DCE is manufactured industrially via the direct chlorination of ethene or via its oxychlorination with hydrogen chloride. Estimated annual DCE production data were compiled biennially by Nolan Sherry Associates (NSA) over the period from 2002 to 2020 (Table ). Data from NSA were reported in the most recent Technology and Economic Assessment Panel (TEAP) report to the parties of the Montreal Protocol (TEAP, 2022) and have been utilized in a range of recent scientific papers (e.g. Chipperfield et al., 2018; Claxton et al., 2020). Analysis by NSA makes use of their extensive database of halocarbon production and production capacities, industry data, and public reports, and is refined through industry dialogue. NSA's analysis of DCE production includes assessment of downstream products (VCM and PVC), accounting for several specific industry and market factors and trade movements. This includes the fact that VCM production may not always occur via the “ethylene route”, which uses DCE at a rolling ratio of 1.6 units of DCE to VCM, but the “acetylene route”, which involves the direct production of VCM from acetylene's reaction with hydrogen chloride (i.e. no DCE involved). The latter approach is prevalent in, for example, China, meaning that Chinese DCE production is relatively modest compared with other major global economies. Note that at the country level, some of the data available to NSA are proprietary in nature and confidential. On this basis and to aid the discussion and presentation, data have been aggregated into 13 broader geographical regions for which we discuss production and emissions. These regions cover all of the world's major industrialized zones, and their boundaries (Fig. ) are based on the region definitions used in Phase 2 of the Hemispheric Transport of Air Pollution (HTAP) project (e.g. Huang et al., 2017).

Evident from the data in Table  is that DCE is produced in large quantities (∼ 52 000 Gg in 2020) and that global production increased by ∼ 29 % between 2002 and 2020. US/Canada; Europe; and S, E, and SE Asia (regions 1–5) were estimated to account for ∼ 86 % of world production in 2020. Noting that the end product of DCE's principal industrial use (i.e. PVC) is closely followed and reported on by both business performance analysts and environmentalists, the production data from NSA are estimated to be accurate to within around ± 5 %. Although estimates of DCE production in peer-reviewed literature are scarce, some independent figures exist with which to compare. For instance, data cited by the US Environmental Protection Agency (EPA) placed the annual volume of DCE produced in the USA at 12 750 Gg in 2011 (EPA, 2020). Interpolating between years on either side (recalling that production data from NSA were provided biennially), the corresponding US production from NSA is 13 145 Gg (2011), i.e. in close agreement with the US EPA figure (within ∼ 3 %). An assessment around health aspects of DCE exposure placed European production at more than 10 000 Ggyr-1 (Cherrie et al., 2011), consistent with the NSA data in Table . Few estimates of DCE production in Asia exist in the peer-reviewed literature. However, an estimate of Chinese DCE production in the year 2010 of 2708 Gg (Chinabaogao, 2012) is very similar to that for China from NSA in the same year (2700 Gg).

The demand for DCE in both producing and non-producing countries was evaluated from trade data. Net imports (gross imports minus gross exports) were calculated for a total of ∼ 150 countries over our study period (2002–2020) using publicly available trade statistics accessed via the online UN Comtrade Database (https://comtradeplus.un.org/, last access: 22 November 2024). Assuming that global imports should equal global exports in a given year, net imports should sum to zero across the globe. However, due to imperfections in reported trade data (known to afflict many commodities besides DCE), this was not found to be the case. Although the imbalance was small (average of ∼ 7 % over our study period, expressed as the difference between gross imports and exports) compared with the large production volumes of DCE, we elected to reconcile the trade data using the method of Zou et al. (2023). Briefly, where a record of DCE trade is recorded by the importer but not the exporter (or vice versa), the missing trade is filled in. Where records match but the trade quantities differ, the larger of the two was adopted (Zou et al., 2023). This approach balances global DCE trade and prevents errors in trade statistics from confounding our subsequent analysis.

In principle, emissions of DCE may arise during its (1) production, (2) use as a feedstock, (3) transportation, and (4) any emissive uses (e.g. as a solvent). Items 1–3 represent fugitive emissions that may arise from, for example, the operation and maintenance of chemical plants, along with bulk storage and other industrial processes where unintended leakage can occur. In a fully explicit bottom-up inventory, production emissions may be calculated as the product of annual DCE production and a suitable emission factor. Similarly, feedstock use emissions, which are additional and additive, may be calculated from the quantity of DCE used as feedstock and a further emission factor (e.g. TEAP, 2022). However, although DCE is principally used as a feedstock in the manufacture of VCM, with some assessments placing this use at > 98 % (CEH, 2023), the precise quantity and how this may have varied over time and across regions is unknown. Note, even if 98 % of DCE use is in producing VCM, this does not imply that the remaining 2 % is used in emissive applications. This is because DCE also finds use as an intermediate in the production of other chemicals, including ethyleneamines and other chlorinated solvents (Sect. 1; TEAP, 2022). Analysis by NSA suggests that these two sectors contribute of the order of 600–800 Ggyr-1 of DCE feedstock use. In our idealized framework for calculating DCE emissions, we assume that 100 % of DCE use (consumption) is as feedstock.

The annual total DCE emission per country in year t was calculated using Eq. (). 1Emission(t)=P(t)α1+C(t)α2+I(t)α3

The first term on the right denotes production emissions calculated from the temporally varying production (P) data provided by NSA. There are 36 producing countries in the NSA database to which this term applies. The second term on the right denotes feedstock use emissions calculated based on consumption (C) data. Consumption (production + net imports) was calculated for ∼ 150 countries in total. The DCE-producing countries dominate global consumption, with non-producers accounting for less than 0.6 % of the global total. Production and consumption data in Tables  and  are aggregated regional totals obtained from country-level analysis. The third term on the right of Eq. () represents fugitive emissions during the supply chain. We have elected to apply these in the country of import, and they are calculated from gross import (I) data.

In Eq. (), emission factors for production, feedstock use, and supply chain emissions are denoted by α1, α2, and α3, respectively. Tight emission controls throughout the whole DCE production cycle and its supply chain, especially in more developed countries, are expected to occur to minimize loss of useful material, to control costs in an extremely competitive industry, and also for possible legal compliance. For DCE, all α values in developed countries are thus expected to be small and likely to lie towards the lower end of the plausible ranges reported in the literature for other gases (TEAP, 2022). For context, fugitive emissions from the production of other halocarbons have typically been estimated at ∼ 0.5 % on production (IPCC/TEAP, 2005). However, due to differences in plant operations and regulatory requirements in different world regions, regional differences in emission factors are likely. For DCE, we examined a range of emission factors around the above value, varying α1 between 0.1 % and 0.6 % for developed countries. For developing countries, approximated as those operating under Article 5 (A5) of the Montreal Protocol, we assume a multiplier of 3. The different scenarios are labelled according to their developed-country α1 emission factor (see Table ). Fugitive emissions from feedstock uses are generally expected to be lower than those from production (TEAP, 2022). We assumed a fixed feedstock emission factor of α2 = 0.1 %, representing the “low” estimate reported by TEAP (2022). Note, where consumption is negative (i.e. exports exceed production plus imports), we assume no feedstock use emission. Similarly, we adopt a fixed supply chain emissions factor of α3 = 0.1 %, representing the low estimate for distribution emissions reported by TEAP (2022). The resulting estimated range of total fugitive emissions was 146–594 Ggyr-1 in 2020 (Table ).

Many modern DCE-producing plants are integrated on-site with VCM/PVC production (Cherrie et al., 2011). If the processes are seamless, the distinction between fugitive losses from production and fugitive losses from feedstock use may be less clear-cut compared with other gases. Thus, an alternative framework (not adopted) might be to consider a single emission factor, applied to production, that encapsulates all possible leakage over DCE's internal lifetime within a plant. As we elected to consider global trade and, hence, use consumption in conjunction with production, it was necessary to treat the two terms separately. We note that our overarching goal is to examine the impact of DCE on stratospheric ozone using a global model calibrated to reproduce tropospheric observations of DCE. The overall magnitude and location of DCE emissions is thus important, but the detail of the fugitive source is secondary.

For inclusion in the CTM, the calculated biennial DCE emissions (Table ) were linearly interpolated to give annual records over the 19-year study period (2002–2020). The emissions were aggregated onto a global 0.5° × 0.5° grid using the country mask of Perrette (2023). This mask was developed for the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP). The within-country DCE distribution was assumed to follow that of ethene. The reaction of ethene with chlorine is the main route by which industrial DCE production occurs; thus, ethene should be a reasonable proxy. Anthropogenic ethene emissions (year 2014) from the “industrial combustion and processes” sector were taken from the gridded (0.5° × 0.5°) datasets produced for CMIP6 (Feng et al., 2020). Figure  illustrates the resulting surface DCE emission distribution and time series of regional and global emissions for “scenario sc05”, i.e. with α1 = 0.5 % (non-A5)/1.5%(A5). In this example case, Asia (sum of regions 3–5) accounts for ∼ 48 % of global emissions in 2020. For other Cl-VSLSs, Asian emissions have been assessed to dominate the global anthropogenic source, such as the estimated ∼ 90 % contribution of Asia to global CH2Cl2 emissions reported by Claxton et al. (2020). For DCE, the approach and information described above give rise to a more even distribution of emissions between continents, including a sizeable source outside of Asia.

The temporally varying DCE emissions described in Sect.  were included in the TOMCAT/SLIMCAT 3-D CTM (Chipperfield, 2006). The model is well evaluated and has been widely used to study the atmospheric budget and impacts of a range of trace gases, including VSLSs (e.g. Claxton et al., 2020; Hossaini et al., 2019). The offline model (hereafter “TOMCAT”) is forced by meteorological fields from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 dataset (Hersbach et al., 2020). The model uses the Prather (1986) scheme for tracer advection and the Holtslag and Boville (1993) scheme to represent boundary layer mixing. For convective transport, the model here utilized archived convective mass fluxes from ERA5. This approach was previously evaluated for ERA-Interim by Feng et al. (2011) and found to perform well. All simulations were performed at a 2.8° × 2.8° (T42 Gaussian grid) horizontal resolution and with 60 hybrid sigma-pressure (σ-p) levels extending from the surface to ∼ 60 km.

To test model–measurement agreement under different DCE scenarios and, thus, provide constraint on the global DCE source strength, a reduced chemistry configuration of the CTM was used. In this configuration, the concentration of the hydroxyl radical (OH) was prescribed from the monthly varying climatology produced for the Atmospheric Tracer Transport Model Intercomparison Project (TransCom) on methane (Patra et al., 2011). Six simulations were performed in which the model DCE tracer was controlled by differing surface emissions (Table ), along with transport, and oxidation by OH. The rate constant for the DCE + OH reaction (kOH) was specified from the latest Jet Propulsion Laboratory (JPL) kinetics evaluation (Burkholder et al., 2020): kOH = 1.14×10-11exp(-1150/T). The reaction was assumed to proceed as follows: DCE+OH→2Cl + products. There is no current recommendation given for DCE absorption cross sections and, like other chlorinated VSLSs, photolysis is expected to be a minor tropospheric sink (Carpenter et al., 2014) and so was not considered. Deposition was not included as a DCE sink but is expected to be relatively minor.

To diagnose DCE's contribution to stratospheric chlorine and its effect on ozone, we used a CTM configuration with “full” stratospheric chemistry. This configuration includes a treatment of all major processes that control polar and extra-polar ozone (e.g. Chipperfield et al., 2018). Full-chemistry simulations considered only the most likely range of DCE emissions (determined in Sect. ), and the DCE tracer was treated as described above. In the troposphere, chlorine atoms released from DCE oxidation will quickly speciate into HCl, the dominant inorganic chlorine (Cly) reservoir (e.g. via CH4+Cl→HCl + products). Tropospheric removal of HCl and other Cly species (HCl, HOCl, and ClONO2) via wet and dry deposition was calculated with the standard tropospheric TOMCAT routines (Giannakopoulos et al., 1999; Monks et al., 2017). Henry's law data used to calculate wet deposition rates were taken from Sander (2023). Full-chemistry model runs were spun up for 10 years and then run over the full 19-year analysis period (2002–2020). The stratospheric chlorine and ozone responses to DCE emissions were diagnosed from paired simulations (i.e. comparing runs with DCE emissions to a no-DCE control run). Temporally varying surface mixing ratios of long-lived gases (e.g. halogenated ODSs, N2O, and CH4) were prescribed from the data given in WMO (2018).

We have utilized a range of aircraft measurements of DCE to evaluate the model and to provide constraints on the global DCE source strength. The HIAPER Pole-to-Pole Observations (HIPPO) mission (e.g. Wofsy et al., 2011) was conducted between 2009 and 2011 and involved measurements of a wide range of trace gases predominately over the Pacific from aboard the National Science Foundation (NSF) Gulfstream V aircraft. Sampling extended over a large latitude range from roughly the North Pole to the Antarctic Ocean and from the surface to ∼ 14 km. The mission comprised five campaigns conducted in different seasons: HIPPO-1 (January 2009), HIPPO-2 (November 2009), HIPPO-3 (March–April 2010), HIPPO-4 (June 2011), and HIPPO-5 (August–September 2011). Measurements of DCE were obtained by the University of Miami based on the analysis of whole air samples collected in flask samples during each campaign.

The more recent NASA Atmospheric Tomography (ATom) mission was conducted between 2016 and 2018 and also involved extensive measurements of trace gases over a wide latitude range spanning both hemispheres (near pole to pole), including over the Pacific and the Atlantic. Measurements were obtained up to an altitude of ∼ 12 km aboard the NASA DC-8 aircraft over four campaigns covering different seasons: ATom-1 (July–August 2016), ATom-2 (January–February 2017), ATom-3 (September–October 2017), and ATom-4 (April–May 2018). An overview of the ATom mission, including some scientific highlights, is given in Thompson et al. (2022). For this study, we have used DCE measurements obtained by NOAA from air samples collected with the Programmable Flask Package (PFP) whole air sampler. The DCE measurement precision is around 1 % on average (for mole fractions of 1–2 ppt), and the detection limit is < 1 ppt. Regarding sample stability, NOAA ATom sampling was conducted via pressurization into glass flasks; note that there has been no indication of systematic growth or destruction of DCE in glass flasks over time within the measurement precision. Like HIPPO, the spatial coverage of ATom makes it an especially useful dataset with which to evaluate global models (e.g. the representation of hemispheric gradients). In our subsequent analysis, data from both HIPPO and ATom have been aggregated into nine latitude bins (> 80° N, 60–80° N, 40–60° N, 20–40° N, 0–20° N, 0–20° S, 20–40° S, 40–60° S, and < 60° S). The mean, standard deviation, and number of data points for each bin (boundary layer only, < 3 km) are given in Table .

To examine model performance over Asia, we also used measurements obtained during the 2016 Korea–United States Air Quality Study (KORUS-AQ) mission. The mission took place in the months of May and June and included 20 research flights of the NASA DC-8 aircraft based from Osan Air Base, approximately 50 km south of Seoul, South Korea (Crawford et al., 2021). Measurements during this campaign targeted local urban sources of photochemical pollutants; therefore, the air sampled differs considerably from that sampled during HIPPO and ATom. Measurements of DCE and other gases were obtained from whole air samples collected by the University of California, Irvine (UCI) from the surface up to an altitude of ∼ 11 km. The measurement detection limit was 0.1 pptv, and the measurement precision was 5 % (Simpson et al., 2020). The UCI group has run extensive tests on the stability of compounds in their canisters in the time between sampling and analysis, and DCE is stable within the canisters. Highly elevated levels of DCE and other volatile organic compounds were reported from KORUS-AQ, especially in air masses originating from China (Simpson et al., 2020). For the analysis here, the DCE mole fractions were aggregated into eight altitude bins (0–8 km) of 1 km depth. Sampling was extensive, and the number of measurements in each bin ranged from 94 to 1323.

To further evaluate model performance over eastern Asia, we used surface DCE measurements made by the University of East Anglia (UEA) at Bachok Marine Research Station, which is located in northeast Malaysia (6.009° N, 102.425° E), and at two sites in Taiwan: (1) Fuguei Cape, on the northern Taiwanese coast (25.297° N, 121.538° E), and (2) Hengchun, on the southern Taiwanese coast (22.0547° N, 120.6995° E). Measurements of DCE and other Cl-VSLSs at each location have been reported by Oram et al. (2017) and show elevated levels with respect to data obtained in other world regions. The same study provides full details of the sampling and instrument method. Briefly, measurements were obtained from whole air samples collected between 2014 and 2020. Sampling is seasonal and primarily targeted at observing emissions from East Asia during the northeast monsoon. Sampling at Bachok occurred in the months of November to March, while sampling at Fuguei Cape (2014 and 2016 onwards) and Hengchun (2013 and 2015 only) mostly occurred in March to May (Table S1 in the Supplement). The latter two sites in Taiwan are combined to give a single time series in our subsequent analysis. UEA samples were collected in silco-treated cylinders (stainless steel with inner surface coated with fused silica, Restek), and no issues with sample loss have been noticed. All collected samples were analysed by gas chromatography–mass spectrometry (GC-MS) at UEA, with a typical precision of 1 %–3 %.

Compared with other Cl-VSLSs, scientific interest in DCE from an ozone depletion perspective is relatively new. As such, an international standard calibration scale has not yet been universally adopted across measuring groups. Historically, the scales among the labs considered in this study have not differed by more than 10 %–30 % for gases similar to DCE. However, in the absence of any formal assessment of calibration scale differences, an informal intercomparison for DCE was performed for this work. Background atmospheric DCE mole fractions from two remote sites (Barrow and Samoa observatories), where both the NOAA and UCI groups sample, were compared (2017–2023). This intercomparison revealed an average offset of up to ∼ 30 % (UCI relatively high and NOAA relatively low), i.e. at the upper end of the above range. While this comparison is limited in scope and will require further effort to refine (beyond the scope of this paper), this uncertainty is highlighted in the ensuing discussion.

Observed boundary layer DCE mole fractions (< 3 km), as averages from all deployments of HIPPO (2009–2011) and ATom (2016–2018), are shown in Fig. a and b. The measurement data were compiled into nine latitude bins (Sect. ), and the mean (± 1 SD) of each bin is shown. Recall that both missions sampled air in various seasons of the year; thus, the SD variability includes seasonal effects. Measurements from both missions show a strong hemispheric gradient, with mean NH mole fractions of ∼ 15 ppt at latitudes greater than 40° N and ∼ 4 ppt or less in the Southern Hemisphere (SH). For comparison, DCE mole fractions of 7.8 (± 1.5) ppt have previously been reported in background air in the NH based on aircraft measurements from the 2006 NASA INTEX-B mission which sampled around the Gulf of Mexico and over the western Pacific off the coast of the USA (Barletta et al., 2009; Singh et al., 2009). The data in Fig.  show that variability is large within the NH, where most DCE production (and emission) is located and where the number of measurements is relatively high. For example, the relative standard deviation of the 20–40° N bin is ∼ 40 % (HIPPO) and ∼ 50 % (ATom). For comparison, the relative standard deviation in the SH is smaller, e.g. ∼ 21 % (HIPPO) and ∼ 12 % (ATom) for the < 60° S bin.

Figure  also includes the TOMCAT modelled DCE abundance using the different emission scenarios, i.e. assuming different α1 emission factors (see Sect.  and Table ). With scenarios sc01 to sc03, the model exhibits a substantial low bias and is unable to reproduce the magnitude of DCE in either hemisphere or the observed hemispheric gradient from each mission. Better model–measurement agreement for both missions is obtained from runs with scenarios sc04 to sc06, shown by the shaded regions in Fig. . The statistics describing model–measurement differences in Table  are based on the central sc05 case, i.e. α1 = 0.5%(non-A5  countries)/1.5%(A5). Under this scenario, the mean bias (model minus observation) varies by latitude and ranges from near zero up to 4.8 ppt. Although underestimating mean DCE observed in the NH during HIPPO, the model falls within the measurement variability, and better agreement is obtained for the comparisons with ATom. Generally, model–measurement biases here are difficult to interpret and could, in part, reflect the model OH field (affecting the DCE lifetime) and/or transport processes; they do not necessarily point to an under- or overestimation of local emissions. Additionally, as for some other VSLSs, differences in calibration scales between measurement groups (recall the discussion in Sect. ) could be a confounding factor (see also Roozitalab et al., 2024, for a more detailed discussed). Importantly, model–measurement agreement is generally good in the tropics (± 20° N/S), the region most relevant for diagnosing transport to the stratosphere, throughout the vertical profile (see Fig. S1 in the Supplement). The low abundance of DCE at SH high latitudes is also well captured.

A comparison of the modelled vertical profile of DCE to that observed during KORUS-AQ (2016) is shown in Fig. c. Compared with HIPPO and ATom, far larger observed levels of DCE are apparent at lower altitudes (up to 2.5 ppbv; not shown in Fig. c), along with very large variability (see filled grey circles). To accommodate the latter, binned measurement data in Fig. c show the median, as opposed to the mean, and the horizontal axis is capped at 80 ppt. (Note that a version of the aforementioned comparison but with the means is shown in Fig. S2.) The maximum observed value of > 2.5 ppb occurred in the 0–1 km bin for air originating from China. A similar maximum of > 2.4 ppb was measured in air originating from an industrial facility in South Korea (Simpson et al., 2020), although we note that a global-scale model is not expected to capture these most extreme values, which included targeted source sampling. Although the model (median ∼ 53 ppt under scenario sc05) underestimates the observations (median ∼ 69 ppt) in the lowest bin (0–1 km), it is evident from comparing Fig. c with Fig. a and b that the model shows significantly elevated DCE in this region. Elevated emissions over East Asia are also a clear feature in Fig. a. Above 1 km, there is very close agreement between the model and measurements using scenario sc05. A full quantitative comparison is given in Table S2. A previous in-depth analysis of the KORUS-AQ data highlighted that DCE was especially elevated in air originating from China (Simpson et al., 2020). Similarly, during the 2006 NASA INTEX-B mission, analysis of air sampled in polluted plumes from Asia (although especially China) revealed substantially elevated DCE relative to background air and to plumes from the USA (Barletta et al., 2009). Indeed, DCE was used as a tracer of air from China during both INTEX-B and KORUS-AQ.

The modelled DCE abundance is compared to the available surface measurements from Bachok and Taiwan in Fig. . As above, the measurements are characterized by large variability, with DCE exceeding 150 ppt at Bachok and 300 ppt at Taiwan on some days. While the model is not expected to capture the most extreme values, the central tendency of the observations appears to be reasonably well captured at both sites under scenario sc05 (see Table S1). Both the measurements and model show especially elevated levels of DCE at Taiwan (median values > 50 ppt in each year of sampling) with respect to levels observed during HIPPO and ATom, suggesting strong regional or local sources and sampling of relatively polluted air in both the Bachok and Taiwanese samples. Samples collected at Bachok, where the model captures the shape of the seasonal cycle well, predominately occur when the site experiences northeasterly winds, and observations are thus likely impacted by emissions occurring from mainland China (Oram et al., 2017). This seasonality is discussed in, for example, Oram et al. (2017) and has a large dynamical component. Briefly, strong northeasterly (NE) winds that form in the NH winter transport polluted air masses from continental East Asia deep into the tropics. The prevailing NE winds may also be strengthened during “cold-surge” events. The effect of such events on various tracers have been observed, including at other sites in Malaysia (e.g. Ashfold et al., 2015, 2017). Similarly, owing to close proximity, measurements at Taiwan are expected to be influenced by emissions from mainland China. Although not exhaustive, the comparisons discussed above (along with those for KORUS-AQ) suggest that the model has a reasonable representation of regional emissions in East and Southeast Asia. Note that the modelled DCE abundance at Taiwan (only) was found to exhibit a strong sensitivity to the choice of model vertical level sampled. Therefore, model data in Fig. b represent the average of the two model levels closest to the surface.

Although the regional variability in atmospheric DCE measurements can be large, the remote atmospheric survey sampling represented by the HIPPO and ATom missions, along with the other comparisons, allows for some constraint on the global DCE source (given our assumptions concerning the DCE emission distribution). Assuming scenario sc05, α1 = 0.5%(non-A5  countries)/1.5%(A5  countries), shown above to provide reasonable model–measurement agreement, we estimate a global DCE source of 349 (± 61) Ggyr-1 in 2002, rising to 505 (± 90) Ggyr-1 in 2020 (i.e. an increase of ∼ 45 %). The mean growth rate of global DCE emissions over this period is ∼ 9.1 Ggyr-2. There are very few estimates of global or regional DCE emissions in the literature with which to compare these findings. Using a simple tracer ratio method, Wang et al. (2014) estimated Chinese DCE emissions of 121.6 (± 89) Ggyr-1 in 2010. For the same year and again utilizing scenario sc05 (range from sc04 to sc06), our inventory produces significantly lower Chinese DCE emissions of ∼ 44 (36–52) Ggyr-1. Oram et al. (2017) estimated Chinese DCE emissions of 203 (± 9) Ggyr-1 for 2015 based on measurements obtained in Taiwan and Malaysia. Our estimate for Chinese emissions in 2016 is ∼ 60 (49–71) Ggyr-1 and is, thus, substantially lower. However, it should be emphasized that inferred emissions from observed tracer correlations, as in the above studies, are based on several assumptions and subject to large uncertainty. For instance, emissions occurring in nearby regions may confound geographical attribution. Although still lower, our inventory provides better agreement with the above estimates if emissions from, for instance, nearby Taiwan are included. We estimate the sum of DCE emissions from China and Taiwan to be 89 (73–106) Gg in 2010 and 107 (87–127) Gg in 2016. A summary of these comparisons of Chinese emissions is given in Table S3. Further factors that may confound the comparison of these different estimates are the month of measurement and the sampling frequency used to infer tracer ratios. For example, note that the DCE measurements at Bachok reported by Oram et al. (2017), an extended time series of which is shown in Fig. a, focus on non-summer months when DCE is relatively abundant at that site.

While the above results provide some constraint on the global DCE source required to reproduce atmospheric observations, some care is needed when interpreting the findings from an emission process standpoint. Our analysis has approximated fugitive emissions arising from production and feedstock use and distribution, assuming that feedstock use accounts for all consumption in every country. However, DCE has known solvent uses (not explicitly accounted for) which may be up to 100 % emissive in the absence of solvent capture and careful disposal. If a non-negligible amount of the observed atmospheric abundance of DCE stems from solvent use, the contribution from fugitive losses could be overestimated. We anticipate that DCE solvent use is most prevalent in developing countries, where it is relatively cheap vs. alternatives and is readily available and where concerns over its toxicity may not yet have resulted in restrictions on its use. Given these uncertainties, we do not overinterpret our findings from an emission process or sectoral standpoint but rather, with more confidence, highlight the overall magnitude of emissions that provide good agreement with the available measurement data in Fig. . In subsequent sections, we present all model quantities assuming scenario sc05 emissions, with reported uncertainties from the sc04 and sc06 cases.

The modelled tropospheric distribution of DCE is shown in Fig.  for the years 2002 and 2020. At the surface (Fig. a and b) DCE exhibits large spatial variability and has a strong hemispheric gradient. Hotspots occur within the industrialized zones of its main source regions (USA, Europe, and East Asia). Recall that the DCE emission distribution within countries is prescribed here to follow that of ethene. In reality, the DCE source may be less dispersed than assumed, particularly if fugitive emissions occur from a relatively small number of point plant locations. Growth of DCE in the NH background is apparent from Fig. , and we estimate that the global DCE burden increased from ∼ 81 (± 15) to ∼ 116 (± 21) Gg between 2002 and 2020 (Table ). Chemical loss of DCE is controlled primarily through reaction with OH, and we calculate an overall global DCE lifetime, defined as the ratio of its global burden over its global loss rate, of ∼ 83 d in 2020 (Table ). This is in very close agreement with the ∼ 82 d reported by Burkholder and Hodnebrog (2022).

It is well established that VSLSs may contribute to stratospheric halogen loading via both source gas injection (SGI) and product gas injection (PGI). Chlorine SGI and PGI from DCE are also shown in Fig. . Defining these quantities at the tropical tropopause (∼ 17 km), the total (SGI + PGI) stratospheric chlorine input from DCE in the year 2020 is estimated to be 12.9 (± 2.4) pptCl, comprising 10.7 (± 2) pptCl from SGI and 2.2 (± 0.4) pptCl from PGI (Table ). For context, the total Cl-VSLS supply to the stratosphere (including VSLSs other than DCE) was estimated to be ∼ 130 (100–160) pptCl in 2019 when total stratospheric chlorine (i.e. including long-lived gases) was around 3240 pptCl (Laube and Tegtmeier et al., 2022). Our DCE SGI estimate in 2020 is similar to the 8.5 (± 1.9) pptCl reported in our previous modelling work (that did not include geographically or temporally varying emissions) for the year 2017 (Hossaini et al., 2019). A notable difference with our previous work is that we assess here that stratospheric chlorine from DCE has increased significantly over time (see Fig. S3), reflecting growth in emissions and hence SGI. Based on an ordinary least-squares regression applied to the model output over the full study period (2002 to 2020; Fig. S3), mean growth rates for SGI and PGI are 0.19 and 0.04 pptClyr-1, respectively.

In the current study, our estimated chlorine PGI is also similar to the value of ∼ 2 pptCl reported in Hossaini et al. (2019). The latter assumed a fixed lifetime of Cly in the troposphere against deposition (∼ 5 d), whereas we adopted an improved, more explicit representation here in which Cly washout was calculated using the standard TOMCAT deposition routines for the component chlorine species (Sect. ). As for all VSLSs, lack of observational constraint means that modelled PGI estimates carry significant uncertainty. A process not considered here is the heterogeneous recycling of Cly on ice crystals in the upper troposphere, for which there is some observational evidence (von Hobe et al., 2011). As demonstrated for iodine (Saiz-Lopez et al., 2015), such a process could plausibly extend the Cly lifetime and thus increase the magnitude of PGI. However, note that the net effect on PGI will likely depend on the interplay between ice uptake followed by sedimentation and also heterogeneous ice-recycling reactions that return species back to the gas phase (e.g. Fernandez et al., 2014, 2021). Such processes and the required parameters with which to treat them in a global model are highly uncertain.

The model estimates of chlorine SGI from DCE that are presented in Fig. c and d and Table  are annual mean quantities at the tropical tropopause, averaged zonally over the whole of the tropics (± 20° N/S). While transport across the tropical tropopause is the main route via which air enters the stratosphere, relatively elevated levels of VSLSs (and other gases) have been reported in the subtropical NH lower stratosphere owing to the effects of the Asian summer monsoon (ASM) and ASM anticyclone (e.g. Fiehn et al., 2017; Keber et al., 2020; Lauther et al., 2022). Forming in boreal summer, ASM dynamics are characterized by a rapid uplift of boundary layer air to the upper troposphere and lower stratosphere by deep convection (e.g. Randel and Park, 2006; Basha et al., 2020), including relatively polluted air masses from South and East Asia (e.g. Li et al., 2005; Randel et al., 2010; Müller et al., 2016). Based on aircraft measurements obtained during the Asian Monsoon Anticyclone 2017 campaign (AMA-17) over the Indian subcontinent, Adcock et al. (2021) reported a mean DCE mole fraction around the tropopause (355–375 K) of ∼ 12 ppt (with a range of 4.5 to 23 ppt), corresponding to a chlorine SGI (i.e. 2× the DCE mole fraction) in the range of 9–47 pptCl. The AMA-17 measurements were obtained in July and August in the latitude range of 21–29° N, longitude range of 79–91° E, and from an altitude of ∼ 10–20 km. The measurements from Adcock et al. (2021) are shown in Fig. S4 along with corresponding model estimates. There is generally good agreement between the two datasets, and the model corroborates the signal of relatively large levels of DCE around the tropopause (∼ 10 ppt) and hence a larger local stratospheric chlorine SGI (∼ 20 pptCl) in this region/season relative to the annual mean quantities around the tropical tropopause reported in Table .

Other recent studies have also highlighted the importance of Asian emissions in contributing to the atmospheric loading of a range of Cl-VSLSs. For example, in a global modelling study, Roozitalab et al. (2024) used a “tagged tracer” approach to show that Asian emissions likely dominate the global CH2Cl2 and C2Cl4 distribution. This was the case not only at the surface but also at high altitudes (150 hPa). The same study also analysed measurements of several Cl-VSLSs (including DCE) during the ATom campaign and tentatively assigned relatively enhanced NH mid-latitude mole fractions of Cl-VSLSs (observed during ATom-1) as being influenced by deep convection associated with the Asian summer monsoon. High-altitude aircraft observations from the Asian Summer Monsoon Chemical and Climate Impact Project (ACCLIP) mission have also revealed that the lower stratospheric abundance of Cl-VSLSs above the East Asian monsoon is at least a factor of 2 larger than previously observed in the tropics (Pan et al., 2024).

The modelled stratospheric ozone change due to DCE under 2020 conditions is shown in Fig. . DCE decreases stratospheric ozone globally, although the effect is generally small. The largest absolute decreases occur in the upper stratosphere (10–1 hPa) and polar lower stratosphere (200–20 hPa), i.e. regions where chlorine-catalysed ozone loss is known to be important (e.g. Chipperfield et al., 2018). The absolute ozone decreases in Fig. a and b are up to ∼ 5 ppb when expressed as an annual average (Fig. a). Larger decreases (up to ∼ 10 ppb) occur within SH high latitudes in spring when the Antarctic ozone hole forms (Fig. b). Corresponding ozone changes (expressed in percent) are shown in Fig. c and d. In most regions, the ozone changes due to DCE represent changes of < 1 %, although reductions of up to ∼ 1.3 % in the lower stratosphere are found in the SH polar spring.

The small (although non-zero) effect of DCE on global stratospheric ozone reflects the relatively small input of chlorine from DCE to the stratosphere (see above). However, as noted, several studies have identified transport via the ASM as a route through which relatively large local injections of various VSLSs (including DCE) to the extratropical lower stratosphere can occur (Keber et al., 2020; Adcock et al., 2021; Lauther et al., 2022; Pan et al., 2024). In principle, this process and its effect on chlorine injection from DCE is represented in our model (see Fig. S4). However, the very large surface levels of DCE, in at least some parts of Asia (see Fig. c), that are not tightly constrained by the data considered here may be underestimated in this analysis and thus too the co-location of emission hotspots with regions of relatively fast vertical ascent. Nonetheless, the local impact of DCE on summertime stratospheric ozone in the ASM region was briefly examined and is shown in Fig. S5. Noting again that chlorine-catalysed ozone destruction is generally efficient in the polar lower stratosphere and upper stratosphere, we find that ozone changes due to DCE in the lower stratosphere above the ASM region/season (i.e. the localized effect) are small (< 0.1 %). Ultimately, the overall significance of the ASM transport pathway for VSLS-driven stratospheric ozone loss is an area of current research that requires further and more detailed investigation considering other VSLS species (including those with predominately natural sources) and would benefit from new measurements in this region.

This study has focussed only on the possible direct impact of DCE emissions on ozone. However, we note that a broader impact assessment (beyond the scope of this work) might also factor in the unintended, although expected to be very minor, formation of other halogenated chemicals that inevitably occur during the DCE production process (the majority of these are destroyed by thermal oxidation or other means). These species are found in the “lights” and “heavies” effluent streams that may range from 0.3 % to 1.0 % of the DCE produced (TEAP, 2022) and include a proportion of ODSs, such as carbon tetrachloride (CCl4), and other chlorinated VSLSs, such as chloroform (CHCl3).