Revealing the global emission gaps for fully fluorinated greenhouse gases

In response to the global trend of climate change, it is important to accurately quantify emissions of fully fluorinated greenhouse gases (FFGHGs, referring to SF6/NF3/CF4/C2F6/C3F8/c-C4F8 here). Atmospheric observation-based top-down methods and activity-based bottom-up methods are usually used together to estimate FFGHG emissions at the global and regional levels. In this work, emission gaps at global and regional levels are discussed among top-down studies, between the top-down and bottom-up FFGHG emissions, and among bottom-up emissions. Generally, trends and magnitudes of individual FFGHG emissions among top-down estimates are close to each other within the uncertainties. However, global bottom-up inventories show discrepancies in FFGHG emissions among each other in trends and magnitudes. The differences in emission magnitudes are up to 93%, 90%, 88%, 83%, 87%, and 85% for SF6, NF3, CF4, C2F6, C3F8, and c-C4F8, respectively. Besides, we reveal the insufficient regional TD studies and the lack of atmospheric observation data/stations especially in areas with potential FFGHG emissions. We make recommendations regarding the best practices for improving our understanding of these emissions, including both top-down and bottom-up methods.


Emission gap among TD from a global perspective
Only the global TD emissions of the individual FFGHG from previous works have been summarized in Fig. 1 and Supplementary Table 2. Figure 1a shows that the mean global SF 6 emissions have increased from 57 Mt CO 2 -eq yr −1 in 1978, reaching a peak (143 Mt CO 2 -eq yr −1 ) in about 1995, then decreased to approximately 118 Mt CO 2 -eq yr −1 in 2000 or so, again consistently rising to 211 Mt CO 2 -eq yr −1 in 2019, followed by a slight decline to 205 Mt CO 2 -eq yr −1 in 2020.During 1978-2008, global SF 6 emissions from three TD studies 1,14,15 showed similar trends as described above and magnitudes with a mean emission of 112 Mt CO 2 -eq yr −1 over this period.Figure 1b shows the significant rise in NF 3 emissions ranging from 0 in 1979 to 68 Mt CO 2 -eq yr −1 in 2020.Throughout 2000-2011, there are consistently NF 3 increasing trends (increasing rate of 1.4 Mt CO 2 -eq yr −2 ) and magnitudes (average emission of about 12.8 Mt CO 2 -eq yr −1 ) among previous global TD results [16][17][18] .
Figure 1c illustrates that global CF 4 emissions 16,18,19 with fluctuation have grown from 0 in 1900 to 111 Mt CO 2 -eq yr −1 in 2020.Note that the global CF 4 emissions among different studies were relatively close within the uncertainties of TD results.For example, the average global CF 4 emissions throughout 1900-1978 from Trudinger et al., 2016 20 using InvE2 and InvEF inversions are both 25 Mt CO 2 -eq yr −1 .Besides, the average global CF 4 emissions during 1979-2014 from Trudinger et al., 2016 20 using InvE1, InvE2, and InvEF inversions are in the range of 94-97 Mt CO 2 -eq yr −1 , close to the values of other works 16,19 (97 Mt CO 2 -eq yr −1 ) over the same period.However, the average global CF 4 emissions (89 Mt CO 2 -eq yr −1 ) over 1975-1989 from Worton et al., 2007 21 were significantly lower than those from all other global TD studies (103-140 Mt CO 2 -eq yr −1 ) over the same period.Figure 1d illustrates that the global C 2 F 6 emissions 16,18,19 20 using InvE1, InvE2, and InvEF inversions are close to 28 Mt CO 2 -eq yr −1 , consistent with values of other works 16,19 (28 Mt CO 2 -eq yr −1 ) over this period.However, the average global C 2 F 6 emissions over 1975-1994 from Worton et al., 2007 21 were only 14 Mt CO 2 -eq yr −1 , about half the values (21-26 Mt CO 2 -eq yr −1 ) from all other studies.The choice of an inversion model may cause the CF 4 &C 2 F 6 emission difference between Worton et al., 2007 21 and the other TD studies.In Worton's work, collecting the firn air samples, they used an iterative approach with a firn physical transport model to obtain emissions.While AGEGE 12-box atmospheric transport model 19,22 or the combination of AGEGE 12-box atmospheric transport model with the iterative approach 20 were used in other TD studies.
Figure 1e indicates that the global C 3 F 8 emissions 16,18,19 ranged from 0.0012 Mt CO 2 -eq yr −1 in 1990 to 5.3 Mt CO 2 -eq yr −1 in 2020 with a peak in 2003 or so.There is an agreement on the global C 3 F 8 emission from different TD studies during 1900-1982 (average value of 0.24 Mt CO 2 -eq yr −1 ) 20 and during 1983-2014 (average value of 5.0 Mt CO 2 -eq yr −1 ) 16,19,20,22 .In Fig. 1f., c-C 4 F 8 emissions 23,24 rose from 0.61 Mt CO 2 -eq yr −1 in 1990 to 23.9 Mt CO 2 -eq yr −1 in 2020 with fluctuation over this period.Before 2000, the global c-C 4 F 8 emissions among previous TD studies were close to each other with a similar average emission of 8-10 Mt CO 2 -eq yr −1 although the global c-C 4 F 8 emissions from Oram et al., 2012 25 showed larger variability.After 2000, the global c-C 4 F 8 emissions among previous TD studies were close to each other with a similar average emission of 12-14 Mt CO 2 -eq yr −1 and an increasing rate of 0.72-0.96Mt CO 2 -eq yr −2 .(5 Mt CO 2 -eq yr −1 ) were slightly lower than results of around 8 Mt CO 2 -eq yr −1 from both Mühle et al., 2019 24 and Mühle et al., 2022 23 .The c-C 4 F 8 datasets from more than one station (including Zeppelin, Mace Head, Jungfraujoch, Monte Cimone, Trinidad Head, Shangdianzi, Gosan, La Jolla, Ragged Point and so on) were employed in other studies 23,24 to derive c-C 4 F 8 emissions, while Oram et al., 2012 25 only used the c-C 4 F 8 dataset from Cape Grim to obtain the global TD c-C 4 F 8 emissions, which may explain the larger variability of its results.

Emission gap among BU from a global perspective
Note that there is a notable difference among these inventory results (Fig. 2). Figure 2a shows that among three inventories, the EDGAR inventory reported the highest global total FFGHG emissions rising from 185 Mt CO 2 -eq   www.nature.com/scientificreports/yr −1 in 1983 to 279 Mt CO 2 -eq yr −1 in 2021 (green solid line).However, global total FFGHG emissions reported by EPA kept relatively stable but with a lower magnitude (166 Mt CO 2 -eq yr −1 in 1990 to 174 Mt CO 2 -eq yr −1 in 2021) (red solid line).The global total FFGHG emissions from UNFCCC dropped from 149 Mt CO 2 -eq yr −1 (1990) to 26 Mt CO 2 -eq yr −1 (2021) (blue solid line).Figure 2b-h illustrates the discrepancies among inventories for individual FFGHG.The global SF 6 emissions reported by the EDGAR rose from 17 Mt CO 2 -eq yr −1 to 217 Mt CO 2 -eq yr −1 from 1970 to 2021 (Fig. 2b).Despite the same increasing trend with the EDGAR, EPA reported the global SF 6 emissions with a lower magnitude (63 Mt CO 2 -eq yr −1 -104 Mt CO 2 -eq yr −1 from 1990 to 2021) (Fig. 2b).However, the global SF 6 emissions submitted to the UNFCCC declined from 60 Mt CO 2 -eq yr −1 in 1990 to 17 Mt CO 2 -eq yr −1 in 2021 (Fig. 2b).In Fig. 2c, EPA has the highest global NF 3 emissions with an average of 6 Mt CO 2 -eq yr −1 (1.5 Mt CO 2 -eq yr −1 for UNFCCC and 2.7 Mt CO 2 -eq yr −1 for EDGAR) and shows the highest increase with the rate of 0.46 Mt CO 2 -eq yr −2 (0.026 Mt CO 2 -eq yr −2 for UNFCCC and 0.093 Mt CO 2 -eq yr −2 for EDGAR) over 2000-2021 among the three inventories.In addition, global NF 3 emissions reported by the EDGAR (0 in 1970 to 2.8 Mt CO 2 -eq yr −1 in 2021) display a similar trend to the UNFCCC results (0.10 Mt CO 2 -eq yr −1 in 1990 to 1.1 Mt CO 2 -eq yr −1 in 2021) but with a slightly higher emission magnitude.For PFCs (Fig. 2d), EDGAR and EPA show relatively similar emission trends (− 1.3 Mt CO 2 -eq yr −2 for both EDGAR and EPA) and magnitudes (the average of 70 Mt CO 2 -eq yr −1 for EDGAR and 63 Mt CO 2 -eq yr −1 for EPA) over 1990-2021, different from those for PFC emissions submitted to UNFCCC (the average of 42 Mt CO 2 -eq yr −1 and the rate of − 2.6 Mt CO 2 -eq yr −2 ).UNFCCC reported the lowest PFC emissions before 2009 despite the same increasing trend.Moreover, the overall trend in PFC emissions after 2009 was decreasing in the UNFCCC reports but increasing in the EDAGR and EPA results.Figure 2e-h illustrates the overall higher emissions in EDGAR for each PFC than those from the UNFCCC (without individual PFC emissions provided in the EPA reports).C 2 F 6 emission gaps between the EDGAR (23 Mt CO 2 -eq yr −1 in 1990 to 9.6 Mt CO 2 -eq yr −1 in 2021) and UNFCCC (17 Mt CO 2 -eq yr −1 in 1990 to 1.6 Mt CO 2 -eq yr −1 in 2021) were the highest, while C 3 F 8 emission gaps between the EDGAR (0.21 Mt CO 2 -eq yr −1 in 1990 to 0.37 Mt CO 2 -eq yr −1 in 2021) and UNFCCC (0.22 Mt CO 2 -eq yr −1 in 1990 to 0.29 Mt CO 2 -eq yr −1 in 2021) were the lowest.
The discrepancies in inventory results may be brought by factors like emission source sector inclusion and country coverage in the inventories.Taking SF 6 emission sources in the EDGAR and EPA as example, EDGAR covered four SF 6 emission sources: chemical industry, metal industry, electronics industry, and other product manufacture and use; EPA covered electric power systems (EPS), electronics (manufacturing of semiconductors, photovoltaics and flat panel displays), and metal industry (magnesium production).EDGAR only provided general emission sector description like electronics industry and other product manufacture and use without detailed subsource, while EPA showed the subsource information of electronics industry.Besides, EPS, the major emission source of SF 6 was not found in the EDGAR; the chemical industry was not contained in the EPA.Thus, it is hard to say which SF 6 emission dataset has the most complete inputs and might therefore be most reliable.Combining multiple datasets makes it possible to obtain reliable emission estimates.In addition, the EPA and EDGAR both reported the NF 3 emissions from non-Annex I countries.However, NF 3 emissions from the non-Annex I countries were not available in the UNFCCC.With these missing data, it is not easy to determine whether there are no emissions or whether emissions were not calculated.This vague cognition would impair the accuracy of the existing BU estimates, which is not conducive to a correct understanding of the causes of the TD-BU differences.The above statements indicate that there is no consensus on the accounting of FFGHG emissions.It seems sectors and/or countries covered by previous inventories are different.Each inventory has its own disadvantages.Thus, more work such as identifying potential emission sources, including NF 3 in national inventories of non-Annex I countries, and strengthening the national inventory reporting mechanism should be developed to further optimize existing BU results for FFGHGs in the future.

Emission gap between TD and BU from a global perspective
Figure 2a shows the significant gap between global TD and BU total FFGHG emission estimates.Total FFGHG emissions here mean the sum of emissions of six individual FFGHG.First, albeit with fluctuations, global total FFGHG emissions from TD have shown an overall upward trend with an increasing rate of 5.3 Mt CO 2 -eq yr −2 .However, three inventories showed diverse emission trends, partially different from the TD result.Among inventories, only the EDGAR inventory displayed a similar increase trend in global total FFGHG emissions but   18 (SF 6 /NF 3 /CF 4 /C 2 F 6 /C 3 F 8 emissions in 2020).The annual global TD total FFGHG emissions were the sum of six FFGHG global TD emissions.The purple shading area represents the 16th-84th percentile range from the AGAGE 12-box model.The hollow squares mean that these values were extrapolated from the recent 5 years' emissions.FFGHG emissions in the United Nations Framework Convention on Climate Change (UNFCCC) are obtained from the following website: https:// di.unfccc.int/ flex_ annex1 and https:// di.unfccc.int/ flex_ non_ annex1.FFGHG emissions in the Emissions Database for Global Atmospheric Research (EDGAR) are from EDGAR v4.2 37 (1970-1989) and EDGAR v7.0 38 (1990-2021) with an increase rate of 3.3 Mt CO 2 -eq yr −2 .However, EPA's report showed relatively steady FFGHG emissions ranging from 166 Mt CO 2 -eq yr −1 (1990) to 174 Mt CO 2 -eq yr −1 (2021).In addition, global total FFGHG emissions reported by the UNFCCC fluctuated widely and a decreasing trend could be found with a decreasing rate of 4.0 Mt CO 2 -eq yr −2 from 1990 to 2021.Noticeably, after 2009 global total FFGHG emissions from TD, the EDGAR inventory, and EPA's report all showed an obvious increase with the rates of 11.6 Mt CO 2 -eq yr −2 , 6.4 Mt CO 2 -eq yr −2 , and 5.3 Mt CO 2 -eq yr −2 , respectively.Figure 2a  www.nature.com/scientificreports/BU estimates may result from emission underestimates of activity-based inventories as well as from substantial emissions from non-reporting countries.However, the causes for the differences between TD and EPA/EDGAR inventories are not fully known.The emission gaps are also found in the individual FFGHG (Fig. 2b-h).From 1978 to 2020, the EDGAR inventory shows the average global SF 6 emission of 124 Mt CO 2 -eq yr −1 and an increasing rate of 3.9 Mt CO 2 -eq yr −2 , consistent with those from TD (the average emission of 134 Mt CO 2 -eq yr −1 and the increase rate of 3.9 Mt CO 2 -eq yr −2 ) (Fig. 2b).However, there are obvious discrepancies between the EDGAR inventory and TD results for both NF 3 (Fig. 2c) and PFCs (Fig. 2d-h).Especially, the discrepancies between the EDGAR and TD results for NF 3 , CF 4 , C 2 F 6 , and c-C 4 F 8 have gradually increased.Figure 2b-d illustrates that despite the similar trend, the EPA estimates for SF 6 (63 Mt CO 2 -eq yr −1 in 1990 to 100 Mt CO 2 -eq yr −1 in 2020), NF 3 (0.24 Mt CO 2 -eq yr −1 in 1990 to 9.9 Mt CO 2 -eq yr −1 in 2020), and PFCs (103 Mt CO 2 -eq yr −1 in 1990 to 57 Mt CO 2 -eq yr −1 in 2020) were significantly lower than the global TD emissions for SF 6 (118 Mt CO 2 -eq yr −1 in 1990 to 205 Mt CO 2 -eq yr −1 in 2020), NF 3 (0.6 Mt CO 2 -eq yr −1 in 1990 to 68 Mt CO 2 -eq yr −1 in 2020), and PFCs (157 Mt CO 2 -eq yr −1 in 1990 to 167 Mt CO 2 -eq yr −1 in 2020), respectively.In addition, UNFCCC estimates for each FFGHG were also lower than those from global TD results.Figure 2 displays the gradual decline in global emissions for SF 6 , CF 4 , and C 2 F 6 reported by the UNFCCC with the decreasing rate of 1.4, 2.1, and 0.50 Mt CO 2 -eq yr −2 , as well as the gradual increasing discrepancies between the UNFCCC and TD results for SF 6 , CF 4 , and C 2 F 6 .It is also worth paying attention to the significant emission gaps for c-C 4 F 8 between the average global TD (12.8 Mt CO 2 -eq yr −1 over 1990-2020) and UNFCCC (0.28 Mt CO 2 -eq yr −1 over 1990-2020).

Emission gap among TD from a regional perspective
The regional TD emission estimation is usually carried out based on the location of existing atmospheric observation stations.Previous regional TD studies for FFGHGs have been gathered in Supplementary Table 3, indicating that existing TD research on FFGHG emissions mainly focused on the following regions: eastern Asia (China; Japan; South Korea; North Korea; and Mongolia), northwest Europe (referring to terms "northwestern Europe/ West Europe/northwest Europe" used in previous studies) [Austria; Belgium, the Netherlands, and Luxembourg (collectively termed Benelux); Denmark; France; Germany; Ireland; Italy; Portugal; Spain; Switzerland; and the United Kingdom (UK)], the US, Australia, India, and Russia.
The FFGHG TD estimates from the above regions except for China (provided by Guo et al., 20023 27 ) are shown in Fig. 3 and Supplementary Figs.1-10.Among these estimates, only one TD result for specific FFGHG is available in regions and countries including China (NF 3 ), Japan (NF 3 ), South Korea (NF 3 ), North Korea (SF 6 / NF 3 /CF 4 /C 2 F 6 /C 3 F 8 ), Mongolia (SF 6 ), northwest Europe (CF 4 /C 2 F 6 /C 3 F 8 /c-C 4 F 8 ), and the US (SF 6 ).These limited TD results are not sufficient to understand FFGHG emissions from these regions.Thus, more work on the emission quantification of FFGHGs in these regions by the TD method should be developed to further verify the previous TD results.Note  ), more than one TD results for each FFGHG are accessible.If considering the emission uncertainties, parts of regional TD FFGHG emissions were relatively consistent.For example, Fig. 3 shows that FFGHG emissions in Japan from different TD studies were close to each other.Supplementary Fig. 1 shows the consistency among different TD studies for SF 6 /CF 4 /C 2 F 6 /C 3 F 8 emissions in South Korea.
Supplementary Figure 4 displayed the relatively close four groups of SF 6 emissions in northwest Europe shown by Simmonds et al., 2020 1 despite using different inverse models and observation data with different number of sampling points and sampling years.However, there are gaps among part of the regional TD results.For example, Supplementary Fig. 10a shows that Australian SF 6 emissions using the interspecies correlation (ISC) method (68 ± 25 t yr −1 in 2005 to 18 ± 6 t yr −1 in 2016) 28 were quite different from those using the InTEM model (29 ± 2 t yr −1 in 2005 to 44 ± 2 t yr −1 in 2016) 28 .Similarly, Supplementary Fig. 10d shows the differences between the Australian C 3 F 8 emissions using the ISC method (7 ± 3 t yr −1 in 2005 to 9 ± 3 t yr −1 in 2016) 28 and those using the NAME model (9 ± 1 t yr −1 in 2005 to 20 ± 2 t yr −1 in 2016) 28 .This means that the selection of TD method covering inversion model, prior emissions, observations, and uncertainties would impact the TD result.In addition, Guo et al., 2023 27 also show obvious discrepancies among TD emissions for SF 6 /CF 4 /C 2 F 6 /c-C 4 F 8 in China.
For previous TD studies, the lack of atmospheric measurement data from existing stations would impede the accurate understanding of long-term FFGHG emissions.For example, Say et al., 2021 19 only reported emissions (2005-2010 for CF 4 ; 2005-2007 for C 2 F 6 and C 3 F 8 ) from the UK, Ireland, and Benelux due to the lack of atmospheric measurements during this period from continental Europe and thus sensitivity to southern France and eastern Germany.In addition, due to the availability of measurements from Jungfraujoch station, reported estimates for France and Germany (and Northwest Europe total) began in 2008 (C 2 F 6 and C 3 F 8 ) and 2010 (CF 4 ) 19 .
Besides, the lack of atmospheric measurement stations would not be conducive to an accurate understanding of FFGHG emissions.For example, Mühle et al., 2019 24 indicated that several large areas such as the US and India where c-C 4 F 8 emissions may occur were not closely monitored by the AGAGE network.c-C 4 F 8 emissions from the continental US were not estimated because two AGAGE stations in California could only catch part of the c-C 4 F 8 emissions from the continental US due to predominant westerly winds 24 .For India, the inversion method played a limited role in identifying distant point sources from a relatively small number of samples 24 .Weiss et al., 2021 29 pointed out that vast blind spots exist in the AGAGE and NOAA measurement networks which include large parts of the developed regions relatively well sampled such as eastern Asia, central North America, and northwest Europe as summarized in our work; however, southern, western, and central Asia, large parts of Southeast Asia, all of South America, portions of North America, Eastern Europe, and New Zealand and most of Africa are not covered and emissions from many of these areas are expected to increase with industrial and economic development 29 .
Overall, more TD research on regional FFGHG emission quantification needs to be carried out to verify the previous results and reduce the uncertainties of FFGHG emissions.Besides, atmospheric measurements from the current regional atmospheric observation should be further completed.More atmospheric observation stations should be developed as well to expand coverage of potential emission areas and then improve the accuracy of atmospheric measurements.3. FFGHG emissions in the United Nations Framework Convention on Climate Change (UNFCCC) are obtained from the following website: https:// di.unfccc.int/ flex_ annex1 and https:// di.unfccc.int/ flex_ non_ annex1.FFGHG emissions in the Emissions Database for Global Atmospheric Research (EDGAR) are from EDGAR v4.2 37 (1970-1989) and EDGAR v7.0 38 (1990-2021).FFGHG CO 2 -equivalent emissions in US Environmental Protection Agency (EPA) were from Global Non-CO 2 Greenhouse Gas Emission Projections & Marginal Abatement Cost Analysis: Methodology Documentation 39 .Note that EPA only provided total PFC emissions instead of individual PFC emissions.All TD and BU data is accessed before 2023-11-10.

Figure 3 .
Figure 3. Summary of TD and BU FFGHG emissions in Japan from previous studies for (a) SF 6 , (b) NF 3 , (c) CF 4 , (d) C 2 F 6 , (e) C 3 F 8 , and (f) c-C 4 F 8 .Unit: tons per year (t yr −1 ).The detailed sources can be found in Supplementary Table3.FFGHG emissions in the United Nations Framework Convention on Climate Change (UNFCCC) are obtained from the following website: https:// di.unfccc.int/ flex_ annex1 and https:// di.unfccc.int/ flex_ non_ annex1.FFGHG emissions in the Emissions Database for Global Atmospheric Research (EDGAR) are from EDGAR v4.237 (1970-1989)  and EDGAR v7.038 (1990-2021).FFGHG CO 2 -equivalent emissions in US Environmental Protection Agency (EPA) were from Global Non-CO 2 Greenhouse Gas Emission Projections & Marginal Abatement Cost Analysis: Methodology Documentation 39 .Note that EPA only provided total PFC emissions instead of individual PFC emissions.All TD and BU data is accessed before 2023-11-10.