean−8.7−2.4Box diffusion model−8.7−1.8Extrapolation of ocean data
Biosphere+3.7+4.61-box model+4.0+3.52-box model
Stratosphere (Transport and Cosmogenic Prod.)+5.9+4.92-box model and scaled neutron flux data+6.0+4.816-box model and sinusoidal approx.
Troposphere (Nuclear and Cosmogenic Prod.)+3.1+3.3Scaled nuclear power prod. and scaled neutron flux data+2.8+3.5Extrapolation of UNSCEAR [2000] and sinusoidal approx.
Total−7.4−2.9 −7.5−3.5 

[27] The biospheric contribution to the trend in tropospheric Δ14C was simulated to be relatively constant, 3.7 ± 1.3 ‰ yr−1 in 1992 and 4.6 ± 1.2 ‰ yr−1 at the end of 2007 (Figure 4a and Table 1). In contrast to the ocean, the disequilibrium between the terrestrial biosphere and the troposphere remained relatively constant from 1992 through 2007 at +85 ± 26 to +99 ± 8 ‰. This is because mean Δ14C in respired carbon has stopped rising and is now decreasing at a similar rate as tropospheric CO2 (Section 5.1 and Naegler and Levin [2009]).

[28] In order to match the bomb-14C inventory of 615 ± 35 ⋅ 1026 atoms, simulations of the biosphere model with relatively high NPP were balanced by high biomass, or vice versa, so that fixing one parameter reduced the range of acceptable values for the other parameters. Therefore, if the uncertainty in one of these parameters was improved, the uncertainty in the other parameters could be tightened using the 14C inventory as a constraint.

[29] Using the global inventory of bomb-derived14C as a constraint also coupled the biospheric parameters to the modeled oceanic inventory. Simulations with slower gas exchange and diffusion reduced the oceanic inventory, requiring longer biospheric residence times to increase the biospheric inventory. Thus, reduced uncertainty in the oceanic bomb 14C inventory or better constraints on the rates of global average air-sea gas exchange and on vertical mixing in the oceanic interior would also tighten the range of acceptable values in the biospheric parameters.

[30] Tropospheric production of 14C by cosmogenic and anthropogenic sources contributed an average of 3.3 ± 1.3 ‰ yr−1 to the trend of Δ14C in the global troposphere (Figure 4a). Cosmogenic production in the troposphere, modeled as 33–50% of total cosmogenic production, comprised an average of 2.5 ± 0.8 ‰ yr−1. Solar variability associated with the sunspot cycle enhanced production in 1993–99 and 2006–08 and reduced production in 1992 and 2000–05 but resulted in only ±0.2 ‰ yr−1 variation in the Δ14C trend. Production of 14C by nuclear power plants and nuclear fuel reprocessing contributed an average of 0.9 ± 0.5 ‰ yr−1. Decay of 14C in the troposphere is also included in the tropospheric production component of Figure 4a, though decay was only −0.1 ‰ yr−1 over the 1992–2007 period.

[31] According to the 2-box model of the stratosphere, the transport of14C-enriched stratospheric air was a positive influence of 5.9 ± 1.0 ‰ yr−1 in 1992 that decreased to 4.9 ± 1.0 ‰ yr−1 at the end of 2007 (Figure 4a and Table 1). The decrease in stratospheric influence was consistent with a slowing in the rate of decrease of tropospheric Δ14C between 1992 and 1997. After this time Δ14C in the troposphere decreased at a relatively steady rate and the stratospheric contribution to the trend remained relatively constant. After 1997, approximately 30% of the positive influence from the stratosphere was due to the lag time for mixing of tropospheric and stratospheric air and 70% was due to cosmogenic production. In the future, the cosmogenic influence on Δ14C in the stratosphere and the troposphere is expected to decline as greater CO2 concentrations will cause stronger dilution of cosmogenic 14C.

[32] The sum of all contributions to the global Δ14C trend is shown by the filled area in Figure 4b. The range of values was calculated by adding together the middle of the range of each contribution (shown as lines in Figure 4a) and computing a quadrature sum using one-half of the range of values for each independent process. Simulated ranges in the trend from fossil fuel combustion,14C production from the nuclear energy industry and the turnover rate of the stratospheric box model each made independent contributions to uncertainty. The range of values for the sum of the biospheric and oceanic components of the trend made a single contribution to the uncertainty in the overall trend, since the biospheric and oceanic parameters of the box models were determined in concert. Similarly, as different estimates shifted cosmogenic 14C production between the stratosphere and troposphere, the range in the sum of cosmogenic production in the stratosphere and troposphere was also considered to be a single contribution to uncertainty.

[33] The global Δ14C trend predicted by the sum of components weakened slightly between 1992 and 1996 then remained fairly steady between 1997 and 2007. The predicted global trend was −7.4 ± 2.3 ‰ yr−1 in 1992 and −2.9 ± 2.4 ‰ yr−1 at the end of 2007, averaging −4.2 ± 2.2 ‰ yr−1 over the entire period.

[34] The observed, smoothed trend at La Jolla is also shown in Figure 4b, repeated from Figure 2b. The observed trend overlaps the modeled trend except when Δ14C decreased more rapidly than average, particularly in 1992–93, 2000, and 2004–05. The average trend in Δ14C observed at La Jolla, −5.5 ± 0.1 ‰ yr−1, lies within the modeled range of values though it is near to the lower end. Consistent with the model, observed trends in Δ14C show a slower rate of decrease in the recent decade (Section 3 and Levin et al. [2010] and Graven et al. [2012]).

5. Discussion

5.1. Tropospheric Δ14C Trend

[35] The correspondence between the modeled global trend and the observed long-term trend at La Jolla suggests that the simple formulations we have used to represent the exchanges of carbon and14C provide a reasonable description of recent Δ14C dynamics. As the observed trend is near the lower end of the modeled range, the majority of our simulations are likely to have overestimated the positive trend contributions of the biosphere, stratosphere and/or production in the troposphere and underestimated the negative trend contributions of the ocean and/or fossil fuels.

[36] Simulated trends may be biased by the simplicity of the models, particularly for the ocean and biosphere. The box diffusion model used here does not simulate intermediate or deep water ventilation in high latitudes by allowing direct exchange between the atmosphere and sub-surface oceanic boxes [Oeschger et al., 1975], unlike some other box model formulations [e.g., Siegenthaler, 1983]. While the model parameters were selected to correspond to oceanic bomb 14C and anthropogenic carbon inventories, neglecting 3-D transports could overestimate surface Δ14C in recent years by excluding high latitude exchange with dense, 14C-depleted water. Additionally, the biospheric enrichment predicted by our one-box model is larger than that estimated byNaegler and Levin [2009]using a two-box model. This suggests the release of14C by terrestrial ecosystems may be overestimated, although the discrepancy with Naegler and Levin [2009]is reduced by the fact that our one-box model neglects the fraction of NPP (roughly 1/3) which involves rapid turnover of assimilated carbon.

[37] A similar study of the recent trends in Δ14C was conducted by Levin et al. [2010] using a different box model setup, summarized in Table 1. In comparison to Levin et al. [2010], our treatment is slightly more complex for air-sea exchange and14C production and simpler for stratospheric and biospheric exchanges. Levin et al. [2010] extrapolated surface Δ14C from oceanic survey data to estimate air-sea14C fluxes. Their estimated fluxes were similar to the box diffusion model in 1992 but changed more rapidly over 1992–2007 so that their oceanic contribution to the Δ14C trend was weaker than our box diffusion model in 2007, though both estimates lie within the other estimate's uncertainty. Levin et al. [2010] also extrapolated 14C production by the nuclear energy industry after 1997 and assumed a sinusoidal solar cycle variation in cosmogenic production. As nuclear production increased by only 7% from 1997 to 2007, Levin et al.'s [2010] extrapolation is likely to have overestimated production from nuclear power plants in recent years. Levin et al.'s [2010] smooth sinusoidal approximation likely underestimated the sharp fluctuations in cosmogenic production, so that production was too strong during solar maxima when cosmogenic production is reduced, particularly the strong solar maximum in 1991–92, and too weak during solar minima when cosmogenic production is enhanced, though the discrepancy is smaller than the total uncertainty. The biospheric 14C flux in Levin et al. [2010] was the same as Naegler and Levin's [2009]two-box model, as previously mentioned, and simulated a positive contribution that was within 1 ‰ yr−1of our one-box model. The biospheric contribution decreased slightly inNaegler and Levin's [2009]model over 1992–2007 while the biospheric contribution increased slightly, on average, in our one-box model (Table 1). The different tendencies reflect a small growth in the troposphere-biosphere disequilibrium in our one-box model versus a small reduction in the troposphere-biosphere disequilibrium in the two-box model between 1992 and 2007, suggesting the discrepancy between the one-box and two-box models is likely to be larger outside of the 1992–2007 interval considered here. Positive contributions of 5–6 ‰ yr−1were simulated over 1992–2007 by both the 2-box stratosphere model we used andLevin et al.'s [2010]16-box stratosphere. Both models used fossil fuel emission data fromMarland et al. [2008] leading to similar contributions to the Δ14C trend. Differences in the fossil fuel component are likely due to small differences in the estimated global mean atmospheric composition or the size of the troposphere.

[38] The simulated components of the Δ14C trend and the total Δ14C trend in this work and in the work by Levin et al. [2010] are nearly the same (Table 1), despite the differences in model formulation. The average global trend simulated over 1992–2007 was −4 ‰ yr−1 in our model and −5 ‰ yr−1 in that of Levin et al. [2010], consistent within the uncertainty of ±2–3 ‰ yr−1. We note however that both model setups have used Naegler and Levin's [2006]estimate of total bomb-derived excess14C as a constraint on biospheric and oceanic Δ14C, and that the simulated trends have rather large uncertainties (Section 4.2 and Levin et al. [2010]). The average global trend over 1992–2007 is smaller in our model, despite matching that of Levin et al. [2010] in 1992 (Table 1), since Levin et al.'s [2010]simulated trend weakens gradually over the whole period while our simulated trend weakens more rapidly in the first few years then remains fairly steady. The largest component of uncertainty to the trend is in air-sea exchange, suggesting that additional constraints on the air-sea flux of14C would particularly improve the uncertainty range of the full modeled trend of Δ14C.

5.2. Rapid Decline of Δ14C in 2000

[39] The most outstanding feature in the seasonally adjusted Δ14C record at La Jolla is the rapid decrease in Δ14C in 2000 (Figures 2b and 4b). The rate of decrease appeared to be nearly twice as rapid as the average. The feature was also observed at Jungfraujoch, Switzerland but not at Cape Grim, Australia [Levin et al., 2010], suggesting it extended through the Northern Hemisphere midlatitudes but not into the Southern Hemisphere. Restriction to the Northern Hemisphere suggests that anomalous regional 14C or carbon fluxes or changes in atmospheric circulation in northern regions may be responsible for the enhanced decline in Δ14C.

[40] One potential cause of the anomalous decrease in Northern midlatitude Δ14C in 2000 may have been a strengthening of the regional air-sea14C flux in the Northern Pacific Ocean. In 2000–01, winter sea surface temperature in the Pacific north of 40° was anomalously cold and high wind speeds and exceptionally high gas exchange velocities were observed in the Western North Pacific [Kawabata et al., 2003]. These high wind speeds may have enhanced the ventilation of deep waters in the North Pacific, exposing aged and 14C-poor water masses [Key et al., 2004] and resulting in rapid, anomalous net exchange with lower-Δ14C CO2entering the atmosphere and higher-Δ14C CO2 entering the ocean. Direct observations are not available to verify that changes in Δ14C occurred in the surface waters of the North Pacific during this time period. However, this process can be investigated with observations of atmospheric O2, since anomalous ventilation would also enhance oceanic uptake of O2, which is depleted in aged water masses due to consumption by organic matter remineralization.

[41] Hamme and Keeling [2008] observed a strong decrease in atmospheric oxygen in the Northern Hemisphere during 1999–2001 that is concurrent with the strong decrease in our Δ14C observations at La Jolla. In Figure 5, we show the trend over 1992–2007 in seasonally adjusted Δ14C and atmospheric O2/N2 ratios, given as atmospheric potential oxygen (APO). The APO notation refers to O2/N2 ratios that have been corrected for terrestrial biospheric exchange, reported as part per million deviations from a standard ratio [Stephens et al., 1998; Keeling et al., 1998]. We also show the year-to-year change in annual mean to demonstrate that the anomalous decline of both Δ14C and APO in 2000 is not likely to be an artifact of curve fitting procedures. The year-to-year change in annual mean also shows an anomalous decrease in both Δ14C and APO, though the anomalies are reduced due to the coarser temporal resolution in the annual mean compared to the spline curves.

Figure 5.

The derivative of the seasonally adjusted observations at La Jolla for (top) Δ14C (repeated from Figure 2b) and (bottom) APO [Hamme and Keeling, 2008], shown as gray lines. The change in annual mean Δ14C and APO at La Jolla is shown by the black squares.

[42] Hamme and Keeling [2008] estimate that anomalous exposure of deep waters with potential density of 1026.6 kg m−3 (the σθ 26.6 isopycnal) in the Western North Pacific could account for the decrease observed in atmospheric O2/N2. Using the approximations of Hamme and Keeling [2008]to account for the air-sea O2 flux, we estimate that the decrease in Δ14C could also be explained by unusual exposure of and rapid exchange with cool, deep waters in the North Pacific. Δ14C in waters of the σθ 26.6 isopycnal was observed to be −20 ‰ in 1992 [Key et al., 2004], approximately 30 ‰ lower than Δ14C in waters of the σθ 26.4 and 26.5 isopycnals that normally outcrop in the Western North Pacific near the Kamchatka Peninsula. Assuming that Δ14C in these isopycnals did not change substantially between 1992 and 2000, the exposure of the σθ26.6 isopycnal would have enhanced the local air-sea Δ14C gradient by 30–40%. The resulting increase in 14C flux could potentially have caused an anomalous decrease of −1 to −4.5 ‰ yr−1in Northern midlatitude air, depending on the spatial extent of anomalous upwelling and the advection and mixing of the low-Δ14C air from the Pacific. A model study resolving high frequency changes to air-sea fluxes and atmospheric transport from the ocean surface is needed to accurately test this hypothesis.

[43] Anomalies in oceanic fluxes occurring in the tropics are not likely to be responsible for the anomalous decrease in Δ14C in 2000 or other significant variability in the recent trend of Δ14C at La Jolla because the tropical air-sea Δ14C gradient observed in the 1990s was weak. Surface waters of the equatorial Eastern Pacific during the 1990s had Δ14C of approximately 70 ‰, as measured in dissolved inorganic carbon by Nydal et al. [1998] and Key et al. [2004] and in shallow corals at Wolf Island by T. Guilderson. This was only 30 ‰ lower than average atmospheric Δ14C in 2000. By contrast, Δ14C in the σθ 26.6 isopycnal water of the North Pacific was roughly 125 ‰ below atmospheric Δ14C. Moreover, the anomalous drop in atmospheric Δ14C in 2000 was not observed at Cape Grim, Australia [Levin et al., 2010], suggesting the cause of the anomaly took place in northern regions.

[44] The rate of decrease in Δ14C observed at La Jolla in 1992–93 also appeared to be quite rapid, with Δ14C dropping at a similar rate as in 2000. The anomaly in 1993 was probably smaller in magnitude than in 2000, however, since an overall slowing in the rate of decrease of Δ14C occurred between 1993 and 2000. Observations at Jungfraujoch show a similar rate of decrease in 1992–93 as at La Jolla (−10 ‰ yr−1), but the trend of Δ14C in the longer record at Jungfraujoch clearly slowed between 1986 and 2000 such that the trend observed in 1992–93 was not especially prominent [Levin et al., 2010].

5.3. Periodic Variation in Δ14C

[45] To identify timescales of periodic variability we performed a spectral analysis on Δ14C at La Jolla after adjusting the observations to regularly spaced monthly values and linearly detrending. The monthly values were calculated by fitting a function to the observations that included a linear trend, an annual harmonic and an error term that was evaluated using a loose spline (cutoff period of 4 months), and then evaluating the function at mid-month. A power spectrum of the Δ14C observations expressed a strong peak at an annual period; we investigate this seasonal variation in an accompanying paper [Graven et al., 2012]. A smaller peak was also present at a period of 28 months.

[46] Variation in Δ14C of atmospheric CO2 at the 28 month (2.3 yr) period has not previously been reported. Variation on periods of 2.6–5.8 yr at Wellington, New Zealand over 1970–1995 was described by Dutta [2002], and related to perturbations in air-sea exchanges [Rozanski et al., 1995] and, potentially, air-land exchanges caused by the El Niño/Southern Oscillation.

[47] The smaller peak at the 28-month period can be investigated by comparing a spectrum of the Δ14C data, after removing the seasonal variation, with the corresponding red noise spectrum with the same one-lag autocorrelation coefficient [Wilks, 1995]. The seasonal variation was removed from the data in two ways. First, by using the spline curve described in Section 3. Second, by applying a low-pass Butterworth filter with 10th order and cutoff period of 24 months. In both methods, the 28-month period was outside the 0.05 confidence interval of the corresponding red noise spectra (Figure 6a), suggesting that observed variability at this period is real and significant. Variation at 2–3 year timescales is also apparent in the observed trend (Figures 2b and 6b) and suggests that Δ14C may be sensitive to climatic variations that operate on similar timescales.

Figure 6.

(a) Power spectra of the QBO index and of linearly detrended, seasonally adjusted Δ14C at La Jolla using low-pass filter and spline techniques as solid lines. Dashed lines show corresponding red noise spectra. The 28-month period is indicated by the vertical gray line. (b) The Δ14C trend anomaly and QBO index for 1992–2007, where the axis of the QBO index is inverted. The QBO index shows the zonally averaged 50 hPa zonal wind anomaly at the equator in m s−1 (available at http://www.cpc.ncep.noaa.gov/data/indices/).

[48] One climatic mode that may be associated with Δ14C variability of a 28 month period is the Quasi-Biennial Oscillation (QBO), a periodic shifting of the zonal winds in the tropical stratosphere that regulate tropical upwelling into the stratosphere and the planetary waves that influence extratropical tropospheric weather patterns and stratosphere-troposphere exchange [Baldwin et al., 2001]. The QBO's dominant period is also 28 months, the same period observed in Δ14C at La Jolla (Figure 6a). In an atmospheric transport modeling study, Hamilton and Fan [2000]found that the QBO had a significant influence on simulated tropospheric growth rates of long-lived trace gases N2O and CH4through its influence on stratosphere-troposphere transport, suggesting the QBO may also have significant effects on Δ14C variability. The effects of the QBO on Δ14C variability are likely to be opposite to the effects on N2O and CH4, since the stratosphere is a source for 14C and a sink for N2O and CH4. Figure 6b shows that positive anomalies in the trend of Δ14C at La Jolla appear to coincide with negative anomalies in the QBO index at 50 hPa, which correspond to easterly winds and enhanced upwelling in the tropical stratosphere. The mechanism by which the QBO could modulate the tropospheric growth rate of Δ14C may be associated with the transport of tropospheric, low-Δ14C air into the stratosphere in the tropics, or with the transport of stratospheric, high-Δ14C air into the troposphere in the extratropics, and may involve substantial lags [Hamilton and Fan, 2000]. Studies using an atmospheric model would allow potential mechanisms of interaction between the QBO and tropospheric Δ14C to be investigated.

6. Summary

[49] Measurements of Δ14C in CO2 were conducted on monthly samples from La Jolla, California, USA collected between July 1992 and December 2007 by the Scripps CO2 Program. Δ14C analysis was conducted at Lawrence Livermore National Laboratory by accelerator mass spectrometry with average measurement uncertainty of ±1.9 ‰.

[50] The Δ14C observations fit an average linear trend of −5.5 ± 0.1 ‰ yr−1. The decrease of Δ14C was highly variable, however, and expressed a 28-month periodicity. A strong decline in Δ14C in 2000 was concurrent with a strong decline in atmospheric O2/N2 ratios at La Jolla, suggesting a mutual cause from enhanced oceanic ventilation in the North Pacific [Hamme and Keeling, 2008].

[51] Using simple models of the contributions to the global tropospheric Δ14C trend, similar to Levin et al. [2010], we showed that fossil fuel emissions were the strongest influence between 1992 and 2007. Oceanic exchange also contributed a negative influence over the entire period 1992–2007; however, the influence weakened by 70%. The modeled air-sea14C flux and Δ14C disequilibrium is now smaller than in the preindustrial state, suggesting the perturbation caused by fossil fuel emissions appears to have become more important than the perturbation from nuclear weapons testing. Negative influences from fossil fuel combustion and oceanic exchange were moderated by 14C production in the troposphere and biospheric and stratospheric exchange, adding up to an overall rate of change of −4.1 ± 2.2 ‰ yr−1, slower than the average observed trend but consistent within the uncertainties.

Appendix A:: Data Table

[52] Measurements of Δ14C in CO2 samples collected by the Scripps CO2 Program and measured at Lawrence Livermore National Laboratory are provided in Table A1. The CO2 mole ratio and δ13C listed are an average of all measurements with the same sample date. The δ13C values footnoted with an “a” in Table A1 are estimates of δ13C when measurements of δ13C in concurrently sampled CO2 were not available. CO2 mole ratios were measured on the ‘SIO 2008A’ Calibration Scale. The SIO calibration scale for CO2 is established by infrared and manometric analysis of primary reference gases [Keeling et al., 2002]. The SIO calibration scale is tied to the historic CO2 measurements at SIO and independent of the WMO scale since 1995. δ13C values are relative to the international V-PDB standard and include the addition of a −0.112 ‰ offset for consistency with measurements performed at the Center for Isotope Research, University of Groningen, Netherlands.σTot is the total measurement uncertainty in Δ14C. Flagged samples (14%) have been removed. Δ14C measurements at 6 additional SIO clean air sites are reported in the companion paper [Graven et al., 2012].

Table A1. Measurements From La Jolla, California, USA
SIO IDLLNL IDSample DateCO2 (ppm)δ13C (‰)Δ14C (‰)σTot (‰)
  • a

    Estimated δ13C values, when direct measurements were not available.



[53] The air sampling and CO2extractions were supported by a grant from BP, by the National Science Foundation (NSF) under grant ATM-0632770, the U.S. Department of Energy (DOE) under grant DE-FG02-07ER64362 as well as by previous awards from NSF and DOE. This work was performed in part under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract W-7405-Eng-48 and DE-AC52-07NA27344. Radiocarbon analyses were funded by grants from NOAA's Office of Global Programs (NA05OAR4311166) and LLNL's Directed Research and Development program (06-ERD-031) to T.P.G. H.D.G. received support from the UC Office of the President and a NASA ESS Fellowship. H.D.G. also thanks Nicolas Gruber for support and helpful discussions. Alane Bollenbacher conducted CO2 and stable isotope analyses. We thank the anonymous reviewers and Jocelyn Turnbull for helpful comments, Ingeborg Levin for sharing recent Δ14C data from Jungfraujoch and Christian Rödenbeck for assistance with the TM3 Model. This research was also presented in H.D.G.'s doctoral dissertation at the University of California, San Diego, USA, 2008.