Here, we describe the measurements used to derive global emissions and mole fraction trends. AGAGE in situ measurements of ODSs and GHGs have historically been made using multiple measurement instruments at each site. Measurements from Medusa gas chromatography mass spectrometry (GC-MS) systems , deployed at each AGAGE site in the early to late 2000s and 2010s, are used to derive global emissions and mole fraction trends for most of the compounds listed in Table . Exceptions are CFC-11, CFC-12, CCl₄, and N₂O, for which measurements by the AGAGE gas chromatography `multidetector' (GC-MD) systems at MHD, THD, RPB, SMO and CGO are preferentially used (in the case of N₂O exclusively) due to higher measurement frequency and longer measurement records see. These compounds are measured using electron capture detection (ECD) . Sites that do not have a GC-MD instrument (JFJ and ZEP) were not used to estimate CFC-11, CCl₄, and N₂O mole fractions and emissions.

Prior to Medusa GC-MS measurements, some compounds had been measured on GC-MS adsorption–desorption systems (ADS), starting out with a prototype-ADS system at MHD in mid-1994, followed by ADS systems at MHD and CGO in late 1997 seefor more information. GC-MS-ADS measurements also commenced at JFJ in 2000 and at ZEP in 2001 . Information about the compounds and time periods for which these GC-MS-ADS data are used to derive trends in emissions and mole fractions is contained in . All GC-MD, ADS, and Medusa measurements are reported on the calibration scales used by AGAGE, as detailed in .

Methane has been historically measured by AGAGE GC-MD systems (and its GAGE predecessor) using flame-ionization detection (FID), reported on the Tohuko 1987 scale maintained at SIO, but at some sites in recent years GC-MD CH₄ measurements have been replaced/superseded by cavity ring-down spectrometer (CRDS) instruments (Picarro) see, reported on the NOAA-2004A scale . Extensive NOAA-AGAGE intercomparisons as well as AGAGE on-site instrument comparisons during instrument overlap have shown that the scale differences are negligible NOAA/AGAGE ratio of 1.0001 ± 0.0007,. Sites that do not have a GC-MD instrument (JFJ and ZEP) were not used for methane (even when CRDS measurements were available).

The typical repeatability of the measurements made by the GC-MS Medusa and GC-MD systems discussed here range from 0.05 % for N₂O, 0.1 % for CF₄ and CFC-12, 0.3 %–1 % for most compounds, and up to 7 % for HFC-236fa of the measured standard value. For CH₄, GC-MDs achieve 0.2% and CRDS systems achieve 0.02 % for more information see. The measurement repeatability is mainly compound and detector-dependent and is largely dominated by the atmospheric abundance of the compound but can also be negatively affected by site specific problems such as lab air contamination, lab temperature problems, trap temperature fluctuations, or MS filament problems.

The measurement data sets available from provide details on the instruments used to measure each compound listed in Table , and for which period. The measurements are used in conjunction with the statistical AGAGE pollution algorithm to determine pollution free monthly mean baseline mole fractions, which are then used as input for the model and inversion to produce the data sets presented here. This method allows for the determination of monthly mean baseline mole fractions, as detailed in Sect. .

For several compounds, measurements of archived air samples are used to extend the in situ measurement record back into the past available from. For the Southern Hemisphere, air collection and archiving began in 1978 with the Cape Grim Air Archive (CGAA), where air samples were taken at CGO during clean air conditions with cryogenic methods and stored in stainless steel tanks , with the intention of reconstructing the historical composition of ambient air once suitable analytical instruments and calibration scales were developed. Early measurements of the CGAA were performed on various instruments summarized in, but here we focus on Medusa GC-MS measurement made in 2007 e.g., and 2011 e.g.,, which were used in many subsequent studies e.g.,. Later CGAA measurement made in 2016 are currently not used here. These Medusa GC-MS measurements were mostly performed at the CSIRO Aspendale laboratory in Australia, but also at the Scripps Institution of Oceanography (SIO), in La Jolla, California USA. The frequency of available CGAA air samples differs, with one or two samples per year typically available before 1994, and up to nine samples available per year between 1994–1999, after which measurements of ongoing archived air samples are no longer used in this work. There is good agreement between the measurements at SIO and CSIRO of identical air samples and air samples with the same or similar fill dates.

To complement the CGAA, archived air samples from the Northern Hemisphere were gathered from several laboratories and mostly measured on Medusa GC-MS systems at SIO. Many of these tanks had been filled at THD or SIO, some at other northern hemispheric locations in the USA (such as Cape Meares in Oregon, Point Barrow in Alaska, and Niwot Ridge in Colorado) between 1973 and 2016 . Unlike the CGAA, many of these samples were not originally intended for future atmospheric archive measurements and required more stringent quality control. For inert and/or volatile or very abundant compounds (such as CF₄, SF₆, NF₃, many HFCs and HCFCs) the resulting measurements were well-suited to reconstruct historic northern hemispheric abundances. Measurements of other compounds (e.g., several minor CFCs, H-2402, HCFC-22, HCFC-124, HFC-43-10mee, PFC-218) produced some anomalous data points during data processing, which resulted in less certain northern hemispheric historic abundances for these compounds. Some archived air samples from the Northern Hemisphere were also measured at CSIRO , again generally confirming that measurements from the instruments at SIO and CSIRO can be combined.

Table  shows the compounds that use archived air measurements in their emissions estimates. The references in Table  are the first publications in which archived air was used to derive emissions, and subsequent relevant publications.

Derived global emissions and mole fraction trends are inferred from monthly mean “baseline” mole fractions for each measurement site in Sect. . A baseline measurement is when the sampled air is well mixed within the air parcel and is not influenced by nearby pollution sources. These monthly mean baseline mole fractions are fed into the inversion framework described in Sect. .

Here, monthly mean baseline measurements are derived using a statistical algorithm . The algorithm identifies measurements that are considered as baseline by taking the following steps.

For a given day, fit a second-order polynomial to the daily minima of measurements over a 121 d window centred on that day (i.e. using 60 d before and after). Subtract the fitted polynomial from all measurements within the window, to detrend the measurements, and calculate the median of these detrended data. Calculate the root mean square error (RMSE) using only the detrended values that fall below this median value. Classify measurements on the given day as baseline if they are within three times the RMSE of the median. Compute this step as a moving window across all days.

Dynamic transport in the 12-box model is represented through a parameterisation of advection and diffusion between the boxes. The change in the mass mixing ratio in surface boxes (j=0,1,2,3), χj, over time is given by the equation 1dχjdt=-1j>0(j)Vj-1,jχ‾j-1,j+(Δχ)j,j-1tj,j-1-1j<3(j)Vj+1,jχ‾j+1,j+(Δχ)j,j+1tj,j+1-Vj+4,jχ‾j+4,j+(Δχ)j,j+4tj,j+4-Lj+EjMj, where Vi,k is an inverse time constant representing the mean meridional transport between boxes i and k, ti,k-1 is the rate of eddy diffusion, Ej is the emissions into box j, the losses are defined collectively by Lj, and Mj is the total mass of air in box j. Other mathematical descriptors are 2χ‾i,k=12(χi+χk),3(Δχ)i,k=χi-χk, and the indicator function, which is defined where 41A(j)=1if Ais  true0otherwise. For tropospheric boxes (j=4,5,6,7), the mass mixing ratio is given by 5dχjdt=-1j>4(j)53Vj-1,jχ‾j-1,j+(Δχ)j,j-1tj,j-1-1j<7(j)53Vj+1,jχ‾j+1,j+(Δχ)j,j+1tj,j+1-53Vj-4,jχ‾j-4,j+53(Δχ)j,j-4tj,j-4-(Δχ)j,j+4tj,j+4-Lj, and finally for the stratospheric boxes (j=8,9,10,11), 6dχjdt=-1j>8(j)(Δχ)j,j-1tj,j-1-1j<11(j)(Δχ)j,j+1tj,j+1-32(Δχ)j,j-4tj,j-4-Lj. Equations (), () and () are solved using a Runge-Kutta (RK4) method see e.g. with a time step of 48 h.

The 12-box model has various loss processes, depending on the compound and box, which are summarised in Table . Where target lifetimes in the literature are given as a range, the median (with respect to the inverse lifetime, or loss frequency) is used. Compounds with total atmospheric lifetimes greater than 10 000 years are considered to have no atmospheric sink.

Losses due to the reaction with the hydroxyl radical (OH), for compounds with non-negligible OH losses, are calculated in the tropospheric boxes (boxes 0–7, see Fig. ). The concentrations of OH taken from are seasonally varying but annually repeating and the OH field is adjusted so that the model mole fraction simulation of CH₃CCl₃, whose loss is dominated by OH, is consistent with AGAGE observations using an approach similar tobut for inter-annually repeating, rather than inter-annually varying OH. All compounds are passive with respect to the OH loss, meaning that there is no loss of OH due to the sink process. Losses due to OH are computed using a first-order rate constant, using the Arrhenius equation and the temperature fields from Sect. . The Arrhenius factor and molar activation energy for each compound are taken from . Where the OH sink is thought to be negligible and is not reported, a default Arrhenius factor of 1 ×10-30 and E/R factor of 1600 is assumed.

Losses in the stratosphere (boxes 8–11 in Fig. ), due to reaction with excited atomic oxygen, stratospheric OH and photolysis are calculated using a single first-order loss frequency that accounts for the overall loss due to these processes. Lyman-α photolysis is treated as a loss in the stratospheric boxes in the model, even though this sink is a mesospheric loss process. A suitable first-order rate constant for stratospheric losses for each compound is found by optimisation, such that the steady-state stratospheric lifetime equals that in . An exception is the stratospheric lifetime of methane, which is taken from . Initial stratospheric losses are distributed latitudinally and temporally following , and these values are adjusted by a single factor in the optimisation, such that the relative temporal and spatial gradients are not altered. Target stratospheric steady state lifetimes are summarised in Table .

Compounds with sink processes due to ocean and/or soil uptake have losses in the lowest layer boxes (0–3 in Fig. ). In a similar manner to the stratospheric losses, ocean and soil losses are treated together as a single loss process. These losses are computed as a first-order loss, where the first order rate constant is optimised to give the steady-state target lifetime of the total soil and ocean loss .

For methane, loss from reaction with the chlorine radical is included in the model. The chlorine distribution is taken from . First order rate constants are calculated using the Arrhenius equation .

Tropospheric loss processes so far not addressed are present for some compounds, such as a non-negligible photolysis sink in the troposphere. These other tropospheric sinks are implemented in boxes 4–7 (Fig. ) using a first-order rate constant that is optimised in a similar way to that of the stratospheric and ocean/soil sinks, using a target tropospheric lifetime , considering the other loss processes present. The spatial and temporal gradient of this loss follows that of loss from tropospheric OH.

The inverse framework here generally follows that of . Emissions are derived based on the comparison of simulated mole fractions in the surface boxes and monthly background mean AGAGE observations. The monthly mean mole fraction in each box is calculated as the mean of the monthly mean mole fractions from all available measurement sites located in that box. We define x as the deviation (in Gg yr⁻¹) from an a priori estimate of emissions taken from available bottom-up estimates of global emissions in the literature, xa (see Sect. ). The corresponding observation y is the deviation of the measured mole fractions from the modelled mole fraction using these a priori emissions. The relationship between the difference in emissions and surface mole fraction from the a priori estimate is 7y=Hx+ϵ, where H is a sensitivity matrix relating emissions to surface mole fractions and ϵ is a stochastic error, resulting from error in the mole fraction measurements. The vector x is derived either monthly, seasonally or annually for each box, depending on the compound in question (see Table S1 in the Supplement for this information). The sensitivity matrix H is derived using the linear relationship between emissions and surface mole fractions by running the model with base (a priori) emissions and again with perturbed emissions (+1 Gg) for each surface box for each time period defined in Table S1. The sensitivity of the mole fraction to emissions is calculated until the end of the period where measurements are available. The surface mole fractions (deviation), y, are baseline monthly means as detailed in Sect. .

An initial first guess at the initial conditions of the mole fraction in the 12 boxes of the model uses a nine-year spin-up period using the a priori emissions field. The initial conditions (and H) are recursively adjusted to approximate the measurements using the a priori emissions. Sensitivities (and derived emissions) begin three years before the earliest measurement. These initial three years account for model spin-up and are not included in the output data sets.

Under the Bayesian framework, we choose to place a prior constraint on the growth in emissions, rather than their absolute value . The growth matrix D operates on the emissions x, to produce an emissions growth value, which we constrain by the a priori emissions growth g. The uncertainty in the a priori emissions growth is taken as a percentage of the maximum a priori emissions, and the percentages used are defined in Table S1 for each compound, where no box will have a growth less than 1 % of the maximum global growth. Uncertainty in this growth is assumed to be independent between boxes and time steps, which we contain in the matrix P.

We separate the random and systematic errors due to the measurements and modelling. Only the random components are included in the statistical evaluation of x and are assumed to be uncorrelated between boxes and time steps. These random errors are contained in the matrix R. The diagonal is composed of the quadratic sum of the typical measurement error (given in Table S1) and the variability of the baseline mole fractions in that month (see Sect. ). This latter term is assumed to be a measure of model error, by accounting for the magnitude of variability not accounted for in the model.

Under an assumption of a normal likelihood and prior probability, the resultant relationship for the probability of the emissions given the measurements (deviation) is, 8-ln⁡P(x∣y)∝(y-Hx)TR-1(y-Hx)+(Dx-g)TP-1(Dx-g), which, by completing the square, allows determination of the maximum a posteriori probability (MAP) estimate of emissions, x^, using 9x^=P^(HTR-1y+DTP-1g), where 10P^=(HTR-1H+DTP-1D)-1, and P^ is the resultant posterior covariance matrix, representing the random error in the emissions estimate. The estimate x^ is added to the initial a priori emissions to give the estimated total emissions. The posterior mean mole fractions are estimated using the relationship 11y^=Hx^+Hxa.

Combined systematic and random uncertainties are derived through the random sampling of systematic uncertainties and the Cholesky decompostion of P^. The systematic uncertainties are due to errors in the calibration, lifetime and transport. Calibration uncertainty is treated as a percentage offset, where the one standard deviation calibration uncertainties for each compound are defined in Table S1. The systematic component of transport error is assumed to be 1 % of emissions for all substances (one standard deviation). Lifetime error variance is calculated as, 12σx2=(Bσ1τ)2, where B is the total atmospheric burden of the compound and σ1τ is the total inverse lifetime error seefor more information. The assumed lifetime uncertainty is shown in Table S1 for each compound. The total uncertainty is then taken as the standard deviation from this ensemble.

An initial set of estimates of the emissions for each compound has been compiled over time. These are from a variety of sources and are included in the AGAGE-derived data sets. Given the longevity of measurements made by the AGAGE network and their widespread use, there could be a lack of independence of the a priori emissions if taken from widely used emissions scenarios, which may have been, at least partly, informed by mole fraction measurements and/or emissions derived by AGAGE e.g.,. We have therefore strived, insofar as possible, to use independent a priori emissions estimates.

A priori emissions for CF₄, C₂F₆, C₃F₈, c-C₄F₈, HFC-23, HFC-32, HFC-125, HFC-134a, HFC-143a, HFC-152a, HFC-227ea, HFC-236fa, HFC-245fa, HFC-365mfc, HFC-43-10mee, HCFC-141b, HCFC-142b and SF₆ are taken from the gridded flux maps from the Emissions Database for Global Atmospheric Research (EDGAR) v8 . These are annual emissions from 1970–2022, with the exception of C₃F₈, which spans 1975–2022. HFC-23 emissions prior to 1970 are taken from . After 2022, emissions for all compounds are repeated using the values from 2022.

The EDGAR v8 inventory also includes NF₃ but its global emissions are erroneously small compared to other literature sources e.g.,. We instead use the PRIMAP-hist v2.6 national historical emissions time series for NF₃ . Emissions are quantified until 2023 in the database and repeated thereafter.

CFC-11, CFC-12 and CFC-113/a a priori emissions are the bottom-up estimates compiled in , which were informed by a variety of inventory compilations and forecasts e.g.,. CH₃CCl₃ was compiled in a similar manner, but emissions have been repeated after 2015 until present. A priori estimates of CFC-114/a and CFC-115 were compiled from a variety of sources seeand its supplementary information. To the best of our knowledge, no comprehensive inventory of global emissions of CFC-13 exists, so we assume that the a priori emissions for CFC-13 are a seventh of those of CFC-115 (see the supplementary information of for the rationale behind this approximation, which is based on available ratios of production of the two compounds).

The a priori emissions estimate for halons – Halon-1211, Halon-1301 and Halon-2402 – are taken from the bottom-up emissions published by . These estimates were originally compiled by but the values were not made available.

The historic a priori estimates of CCl₄ emissions are based on information from industry on the production of CFCs (Archie McCulloch, personal communication, 2013). Estimates since 2010 are taken from . The use of carbon tetrachloride as a feedstock for the production of CFCs was its major source of emissions prior to the phase out of the production of CFCs for dispersive uses. There are continued unexplained global emissions of carbon tetrachloride, which makes a comprehensive bottom-up inventory of its emissions difficult .

Little is known about the source of emissions of HCFC-132b and HCFC-133a . These two substances are assumed a priori to have constant emissions, with an annual mean of 1 Gg yr⁻¹ with a small seasonal trend, following . HCFC-22 a priori emissions are an extension of the bottom-up emissions derived by , using the HCFC-22 production data from multiple sources . These emissions estimates extend through 2009, after which the emissions are assumed to remain constant. A bottom-up inventory of emissions for HCFC-124 was compiled for 1990–2001 and projected to 2015 under three scenarios . We use Scenario 2, which envisaged improvements to reduce HCFC emissions over the policy at the time of compilation. Emissions are repeated after 2015.

PCE (CCl2=CCl₂) a priori emissions are taken from , which was compiled from multiple sources . CH₂Cl₂, CHCl₃ and CH₃Cl estimates are based on the emission estimates compiled in and methyl bromide in . Given the lack of a priori information about emissions of these gases, they are assumed to be constant in time.

The a priori estimates for SO₂F₂ were described in , which were compiled using information on the production of SO₂F₂. Given the use of SO₂F₂ as a fumigant, it can be assumed that its emissions will be approximately equal to its production in a given year. Our a priori emissions estimate for this compound was constant after 2007.

Emissions for N₂O are built from various sources. Anthropogenic emissions are from EDGARv8.0 for the years 1970–2022, and the 2022 emissions are repeated thereafter . Ocean emissions are from the ECCO-Darwin model for the period 2009–2013 , and annually repeating before 2009 using the 2009 emissions and after 2013 using the 2013 emissions. Other natural emissions are from for 1990–2008, with annually repeating values using emissions from 1990 and 2008 before and after these dates. This estimate is within the range of bottom-up derived N₂O net flux .

Anthropogenic a priori emissions of CH₄ are taken from EDGAR v8.0 , wetland emissions are taken from WetCHARTS v1.3.1 , biomass burning from the Global Fire Emissions Database (GFED) , freshwater emissions from , rice from , and other natural sources from . Years without estimates from a particular source are filled with the closest year of available data. The total emissions are within the uncertainty of other well-used bottom-up total flux estimates e.g..

The presented mole fractions for each semihemisphere are those of the MAP estimate of the emissions, and their uncertainty, forward simulated through the 12-box model. The global annual mole fraction is the mean of the four surface-level boxes. The growth rates of the mole fraction are derived from y^ (see Sect. ) using backward differencing. These growth rates are then smoothed using a Kolmogorov–Zurbenko (KZ) filter with a window size of 9 and a filtering degree of 4. The uncertainties for the estimated mole fractions and their growth rates are derived from the Monte Carlo ensemble described in Sect. .

The annual global emissions of synthetic greenhouse gases are presented in Fig. . We group compounds as CFCs, HCFCs, HFCs, Halons, “other F-gases” (SF₆, NF₃ and SO₂F₂), PFCs and “Solvents” (CCl₄ and CH₃CCl₃). These emissions are shown in terms of the direct climate warming potential of the emissions over a 100 year time horizon (GWP-100, CO₂-eq) and their ozone depletion potential (ODP, CFC-11-eq) , respectively.

The total CO₂-equivalent emissions from synthetic greenhouse gases were 3.2 ± 0.6 Pg CO₂-eq in 2023, and have remained relatively unchanged over the last 20 years of the record (Fig. ). There has been an overall reduction in CO₂-equivalent emissions due to reduction of CFC emissions since the 1990s. This reduction has been partially offset by a growth and recent decline in emissions of HCFCs e.g.,, and the continuing growth of HFCs and, to a lesser extent, PFCs and other F-gases. It can reasonably be expected that the emissions of synthetic greenhouse gases will decrease in the coming years given the controls on HFCs under the Kigali Amendment to the Montreal Protocol and other commitments as part of the Paris Agreement and other climate policies .

The emissions of the ozone-depleting synthetic greenhouse gases (CFCs, HCFCs, halons and chlorinated solvents) were 152 ± 66 Gg CFC-11-eq in 2023. Emissions of these ozone-depleting substances are now at their lowest since measurement-derived emission records began, despite some small, but impactful, fluctuations in total emissions, meaning that this decline has not been entirely monotonic see, e.g.,.

The global abundances of the synthetic greenhouse gases are shown in Fig. , using the same compound groupings as Fig. . The impact on climate is shown in terms of direct radiative forcing, which considers stratospheric adjustments to the instantaneous radiative forcing, and also tropospheric adjustments for CFC-11 and CFC-12 . The impact on ozone depletion is shown in terms of equivalent effective chlorine , which is the global mean surface chlorine mole fraction (number of chlorine atoms in a given species multiplied by its mole fraction).

The direct radiative forcing of synthetic greenhouse gases has increased since measurements of their atmospheric abundance began, reaching 411 ± 5 mW m⁻² in 2023. Global abundances of CFCs, chlorinated solvents and HCFCs are now decreasing , and are being offset by an increasing abundance of HFCs, PFCs (foremost CF₄) and other F-gases (foremost SF₆).

Conversely, the equivalent effective chlorine has declined since its peak in 1994, to 2920 ± 30 ppt in 2023. This is driven by the decreasing abundances of CFCs, halons and solvents (foremost CH₃CCl₃) in the atmosphere and also more recently by a fall in the abundance of HCFCs .

Figure  shows the global and semihemispheric mole fractions of methane and nitrous oxide, the growth rate of these mole fractions and the north-south inter-hemispheric difference. Figure  shows their global mean emissions.

The global mean mole fraction of methane reached 1917 ± 10 ppb in 2023, following an accelerating rate of growth since a plateau in the global mole fraction during the mid-2000s . This increase in the global mole fraction is coupled with an increase in the north-south interhemispheric difference. Global emissions of methane in 2023 were 579 ± 73 Tg yr⁻¹. In 2020 these emissions were 564 ± 78 Tg yr⁻¹, which falls within the range of top-down emissions derived for methane in multiple studies, of 608 [561–650] Tg . The main drivers behind the increases and times of stagnation in direct methane emissions and the resulting global mole fraction are uncertain and may come from a mixture of natural and anthropogenic source and sink processes .

The global mean mole fraction of nitrous oxide reached 337 ± 2 ppb in 2023, following an increasing rate of growth since direct measurement records began. The north-south interhemispheric difference has shown substantially interannual variability but no obvious overall trend during this time. Global annual emissions reached 29 ± 3 Tg yr⁻¹ in 2023. In 2020 global emissions were 31 ± 2 Tg yr⁻¹, which is larger than the range of 26.7 [26.1–27.3] Tg yr⁻¹ (mean and range) for 2020 from four top-down approaches , but close to the estimate of 30.4 [29.7–31.6] Tg yr⁻¹ in . The emissions in 2020 are within the range of bottom-up estimates of 29.1 [16.7–42.4] Tg yr⁻¹ presented in .

Very short-lived substances (VSLSs) have total atmospheric lifetimes of less than around 6 months. Dichloromethane (CH₂Cl₂) and chloroform (CHCl₃) have natural as well as anthropogenic sources, although the natural emissions of CH₂Cl₂ are relatively minor . The atmospheric mole fraction of CH₂Cl₂, and its emissions, have been continuously increasing during the atmospheric record (Fig.  and see, e.g., ). Its global mean mole fraction has approximately doubled between 2008 and 2023, when it reached 41.5 ± 1.3 ppt, which has been driven mostly by increased emissions from China . Mole fraction and emissions of CHCl₃ increased rapidly until 2015 , after which they fell to the levels seen in the 2000s. Perchloroethylene (CCl2=CCl₂) has only anthropogenic sources and its atmospheric abundance and emissions have been falling since the measurement record began.

Methyl chloride (CH₃Cl) has an atmospheric lifetime of 10–11 months. It has a mixture of natural and anthropogenic sources . Its atmospheric mole fraction has remained fairly constant at around 550 ppt and its emissions at 4500–5000 Gg yr⁻¹ over the measurement record. Methyl bromide (CH₃Br) has an atmospheric lifetime of 9–10 months. It has a mixture of natural and anthropogenic (foremost fumigation) sources . Its production for all applications other than quarantine and pre-shipment purposes has been phased out under the Montreal Protocol, and its atmospheric mole fraction and emissions have been declining over the measurement record (Fig. ).