Recent increases in global HFC-23 emissions

[ 1 ] Firn-air and ambient air measurements of CHF 3 (HFC-23) from three excursions to Antarctica between 2001 and 2009 are used to construct a consistent Southern Hemisphere (SH) atmospheric history. The results show atmospheric mixing ratios of HFC-23 continuing to increase through 2008. Mean global emissions derived from this data for 2006–2008 are 13.5 ± 2 Gg/yr (200 ± 30 (cid:2) 10 12 gCO 2 -equivalent/yr, or MtCO 2 -eq./yr), (cid:3) 50% higher than the 8.7 ± 1 Gg/yr (130 ± 15 MtCO 2 -eq./yr) derived for the 1990s. HFC-23 emissions arise primarily from over-fluorination of chloroform during HCFC-22 production. The recent global emission increases are attributed to rapidly increasing HCFC-22 production in developing countries since reported HFC-23 emissions from developed countries decreased over this period. The emissions inferred here for developing countries during 2006–2008 averaged 11 ± 2 Gg/yr HFC-23 (160 ± 30 MtCO 2 -eq./yr) and are larger than the (cid:3) 6 Gg/yr of HFC-23 destroyed in United Nations Framework Convention on Climate Change Clean Development Mechanism projects during 2007 and 2008.


Introduction
[2] Trifluoromethane (HFC-23) has an atmospheric lifetime of 270 yr, a 100-yr global warming potential (GWP) of 14,800 [Forster et al., 2007], and is an unavoidable by-product of chlorodifluoromethane (HCFC-22) production. Climate concerns have prompted efforts to reduce HFC-23 emissions by optimizing conditions during production of HCFC-22 and by destroying HFC-23 before it escapes to the atmosphere. Through voluntary and regulatory efforts in developed (Annex 1) countries [e.g. Approved CDM projects in non-Annex 1 countries generated Certified Emission Reductions (CERs) of 5.7 and 6.5 Gg of HFC-23 (84 and 97 MtCO 2 -eq.), in and 2008, respectively [UNFCCC, 2009. These CDM projects had a value during 2007 and 2008 of nearly US$1 billion annually (at US$13 per ton CO 2 -eq.), which is substantially higher than the estimated industry cost of this HFC-23 emission abatement alone [Wara, 2007].
[3] The importance of understanding the influence of HFC-23 emission abatement efforts has increased with rapid growth in recent production of HCFC-22 in developing countries for both dispersive and feedstock uses [United Nations Environment Programme (UNEP), 2009]. Atmosphere-based estimates of HFC-23 emissions are relevant to ongoing discussions under the UNFCCC and its Kyoto Protocol regarding renewing existing CDM projects and approving additional projects for HCFC-22 facilities that are not currently eligible to participate in this program. In this paper global HFC-23 emissions are estimated from measurements of HFC-23 in ambient air and air from the perennial snowpack (firn) during three separate excursions to Antarctica between 2001 and 2009. The analysis of air trapped in firn provides a robust record of atmospheric trace-gas changes during the past 50 -100 years [Bender et al., 1994;Battle et al., 1996;Butler et al., 1999].

Firn-Air Analysis
[5] Flask air was analyzed using gas chromatography with mass spectrometry and sample cryo-trapping techniques [Montzka et al., 1993]. Separation was performed on a 30-m Gas-Pro column. Both HFC-23 and HCFC-22 were detected with the CHF 2 + ion (m/z = 51) eluting at different times. Calibration is based upon static HFC-23 standards at 8.53 and 25.12 ppt that were prepared with gravimetric techniques. Calibration for HCFC-22 has been discussed previously [Montzka et al., 1993]. Consistency in HFC-23 calibration was checked by periodic analyses of 4 archived air tanks. Results from these analyses showed no significant secular trend in HFC-23 mixing ratios (0.1 ± 0.1 ppt/yr) during 2007-2009. Based on repeat analyses of ambient air and differences between simultaneously filled flasks, the uncertainty on HFC-23 measurements is estimated to be 0.3 ppt.

Firn Modeling
[6] Diffusive air movement within firn was simulated with two different firn models: the Bowdoin model for SPO'01 and WAIS-D [Mischler et al., 2009], and the UCI model for SPO'08-09 [Aydin et al., 2004]. These models allow the consistency between a given trace-gas atmospheric history and firn-air measurements to be tested. The modeled diffusivity vs. depth relationships for each of the field studies were empirically determined by optimizing the agreement between modeled and measured CO 2 depth profiles and the known Antarctic atmospheric CO 2 history [Etheridge et al., 1996;Conway et al., 2004].
[7] An initial atmospheric history for HFC-23 from the 1940s to 2009 (history C) was derived from consideration of multiple inputs: during 1943 to 1995 with an atmospheric box model [Montzka et al., 2009] in which HFC-23 emissions were derived as a constant percentage of past HCFC-22 production (Alternative Fluorocarbons Environmental Acceptability Study, data tables, 2009, available at http:// www.afeas.org) and scaled to fit published measurements of HFC-23 from 40°S during the early 1990s [Oram et al., 1998]; during 1996 -2006 with firn-model-based dating of HFC-23 and HCFC-22 firn data using the ''effective age technique'' [Trudinger et al., 2002]; and with ambient measurements made during the firn-air collections in Jan. 2001, Dec. 2005, and Dec. 2008-Jan. 2009and constant emissions during 2006-2008 [8] Nineteen additional trial mixing ratio histories were considered for HFC-23 (Table 1 and Text S1 of the auxiliary material). 1 Most differed from C only in years after 1995 and were derived with an atmospheric box model incorporating HFC-23 emissions as different and variable fractions of reported HCFC-22 production (F, G, and K histories). A constant emissions scenario was also tested (history H) as were emissions histories derived from updated Cape Grim observations [McCulloch and Lindley, 2007; Intergovernmental Panel on Climate Change (IPCC), 2005] (E histories). Histories were also derived from constant HFC-23 emission to HCFC-22 production (E 23 /P 22 ) fractions to match observed atmospheric HFC-23 mixing ratios at certain dates and as modifications to good-fitting histories, but these trial histories gave poor fits to firn-air results (J and L histories in Text S1). Reduced c 2 values for SPO'01 were calculated with all firn data, but for WAIS-D and at SPO'08-09 with samples only from the mid-to-upper firn (see text). For the eight degrees of freedom associated with the nine samples used to assess histories at both WAIS-D and SPO'08-09 (HCFC-22 > 90 ppt), P < 0.1 for c 2 ! 1.67 (P < 0.05 for c 2 ! 1.938). For SPO'01 (degrees of freedom = 10), P < 0.1 for c 2 > 1.6 [Bevington and Robinson, 2003]. e UNEPa = UNEP HCFC-22 production amounts for dispersive uses only. Fractions of 2.8, 3.0, and 3.2% of UNEPa production correspond approximately to 1.8, 1.9, and 2.0% of total UNEP HCFC-22 production. UNEP(P 22, A5 total ) and UNEP(P 22, nA5 total ) correspond to total HCFC-22 production reported for all uses by developing (A5) and developed (nonA5) countries, respectively (terms used as defined in the Montreal Protocol) (see Text S1).
[9] The well known atmospheric history of HCFC-22, derived from ongoing and archived surface flask measurements [Montzka et al., 1993;2009;Miller et al., 1998] (see Text S1), provides the basis here for deriving accurate HFC-23 histories from firn air. The consistency between trial HFC-23 histories and firn-air data was objectively assessed by calculating reduced c 2 between the modeled and measured HFC-23 vs. HCFC-22 relationship in firn air (Table 1). Reduced c 2 is calculated as P [(modelÀobserved) 2 /error 2 ]/ (degrees of freedom); a c 2 of 1.0 indicates that residuals and uncertainties are similar. The HFC-23 vs. HCFC-22 relationship was used to assess trial HFC-23 histories in order to minimize the influence of errors in the firn diffusivity vs. depth parameterization [Battle et al., 1996]. The accuracy of these models was validated using firn-air measurements of other compounds having well known atmospheric histories (HCFC-22, CFC-12, HFC-134a, and CH 3 CCl 3 ). Consistent results were obtained for all these gases despite their very different histories (see Text S1). Similar conclusions regarding which HFC-23 trial histories are most consistent with the firn data are reached when trial histories are evaluated with the SH atmospheric history and firn data for CO 2 .

Results and Discussion
[10] Results from all three Antarctic firn-air samplings show tight correlations between HFC-23 and HCFC-22 mixing ratios that are nearly linear, suggesting similar relative atmospheric changes for these trace gases in the past (Figure 1). This observation is consistent with emissions of HFC-23 arising primarily from HCFC-22 production at a fairly constant yield. Yields of 1.5 to 4% (by mass) of HFC-23 are typical during the production of HCFC-22, depending upon how well this process is optimized [McCulloch and Lindley, 2007].
[11] Firn air diffusion models provide a means to compare trial atmospheric histories with firn-air observations. A rough estimate of 20th-century changes in HFC-23 mixing ratios was initially provided with history C. This history, when modeled with the Bowdoin and UCI models, yields an expected firn profile that is highly consistent with the entire measured firn profile from SPO'01 and SPO'08-09 (c 2 = 0.7 for SPO'01 and 0.8 for SPO'08-09). This history is also reasonably consistent with Oram et al.'s [1998] results. Contamination of the deepest samples collected at WAIS-D by the KNF pump prevented an assessment of the older part of history C with the WAIS-D data (see Text S1).
[12] To improve our understanding of atmospheric HFC-23 changes since the mid-1990s, a set of trial histories was derived as modifications of history C in years after 1995. These histories were also assessed with the reduced c 2 metric but only against firn samples in the mid-to-upper firn profile having HCFC-22 mixing ratios >90 ppt (>68 m depth at WAIS-D and >62 m depth at SPO'08-09) ( Table 1). HCFC-22 mixing ratios of >90 ppt are representative of high-latitude SH sites since the early 1990s [Montzka et al., 1993;Miller et al., 1998]. Calculated in this way, the reduced c 2 metric reflects model-data agreement for the past two decades.
[13] Among these trial atmospheric histories, only a few provided a good fit (P < 0.1 for reduced c 2 > 1.67) to results from WAIS-D and SPO'08-09 in the mid-to-upper firn (Table 1 and Figure 1). All of these best-fit histories suggest an increase in the growth rate of HFC-23 in the atmosphere after 2005. Trial history H was derived as a linear increase to match ambient mixing ratios in 2001 and at the end of 2005. This history provides a good fit to the WAIS-D firn profile collected in December 2005 (c 2 = 0.6), but, when extrapolated to January 2009, underestimates the surface mixing ratio measured during SPO'08-09 in three different flasks by $1 ppt (Figure 1). History H also gives a poor fit to the SPO'08-09 firn results (c 2 = 2.3; Table 1), providing Multiple trial histories were derived (lines) and incorporated into the firn models to assess their consistency to firn-air measurements (points) (see Table 1 and Text S1 for history descriptions). Best-fitting histories (C, F1, F2, G, K2) are shown as red lines; others are shown in gray, except history H (green line). Results from WAIS-D showing substantial pump contamination are indicated as plus symbols. Insets are expanded views of results from the upper firn. Uppermost points are ambient air samples filled through firnsampling apparati.
further evidence that the atmospheric growth rate of HFC-23 increased in recent years.
[14] The range of trial atmospheric histories considered here leads to a wide range of past global HFC-23 emissions (Figure 2a). The atmospheric histories giving the lowest c 2 all suggest fairly constant emissions from 1990 to 2003 and increased emissions thereafter. A best estimate HFC-23 emissions record was derived from the mean of the five best-fitting SH atmospheric histories and indicates global HFC-23 emissions of 8.7 ± 1 Gg/yr during the 1990s and 13.5 ± 2 Gg/yr (200 MtCO 2 -eq./yr) during 2006-2008 (Figure 2b). By comparison, HCFC-22 emissions during 2006-2008 averaged 610 MtCO 2 -eq./yr [Montzka et al., 2009]. The best estimate HFC-23 emissions history is consistent with one derived from all 20 trial histories after weighting annual emissions by the sum of 1/c 2 from WAIS-D and SPO'08-09. It is also consistent with the mean emissions implied by measured HFC-23 changes in ambient air since 2001 (Figure 2b; see also Text S1). When considered with global HCFC-22 production data (including feedstocks), these results suggest a global mean E 23 /P 22 fraction of 1.7% by mass for 2003 -2008, which is slightly less than observed in the 1990s (Figure 2c) [Oram et al., 1998;McCulloch and Lindley, 2007].
[15] HFC-23 emissions from Annex 1 countries reported to the UNFCCC indicate a substantial decline beginning in 1998 as a result of voluntary and regulatory efforts (Figure 2b) [UNFCCC, 2009] (see Table 2 of Text S1). The decline in Annex 1 emissions stems from reduced HCFC-22 production and a decrease in the E 23 /P 22 fraction from approximately 2% in the 1990s to 0.9% during 2003-2007 (Figure 2c). Reported reductions in Annex 1 HFC-23 emissions and in the E 23 /P 22 fraction cannot be directly verfied with our atmospheric data because during this same period HFC-23 emissions were changing as HCFC-22 production was increasing rapidly in non-Annex 1 countries (Figure 2d).
[16] The difference between global emissions derived here and those reported to the UNFCCC from Annex 1 countries provides an estimate of HFC-23 emissions from non-Annex 1 countries, which are not reported to the UNFCCC (Figure 2b). This analysis suggests steady increases in HFC-23 emissions from non-Annex 1 countries at the same time their HCFC-22 production was increasing on average by $50 Gg/yr (from 2000 to 2007) (Figures 2b  and 2d). Mean HFC-23 emissions from non-Annex 1 countries are estimated to have been 11 ± 2 Gg/y during 2006-2008. A mean E 23 /P 22 of 2.4 ± 0.3% is derived for this same period using total non-Annex 1 HCFC-22 production (Figure 2c).
[17] UNFCCC data show that 5.7 and 6.5 Gg of HFC-23 (84-97 MtCO 2 -eq.) were destroyed in 2007 and 2008, respectively, through the execution of CDM projects approved by the UNFCCC (Figure 2d; see Table 2 of Text S1). This represents the destruction of HFC-23 emissions from 43-48% of the HCFC-22 produced in non-Annex 1 Results are shown for the globe (red lines), for Annex 1 countries (blue lines) and for non-Annex 1 countries (black lines). Figure 2b includes a global best-estimate HFC-23 emissions history calculated from the mean of the best-fit trial histories in Figure 2a (bold red lines; other histories shown as different colors). Global emissions derived from surface measurements alone are indicated as shaded gray regions (Figure 2b, see Text S1). HFC-23 emissions from non-Annex 1 countries are calculated from the difference between the bestestimate global emissions and HFC-23 emissions reported by Annex 1 countries [UNFCCC, 2009] (Figure 2b). E 23 /P 22 values are derived from emissions in Figure 2b and HCFC-22 production data including unrestricted amounts for feedstocks, which accounted for 37% of global production in 2007 [UNEP, 2009]. Adding CDM-related CER quantities to the best-estimate global HFC-23 emissions shows the world avoided by CDM projects (green dotdot-dashed lines (Figure 2b; see Table 2 in Text S1). The green dot-dot-dashed line in Figure 2c is calculated from total non-Annex 1 HFC-23 emissions divided by non-Annex 1 HCFC production not covered by CDMs. Firn and ambient air results yield only a single average for 2006-2008 emissions and quantities derived from these emissions. Global quantities estimated elsewhere are also shown (red circles and lines [Oram et al., 1998] (Figures 2b and  2c). Production and Annex 1 emission data for 2008 are projections (dashed lines in Figures 2b-2d). Uncertainties on firn-derived global emissions represent the spread of best-fit trial histories plus a modeling uncertainty of 10%. Uncertainties of ± 5% are applied to production data and ±10% on reported Annex 1 HFC-23 emissions (see Text S1). Though a 100-yr GWP of 14800 is used here to convert HFC-23 emissions to CO 2 -eq. emissions [Forster et al., 2007], the UNFCCC [2009] uses a GWP of 11700. Annual values are plotted at mid-year. countries during these years. In the world avoided defined by the absence of HFC-23 destruction by CDM projects, global emissions of HFC-23 would have doubled from $9 Gg/yr to $18 Gg/yr during the past decade as HCFC-22 production increased in non-Annex 1 countries (Figure 2b).
[18] Our results indicate that 11 ± 2 Gg/yr of HFC-23 (160 ± 30 MtCO 2 -eq./yr) was emitted during 2006 -2008 from non-Annex 1 countries. These emissions are associated with HCFC-22 production not covered by CDM projects and have an inferred E 23 /P 22 ratio of 3.7 ± 0.3% (Figure 2c; Table 2 of Text S1). This ratio is slightly higher, on average, than inferred for non-Annex 1 countries in most other years and is substantially larger than reported by Annex 1 countries. There are uncertainties in this ratio related to the precise timing of the inferred global emission changes and the extrapolation to 2008 of the Annex 1 reported emission and HCFC-22 production magnitudes. However, these uncertainties do not appreciably affect our derived 2006 -2008 emission and E 23 /P 22 estimates because these estimates represent averages over a 3-year period. The rather high yield ratio inferred for non-Annex 1 HCFC-22 production not currently covered by CDM projects explains why the global E 23 /P 22 fraction did not decrease between 2003 and 2008, even though HFC-23 emissions associated with $30% of total global HCFC-22 production were abated by CDM projects during 2007-2008 (Figures 2c  and 2d).
[19] In summary, the new atmospheric and firn air observations presented here indicate a substantial increase in global HFC-23 mixing ratios and emissions during the early 2000s. These increases are derived for a period when Annex 1 countries reported decreasing emissions to the UNFCCC, indicating that HFC-23 emissions from non-Annex 1 countries increased as they produced more HCFC-22. Although CDM projects destroyed a large fraction of HFC-23 emissions from non-Annex 1 countries during 2007 -2008, both HCFC-22 production data and the non-Annex 1 HFC-23 emissions inferred here suggest that a substantial amount of HCFC-22 production and associated HFC-23 emission continued unabated during these years.