Main

As the primary atmospheric oxidant, the OH radical has a key role in the removal or production of major air pollutants, greenhouse gases and many ozone-depleting substances1,2,3. A better understanding of the NH/SH OH ratio will lead to significantly improved source estimates of reactive species and to an improved prediction of chemistry–climate interactions from changing human activities. Because of its very short lifetime (∼1 s), OH manifests high spatiotemporal variability. Moreover, because OH concentrations are typically very low (∼106 molecules cm−3), in situ measurements are challenging, and large differences between observations have prevented a direct validation of model-simulated OH distributions or an evaluation of uncertainties in the chemical mechanisms responsible for OH recycling under different environmental conditions14,15,16,17. Rather, indirect estimates of the total abundance and interannual variations of global OH have been made with methyl chloroform (CH3CCl3) or 14CO measurements and simulations by chemistry-transport models (CTMs)8,9,10,12,18,19,20.

Constraints on the meridional OH gradient are needed for estimating hemispheric source and sink magnitudes of gases and aerosols that are produced or destroyed through reactions with OH (see Methods). For example, large increases in NOx (NO + NO2) emissions from China compared with emission inventories are estimated by using a data assimilation system5 and the CHASER (Chemical AGCM for Studies of Atmospheric Environment and Radiative forcing21) CTM. An overestimate of NH OH could account for such discrepancies, because it affects the hemispheric budgets of NOx and other reactive species such as carbon monoxide (CO) and methane (CH4). CHASER-simulated OH is ∼26% higher in the NH than in the SH, and this model predicts a smaller than observed NH–SH CH4 gradient6. In the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP), the multi-model average and 1σ spread of annual mean NH/SH OH ratios was 1.28 ± 0.10 (range 1.13–1.42). The simulated NH/SH OH ratio being greater than 1 is related primarily to OH production from modelled ozone, which is biased high in the NH and low in the SH, compared with observations4. Other evidence exists for significantly lower NH/SH OH ratios. On the basis of 14CO and CH3CCl3 observations, the NH/SH ratio of OH is suggested to be significantly lower than 1 (refs 7, 8). A NH/SH OH ratio of 0.98 using models has been shown9, but it was concluded that accurate estimation of the ratio would require more accurate CH3CCl3 emissions (also in ref. 8) and models that accurately describe interhemispheric exchange.

We constrain the NH/SH ratio of OH by using an atmospheric general circulation model (AGCM)-based CTM, the Japan Agency for Marine-Earth Sciences and Technology (JAMSTEC) Atmospheric Chemistry Transport Model (ACTM)6,22. CH3CCl3 simulations are performed with two spatially distinct tropospheric OH distributions, namely, ACTM_0.99 and ACTM_1.26, which have annual mean NH/SH OH ratios of 0.99 and 1.26, respectively21,23 (Extended Data Fig. 1), with the NH and SH separated at the geographical Equator. The NH/SH OH ratios are 0.87 for ACTM_0.99 and 1.18 for ACTM_1.26, when the two hemispheres are divided in accordance with the monthly locations of the Intertropical Convergence Zone (ITCZ). The emissions of CH3CCl3 and sulphur hexafluoride (SF6) are taken from the transport model intercomparison (TransCom)-CH4 experiment6,24,25 and extrapolated for the years after 2008 (Extended Data Fig. 2 and Extended Data Table 1). We show below that the simulated NH–SH gradient of CH3CCl3 is equally sensitive to the NH/SH OH ratio in the model with regard to the emissions since 2000. Thus we can use the observed NH–SH CH3CCl3 gradient to constrain the NH/SH OH ratio in the model, provided that little uncertainties are contributed by emission estimates and transport parameterization.

Results from both the ACTM_0.99 and ACTM_1.26 simulations reproduce the temporal evolution of CH3CCl3 for the period 1994–2011 (Fig. 1), confirming that the global annual mean OH concentrations are in good balance with the ‘Control’ surface emissions used in the simulations. Concentrations of CH3CCl3 have decreased exponentially since the late 1990s as a result of the stringent implementation of the Montreal Protocol and its amendments to mitigate stratospheric ozone depletion3,12 (Fig. 1a). The simulated CH3CCl3 concentration decay rates are not significantly different, namely 18.28 ± 0.14% per year (average ± 1σ as interannual variability) for ACTM_0.99 and 17.27 ± 0.13% per year of the annual mean concentrations for ACTM_1.26, compared with the observed 17.85 ± 0.29% per year during 2002–2011 at five Advanced Global Atmospheric Gases Experiment (AGAGE) sites and nine National Oceanic and Atmospheric Administration (NOAA) sites (Extended Data Table 2a). The average CH3CCl3 lifetimes are calculated to be 4.91 ± 0.03 years for ACTM_0.99 and 5.19 ± 0.03 years for ACTM_1.26. These lifetimes agree well with the observation-based lifetimes for given inventory emissions of 5.0 (range 4.87–5.23) years3,26.

Figure 1: Temporal evolution of measured (symbols) and simulated (lines) CH3CCl3 in the atmosphere.
figure 1

a, Monthly mean concentrations at MHD (blue), RPB (green), SMO (red) and CGO (black). Observations (symbols) are taken at four AGAGE sites using GC–MD, and the ACTM simulations (lines) correspond to the ‘Control’ case of total emissions and annual mean OH. b, MHD–CGO concentration differences are shown in comparison with ACTM_0.99 (red) and ACTM_1.26 (blue) simulations. Note that because of the coarse horizontal resolution (T42 spectral truncations; ∼2.8° × 2.8°) site representation errors are large for Mace Head when intense emissions occurred over Western Europe, for example until 2000 for CH3CCl3. ACTM_0.99 simulation at a horizontal resolution of T106 spectral truncations (∼1.1° × 1.1°; ACTM_T106 in inset to b; green) for the period 2002–2011 shows no significant difference for CH3CCl3 from the T42 resolution run, indicating that site representation error does not affect our results. The inset to b also shows a model simulation (ACTM_UNEP; purple) using a different emission distribution, based on countries reporting to the United Nations Environment Programme (UNEP), but with identical global emission totals and OH distribution as for ACTM_0.99. The model lines are broken because of a missing observation in August 2010 at CGO. Representative CH3CCl3 emission distributions for the ‘Control’ and ‘UNEP’ cases are shown in Extended Data Fig. 2. Similar CH3CCl3 concentration gradients, based on a greater number of NOAA flask sampling sites, are shown in Extended Data Fig. 3.

PowerPoint slide

Source data

Full size image

Given specified source distribution and magnitude, the meridional CH3CCl3 concentration gradients are controlled mainly by loss due to reaction with OH and by meridional transport23. This is because the local lifetimes of 1–3 years in the tropical troposphere are of similar magnitude as the interhemispheric transport time of 1.3 years in the ACTMs6,22. Figure 1b shows CH3CCl3 concentration gradients between Mace Head, Ireland (MHD), and Cape Grim, Australia (CGO). Results from ACTM_0.99 reveal a closer agreement with the observed MHD–CGO CH3CCl3 concentration gradients than with those for ACTM_1.26, given the set of ‘Control’ global emissions and global mean OH concentrations (also noted using NOAA data from multiple sites; see Extended Data Fig. 3). The differences between ACTM_1.26 and ACTM_0.99 are readily apparent for the 2000s, when the yearly emissions of CH3CCl3 are less than 3% of the atmospheric burden, compared with the early 1990s, when yearly emissions were as large as ∼20% of the burden. Sensitivity simulations are conducted using ‘Control’ global emissions and mean OH at increased horizontal resolution (ACTM_T106) and with a different spatial distribution of CH3CCl3 emissions (ACTM_UNEP). ACTM_T106 and ACTM_UNEP show equally good agreement with observed concentration gradients between MHD and CGO (Fig. 1b, inset), suggesting model resolution and source distribution are, unlike the NH/SH OH ratio, not strong drivers for the NH–SH CH3CCl3 concentration gradient. Generally, the ratio of annual mean MHD–CGO CH3CCl3 gradients to annual mean concentrations is extremely stable at 2.87 ± 0.41% for AGAGE gas chromatograph–multi detector (GC–MD) data during the 2000s. The ACTM_0.99 simulation well predicted this ratio (2.88 ± 0.19%), whereas it is only 0.54 ± 0.22% for ACTM_1.26. The ratio of the MHD–CGO gradient to the mean CH3CCl3 concentration decreased rapidly in the 1990s, from ∼34% in 1990 to ∼2.9% in 1999 and the 2000s, in proportion to the decrease in global total emissions, mostly occurring in the NH.

Because no formal emission inventory of CH3CCl3 exists, uncertainties in the NH/SH emission ratio and in the global totals should be quantified. We find that the MHD–CGO CH3CCl3 differences are not particularly sensitive to NH/SH emission ratios of 10 or greater (∼16.6 in the ‘Control’ case) in the period 2004–2011. To assess the uncertainties contributed by the global mean OH concentration and total CH3CCl3 emission jointly, we derive a linear relationship between the two (percentage lifetime change = −3.9 × percentage emission change) for simulating the observed CH3CCl3 growth rate (Extended Data Fig. 4). We find that ACTM_0.99 produces a minimum for the model-observation mismatch (J) for the magnitude of global emission and chemical loss given as the ‘Control’ case (Fig. 2). A slightly larger mismatch minimum is observed for ACTM_1.26 when we consider +20% chemical loss and +78% global CH3CCl3 emissions during the period 2004–2011 (Fig. 2a, b). However, a significant increase in emission and loss deteriorates the agreement between simulated and observed seasonal cycle amplitudes in the CH3CCl3 concentration difference between MHD–CGO or Alert, Canada (ALT), and Palmer Station, Antarctica (PSA) (Fig. 2c). Thus significantly larger global CH3CCl3 emissions than considered in the ‘Control’ case can be ruled out for the period 2004–2011, as opposed to the period of the late 1990s (ref. 8), and therefore the possibility of significantly more OH in the NH than in the SH is also deemed inconsistent with these observations and their simulation.

Figure 2: Role of global total emissions and chemical loss on inter-site CH3CCl3 differences.
figure 2

a, Model–measurement mismatch (derived as J = √{[(CN − CS)model − (CN − CS)measured]2/N}; CN and CS are CH3CCl3 concentrations in the NH and SH, respectively, and N is the number of data points) as a function of total emissions (E) and chemical loss (CL) due to global mean OH abundance varied together in a manner consistent with the observed global decline in CH3CCl3 concentrations. The mismatch is shown in terms of standard deviations of simulated ALT–PSA (black, ACTM_0.99; red, ACTM_1.26) and MHD–CGO (green, ACTM_0.99; blue, ACTM_1.26) CH3CCl3 concentration differences with respect to measurements as monthly averages over the period 2004–2011. b, c, Annual means of inter-site difference at monthly intervals (b) and peak-to-trough seasonal cycle amplitude in the inter-site difference (c), to decompose the contribution of E and CL to the model–measurement mismatch. The observed values are shown by horizontal purple lines for ALT-PSA and light blue lines for MHD-CGO. These results are independent of sampling network (MHD–CGO from AGAGE GC–MD and ALT–PSA from NOAA). All the sensitivity simulations were for 2001–2011. Simulations for 2001–2003 have been considered as spin-up, and are excluded for calculating statistics.

PowerPoint slide

Source data

Full size image

To exclude erroneous interhemispheric transport in the ACTM model, we briefly present results for SF6, which is purely anthropogenic and chemically inert in the troposphere, and thus comprises an excellent tracer of atmospheric transport given its relatively high emission rate relative to the global burden and well-known meridional emission distribution22,27,28. The meridional transport of SF6 in ACTM has been validated extensively using surface sites6,22. Further, fine-grained measurements of various species during High-performance Instrumented Airborne Platform for Environmental Research (HIAPER) Pole-to-Pole Observation (HIPPO) campaigns13 represent zonal mean cross-sections of the troposphere (Extended Data Fig. 5). In general, ACTM-simulated SF6 matches the observations well along all flight paths: the simulated SF6 values frequently lie within the variability observed by the three instruments that recorded SF6 on the HIPPO aircraft (Extended Data Fig. 6). The ACTM simulated NH–SH differences agree within ±0.02 parts per trillion (p.p.t.) or ±6% of the observed values for each HIPPO campaign, averaged over all latitudes and 1–3 km altitude for the three instruments (Extended Data Table 3).

The effect of the NH/SH OH ratio on NH–SH CH3CCl3 concentration gradients is well supported by the HIPPO measurements over the Pacific Ocean. The ACTM_0.99 simulated CH3CCl3 meridional gradients, given the ‘Control’ emissions (NH–SH = 0.19 p.p.t. averaged over all HIPPO campaigns), are in close agreement with the HIPPO observations (0.21 p.p.t.), with model-observation differences mostly <0.2 p.p.t. (or <2%) at all latitudes, during all seasons (Fig. 3). However, the results using ACTM_1.26 show systematically smaller NH–SH CH3CCl3 differences (0.05 p.p.t. averaged over all HIPPO campaigns), suggesting that the loss due to the CH3CCl3 + OH reaction is too high in the NH troposphere or too low in the SH. The poleward increase in CH3CCl3 from the SH subtropics is caused mainly by the greater abundance of OH in the SH tropics than in the subtropics in austral summer (HIPPO 1, 2 and 3).

Figure 3: Meridional gradients of CH3CCl3 during five HIPPO campaigns suggest that the NH/SH OH ratio is close to 1.
figure 3

Latitudinal distributions of CH3CCl3 are shown as measured from the Advanced Whole Air Sampling (AWAS) flask air (black), and as simulated by ACTM_0.99 (red) and ACTM_1.26 (blue) with ‘Control’ global emissions and global mean OH concentrations. a, HIPPO 1, 12–23 January 2009; b, HIPPO 3, 26 March to 15 April 2010; c, HIPPO 4, 16 June to 10 July 2011; d, HIPPO 5, 19 August to 8 September 2011; e, HIPPO 2, 2–21 November 2009. The panels are arranged in seasonal order. The median concentrations are shown at 5° latitude intervals for a 1–4-km altitude range for the meridional gradients. The y-axis range is maintained at 1.5 p.p.t., however, the absolute values differ, reflecting time differences between the campaigns. Both ACTM results are adjusted to the mean observed values corresponding to >25° S and the altitude range 1–4 km for each of the HIPPO campaigns separately (+0.07, +0.05, −0.24, +0.05 and −0.05 for ACTM_0.99, and −0.90, −0.85, −1.09, −0.75 and −0.80 for ACTM_1.26 for HIPPO 1–5, respectively), to allow for uncertainties in decadal emissions and lifetimes of CH3CCl3 (and bias in concentration gradients for ACTM_1.26), but this systematic shift with the SH reference does not affect the meridional gradient northward of 25° S.

PowerPoint slide

Source data

Full size image

For estimating the possible range of the annual mean NH/SH OH ratio, we performed nine sensitivity simulations using synthetic OH distributions (Extended Data Table 2b) for the period 2001–2011. The synthetic OH distributions are prepared by adding and subtracting sine functions with a phasing of 2 × latitude for ACTM_0.99 and mixing the two OH distributions. Figure 4 shows the dependence of average CH3CCl3 concentration gradients between NH and SH on the NH/SH OH ratio. Using the linear fits (shown as lines in Fig. 4) to the model simulations (open symbols), annual NH/SH OH ratios are calculated on the basis of the observed CH3CCl3 concentration gradients (filled symbols) averaged over the period 2004–2011 (2009–2011 for HIPPO). The average (±1σ of annual values) of NH/SH OH ratios are estimated to be 0.98 ± 0.12, 0.97 ± 0.11 and 0.96 ± 0.63 using the CH3CCl3 concentration gradients between MHD and CGO from AGAGE (average of GC–MD and Medusa instruments), ALT and PSA from NOAA flask samples, and NH and SH averages (1–4 km altitude, latitudes >30°) from HIPPO, respectively. Combining the results from the AGAGE and NOAA surface network, the decadal average NH/SH OH ratio is 0.97 ± 0.12. The decadal average NH/SH OH ratio estimated from the surface network is in excellent agreement with that estimated from five HIPPO campaigns covering greater geographical areas and vertical extents of both hemispheres but sampling over a briefer period. The uncertainty of about ±13% for the surface networks includes the measurement and model errors (<3%), emission uncertainties, and interannual and seasonal variations in OH within each of the hemispheres, although relative contributions of emission uncertainty and OH variations cannot be quantified.

Figure 4: Estimation of the NH/SH OH concentration ratio from CH3CCl3 interhemispheric gradients.
figure 4

NH–SH CH3CCl3 concentration differences for different measurement data sets (black, AGAGE GC-MD, MHD–CGO; red, AGAGE Medusa, MHD–CGO; blue, NOAA flask, ALT–PSA; green, HIPPO, between 30° N and 30° S) based on the ACTM sensitivity simulations for various NH/SH OH ratios during the period 2004–2011 considering the case of ‘Control’ global total emissions and global mean OH concentrations. ACTM simulation results (open symbols) using different OH distributions (ACTM_0.99; ACTM_0.99 ± sine functions; ACTM_0.99 and ACTM_1.26 mixtures; and ACTM_1.26; Extended Data Table 2b) are sampled for the AGAGE GC–MD, AGAGE Medusa, NOAA flasks and HIPPO sampling locations. The observed NH–SH CH3CCl3 concentration gradients (closed symbols) are calculated using MHD and CGO for AGAGE, ALT and PSA for NOAA flasks, and averages of data in the latitudes polewards of 30° in the altitude range 1–4 km for HIPPO, which are then used for calculating the NH/SH OH ratio with the fitted lines (GC–MD, y = 1.556 − 1.191x; Medusa, y = 1.589 − 1.188x; flask, y = 1.589 − 1.188x; HIPPO, y = 0.899 − 0.481x). Model outputs for the Medusa and GC–MD measurements differ slightly because gaps in the records from the two instruments during the 2004–2011 time period are not coincident.

PowerPoint slide

Source data

Full size image

The precise and well-calibrated measurements from different networks, combined with a transport model that accurately describes interhemispheric transport and modelled emissions, allow us to conclude that a global OH distribution with substantially more OH in the NH is inconsistent with CH3CCl3 observations. Our result of an NH/SH OH ratio of close to 1 is in strong contrast to higher modelled OH in the NH4. Our results may be explained in various ways. Either NH OH sources are overestimated, possibly owing to O3 that is biased high in the NH4, or NH OH sinks are underestimated29. Alternatively, SH OH sources may be underestimated30 or SH OH sinks are overestimated (less likely). Further refinements of OH distributions and OH budgets are required for an accurate estimation of surface emissions of many important short-lived species that affect the Earth’s radiative budget and air pollution chemistry. For example, to match the observed interhemispheric gradient in atmospheric CH4 (based on data from the TransCom experiment6), CH4 emissions in the NH have to be increased from 398 to 430 Tg of CH4 per year, and decreased from 151 to 119 Tg of CH4 per year in the SH if ACTM_1.26 OH is used instead of ACTM_0.99. Our results also imply that top-down emission estimates of reactive species (for example, CO, NOx and SOx) in key emitting countries in the NH are probably overestimated if OH fields are used with an NH/SH OH ratio much larger than 1.

Methods

The JAMSTEC ACTM

The CCSR/NIES/FRCGC (Center for Climate System Research/National Institute for Environmental Studies/Frontier Research Center for Global Change) atmospheric general circulation model (AGCM)-based ACTM is developed for simulations of long-lived gases in the atmosphere6,22. Scenarios of surface emissions are used in ACTM to simulate atmospheric concentrations of methyl chloroform and SF6 to derive constraints on the model transport and to determine the NH/SH ratio of OH abundance. The ACTM simulations analysed here have a horizontal resolution of T42 spectral truncation (∼2.8° × 2.8°) with 67 sigma-pressure levels in the vertical and model top at ∼90 km. The horizontal winds and temperature of ACTM are nudged with Japan Meteorological Agency Reanalysis (JRA) data products31.

We use predefined tropospheric OH concentrations23, after scaling the global totals by 0.92 to simulate global CH3CCl3 decay rates as in the TransCom-CH4 experiment6 (Extended Data Fig. 1a; hereafter called ACTM_0.99 because the NH/SH OH ratio is 0.99). The hemispheres are delineated at the Equator. A previous study23 used a semi-empirical method to estimate seasonally varying OH fields by using observed distributions of O3, H2O, NOt (NO2 + NO + 2N2O5 + NO3 + HNO2 + HNO4), CO, hydrocarbons, temperature and cloud optical depth. To test the sensitivity to different interhemispheric OH distributions, we also consider tropospheric OH simulated by CHASER with online chemistry21, scaled by 0.88 to simulate global CH3CCl3 decay rates more accurately6 (Extended Data Fig. 1b; hereafter called ACTM_1.26 because the NH/SH OH ratio is 1.26). For both cases, stratospheric OH distributions are adopted from the full stratospheric chemistry model version of the ACTM32. The NH/SH OH ratios for ACTM_0.99 and ACTM_1.26 are estimated to be 0.87 and 1.18, respectively, when the hemispheres are separated at the ITCZ (named observed ‘dynamical’ Equator) following the location of maximum tropical rainfall33 or the maximum in meridional gradient of ACTM_0.99 simulated concentration of HFC-134a (CH2FCF3) (named modelled ‘chemical’ Equator). Annual mean locations of the ITCZ defined by CMAP and HFC-134a are at 4.2° N and 4.3° N, respectively, suggesting good performance of ACTM interhemispheric transport.

ACTM simulation setup for SF6, CH3CCl3 and HFC-134a.

We took SF6 and HFC-134a emission distributions from the Emissions Database for Global Atmospheric Research (EDGAR; version 4.2)25 for 1980–2008 (Extended Data Table 1). Global total SF6 emissions are scaled to those estimated previously34. The 2008 emission distributions are repeated for 2009–2011, but global total emissions are scaled to maintain the 2007–2008 increase rate for HFC-134a, and for SF6 in proportion with fossil fuel emissions trends (Carbon Dioxide Information Analysis Center, http://cdiac.ornl.gov/trends/emis/meth_reg.html). The spatial distributions in EDGAR4.2 emissions are based on proxy data sets, mainly urban population (G. Maenhout, personal communication, April 2013). The EDGAR4.2 CH3CCl3 emissions are from ref. 24 (1980–2000), and are assumed to decrease exponentially in subsequent years, following E(t) = E0 exp(−t/τ), where emissions E0 are at t = 0 (year 2000) and lifetime τ = 5 years. The global totals and spatial distributions of CH3CCl3 emissions are identical to the TransCom-CH4 experiment for 1987–2008 (‘Control’ case) (prepared by M. Krol)6. We also prepared alternative CH3CCl3 emission distribution maps using yearly country consumption (United Nations Environment Programme, UNEP). Emissions within a country are gridded at 1° × 1° on the basis of population distributions. Contrasting CH3CCl3 emission distributions used in the ‘Control’ and UNEP cases are depicted in Extended Data Fig. 2a, b. The time evolution of global total emissions is shown in Extended Data Fig. 2c, d and Extended Data Table 1.

We have considered these equations to simulate OH chemistry, deposition and stratospheric photolysis and rate constants for CH3CCl3 and HFC-134a:

Photochemical losses are calculated online in ACTM in accordance with recommended temperature-dependent reaction and wavelength-dependent photolysis rates35. We have not considered uncertainties in reaction rates, which can be as high as 50% for some of the species. CH3CCl3 deposition on the oceanic surface is parameterized6 to give a partial global lifetime of 197 years. The O(1D) concentrations are calculated in ACTM, and the monthly mean Cl climatology is from ref. 32. Zonal mean gridded CH3CCl3 lifetimes due to equations (1) and (2) are shown in Extended Data Fig. 1c. The species lifetime at steady state (τSS)36 can be defined as burden/loss (M × ΣNgas/M × ΣLgas), where N and L are the moles present and lost in a model grid box, respectively, and M is the molecular mass. We calculated τSS for CH3CCl3 (all loss by reaction with OH and photolysis) as 4.91 ± 0.03 (s.d. over multiple years) and for HFC-134a as 13.65 ± 0.11 years in ACTM_0.99 (average 2001–2010). Their lifetimes are slightly longer (5.19 ± 0.03 and 14.23 ± 0.11 years, respectively) in ACTM_1.26. Global CH3CCl3 lifetimes are also estimated with a different method: d(burden)/dt = emission − burden/τtotal, as compared in Extended Data Fig. 2c with τSS. Good agreements between the two lifetimes after 2000 indicate that although some CH3CCl3 emissions still occur, τtotal estimation depends less on emission trends during our analysis period (2004–2011). Small interannual lifetime variations after ∼2000 are caused by the temperature dependence of equation (1). Both ACTM estimated CH3CCl3 lifetimes agree within 0.2 years with the 5.0-year observationally defined lifetime3 and with other independent estimates, for example, 4.87–5.23 years26.

Measurements and data processing

Simulations span 1980–2011, after a 10-year spin-up to stabilize the stratospheric distribution. Three-dimensional model output is sampled hourly for comparison with observations. We use in situ data from five AGAGE observatories11 (from two types of instrument: the GC–MD11 and the Medusa gas chromatograph–mass spectrometer (GC–MS)37) and flask measurements from the NOAA halocarbon and other trace species (HATS) network6,8. For surface measurement sites see Extended Data Fig. 2b and Extended Data Table 2a. We calculate concentration gradients for each network separately and therefore do not consider calibration scale differences.

We also use data from five HIPPO campaigns13 (http://hippo.ornl.gov; HIPPO_all_missions_merge_10s_20121129.tbl). Model results are interpolated in space and time along the aircraft trajectory. HIPPO SF6 data are from three instruments (NOAA whole air sampler (NWAS) using GC–MS analysis, Unmanned Aircraft Systems (UAS) Chromatograph for Atmospheric Trace Species (UCATS), PAN and other Trace Hydrohalocarbon Experiment (PANTHER) GC-ECD). CH3CCl3 data are from the AWAS system using GC–MS analysis (University of Miami). Simulated and measured values are binned by 1 km vertically and 2.5° latitudinally.

The model was sampled at hourly intervals and we chose the nearest model grid (within ∼1.45 degrees) for each site. Monthly or annual means are calculated by including model results at the time of measurements, unless stated otherwise.

Model–measurement comparison (surface sites)

Because the NH/SH SF6 emission ratio is ∼36 and SF6 has no tropospheric loss, the observed and simulated SF6 concentrations in the NH exceed SH concentrations. Given accurate emission estimates, SF6 observations can therefore be used to verify the interhemispheric mixing times22,34. Meridional CH3CCl3 gradients are more complex than those of SF6 as a result of reaction with seasonally varying OH. The higher NH CH3CCl3 concentrations than those of the SH for ACTM_0.99 are caused by small lingering NH emissions combined with equal OH removal in the SH and NH (see Extended Data Fig. 3). In contrast, higher NH emissions are compensated for by higher NH OH loss in ACTM_1.26.

Considering ‘Control’ global emissions and global OH concentrations, ACTM_0.99 CH3CCl3 simulations are in very good agreement with AGAGE and NOAA surface measurements (Fig. 1 and Extended Data Fig. 3c, d); however, ACTM_1.26 always underestimates measured meridional CH3CCl3 gradients. These comparisons show that the CH3CCl3 change rate is well simulated from the 1990s to the present, suggesting that assumed global total emissions comply well with assumed global OH totals in this ‘Control’ scenario (note that annual mean OH levels are assumed constant from year to year, with only seasonal variation). Spatial distributions of emissions and emission magnitudes are more difficult to validate (this is addressed later). Extended Data Fig. 3c, d show concentration differences for eight NOAA sites relative to Palmer Station (65° S), suggesting that spatial emission distributions are reasonably estimated for the past 15 years. CH3CCl3 at SMO is found to be consistently lower than at PSA (Extended Data Fig. 3c, d, middle column)8, as a result of higher CH3CCl3 loss in the SH tropics (SMO at 14° S) than in the SH mid and high latitudes (PSA at 65° S). Both ACTM_0.99 and ACTM_1.26 capture the intra-SH gradient, suggesting that intra-SH OH gradients and intra-hemispheric transport of CH3CCl3 are well represented in both ACTM versions.

CH3CCl3 interhemispheric gradients are sensitive to the NH/SH OH ratio and emission magnitudes and may be calculated differently depending on the model’s horizontal resolution. We have performed test simulations using, first, two different emission maps (‘Control’ and UNEP-based) with geographically differing emissions, for example India and China versus Europe and the USA (Extended Data Fig. 2a, b), and second, varied horizontal resolutions (T42 and T106 spectral truncations). MHD–CGO CH3CCl3 gradients are simulated consistently using ACTM_0.99 since the later 2000s for all four test simulation cases (Fig. 1).

Sensitivity of the NH/SH OH ratio to assumed CH3CCl3 emission magnitudes and NH/SH emission ratio (ERs)

We have simulated MHD–CGO CH3CCl3 gradients using ACTM_0.99 for NH/SH CH3CCl3 ERs from 0.23 (more SH emissions) to infinity (only NH emissions). Extended Data Fig. 2e suggests that MHD–CGO CH3CCl3 gradients are not very sensitive to NH/SH ER, when the ER is greater than that assumed for the ‘Control’ emission case. The NH/SH ER is maintained at 16.6 for our sensitivity simulations (2001–2011).

We have estimated how much change in global total emissions is needed for a given change in chemical loss rate, to fit the observed global mean CH3CCl3 decline. Since 2001, the atmosphere still contains a significant CH3CCl3 burden and one would expect the OH estimate to be relatively independent of emissions8. Simulations using ACTM_0.99 (Extended Data Fig. 4a, b) suggest that if emissions are wrong by 10% in 2001, the lifetime will be in error by 10/3.9 = 2.6%. A consistency in emission-lifetime relative errors (2001, 2006 and 2011) is also observed because the emission/burden ratios remained almost constant (Extended Data Table 1) due to the same e-folding time for emission and burden after 2001.

Both the observed global mean CH3CCl3 decline and MHD–CGO gradients are simulated well with ACTM_1.26 when global total emissions and global mean OH are adjusted by roughly +78% and +20%, respectively (Fig. 2). The seasonal cycle amplitude in NH–SH CH3CCl3 difference is an additional constraint on global emissions, which is due mainly to the seasonality in loss through reaction with OH (Extended Data Fig. 4c, d). Although this simulation (+78% emission; +20% OH) fits the average NH–SH concentration difference in CH3CCl3 only slightly worse than ACTM_0.99 (‘Control’) (see Fig. 2a, b), the simulation seasonal cycle amplitude (peak-to-trough height within each year) clearly deteriorates (Fig. 2c). Too high emissions and chemical loss strengths increase the seasonality of the concentration differences substantially more than observed for both site pairs. This is primarily because NH sites (MHD and ALT) experience larger increases in concentration due to emissions during the NH winter (lower OH) and larger decreases in concentration due to chemical loss during the NH summer (higher OH).

The best model–observation agreement is found for ACTM_0.99 with ‘Control’ cases of global total emissions and global mean OH for both the MHD–CGO and ALT–PSA CH3CCl3 concentration differences.

Apart from the poor simulation of the observed seasonal cycle, 78% higher global total emissions (compared with ‘Control’) are unlikely on the basis of CH3CCl3 production and consumption reported to UNEP (http://ozone.unep.org/new_site/en/ozone_data_tools_access.php). The ‘Control’ emissions derived here are typically 10–20% higher than those estimated from the UNEP database for individual years between 1990 and 2009, but not for 2010 and 2011, when UNEP reported a factor of 4.8 and 31.4, respectively, lower than the ‘Control’ global total emissions.

SF6 and CH3CCl3 interhemispheric and vertical gradients (HIPPO campaigns)

Measurements and modelling of long-lived gases have been widely used to infer air mass transport and mixing38,39,40,41. HIPPO provides a unique opportunity to test model transport, emission and chemistry of multiple chemical species of widely varying lifetimes13. We have used flights only over the central Pacific Ocean for this analysis, to avoid greater site representation error in coarse-resolution transport models (for example, 2.8° × 2.8° and ∼500 m height intervals in the troposphere for ACTM) near the source regions. The locations of all research flights (denoted by numbers) are plotted on top of precipitation maps (Extended Data Fig. 5a). Because the chosen measurement locations are away from major sources of anthropogenic species, the site representation errors in our coarse-resolution ACTM are minimal, and measurements over the middle of the Pacific Ocean are representative of zonal means (Extended Data Fig. 5b, c).

Comparison of the location of the sharp SF6 decrease (Extended Data Fig. 6a–h) with observed precipitation (Extended Data Fig. 5a) shows it to be coincident with the Pacific ITCZ. Over the NH mid to high latitudes, SF6 concentrations at 6 km are within 0.1 p.p.t. of surface values. With deeper penetrating convection and a higher tropopause in the tropics, values comparable to the surface occur up to 12 km. We observed a reversed vertical gradient in the SH low to mid latitudes, with SF6 concentrations aloft exceeding those at the surface. This reversed vertical gradient is interpreted in terms of the bulk of interhemispheric transport occurring in the mid to upper troposphere22,28,42,43,44.

Both observed meridional and vertical SF6 gradients are well simulated by ACTM (Extended Data Fig. 6i–r), mostly within 0.1 p.p.t., confirming that large-scale transport processes are well represented. Previously, such conclusions were drawn on the basis of limited data sets, measured at only a few surface sites6,22. CH3CCl3 simulation results using ACTM_0.99 agree well with HIPPO AWAS aircraft data (Fig. 3) when the ‘Control’ global emissions and global mean OH concentrations are considered, whereas ACTM_1.26 always underestimates meridional CH3CCl3 gradients (Extended Data Table 3), as for surface measurements.

Estimation of the NH/SH OH ratio from CH3CCl3 interhemispheric gradients

Sensitivity simulations are also carried out using nine synthetic OH distributions cases: three scenarios derived by combining the OH distributions used in ACTM_0.99 and ACTM_1.26, and six scenarios derived by manipulating the OH distributions used in ACTM_0.99 with a sine function (Extended Data Table 2b).

Extended Data Fig. 7 shows the differences in monthly and annual mean CH3CCl3 concentrations between MHD and CGO measured by AGAGE compared with ACTM simulations using the synthetic NH–SH OH distributions. It is clear that simulations with a NH/SH OH ratio of ∼1 show the closest agreement with the annual mean measurements for most years, except for 2007, when anomalously high (>0.7 p.p.t.) MHD–CGO CH3CCl3 differences were measured with the Medusa instrument. We performed the same analysis using ALT and PSA sites from the NOAA flask network, and differences in averaged HIPPO measurements in the NH (north of 30° N) and SH (south of 30° S) below 4 km. A summary of these results is shown in Fig. 4. The 2010 anomaly in AGAGE Medusa CH3CCl3 differences is not found for ALT–PSA (or MHD–CGO) using NOAA data (Extended Data Fig. 3c, d), underlining the advantages of two cooperating yet independent observational networks.

The ACTM sensitivity runs derived with ‘Control’ global emissions and global mean OH concentrations form tight linear relationships for each year, although NH–SH CH3CCl3 gradients vary significantly with time in proportion to total CH3CCl3 emissions (Extended Data Fig. 7e, f). Using the ‘Control’ CH3CCl3 emission scenario, we have estimated 2004–2011 average (±1σ of annual values) NH/SH OH ratios of 0.99 ± 0.10, 0.97 ± 0.13, 0.97 ± 0.11 and 0.96 ± 0.63 using CH3CCl3 concentrations from AGAGE GC–MD, AGAGE Medusa, NOAA flasks and the five HIPPO campaigns, respectively. The large variability (±0.63) between the HIPPO campaigns is caused by the seasonal OH cycle and transport as well as uncertainties in emissions. However, because five HIPPO campaigns cover the full seasonal cycle, the mean value can be considered as annually representative. The average NH/SH OH ratio estimated from surface observation networks (0.97) agrees very well (within in 2%) with the mean value estimated from the five HIPPO campaigns (0.96).

We use the standard deviation of individual annual mean values as an estimate of uncertainty in the long-term mean value. The uncertainty of about ±13% in the NH/SH OH ratio for the surface networks includes the measurement and model errors (<3%), emission uncertainties and interannual and seasonal variations in OH within both hemispheres, but the relative contributions of emission uncertainty and OH variations cannot be quantified. The model transport errors from SF6 interhemispheric gradients are estimated to be 3% when averaged over five HIPPO campaigns, which is likely to decrease for longer time-averaging to ∼1% (see ref. 6 for 2004–2007). An integrated estimate of measurement uncertainty has been produced26, essentially using the standard deviation of the differences between global NOAA and AGAGE data. Their quoted 1% gives us an upper limit to the attainable measurement uncertainty, which is consistent across years. It includes the error in calibration propagation. However, because we are unable to quantify interannual variations in CH3CCl3 emissions accurately45, the effects of interannual OH variations46 within each hemisphere on the NH/SH ratio are not addressed here. For example, a recent study47 shows that lightning mainly affects the air chemistry at ∼300 mbar over the tropics, although the systematic difference in ACTM_0.99 and ACTM_1.26 mainly occur below 500 mbar in the extratropical to mid-latitude regions of the NH.