We use measurements of dry-air atmospheric mole fractions from two global monitoring networks, the Advanced Global Atmospheric Gases Experiment (AGAGE, ) and the United States National Oceanic and Atmospheric Administration (NOAA) Global Greenhouse Gas Reference Network . Measurements from AGAGE and NOAA stations in the remote atmosphere were used separately to estimate global emissions. Measurements from AGAGE stations provide regional emissions estimates for Europe, Australia and east Asia and NOAA stations for the United States. Figure  shows the locations of the measurement stations, and further information is summarised in Tables S1 and S2.

AGAGE HCFC-141b measurements are reported on the Scripps Institution of Oceanography (SIO) 2005 calibration scale. The Medusa GC–MS instruments in Table S1 are cryogenic pre-concentration systems coupled with a gas chromatograph (GC, Agilent) and quadrupole mass selective detector (MSD) . The ADS–GC–MS is an adsorption–desorption system with a gas chromatograph and mass spectrometer . AGAGE in situ atmospheric measurements were made with these systems approximately every 2 h (Medusa GC–MS) or 4 h (ADS–GC–MS). Paired flask samples were collected at the Taunus observatory (Table S1) and analysed on a GC-quadrupole MSD . Before in situ measurements were available (1994 in the Northern Hemisphere and 1998 in the Southern Hemisphere), global emission estimates are based on archived air samples (historic air samples collected and stored). Archived air samples from the Cape Grim Air Archive (CGAA, collected 1978–2009) for the Southern Hemisphere were measured in 2011 using Medusa GC–MS technology, also on the SIO-2005 scale . Several CGAA and 126 archived air samples taken at Trinidad Head and other locations in the Northern Hemisphere between 1973–2016 were measured at the Scripps Institution of Oceanography using Medusa GC–MS technology . These archived air samples can therefore be easily integrated with the later AGAGE in situ measurements.

NOAA estimates of the global mean, remote atmospheric abundance of HCFC-141b considered here are derived from measurements of air samples pressurised into paired stainless-steel flasks that have been collected approximately weekly at eight remote sites since the earlier 1990s (see Table S2). The flasks are shipped to Boulder for analysis on customised GC–MS instruments . For deriving continental US emissions from atmospheric measurements, additional flask samples are regularly collected from tall towers (100–400 m a.g.l.; a single flask per sample, typically) and as profiles from aircraft (6 to 12 flasks collected at regular altitude intervals in a profile up to 8 km a.s.l.) at 17 profiling locations .

Regional emission estimates for east Asia were derived using four inverse methods: the Bristol Markov chain Monte Carlo (MCMC) inversion, Sect. ; InTEM, Sect.  ; EBRIS, Sect. ; and FLEXPART-MIT, Sect. . Regional emissions for northwest Europe (NW Europe) and Australia were derived using InTEM, Sect. . Regional emissions for the contiguous United States were derived using the NOAA framework, Sect. . Inverse methods were run with independent choices to a priori emissions, statistical models, transport set-ups and treatment of measurement data sets, in instances where multiple estimates were performed for the same region.

As expected from the Montreal Protocol-mandated HCFC phase-out schedule, global emissions of HCFC-141b had been declining since 2012 . However, since 2017, an increase in emissions is evident from recent changes in the distribution and global mean mole fraction of HCFC-141b. The growth rate of the global mole fractions had continuously slowed since 2012, and the global mole fractions started to decline during 2017 (Fig. a and b for AGAGE and NOAA, respectively). Such a slow-down and decline are expected as emissions decrease and HCFC-141b present in the atmosphere is destroyed by reactions with the OH radical and photolysis in the stratosphere. Since 2018, however, the rate of decline in global mole fraction slowed and, as of 2019, annual mean growth rates are positive again. This increase is likely being driven by increased emissions in the Northern Hemisphere, as the increase in the growth rate in the Northern Hemisphere leads that in the Southern Hemisphere in recent years, just as it did in earlier periods when the global HCFC-141b growth rate and emissions were rapidly increasing (e.g. 1992–1996 and 2009–2012). As of 2019 there is a growing difference between the observed mole fraction in the Northern Hemisphere and Southern Hemisphere (mean difference of 1.8 ppt in 2018 and 2.2 ppt in 2021).

Using AGAGE and NOAA measurements, we show that emissions of HCFC-141b started to increase again in 2018, despite the reported reduction in global dispersive production and consumption expected from the Montreal Protocol control schedules (Figs.  and a). Aggregated emissions from 2017 to 2021 have increased by an additional 4.4 ± 1.9 (AGAGE) or 7.5 ± 0.5 Gg (NOAA), an average linear increase at a rate of 1.1 ± 0.2  Gg yr-1 per year (AGAGE) or 1.6 ± 0.1 Gg yr-1 per year (NOAA). This assumes that all uncertainty is only due to the random error component of the emissions estimates (i.e. systematic uncertainty causes an offset in emissions and does not impact year-to-year variability). Note that Fig.  shows the uncertainty due to the combined random and systematic error components.

To understand this reversal, we estimate global emissions of HCFC-141b using reported consumption data by adapting the approach taken by , as detailed in Sect. . We estimate the prompt release fraction to be 24 (17–31) % and 37 (36–39) % of the annual consumption for Article 5 and non-Article 5 countries, respectively (68 % uncertainty interval). Our estimated annual bank release fraction is 3.6 (3.0–4.2) % from the remaining bank and 2.8 (1.9–3.9) % for Article 5 and non-Article 5 countries, respectively, and is assumed to be unchanged since 1989. used an aggregated HCFC-141b prompt release fraction of 34.5 ± 4.0 %, along with a bank release fraction of 2.2 %. Our estimated released fractions thus broadly agree with that used by and that previously estimated for closed foams containing CFC-11 , which is 25 %–35 % for the prompt release fraction and 1.5 %–4.2 % for the bank release fraction. Using the reported consumption and estimated release fraction, we derive the respective emissions using the top-down-informed bank model. Under the constraints of the top-down-informed bank model, we are able to estimate the necessary consumption needed to produce the observed top-down emissions in 2021, which has not yet been fully reported. This uses the reported consumption until 2020 and the estimated emissions rate from the bank to estimate the needed 2021 consumption to reproduce the top-down emissions estimate. This analysis, under the constraints of the model, suggests that consumption likely rose between 2020 and 2021, from 39 Gg yr-1 in 2020 to 67 (28–95) Gg yr-1 in 2021. This consumption estimate for 2021 is very uncertain due to the uncertainty in the top-down-derived emissions, and so a decline in global emissions, and thus decline in consumption, cannot be excluded. Increasing consumption between 2017–2020 is not possible in this model, due the constraints imposed by the reported production and bank behaviour. An alternative conclusion that could be drawn from this analysis, under the assumption that reported consumption should decrease in 2021, is that the emissions rate from the bank has increased in more recent years, meaning that the inferred emissions rate, which is constant in time, is no longer representative and cannot adequately model emissions.

Figure a shows a comparison of the top-down atmospheric measurement-derived emissions, using the 12-box model (and without using information about reported consumption), and the emissions estimated using the top-down-informed bank model. Although these estimates are unique, they are not independent due to both top-down estimates using the same methodology to derive emissions using AGAGE and NOAA measurements and the dependence of the top-down-informed bank model on the estimates using AGAGE measurements. Figure b shows a breakdown of the mean emissions from non-Article 5 and Article 5 countries under the top-down-informed bank model using reported consumption data and assumptions inherent in the model, where 2021 consumption has been estimated. This imposes that all consumption in 2021 (over a maximum of 1 Gg yr-1) must come from Article 5 countries (see Sect. ).

The timing of this increase in HCFC-141b emissions is nearly coincident with the rapid post-2018 global decline in CFC-11 emissions , which may suggest that the enhanced apparent use and production of CFC-11 during 2012–2018, which was likely used for foam production, may have transitioned to HCFC-141b instead of transitioning to newer replacements. An explanation for the recent rise in global HCFC-141b emissions is, however, further complicated by a lack of understanding of the time-dependent changes in emissions from the existing HCFC-141b bank. suggest that HCFC-141b emissions in China are expected to peak in 2025–2027, 15 years after peak consumption in China. This is attributed to the release of HCFC-141b from foams during the disposal of insulated refrigerators and electric water heaters at their end of life around 15 years after peak consumption. If such a pattern predicted in China is applied to all developing countries, peak emissions from all developing countries would be expected around 2026–2027, 15 years after their peak consumption in 2011–2012, and could perhaps explain the observed emissions increase. However, a study in Lahore, Pakistan, showed that HCFC-141b emissions due to waste and disposal of HCFC-141b-containing refrigerators fell between 2005 and 2013 . It may therefore not be appropriate to extrapolate the bottom-up-predicted bank emission trends in China to other developing countries, and there is likely considerable uncertainty and variation in the life cycle of appliances containing HCFC-141b foam and their disposal practices.

The top-down-informed bank model neglects emissions from feedstock and other so-called non-dispersive production. A typical emissions factor associated with fluorochemical production is around 4 % . If applied to global feedstock production, additional feedstock-related emissions would be 0.5 Gg yr-1 between 2017–2020, not enough to explain the global increase, with no increasing trend. Emissions factors from individual production facilities may vary considerably (0.1 %–20 %, 95 % uncertainty, ). A universal feedstock production leakage rate of 20 % in 2020 compared to 0.1 % in 2017 would be needed to explain the observed global increase in emissions. This would result in an additional 3.0 Gg yr-1 of emissions, compared to a mean global increase in the NOAA and AGAGE estimates of 3.0 ± 1.2 Gg yr-1 during the same period. A sustained rapid deterioration in feedstock production losses since 2017 seems an unlikely scenario to explain the global increase in emissions.

In the following section we turn to regional emissions estimates of HCFC-141b, where observations are available, to further diagnose the driver behind the global rise.

Atmospheric measurement-based emissions estimates for east Asia are based on four different inverse methods using measurements from Gosan station, South Korea (see Sect. ), and are combined into a single estimate using Monte Carlo sampling. These results provide emissions estimates for South Korea, North Korea and two regions that we denote eastern China (which includes the provinces of Anhui, Beijing, Hebei, Jiangsu, Liaoning, Shandong, Shanghai, Tianjin and Zhejiang) and western Japan (which includes the regions Chūgoku, Kansai, Kyūshū and Okinawa, and Shikoku), as measurements at Gosan have little sensitivity to emissions in western China or eastern Japan.

Emissions of HCFC-141b from eastern China show an increase in emissions from 3.7 (1.8–4.8) Gg yr-1 in 2008 to 6.8 (3.7–9.4) Gg yr-1 in 2020 and an increase of 1.0 (-2.9, 3.8) Gg yr-1 between 2017–2020. The 2017–2020 increase in eastern China could account for around a third of the global increase over the same period (3.0 ± 1.2 Gg yr-1, NOAA and AGAGE mean), but given the large uncertainties this could explain all or none of the global increase. The mean emissions of the inverse frameworks (Fig. ) generally agree well with bottom-up inventory estimates for China scaled down to the eastern China region by population or gross domestic product from emissions from the whole of China (the range is shown by the orange shading in Fig. ), given the uncertainties in the top-down modelling and the uncertainties likely present in the bottom-up model (which were not given in the study). The top-down-derived emissions trend is also supported by the projected emissions of , although emissions projected by were expected to peak in 2018. If the bottom-up-predicted emissions by and come to fruition, HCFC-141b emissions from China will continue to rise until the mid-2020s. Our results presented here for eastern China contrast with top-down estimates presented for the whole of China in previous studies , which suggest a continual fall in emissions.

Emissions from South Korea (around 1–3 Gg yr-1) are smaller than from eastern China, and it is uncertain whether they are increasing or decreasing. Emissions from western Japan and North Korea are even smaller (generally less than around 1 Gg yr-1). Japan, in contrast to other east Asian countries, legislates the removal and destruction of HCFC refrigerants and HCFCs in foams within appliances through the Home Appliance Recycling Act .

Emissions for northwestern Europe (Belgium, France, Germany, Ireland, Luxembourg, the Netherlands and the United Kingdom) were estimated using the InTEM inversion system, based on measurements from Mace Head in Ireland, starting in 1994, with additional measurements from Jungfraujoch in Switzerland from 2008, Monte Cimone in Italy from 2012, Tacolneston in the UK from 2012 and Taunus in Germany from 2013. NW Europe sees a sharp fall in emissions after 2003, Fig. a, from 3.6 ± 0.9 in 2003 to 2.2 ± 1.0 Gg yr-1 in 2004 and then 1.3 ± 0.9 Gg yr-1 in 2005, with timing consistent with the phase-down of HCFC production in those countries. Emissions have steadily fallen since, reaching 0.4 ± 0.1 Gg yr-1 in 2020. Emissions from individual countries are presented in the Supplement.

Emissions in NW Europe peak coincidentally with high consumption in developed countries. The European Union's Regulation (EC) 1005/2009 (and similar legislation in the UK) requires that appliance foam insulation controlled under the Montreal Protocol is recovered at disposal to be recycled or properly destroyed. Therefore we expect relatively low end-of-life emissions in NW Europe, in comparison to the emissions rates in China, and our results provide evidence that such legislation has, likely, been effective.

Top-down emissions estimates for the contiguous United States from 2015 to 2020 (Fig. a) are derived from measurements made by NOAA's North American sampling network (see Sect.  and ) using two atmospheric transport models and the NOAA inversion framework (Sect. , ). Taking an average of the two top-down estimates for the United States, emissions likely increased slightly between 2015 and 2017, with 5.5 ± 0.6 in 2015 and 6.4 ± 0.7 Gg yr-1 in 2017. The HYSPLIT-NAM model estimate suggests that emissions remain unchanged in 2018 at 6.3 ± 0.6 and fall to 5.3 ± 1.0 Gg yr-1 in 2020. Bottom-up emissions estimates by the U.S. Environmental Protection Agency over the same period (Fig. a) suggest that emissions fell from 8.6 in 2015 to 7.1 Gg yr-1 in 2017 to 6.7 Gg yr-1 in 2020. The reason for the discrepancy in the trend between the top-down and bottom-up estimates in 2015–2016 is unclear; however both show only small changes in emissions between 2017–2020. The NOAA top-down estimates are 10 %–40 % smaller than the U.S. EPA bottom-up estimates (assuming that the non-contiguous states and territories are not responsible for the discrepancy).

Consumption of HCFCs in the United States from 2002 to 2003 fell to around half and has continued to decline. The U.S. EPA emission estimate suggests peak emissions many years after a peak consumption, different from what the NW European top-down emissions show and more similar to bottom-up projections for China. The reason for a slowing in the decline of emissions since 2017 is not fully clear, but the bottom-up model suggests that it may be due to post-disposal emissions from polyurethane foams used for domestic refrigerator and freezer insulation, occurring during the 26 years following disposal . This assumes that domestic refrigerators and freezers have a lifetime of 14 years and that the leakage post-disposal (2 %) is higher than that during the appliance lifetime (0.25 %). In the United States, a gradual increase in HCFC-141b emissions due to the disposal of domestic refrigerants following a peak in consumption may be occurring, much like that predicted to occur in China, offsetting reductions in other banks. Unlike the European Union, which mandates destruction of HCFC-141b within refrigerator foams, the U.S. EPA's Responsible Appliance Disposal (RAD) Program is voluntary. This may explain different US and European HCFC-141b emission patterns, despite similar consumption patterns.

Top-down HCFC-141b emissions estimates using InTEM show that emissions in Australia have remained largely constant from 2000 to 2021, perhaps showing a slight decline (Fig. b).

Australia has not produced or exported HCFC-141b, and therefore HCFC-141b consumption is governed by imports, assuming that no significant stockpiling for later use occurred. Australian HCFC-141b imports/consumption commenced in the early 1990s, reaching 0.8–0.9 Gg in 1999 and then declining to zero by 2012, well ahead of the mandated HCFC phase-out schedule . From 1991 to 2012, Australian HCFC-141b imports totalled 7.8 Gg.

Assuming that emissions have therefore only come from the bank, and using the emissions factor derived for non-Article 5 countries from Sect. , we derive a bottom-up emissions estimate based on consumption in Australia. The bottom-up model likely underestimates the emissions release rate because prompt emissions, which we assume to include emissions during installation and immediate usage, have not been included. The top-down and bottom-up estimates for Australia generally agree, given the large uncertainties. Australia has no laws to mandate the destruction of HCFCs within appliance-insulating foams at disposal. However, there is no obvious increase in HCFC-141b emissions in Australia in recent years due to an increase in the rate of emissions from foam-containing appliances when they reach their end of life (within uncertainties).