Introduction
Chlorofluorocarbons (CFCs) were first developed in the 1930s as safe,
reliable, and non-toxic refrigerants for domestic use.
Trichlorofluoromethane, known as CFC-11 or Freon-11, and
dichlorodifluoromethane, known as CFC-12 or Freon-12, were the two most
widely used CFCs in applications ranging from refrigerators and air
conditioners to propellants in spray cans and blowing agents in foam
production.
Ultimately, however, CFCs proved too good to be true. The explosion in their
use led to a steady increase in their atmospheric abundances. While they are inert in
the troposphere, it was this stability which enabled them to reach the
stratosphere, where dissociation by ultraviolet radiation released chlorine
atoms, which catalyse the destruction of stratospheric ozone (Solomon,
1999). The realisation of this impending environmental disaster prompted
international action, and in 1987 the Montreal Protocol was ratified; this
led to the phasing out of the worldwide production and use of CFCs. CFCs are
still released into the atmosphere from “banks”, such as old
refrigerators; however these are not regulated by the protocol (Harris et
al., 2014). Banks are the major source of emissions for many ozone-depleting substances (ODSs), including
CFC-11, which has a long atmospheric lifetime of 52 years (Harris et al.,
2014).
Summary of limb sounders past and present capable of measuring
CFC-11.
Instrument
Platform
Years
ATMOS (Atmospheric Trace MOlecule Spectroscopy)
Space shuttle
1985, 1992, 1993, 1994
Chang et al. (1996), Irion et al. (2002)
CIRRIS 1A (Cryogenic InfraRed Radiance Instrumentation for Shuttle)
Space shuttle
1991
Bingham et al. (1997)
CRISTA (CRyogenic Infrared Spectrometers and Telescopes for the Atmosphere)
Space shuttle
1994, 1997
Offermann et al. (1999)
CLAES (Cryogenic Limb Array Etalon Spectrometer)
UARS (Upper Atmosphere Research Satellite)
1991–1993
Roche et al. (1993)
ILAS (Improved Limb Atmospheric Spectrometer)
ADEOS (ADvanced Earth Observing Satellite)
1996–1997
Yokota et al. (2002)
ILAS II
ADEOS II
2003
Wetzel et al. (2006)
HIRDLS (HIgh Resolution Dynamics Limb Sounder)
Aura
2004–2008
Hoffmann et al. (2014)
MIPAS (Michelson Interferometer for Passive Atmospheric Sounding)
ENVISAT (ENVIronmental SATellite)
2002–2012
e.g. Hoffmann et al. (2005), Dinelli et al. (2010), Kellmann et al. (2012)
ACE-FTS (Atmospheric Chemistry Experiment – Fourier transform spectrometer)
SCISAT (Scientific Satellite)
2004–
Brown et al. (2011)
At present, CFC-11 is the second-most-abundant CFC in the atmosphere and
contributes the second-highest amount of chlorine to the stratosphere,
behind CFC-12. In addition to its role in stratospheric ozone destruction –
it has the highest ozone depletion potential (1.0) (Harris et al., 2014) of
all the CFCs – CFC-11 is a particularly strong greenhouse gas: it has a
100-year global warming potential of 5160 (Harris et al., 2014).
As a key species in stratospheric ozone destruction, CFC-11 atmospheric
concentrations are monitored in situ at the surface; e.g. the annual global
mean mole fraction of CFC-11 measured by the AGAGE (Advanced Global
Atmospheric Gases Experiment) network in 2012 was 235.5 ppt (Carpenter et
al., 2014). However, in order to measure concentrations in the stratosphere
where ozone destruction occurs, remote-sensing techniques are required.
Table 1 contains a listing of limb sounders capable of measuring CFC-11, as
described in the literature.
The infrared (IR) spectra for large molecules like trichlorofluoromethane
are highly complex, consisting of very many closely spaced spectroscopic
lines, making the task of generating line parameters from measurements an
almost impossible one. For the purposes of atmospheric remote sensing, it is
possible to use absorption cross sections in forward models instead of line
parameters; however this requires laboratory measurements of air-broadened
spectra over a range of temperatures and pressures. The accuracy of
retrievals of CFC-11 abundances for the limb sounders in Table 1 is very
much dependent on the quality of the underlying spectroscopy; ideally
absorption cross-section datasets should cover a range of atmospherically
relevant pressure–temperature (PT) combinations, with accurate wavenumber
scales and band intensities, and properly resolved spectral features. This
work presents new spectroscopic data, optimised for limb sounding
instruments, which improve upon those currently available in the HITRAN
(HIgh-resolution TRANsmission; Gordon et al., 2017) and GEISA (Gestion et Étude des Informations
Spectroscopiques Atmosphériques; Jacquinet-Husson et al., 2016) databases.
Infrared spectroscopy of trichlorofluoromethane
Spectroscopic background
There are two stable isotopes of carbon and chlorine, and one of fluorine,
resulting in eight stable isotopologues of trichlorofluoromethane, namely
12/13C35Cl3F, 12/13C35Cl237ClF,
12/13C35Cl37Cl2F, and 12/13C37Cl3F; these
belong to the point groups C3v, Cs, Cs and
C3v, respectively. Taking into account the natural abundances of
12C and 13C (∼99 % and ∼1 %, respectively), and
35Cl and 37Cl (∼76 % and ∼24 %, respectively), the most
abundant isotopologues are therefore 12C35Cl3F,
12C35Cl237ClF, and 12C35Cl37Cl2F, with
abundances of 43 %, 41 %, and 13 %, respectively.
The absorption cross section of trichlorofluoromethane/dry synthetic
air at 191.7 K and 7.535 Torr (this work), with vibrational band
assignments for the main band systems in the 710–1290 cm-1 spectral
region.
As a non-linear molecule with five atoms, trichlorofluoromethane possesses
nine normal vibrational modes; in the C3v point group there are
three non-degenerate fundamentals of A1 symmetry (ν1, ν2,
and ν3) and three doubly degenerate fundamentals of E symmetry
(ν4, ν5, and ν6). For the Cs point group, the
ν1, ν2, and ν3 modes possess A′ symmetry, with the
doubly degenerate ν4, ν5, and ν6 modes each splitting
into one A′ and one A′′ mode (Snels et al., 2001). Since the splittings
in the ν4, ν5, and ν6 levels are small, it is normal to
label these bands assuming C3v symmetry. The 710–1290 cm-1
spectral range covered in the present work contains two strong fundamental
bands, ν1∼1081.28 cm-1 and ν4∼849.5 cm-1,
and a weaker combination band, ν2+ν5∼936.5 cm-1;
reported frequencies are those for the most abundant isotopologue,
12C35Cl3F (von Lilienfeld et al., 2007; Snels et al., 2001).
Isotopologues complicate the already dense CCl3F rotation–vibration
spectrum; each has slightly different molecular parameters, with bands
shifted by small amounts relative to each other. These main band systems are
shown in Fig. 1 in the plot of the new absorption cross section at 191.7 K
and 7.535 Torr. Details on the measurement conditions and derivation of this
cross section are given in Sect. 3.
A brief history of trichlorofluoromethane absorption cross
sections
High resolution (0.03 cm-1) absorption cross sections of pure
trichlorofluoromethane at 296 K were first included in HITRAN as part of the
1986 compilation (Massie et al., 1985; Rothman et al., 1987). The HITRAN
1991/1992 compilation saw the first introduction of temperature-dependent
cross sections (203–293 K) for CFC-11 (McDaniel et al., 1991; Rothman et
al., 1992; Massie and Goldman, 1992); as before these were derived from
measurements of pure CCl3F at 0.03 cm-1 resolution.
While the two previous HITRAN editions (1986 and 1991/1992) neglected
pressure-broadening effects on the CCl3F spectra, cross sections for 33
distinct PT combinations (201–296 K and 40–760 Torr N2-broadened)
over two wavenumber ranges, 810–880 and 1050–1120 cm-1,
were introduced into HITRAN 1996 (Li and Varanasi, 1994; Rothman et al.,
1998). Another 22 PT combinations covering lower pressures and temperatures
over the same wavenumber ranges were added to HITRAN 2000 (provided by
Varanasi, cited within Rothman et al., 2003), bringing the overall PT
coverage to 190–296 K and 8–760 Torr. Out of these 55 PT combinations,
four pairs possess both temperature and pressure within 1 K and 5 Torr,
respectively. This dataset, henceforth referred to as the Varanasi dataset,
has been used widely for remote-sensing applications since it was first
introduced; it is still the dataset included in the most recent GEISA 2015
(Jacquinet-Husson et al., 2016) and HITRAN 2016 (Gordon et al., 2017)
spectroscopic databases. Despite its widespread use, the Varanasi dataset
has some deficiencies, which will be discussed in Sect. 4, alongside a
comparison with the new spectroscopic data taken as part of the present
work.
FTS parameters, sample conditions, and cell configuration for all
measurements.
Spectrometer
Bruker Optics IFS 125HR
Mid-IR source
Globar
Detector
Mercury–cadmium–telluride (MCT) D313∗
Beam splitter
Potassium bromide (KBr)
Optical filter
∼700–1400 cm-1 bandpass
Spectral resolution
0.01 to 0.03 cm-1
Aperture size
3.15 mm
Apodisation function
Boxcar
Phase correction
Mertz
CCl3F (Supelco)
99.9 % purity, natural-abundance isotopic mixture; freeze–pump–thaw-purified multiple times prior to use
Air, zero grade (BOC Gas)
Total hydrocarbons <3 ppm, H2O<2 ppm, CO2<1 ppm, CO<1 ppm; used “as is”
Cell pathlength
26 cm
Cell windows
Potassium bromide (KBr) (wedged)
Pressure gauges
3 MKS-690A Baratrons (1, 10 & 1000 Torr) (±0.05 % accuracy)
Refrigeration
Julabo F95-SL Ultra-Low Refrigerated Circulator (with ethanol)
Thermometry
4 platinum resistance thermometers (PRTs), Labfacility IEC 751 Class A
Wavenumber calibration
N2O
∗ Due to the non-linear response of MCT detectors to the
detected radiation, all interferograms were Fourier-transformed using
Bruker's OPUS software with a non-linearity correction applied.
New absorption cross sections of air-broadened
trichlorofluoromethane
Experimental
The experimental setup at the Molecular Spectroscopy Facility (MSF),
Rutherford Appleton Laboratory (RAL), and the experimental procedures have
been described previously for related measurements (e.g. Harrison et al.,
2010; Harrison, 2015b, 2016); the reader is referred to one of
these previous studies for more information. Instrumental parameters
associated with the Fourier transform spectrometer (FTS) used for the
measurements, sample details, and the cell configuration are summarised in
Table 2. The sample pressures and temperatures for each air-broadened
spectrum, along with their experimental uncertainties and associated
spectral resolutions, are listed in Table 3.
Summary of the sample conditions for all measurements.
Temperature
Initial CCl3F
Total pressure
Spectral resolution
(K)
pressure (Torr)a
(Torr)
(cm-1)b
191.7±0.8
0.266
7.535±0.035
0.0100
191.5±0.8
0.302
49.83±0.13
0.0150
191.6±0.8
0.302
98.14±0.68
0.0225
191.6±0.8
0.266
200.0±0.3
0.0300
202.3±0.5
0.319
7.508±0.006
0.0100
202.4±0.5
0.309
50.28±0.13
0.0150
202.3±0.5
0.318
99.85±0.30
0.0150
202.4±0.5
0.309
200.4±0.2
0.0225
202.3±0.5
0.318
301.6±0.3
0.0300
216.7±0.5
0.347
7.496±0.018
0.0100
216.7±0.5
0.358
49.93±0.09
0.0150
216.7±0.5
0.357
99.94±0.07
0.0150
216.6±0.5
0.375
201.0±0.2
0.0225
216.7±0.5
0.383
360.4±0.3
0.0300
232.6±0.4
0.407
7.500±0.020
0.0100
232.6±0.4
0.395
49.80±0.15
0.0150
232.6±0.4
0.544
99.67±0.16
0.0150
232.6±0.4
0.417
201.0±0.1
0.0225
232.6±0.4
0.413
399.8±0.3
0.0300
252.5±0.2
0.503
7.477±0.003
0.0100
252.5±0.2
0.486
50.06±0.05
0.0150
252.5±0.2
0.516
200.9±0.1
0.0225
252.5±0.2
0.544
399.9±0.2
0.0300
252.5±0.2
0.607
600.2±0.3
0.0300
273.9±0.2
0.475
7.501±0.001
0.0100
273.8±0.2
0.613
201.6±0.1
0.0225
273.8±0.2
0.598
355.8±0.1
0.0300
273.8±0.2
0.607
760.1±0.2
0.0300
293.1±0.1
0.548
355.8±0.1
0.0300
293.0±0.1
0.566
760.0±0.1
0.0300
a MKS-690A Baratron readings are accurate to ±0.05 %. b Using the Brukerdefinition of 0.9/MOPD.
Generation of absorption cross sections
The procedure used to generate absorption cross sections from measured
spectra has been reported previously (e.g. Harrison et al., 2010; Harrison,
2015b, 2016), so the full details are not provided here. The
wavenumber scale of the cross sections is calibrated against the positions
of isolated N2O absorption lines taken from the HITRAN 2012 database
(Rothman et al., 2013). The absorption cross sections, σ(υ,Pair,T), in units of square centimetres per molecule and at wavenumber
υ (cm-1), temperature T (K), and synthetic air pressure
Pair are normalised according to
∫710cm-11290cm-1συ,Pair,T∂υ=9.9515×10-17cmmolecule-1,
where the value on the right-hand side is the average integrated band
intensity over the spectral range 710–1290 cm-1 for three
760 Torr N2-broadened trichlorofluoromethane spectra (at 278, 298, and
323 K) from the Pacific Northwest National Laboratory (PNNL) IR database
(Sharpe et al., 2004). This intensity calibration procedure counters
problems with trichlorofluoromethane adsorption in the vacuum line and on
the cell walls, and furthermore assumes that the integrated intensity over
each band system is independent of temperature. The reader is referred to
Harrison et al. (2010) for a more complete explanation of the underlying
assumption and references cited within Harrison (2015a, b, 2016)
for details on previous successful uses of this approach.
The new absorption cross sections of trichlorofluoromethane/dry
synthetic air at a total pressure of ∼200.0 Torr over a range of
temperatures (191.6, 202.4, 216.6, 232.6, 252.5, and 273.8 K). The observed
narrowing of the ν4 band as the temperature decreases is due to the
decline in Boltzmann populations of the upper rovibrational levels of the
ground state.
A selection of the derived absorption cross sections is presented in Fig. 2, showing the expected behaviour with temperature at a total pressure of
∼200 Torr; the wavenumber range covers the microwindow for
the ACE-FTS v3.6 retrieval scheme.
Absorption cross-section uncertainties
The accuracy of the wavenumber scale for the new absorption cross sections
is comparable to the accuracy of the N2O lines used in the calibration;
according to the HITRAN error codes, this is between 0.001 and 0.0001 cm-1. The uncertainty in the intensity is dominated by systematic
errors. A true measure of the random errors as a function of wavenumber would
ideally require multiple concentration-pathlength burdens at each PT
combination, but only one is available for each; however, as indicated
in Sect. 4.4, these are small and make minimal contribution to the overall
error budget. The maximum systematic uncertainties in the sample
temperatures (μT) and total pressures (μP) are 0.4 %
and 0.7 %, respectively (see Table 3). The photometric uncertainty (μphot), associated with the detection of radiation by the MCT detector
and systematic error arising from the use of Bruker's non-linearity
correction for mercury–cadmium–telluride (MCT) detectors, is estimated to be ∼2 %. The
pathlength error (μpath) is estimated to be negligibly small,
lower than 0.1 %. According to the PNNL metadata, the systematic error in
the PNNL CCl3F spectra used for the intensity calibration is estimated
to be less than 3 % (2σ). Equating the error, μPNNL,
with the 1σ value, i.e. 1.5 %, and assuming that the systematic
errors for all the quantities are uncorrelated, the overall systematic error
in the dataset can be given by
μsystematic2=μPNNL2+μT2+μP2+μphot2.
Note that using PNNL spectra for intensity calibration effectively nullifies
the errors in the trichlorofluoromethane partial pressures and cell
pathlength, so these do not have to be included in Eq. (2). According to Eq. (2), the systematic error contribution, μsystematic, to the new
cross sections is ∼3 % (1σ).
Comparison between absorption cross-section datasets
In this section the new dataset presented in this work is compared with the
older Varanasi dataset, which has a stated uncertainty of 2 % (Li and
Varanasi, 1994). The comparison focuses on their wavenumber scales,
integrated band strengths, artefacts such as channel fringing,
signal-to-noise ratios (SNRs), spectral resolution, and PT coverage. Given the
various problems identified in sections below, the 2 % uncertainty is a
significant underestimate. Overall, these new data will provide a more
accurate basis for retrieving CFC-11 from atmospheric spectra recorded in
the limb. In addition, the new dataset includes the weak combination band,
ν2+ν5 (not present in the Varanasi measurements),
which will help improve the retrievals of other species, in particular
SF6.
Wavenumber scale
It is likely that the wavenumber scale for the Varanasi dataset was never
calibrated; this has been observed in a number of recent studies for other
halogenated species in which new datasets have been compared with older
Varanasi datasets, e.g. HFC-134a (Harrison, 2015a), CFC-12 (Harrison, 2015b),
and HCFC-22 (Harrison, 2016). As explained earlier, the absolute accuracy of
the wavenumber scale for the new dataset lies between 0.001 and
0.0001 cm-1. In comparison, the ν4 band in the Varanasi cross
sections is shifted too low in wavenumber; this shift varies between cross
sections, e.g. by ∼0.002 cm-1 (a correction factor of ∼1.000002) for the 190 K, 7.5 Torr ν1 Varanasi measurement and by
∼0.007 cm-1 (a correction factor of ∼1.000007) for
the 216.1 K, 100.0 Torr ν1 measurement.
Integrated band strengths
Integrated band strengths for the Varanasi cross sections have been
calculated over the spectral ranges of the cross-section files, 810–880
and 1050–1120 cm-1, covering the ν4 and ν1 bands
respectively, and compared with those for the new absorption cross sections
calculated over the same ranges; plots of integrated band strength against
temperature for each dataset, including the PNNL spectra, and wavenumber
range can be found in Fig. 3. At each temperature the Varanasi integrated
band strengths display a small spread in values, most notably for the ν4 band; however there is no evidence for any temperature dependence,
backing up the assumption in Sect. 3.2 that the integrated intensity over
each band system is independent of temperature. The small spread in values
is likely due to inconsistencies in the baselines for the Varanasi cross
sections, which are larger for the ν4 band. Additionally,
according to the PNNL spectra and the new measurements, the ν4
cross section at 810 cm-1 is non-zero due to the presence of a weak hot
band. Therefore, calculating integrated band strengths for the new dataset
over the 810–880 cm-1 range creates a very small temperature
dependence in the ν4 integrated band strengths. Unfortunately, the
wavenumber ranges do not extend far enough to obtain an unambiguous measure
of the baseline position for the Varanasi data, and the cross sections in
the HITRAN and GEISA databases have had all negative cross-section values
set to zero, which has the effect of adjusting the baseline positions by a
small amount near the band wings.
Integrated band strength as a function of temperature for the new,
Varanasi, and PNNL cross-section datasets over the wavenumber ranges 810–880
and 1050–1120 cm-1.
The Varanasi absorption cross section of trichlorofluoromethane/dry
synthetic air at 232.7 K and 250.0 Torr (black), with the new cross section
at 232.6 K and 201.0 Torr overlaid (red; this work). Channel fringes in the
Varanasi cross section are clearly visible.
Channel fringes
Most of the absorption cross sections in the Varanasi CFC-11 dataset contain
noticeable channel fringes above the noise level (refer to Fig. 4 for an
example of this); in transmittance these would equate to peak-to-peak
amplitudes as high as ∼2–3 %. For the measurements described in
the present work, wedged cell windows were used to avoid channel fringes by
preventing reflections from components in the optical path of the
spectrometer.
The Varanasi absorption cross section of trichlorofluoromethane/dry
synthetic air at 216.5 K and 7.50 Torr (black), with the new cross section
at 216.7 K and 7.50 Torr overlaid (red; this work). Additional noise in the
Varanasi cross section is clearly visible.
Signal-to-noise ratios (SNRs)
The SNRs of the transmittance spectra measured in the present work have been
calculated using Bruker's OPUS software at ∼990 cm-1,
where the transmittance is close to 1; the values range from 2600 to 4700
(rms), equivalent to percentage transmittances between 0.04 and 0.02 %. A
direct comparison with the Varanasi dataset, however, is not possible
without the original transmittance spectra or, at the very least,
information on the experimental mixing ratios. Further complicating issues,
the Varanasi cross sections are missing negative values near the baselines
(refer to Sect. 4.2), and many have channel fringes superimposed. However,
it is apparent from a direct inspection that the new cross sections have
improved SNR, in some cases substantially so, such as shown in Fig. 5.
Spectral resolution
All spectra used to create the Varanasi cross-section dataset were recorded
at either 0.01 (for sample mixtures of 75 Torr and below) or 0.03 cm-1
spectral resolution (defined as 0.9/MOPD; MOPD: maximum optical path
difference). In the present work 0.01 cm-1 resolution was used for
mixtures below 10 Torr, 0.03 cm-1 for 300 Torr and above, and 0.015
and 0.0225 cm-1 for intermediate pressures. The spectra recorded at
191.6 K, 98.14 Torr and 191.6 K, 200.0 Torr were mistakenly recorded at spectral
resolutions of 0.0225 and 0.0300 cm-1, respectively, instead of the planned
0.015 and 0.0225 cm-1, respectively.
However, careful inspection indicated that there was no under-resolving of
spectral features for these two measurements. Overall, the dataset comparison
indicates that the spectral resolutions chosen for the Varanasi measurements
were suitable.
Pressure–temperature coverage
An absorption cross-section dataset used in remote sensing should cover all
possible combinations of pressure and temperature appropriate for the region
of the atmosphere being observed; in this case the focus is on the
mid-troposphere (∼5 km) up to the stratosphere. Extrapolating
beyond the temperatures and pressures represented within the dataset is
generally unreliable, so forward-model calculations should at the very least
use a four-point interpolation scheme. With this in mind, the P and T of the
laboratory measurements were chosen to cover the range of P and T from
ACE-FTS v3.0 data. The additional cross sections outside the range of the
Varanasi P and T will ensure a better coverage for analysing atmospheric
limb spectra. Figure 6 provides a graphical representation of the PT
combinations for both datasets, illustrating the improved PT coverage (30 PT
combinations in total) relative to the Varanasi dataset. The sampling
density in PT space is lower than for the Varanasi dataset; however, due to
the congestion and lack of any strong sharp features in the spectra, it is
not anticipated that this will have any noticeable effect for remote-sensing
applications.
A graphical representation of the PT coverage for both the new and
Varanasi datasets.