The majority of AGAGE DCE measurements were conducted using a Medusa preconcentration unit, coupled to gas chromatography and mass spectrometry (Medusa-GC/MS). The method is described in detail by Miller et al. (2008), Arnold et al. (2012), and Prinn et al. (2018) and only the most important aspects are summarized here. For some additional operational and analytical information see Appendix C.

With Medusa-GC/MS systems, one measurement is made every 60–65 min. Working standard measurements (std_sample) are either run after every air measurement (i.e. air_sample, std_sample, air_sample, std_sample) or after every two air measurements (i.e. air_sample, air_sample, std_sample, air_sample, air_sample, std_sample), with the latter approach used at JFJ, CMN, TOB, TAC, and in later years at CGO. For AGAGE in situ measurements, ambient air is sampled through inert, continuously flushed Synflex (Eaton) or stainless-steel inlet lines pumped from the top of an air sampling tower. From these continuous air streams, the air is sampled into the instruments using either a clean diaphragm or metal bellows pump. The sampling setups were found to not influence ambient levels of DCE.

With a Medusa-GC/MS, 2 L of air are sampled at a flow rate of 100 mL min⁻¹ (i.e. total sampling time 20 min), dried using two Nafion membrane dryers (Perma Pure), and pre-concentrated and focused cryogenically with a two-trap system (Hayesep D adsorbent). For this, initially Polycold “Cryotiger”/PCC cryocoolers were used, which were eventually replaced by Stirling coolers (AMETEK, Inc./Sunpower Inc., CryoTel GT) at most sites, creating sample trapping temperatures of -165 °C or colder for both micro traps. More abundant bulk gases such as N₂, O₂, CO₂, and some noble gases, are largely removed from the sample via fractionated distillation steps (by controlled heating of each micro-trap) and purging with research grade helium. After pre-concentration, the analytes are cryo-focused and then transferred into a gas chromatograph (Agilent 6890, 7890). Chromatographic separation of DCE from other analytes is achieved with a CP-PoraBOND Q column (0.32 mm ID × 25 m, 5 µm film thickness; Agilent). The analytes are detected by electron ionization quadrupole mass spectrometry (EI)-qMS (Agilent 5975, 5977) in the selected ion monitoring (SIM) mode.

At CMN, the first years of the DCE record were measured using a simpler Adsorption Desorption System (ADS) (Maione et al., 2013), before the installation of a Medusa-GC/MS in December 2023. More details are described in Appendix C and D.

Helium (grade 5.0 or 6.0), which is further purified with a helium purifier (HP2, VICI), is generally used as the carrier gas. GCWerks (GCWerks, 2026) is used as instrument control and data processing software. Measured DCE mole fractions are reported as dry air mole fractions in ppt (parts per trillion, pmol mol⁻¹).

Laboratory air and “blank” samples are typically measured on a weekly basis. Laboratory air is monitored to identify cases where there is the potential for contamination of the ambient air samples. The blank measurements are performed by running a measurement cycle without trapping any sample air. In general, non-zero blanks can be indicative of contamination in the carrier gas, contamination or leakage within the instrument, or carry-over from the previous sample. For DCE we typically observe non-zero blank values at all sites within the network except for CGO (this site only recorded a very small number of non-zero blanks over the entire timeseries). Carry-over has been identified as the principal cause of non-zero blanks for this species.

A correction was applied within the GCWerks software to account for non-zero blanks. This correction assumes that a constant blank mole fraction is added to all run types (e.g. blank, air, working standard). Typically, this is calculated as the mean mole fraction for the blank runs over each working standard lifetime (i.e. each physical working standard tank is associated with its own blank mole fraction). This correction performs well when the mole fraction of the ambient air is similar in magnitude to the mole fractions of the working standard and tertiary tank used to propagate the calibration scale to the site (see Sect. 2.3). For sites that experience very high pollution events, the assumption of a constant blank mole fraction breaks down. For example, the carry-over from a highly polluted air sample can significantly increase the peak area measured by the subsequent standard, such that the mole fraction calculated for the polluted air sample is underestimated. For this reason, data from the highly polluted AGAGE site at Gosan were excluded from this study, as the blank correction we applied could not account for carry-over in this case. The second most polluted site within the network for which DCE data exists was TAC. At this site the root-mean-squared error on the reported mole fractions due to carry-over was found to be 0.7 %. For 99 % of samples at this site, the impact of carry-over was found to be less than 2 %.

For NOAA's flask measurements, samples were collected approximately weekly during clean air conditions. Sample collection systems vary between sites, but the equipment used was found to not influence ambient air mole fractions of DCE during sampling. Intake heights vary from just a few meters above ground level (a.g.l.) at a high alpine tundra location (NWR), to 300 m a.g.l. (LEF). Most of the sampling inlets at the remote sites are between 10 and 20 m a.g.l. The intake lines are either continuously flushed Synflex (similar to the Medusa systems described above), 316 Stainless Steel, or polyvinylidene fluoride (PVDF; Kynar^®) tubing that has been cleaned and baked in the laboratory prior to deployment to the field. These flushed intake lines are connected to a diaphragm pump (N86, KNF) with a 316 stainless-steel head and typically flow at 4–5 SLPM. The lines are flushed for 5–10 min (depending on measured flows at the site and inlet length) prior to connecting flasks to adequately flush out stagnant air before sampling is started. Paired 3 L internally electropolished stainless steel flasks are connected, and flushed for 12–20 min, again depending on the flow rate measured at the time of sampling (approximately 10 times the volume for two flasks combined). The flasks are flushed either in parallel, or in series (again site dependent) and subsequently pressurized to 3 bar absolute. The flasks are then shipped back to NOAA's Global Monitoring Laboratory in Boulder, Colorado, USA for analysis on the NOAA “M3” GC/MS (Agilent 6890 GC and Agilent 5973 MS) or, after November of 2023, the “M4” instrument (Agilent 8890 GC and 5977B MS).

For analysis, air from a flask is used to pre-flush the instrument inlet and approximately 200 mL (STP) of sample is directed onto a 10 cm length of a 0.53 mm ID uncoated fused silica tube maintained at reduced pressure and -165 °C. Prior to cryo-condensation of DCE and other trace analytes, the sample air is dried by passing it through a tube (∼ 1/4^′′ SS by 10 cm long) containing powdered magnesium perchlorate held in place with silanised glass wool. For the manually filled glass and stainless steel flasks, two injections are analysed from each flask, and the mean of four injections is reported for each sampling event. When flask pair differences are larger than 0.5 ppt or 10 % the results are discarded as being unreliable (less than 3 % of flask pairs). Only one injection is taken per flask from the automatically filled glass flasks, since a smaller amount of air is captured (0.7 L flasks are filled to 40 psia). Air analysis proceeds on a fused silica capillary column (0.25 mm ID × 60 m, with a 1µm DB-5 film) that is cycled from -55 to 200 °C for each sample and by monitoring the ion at m/z = 62 (C₂H₃Cl⁺). Peaks are integrated and peak area data are processed with custom software. Replicate injections from flasks are bracketed by analyses of working standards containing whole air, and instrument blanks (synthetic “zero” air) are analysed daily and have repeatedly shown no measurable levels of DCE. The median replicate injection precision (as 1 standard deviation of the two injections) in the ∼ 10 000 flasks analysed over the measurement period has been 0.04 ppt (or 0.4 %); in 95 % of the analyses, it has been smaller than 0.16 ppt (2.3 %).

AGAGE's DCE measurements are linked to the recently established METAS-2021 primary calibration scale, which was produced based on permeation and dynamic dilution at the Swiss Federal Institute of Metrology (METAS). The measurement uncertainty on the SI-traceable mole fraction is 0.9 % (1-sigma confidence level, for cylinder MP21-001) (Bühlmann et al., 2026; Guillevic et al., 2018). The integration of the METAS-2021 primary scale into the AGAGE R1 relative calibration scale (maintained by the Scripps Institution of Oceanography, SIO), allows full, network-wide intercalibration of DCE measurements. The primary scale is propagated to each AGAGE measurement site through a calibration scheme (Miller et al., 2008; Prinn et al., 2000, 2018) involving a set of three-level hierarchically linked standards. This calibration hierarchy comprises of: whole-air secondary standards, based at the SIO central calibration centre; whole-air tertiary standards, distributed to the sites by the calibration centre; and whole-air quaternary (working) standards, mostly collected by each group operating a measurement site. All whole-air standards are ambient moist air compressed into 34 L internally electro-polished stainless-steel tanks (Essex Industries, Inc.), using an oil-free diving air compressor (SA-6, RIX Industries), or collected cryogenically, during relatively clean air conditions. From the beginning of whole-air standard collection for the routine measurement of DCE in AGAGE (starting end of 2017), the ambient air at all sites contained DCE at mole fractions well above the limit of detections of onsite instrumentation. The accuracy associated with the propagation of the primary scale through the AGAGE calibration hierarchy is usually derived from propagating the measurement precisions.

The instrumental precisions for the in situ air measurements are derived from the standard deviations of continuously bracketing working standard measurements (see Sect. 2.2). During routine operation at each field site, working standard measurements bracket each or each pair of ambient air sample measurements to account for changes of the MS sensitivity. For Medusa-GC/MS instruments the average measurement precision for DCE was 0.5 % (taken as the mean of the average precisions for each site). For the ADS instrument the respective average measurement precision was 1.6 %. Combining these on-site measurement precisions with uncertainties associated with the production and propagation of the calibration scale, including an estimate for the impact of carry-over of 0.7 % for each propagation step, we derive overall 1-sigma uncertainties of 2.4 % for the Medusa-GC/MS instruments; and 3.3 % for the CMN-ADS.

Weekly measurements of the working standard against the tertiary standard (typically 4 repeat sets of measurements) allow for tracking of potential drift of DCE in the standard cylinders. When a new tertiary standard arrives at a field site, it is compared to the old tertiary, and the old tertiary is remeasured at SIO, resulting in three sets of comparisons (OUT value at SIO, on-site value, IN value at SIO). Combined, these comparisons usually allow SIO to correct any drift in DCE (if it occurred). No drift has been observed in the secondary standards at SIO.

NOAA's DCE measurements are referenced to the NOAA-2003 absolute calibration scale, which consists of five different primary standards prepared using gravimetric techniques at mole fractions ranging from 13 to 30 ppt in Essex cannisters initially pressurized with humidified synthetic air to 900 psi. The response per mole of DCE injected for these standards in repeat analyses at different times has been within ±5 % and has been consistent overall since they were initially prepared at different times between 2018 and 2022. Consistency in the measurement scale throughout the entire record is supported by repeat analyses of archived whole-air samples pressurized in Essex cryogenics 34 L cannisters. Repeat analyses for these archived samples has been within 5 % for the entire suite of samples, whose mole fractions range from 2 to 35 ppt. For more details on standard preparation in general at NOAA/GML, see Hall et al. (2007).

The measurements of both networks at five shared sites (CGO, SMO, RPB, THD, MHD) are regularly compared at the twice-yearly AGAGE meetings. For DCE, the average ratio between the NOAA (NOAA-2003 scale) and AGAGE (METAS-2021 scale) measurements is 1.0406 ± 0.0103 (NOAA/AGAGE) for the period 2017–2025.

The global and regional emission estimates used a prior emissions map taken from Hossaini et al. (2024). In that study, several emissions maps were constructed by combining proprietary country-level data for DCE production with publicly available data for country-level imports and exports, then applying a range of different assumed production emission factors. Hossaini et al. (2024) established the scenario “sc05” map as their central case, as derived model results using sc05 were in overall better agreement with previous aircraft observations than for their other scenarios. Hence, in this study we use their scenario sc05 map, corresponding to emission factors from production of 0.5 % for developed countries and 1.5 % for developing countries (as defined by Article 5 of the Montreal Protocol), with an emission factor of 0.1 % applied to imports (i.e. supply chain emissions) and consumption (assumed to be feedstock use). Emissions within each country were spatially distributed using an ethene proxy, taken from the “industrial combustion and processes” sector of the gridded (0.5° × 0.5°) CMIP6 dataset (Feng et al., 2020). The Hossaini et al. (2024) emissions estimates only include years up to 2020, so for 2021–2023 we have used the 2020 emissions map.

Global emissions of DCE have been quantified using the approach described in detail in Western et al. (2025). In brief, an inverse method is used to derive emissions based on semi-hemispheric monthly averaged mole fractions and a 12-box model of the atmosphere (Cunnold et al., 1983; Rigby et al., 2014). The atmospheric lifetime of DCE is assumed to be controlled by reaction with the hydroxyl radical, with rate constants taken from Burkholder et al. (2020), and an ocean sink with a partial lifetime of 12.9 years (Yvon-Lewis and Butler, 2002). Given the errors likely introduced by using a coarse resolution atmospheric model for a short-lived substance, we assume a 1-sigma error of 50 % in the derived total effective lifetime, given the variability within each semi-hemisphere. The inverse method requires a priori emissions, the “sc05” scenario from Hossaini et al. (2024) as described above, and we assume that the uncertainty in the year-to-year growth is 20 %. The measurement uncertainty is split into two components: a random 1-sigma uncertainty on the monthly mean mole fractions that combines uncertainties associated with propagating the scale from SIO to the AGAGE working standards (1.2 %) and the variability of the background measurements each month, and a systematic 1-sigma uncertainty associated with the calibration scale of 2.9 % (derived from the comparison between METAS-2021 and NOAA-2003 scales). In addition to providing posterior emissions estimates, posterior semi-hemispheric monthly mean mole fractions are also output by the model.

The global model results presented in the main text used only data collected as part of the core AGAGE and NOAA networks, with separate model runs performed using data from each network to independently derive emissions for the period 2017–2023. For this period the global emission growth rate is well constrained by the observational network. A separate model run was performed using the archived air measurements to derive emissions from 2002 onwards. Due to the sparsity of the air archive data, AGAGE and NOAA air archive measurements were both used simultaneously alongside core AGAGE measurements (after converting the NOAA data to the METAS-2021 scale using the average ratio of time-matched NOAA/AGAGE measurements at the co-located sites). Using these archive measurements allows for a comparison against the full emission timeseries presented by Hossaini et al. (2024), but the data sparsity leads to large uncertainty in the annual emission growth rate derived using the 12-box model. The results from these emissions estimates using the box model and air archive measurements are presented in Appendix E.

DCE emissions in northwest (NW) Europe (Belgium, Germany, France, UK, Ireland, Luxembourg, and the Netherlands) and California were estimated using a regional modelling approach. For NW Europe, observations from the following sites were used: CMN, JFJ, MHD, TAC, and TOB. The seven countries for which we report aggregate emissions were selected based on the country-level error reduction in the posterior relative to the prior. The California emission estimates used observations from the sites at SIO and THD. Four different inverse modelling systems were used for NW Europe: InTEM (Inverse Technique for Emission Modelling) (Arnold et al., 2018; Manning et al., 2021), RHIME (Regional Hierarchical Inverse Modelling Environment) (Ganesan et al., 2014), ELRIS (Empa Lagrangian Regional Inversion System) (Henne et al., 2016; Katharopoulos et al., 2023) and FLEXINVERT (Thompson and Stohl, 2014). Only InTEM and RHIME were used to estimate California emissions.

All four models employ Bayesian methods to infer emissions, subject to constraints provided by model and measurement uncertainties as well as uncertainties associated with a prior emissions map. The InTEM, RHIME, and ELRIS runs in this study all used the NAME (Numerical Atmospheric dispersion Modelling Environment) (Jones et al., 2007) dispersion model to represent the transport of emissions from their source to the measurement sites. FLEXINVERT used a different dispersion model: FLEXPART (FLEXible PARTicle dispersion model) (Bakels et al., 2024). All models used the Hossaini et al. (2024) prior map described above. Differences between the models are summarized in Appendix F (Table F1). In addition, a sensitivity test using a prior with uniform emissions over land was also performed for the NW Europe emission estimates, as described in Appendix F.

Details of the modelling approach used by InTEM, RHIME, and ELRIS are presented by Vollmer et al. (2026). Here we describe only the differences in model configuration for this specific study. In this case, the lifetime of DCE in the atmosphere is sufficiently long relative to the transport represented by the NAME model (up to 30 d backwards in time from each measurement point) that the species was considered inert within the model, in contrast to Vollmer et al. (2026), who applied a decay function to the model output to account for atmospheric degradation of the shorter-lived halogenated olefins in their study. The measurement uncertainties used by the models were taken to be the working standard precisions (as described in Sect. 2.3). Measurements were averaged into 4 h periods with the variability within each 4 h period used as a proxy for model representation uncertainty. The model and prior uncertainties used by each inversion system are described by Vollmer et al. (2026). In this study, RHIME used a scaling factor on the prior fluxes sampled from a lognormal distribution with a mean of 1 and a standard deviation of 8 (as opposed to a standard deviation of 4 used by Vollmer et al., 2026), reflecting the higher uncertainty associated with the DCE prior flux map. Similarly, ELRIS used a spatial correlation length of 250 km for the prior emissions, as opposed to 500 km used by Vollmer et al. (2026), reflecting the expectation of more point-like sources and greater uncertainty on the prior spatial distribution for DCE.

In InTEM, boundary conditions were determined from the fraction of air entering from the 11 boundaries surrounding the computational domain (calculated by the NAME transport model) as detailed by Manning et al. (2021). Air entering from the southern boundaries is strongly influenced by the tropics and Southern Hemisphere, hence an average annual ratio of the background mole fraction between the northern tropics and the northern mid-latitudes was calculated from the observations at RPB, and the observations at MHD. The same estimated tropics ratio values were used for both the European and Californian domains. In RHIME and ELRIS, boundary conditions for the NW Europe estimates followed the same approach as Vollmer et al. (2026). RHIME used prior boundary conditions for the California emission estimates taken from the results of the 12-box model (see Sect. 2.4.2).

In contrast to the other three models, FLEXINVERT employed the FLEXPART dispersion model with 10 d back-trajectory calculations (with DCE again assumed inert over this time frame). The measurements used in the inversion were averaged into 3 h periods (different from the 4 h averaging of the other models). The FLEXINVERT inversion framework used is detailed by Annadate et al. (2025); here, we focus only on the details specific to this study. For the estimation of background mole fractions, we applied the Robust Extraction of Baseline Signal (REBS) (Ruckstuhl et al., 2012) method. Within the FLEXINVERT+ framework, we utilised a log-normal prior emission distribution and the M1QN3 solver to iteratively minimize the cost function. To estimate posterior uncertainties, we ran a 20-member Monte Carlo ensemble of inversions.