A new high precision 14CO2 time series for North American continental air



[1] We develop a high precision Δ14CO2 measurement capability in 2-5 L samples of whole air for implementation within existing greenhouse gas flask sampling networks. The long-term repeatability of the measurement is 1.8‰ (1-sigma), as determined from repeated analyses of quality control standards and replicate extraction and measurement of authentic field samples. In a parallel effort, we have begun a Δ14CO2 measurement series from NOAA/ESRL’s (formerly NOAA/CMDL) surface flask sampling site at Niwot Ridge, Colorado, USA (40.05°N, 105.58°W, 3475 masl) in order to monitor the isotopic composition of carbon dioxide in relatively clean air over the North American continent. Δ14CO2 at Niwot Ridge decreased by 5.7‰/yr from 2004 to 2006, with a seasonal amplitude of 3-5‰. A comparison with measurements from the free troposphere above New England, USA (41°N, 72°W) indicates that the Δ14CO2 series at the two sites are statistically similar at timescales longer than a few days to weeks (i.e., those of synoptic scale variations in transport), suggesting that the Niwot Ridge measurements can be used as a proxy for North American free tropospheric air in future carbon cycle studies.

1. Introduction

[2] The Northern Hemisphere tropospheric radiocarbon burden almost doubled from natural levels in the early 1960s due to atmospheric nuclear weapons testing, but has since decreased markedly as the atmospheric overburden of 14C has been absorbed into the oceans and terrestrial biosphere [Levin et al., 1985]. Soon after above ground weapons testing ceased, the radiocarbon content of atmospheric carbon dioxide (Δ14CO2) fell by up to 100 permil per year (‰/yr), reflecting the rapid uptake of 14C by surface reservoirs. Most “bomb 14C” was initially injected into the stratosphere, leading to seasonal variations of Δ14CO2 within the troposphere as large as 200‰ in response to seasonal variation in cross-tropopause exchange [Levin and Kromer, 1997; Nydal and Lövseth, 1996; Manning et al., 1990]. As the atmospheric 14C burden has decreased and become more uniform, both the annual secular change and seasonal cycle amplitudes have fallen to current values of ∼5‰/yr and ∼10‰, respectively, so that the detection of annual and seasonal changes now demands high precision measurement. Conventional gas counting can obtain precisions of up to 1.2‰, but requires relatively large samples (of order 15m3), which are typically collected over a period of days to weeks [Levin and Kromer, 2004; Levin et al., 2003; Manning et al., 1990; Tans et al., 1979]. Here we describe improvements in 14C measurement by accelerator mass spectrometry (AMS) yielding a long-term Δ14C measurement repeatability of as good as 1.8‰ (1-sigma) in 2-5 L samples of air. The small sample size allows measurement in existing greenhouse gas flask sampling networks (such as operated by NOAA/ESRL, http://www.cmdl.noaa.gov/ccgg/) and provides the opportunity for direct comparison with measurements of other species from the same samples (see below). The increased precision permits resolution of annual and seasonal trends in Δ14CO2, and provides for a fossil fuel CO2 detection capability of better than 1 ppm of CO2 [Turnbull et al., 2006].

[3] Long-term atmospheric Δ14CO2 records exist for several sites in Europe and in the Southern Hemisphere [e.g. Levin and Kromer, 2004; Nydal and Lövseth, 1996; Manning et al., 1990], but there are no such records for North America. The existing records have been used to estimate the exchange rate of CO2 with the ocean [e.g. Peacock, 2004; Orr et al., 2001; Broecker et al., 1995; Hesshaimer et al., 1994; Oeschger et al., 1975], the turnover time of carbon in the terrestrial biosphere [e.g. Trumbore, 1997; Gaudinski et al., 2000], and cross-tropopause exchange [Randerson et al., 2002; Nakamura et al., 1994]). In addition, because fossil fuel CO2 emissions are uniquely characterized by the absence of 14C, Δ14CO2 measurements have been used to quantify fossil fuel CO2 mixing ratios and emissions at regional scales in Europe [Levin et al., 2003; Meijer et al., 1996; Zondervan and Meijer, 1996; Mook, 1980] and, most recently, in aircraft samples from New England [Turnbull et al., 2006].

[4] In 2003 we began measurement of a new Δ14CO2 time series at NOAA/ESRL’s Niwot Ridge, Colorado, USA sampling site (site code NWR, 40.05°N, 105.58°W, 3475 masl), which we envision will be continued as a long-term time series, to complement the existing European and Southern Hemisphere records. CO2 mixing ratio measurements were begun at NWR in 1968 and measurements of other greenhouse gases (CH4, CO, SF6, N2O, H2) and the stable isotopes of CO2 and CH4 have been added over the years [Schnell et al., 2004; Trolier et al., 1996; Miller et al., 2002]. In addition, previous regional scale studies (in Wisconsin and Massachusetts) have made use of Niwot Ridge air as a proxy for North American free tropospheric “background” air against which regional trace gas anomalies have been quantified [Helliker et al., 2004; Bakwin et al., 2004]. Ideally, the new NWR Δ14CO2 series can be used analogously in order to provide the background observation needed to quantify local enhancements of 14C-free fossil fuel derived CO2 around North America. We therefore compare the new Niwot Ridge Δ14CO2 measurements with free tropospheric Δ14CO2 measurements from over New England (41°N, 72°W) and evaluate the skill of Niwot Ridge Δ14CO2 in quantifying the fossil fuel CO2 contributions to boundary layer air over New England previously determined by Turnbull et al. [2006].

2. Methods

2.1. Methods: Sample Collection

[5] Measurements are presented from both repeat analyses of a single tank of air and from authentic atmospheric samples. The single tank (nominally 4200 L at STP) was collected at Niwot Ridge, Colorado, USA in November 2002 to provide replicate extraction aliquots for evaluation of long term measurement repeatability and is designated NWTstd. Authentic samples are collected each week at Niwot Ridge, Colorado, USA (site code NWR, 40.05°N, 105.58°W, 3475masl) as part of the National Oceanic and Atmospheric Administration Earth System Research Laboratory (NOAA/ESRL) Global Co-operative Air Sampling Network [Schnell et al., 2004]. Two flasks are filled in series, which are measured individually for several trace gas species including CO2 and CO. Flasks are flushed with air for about nine minutes, followed by a fill time of about one minute. On alternate weeks, an additional pair of flasks is flushed and filled in a second ten minute period immediately following the filling of the first flask pair. The second flask pair is measured for CO2 mixing ratio only and provides additional material for Δ14CO2 quality control purposes. On sampling dates when only two flasks are collected, we combine the air from both flasks to provide a single sample, filled at the same time. When four flasks are collected, we obtain two theoretically identical samples by combining two flasks (one from each simultaneously filled pair) for each sample. The resulting authentic replicate samples each represent the same time-averaged filling period. Sample sizes for Δ14CO2 measurement are 2-3 L of whole air for single samples, and 3-4 L each for replicated samples.

2.2. Methods: 14C Preparation and Measurement

[6] Sample preparation is undertaken at the University of Colorado Laboratory for AMS Radiocarbon Preparation and Research. Extraction of CO2 is performed cryogenically, using a method based on that of Zhao et al. [1997], with a scaled-up system to allow faster extraction for the large sample size required for 14C measurement (Figure 1). Briefly, the sample air is allowed to flow through a vacuum system at 200 standard milliliters per minute (mL/min (STP)). Water is removed by an ethanol trap cooled to −90°C with liquid nitrogen, and CO2 and N2O are frozen into a liquid nitrogen trap (−196°C) while other gases pass through to the vacuum pump. The internal pressure is controlled to stay below six Torr to ensure complete freezing of CO2 without freezing O2 or CH4 (no attempt is made to separate non-interfering N2O).

Figure 1.

Vacuum system for extraction of CO2 from air. Sample flasks containing whole air are attached to the left side of the line. Sample air is released through the flow controller at 200 mL/min (STP) and while continuously pumping with a rotary vane pump. Water is frozen out using ethanol at −90°C in the left hand cold trap. CO2 is quantitatively recovered by freezing in liquid nitrogen in the second trap. Two trap loops are included to ensure quantitative extraction of CO2. No attempt is made to remove N2O, which freezes out along with CO2, but does not interfere with the 14C measurement. The pressure in the CO2 trap is kept below six Torr, to ensure that methane is not also collected. Extraction is continued either for a set time, or until all the sample air has passed through the extraction system. The CO2 is sublimed and quantified in the known volume cold finger, then flame-sealed into a Pyrex ampoule.

[7] For NWTstd samples, extraction is continued for 20–60 minutes (depending on the final aliquot size required), and then the tank is closed off. For authentic flask samples, the extraction is continued until all the air has been extracted from the flask, with typical extraction times ranging from 20–40 minutes. In all cases, the CO2 yield is quantified in a known volume manifold and then transferred to a flame-seal tube for graphitization on a separate vacuum system. Sample size is 0.4–1.0 mg of carbon (mgC). For NWTstd, we commonly extract twice this amount of CO2, and split it into two aliquots under equilibrium conditions.

[8] To test for variability associated with the CO2 extraction process, we measured the δ13C of CO2 aliquots extracted from our NWTstd tank. δ13C measurements were made at the INSTAAR Stable Isotope Laboratory, with a reported precision of 0.01‰ [http://instaar.colorado.edu/sil, Trolier et al., 1996]. Firstly, we tested for complete extraction of CO2 by varying the flow rate during extraction from 40–200 mL/min (STP). Secondly, we repeated extractions at 200 mL/min (STP) to estimate the uncertainty due to the extraction process. Thirdly, we filled a series of sample flasks with air from the NWTstd tank, and extracted these following the process used for authentic samples to check for variability due to flask filling and extraction. δ13C values for all these tests differed by less than 0.05‰, which equates to less than 0.1‰ in Δ14C. These values suggest that extraction of CO2 is complete at the 200 mL/min (STP) flow rate, and that variability due to flask filling and extraction is not significant relative to our Δ14C measurement precision.

[9] Graphite is produced by reduction of CO2 with hydrogen over an iron catalyst, shown schematically in Figure 2 and based on the method of McNichol et al. [1992]. The iron catalyst is pre-baked at 400°C for 30 minutes, except for samples analyzed prior to December 2004, for which the pre-bake was done at 800°C. The sample CO2 is then introduced into a small reaction manifold (4–7mL volume) and hydrogen gas is added in a ratio of 2.5–3 times the number of moles of sample CO2 to provide a slight stoichiometric excess and ensure complete reaction. The reduction is performed at 625°C, precipitating graphite onto the iron catalyst. Water produced by the reaction is frozen using a thermo-electric cooler at −15 to −17°C, except for samples analyzed prior to July 2005, when water was frozen using an ethanol-liquid nitrogen slush bath at temperatures between −5 and −20°C. Reaction progress is monitored by pressure; reactions typically take four to five hours, but are allowed to continue for eight hours to ensure completion.

Figure 2.

Graphitization system. The quartz side arm is loaded with iron powder as a catalyst. Sample CO2 is introduced from the right hand side, and frozen with liquid nitrogen as hydrogen gas is introduced. A furnace at 625°C is placed over the quartz side arm, and a thermoelectric cooling system is attached to the vertical Pyrex tube. Graphite is precipitated onto the iron catalyst, and water is frozen into the Pyrex tube, allowing the reaction to go to completion. Reaction progress is monitored via the attached pressure transducer.

[10] Minimally different reaction conditions are used depending on the AMS laboratory conducting the 14C measurement. Low lithium quartz glass reaction tubes were used for measurements made on the Rafter Radiocarbon Laboratory (RRL) AMS system where di-lithium contamination interferes with the measurement, and for RRL the Fe catalyst to graphite mass ratio was 1.2 to 1.3. At the University of California, Irvine (UCI) AMS facility, we use 3–4 mg of Fe catalyst regardless of sample size. This choice avoids an apparent mass-independent fractionation observed when sample targets are near exhaustion in the ion source, which we believe is related to a small loss at the edges of the 12C ion beam as it passes through apertures in the AMS; this effect is determined by the abundance of each isotope, rather than the isotopic mass, and substantially impacts only the large 12C ion beam. For both AMS systems, the resulting graphite/catalyst mixture is packed into an aluminum target holder for AMS 14C measurement.

[11] We use NBS Oxalic Acid I (Ox-I) as our primary 14C reference standard [Stuiver and Polach, 1977; Olsson, 1970]. Aliquots of Ox-I CO2 are split under equilibrium conditions (whereby the CO2 is allowed to equilibrate at room temperature into a single large volume, then divided into two aliquots by closing a valve to separate the large volume into two smaller volumes) from a single large parent flask and graphitized in the same manner as unknowns. We use NBS SRM 4990 Oxalic Acid II (Ox-II) [Stuiver, 1983], prepared in the same manner as Ox-I as a secondary quality control check. The total processing and measurement blank is estimated by regular extraction and measurement of CO2 from a tank of synthetic 14C-free air. Preparation and measurement follow the same protocols as for unknowns. Further quality control is provided by repeated measurements of NWTstd, and from replicate measurements of authentic samples (see section 3).

[12] The 14C content of each graphite sample is measured by AMS at either UCI (analyses since June 2004) or RRL (prior to June 2004). A batch or “wheel” of samples is measured with concurrent standards in a single run. At UCI, a wheel of up to 40 targets is measured over a 24-hour period, including eight Ox-I primary standards, a blank, two Ox-II and three to five NWTstd quality control targets, and 24–26 unknowns. Each target is measured to 300,000–750,000 14C counts (for blanks, a time limit is used) in 6–15 separate exposures, with standards interspersed throughout the wheel. The 14C content measured in each exposure is calculated firstly as the 14C/13C ratio each sample exposure divided by the 14C/13C of the 6 closest Ox-I standard exposures. The mean of this ratio for all exposures for the sample is determined to obtain the “ratio to standard”. The fraction modern (ratio of sample to the absolute radiocarbon standard) is calculated from the “ratio to standard”, following calculations described in detail in Donahue et al. [1990] and Stuiver and Polach [1977]. This calculation includes a correction for the process blank, which is measured in the same wheel. The result is also corrected for isotopic fractionation and normalized to a δ13C of −25‰ using the δ13C measurement obtained in the AMS system concurrently with the 14C measurement. This value reflects any fractionation during graphitization and AMS measurement, and may deviate from the environmental δ13CO2 value by up to several permil.

[13] AMS measurement and data analysis procedures differed slightly for measurements at RRL. A “wheel” consisted of eight unknowns along with three to four Ox-I standards, and one or two NWTstd targets. We measured two or three such wheels over a consecutive two or three day period. Each target was measured to ∼300,000 14C counts in ten exposures, and the “ratio to standard” for each exposure was determined from a line of best fit to all Ox-I exposures in the wheel. The δ13C value was obtained offline from a stable isotope mass spectrometer measurement of δ13CO2 from the same flask, performed at the INSTAAR Stable Isotope Laboratory [http://instaar.colorado.edu/sil, Trolier et al., 1996]. This measurement does not account for any fractionation during graphitization and AMS measurement. Fraction modern was then determined using the same method as for UCI, but using the offline δ13C value. Use of the latter may contribute to the larger spread in measurements at RRL.

[14] A tank of synthetic 14C-free air (CO2-free air spiked with 14C-dead CO2) is used as a process blank. An aliquot is extracted and measured for 14C in every measured wheel (at RRL, one blank was included in each set of 2–3 wheels). The measured Δ14CO2 ranges from −997 to −999‰ with an average of −998‰. The effect of this level of contamination on the Δ14C of modern air samples is insignificant, but we continue these measurements in each measured wheel as a quality control measure to monitor for contamination during sample preparation and AMS analysis.

[15] All results from either lab are corrected for radioactive decay since the date of collection and reported as Δ14C, the permil deviation from the absolute radiocarbon standard, such that

equation image

where FN is the normalized fraction modern, λ is the decay constant of 14C and x is the date of collection for each sample [Stuiver and Polach, 1977].

[16] The AMS single sample precision is determined from the statistical uncertainties on both sample and associated standards, accounting for both the “internal” statistical uncertainty determined from the number of 14C counts, and the “external” uncertainty determined from the variability amongst the different exposures for that sample. Occasionally, unstable operating conditions in the UCI AMS system occur which are readily identified by the larger scatter in the measured Ox-I values. For these wheels, we adjust the reported uncertainty to reflect the scatter in the Ox-I values. This has occurred in six of 28 wheels measured at UCI since April 2004, and these are treated as a separate dataset for quality control purposes (designated “UCI Poorly Performing” or “UCIPP”). At RRL, the reported single sample precision was 2.1–2.4‰ in Δ14C. At UCI, where we are able to obtain more 14C counts, the single sample precision is 1.4–3.0‰, with the higher uncertainties reflecting samples where instability problems occurred. We assess how well these reported precisions reflect the true measurement repeatability below.

3. 14C Analysis Precision and Repeatability

[17] The true repeatability of our 14C measurements is examined primarily by preparing and measuring replicate aliquots of NWTstd (Figure 3). Individual samples are prepared by extracting sufficient air to obtain either: (a) 1.0–1.5mg C and then splitting the extracted CO2 into two aliquots under equilibrium conditions; or (b) by extracting 0.4–1.0mg C for a single sample. Each sample is then graphitized individually and measured by AMS at RRL or UCI. No significant inter-laboratory difference is observed (mean values and standard errors are 73.5 ± 0.5‰ for RRL and 73.3 ± 0.2‰ for UCI). Some wheel-to-wheel offsets can be seen in the UCI data, but there has been no long-term drift in the mean value. The 1-sigma standard deviation of the NWTstd measurements is 2.5‰ for RRL and 1.8‰ for typical (non-UCIPP) UCI wheels, with the larger spread in RRL measurements resulting mainly from the lower number of 14C counts and the offline δ13C measurement. The UCIPP wheels have a 1-sigma standard deviation of 2.7‰. We therefore assign uncertainties to our authentic sample measurements as the repeatability for the given dataset (RRL, UCI or UCIPP), or the AMS reported single sample precision, whichever is larger.

Figure 3.

Measurements of Ox-I, OX-II and NWTstd for measurements made between November 2003 and March 2006. Measurements are shown in order of date of measurement, and the y-axis is the same for all panels, but is not linear. Squares indicate measurements at RRL (only NWTstd measurements are shown for RRL); diamonds indicate measurements at UCI. Poorly performing wheels at UCI are shown as open diamonds. Error bars are the one-sigma AMS single sample precision as described in the text. The solid line indicates the expected or mean value for each sample type. For the primary standard Ox-I, the mean for each wheel is set at the book value [Stuiver and Polach, 1977], and the scatter represents the repeatability of the measurements within a given wheel. For Ox-II, the expected value is 340.66‰ [Stuiver, 1983]. For NWTstd, the mean value is 73.4‰, with values of 73.5 ± 0.5‰ for RRL and 73.3 ± 0.2‰ for UCI. Note that while NWTstd and Ox-II are measured in every wheel containing air samples, we also include additional Ox-I measurements from wheels containing other sample types.

[18] As a secondary check, we also measure aliquots of Ox-II at UCI (Figure 3) prepared in the same manner as our primary Ox-I standard, and these have a mean value of 339.0‰ (decay corrected to the 1950 value) and a one-sigma repeatability of 1.6‰. The book value for Ox-II is 340.66‰ [Stuiver, 1983].

[19] The variability (but not absolute value) of Ox-I targets within each wheel provides a third measure of the intra-wheel repeatability (Figure 3), and we obtain 2.5‰ for RRL and 1.8‰ for regular UCI wheels, consistent with the NWTstd estimates of repeatability. For poorly performing UCI wheels the repeatability is 2.9‰. Because the Ox-I values are always normalized to the book value of Ox-I (Fraction modern of 1.03290 in 2005), the repeatability of this measurement indicates the intra-wheel repeatability, rather than the inter-wheel variability reflected by the other quality control standards. There is no significant difference between the intra-wheel and inter-wheel variances, suggesting that we can legitimately compare results for different wheels.

[20] Finally, we use authentic field sample replicates to confirm that we have correctly assigned the uncertainties using the methods described above (Figure 4). Two types of authentic sample replicates are obtained: (1) two sets of flasks are filled simultaneously, and then treated as completely different samples throughout sample preparation and analysis; and (2) a single CO2 extraction is performed, and the resulting CO2 gas is split (under equilibrium conditions) into two replicates for graphitization and measurement. Sample replicates have come from a variety of locations, including 35 pairs from NWR. We have measured 74 replicate sample pairs since 2004, and obtain a reduced chi-square (χν2) of 1.07 for all replicate pairs. The 35 replicate pairs of NWR (all of which except one were obtained using method 1) yield an χν2 = 0.79. These values indicate that we have accurately represented the uncertainties for these samples.

Figure 4.

Pair differences for replicate analyses of authentic samples. Each pair Δ14CO2 is normalized to a mean of 0 ‰, and the symbols indicate the deviation of each measurement from the pair mean. Error bars are the 1-sigma uncertainty on each individual measurement as described in the text. Squares are measurements made at RRL; circles are measurements made at UCI. Closed symbols indicate the first sample from each pair; open symbols are the second sample from the pair. Xν2 = 1.07 for the 74 replicate pairs.

4. Results and Discussion

4.1. NWR Results

[21] We began biweekly Δ14CO2 measurements at Niwot Ridge, Colorado, USA (NWR, 40.05°N, 105.58°W, 3475masl) in May of 2003 and present results up to January 2006 (Table 1, Figure 5a). Samples were collected over periods of 1–20 minutes as described above, typically during the afternoon. The high altitude location on the continental divide and generally westerly air stream suggest that NWR measurements should be representative of relatively clean free tropospheric air over North America. However, upslope easterly wind events occasionally bring locally polluted air from the Denver metropolitan region 40 miles to the east. To obtain a “clean troposphere” dataset, we use measurements of carbon monoxide (CO) from the same sample flasks to identify polluted air (the CO observations are available digitally at ftp.cmdl.noaa.gov/ccg/co/flask/event). CO is produced by incomplete fossil fuel combustion, so high CO values are a qualitative indicator of polluted air. Additional CO measurements from samples collected at NWR, but not measured for Δ14CO2, are included in the CO dataset. A curve is fitted to the CO data [Thoning et al., 1989] to define the (seasonally varying) background and samples where the CO value is more than 15 parts per billion (ppb) above the fitted curve are flagged as polluted (Figure 5b). Using these criteria, 11 of the 79 14CO2 sampling dates are excluded from the original dataset. Alternately, we flagged all points where the wind direction was between 0° and 180° (i.e. from the east). This excluded 22 of the 79 data points, including all the points that were flagged using the CO method. We chose the CO flagging method, as it appeared to best identify the polluted samples.

Figure 5.

(a) Measured Δ14CO2 values for NWR. Open symbols indicate samples that were flagged as polluted, as indicated by high CO values shown in panel b and described in the text. Error bars are one sigma uncertainties as described in the text. The curve is fitted to the remaining NWR Δ14CO2 background time series (solid symbols), using a smoothed trend obtained from a linear fit with two harmonics and smoothed residuals from a low-pass filter with a 180-day cutoff in the frequency domain [Thoning et al., 1989]. (b) CO values measured at NWR, with a fitted curve [Thoning et al., 1989]. Open symbols indicate CO values more than 15 ppb higher than the fitted curve.

Table 1. Δ14CO2 Measurements From Niwot Ridge, Colorado, USA
CURL lab codeadate collectedbtime collectedbAMS labcΔ14CO2unc.dFlage
  • a

    CURL is the laboratory code for each analysis.

  • b

    Date and time collected are in UTC.

  • c

    AMS indicates which AMS facility was used for the AMS measurement; all sample preparation was performed at the University of Colorado.

  • d

    Uncertainty is the one-sigma standard deviation as described in the text.

  • e

    An F in the flag column indicates samples where local fossil fuel pollution was identified (the CO mixing ratio exceeded 15 ppb above the baseline level); these data points are excluded from the background dataset.


[22] The cleaned dataset is shown in Figure 5a, and the curve shown is fitted following Thoning et al. [1989], using one linear and two harmonic terms, with fit residuals being added back using a low-pass cutoff filter with a 180 day cutoff in the frequency domain, equivalent to an averaging filter in the time domain with a full width half-maximum of 106 days. Adding additional polynomial or harmonic components did not improve the fit. The residual scatter about the fitted curve (of 2.6‰ at one standard deviation) likely results from the individual measurement uncertainties and authentic, short-term environmental variability. A possible example of the latter effect is the high Δ14C value observed at NWR on January 7th, 2004 (77.7 ± 2.6‰), which may reflect influence of air originating in the upper troposphere where substantial production of 14CO2 occurs (consistent with analysis of the back trajectory for this sample (data not shown)).

[23] There is a consistent decrease in Δ14CO2 through the record, with Δ14CO2 changing by −5.7‰/yr between the beginning of 2004 and 2006 according to the fitted curve shown in Figure 5a. The simple linear trend component of the fitting process is similar at −5.2 ± 0.1‰/yr, although the short length of the record may result in bias in this fit. Given current fossil fuel carbon emission rates of about 7.5 GtC/yr [Marland et al., 2003] and the current atmospheric loadings of about 380 ppm CO2 and 60‰ Δ14CO2, global emissions of 14C-free fossil fuel CO2 are expected to reduce Δ14CO2 by 9.7‰/yr. Therefore, the effect of all other sources and sinks must be to increase the atmospheric Δ14CO2 by 4–4.5‰/yr. This is consistent with only a small remaining 14C disequilibrium between the atmosphere and surface reservoirs, and a change in sign (from negative to positive) of the biospheric disequilibrium flux as bomb 14C taken up by the biosphere in the last few decades returns to the atmosphere [Naegler and Levin, 2006; Randerson et al., 2002]. These changes imply that fossil fuel emissions are now a dominant control on the temporal trend in Δ14CO2.

[24] An analysis of the harmonic component of the fit reveals an apparent seasonal cycle, with a maximum in August, a minimum in February–March, and peak-to-trough amplitudes of 3–5‰, very similar to the average behavior at Jungfraujoch, Switzerland (46.55°N, 7.7°E, 3450 masl) between 1986 and 2003 [Levin and Kromer, 2004]. Fossil fuel CO2 emissions have a maximum in the Northern Hemisphere winter, with an estimated seasonal amplitude of 20% of the total emissions for the United States and likely a similar or slightly larger seasonal amplitude for other regions [Blasing et al., 2005]. This seasonal change has been modeled to produce a 3–4‰ seasonal amplitude in Δ14CO2 in the Northern Hemisphere mid-latitudes [Randerson et al., 2002] with a Δ14CO2 maximum in the Northern Hemisphere summer. This is consistent with the NWR Δ14CO2 observations for 2004 and 2005, both in the phase of the seasonal cycle and in its amplitude. The 2003 results (which do not contribute significantly to the harmonic fit) appear to oppose this seasonal trend, showing a minimum in August of 2003, but the time series is too short to determine whether this is a real anomalous seasonal signal. The seasonal amplitude is larger in 2005 than in 2004, and this year-to-year variability in seasonal amplitudes is consistent with past variability in other Δ14CO2 records [Levin and Kromer, 2004], suggesting that this difference is not simply due to the small number of samples measured. It is also unlikely that fossil fuel emissions vary sufficiently from year to year to explain this. The observed changes in amplitude of the seasonal signal are more likely due to differences in atmospheric transport or to other seasonally varying sources, such as the magnitude of net cross-tropopause exchange or terrestrial biospheric respiration.

4.2. Comparison With Other North American Sites

[25] Previous studies [Bakwin et al., 2004; Helliker et al., 2004] have used NWR as a proxy for well-mixed free tropospheric air across much of the North American continent, which was a primary motivation for establishing an initial Δ14CO2 monitoring effort at this location. In order to begin to evaluate how well NWR Δ14CO2 represents North American free tropospheric Δ14CO2, we compare the NWR results with a second free tropospheric dataset collected over New England beginning in 2004 [Turnbull et al., 2006]. The New England samples were collected via aircraft at varying altitudes between 3000 and 5000 m attitude at two locations: over Harvard Forest, Massachusetts (HFM, 42°32′N, 72°10′W) and above Portsmouth, New Hampshire (NHA, 42°57′N, 70°37′W). Sampling altitudes for each date were chosen to obtain the “cleanest” air in the vertical profile (the selection criteria are described in more detail in Turnbull et al. [2006]).

[26] The NWR and New England datasets agree well, both in terms of absolute value and in temporal trends (Figure 6). The dip in Δ14CO2 values from 67‰ to 62‰ in May 2004 is present in both records, as well as the strong downward trend beginning in August 2004. The absolute values are also similar, with six-monthly mean Δ14CO2 values overlapping at one standard deviation (Table 2). Higher frequency variability in the two datasets does not appear to be shared, but most of this difference may be attributed to the snapshot sampling method and to measurement uncertainty, as the scatter in the residuals from the fitted data approaches the measurement uncertainty for both datasets.

Figure 6.

Comparison of NWR and New England Δ14CO2 observations. Diamonds are the NWR Δ14CO2 background time series as shown in Figure 5, and upper troposphere measurements from over New England are shown as open squares. In some cases, two measurements were made at different altitudes in the New England free troposphere, or measurements were made at both New England sites on the same day; in either case, both results are shown. The solid line is a best fit to the NWR data, using the same method as for Figure 5, but using a shorter 90-day cutoff filter. The dashed line is a fit to the New England data using the same fitting procedure.

Table 2. Six Monthly Mean Values for NWR and New Englanda
 NWRNew England
  • a

    Uncertainties are the standard error for the averaged values.

Jan–Jun 200465.9 ± 0.665.3 ± 0.6
Jul–Dec 200464.3 ± 0.564.5 ± 0.5
Jan–Jun 200558.6 ± 0.659.9 ±0.8

[27] In order to demonstrate the usefulness of NWR as a proxy for background air we re-calculate the fossil fuel CO2 contribution in New England boundary layer air samples, which were previously determined in Turnbull et al. [2006]. The fossil fuel CO2 contribution (Cff) is related to the observed Δ14CO2obs) and the background Δ14CO2bg) value, such that

equation image

Δ14CO2 of fossil fuels (Δff) is, by definition, −1000‰. There is also a small correction term for heterotrophic respiration (indicated as Cr and Δr), which is typically smaller than the uncertainty in Cff. In the calculation performed by Turnbull et al. [2006], Δbg was obtained from free tropospheric samples collected in the same vertical profile as the continental boundary layer samples for which Cff was calculated. Here we recalculate the fossil fuel CO2 contribution using the monthly mean Δ14CO2 values from NWR as Δbg (Figure 7). There is no significant or systematic difference between the estimates of Cff obtained using the local, real-time value of Δbg vs. that based on NWR monthly means.

Figure 7.

Estimates of the recently added fossil fuel CO2 mixing ratio (Cff) in the boundary layer over New England. Open symbols are the values from Turnbull et al. [2006], where background Δ14CO2 values were obtained from free tropospheric observations made on the same day in the same vertical profile. Closed symbols are Cff estimates using the same method, except that background Δ14CO2 values were from the NWR monthly mean.

[28] While not conclusive, because the New England samples examined here are deliberately biased to clean air sampling, the observed agreement in Figure 6 suggests that the free troposphere is well mixed over North America, and that venting of 14C-free emissions through the top of the continental boundary layer is small relative to the free tropospheric mixing.

5. Conclusions

[29] Improved precision in Δ14CO2 measurements allows us to resolve Δ14CO2 differences of a few permil. The ability to measure samples as small as two liters of whole air allows direct comparison with other trace gases measured in the same flasks, and provides a complementary method to the larger, time-averaged samples traditionally used for atmospheric Δ14CO2 studies.

[30] Our high-precision measurements from Niwot Ridge, Colorado, USA provide (when screened for occasional, easily identified, local pollution events) a new background Δ14CO2 record for North America for application to carbon cycle studies. Changes in Δ14CO2 during the two-year record appear to be dominated by the impacts of fossil fuel combustion, both in the secular decrease (of 5.7‰/yr), and in the seasonal cycle. Comparison of the results from NWR and the New England free troposphere suggest that the free troposphere is well mixed zonally across North America with respect to Δ14CO2 on annual and seasonal time scales, suggesting that the NWR record may be useful as a proxy for background Δ14CO2 for North America.


[31] Funding for this research was provided by NOAA Office of Global Programs (OAR4310098), the ISAT Linkages Fund (03-CSP-20-SPAR) and the Colorado Mountain Club Foundation. We thank Chad Wolak for assistance with sample preparation, Valerie Claymore for assistance with stable isotopic measurements, and Duane Kitzis and Mark Losleben for collection of additional samples and standards at Niwot Ridge.