The global SF 6 source inferred from long-term high precision atmospheric measurements and its comparison with emission inventories

Emissions of sulphur hexafluoride (SF 6), one of the strongest greenhouse gases on a per molecule basis, are targeted to be collectively reduced under the Kyoto Protocol. Because of its long atmospheric lifetime (estimated as 800 to 3200 years), the accumulation of SF 6 in the atmosphere is a direct measure of its global emissions. Examination of our extended data set of globally distributed high-precision SF6 observations shows an increase in SF 6 abundance from near zero in the 1970s to a global mean of 6.7 ppt by the end of 2008. In-depth evaluation of our long-term data records shows that the global source of SF 6 decreased after 1995, most likely due to SF 6 emission reductions in industrialised countries, but increased again after 1998. By subtracting those emissions reported by Annex I countries to the United Nations Framework Convention of Climatic Change (UNFCCC) from our observation-inferred SF 6 source leaves a surprisingly large gap of more than 70–80% of non-reported SF6 emissions in the last decade. This suggests a strong under-estimation of emissions in Annex I countries and underlines the urgent need for independent atmospheric verification of greenhouse gases emissions accounting. Correspondence to: I. Levin (ingeborg.levin@iup.uni-heidelberg.de)


Introduction
SF 6 is an extremely stable mainly anthropogenic gas, having a very high global warming potential of 23 900 (Forster et al., 2007).Traces of SF 6 have been shown to be produced in the Earth' crust (Harnisch and Eisenhauer, 1998), but natural fluxes into the atmosphere are negligible in the context of the present study (Busenberg and Plummer, 2000).The industrial production of SF 6 began in 1953 for use as an insulation gas in high voltage installations (Ko et al., 1993;Maiss and Brenninkmeijer, 1998).Emissions from the electricity sector (through leakage and venting) continue to form the largest source of SF 6 to the atmosphere, with additional contributions from magnesium production, semiconductor manufacturing as well as other minor sources (Olivier et al., 2005).SF 6 is primarily destroyed in the mesosphere; its atmospheric lifetime is estimated to range from 800 to 3200 years (Ravishankara et al., 1993;Morris et al., 1995) with the latter number being adopted by IPCC (Forster et al., 2007).Therefore, more than 96-99% of all SF 6 emitted to the atmosphere accumulates there, allowing us to directly infer its global emissions from the observed atmospheric concentration increase (Maiss and Levin, 1994).Assuming that the distribution of emissions is well known (e.g. from inventories such as EDGAR, 2009), SF 6 has been widely used as Table 1.Characteristics of SF 6 measurement stations; the column "type" distinguishes between long-term background stations (LTB), regular aircraft sampling (RAS) and balloon sites (B).For the balloon sites, the figure in brackets after the "B" denotes the number of vertical SF 6 profiles taken.a tracer to compare and validate atmospheric transport models (e.g., Levin and Hesshaimer, 1996;Denning et al., 1999;Kjellström et al., 2000;Waugh and Hall, 2002;Peters et al., 2004;Gloor et al., 2007;Bönisch et al, 2008;Patra et al., 2009).
In this study, global SF 6 emissions from 1978 to 2008 are estimated using its accumulation rate in the atmosphere based on new observational data.These top-down source estimates, hereafter called inferred emissions, are compared to global annual emissions published in the most recent compilation (V4.0) of the EDGAR data base (EDGAR, 2009) which uses a so-called bottom-up approach, based on statistical information on the sources and their global distribution.Both, top-down inferred and bottom-up estimates are further compared to SF 6 emissions reported by Annex I countries to UNFCCC (2009).Annex I countries include all major industrial countries of Western Europe, Canada, the United States, Japan, Australia, and New Zealand, as well as Eastern European countries, the Russian Federation and Turkey.At least until the mid 1990s, emissions from Annex I countries should comprise the major share of global SF 6 emissions.After 2000 emissions from newly industrialised countries also significantly contribute to the increasing global atmospheric SF 6 burden (EDGAR, 2009).The difficulties to validate, on the regional scale, reported bottom-up emissions with spatially distributed observations and atmospheric transport modelling will also be discussed.
The tropospheric SF 6 records from Neumayer, Cape Grim, Izaña, Alert and the aircraft sites (altitude >2500 m) are displayed in Fig. 1.From 1998 to 2006 an almost constant increase rate (solid lines) is observed, suggesting near constant global SF 6 emissions; only in the last four years emissions are increasing again.The measurement records from Izaña and Alert, starting in 1991 and 1993 respectively, show that mixing ratios from the Northern Hemisphere are about 0.3 to 0.4 ppt higher than the Southern Hemispheric data from Cape Grim and Neumayer.The observed inter-hemispheric difference is due to the uneven distribution of sources (more than 95% of SF 6 emissions originate in the Northern Hemisphere; Olivier et al., 2005;EDGAR, 2009), combined with the ca. 1 year inter-hemispheric exchange time of air masses.2   0 . 3   1 9 8 0  1 9 8 5  1 9 9 0  1 9 9 5  2 0 0 0  2 0 0 5  2 0 1

A new top-down estimate of global SF 6 emissions
Observations at the four globally distributed stations and along meridional transects over the Atlantic Ocean show relatively uniform SF 6 mixing ratios north of 30 • N, and south of about 15 • S. A nearly linear north-to-south decrease is observed in the tropics (Maiss et al., 1996;Geller et al., 1997).If we assume that observations from these background stations and the Syktyvkar aircraft location represent the zonal mean tropospheric SF 6 mixing ratios in their respective latitudinal bands, it is possible to reconstruct the temporal evolution of the tropospheric SF 6 distribution from this network.Further, combination with observed stratospheric SF 6 profiles (Supplementary Fig.With an atmospheric lifetime of 800-3200 years (Ravishankara et al., 1993;Morris et al., 1995), the total atmospheric SF 6 sink in 2005 was 0.04-0.17Gg (1 Gg = 10 9 g), i.e. approximately 1-4% of the observed annual atmospheric increase of ca.5-6 Gg.Considering that the oceanic sink is one order of magnitude smaller than the atmospheric sink (Ko et al., 1993), the total SF 6 sink can be neglected.Therefore, we claim that the first tempo- Newly industrialized countries not included in Annex I of the UNFCCC, such as China, India, Brazil, and others, are not required to report SF 6 emissions to UNFCCC.Their SF 6 emissions as estimated by EDGAR are displayed in Fig. 2b (dashed-dotted black line).These data are compared here with the residual emissions as calculated from the difference between the total observation-inferred SF 6 source and Annex I reported SF 6 emissions.These so-called Non-reported emissions are also plotted in Fig. 2b, either based on originally reported or Japan-corrected emissions.We call these "scenarios" of SF 6 emissions "UNFCCC-based".From 1995 to 2000 Non-reported emissions in the UNFCCC-based scenario are more than three times higher than EDGAR estimates for Non-Annex I countries, and in 2000 they already account for about 2/3 of the global SF 6 source.Both, the increase of Non-reported emissions and their respective share in the global emissions are surprisingly large, in particular if all these Non-reported emissions would have to be assigned to emissions from Non-Annex I countries.The dominant sector for SF 6 use and emission is electricity production (Olivier and Berdowski, 2001).Interestingly, the share of Non-Annex I countries in global electricity production was only 1/3 in 2000 (BP, 2009).Assuming that the UNFCCC-based SF 6 emission scenario is correct therefore implies that the annual SF 6 emissions per GWh electricity production would be two to more than three times larger in Non-Annex I countries than in Annex I countries.The large discrepancies between the two emission scenarios, EDGAR and UNFCCC-based are further investigated in the next section.

Comparison of observed mixing ratios with model simulations
A comparison of observed mixing ratios with atmospheric transport model estimates that are based on different emission scenarios may help to decide which of these scenarios is more likely correct.Here we have deployed the two emission distribution estimates, EDGAR and the corrected UNFCCC-based in a coarse-resolution two-dimensional atmospheric box model GRACE (Levin et al., 2010).Thereby, simulated tropospheric SF 6 mixing ratios were compared with observations at Alert (82 extra-tropical tropopause.While the meridional distribution of SF 6 emissions for the EDGAR scenario can be directly taken from the data base (EDGAR, 2009), we used for the source distribution of the UNFCCC-based scenario for Annex I countries the officially reported emissions corrected for Japan (UNFCCC, 2009).For Non-Annex I countries the residual from the global inferred SF 6 source (dotted blue line in Fig. 2b) was distributed according to the electrical power production of these countries (see Supplementary Material Sect.4.3 and Fig. A4, http://www.atmos-chem-phys.net/10/2655/2010/acp-10-2655-2010-supplement.pdf).Simulated SF 6 mixing ratios from 1976 to 2009 for the mid latitude box of GRACE in the southern hemisphere are displayed in Fig. 3a, together with the observations at Cape Grim (41 • S) starting in 1978.There is generally good agreement of the long-term trends between EDGAR-based simulations and the observations, with model results being slightly lower than observations up to 1992 and again from about 2003 onwards.When using the relative latitudinal distribution of SF 6 emissions from EDGAR adjusted to our observation-inferred annual totals, we obtain almost perfect agreement with observations for the whole period from 1978 until present.This confirms our top-down method, but also shows how sensitively global tropospheric SF 6 mixing ratio trends can reflect the underlying source strengths.
With the adjusted EDGAR distribution, the mixing ratios at Alert (82 • N) and Neumayer (71 • S) (inlay in Fig. 3a) as well as the north-south gradient (Fig. 3b, thick red line) are also correctly reproduced.However, model simulations using the UNFCCC-based emission scenario slightly underestimate the observed north-south gradient for the period of 1995 to the present (Fig. 3b, thick blue line).This could be an indication that in the UNFCCC-based scenario the distribution of sources is not correct, i.e. that emissions are overestimated in the southern hemisphere or in the tropics and underestimated in mid latitudes of the northern hemisphere.Except for Australia and New Zealand, inhabited regions of all Annex I countries are located in northern mid-latitudes (30 • N-60 • N), while large areas of Non-Annex I countries (i.e.Southern China, India and Brazil) are located in subtropical and tropical regions of the Northern and Southern Hemispheres.Consequently, a shift of SF 6 emissions from Annex I to Non-Annex I countries (e.g. after the beginning of the 1990s) would imply a southward redistribution of the global SF 6 source.This in turn should result in a smaller SF 6 mixing ratio difference between the north and the south, as simulated with the UNFCCC-based emission scenario after about 1992 (Fig. 3b).
Besides incorrect distribution of sources, model-data mismatch can, however, also be caused by several other factors: Meridional mixing in the model may be over-estimated, resulting in under-estimation of inter-hemispheric SF 6 differences or vice versa.Also, the observations at Alert and Neumayer may not necessarily be representative for the large GRACE model boxes.Atmospheric transport between the different boxes in GRACE, which was kept constant from year to year, has been optimized using bomb 14 CO 2 , 10 Be/ 7 Be, but also SF 6 with estimates of the total SF 6 source taken from Levin and Hesshaimer (1996), assumed to be spatially distributed according to electricity production (Prather et al., 1987).However, this transport optimization still allows for an uncertainty of hemispheric residence times on the order of ±25%.Indeed, increasing hemispheric residence times by 25% would bring model simulations with the UNFCCC-based scenario almost into agreement with the observations (upper boundary of the blue hatched area in Fig. 3b; the mean inter-hemispheric exchange time was increased here from 0.95 to 1.19 years); in this case the EDGAR distribution would over-estimate the northsouth difference (upper boundary of the red hatched area in Fig. 3b).At this stage, using the observed north-south difference of SF 6 we thus cannot firmly reject one of the two SF 6 emission scenarios.This is mainly because of the lack of really independent validation of transport properties in our model, i.e. independent from any prior calibration with SF 6 or other tracers, whose emissions also have source distributions similar to that of electricity production, such as fluorocarbons (Prather et al., 1987).A possible tracer for transport validation may be 85 Krypton (Jacob et al., 1987;Levin and Hesshaimer, 1996) which is mainly emitted from nuclear fuel reprocessing plants.However, these 85 Krypton emissions are essentially confined to a few large point sources (Winger et al., 2005), a distribution not suitable for transport validation of our coarse resolution GRACE model.

Conclusions
Rigorously assessing the reliability of the EDGAR or the UNFCCC-based emission distribution, and validation of reported emissions by Annex I countries may be possible with a high resolution atmospheric transport model; however, such a model needs to be very well validated to correctly simulate atmospheric transport and mixing, i.e. with different tracers that have different but well-known source characteristics and distribution.Also, this would require a denser observational network.At present we are thus left with only the evidence from total Non-reported SF 6 emissions as well as from specific SF 6 emissions per electricity production, which suggests, that Annex I reported UNFCCC emissions during the 1990s (and possibly until today) are too low.This suggestion is confirmed by the EDGAR (2009) data base which assumes much lower emission factors for leakage and maintenance for Annex I than for Non-Annex I countries, but which results in similar emission intensities of SF 6 per GWh electricity produced (J.G. J. Olivier, personal communication, 2010).Annex I reported emissions may be too low because of intrinsic uncertainties of estimated SF 6 stored in end-use applications in the US and Europe (Maiss and Brenninkmeijer, 1998), and possibly due to underestimated emissions from economies in transition.The accelerating increase of global SF 6 emissions since the end of the 1990s may be linked to rising emissions from Non-Annex I countries, which qualitatively agree with their economic growth (e.g.China) (RHGDP, 2008).However, as long as these countries are not obliged to report their emissions to UNFCCC, there will remain large inaccuracies in related bottom-up emission estimates.
Our study clearly shows that top-down verification of reported emissions is without alternative for greenhouse gases budgeting.On a country level, such validation can, however, only be achieved with a dense network of high-precision atmospheric observations in combination with adequately calibrated atmospheric transport models.Verification of total global SF 6 emissions by atmospheric measurement is accurately possible without a sophisticated transport model, even with only one or a few globally distributed background stations.This mechanism should, therefore, be included as an additional measure in the Kyoto reporting process, as it provides the only ultimate proof of total reported emissions (changes), at least for gases with well-defined sinks such as SF 6 , and other fluorinated and chlorinated compounds.

Fig. 1 .
Fig. 1.Global observations of tropospheric SF 6 : Symbols and left axis: Atmospheric SF 6 mixing ratios (given in ppt = parts per trillion, i.e. pico moles of SF 6 per mole of dry air) observed in the Northern and Southern Hemispheres.The overlapping flask and cylinder data from Neumayer are virtually indistinguishable (note that 12 data points in total have been rejected as outliers from the figure).Lines and right axis: SF 6 growth rates calculated for individual stations from de-seasonalized measurements using a fit routine fromNakazawa et al. (1997) (colour codes: same as original data).

Fig. 3 .
Fig. 3. Comparison of observed and simulated SF 6 mixing ratios and north-south differences: (a) SF 6 observations from Cape Grim (41 • S), Alert (82 • N) and Neumayer (71 • S) (inlay) in comparison with GRACE simulations for the respective model box, based on the original EDGAR (2009) emissions (black lines) as well as simulations obtained when total EDGAR emissions were adjusted to the inferred emissions from Figure 2a (red line).(b) SF 6 differences between Alert and Neumayer (taken from the fitted curves through the data and 1σ error estimates, grey line and hatched area) in comparison with GRACE simulations.Hatched areas show model results when inter-hemispheric transport is varied by ±25%.

Table 2 .
Observed global atmospheric SF 6 inventory and inferred annual SF 6 source.The 1σ uncertainties of the inventory are ±3-4Gg, while the uncertainties of the annual emissions are ±6%, neglecting oceanic and atmospheric sinks.A total atmospheric mass of 5.1×10 21 g of air was used for the source estimates.Note that the inventory refers to the middle of each year, whereas the source refers to the period 1 January-31 December Comparison of annual observation-inferred global SF 6 emissions (red line with ±1σ uncertainty range) with global emissions estimated byEDGAR (2009).Also included are the (Japancorrected and original, see main text) emissions reported by Annex I countries to UNFCCC in 2009 (dashed resp.solid blue line) as well as estimated Annex I emissions from the EDGAR data base.