For the present study, archived and urban ambient air samples were analyzed at the Commonwealth Scientific and Industrial Research Organization (CSIRO) laboratory at Aspendale (Victoria, Australia) using Medusa gas chromatographic (GC) mass spectrometric (MS) techniques . The archived air samples consisted primarily of the Cape Grim Air Archive (CGAA) samples collected under clean air baseline conditions for archival purposes since 1978 at the Cape Grim Baseline Air Pollution Station (Tasmania, Australia; 40.7∘ S, 144.7∘ E). These >100 samples were collected into 34 L internally electropolished stainless steel canisters (Essex Industries, USA) using cryogenic techniques . The CGAA record was complemented with a few samples collected in the Northern Hemisphere (33–53∘ N) mostly using oil-free diving compressors.

Two firn air samples were also analyzed, which were collected at the Aurora Basin North site in Antarctica (71.1∘ S, 111.4∘ E). The site is located 550 km inland from Australia's Casey Station, at 2710 m a.s.l., and has a low mean annual air temperature of -44∘C. Samples were collected in December 2013; those for the halocarbon measurements were collected into internally electropolished stainless steel containers using a two-stage Teflon-coated Viton diaphragm pump. Only two samples were available for the present study as other samples from this site were used for a different halocarbon study.

In situ measurements of c-C4F8O at Aspendale (38.0∘ S, 145.1∘ E) were started in February 2017. These samples are collected from the rooftop at CSIRO (at 11 m height from the ground) through a 3/8 in. OD Synflex 1300 tube (Saint-Gobain, France) using a continuous-flow air sampling module with a diaphragm sampling pump fitted with stainless steel heads and a neoprene membrane (KNF Neuberger, Germany).

All archived air samples were analyzed on the Medusa-GCMS “Medusa-9” in December 2016 along with a suite of other trace gases. The instrument is based on the original design of the Medusa-GCMS used in the Advanced Global Atmospheric Gases Experiment (AGAGE) network but fitted with different chromatography columns . A GS-GasPro main capillary column (0.32 mm ID × 60 m, Agilent Technologies) was used for the main separation and a column of the same type (5 m) was fitted as a precolumn, allowing for a backflushing of late-eluting compounds. In this GCMS setup (Agilent 6890 GC, 5975 MS), c-C4F8O was identified using a multicomponent diluted mixture of known composition with the MS in scan and selected ion modes. The choice for the two fragments used in the analysis of our air samples was based on the mass spectrum, which we measured for c-C4F8O, to the best of our knowledge the first one published for this compound (see the Supplement).

Analytes from the samples were cryogenically preconcentrated on a first microtrap of the GCMS and subsequently transferred to a second microtrap, both filled with HayeSep D and held at ∼-155 ∘C. During this process, water vapor was largely removed using Nafion dryers; nitrogen, oxygen, and a large fraction of noble gases were removed due to their trap breakthroughs, and carbon dioxide was removed using a molecular sieve (4Å) packed column between the traps. To enhance the signal size of the measured compounds, 3 L sample sizes were used for each measurement (compared to normally 2 L) and the MS electron multiplier voltage was increased by 50 V compared to what was given by the autotune algorithm. Analysis of a single sample lasted 65 min. Archived air sample measurements were bracketed by measurements of a standard (E-146S) to track and correct for MS sensitivity changes. This standard was air compressed into a 34 L tank at the remote Rigi-Seebodenalp station (Switzerland) using an oil-free compressor, and was additionally spiked with small amounts of c-C4F8O and other compounds to enhance the GCMS peak size and signal-to-noise ratio. In general, three measurements of each archived air sample were made. For some, no standard measurement was made between the second and third sample to assess potential memory effects of the system. For c-C4F8O, no memory effect and no signal in the blank runs could be detected. Detection limits are estimated at 5 ppq (parts per quadrillion, femtomole per mole). Mean precisions (2σ) for the measurements of the archived air samples ranged 3–4 ppq (20 %–5 %) for the low (∼15 ppq) to high (∼70 ppq) mole fractions, respectively. Based on two different types of experiments, a linear system response for the relevant mole fraction range was found (see the Supplement). In situ urban air measurements at Aspendale are based on 2 L samples and without alteration of the MS electron multiplier voltage. Consequently the precisions are slightly poorer for these measurements. These air precisions were estimated at ∼12 ppq (∼17 %, 2σ) under the assumption that c-C4F8O remains constant in the air measured in situ at Aspendale on a daily basis.

A primary calibration scale was prepared based on a commercially obtained multicomponent mixture in dry synthetic air (Carbagas, Switzerland, HCP-04Carba), with a mole fraction of c-C4F8O at 10 ppm (parts per million, micromole per mole). This mixture was diluted manometrically and using a bootstrap technique, resulting in a primary calibration standard (EP-001) with c-C4F8O at 1.81 ppt (parts per trillion, picomole per mole). Three secondary standards were additionally prepared from ambient air compressed into cylinders (Essex Industries, USA) and spiked with small quantities of c-C4F8O resulting in mole fractions of ∼0.5 ppt. These secondary standards were the base for propagating the calibration scale to other calibration standards, in particular that used for the Cape Grim Air Archive measurements (E-146S). They define the Empa-2013 calibration scale for c-C4F8O  on which our results are reported. The systematic uncertainty of the preparation of this primary calibration scale (including its propagation to the working standards), which defines its accuracy, is estimated at 15 % (2σ). Details of the dilution technique and the primary calibration scale are provided by .

We use a numerical firn air model to quantify the movement of c-C4F8O in firn air in order to determine the time period for which c-C4F8O in the firn samples is representative of the atmosphere. Vertical diffusion in the firn and other physical processes cause a tracer in a firn air sample to correspond to an age spectrum relative to the atmosphere, rather than a discrete age. Green's functions from the firn model represent the age spectrum of a tracer in each firn sample and are used in this work to relate the measured mole fractions of c-C4F8O in firn to the time range of the corresponding atmospheric mole fractions.

For Aurora Basin North, the firn model uses an accumulation rate of 97 kg m-2 yr-1, a temperature of -44 ∘C, and pressure of 695 hPa. The density profile used was based on a spline fit to density measurements. Diffusion parameters in the firn model are calibrated for Aurora Basin North using 12 tracers at between 5 and 11 depths each throughout the firn. The diffusion coefficient used in this work for c-C4F8O in air relative to CO2 in air (for a temperature of 253 K) is 0.460. This value was determined using Eq. (4) from with Le Bas volume increments (e.g., Table 1.3.1, ) and a multiplier for the Le Bas increments of 0.97 (this value minimizes the difference of calculated relative diffusion coefficients of a number of compounds from values measured by ). Further details on diffusivity in the firn model calculations for Aurora Basin North are given in the Supplement.

We use the AGAGE 12-box atmospheric model to relate the atmospheric mole fractions to surface emissions. Briefly, in this model, the atmosphere is divided into four zonal bands, separated at the Equator and at the 30∘ latitudes, thereby creating boxes of similar air masses. There are also vertical separations, at altitudes represented by 500 and 200 hPa, resulting in the overall 12 boxes. Model transport parameters and stratospheric photolytic loss vary seasonally and repeat interannually . We anticipate that variations in emissions dominate atmospheric trends, particularly over the longer (multi-annual) timescales that are our primary focus, so interannual variation in transport is not expected to be important here. Loss processes other than those in the atmosphere, such as uptake by land and oceans – and potential natural sources – are not included in the model. Green's functions derived from the 12-box atmospheric model relate atmospheric mole fraction in the high-latitude Northern Hemisphere and Southern Hemisphere to annual global emissions in preceding years, and are used in the inversion .

To estimate global emissions to the atmosphere from the mole fraction measurements, we employ an inverse calculation (inversion InvE2 from , and termed “CSIRO” inversion in ) that was developed to focus on sparse observations from air archives, and firn air and ice core samples that are associated with age spectra. The inversion combines Green's functions from both the firn model and the AGAGE 12-box atmospheric model described above to relate firn and tropospheric mole fraction to c-C4F8O surface emissions. The Green's functions from the 12-box model were calculated using a constant distribution of emissions into the four zonal boxes at the surface, and for this we used the relative contributions 0.675, 0.325, 0.0, and 0.0 in the northernmost to southernmost zonal bands. Results are fairly insensitive to emissions distributions that have most emissions in the Northern Hemisphere (see the Supplement). The characteristics of sparse atmospheric, firn, and ice core data necessitate the use of constraints on the inversion to avoid unrealistic oscillations in the reconstructed mole fractions or negative values of mole fraction or emissions. The inversion uses non-negativity constraints and favors relatively small changes in annual emissions between adjacent years over large, unrealistic fluctuations. A prior emissions history is needed as a starting point for the inversion. A nonlinear constrained optimization method (constrained_min routine in IDL (Harris Geospatial Solutions Inc., Broomfield, Colorado) is used to find the solution that minimizes a cost function consisting of the model–data mismatch weighted by the observation uncertainties, plus the sum of the year-to-year changes in emissions . Given the lack of industry-based bottom-up emission estimates for c-C4F8O, we use emissions derived from observations of perfluorooctane, which was found present for many decades and at low abundances in the global atmosphere . Because the prior is not based on information on c-C4F8O, we do not include the prior in the cost function. The emissions derived from the inversion are rather insensitive to the choice of the prior (see the Supplement) because the prior is used here as a starting point for the inversion only, and not as a constraint. Our observations used in the inversion are the firn measurements and annual values of mole fraction from a smoothing spline fit (50 % attenuation at 10 years) to measurements of the CGAA and in situ measurements at Aspendale. Northern Hemisphere measurements were compared with the reconstructed mole fractions for that hemisphere, but were not used in the inversion. Uncertainties in the emissions are estimated using a bootstrap method that incorporates temporally correlated uncertainties in the annual values derived from the atmospheric data (see the Supplement), uncertainty in firn measurements, uncertainty of ±15 % in the calibration scale, and uncertainties in the firn model parameters through the use of an ensemble of firn Green's functions.

Absorption spectra, A(v) (base e), or integrated band strengths, were quantified using Beer's law A(λ)=-ln⁡I(λ)I0(λ)=σ(λ)×L×c-C4F8O, where A is the measured absorbance at wavelength λ; I(λ) and I0(λ) are the measured light intensities with and without the sample present in the absorption cell, respectively; L is the optical absorption path length; σ(λ) is the infrared or UV cross section of c-C4F8O; and [c-C4F8O] is the concentration of c-C4F8O. In total, 11 independent absorption spectrum measurements were used in the linear least-squares fit. The c-C4F8O concentration was determined using the ideal gas law and absolute pressure measurements of either the pure compound or of a dilute mixture of the compound in a helium (He) bath gas.

The c-C4F8O sample was obtained from SynQuest Laboratories Inc. (Alachua, Florida, USA, 99 % purity). For the experiments described below, c-C4F8O was introduced into the absorption cells as a pure sample or in a dilute mixture prepared off-line. The dilute mixtures of c-C4F8O in a He (UHP, 99.999 %) bath gas were prepared manometrically in a 12 L Pyrex bulb with an estimated accuracy of ∼1 %, as derived from the accuracy of the pressure measurements. The pressure measurement uncertainty includes the uncertainty in the sample pressure, the total mixture pressure, and the linearity of the pressure gauge (estimated to be 0.2 %). Pressures were measured with 100 and 1000 Torr (130 and 1300 hPa, respectively) capacitance manometers. Quoted uncertainties are 2σ.

Infrared absorption spectra were measured at 296 K using Fourier transform infrared (FTIR) spectroscopy over the 500–4000 cm-1 spectral region at 1 cm-1 resolution with boxcar apodization. The apparatus has been used extensively in previous studies . The FTIR was coupled to a 15 cm path length single-pass absorption cell with potassium bromide (KBr) windows. A liquid-nitrogen-cooled HgCdTe/B semiconductor detector was used. Infrared spectra were recorded in 100 or 500 co-added scans. Absorption spectra were recorded under static conditions using a dilute mixture of c-C4F8O in He with a 0.00180 mixing ratio. The c-C4F8O concentration used in the absorption measurements was in the range 1.75 × 1015 to 2.34 × 1016 molecule cm-3. Integrated band strengths (IBSs) were obtained from the measurement of 11 individual IR spectra.

The UV absorption spectrum of c-C4F8O was measured at 296 K using a 0.5 m spectrometer (1 nm resolution) equipped with a charge-coupled device (CCD) detector. The collimated output of a 30 W deuterium lamp passed through a 100 cm long and 2.5 cm diameter Pyrex absorption cell with quartz windows. Spectral measurements were made over the wavelength region 200–350 nm. The wavelength scale of the spectrometer was calibrated using the emission lines from a low-pressure Hg Pen-Ray lamp. c-C4F8O was added to the absorption cell in pure form from the original sample. Measurements were performed over a range of c-C4F8O concentrations from 2.51 × 1018 to 2.16 × 1019 molecule cm-3. Eleven independent UV absorption spectrum measurements were used in the final linear least-squares fit.

The reactive rate coefficient, k1, for the reaction O(1D)+c-C4F8O→Products,O(1D)+c-C4F8O→c-C4F8O+O(3P), i.e., the channel resulting in the loss of c-C4F8O, was measured at 294 K using a relative method e.g.,. The loss of c-C4F8O was measured relative to the loss of the reference compound CHF3 during the same experiment: O(1D)+CHF3→Products,O(1D)+CHF3→CHF3+O(3P).

The recommended total rate coefficient for Reaction (R2), k2, is (9.60±0.5)×10-12 cm3 molecule-1 s-1 and the recommended reactive channel branching ratio, k2a/k2, is 0.25, i.e., k2a=2.4×10-12 cm3 molecule-1 s-1 .

Provided c-C4F8O and the reference compound are removed solely by reaction with O(1D), the rate coefficient for Reaction (R1a) is related to the reference compound rate coefficient by the equation ln⁡[c-C4F8O]t0[c-C4F8O]t=k1ak2aln⁡[CHF3]t0[CHF3]t, where [c-C4F8O]t0, [CHF3]t0, [c-C4F8O]t, and [CHF3]t are the concentrations of c-C4F8O and CHF3 at times zero (t0) and t, respectively. The k1a and k2a are the reactive rate coefficients for the reaction of O(1D) with c-C4F8O Reaction (R1a) and CHF3 Reaction (R2a), respectively.

The Pyrex reactor, which was 100 cm long and with a 2.2 cm internal diameter, was coupled with a Teflon circulating pump to an absorption cell where the losses of c-C4F8O and CHF3 were measured using FTIR spectroscopy. The FTIR absorption cell was equipped with a multi-pass cell (485 cm path length) with KBr windows. Spectra were recorded in 100 co-adds at a spectral resolution of 1 cm-1.

O(1D) was produced by KrF (248 nm) excimer pulsed laser photolysis of ozone: O3+hν(248nm)→O(1D)+O2(1Δ),→O(3P)+O2(3Σ).

The yield of the O(1D) channel is 0.9 . After thoroughly mixing the gas mixture in the system, a time zero infrared spectrum was recorded. Ozone was then slowly added to the reactor with the photolysis laser and gas circulation on. The photolysis laser fluence was in the range ∼2–7.4 mJ cm-2 pulse-1. The laser was operated at 10 or 20 Hz. The total pressure in the cell increased during an experiment by ∼300 Torr, mostly due to the addition of He carrier gas used to flush ozone into the reactor. Infrared spectra were recorded at regular intervals with approximately 10 spectra recorded over the course of an experiment. Experiments performed separately demonstrated that there was no significant loss of c-C4F8O or CHF3 under identical conditions in the absence of photolysis. The initial c-C4F8O and CHF3 concentrations were in the range 6.4–6.8 × 1014 molecule cm-3 and 4.5–5.0 × 1014 molecule cm-3, respectively.