The gas chromatography (GC) analytical system (Fig. 1) employed at Portland State University for measuring N2O and SF6 in archived air samples is based on the configuration used by Hall et al. (2007) and references therein. We use an Agilent 6890N gas chromatograph fitted with a micro-electron capture detector (µECD, Agilent Technologies, Santa Clara, CA). Two Poropak Q 80/100 mesh columns (1.8 m × 2 mm i.d. pre-column, 3.7 × 2 mm i.d. analytical column) achieve peak separation. The carrier gas is P5 (99.999 %, Airgas, Portland, OR) equipped with O2 and hydrocarbon traps (Restek, Bellefonte, PA) to further reduce impurities and to significantly improve baseline signal stability. Two 6-port switching valves (V1 and V2), a 4-port switching valve (V3), and a 16-port multi-position valve (Valvo Instrument Company Inc., Houston, TX) are controlled by Chemstation (V1.A, Agilent Technologies Inc., Santa Clara, CA).

A sample run begins in “back-flush” mode, with the carrier gas flushing the pre-column in the reverse analytical direction to remove the buildup of water on the analytical column that would otherwise eventually elute to the µECD and affect signal baseline. A 16-port multi-position valve is used to introduce pressurized samples into the system; a two-way electric valve (Clippard, Cincinnati, OH) is used to stop sample flow to the sample loop and prevent sample loss. Samples initially pass through a desiccant trap (Perma Pure, Toms River, NJ), before flushing a 10 mL sample loop at 60 mL min-1 for 1.5 min. At this time, V3 rotates, which places the system in “front-cut” mode and allows the sample loop to equilibrate. V1 rotates at 1.75 min and allows the carrier gas to carry the sample N2O and SF6 to the pre-column where separation from O2 and H2O occurs. After O2 elutes through the pre-column to vent, at 3 min V2 rotates and places the pre-column in line with the analytical column, transferring N2O and SF6 to the analytical column. At 4.25 min, the sample has reached the analytical column and V1, V2, and V3 rotate. This begins the back-flush of the pre-column while the analytes are carried to the µECD on the analytical column.

Oven and detector temperatures are maintained at 56 and 310 ∘C, respectively. Carrier gas flow rates are 40 mL min-1 and are maintained by the electronic pressure control of the 6890N. The N2O peak retention time is 6.1 min and the SF6 peak retention time is 7.0 min (Fig. 2). Peak integration is accomplished via Chemstation based on peak height.

All measurements of N2O and SF6 are made relative to a calibrated whole air sample on the NOAA-06A N2O scale and NOAA-14 SF6 scale (NOAA Tank CB11406-A, 328.71±0.5 ppb N2O, 8.76±0.06 ppt SF6), hereafter referred to as the NOAA reference gas. Each sample is analyzed six times and bracketed by six reference gas runs used to measure instrument response and track signal drift. The GC-µECD analytical system was evaluated for precision, reproducibility, and linearity prior to its application to measure mole fractions in historic archive samples.

Precision of measurement was determined by repeated analysis of the reference standard. Histogram distributions in Fig. 3 show 180 residuals (expressed as a percent relative standard deviation) collected from 30 sets of 6 measurements of N2O (Fig. 3a) and SF6 (Fig. 3b) of the NOAA reference gas. Both N2O and SF6 compare well to a normal distribution (black dashed lines), with chi-square goodness of fit p values of 0.16 and 0.35, respectively. For N2O, 1σ=0.16 % whereas for SF6, 1σ=1.1 %. This corresponds to an uncertainty of ±0.52 ppb for N2O and ±0.10 ppt for SF6. The mean measurement uncertainty (1σ) of OHSU-PSU air archive samples for N2O is 0.23 %. The mean measurement uncertainty (1σ) of SF6 in the OHSU-PSU air archive samples ranges between 6.5 % for samples below 1 ppt and 2.5 % for samples at 4 ppt.

Measurement reproducibility was evaluated by repeatedly measuring a dry air sample (Breathing Air, Airgas, Portland, OR) against the NOAA reference gas and evaluating consistency from the standard deviation of the results. The sample was measured 18 times over 2 weeks with mean measured mixing ratios of N2O and SF6 of 390.9 ppb and 13.2 ppt, respectively. The standard deviations in N2O and SF6 measurements are 0.46 ppb and 0.11 ppt, respectively, which are indistinguishable from 1σ precision for a set of six NOAA reference gas measurements.

To ensure accurate results for this work, the detector response was evaluated over the mole fraction range expected for N2O and SF6 in the OHSU-PSU air archive. The range in the Northern Hemisphere N2O mole fraction between 1978 and 1996 is between 295 and 314 ppb (Prinn et al., 2000; Ciais et al., 2013). Archived air sample measurements of the Northern Hemisphere SF6 mole fraction from Trinidad Head, CA, measure below 1 ppt in the 1970s and rise to nearly 4 ppt in 1997; Southern Hemisphere measurements from Cape Grim, Tasmania, and the South Pole show a similar range (Levin et al., 2010; Rigby et al., 2010).

A series of manometric dilutions were prepared from the NOAA reference gas at Portland State University to evaluate the µECD response over the historical N2O and SF6 mole fraction sample range. To characterize the N2O response, the N2O reference gas was diluted with ultrapure air (zero grade, Airgas, Portland, OR; N2O and SF6 at mixing ratios below detection limits) using capacitance manometers (MKS Instruments, Andover, MA; range 0–10 and 0–1000 torr) into 3 L electropolished stainless steel canisters (precision ±0.01 %). The range of N2O mixing ratios produced in the 3 L canisters was 32.2–321.4 ppb. The error introduced from the manometric process is small when compared with measurement uncertainty (maximum 1σ error of ±0.07 ppb for N2O).

To characterize the SF6 response at low part per trillion ( ppt) mixing ratios requires consideration of the effect of the falling N2O tail on the chromatogram baseline. To properly account for this interference, SF6 dilutions at low mixing ratios (0.6–6.0 ppt) must have N2O mole fractions that reflect expected mixing ratios in archived samples (300–315 ppb). Prepared dilutions of SF6 included the addition of an aliquot of 1 ppm N2O (±5 %, Scott Specialty Gases, St. Louis, MO) into the canister prior to dilution with ultrapure air. The maximum error (1σ) in SF6 introduced from the manometric process is small (0.001 ppt) compared with the measurement uncertainty. However, SF6 present in either ultrapure air dilution gas or the N2O aliquot at trace levels below the detection limit of our measurement (<0.1 ppt) contribute to the uncertainty in prepared samples. All dilution samples were measured at PSU on the GC-µECD system over several weeks to account for instrument drift. Tables 1 and 2 provide dilution sample pressures, calculated and observed µECD response, and measured N2O and SF6 mole fractions with the error in measurement used to characterize the GC-µECD linearity.

Results of linearity experiments are shown in Fig. 4. For N2O, a slope of 0.8747±0.028 (95 % CI) is found over the data range from 289.7 to 328.7 ppb, which is most relevant for this work. A linear fit is a good model for the deviation from expected over this range (R2=0.964); additional polynomial terms are not statistically robust. This results in sample measurements deviating from expected by ∼0.14 ppb ppb-1 N2O difference from the NOAA reference. For the range of the N2O in the OHSU-PSU air archive, all N2O samples are adjusted for a linear correction of the following form: 1[N2O]X=a1[N2O]Y+a22a1=1.143±0.037(95%CI)3a2=-47.24±11.49(95%CI), where [N2O]Y is the response evaluated N2O mole fraction and [N2O]X is the corrected value. The slope and y intercept, as well as their 95 % confidence intervals, are represented by a1 and a2, respectively. This correction is applied to all sample N2O measurements (corrected values ranging between 298.9 and 314.8 ppb).

The entire NOAA reference gas dilution range for N2O (32–321 ppb) results in a deviation that can be adequately modeled using a third degree polynomial. The linear fit discussed above is indistinguishable from the full third degree polynomial over the N2O mixing ratio range of the OHSU-PSU air archive. However, if measuring N2O samples with a difference of more than 80 ppb compared with the NOAA reference gas, the full third degree polynomial is necessary to correct for the nonlinear response in the µECD.

For SF6, the prepared sample range over which the linear correction is applied is 0.59–8.76 ppt, which is most relevant for this work. The slope of the SF6 linear fit is 0.9728±0.017 (95 % CI) and is a good model for the deviation from expected over this range (R2=0.9995). This results in a deviation from expected of ∼0.03 ppt ppt-1 SF6 difference from the NOAA reference when measuring samples. All SF6 measurements are adjusted for a linear correction of the following form: 4[SF6]X=b1[SF6]Y+b25b1=1.028±0.018(95%CI)6b2=-0.294±0.099(95%CI), where [SF6]Y is the response evaluated SF6 mole fraction and [SF6]Y is the corrected value. The slope and y intercept, as well as their 95 % confidence intervals, are represented by b1 and b2, respectively. This correction is applied to all sample SF6 measurements (corrected values ranging between 0.6 and 4.3 ppt).

Detector response nonlinearity has been evaluated in previous work by other groups on GC-ECD systems. For N2O, deviations from expected of ∼0.2 ppb ppb-1 difference from the reference gas are are typical when in the linear range (Schmidt et al., 2001; Hall et al., 2007). These are similar to the value reported here for the µECD. Over larger ranges, a similar nonlinear response curve is also reported. SF6 nonlinearity reported in Levin et al. (2010) has a similar curvature to the full N2O nonlinear response previously discussed. However, this curvature is not observed to be significant over the range of SF6 dilutions conducted here.

Measurements of N2O and SF6 mole fractions from 159 samples of the OHSU-PSU air archive were initially filtered for analysis using a 7 median absolute deviation (7MAD) noise filter to remove far outliers. Polynomial fits (first degree for N2O and second degree for SF6) were then applied to the data. Residual values outside of 2σ for N2O and 3σ for SF6 were removed for further data analysis. The entire process removed 12 data points for N2O and 4 data points for SF6 used in the analysis.

Deseasonalized measurements of N2O and SF6 from Cape Meares are shown in Fig. 5a and b, respectively. A locally weighted linear regression (LOWESS) is utilized to smooth the data using a 3 year smoothing window (Cleaveland and Devlin 1988). The confidence intervals around regressions are calculated by bootstrapping residual variability 1000 times. The regression results in a N2O mole fraction of 301.5±0.3 ppb (1σ) in 1980 that increases roughly linearly to the mid-1990s, where the mixing ratio is 313.5±0.3 ppb (1σ) in 1996.

Observations of the N2O mole fraction match well with previously published measurements of N2O from Cape Meares between 1978 and 1998 of 301.2 ppb in 1980 and 313–314.5 ppb in 1996 on the SIO-1998 N2O scale (Prinn et al., 1990, 2000; Khalil et al., 2002). The N2O scale difference between SIO-1998 and NOAA-06 is minimal (Hall et al., 2007). Additional measurements by the Advanced Global Atmospheric Gases Experiment (AGAGE) and NOAA/ESRL (on the SIO-1998 N2O and NOAA-06 N2O scales, respectively) are reported from comparable sample locations. Trinidad Head, CA (41∘ N, 121∘ W), Mace Head, Ireland (53∘ N, 10∘ W), and Niwot Ridge, CO (40∘ N, 106∘ W), all measure ∼313 ppb in 1996 (Prinn et al., 2000; Hall et al., 2007). Together, these comparisons indicate that the N2O in the archived samples has stored well.

The measured SF6 mixing ratio in archived Cape Meares samples is determined to be 0.85±0.03 ppt (1σ) in 1980 and increases to a mixing ratio of 3.83±0.03 ppt (1σ) in 1996. Cape Meares does not have previously reported measurements of SF6 to compare with directly. Measurements of SF6 from Trinidad Head, CA, are reported to be ∼0.85 ppt in 1980 and ∼3.73 ppt in 1996 on the SIO-2005 SF6 scale (Rigby et al., 2010). To convert to the NOAA-06 SF6 scale, values measured on the SIO-2005 SF6 scale are divided by a conversion factor of 0.9991 (Hall et al., 2014). In 1996, values of 3.87, 3.87, and 3.78 ppt are reported for Alert, Canada (82∘ N, 62∘ W), Barrow, AK (71∘ N, 157∘ W), and Niwot Ridge, CO, respectively, on the NOAA-06 SF6 scale (Hall et al., 2011). At these SF6 mixing ratios, the difference between the NOAA-06 scale and the NOAA-14 scale is minimal. Cape Meares SF6 measured values compare well with these Northern Hemisphere locations.

In the Northern Hemisphere, maximum background mixing ratio measurements of SF6 are reported from mid-to-high latitudes (Geller et al., 1997). For the year 1994, measurements from Fraserdale, Canada (50∘ N, 82∘ W), are reported to be 0.14 ppt higher than samples measured from Izaña, Tenerife (28∘ N, 16∘ W) (Maiss et al., 1996). This difference is explained by the vast majority of SF6 emissions coming from the midlatitudes in the Northern Hemisphere (Maiss and Brenninkmeijer 1998; Levin et al., 2010; Rigby et al., 2010). The measured SF6 mixing ratios from Cape Meares, also a midlatitude Northern Hemisphere site, appear to fit in well with the expected meridional gradient when comparing to previously mentioned reported values.

Southern Hemisphere measurements of SF6 from archived atmospheric samples from Cape Grim, Tasmania (41∘ S, 145∘ E), and Neumayer, Antarctica (70∘ S, 8∘ W), are ∼0.6–0.7 ppt in 1980 and ∼3.4–3.5 ppt in 1996 on the SIO-2005 and University of Heidelberg SF6 scales (Levin et al., 2010, Rigby et al., 2010). As with the SIO-2005 SF6 scale, the NOAA-06 and University of Heidelberg scale differences are small. To convert to the NOAA-06 SF6 scale, values measured on the University of Heidelberg SF6 scale are divided by a conversion factor of 0.9954 (Hall et al., 2014). Including a scale correction, Cape Meares SF6 measurements are higher than Cape Grim and Neumayer during this period by 0.2–0.4 ppt. Much or all of this difference can be explained by an interhemispheric north–south difference of 0.3–0.4 ppt (Levin et al., 2010).

The mean secular trend between 1978 and 1996 for N2O and SF6 is 0.78±0.03 ppb yr-1 (95 % CI) and 0.17±0.01 ppt yr-1 (95 % CI), respectively, and is determined by applying a linear fit to deseasonalized data over this time period. These trends translate to annual increases of ∼0.25 % and ∼0.1 % for N2O and SF6, respectively. Annual trends for N2O and SF6 at Cape Meares, Oregon, are determined from the derivative of the deseasonalized localized regression (Fig. 5c, d). Uncertainty bands are generated from regressions of bootstrapped variability. Data points represent the mean annual trend with error bars equal to ±1σ of the trend over the year.

The mean annual trend in N2O (Fig. 5c) ranges between 0.6 and 1.2 ppb yr-1. All years between 1980 and 1996 show a positive rate of change significant at the 95 % confidence level. The uncertainty in the annual trend is smallest in the early 1980s, at ±0.15 ppb yr-1 (95 % CI), where the largest amount of data is available (∼50 % of samples are between 1980 and 1985). After 1985, uncertainty in the annual trend becomes ±0.5 ppb yr-1 (95 % CI). This relatively large uncertainty results in an annual growth rate that is statistically indistinguishable between years.

A previously reported secular trend of N2O found between 1978 and 1998 for Cape Meares is 0.74±0.02 ppb yr-1, which is indistinguishable from our result (Prinn et al., 2000). The global secular trend of N2O for the period from 1985 to 1996 reported by Khalil et al. (2002) is 0.69±0.03 ppb yr-1, which is also compatible with our trend at Cape Meares.

The SF6 annual trend (Fig. 5d) from the Cape Meares analysis increases from 0.07±0.03 ppt yr-1 (95 % CI) in 1980 to 0.26±0.05 ppt yr-1 (95 % CI) in 1994. The average rate of change in the growth rate (second derivative of mole fraction vs. time) over this period is 0.014 ppt yr-2. The increase in the growth rate over this period is statistically significant at high levels of confidence (95 %). After 1994, we measure a decrease in the growth rate, although this decline is not statistically significant at high levels of confidence over this short time interval.

Comparable trends in SF6 measured at other locations are available for the mid-1990s. The average global growth rate of SF6 in 1994 was reported at 0.23 ppt yr-1 in the Northern Hemisphere (Maiss et al., 1996). Alert, Canada, and Izaña, Tenerife, are observed to have maximum trends of 0.26 ppt yr-1 in mid-1994 and at the beginning of 1995 (Levin et al., 2010), respectively, which are compatible with results presented here. This localized maximum in the growth rate is also present in some Southern Hemisphere observations of SF6 at a similar time; Neumayer, Antarctica shows a maximum trend of 0.25 ppt yr-1 in 1995–1996 (Levin et al., 2010). This finding is consistent with a peak in SF6 emissions as reported by the European Database for Global Atmospheric Research (EDGAR, v4.2).

Another feature observed in the SF6 trend from Cape Meares is a local maximum in the growth rate near 1987 (Fig. 5d). Notably, however, not all datasets agree. The growth rate reported from Neumayer, Antarctica, has this feature during a similar period (Levin et al., 2010), but the trend reported at Cape Grim, Tasmania, does not show this local maximum (Rigby et al., 2010). Due to the large uncertainty from the few archived samples available during that time period, this local maximum is not statistically distinguishable from surrounding years at high levels of confidence in the Cape Meares analysis; thus, this result is merely suggestive. Additional evidence is needed to corroborate this finding.

Seasonal behavior for N2O and SF6 are shown in Fig. 6 and are determined from residuals to the secular trend. The N2O seasonal cycle at Cape Meares shows a maximum near April and May of 0.3 ppb and an extended minimum from September through December of -0.4 ppb. Although there is considerable uncertainty surrounding monthly means, the difference between the spring maximum and fall minimum is statistically robust at high levels of confidence (two-sample KS test, p value = 0.003).

The seasonal amplitude matches well with previously reported Northern Hemisphere magnitudes of ±0.4 ppb (Liao et al., 2004). Other midlatitude Northern Hemisphere sites also show a seasonal phase similar to that observed at Cape Meares. N2O seasonality reported at Mace Head, Ireland, has a maximum near April and a minimum near August and September (Nevison et al., 2004; Jiang et al., 2007), and Trinidad Head, CA, seasonality has a maximum near late May and a broad minimum from September to January (Nevison et al., 2007).

In general, the N2O seasonal amplitude is known to vary strongly with latitude, e.g., 0.29 ppb at the South Pole (90∘ S, 102∘ W) and 1.15 ppb at Alert, Canada (Jiang et al., 2007). This is attributed in part to the stronger branch of the Brewer–Dobson circulation in the Northern Hemisphere which also explains the high-latitude minimums in late-summer months related to the influx of N2O-depleted air from the stratosphere during the spring (Liao et al., 2004; Nevison et al., 2004). Aside from atmospheric circulation, N2O seasonality may also be influenced by regional sources. Lueker et al. (2003) suggested that local maximums at Trinidad Head may reflect the influence of strong coastal upwelling. Similarly located in the eastern Pacific, Cape Meares may also be subject to coastal upwelling influences. Isotopic analysis or modeling of transport effects and source influence would be useful to help interpret seasonal behavior of N2O at Cape Meares.

Seasonality for SF6 shows a maximum between December and February of 0.04 ppt and a minimum near July of -0.03 ppt. The difference between the winter maximum and summer minimum is statistically significant (two-sample KS test, p value = 0.004). SF6 seasonality has not previously been reported for Cape Meares.

Some seasonality in Northern Hemisphere observations of SF6 is reported in the literature at select locations. Barrow, AK, has a minimum in September and October with a broad maximum from December to June (Patra et al., 2009). Alert, Canada, shows a strong minimum in October, although a maximum is not clearly defined (Wilson et al., 2014). Continental sites such as Niwot Ridge show large interannual variability (IAV) but have little distinguishable seasonality (Patra et al., 2009).

SF6 seasonality at Cape Grim has been reported to have an amplitude of ±0.01 ppt with a maximum in September and October and a minimum near February (Nevison et al., 2007; Wilson et al., 2014). The seasonality phase of Cape Grim is nearly antiphase of the Cape Meares results reported here, although the amplitude is a factor of 4 smaller at Cape Grim. Similar to N2O, the seasonal amplitude is expected to be larger in the Northern Hemisphere than in the Southern Hemisphere (Nevison et al., 2007). Because sources of SF6 are aseasonal and sinks are essentially zero in the troposphere; the driving force behind the observed seasonality in SF6 is considered to be atmospheric transport (Patra et al., 2009). Processes such as convection, boundary layer mixing, stratosphere–troposphere exchange (STE), and shifts in the intertropical convergence zone can potentially influence the observed seasonality at a location. Seasonal transport from STE adds relatively depleted SF6 air into the troposphere from the stratosphere. The seasonal phase of SF6 observed at Cape Meares closely reflects seasonality phasing observed in CFCs in the Northern Hemisphere driven by STE (Liang et al., 2008). Modeling atmospheric transport effects on SF6 at Cape Meares could help confirm amplitude and phase reported here.