Article

Effective realization of abatement measures

can reduce HFC-23 emissions

https://doi.org/10.1038/s41586-024-07833-y Dominique Rust1,2,5, Martin K. Vollmer1 ✉, Stephan Henne1, Arnoud Frumau3,

Pim van den Bulk3, Arjan Hensen3, Kieran M. Stanley4, Renato Zenobi2,
Received: 19 October 2023
Lukas Emmenegger1 & Stefan Reimann1
Accepted: 16 July 2024

Published online: 21 August 2024

HFC-23 is a potent greenhouse gas, predominantly emitted as an undesired by-
Check for updates product during the synthesis and processing of HCFC-22 (ref. 1). Previously, the
Clean Development Mechanism and national efforts called for the implementation
of abatement technology for reducing HFC-23 emissions2,3. Nevertheless, between
2015 and 2019, a divergence was found between the global emissions derived from
atmospheric observations and those expected from reported abatement1,2. Primarily,
this points to insufficient implementation of abatement strategies2,4, calling for
independent verification of the emissions at the individual chemical facility level.
Here we use regional atmospheric observations and a new, deliberately released tracer
to quantify the HFC-23 emissions from an HCFC-22 and fluoropolymer production
facility, which is equipped with waste gas destruction technology. We find that our
inferred HFC-23/HCFC-22 emission factor of 0.19% (0.13–0.24%) broadly fits within
the emission factor considered practicable for abatement projects5,6. Extrapolation to
global HCFC-22 production underscores that the operation of appropriate destruction
technology has the potential to reduce global HFC-23 emissions by at least 84%
(69–100%) (14 (12–16) Gg yr−1). This reduction is equivalent to 17% CO2 emissions
from aviation in 2019 (ref. 7). We also demonstrate co-destruction of PFC-318,
another by-product and greenhouse gas. Our findings show the importance of the
2016 Kigali Amendment to the Montreal Protocol, which obligates parties to destroy
HFC-23 emissions from facilities manufacturing hydrochlorofluorocarbons and
hydrofluorocarbons “to the extent practicable” from 2020 onwards8.

HFC-23 (CHF3) is the longest-lived and the most potent greenhouse gas and Annex I under the United Nations Framework Convention on Cli-
among the hydrofluorocarbons (HFCs), with an atmospheric lifetime of mate Change (UNFCCC), including Turkey but excluding Israel) through
222 years (ref. 9) and a 100-year global warming potential (GWP100) of process optimization and incineration of waste gases. Through the
12,400 (ref. 9). In 2020, HFC-23 was the third largest (15%) contributor to UNFCCC Clean Development Mechanism (CDM), destruction capaci-
the total global radiative forcing of all HFCs1. Global HFC-23 emissions ties were installed at HCFC-22 production facilities in participating
in 2020 amounted to 16.5 (15.0–18.1; ranges refer to 95% confidence developing countries (defined here as Article 5 countries under the
bands) Gg yr−1 (ref. 1) or 205 (186–224) Tg CO2eq yr−1. The main known Montreal Protocol and non-Annex I under the UNFCCC, including
source to the atmosphere of this entirely anthropogenic substance is the Israel). Despite these efforts, global HFC-23 emissions could not be
formation and release as an undesired by-product in the synthesis of the reduced over the long term2,3,12,21. Emissions, inferred from atmospheric
hydrochlorofluorocarbon (HCFC) HCFC-22 (CHClF2) (refs. 2,3,10–14). observations (top-down), have recurrently increased over the past
Additional releases of HFC-23 may occur from further processing of decades to the highest on record estimated for the 3 years at the end of
HCFC-22 to tetrafluoroethylene (TFE, C2F4) and hexafluoropropylene the published data record (2018–2020) (ref. 1). Renewed agreements to
(HFP, C3F6) (refs. 1,13,15,16), largely used as feedstock monomers for reduce HFC-23 emissions and thus mitigate their environmental impact
the production of fluoropolymers such as polytetrafluoroethylene were made in the 2016 Kigali Amendment to the Montreal Protocol on
(PTFE, (C2F4)n). Comparatively smaller HFC-23 emissions occur from Substances that Deplete the Ozone Layer8,12. From 2020 onwards, par-
other sources14,17–20. ties to the amendment are obligated to ensure the destruction of HFC-23
Past endeavours to reduce HFC-23 by-product emissions from emissions from HCFC and HFC manufacturing facilities ‘to the extent
HCFC-22 synthesis were predominantly made in developed countries practicable’ by ‘using technologies approved by the parties’8. Remain-
(defined here as non-Article 5 countries under the Montreal Protocol ing HFC-23 emissions are to be reported annually8, including those
1
Laboratory for Air Pollution/Environmental Technology, Empa, Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, Switzerland. 2Department of Chemistry and
Applied Biosciences, ETH, Swiss Federal Institute of Technology, Zurich, Switzerland. 3Department of Environmental Modelling, Sensing and Analysis, TNO, Netherlands Organisation for
Applied Scientific Research, Petten, The Netherlands. 4Atmospheric Chemistry Research Group, School of Chemistry, University of Bristol, Bristol, UK. 5Present address: Atmospheric Chemistry
Research Group, School of Chemistry, University of Bristol, Bristol, UK. ✉e-mail: martin.vollmer@empa.ch

96 | Nature | Vol 633 | 5 September 2024

 a 2 HFC-161 (tracer) b
N = 164
m = 47 (36–55)
1
k = 35 (31–40)
150
0
135 HFC-23

HFC-23 (ppt)
85
100
35
8 PFC-318

6
50 R2 = 0.8
4 P < 0.01
Mole fraction (ppt)

2
450 HCFC-22 0 1 2 3

N = 164
350
m = 3.2 (2.3–3.9)
5 k = 2.0 (1.8–2.3)
250
HCFC-21
4.5

PFC-318 (ppt)
4
2.5

0.5
1.75 HFC-4310mee 3

1.00 R2 = 0.8
P < 0.01
0.25 2
20 Jun 27 Jun 04 Jul 11 Jul 18 Jul 25 Jul 1 Aug 8 Aug
2022 0 0.5 1.0
HFC-161 (ppt yr t–1)

Fig. 1 | Atmospheric abundances of industrially emitted waste gases emissions of HFC-161 (Methods). The annual emissions of HFC-23 and PFC-318
and the tracer HFC-161, recorded at the Cabauw tall tower station, are determined as the slopes (m) of the linear regressions using weighted total
the Netherlands, and tracer ratio interspecies correlation for HFC-23 and least squares (Methods). The intercepts (k) represent the background mole
PFC-318. a, Dry-air mole fractions in ppt (pmol mol−1) are shown for the duration fractions of HFC-23 and PFC-318. N is the total number of measurements
of the tracer release experiment (16 June to 7 August 2022). HFC-161 mole included in the calculation, R 2 is the coefficient of determination for the linear
fractions are normalized to a reference release rate during the experiment. regression. The correlations are highly significant (P < 0.01). Uncertainties
Pollution events showed high temporal correlation for all six substances (see Methods for derivation) are given at 95% confidence level. Uncertainties
emitted from the HCFC-22 and fluoropolymer production plant located at a are sometimes smaller than the symbol size. For better visualization, some of
distance of 22 km from the Cabauw tall tower. b, Scatterplot between dry-air the largest mole fractions are omitted from the plots.
mole fractions of HFC-23 and PFC-318 and the dry-air mole fractions per yearly

during interruptions of the destruction systems because of failures or suggested for the discrepancy were that the reported abatement targets
maintenance; that is, when HFC-23 emissions are largely unabated. For were not met or emission reductions were inaccurately quantified, par-
abatement projects supported by the Multilateral Fund for the Imple- ticularly in China and India2,4. Regional top-down estimates of emissions
mentation of the Montreal Protocol, HFC-23 destruction to the extent from these countries4,25,33 provided indirect partial evidence for these
practicable is assumed if a weight-based emission factor (EF) of HFC-23 hypotheses for recent years. Direct clarification of these hypotheses
emissions relative to produced HCFC-22 of at least 0.1% is achieved6. could be obtained by independent facility-scale top-down assessment
Many HCFC-22-producing countries17,18 ratified the Kigali Amend- of emissions, but this approach is currently lacking.
ment before 2020; however, some with large HCFC-22 production Here we quantify HFC-23 and other by-product and intermediate
adopted it only after 2020, when HFC-23 destruction should have product emissions from an HCFC-22 and fluoropolymer production
already been implemented. Nevertheless, for developed countries, it facility. By combining a deliberate tracer release with regional atmos-
was already reported in 2013 that most of the HCFC-22 manufacturers pheric observations, we introduce a new facility-selective method for
introduced measures to reduce HFC-23 emissions22, and concordantly top-down emission estimates, which is independent of the methods
this was reflected in their statistical national inventories (bottom-up), used to derive bottom-up emissions using factory-internal infor­mation.
as reported to the UNFCCC1. For most developing countries, partial Moreover, this method covers both the main stack and the fugitive23
abatement, or post-CDM cessation of abatement, was reported in emissions in an integrative approach. We calculate a facility-level
2017 (refs. 2,23–26). However, comprehensive abatement efforts mass-based EF of HFC-23 relative to the produced and processed HCFC-
were mandated and reported since 2016 from China27–29 and India30, 22. Applying this EF to global HCFC-22 production allows for deriving
which produce most of the global HCFC-22 (approximately 69% in 2018; global HFC-23 emissions based on practicable abatement. The same
ref. 18). Also, funding was awarded by the Executive Committee of the approach as used for HFC-23 was applied to assess the emissions of
Multilateral Fund for abatement in Argentina and Mexico (contributing the long-lived, potent greenhouse gas perfluorocarbon (PFC)−318
approximately 1% of the global HCFC-22 production in 2018; ref. 18), (c-C4F8, atmospheric lifetime = 3,200 years, GWP100 = 9,540) (ref. 9),
starting in 2022 (refs. 31,32). Based on these cumulative efforts, a large the ozone-depleting HCFC-21 (CHCl2F) and HCFC-22, and HFC-4310mee
reduction in global HFC-23 emissions is expected. (C5H2F10), all of which were also emitted from the same facility. Although
Despite previous efforts, a persistently increasing discrepancy unabated emissions of these substances contribute to radiative forcing,
between the expected low emissions from reported abatement meas- they have never been assessed top-down at the facility scale.
ures and the increasing global top-down emissions was found2,12. The Our method was applied to a production facility at Dordrecht, in
latter were estimated at 16.5 (15.0–18.1) Gg yr−1 for 2020, approximately the Netherlands. This is a main European HCFC-22 facility and the only
eight times as high as the expected emissions1. The main reasons one in our study area (Belgium and the Netherlands). At this facility,

Nature | Vol 633 | 5 September 2024 | 97

Article
the produced HCFC-22 is further processed to produce TFE and HFP, Table 1 | Emissions and EFs
which are then polymerized to PTFE, fluorinated ethylene propylene
(FEP) and chlorotrifluoroethylene vinylidene fluoride (FPM/FKM). Inferred Inferred EF Inferred Global
emissions (% by weight)a averaged annually
Emissions are abated through adjusted production processes and from the global averaged
on-site destruction of waste gases. However, besides potential residual Dordrecht emissions emissions
and fugitive emissions, additional release may result from thermal production (Gg yr−1 for (Gg yr−1 for
converter/destruction failures, which may occur during ongoing pro- facility 2019–2021) 2019–2020)d
(Mg yr−1)
duction and are currently not caught by stop protocols or sufficient
waste gas storage capacity13. HFC-23 47 0.19 1.9 16.9
(36–55) (0.13–0.24) (1.3–2.4)b (15.3–18.5)
The tracer experiment was conducted during a seven-week period
PFC-318 3.2 0.013 0.096 2.3
as part of a longer measurement campaign (19 November 2021 to 8
(2.3–3.9) (0.0084–0.017) (0.062–0.13)c (2.1–2.5)
August 2022) at the Cabauw atmospheric tall tower station (western
HCFC-22 62 0.25 2.5 356
Netherlands). HFC-161 (fluoroethane, C2H5F) was released as a new
(44–76) (0.16–0.32) (1.6–3.2)b (247–465)
gaseous tracer at the Dordrecht industrial facility at times of favourable
HCFC-21 1.5 6.0 0.061 –
wind conditions, allowing the tracer to reach the Cabauw observational (0.066–3.0) (0.14–12) × 10−3 (0.0014–0.12)b
site 22 km downwind. Air was sampled from the top of the tower (207 m
HFC-4310mee 1.3 – 1.2
above ground level) for approximately hourly quantitative analysis (0.59–1.7) (0.5–1.9)
of >60 target analytes, including HFC-161, using pre-concentration
Emissions and weight-based emission factors (EFs) of substance emissions relative to
gas chromatography–mass spectrometry (GC–MS)34,35. The obser- produced HCFC-22 from an HCFC-22 and fluoropolymer production facility at Dordrecht,
vations frequently captured HFC-161 plumes, and they were always the Netherlands. Global-scale emissions derived from the EFs of this study are presented in
accompanied by industrially emitted gases (Fig. 1a). Measurements comparison with other global emission estimates. Uncertainties are given at 95% confidence
that represent times of normal operation of the facility (that is, no level, assuming a normal distribution.
a
Relative to HCFC-22 feedstock production at the Dordrecht production plant: 25 (20–30) Gg yr−1.
failures of abatement technology; Methods) were used for emissions b
Relative to averaged total global HCFC-22 production as reported to the UNEP: 1,009 Gg yr−1
quantification. (2019–2021). cRelative to 97% of averaged global HCFC-22 feedstock production as
Based on our atmospheric measurements and the known quanti- reported to the UNEP: 758 Gg yr–1 (2019–2021). dModelled top-down, based on atmospheric
ties of the released tracer, we determined the industrial emissions of observations and a global 12-box model1,38,47.
HFC-23 and other substances using an interspecies correlation tracer
ratio method (Fig. 1b, Table 1, Methods and Extended Data Fig. 1). We cal-
culated HFC-23 emissions of 47 (36–55) Mg yr−1 from the Dordrecht facil- considered practicable5,6 and anticipated in future scenarios37, there
ity. Relative to the annual production of HCFC-22 of 25 (20–30) Gg yr−1, is potential for further reductions.
this results in an HFC-23/HCFC-22 EF of 0.19% (0.13–0.24%) and includes Using the tracer method, we also calculated the emissions and opera-
the HFC-23 emissions from the further processing of HCFC-22 to TFE tive abatement efficiencies of other substances involved in the PTFE/
and HFP. This EF is an order of magnitude lower than that of 1.5% con- FEP manufacturing process (Fig. 1 and Table 1). For PFC-318, which is
sidered achievable for production plants with optimized production emitted from the pyrolysis of HCFC-22 to produce TFE and HFP, we
but without waste gas destruction36; however, this is higher than the EF derived emissions of 3.2 (2.3–3.9) Mg yr−1. Similar to HFC-23, the inferred
expected from a destruction efficiency of 99.99% (ref. 17). Our inferred PFC-318/HCFC-22 EF of 0.013% (0.0084–0.017%) is notably lower than
EF broadly fits with the “practicable” EF of 0.1% that is required to be previously reported PFC-318 EFs at the global scale36,38,39 or of a reported
achieved by abatement projects (with destruction) recently funded by EF relative to TFE and HFP yield from optimized production condi-
the Multilateral Fund5,6. It also broadly fits with the EF of 0.08%, which tions40. We apply our EF to the global HCFC-22 feedstock production
was assumed for a future emissions scenario of HCFC-22 production, (average 758 Gg yr−1 for 2019–2021), corrected for the 97% (ref. 38)
accounting for effective destruction of generated HFC-23 and small share used for further processing to TFE and HFP. The resulting global
remaining emissions because of failures and maintenance of destruc- PFC-318 emissions of 0.096 (0.062–0.13) Gg yr−1 account for 4% of the
tion units37. global emissions derived from global PFC-318 observations38,41 (Fig. 2).
Scaling our derived EF with total global HCFC-22 production (average Under the assumption that PFC-318 is destroyed by the same route as
1,009 Gg yr−1 for 2019–2021) as reported to the United Nations Environ- HFC-23 (refs. 13,26), co-abatement of PFC-318 and HFC-23 would consid-
ment Programme (UNEP), global HFC-23 emissions would amount to erably contribute to avoiding greenhouse gas emissions. In a scenario
1.9 (1.3–2.4) Gg yr−1, which corresponds to 11% of the latest published in which more than 60% (Methods) of the global PFC-318 emissions are
top-down global HFC-23 emissions (average 16.9 (15.3 – 18.5) Gg yr−1 attributed to TFE and HFP production, the Dordrecht-inferred global
for 2019–2020) (ref. 1) (Fig. 2). PFC-318 emissions suggest the potential for globally avoided emissions
To our knowledge, this study provides the first independent experi- of more than 1.4 Gg yr−1 PFC-318 (>13 Tg CO2eq yr−1) in 2020.
mentally derived top-down-based evidence of practicable abatement. For HCFC-22 (atmospheric lifetime = 11.9 years, GWP100 = 1,760,
If this practice had been applied globally, and also accounting for ozone-depleting potential = 0.038) (refs. 9,42) the EF of 0.25% (0.16–
HFC-23 emissions of 0.77 (0.37–1.9) Gg yr−1 from other sources than 0.32%) from the present study is considerably lower than the sug-
the production and processing of HCFC-22 (Methods), it would likely gested EF without abatement (2–4%) (refs. 36,37). Also, for the first
have resulted in an average HFC-23 abatement of 14 (12–16) Gg yr−1 time, we report fully calibrated atmospheric measurements of HCFC-21
for 2019–2020, corresponding to an 84% (69–100%) reduction of the (atmospheric lifetime = 1.7 years, GWP100 = 148, ozone-depleting poten-
HFC-23 emissions inferred from global observations1. This reduction tial = 0.036) (refs. 9,42) in the peer-reviewed literature and provide
is equivalent to 170 (150–200) Tg CO2eq yr−1, which compares with 17% direct atmospheric evidence of this known intermediate in HCFC-22
of the carbon dioxide (CO2) emissions from pre-COVID-19 pandemic production. We determine an EF of 6.0 (0.14–12) × 10–3%, which is con-
aviation (1,036 Tg CO2 yr−1 for 2019)7. siderably lower than the reported range without abatement (2–4%)
Given that the abatement technology targeted by our case study (refs. 36,37), suggesting that this compound can be efficiently pre-
is long proven, it suggests a shortfall in global abatement in the past vented from being released to the atmosphere.
decades and the potential of an immediate global technological imple- Other substances that were observed and related to the Dordrecht
mentation, leading to greatly reduced emissions of HFC-23 (refs. 1,2,22). production plant were TFE (atmospheric lifetime = 1.1 days, GWP100 = < 1)
Furthermore, because the EF we determined is higher than those (ref. 9) and HFP (atmospheric lifetime = 4.9 days, GWP100 = <1) (ref. 9).

98 | Nature | Vol 633 | 5 September 2024

 Based on this study Global HCFC-22 production for feedstock use, currently accounting
Based on this study, applied to earlier years for about 80% of the total global HCFC-22 production, is anticipated
Bottom-up + reported abatement
Global top-down to increase by approximately 5 % yr−1 in the next 10 years (refs. 37,44).
Driven by the increasing global HFC-23 emissions in recent years, inter-
HFC-23 national bodies have refocused their attention on HFC-23 formation
20 250 and abatement1,8,14,17,18,23,37,45,46. Here we have shown a current potential
to substantially reduce global HFC-23 emissions, which, in the spirit of
Global emissions (Gg yr–1)

CO2eq emissions (Tg yr–1)

200 the Kigali Amendment, could be exploited rapidly. The failures in HFC-
15
23 abatement in the past years call for improved independent control
150
and verification of abatement practices and emissions23. Atmospheric
10
observations (top-down) coupled with a tracer release are an efficient
100
tool to assess sub-national to facility-scale emissions. It could also
5 be applied to HFC-23 sources from other fluorochemical (for exam-
50
ple, HFC-32) production and from electronics industries, as well as to
0 0
sources of other greenhouse gases.

2002 2006 2010 2014 2018 2022

Year Online content
PFC-318
Any methods, additional references, Nature Portfolio reporting summa-
ries, source data, extended data, supplementary information, acknowl-
Global emissions (Gg yr–1)

CO2eq emissions (Tg yr–1)

20 edgements, peer review information; details of author contributions
2
and competing interests; and statements of data and code availability
are available at https://doi.org/10.1038/s41586-024-07833-y.

1 10 1. Liang, Q. et al. in Scientific Assessment of Ozone Depletion: 2022. GAW Report

No. 278, Ch. 2 (World Meteorological Organization, 2022).
2. Stanley, K. M. et al. Increase in global emissions of HFC-23 despite near-total expected
reductions. Nat. Commun. 11, 397 (2020).
3. Miller, B. R. et al. HFC-23 (CHF3) emission trend response to HCFC-22 (CHClF2) production
0 0 and recent HFC-23 emission abatement measures. Atmos. Chem. Phys. 10, 7875–7890
2002 2006 2010 2014 2018 2022 (2010).
4. Park, H. et al. A rise in HFC-23 emissions from eastern Asia since 2015. Atmos. Chem.
Year
Phys. 23, 9401–9411 (2023).
5. UNEP. Key Aspects Related to HFC-23 By-product Control Technologies (Decision 83/67(d)).
Fig. 2 | Global HFC-23 and PFC-318 emissions based on an abatement
Report No. UNEP/OzL.Pro/ExCom/89/13 (United Nations Environment Programme, 2022).
scenario. Global emissions (red lines and red diamonds) are calculated based 6. UNEP. Report of Part II of the Eighty-Ninth Meeting of the Executive Committee. Report No.
on the weight-based emission factors (EFs) of this study for HFC-23 emissions/ UNEP/OzL.Pro/ExCom/89/16 (United Nations Environment Programme, 2022).
HCFC-22 production and PFC-318 emissions/HCFC-22 production from an 7. International Energy Agency. Aviation. IEA https://www.iea.org/reports/aviation (2022).
HCFC-22 and fluoropolymer production plant with destruction technology 8. UNEP. Amendment to the Montreal Protocol on Substances that Deplete the Ozone
Layer. United Nations Treaty Collection https://treaties.un.org/Pages/ViewDetails.
in operation. These EFs were determined for 2022 and multiplied with global aspx?src=IND&mtdsg_no=XXVII-2-f&chapter=27&clang=_en (2016).
HCFC-22 production (97% of HCFC-22 feedstock production for PFC-318) as 9. Myhre, G. et al. in Climate Change 2013: The Physical Science Basis. Contribution of
reported to the UNEP. Global emissions (black lines) derived from global Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate
atmospheric observations are shown for comparison1,38. Expected HFC-23 Change Ch. 8 (Cambridge Univ. Press, 2013).
10. Oram, D. E., Sturges, W. T., Penkett, S. A., McCulloch, A. & Fraser, P. J. Growth of fluoroform
emissions assuming full implementation of reported abatement are shown (CHF3, HFC-23) in the background atmosphere. Geophys. Res. Lett. 25, 35–38 (1998).
as the blue line2 . Uncertainties (grey and red shaded bands) are given at 95% 11. Miller, B. R. & Kuijpers, L. J. M. Projecting future HFC-23 emissions. Atmos. Chem. Phys. 11,
confidence level and are in some cases smaller than the symbol size. 13259–13267 (2011).
12. Simmonds, P. G. et al. Recent increases in the atmospheric growth rate and emissions of
HFC-23 (CHF3) and the link to HCFC-22 (CHClF2) production. Atmos. Chem. Phys. 18,
4153–4169 (2018).
13. Rodriguez, M. A., van Dril, T. & Gamboa Palacios, S. Decarbonisation Options for the
However, as our measurements were not calibrated on an absolute Dordrecht Chemical Cluster, MIDDEN: Manufacturing Industry Decarbonisation Data
scale, emissions could not be calculated. HFC-4310mee (atmospheric Exchange Network (PBL Netherlands Environmental Assessment Agency, 2021).
lifetime = 16.1 years, GWP100 = 1,650) (ref. 9), used as a processing aid, 14. TEAP. Volume 6: Response to Decision XXXIV/7: Strengthening Institutional Processes with
Respect to Information on HFC-23 By-Product Emissions, Report of the Technology and
was also detected and quantified based on fully calibrated measure- Economic Assessment Panel, Vol. 6 (United Nations Environment Programme, 2023).
ments (Fig. 1a and Table 1). 15. Ebnesajjad, S. Fluoroplastics, Vol. 2, Melt Processible Fluoropolymers, The Definitive
On a national level, we compare our estimated annual emissions to User’s Guide and Data Book 2nd edn, Ch. 6 (Elsevier, 2015).
16. Sung, D. J., Moon, D. J., Moon, S., Kim, J. & Hong, S. I. Catalytic pyrolysis of
the Dutch National Inventory Report (NIR) to UNFCCC (Extended Data chlorodifluoromethane over metal fluoride catalysts to produce tetrafluoroethylene.
Fig. 2), which attributes more than 99% of the reported HFC-23 emis- Appl. Catal. A Gen. 292, 130–137 (2005).
sions (2019–2021) to the Dordrecht facility. For HFC-23, our estimates 17. UNEP. Medical and Technical Options Committee (MCTOC) 2022 Assessment Report
(United Nations Environment Programme, 2022).
for the Dordrecht facility (47 (36–55) Mg yr−1), assuming no interrup- 18. TEAP. Volume 6: Assessment of the Funding Requirement for the Replenishment of the
tions in the abatement system, exceed those of the latest NIR (2021) Multilateral Fund for the Period 2021–2023, Report of the Technology and Economic
of 17.45 Mg yr−1. Our results are comparable to a top-down estimate of Assessment Panel (United Nations Environment Programme, 2021).
19. Pérez-Peña, M. P., Fisher, J. A., Hansen, C. & Kable, S. H. Assessing the atmospheric fate of
Dutch emissions for 2008–2010 based on observations at more distant trifluoroacetaldehyde (CF3CHO) and its potential as a new source of fluoroform (HFC-23)
European measurement sites43. For PFC-318, our estimate suggests using the AtChem2 box model. Environ. Sci. Atmos. 3, 1767–1777 (2023).
that the Dordrecht facility contributes approximately 50% to the 2021 20. McGillen, M. R. et al. Ozonolysis can produce long-lived greenhouse gases from
commercial refrigerants. Proc. Natl Acad. Sci. USA 120, e2312714120 (2023).
Dutch NIR emissions of an unspecified mix of PFCs (64.9 Gg CO2eq yr−1). 21. McCulloch, A. & Lindley, A. A. Global emissions of HFC-23 estimated to year 2015. Atmos.
Overall, our HFC-23 and PFC-318 emission estimates for the Dordrecht Environ. 41, 1560–1566 (2007).
22. US EPA. Global Mitigation of Non-CO2 Greenhouse Gases: 2010-2030 (United States
facility total 630 Gg CO2eq yr−1 or approximately 0.4% of the 2021 total
Environmental Protection Agency, Office of Atmospheric Programs, 2013).
Dutch greenhouse gas emissions (including Land Use, Land-Use Change 23. UNEP. Key Aspects Related to HFC-23 By-Product Control Technologies. Report No. UNEP/
and Forestry). OzL.Pro/ExCom/78/9, Vol. 1 (United Nations Environment Programme, 2017).

Nature | Vol 633 | 5 September 2024 | 99

Article
24. UNEP. Corrigendum Key Aspects Related to HFC-23 By-Product Control Technologies. 38. Mühle, J. et al. Global emissions of perfluorocyclobutane (PFC-318, c-C4F8) resulting from
Report No. UNEP/OzL.Pro/ExCom/78/9/Corr.1 (United Nations Environment Programme, the use of hydrochlorofluorocarbon-22 (HCFC-22) feedstock to produce
2017). polytetrafluoroethylene (PTFE) and related fluorochemicals. Atmos. Chem. Phys. 22,
25. Say, D. et al. Emissions of halocarbons from India inferred through atmospheric 3371–3378 (2022).
measurements. Atmos. Chem. Phys. 19, 9865–9885 (2019). 39. Harnisch, J. et al. 2006 IPCC Guidelines for National Greenhouse Gas Inventories Ch. 3,
26. UNEP. Key Aspects Related to HFC-23 By-product Control Technologies (Decision 78/5). Vol. 3 (Intergovernmental Panel of Climate Change, 2006).
Report No. UNEP/OzL.Pro/ExCom/79/48 (United Nations Environment Programme, 2017). 40. Murphy, P. M., Schleinitz, H. M. & Van Bramer, D. J. Synthesis of tetrafluoroethylene.
27. UNEP. Cost-Effective Options for Controlling HFC-23 By-Product Emissions (Decision US patent 5672784 (1997).
81/68(e)). Report No. UNEP/OzL.Pro/ExCom/82/68 (United Nations Environment 41. Mühle, J. et al. Perfluorocyclobutane (PFC-318, c-C4F8) in the global atmosphere. Atmos.
Programme, 2018). Chem. Phys. 19, 10335–10359 (2019).
28. UNEP. Corrigendum: Cost-effective Options for Controlling HFC-23 By-Product Emissions 42. Burkholder, J. B. & Hodnebrog, Ø. in Scientific Assessment of Ozone Depletion 2022. GAW
(Decision 81/68(e)). Report No. UNEP/OzL.Pro/ExCom/82/68/Corr.1 (United Nations Report No. 278 (World Meteorological Organization, 2022).
Environment Programme, 2018). 43. Keller, C. A. et al. Evidence for under-reported Western European emissions of the potent
29. UNEP. Report of the Sub-group on the Production Sector. Report No. UNEP/OzL.Pro/ greenhouse gas HFC-23. Geophys. Res. Lett. 38, L15808 (2011).
ExCom/84/74* (United Nations Environment Programme, 2019). 44. US EPA. Global Non-CO2 Greenhouse Gas Emission Projections & Mitigation Potential:
30. MEFCC. Ministry of Environment, Forest and Climate Change. Order No. F. No. 2015–2050 (United States Environmental Protection Agency, 2019).
10/29/2014-OC (Ozone Cell, 2016). 45. UNEP. Decisions adopted by the Thirty-Fourth Meeting of the Parties to the Montreal
31. UNEP. Key Aspects Related to HFC-23 By-product Control Technologies: Mexico (Decision Protocol on Substances that Deplete the Ozone Layer. Report No. UNEP/OzL.Pro.34/9/Add.1/
86/96). Report No. UNEP/OzL.Pro/ExCom/87/54 (United Nations Environment Programme, Rev.1 (United Nations Environment Programme, 2022).
2021). 46. UNEP. Report of the Thirty-Fourth Meeting of the Parties to the Montreal Protocol on
32. UNEP. Report of the Eighty-Seventh Meeting of the Executive Committee. Report No. UNEP/ Substances that Deplete the Ozone Layer Introduction. Report No. UNEP/OzL.Pro.34/9
OzL.Pro/ExCom/87/58 (United Nations Environment Programme, 2021). (United Nations Environment Programme, 2022).
33. Yi, L. et al. In situ observations of halogenated gases at the Shangdianzi background 47. Laube, J. C. et al. in Scientific Assessment of Ozone Depletion 2022. GAW Report No. 278,
station and emission estimates for northern China. Environ. Sci. Technol. 57, 7217–7229 Ch. 1 (World Meteorological Organization, 2022).
(2023).
34. Miller, B. R. et al. Medusa: a sample preconcentration and GC/MS detector system for Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
in situ measurements of atmospheric trace halocarbons, hydrocarbons, and sulfur published maps and institutional affiliations.
compounds. Anal. Chem. 80, 1536–1545 (2008).
35. Arnold, T. et al. Automated measurement of nitrogen trifluoride in ambient air. Anal. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this
Chem. 84, 4798–4804 (2012). article under a publishing agreement with the author(s) or other rightsholder(s); author
36. Dewi, R. G. et al. in 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse self-archiving of the accepted manuscript version of this article is solely governed by the
Gas Inventories Ch. 3, Vol. 3 (Intergovernmental Panel of Climate Change, 2019). terms of such publishing agreement and applicable law.
37. Daniel, J. S. et al. in Scientific Assessment of Ozone Depletion 2022. GAW Report No. 278,
Ch. 7 (World Meteorological Organization, 2022). © The Author(s), under exclusive licence to Springer Nature Limited 2024

100 | Nature | Vol 633 | 5 September 2024

Methods For HFC-161, for the period of the tracer campaign, a limit of detec-
tion (LOD) of 0.03 ppt was determined from the smallest integrated
Measurement site and in situ measurements chromatographic peaks, resulting in a minimum signal-to-noise level
From 19 November 2021 to 7 August 2022, in situ measurements of of around 2.
halogenated greenhouse gases were conducted at the Cabauw tall
tower station in the Netherlands (51.972 °N, 4.927 °E, altitude −0.7 m Tracer release
above sea level). The site is located on rural flat terrain and is most The tracer release experiment was conducted between 16 June 2022
sensitive to emissions from the southern part of the country with the and 7 August 2022, using approximately 30 kg of HFC-161 (fluoroeth-
largest urban agglomerations. ane, C2H5F, atmospheric lifetime = 66 days, GWP100 = 4) (ref. 9), CAS
Atmospheric measurements covered a suite of more than 60 halogen- 353-36-6, 99% (97–100%) purity (SynQuest Laboratories) as the tracer.
ated greenhouse gases, including substances of all classes addressed The release setup was installed inside a temperature-controlled trailer
by the Montreal Protocol. In situ measurements of ambient air were that was parked in the parking lot of the industrial facility at Dordrecht
conducted with a Medusa pre-concentration unit, coupled to gas chro- at 51.820 °N, 4.724 °E, 4 m above sea level, with the main industrial area
matography (GC; Agilent 6890N) and mass spectrometry (MS; Agilent lying at a distance of 300–600 m East (90°) of the trailer. HFC-161 was
5975)34,35. Air samples were drawn from a 207-m tower height to the released through ¼″ tubing at the top of a 10-m high pole attached
temperature-controlled station building that hosted the instrument, to the trailer. The 60-m high industrial stack, which is the designated
using a 0.38″ inner diameter (ID) sampling line (Synflex). Two-litre release path for the facility off-gases, was at approximately 400 m
samples were collected for 20 min and analysed in a measurement distance from the trailer location. The HFC-161 release was regulated
routine of two consecutive air samples bracketed by a whole-air work- through a remotely controlled high-precision thermal mass flow con-
ing standard, each lasting 75 min. During tracer release events, this troller (MFC, 0–6 l min−1 for air, Vögtlin Instruments). It was calibrated
routine was temporarily altered to fewer standard measurements. for HFC-161 using a volumeter (DC-2 flow calibrator, DryCal, BIOS),
Chromatographic separation was obtained with a CP-PoraBOND Q which in turn was calibrated by the Swiss Federal Institute of Metrol-
column (0.32 mm ID × 25 m; 5 μm film thickness; Agilent). Analytes ogy (METAS). A calibration factor of 0.43 was determined for HFC-161
were detected by quadrupole electron ionization MS in the selected against air. During the experiments, various flow rates were set (0.41–
ion mode. Weekly laboratory air and instrument blank measurements 2.56 l min−1 HFC-161) to adjust for various meteorological conditions.
showed the absence of local contamination, system blank and carryover During the tracer release campaign, HFC-161 was released only during
for the substances of this study. periods of favourable advection from Dordrecht to the Cabauw station,
HFC-161 was identified on the GC–MS system based on a dilution located 22 km Northeast (40°) of the industrial facility. For this, plume
of the pure substance (SynQuest Laboratories) with synthetic air. For forecasts, generated by the NAME model50 (v.7.2), and wind forecasts
the duration of the experiment, HFC-161 was detected based on frag- from the Windfinder (https://www.windfinder.com) and the Wetter-
ments with mass/charge (m/z) 47 ([CH2CH2F]+) as the target mass, and zentrale (https://www.wetterzentrale.de/) websites, were monitored.
m/z 33 ([CH2F]+) and m/z 27 ([CH2CH]+) as qualifying masses. Also, to To dynamically track the tracer and industrial waste gas plume and to
increase the MS sensitivity and baseline quality, the ion fragments of investigate potential influence from sources other than the Dordrecht
most other compounds in the acquisition region of HFC-161 were omit- industrial site, a 2-day flask sampling experiment was conducted. Dur-
ted. Abundances are reported as dry-air mole fractions in ppt (parts per ing favourable direction of plumes to the Cabauw station, 22 flask sam-
trillion, pmol mol−1) for all substances except for TFE and HFP, which ples were collected at various locations up- and downwind of Dordrecht.
are reported relative to a working standard. Observations of HFC-23, Results from their analyses and correlation to the tracer confirmed the
PFC-318 and HFC-4310mee, and HCFC-22 are reported on the primary absence of relevant other sources of our targeted substances.
calibration scales of the Scripps Institution of Oceanography (SIO),
SIO-07, SIO-14 and SIO-05, respectively48, whereas those of HCFC-21 Emission calculations
and HFC-161 are reported on the Swiss Federal Laboratories for Materi- Emissions from Dordrecht were calculated by interspecies correlation,
als Science and Technology (Empa)-2013 primary calibration scales49. based on the tracer and target substance pollution events captured
The synthetic standards used to define the primary calibration scales at the Cabauw station. Emissions were calculated for the target sub-
were linked to the working standards through additional hierarchical stances HFC-23, PFC-318, HCFC-22, HCFC-21 and HFC-4310mee, for
levels of whole-air standards. These, and the working standards, are which calibrated air measurements were made. In general, for emission
whole-air samples collected into 34 l internally electropolished stain- calculations by interspecies correlation, observed pollution events of
less steel tanks (65 bar, Essex Industries) using oil-less diving compres- target gases are referenced against the observed pollution events of a
sors. Because HFC-161 is absent in ambient air, it was spiked into the tracer gas and linked by the known emitted quantity of the tracer gas.
standards to a target abundance of 0.5 ppt, to track the sensitivities of Interspecies correlation calculation is based on several assumptions,
the MS and to propagate the measurements through the calibration which were met for the present experiment; the reference tracer and
chain of the various standards. the industrial waste gases originated from the same temporal source
The following uncertainties are attributed to the measurements and approximately the same spatial source. They were transported
of the main analytes of this study. The instrumental precisions are to the Cabauw station (the receptor) without relevant contributions
determined from the measurements of the whole-air working stand- from other sources, as was found from our 2-day flask experiment
ards. These were 1.5% for HFC-23, 2% for PFC-318, 0.5% for HCFC-22, and in line with the fact that the Dordrecht facility is the only HCFC-22
3% for HCFC-21, 7% for HFC-4310mee and 5% for HFC-161 (95% con- production plant in the area. For HCFC-22, contributions from legacy
fidence level). Another uncertainty is attributed to the propagation applications are regarded as negligible given the rural land use in the
of the primary calibration, generally using two additional standards air flow path. Contributions from sources distributed on a larger scale
and is approximated the same as the instrument precision for each are intercepted by the general background air mole fraction. For the
of the propagation steps. All these uncertainties are independent tracer HFC-161, the background air was free of this substance support-
and lead to an overall random uncertainty of 2.6%, 3.5%, 0.9%, 5.2%, ing our assumption of the deliberate release being the only source. For
12% and 8.7%, for these substances, respectively. The accuracy of all the main substances of the study, atmospheric loss during transport
the primary calibration scale is estimated at 3% for HFC-23, PFC- (0.7–8.5 h) was considered negligible given their atmospheric lifetimes.
318, HCFC-22 and HFC-4310mee, and 10% for HCFC-21 and HFC-161 Also, the transport distance from Dordrecht to the Cabauw station is
(95% confidence level). considered sufficiently long enough to neglect the small horizontal
Article
(approximately 400 m) and vertical (50 m, tracer release level versus with 1,000 iterations of non-parametric resampling. This provided
industrial stack height) difference in release location, such that we larger uncertainties than derived from the regression itself and thus
consider the two emerging plumes to have been well-mixed on arrival at gives a rather conservative estimate of the final regression uncertainty.
the Cabauw station. The potential fugitive emissions of the production For the final uncertainties, the measurement calibration scale accu-
facility were located even closer in height to the tracer release position. racy and the propagation of calibration scale accuracy (see section
Emission calculation by interspecies correlation is based on the fol- ‘Measurement site and in situ measurements’) were added. Final uncer-
lowing underlying equation (1): tainties of calculated emissions are given at the 95% confidence level.

Mt χt, i − χb,t Non-HCFC-22-related HFC-23 emissions

Et, i = Er, i (1)
Mr χr, i − χb,r For our estimate of HFC-23 emissions from other sources than HCFC-22
production and processing, we reference the Technology and Eco-
where χ (in unit ppt) is the measured enhancement of the target analyte nomic Assessment Panel (TEAP) report14, which assesses these sources.
(t) and of the reference tracer (r) above background (b) at different Sources that we take into consideration are HFC-23 formation as a
times (i), assuming constant background mole fractions. M (in g mol−1) by-product during some production pathways of HFC-32 and HFC-
are the respective molecular weights. Er and Et are the emissions of the 125, or other controlled substances, its use as feedstock, its use for
reference and target tracer, respectively. Instead of estimating the very low-temperature refrigeration, speciality fire suppression, the
target emissions individually, we obtain their average through linear semiconductor and electronics industry or etching. We conservatively
regression (Extended Data Fig. 1), rearranging equation (1): assume that all HFC-23 generated from these sources is emitted to the
atmosphere without abatement. The TEAP identified large uncertain-
Mr(χr, i − χb,r ) ties in global estimates of HFC-23 generation and thus emissions from
χb,t + E¯t = χt, i
ME (2) these processes other than HCFC-22 and TFE/HFP production. Using
⏟
⏟ ⏟
t r, i
m   yi
k
xi their uncertainty estimates and applying Monte Carlo error propa-
gation, we derive global emissions of 0.77 (0.37–1.9, 95% confidence
Here we set χb,r = 0, because HFC-161 was not detectable in background range) Gg yr−1. Additional atmospheric production of HFC-23 is sus-
air. The average emission of a target substance (Ēt), was then determined pected from the degradation of hydrohaloolefins19,20. However, on a
as the slope, m, of the given linear expression and determined by global level, this may become relevant only in the future and was thus
weighted total least squares regression analysis through all included not considered here.
measurements of all detected pollution events. Similarly, the atmos-
pheric background mole fraction level of the target substance is given PFC-318 emission estimates
as the intercept k of the regression. Our estimate for the potential of avoided global emissions of PFC-318,
For the calculations, all measurements of the tracer release peri- with functional abatement technology at HCFC-22 facilities, is based
ods, excluding abatement failure periods (see section ‘Discussion of on the assumption that more than 60% of the current global PFC-318
robustness of results’), were included. Furthermore, to account for the emissions is derived from TFE and HFP production. This assumption
time-lagged arrival of the tracer at the Cabauw station, measurements is derived from the strong correlation of global PFC-318 emissions
after the end of tracer release periods and at least three subsequent with global HCFC-22 production for feedstock use38 and co-location
measurements without a tracer signal were also included. For measure- of large PFC-318 emissions with potential production facilities of TFE
ments with undetectable tracer mole fraction, the tracer mole fraction and HFP, whereas no major PFC-318 emissions were found in countries
was set to LOD/2. with large semiconductor production41. If we assume that the PFC-318
To account for the various tracer release rates during the experi- emissions for Eastern China41 originate from the HCFC-22 processed to
ment (see section ‘Tracer release’), we conducted a normalization to TFE and HFP38,41, then an average EF of approximately 0.2% for PFC-318
a reference tracer emission Er,i (in unit t yr−1) and assigned the latter from TFE/HFP production can be derived18. If applied to global HCFC-
to each measured mole fraction by linear interpolation. For this, the 22 feedstock for TFE and HFP18,38, this accounts to around 60% of the
air transport times were taken into account, which were calculated global PFC-318 emissions. We take this as a conservative lower estimate
from the distance between the tracer source and the Cabauw station for calculating the potential of avoiding global PFC-318 emissions.
(21.7 km) and the mean wind speed over the different meteorological
measurement heights at the Cabauw station. Discussion of robustness of results
For weighted total least squares regression analyses, different Owing to the limited time frame of the tracer release experiment of
sources of error were taken into account. Where appropriate, uncer- 1.5 months, the emission estimates are based on a limited number of
tainties were propagated by standard Gaussian error propagation. captured pollution events. However, in consultation with the Dordrecht
Uncertainties associated with the amount of tracer released are the industry, we consider our results representative under regular opera-
volumeter calibration uncertainty determined by METAS (0.3%, 68% tion conditions, with continuous production at an average rate and
confidence level), the uncertainty of the calibration curve determined optimal operation and destruction conditions of the facility during
with the used MFCs for HFC-161 (3%, 68% confidence level), and the this period, excluding, for example, converter failure events. Pollution
purity of the tracer gas (99% (97–100%), uncertainty 1.5%, 68% con- events were captured at the Cabauw station during each episode of wind
fidence level). Instead of the relatively attributed air measurement arriving from the direction of Dordrecht, suggesting the continuous
precision, an absolute precision was assigned to every measurement release of industrial gases from the facility. Therefore, we consider the
for use in the regression analysis. For the target analytes, the absolute experiment time frame representative of the long-term emission rate
precision was determined as the mean relative precision (see section from this facility under the operation conditions encountered, allowing
‘Measurement site and in situ measurements’) for the duration of the for the upscaling of the emissions results to 1 year.
measurement campaign, assigned to the mean atmospheric back- Ten abatement failure events were reported by the production facil-
ground mole fraction level of a substance. For HFC-161, the relative ity for the time of the tracer release experiment. However, all but two
precision was assigned to the LOD/2. of these were not captured by our measurements due to unfavourable
To account for additional methodological uncertainty and atmos- wind directions towards the station.
pheric variability, the uncertainty range of the emission result was This is considered an insufficient representation of failure events
determined by bootstrapping the weighted total least square regression and precludes us from assessing their effects on the overall HFC-23
EF. Nevertheless, a rough estimate is made based on the information rates are available from the Zenodo data repository (https://doi.
received from the production facility and from literature data. Based org/10.5281/zenodo.11208167). HCFC-22 inventory data are avail-
on this, we estimate failing abatement at 2% of the HCFC-22 production able from the Ozone Secretariat (https://ozone.unep.org/). Dutch
time and that all generated HFC-23 (at an assumed generation rate of National Inventory Reports to the United Nations Framework Con-
1.5% of HCFC-22 production14,36) is vented. Under these conditions, the vention on Climate Change are available from https://unfccc.int/
overall annual EF for the facility changes from 0.19% (reported in this ghg-inventories-annex-i-parties/2023.
paper, assuming fully functional abatement) to 0.21%. Applying this
EF to global HCFC-22 production, HFC-23 emissions would be approxi- 48. Prinn, R. G. et al. History of chemically and radiatively important atmospheric gases from
the Advanced Global Atmospheric Gases Experiment (AGAGE). Earth Syst. Sci. Data 10,
mately 10% higher than our reported value of 1.9 Gg yr−1. 985–1018 (2018).
In contrast to the high temporal correlation of pollution events 49. Vollmer, M. K., Reimann, S., Hill, M. & Brunner, D. First observations of the fourth generation
observed at the Cabauw station, the observed mole fraction ratios synthetic halocarbons HFC-1234yf, HFC-1234ze(E), and HCFC-1233zd(E) in the atmosphere.
Environ. Sci. Technol. 49, 2703–2708 (2015).
varied considerably between pairs of substances and over time. This 50. Jones, A., Thomson, D., Hort, M. & Devenish, B. The U.K. Met Office’s Next-Generation
also leads to the relatively large scatter in the target versus tracer Atmospheric Dispersion Model, NAME III. In Air Pollution Modeling and Its Application XVII
(eds Borrego, C. & Norman, A. L.) 580–589 (Springer, 2007).
plots and partly unexplained variance (approximately 20% for HFC-
23). A reason for this observation might be substance-specific vari-
able emission rates potentially caused by variabilities in the process Acknowledgements Support for the measurement campaign at the Cabauw tall tower and
for the tracer experiment was provided by staff from Empa, TNO and the University of Bristol.
steps and destruction rates. Temporally fluctuating generation rates The Royal Netherlands Meteorological Institute (KNMI) hosted the measurement campaign at
of HFC-23 have previously been observed at other facilities26. As for the Cabauw tall tower and provided meteorological observations. The industrial production
the variable ratios to the tracer, we explored the potential violation company at Dordrecht approved and facilitated the tracer experiment on their premises and
provided background information on internal production processes. The Dienst Centraal
of our assumption of a well-mixing with the industrial waste gases, Milieubeheer Rijnmond (DCMR) permitted the release of the tracer. The UK Met Office provided
especially at times with fast transport from Dordrecht to the Cabauw daily plume forecasts. The Swiss Federal Institute of Metrology (METAS) supported the MFC
station. The main influencing meteorological factor may be direc- calibrations for HFC-161. The research of Empa was funded by the Swiss National Science
Foundation (SNSF) (grant no. 200020_175921). Funding for TNO was provided by the Ministry
tional wind shear between our 10-m release height and the stack emis- of Economic Affairs and Climate Policy for the maintenance of greenhouse gas measurement
sions at 60 m, leading to initially different transport directions and, systems at the Cabauw tower in 2022.
hence, non-concurrent plume interceptions at the Cabauw station.
Author contributions D.R., M.K.V., A.F. and P.v.d.B. made the measurements. A.F., P.v.d.B. and
However, this influence is considered non-systematic and, hence, A.H. provided campaign support. D.R., S.H., M.K.V. and S.R. processed the data. M.K.V., D.R.,
only adds to the uncertainty in our emission estimates. Moreover, S.R., S.H., A.F., A.H. and K.M.S. designed the experiment. R.Z. and L.E. provided guidance on
similar correlations among the industrial substances are found, as the paper structure and review. D.R., M.K.V., S.H. and S.R. wrote the paper with contributions
from all the co-authors.
with the tracer, supporting our assumption that the variable ratios
to the tracer are caused by variable emission rates rather than by Competing interests The authors declare no competing interests.
deficient mixing.
Additional information
Supplementary information The online version contains supplementary material available at
https://doi.org/10.1038/s41586-024-07833-y.
Data availability Correspondence and requests for materials should be addressed to Martin K. Vollmer.
Peer review information Nature thanks Nick Campbell, Deborah Ottinger and Guus Velders for
Atmospheric measurement data acquired at the Cabauw station their contribution to the peer review of this work. Peer reviewer reports are available.
for all substances subject to this paper as well as tracer emission Reprints and permissions information is available at http://www.nature.com/reprints.
Article

Extended Data Fig. 1 | Tracer ratio interspecies correlation. Scatterplot total number of measurements included in the calculation, R 2 is the coefficient
between the dry-air mole fractions of the target substances (y-axis) and the of determination for the linear regression. The correlations are highly
dry-air mole fractions per yearly emissions of the tracer HFC-161 (see equation 2) significant (p < 0.01). Uncertainties are given at 95 % confidence level.
(x-axis). The annual target substance emissions are determined as the slopes Uncertainties are sometimes smaller than the symbol size. For better
(m) of the linear regressions using weighted total least squares. The intercepts visualization some of the largest mole fractions are omitted from the plots.
(k) represent the background mole fractions of the target substances. N is the
Extended Data Fig. 2 | Dutch national HFC-23 and PFC-318 emissions.
Experimentally inferred HFC-23 and PFC-318 emissions (red diamonds) from an
HCFC-22 and fluoropolymer production facility at Dordrecht. Uncertainties
are given at 95 % confidence level. The Dutch national bottom-up inventory
of HFC-23 and aggregate PFC emissions (black lines) as annually reported
to the UNFCCC is shown for comparison. HFC-23 emissions of the Benelux
(BLX; Belgium, the Netherlands, Luxembourg) countries for 2008–2010,
derived top-down by Keller et al.43, are shown in purple and turquoise.