A multi-model intercomparison of halogenated very short-lived substances (TransCom-VSLS): linking oceanic emissions and tropospheric transport for a reconciled estimate of the stratospheric source gas injection of bromine

The first concerted multi-model intercomparison of halogenated very short-lived substances (VSLS) has been performed, within the framework of the ongoing Atmospheric Tracer Transport Model Intercomparison Project (TransCom). Eleven global models or model variants participated (nine chemical transport models and two chemistry– climate models) by simulating the major natural bromine VSLS, bromoform (CHBr3) and dibromomethane (CH2Br2), over a 20-year period (1993–2012). Except for three model simulations, all others were driven offline by (or nudged to) reanalysed meteorology. The overarching goal of TransComVSLS was to provide a reconciled model estimate of the stratospheric source gas injection (SGI) of bromine from these gases, to constrain the current measurement-derived range, and to investigate inter-model differences due to emissions and transport processes. Models ran with standardised idealised chemistry, to isolate differences due to transport, and we investigated the sensitivity of results to a range of VSLS emission inventories. Models were tested in their ability to reproduce the observed seasonal and spatial distribution of VSLS at the surface, using measurements from NOAA’s long-term global monitoring network, and in the tropical troposphere, using recent aircraft measurements – including high-altitude observations from the NASA Global Hawk platform. The models generally capture the observed seasonal cycle of surface CHBr3 and CH2Br2 well, with a strong model– measurement correlation (r ≥ 0.7) at most sites. In a given model, the absolute model–measurement agreement at the surface is highly sensitive to the choice of emissions. Large inter-model differences are apparent when using the same emission inventory, highlighting the challenges faced in evaluating such inventories at the global scale. Across the ensemble, most consistency is found within the tropics where most of the models (8 out of 11) achieve best agreement to surface CHBr3 observations using the lowest of the three CHBr3 emission inventories tested (similarly, 8 out of 11 models for CH2Br2). In general, the models reproduce observations of CHBr3 and CH2Br2 obtained in the tropical tropopause layer (TTL) at various locations throughout the Pacific well. Zonal variability in VSLS loading in the TTL is generally consistent among models, with CHBr3 (and to a lesser extent CH2Br2) most elevated over the tropical western Pacific during boreal winter. The models also indicate the Asian monsoon during boreal summer to be an important pathway for VSLS reaching the stratosphere, though the strength of this signal varies considerably among models. We derive an ensemble climatological mean estimate of the stratospheric bromine SGI from CHBr3 and CH2Br2 of 2.0 (1.2–2.5) ppt, ∼ 57 % larger than the best estimate from the most recent World Meteorological Organization (WMO) Ozone Assessment Report. We find no evidence for a long-term, transport-driven trend in the stratospheric SGI of bromine over the simulation period. The transport-driven interannual variability in the annual mean bromine SGI is of the order of ±5 %, with SGI exhibiting a strong positive correlation with the El Niño–Southern Oscillation (ENSO) in the eastern Pacific. Overall, our results do not show systematic differences between models specific to the choice of reanalysis meteorology, rather clear differences are seen related to differences in the implementation of transport processes in the models.

Models were tested in their ability to reproduce the observed seasonal and spatial distribution of VSLS at the surface, using measurements from NOAA's long-term global monitoring network, and in the tropical troposphere, using recent aircraft measurements -including high altitude observations from the NASA Global Hawk platform. 15 The models generally capture the observed seasonal cycle of surface CHBr 3 and CH 2 Br 2 well, with a strong model-measurement correlation (r ≥0.7) at most sites. In a given model, the absolute model-measurement agreement at the surface is highly sensitive to the choice of emissions. Large inter-model differences are apparent when using the same emission inventory, highlighting the challenges faced in evaluating such inventories at the global scale. Across the ensemble, most 20 consistency is found within the tropics where most of the models (8 out of 11) achieve best agreement to surface CHBr 3 observations using the lowest of the three CHBr 3 emission inventories tested (similarly, 8 out of 11 models for CH 2 Br 2 ). In general, the models reproduce well observations of CHBr 3 and CH 2 Br 2 obtained in the tropical tropopause layer (TTL) at various locations throughout the Pacific. Zonal variability in VSLS loading in the TTL is generally consistent among models, 25 with CHBr 3 (and to a lesser extent CH 2 Br 2 ) most elevated over the tropical West Pacific during boreal winter. The models also indicate the Asian Monsoon during boreal summer to be an important pathway for VSLS reaching the stratosphere, though the strength of this signal varies considerably among models. teorology, rather clear differences are seen related to differences in the implementation of transport processes in the models.

Introduction
Halogenated very short-lived substances (VSLS) are gases with atmospheric lifetimes shorter than, 40 or comparable to, tropospheric transport timescales (∼6 months or less at the surface). Naturallyemitted VSLS, such as bromoform (CHBr 3 ), have marine sources and are produced by phytoplankton (e.g. Quack and Wallace, 2003) and various species of seaweed (e.g. Carpenter and Liss, 2000) -a number of which are farmed for commercial application (Leedham et al., 2013). Once in the atmosphere, VSLS (and their degradation products) may ascend to the lower stratosphere (LS), where 45 they contribute to the inorganic bromine (Br y ) budget (e.g. Pfeilsticker et al., 2000;Sturges et al., 2000) and thereby enhance halogen-driven ozone (O 3 ) loss (Salawitch et al., 2005;Feng et al., 2007;Sinnhuber et al., 2009;Sinnhuber and Meul, 2015). On a per molecule basis, O 3 perturbations near the tropopause exert the largest radiative effect (e.g. Lacis et al., 1990;Forster and Shine, 1997;Riese et al., 2012) and recent work has highlighted the climate relevance of VSLS-driven O 3 loss in 50 this region (Hossaini et al., 2015a).
Quantifying the contribution of VSLS to stratospheric Br y (Br VSLS y ) has been a major objective of numerous recent observational studies (e.g. Dorf et al., 2008;Laube et al., 2008;Brinckmann et al., 2012;Sala et al., 2014;Wisher et al., 2014) and modelling efforts (e.g. Warwick et al., 2006;Hossaini et al., 2010;Liang et al., 2010;Aschmann et al., 2011;Tegtmeier et al., 2012;Hossaini et al., 2012bHossaini et al., , 201355 Aschmann and Sinnhuber, 2013;Fernandez et al., 2014). However, despite a wealth of research, Br VSLS y remains poorly constrained, with a current best-estimate range of 2-8 ppt reported in the most recent World Meteorological Organization (WMO) Ozone Assessment Report (Carpenter and Reimann, 2014). Between 15% and 76% of this supply comes from the stratospheric source gas injection (SGI) of VSLS; i.e. the transport of a source gas (e.g. CHBr 3 ) across the tropopause, followed by its 60 breakdown and in-situ release of Br VSLS y in the LS. The remainder comes from the troposphere-tostratosphere transport of both organic and inorganic product gases, formed following the breakdown of VSLS below the tropopause; termed product gas injection (PGI).
Owing to their short tropospheric lifetimes, combined with significant spatial and temporal inhomogeneity in their emissions (e.g. Carpenter et al., 2005;Archer et al., 2007;Orlikowska and Schulz-Bull, global-scale emission inventories of CHBr 3 and CH 2 Br 2 are poorly constrained, owing to a paucity of observations used to derive their surface fluxes , contributing significant uncertainty to model estimates of Br VSLS y (Hossaini et al., 2013). Given the uncertainties outlined above, it is unclear how well preferential transport pathways of VSLS to the LS are represented in 75 global scale models.
Strong convective source regions, such as the tropical West Pacific during boreal winter, are likely important for the troposphere-to-stratosphere transport of VSLS (e.g. Levine et al., 2007;Aschmann et al., 2009;Pisso et al., 2010;Hossaini et al., 2012b;Liang et al., 2014). The Asian Monsoon also represents an effective pathway for boundary layer air to be rapidly transported to the LS 80 (e.g. Randel et al., 2010;Vogel et al., 2014;Orbe et al., 2015;Tissier and Legras , 2016), though its importance for the troposphere-to-stratosphere transport of VSLS is largely unknown, owing to a lack of observations in the region. While global models simulate broadly similar features in the spatial distribution of convection, large inter-model differences in the amount of tracers transported to the tropopause have been reported by Hoyle et al. (2011), who performed a model intercomparison 85 of idealised ("VSLS-like") tracers with a uniform surface distribution. In order for a robust estimate of the stratospheric SGI of bromine, it is necessary to consider spatial variations in VSLS emissions, and how such variations couple with transport processes. However, a concerted model evaluation of this type has yet to be performed.
Over a series of two papers, we present results from the first VSLS multi-model intercompar- 90 ison project (TransCom-VSLS). The TransCom initiative was setup in the 1990s to examine the performance of chemical transport models. Previous TransCom studies have examined non-reactive tropospheric species, such as sulphur hexafluoride (SF 6 ) (Denning et al., 1999) and carbon dioxide (CO 2 ) (Law et al., 1996(Law et al., , 2008. Most recently, TransCom projects have examined the influence of emissions, transport and chemical loss on atmospheric CH 4 (Patra et al., 2011) and N 2 O 95 (Thompson et al., 2014). The overarching goal of TransCom-VSLS was to constrain estimates of Br V SLS y , towards closure of the stratospheric bromine budget, by (i) providing a reconciled climatological model estimate of bromine SGI, to reduce uncertainty on the measurement-derived range (0.7-3.4 ppt Br) -currently uncertain by a factor of ∼5 (Carpenter and Reimann, 2014) -and (ii) quantify the influence of emissions and transport processes on inter-model differences in SGI. In 100 this regard, we define transport differences between models as the effects of boundary layer mixing, convection and advection, and the implementation of these processes. The project was not designed to separate clearly the contributions of each transport component in the large model ensemble, but can be inferred as the boundary layer mixing affects tracer concentrations mainly near the surface, convection controls tracer transport to the upper troposphere and advection mainly distributes 105 tracers horizontally (e.g. Patra et al., 2009). Specific objectives were to (a) evaluate models against measurements from the surface to the tropical tropopause layer (TTL) and (b) examine zonal and seasonal variations in VSLS loading in the TTL. We also show inter-annual variability in the strato-spheric loading of VSLS (limited to transport) and briefly discuss possible trends related to the El Niño Southern Oscillation (ENSO). Section 2 gives a description of the experimental design and 110 an overview of participating models. Model-measurement comparisons are given in Sections 3.1 to 3.3. Section 3.4 examines zonal/seasonal variations in the troposphere-stratosphere transport of VSLS and Section 3.5 provides our reconciled estimate of bromine SGI and discusses inter-annual variability.

115
Eleven models, or their variants, took part in TransCom-VSLS. Each model simulated the major bromine VSLS, bromoform (CHBr 3 ) and dibromomethane (CH 2 Br 2 ), which together account for 77-86% of the total bromine SGI from VSLS reaching the stratosphere (Carpenter and Reimann, 2014). Participating models also simulated the major iodine VSLS, methyl iodide (CH 3 I), though results from the iodine simulations will feature in a forthcoming, stand-alone paper (Hossaini et al. 120 2016, in prep). Each model ran with multiple CHBr 3 and CH 2 Br 2 emission inventories (see Section 2.1) in order to (i) investigate the performance of each inventory, in a given model, against observations and (ii) identify potential inter-model differences whilst using the same inventory. Analogous to previous TransCom experiments (e.g. Patra et al., 2011), a standardised treatment of tropospheric chemistry was employed, through use of prescribed oxidants and photolysis rates (see Section 2.2).

125
This approach (i) ensured a consistent chemical sink of VSLS among models, minimising the influence of inter-model differences in tropospheric chemistry on the results, and thereby (ii) isolated differences due to transport processes. Long-term simulations, over a 20 year period (1993-2012), were performed by each model in order to examine trends and transport-driven inter-annual variability in the stratospheric SGI of CHBr 3 and CH 2 Br 2 . Global monthly mean model output over 130 the full simulation period, along with output at a higher temporal resolution (typically hourly) over measurement campaign periods, was requested from each group. A brief description of the models is given in Section 2.3 and a description of the observational data used in this work is given in Section 2.4. Figure 1 summarises the approach of TransCom-VSLS and its broad objectives.

135
Owing to significant differences in the magnitude and spatial distribution of VSLS emission fluxes, among previously published inventories (Hossaini et al., 2013), all models ran with multiple CHBr 3 and CH 2 Br 2 tracers. Each of these tracers used a different set of prescribed surface emissions. Tracers named "CHBr 3 _L", "CHBr 3 _O" and "CHBr 3 _Z" used the inventories of Liang et al. (2010), Ordóñez et al. (2012) and Ziska et al. (2013), respectively. These three studies also reported emis-140 sion fluxes for CH 2 Br 2 , and thus the same (L/O/Z) notation applies to the model CH 2 Br 2 tracers, as summarised in Table 1. As these inventories were recently described and compared by Hossaini et al. (2013), only a brief description of each is given below. Surface CHBr 3 /CH 2 Br 2 emission maps for each inventory are given in the Supplementary Information (Figures S1 and S2).
The Liang et al. (2010) inventory is a top-down estimate of VSLS emissions based on aircraft 145 observations, mostly concentrated around the Pacific and North America between 1996 and 2008.
Measurements of CHBr 3 and CH 2 Br 2 from the following National Aeronautics and Space Administration (NASA) aircraft campaigns were used to derive the ocean fluxes: PEM-Tropics, TRACE-P, INTEX, TC4, ARCTAS, STRAT, Pre-AVE and AVE. This inventory is aseasonal and assumes the same spatial distribution of emissions for CHBr 3 and CH 2 Br 2 . The Ordóñez et al. (2012) inventory 150 is also a top-down estimate based on the same set of aircraft measurements with the addition of the NASA POLARIS and SOLVE campaigns. This inventory weights tropical (±20 • latitude) CHBr 3 and CH 2 Br 2 emissions according to a monthly-varying satellite climatology of chlorophyll a (chl a), a proxy for oceanic bio-productivity, providing some seasonality to the emission fluxes. The Ziska et al. (2013) inventory is a bottom-up estimate of VSLS emissions, based on a compilation of 155 seawater and ambient air measurements of CHBr 3 and CH 2 Br 2 . Climatological, aseasonal emission maps of these VSLS were calculated using the derived sea-air concentration gradients and a commonly used sea-to-air flux parameterisation; considering wind speed, sea surface temperature and salinity (Nightingale et al., 2000).

160
Participating models considered chemical loss of CHBr 3 and CH 2 Br 2 through oxidation by the hydroxyl radical (OH) and by photolysis. These loss processes are comparable for CHBr 3 , with photolysis contributing ∼60% of the CHBr 3 chemical sink at the surface (Hossaini et al., 2010). For CH 2 Br 2 , photolysis is a minor tropospheric sink, with its loss dominated by OH-initiated oxidation.
These values are calculated based on [OH] = 1×10 6 molecules cm −3 , T = 275 K and with a global annual mean photolysis rate. For completeness, models also considered loss of CHBr 3 and CH 2 Br 2 by reaction with atomic oxygen (O( 1 D)) and chlorine (Cl) radicals. However, these are generally very minor loss pathways owing to the far larger relative abundance of tropospheric OH and the 170 respective rate constants for these reactions. Kinetic data (Table 1) was taken from the most recent Jet Propulsion Laboratory (JPL) data evaluation (Sander et al., 2011). Note, the focus and design of TransCom-VSLS was to constrain the stratospheric SGI of VSLS, thus product gases -formed following the breakdown of CHBr 3 and CH 2 Br 2 in the TTL (Werner et al. 2016, in prep) -and the stratospheric PGI of bromine was not considered.

175
Participating models ran with the same global monthly-mean oxidant fields. For OH, O ( 1 D) and Cl, these fields were the same as those used in the previous TransCom-CH 4 model intercomparison (Patra et al., 2011). Within the TransCom framework, these fields have been exten-sively used and evaluated and shown to give a realistic simulation of the tropospheric burden and lifetime of methane and also methyl chloroform. Models also ran with the same monthly-mean 180 CHBr 3 and CH 2 Br 2 photolysis rates, calculated offline from the TOMCAT chemical transport model (Chipperfield, 2006). TOMCAT has been used extensively to study the tropospheric chemistry of VSLS (e.g. Hossaini et al., 2010Hossaini et al., , 2012bHossaini et al., , 2015b and photolysis rates from the model were used to evaluate the lifetime of VSLS for the recent WMO Ozone Assessment Report (Carpenter and Reimann, 2014).

Participating models and output
Eight global models (ACTM, B3DCTM, EMAC, MOZART, NIES-TM, STAG, TOMCAT and UKCA) and 3 of their variants (see Table 2) participated in TransCom-VSLS. All the models are offline chemical transport models (CTMs), forced with analysed meteorology (e.g. winds and temperature fields), with the exception of EMAC and UKCA which are free-running chemistry-climate models 190 (CCMs), calculating winds and temperature online. The horizontal resolution of models ranged from In the vertical, the number of levels varied from 32 to 85, with various coordinate systems. A summary of the models and their salient features is given in Table 2. Note, these features do not necessarily link to model performance as evaluated in this work. Note also, approximately half of the models used ECMWF ERA-Interim meteorological data.

195
In terms of mean upwelling in the tropics, where stratospheric bromine SGI takes place, there is generally good agreement between the most recent major reanalysis products from ECMWF, JMA and NCEP (e.g. Harada et al., 2015). Therefore, we do not expect a particular bias in our results from use of ERA-Interim.
Three groups, the Karlsruhe Institute of Technology (KIT), the University of Leeds (UoL) and 200 the University of Cambridge (UoC), submitted output from an additional set of simulations using variants of their models. KIT ran the EMAC model twice, as a free running model (here termed "EMAC_F") and also in nudged mode (EMAC_N). The UoL performed two TOMCAT simulations, the first of which used the model's standard convection parameterisation, based on the mass flux scheme of Tiedtke (1989). The second TOMCAT simulation ("TOMCAT_conv") used archived 205 convective mass fluxes, taken from the ECMWF ERA-Interim reanalysis. A description and evaluation of these TOMCAT variants is given in Feng et al. (2011). In order to investigate the influence of resolution, the UoC ran two UKCA model simulations with different horizontal/vertical resolutions.
The horizontal resolution in the "UKCA_high" simulation was a factor of 4 (2 in 2 dimensions) greater than that of the standard UKCA run (Table 2).

210
All participating models simulated the 6 CHBr 3 and CH 2 Br 2 tracers (see Section 2.1) over a 20 year period; 01/01/1993 to 31/12/2012. This period was chosen as it (i) encompasses a range of field campaigns during which VSLS measurements were taken and (ii) allows the strong El Niño event of 1997/1998 to be investigated in the analysis of SGI trends. The monthly mean volume mixing ratio (vmr) of each tracer was archived by each model on the same 17 pressure levels, extending from 215 the surface to 10 hPa over the full simulation period. The models were also sampled hourly at 15 surface sites over the full simulation period and during periods of recent ship/aircraft measurement campaigns, described in Section 2.4 below. Note, the first two years of simulation were treated as spin up and output was analysed post 1995.  Table 3), were also considered. A description of these data is given in 235 Robinson et al. (2014). Briefly, in-situ measurements were made using the µ-Dirac gas chromatograph instrument with electron capture detection (GC-ECD) (e.g. Pyle et al., 2011). Measurements at TAW are for a single year (2009) only, making the observed record at this site far shorter than that at NOAA/ESRL stations discussed above.

Observational data and processing
A subset of models also provided hourly output over the period of the TransBrom and SHIVA 240 (Stratospheric Ozone: Halogen Impacts in a Varying Atmosphere) ship cruises. During both campaigns, surface CHBr 3 and CH 2 Br 2 measurements were obtained on-board the Research Vessel (R/V) Sonne. TransBrom sampled along a meridional transect of the West Pacific, from Japan to Australia, during October 2009 . SHIVA was a European Union (EU)funded project to investigate the emissions, chemistry and transport of VSLS (http://shiva.iup.uni-245 heidelberg.de/). Ship-borne measurements of surface CHBr 3 and CH 2 Br 2 were obtained in November 2011, with sampling extending from Singapore to the Philippines, within the South China Sea and along the northern coast of Borneo (Fuhlbrügge et al., 2015). The ship track is shown in Figure   2.

Aircraft
Observations of CHBr 3 and CH 2 Br 2 from a range of aircraft campaigns were also used ( Figure 2). As (i) the troposphere-to-stratosphere transport of air (and VSLS) primarily occurs in the tropics, and (ii) because VSLS emitted in the extratropics have a negligible impact on stratospheric ozone  (Sala et al., 2014;Fuhlbrügge et al., 2015). VSLS mea-  In this section, we evaluate the models in terms of (i) their ability to capture the observed seasonal cycle of CHBr 3 and CH 2 Br 2 at the surface and (ii) the absolute agreement to the observations. We focus on investigating the relative performance of each of the tested emission inventories, within a given model, and the performance of the inventories across the ensemble.

300
We first consider the seasonal cycle of CHBr 3 and CH 2 Br 2 at the locations given in Table 3.  Table 4, are >0.7 for at least 80% of the models at 7 of 11 NH sites. Second, at some sites, notably MHD, THD, CGO and PSA, the observed seasonal cycle of CHBr 3 is not captured well by virtually all of the models (see discussion below). Third, at most sites the amplitude of the seasonal cycle is 315 generally consistent across the models (within a few percent, excluding clear outliers). The cause of outliers at a given site are likely in part related to the model sampling error, including distance of a model grid from the measurement site and resolution (as was shown for CO 2 in Patra et al. (2008)).
These instances are rare for VSLS but can be seen in B3DCTM's output in Figure 3 for CHBr 3 at SMO. B3DCTM ran at a relatively coarse horizontal resolution (3.75 • ) and with less vertical layers 320 (40) compared to most other models. Note, it also has the simplest implementation of boundary layer mixing ( Table 2). The above behaviour is also seen at SMO but to a lesser extent for CH 2 Br 2 , for which the seasonal cycle is smaller (see below). The STAG model also produces distinctly different features in the seasonal cycle of both species at some sites (prominently at CGO, SMO and HFM).
We attribute these deviations to STAG's parameterisation of boundary layer mixing, noting that dif-325 ferences for CHBr 3 are greater at KUM than at MLO -two sites in very close proximity but with the latter elevated at ∼3000 metres above sea level (i.e. above the boundary layer). With respect to the observations, the amplitude of the seasonal cycle is either under-(e.g. BRW) or over-estimated (e.g. KUM) at some locations, by all of the models. This possibly reflects a more systematic bias in the prescribed CHBr 3 loss rate and/or relates to emissions, though this effect is generally small and 330 localised.
A similar analysis has been performed to examine the seasonal cycle of surface CH 2 Br 2 . Observed and simulated monthly mean anomalies, calculated in the same fashion as those for CHBr 3 above, are shown in Figure 4 and correlation coefficients are given in Table 5. The dominant chemical sink of CH 2 Br 2 is through OH-initiated oxidation and thus its seasonal cycle at most stations 335 reflects seasonal variation in [OH] and temperature. At most sites, this gives rise to a minimum in the surface mole fraction of CH 2 Br 2 during summer months, owing to greater [OH] and temperature, and thereby a faster chemical sink. Relative to CHBr 3 , CH 2 Br 2 is considerably longer-lived (and thus well mixed) near the surface, meaning the amplitude of the seasonal cycle is far smaller.
At most sites, most models capture the observed phase and amplitude of the CH 2 Br 2 seasonal cy-340 cle well, though as was the case for CHBr 3 , agreement in the southern hemisphere (SH, e.g. SMO, CGO, PSA) seems poorest. For example, at SMO and CGO only 40% of the models are positively correlated to the observations with r >0.5 ( Table 5). The NIES-TM model does not show major differences from other models for CHBr 3 , but outliers for CH 2 Cl 2 at SH sites (SMO to SPO) are apparent. We were unable to assign any specific reason for the inter-species differences seen for this 345 model.
At two sites (MHD and THD) almost none of the models reproduce the observed CHBr 3 seasonal cycle, exhibiting an anti-correlation with the observed cycle (see bold entries in Table 4). Here, the simulated cycle follows that expected from seasonality in the chemical sink. At MHD, seasonality in the local emission flux is suggested to be the dominant factor controlling the seasonal cycle of 350 surface CHBr 3 (Carpenter et al., 2005). This leads to the observed summer maximum (as shown in Figure 3) and is not represented in the models' CHBr 3 _L tracer which, at the surface, is driven by the aseasonal emission inventory of Liang et al. (2010). A similar summer maximum seasonal cycle is observed for CH 2 Br 2 , also not captured by the models' CH 2 Br 2 _L tracer. To investigate the sensitivity of the model-measurement correlation to the prescribed surface fluxes, multi-model mean 355 (MMM) surface CHBr 3 and CH 2 Br 2 fields were calculated for each tracer (i.e. for each emission inventory considered) and each site. Figure 5 shows calculated MMM r values at each site for CHBr 3 and CH 2 Br 2 . For CHBr 3 , r generally has a low sensitivity to the choice of emission fluxes at most sites (e.g. ALT, SUM, BRW, LEF, NWR, KUM, MLO, SPO), though notably at MHD, use of the Ziska et al. (2013) inventory (which is aseasonal) reverses the sign of r to give a strong positive 360 correlation (MMM r >0.70) against the observations. Individual model r values for MHD are given in Table S1 of the Supplementary Information. With the exception of TOMCAT, TOMCAT_CONV and UKCA_HI, the remaining 7 models each reproduce the MHD CHBr 3 seasonality well (with r >0.65). That good agreement is obtained with the Ziska aseasonal inventory, compared to the other aseasonal inventories considered, highlights the importance of the CHBr 3 emission distribution, with 365 respect to transport processes, serving this location. We suggest that the summertime transport of air that has experienced relatively large CHBr 3 emissions north/north-west of MHD is the cause of the apparent seasonal cycle seen in most models using the Ziska inventory (example animations of the seasonal evolution of surface CHBr 3 are given in the Supplementary Information to visualise this). Note also, the far better absolute model-measurement agreement obtained at MHD for models 370 using this inventory (Supplementary Figure S3). At other sites, such as TAW, no clear seasonality is apparent in the observed background mixing ratios of CHBr 3 and CH 2 Br 2 .
Here, the models exhibit little or no significant correlation to measured values and are unlikely to capture small-scale features in the emission distribution (e.g the contribution from local aquaculture) that conceivably contribute to observed levels of CHBr 3 and CH 2 Br 2 in this region (Robinson et al.,375 2014).

Absolute agreement
To compare the absolute agreement between a model (M) and an observation (O) value, for each monthly mean surface model-measurement comparison, the mean absolute percentage error (MAPE, equation 1) was calculated for each model tracer. Figure 6 shows the CHBr 3 and CH 2 Br 2 tracer that 380 provides the lowest MAPE (i.e. best agreement) for each model (indicated by the fill colour of cells).
The numbers within the cells give the MAPE value itself, and therefore correspond to the "best agreement" that can be obtained from the various tracers with the emission inventories that were tested.
For both CHBr 3 and CH 2 Br 2 , within any given model, no single emission inventory is able to provide the best agreement at all surface locations (i.e. from the columns in Figure 6). This was previously noted by Hossaini et al. (2013) using the TOMCAT model, and to some degree likely reflects the geographical coverage of the observations used to create the emission inventories. Hossaini et al. (2013) also noted significant differences between simulated and observed CHBr 3 and CH 2 Br 2 , using the same inventory; i.e. at a given location, low CHBr 3 MAPE (good agreement) does not necessarily accompany a corresponding low CH 2 Br 2 MAPE using the same inventory.
A key finding of this study is that significant inter-model differences are also apparent (i.e. see rows in Figure 6 grid). For example, for CHBr 3 , no single inventory performs best across the full range of models at any given surface site. TOMCAT and B3DCTM -both of which are driven by 395 ERA-Interim -agree on the best CHBr 3 inventory (lowest MAPE) at approximately half of the 17 sites considered. This analysis implies that, on a global scale, the "performance" of emission inventories is somewhat model-specific and highlights the challenges of evaluating such inventories.
Previous conclusions as to the best performing VSLS inventories, based on single model simulations (Hossaini et al., 2013), must therefore be treated with caution. When one considers that previ-400 ous modelling studies (Warwick et al., 2006;Liang et al., 2010;Ordóñez et al., 2012), each having derived different VSLS emissions based on aircraft observations, and having different tropospheric chemistry, report generally good agreement between their respective model and observations, our findings are perhaps not unexpected. However, we note also that few VSLS modelling studies have used long-term surface observations to evaluate their models, as performed here. This suggests any 405 attempts to reconcile estimates of global VSLS emissions, obtained from different modelling studies, need to consider the influence of inter-model differences.
As the chemical sink of VSLS was consistent across all models, the inter-model differences discussed above are attributed primarily to differences in the treatment and implementation of transport processes. This includes convection and boundary layer mixing, both of which can significantly in-410 fluence the near-surface abundance of VSLS in the real (Fuhlbrügge et al., 2013(Fuhlbrügge et al., , 2015 and model (Zhang et al., 2008;Feng et al., 2011;Hoyle et al., 2011) atmospheres, and are parameterised in different ways (Table 2). On this basis, it is not surprising that different CTM setups lead to differences in the surface distribution of VSLS, nor that differences are apparent between CTMs that use the same meteorological input fields. Indeed, such effects have also been observed in previous model 415 intercomparisons (Hoyle et al., 2011). Large-scale vertical advection, the native grid of a model and its horizontal/vertical resolution may also be contributing factors, though quantifying their relative influence was beyond the scope of TransCom-VSLS. At some sites, differences among emission inventory performance are apparent between model variants that, besides transport, are otherwise identical; i.e. TOMCAT and TOMCAT_CONV entries of Figure 6. This is significant as troposphere-to-stratosphere transport primarily occurs in the tropics and the when using the Liang et al. (2010) inventory, which also has the lowest global flux of the three inventories tested. For a number of models, a similar agreement is also obtained with Ordóñez et al.

430
For CH 2 Br 2 , the Ziska et al. (2013) inventory performs poorest across the ensemble (models generally overestimate CH 2 Br 2 with this inventory). Overall, the tropical MAPE for a given model is more sensitive to choice of emission inventory for CHBr 3 than CH 2 Br 2 (Figure 7). Based on each model's preferred inventory (i.e. from Figure 7), the tropical MAPE is generally ∼40% for CHBr 3 and <20% for CH 2 Br 2 (in most models). One model (STAG) exhibited a MAPE of >50% for both 435 species, regardless of the choice of emission inventory, and was therefore omitted from the subsequent model-measurement comparisons to aircraft data and also from the multi-model mean SGI estimate derived in Section 3.5.
For the 5 models that submitted hourly output over the period of the SHIVA (2011)   estimated in global-scale models. Note, TransBrom also sampled sub-tropical latitudes (see Figure   2).
Overall, our results show that most models capture the observed seasonal cycle and the magnitude of surface CHBr 3 and CH 2 Br 2 reasonably well, using a combination of emission inventories.
Generally, this leads to a realistic surface distribution at most locations, and thereby provides good 455 agreement between models and aircraft observations above the boundary layer; see Section 3.2 below.

Model-observation comparisons: free troposphere
We now evaluate modelled profiles of CHBr 3 and CH 2 Br 2 using observations from a range of recent aircraft campaigns (see Section 2.4). Note, for these comparisons, and from herein unless noted, 460 all analysis is performed using the preferred CHBr 3 and CH 2 Br 2 tracer for each model (i.e. preferred emissions inventory), as was diagnosed in the previous discussion (i.e. from Figure 7, see also Section 3.1.2). This approach ensures that an estimate of stratospheric bromine SGI, from a given model, is based on a simulation in which the optimal CHBr 3 /CH 2 Br 2 model-measurement agreement at the surface was acheived. The objective of the comparisons below is to show that the models 465 produce a realistic simulation of CHBr 3 and CH 2 Br 2 in the tropical free troposphere and to test model transport of CHBr 3 and CH 2 Br 2 from the surface to high altitudes, against that from atmospheric measurements. Intricacies of individual model-measurement comparison are not discussed.
Rather, Figure 10 compares MMM profiles (and the model spread) of CHBr 3 and CH 2 Br 2 mixing ratio to observed campaign means within the tropics (±20 • latitude). Generally model-measurement 470 agreement, diagnosed by both the campaign-averaged MAPE and the correlation coefficient (r) is excellent during most campaigns. For all of the 7 campaigns considered, the modelled MAPE for CHBr 3 is ≤35% (≤20% for CH 2 Br 2 ). The models also capture much of the observed variability throughout the observed profiles, including, for example, the signature "c-shape" of convection in the measured CHBr 3 profile from SHIVA and HIPPO-1 (panel (a), 2nd and 3rd rows of Figure 10).

475
Correlation coefficients between modelled and observed CHBr 3 are ≥0.8 for 5 of the 7 campaigns and for CH 2 Br 2 are generally >0.5.
It is unclear why model-measurement agreement (particularly the CHBr 3 MAPE) is poorest for the HIPPO-4 and HIPPO-5 campaigns. However, we note that at most levels MMM CHBr 3 and CH 2 Br 2 falls within ±1 standard deviation (σ) of the observed mean. Note, an underestimate of 480 surface CHBr 3 does not generally translate to a consistent underestimate of measured CHBr 3 at higher altitude. Critically, for the most part, the models are able to reproduce observed values of both gases well at ∼12-14 km, within the lower TTL. Recall that the TTL is defined as the layer between the level of main convective outflow (∼200 hPa, ∼12 km) and the tropical tropopause (∼100 hPa, ∼17 km) (Gettelman and Forster, 2002). For a given model, simulations using the non-485 preferred tracers (i.e. with different CHBr 3 /CH 2 Br 2 emission inventories, not shown), generally lead to worse model-measurement agreement in the TTL. This is not surprising as model-measurement agreement at the surface is poorer in those simulations (as discussed in Section 3.1.2.).
Overall, given the large spatial/temporal variability in observed VSLS mixing ratios, in part due to the influence of transport processes, global-scale models driven by aseasonal emissions and using 490 parameterised sub-grid scale transport schemes face challenges in reproducing VSLS observations in the tropical atmosphere. Yet despite this, we find that the TransCom-VSLS models generally provide a very good simulation of the tropospheric abundance of CHBr 3 and CH 2 Br 2 , particularly in the important tropical West Pacific region (e.g. SHIVA comparisons).  that the TransCom-VSLS models implement these processes in different ways (Table 2), it was not possible to detangle transport effects within the scope of this project. However, no systematic similarities/differences between models according to input meteorology were apparent. Examining the difference between UKCA_HI and UKCA_LO reveals that horizontal resolution is a significant factor. The UKCA_HI simulation shows a greater role of the Monsoon region, likely due to differences 605 in the distribution of surface emissions (e.g. along longer coastlines in the higher resolution model) with respect to the occurrence of convection, as shown by Russo et al. (2015). Overall, aircraft VSLS observations within this poorly sampled region are required in order to elucidate further the role of the Monsoon in the troposphere-to-stratosphere transport of brominated VSLS.

Stratospheric source gas injection of bromine and trends
In this section we quantify the climatological SGI of bromine from CHBr 3 and CH 2 Br 2 to the tropical LS and examine inter-annual variability.  (2014), though our best estimate is 0.72 ppt (57%) larger. The spread in the TransCom-VSLS models is also 37% lower than the Carpenter and Reimann (2014) range, suggesting that their measurement-derived range in 625 bromine SGI from CHBr 3 and CH 2 Br 2 is possibly too conservative, particularly at the lower limit ( Figure 16), and from a climatological perspective. We note that (i) the TransCom-VSLS estimate is based on models, shown here, to simulate the surface to tropopause abundance of CHBr 3 and CH 2 Br 2 well and (ii) represents a climatological estimate over the simulation period, 1995-2012.
The measurement-derived best estimate and range (i.e. that from Carpenter and Reimann (2014) because of the constraint on the contribution from CHBr 3 and CH 2 Br 2 , as discussed above.
Our uncertainty estimate on simulated bromine SGI (from the model spread) reflects inter-model variability, primarily due to differences in transport, but does not account for uncertainty on the chemical factors influencing the loss rate and lifetime of VSLS (e.g. tropospheric [OH]) -as all of the models used the same prescribed oxidants. However, Aschmann and Sinnhuber (2013)  We found no clear long-term transport-driven trend in the stratospheric SGI of bromine. Clearly, this result is limited to the study period examined and does not preclude potential future changes due to climate change, as suggested by some studies (e.g. Hossaini et al., 2012b). In terms of interannual variability, the simulated annual mean bromine SGI varied by ±5% around the climatological mean (panel (b) of Figure 16) over the simulation period (small in the context of total stratospheric 660 Br y , see above). Naturally, this encompasses inter-annual variability of both CHBr 3 and CH 2 Br 2 reaching the tropical LS. The latter of which is far smaller and given that CH 2 Br 2 is the larger contributor to SGI, dampens the overall inter-annual variability. Note, inter-annual changes in emissions, [OH] or photolysis rates were not quantified here (only transport). On a monthly basis, the amount of CHBr 3 reaching the tropical LS can clearly exhibit larger variability. CHBr 3 anomalies 665 (calculated as monthly departures from the climatological monthly mean mixing ratio) at the tropical tropopause are shown in Figure 17. Also shown in Figure 17 is the Multivariate ENSO Index (MEI) -a time-series which characterises ENSO intensity based on a range of meteorological and oceanographic components (Wolter and Timlin, 1998). See also: http://www.esrl.noaa.gov/psd/enso/mei/.
The transport of CHBr 3 (and CH 2 Br 2 , not shown) to the tropical LS is strongly correlated (r val-670 ues ranging from 0.6 to 0.75 across the ensemble) to ENSO activity over the Eastern Pacific (owing to the influence of sea surface temperature on convection). For example, a clear signal of the very strong El Niño event of 1997/1998 is apparent in the models (i.e. with enhanced CHBr 3 at the tropopause) supporting the notion that bromine SGI is sensitive to such climate modes, in this region (Aschmann et al., 2011). However, when averaged over the tropics no strong correlation be-675 tween VSLS loading in the LS and the MEI (or just sea surface temperature) was found across the ensemble. We suggest that zonal variations in SST anomalies (and convective activity) associated with ENSO, with warming in some regions and cooling in others, has a cancelling effect on the tropical mean bromine SGI. Indeed, previous model studies have showed a marked zonal structure in CHBr 3 /CH 2 Br 2 loading in the LS in strong ENSO years, with relative increases and decreases 680 with respect to climatological averages depending on region (Aschmann et al., 2011). Further investigation, beyond the scope of this work, is needed to determine the sensitivity of total stratospheric Br VSLS y (i.e. including the contribution from product gas injection), to this and other modes of climate variability.

685
Understanding the chemical and dynamical processes which influence the atmospheric loading of VSLS in the present, and how these processes may change in the future, is important to understand the role of VSLS in a number of issues. In the context of the stratosphere, it is important to (i) determine the relevance of VSLS for assessments of O 3 layer recovery timescales (Yang et al., 2014), (ii) assess the full impact of proposed stratospheric geoengineering strategies (Tilmes et al.,690 2012) and (iii) accurately quantify the ozone-driven radiative forcing (RF) of climate (Hossaini et al., 2015a). Here we performed the first concerted multi-model intercomparison of halogenated VSLS.
The overarching objective of TransCom-VSLS was to provide a reconciled model estimate of the SGI of bromine from CHBr 3 and CH 2 Br 2 to the lower stratosphere and to investigate inter-model differences due to emissions and transport processes. Participating models performed simulations 695 over a 20-year period, using a standardised chemistry setup (prescribed oxidants/photolysis rates) to isolate, predominantly, transport-driven variability between models. We examined the sensitiv- is somewhat poorly constrained. We find that at a number of locations, no consensus among models as to which emission inventory performs best can be reached. This is due to differences in the representation/implementation of transport processes between models which can significantly influ-710 ence the boundary layer abundance of short-lived tracers. This effect was observed between CTM variants which, other than tropospheric transport schemes, are identical. A major implication of this finding is that care must be taken when assessing the performance of emission inventories in order to constrain global VSLS emissions, based on single model studies alone. However, we also find that within the tropics -where the troposphere-to-stratosphere transport of VSLS takes place -most models (∼70%) achieve best agreement with measured surface CHBr 3 when using a bottom-up derived inventory, with the lowest CHBr 3 emission flux (Ziska et al., 2013). Similarly for CH 2 Br 2 most (also ∼70%) of the models achieve optimal agreement using the CH 2 Br 2 inventory with the lowest tropical emissions (Liang et al., 2010), though agreement is generally less sensitive to the choice of emission inventory (compared to CHBr 3 ). Recent studies have questioned the effectiveness of us-720 ing aircraft observations and global-scale models (i.e. the top-down approach) in order to constrain regional VSLS emissions (Russo et al., 2015). For this reason and given growing interest as to possible climate-driven changes in VSLS emissions (e.g. Hughes et al., 2012), online calculations (e.g. Lennartz et al., 2015) which (i) consider interactions between the ocean/atmosphere state (based on observed seawater concentrations) and (ii) produce seasonally-resolved sea-to-air fluxes may prove a 725 more insightful approach, over use of prescribed emission climatologies, in future modelling work. studies (e.g. Aschmann et al., 2009;Pisso et al., 2010;Hossaini et al., 2012b;Liang et al., 2014).
Owing to significant inter-model differences in transport processes, both the absolute tracer amount transported to the stratosphere and the amplitude of the seasonal cycle varies among models. However, of the above regions, the tropical West Pacific is the most important in all of the models (regardless of the emission inventory), due to rapid vertical ascent of VSLS simulated during boreal winter. In the free troposphere, the models reproduce observed CHBr 3 and CH 2 Br 2 from the recent SHIVA and CAST campaigns in this region to within ≤16% and ≤32%, respectively. However, at higher altitudes in the TTL the models generally (i) underestimated CHBr 3 between 14-16 km observed during the 2014 NASA ATTREX mission in this region but (ii) fell within ±1 σ of the observed mean around the tropical tropopause (∼17 km). Generally good agreement was also 740 obtained to high altitude aircraft measurements of VSLS around the tropopause in the Eastern Pacific. During boreal summer, most models show elevated CHBr 3 around the tropopause above the Asian Monsoon region. However, the strength of this signal varies considerably among the models with a spread that encompasses virtually no CHBr 3 enhancement over the Monsoon region to strong (85%) CHBr 3 enhancements at the tropopause, with respect to the zonal average. Measure-745 ments of VSLS in the poorly sampled Monsoon region from the upcoming StratoClim campaign (http://www.stratoclim.org/) will prove useful in determining the importance of this region for the troposphere-to-stratosphere transport of VSLS.
-Climatologically, we estimate that CHBr 3 and CH 2 Br 2 contribute 2.0 (1.2-2.5) ppt Br to the lower stratosphere through SGI, with the reported uncertainty due to the model spread. The TransCom-750 VSLS best estimate of 2.0 ppt Br is (i) ∼57% larger than the measurement-derived best estimate of 1.28 ppt Br reported by Carpenter and Reimann (2014), and (ii) the TransCom-VSLS range (1.2-2.5 ppt Br) is ∼37% smaller than the 0.6-2.65 ppt Br range reported by Carpenter and Reimann (2014).
From this we suggest that, climatologically, the Carpenter and Reimann (2014) measurement-derived SGI range, based on a limited number of aircraft observations (with a particular paucity in the tropical West Pacific), is potentially too conservative at the lower limit. Although we acknowledge that our uncertainty estimate (the model spread) does not account for a number of intrinsic uncertainties within global models, for example, tropospheric [OH] (as the models used the same set of prescribed oxidants). No significant transport-driven trend in stratospheric bromine SGI was found over the simulation period, though inter-annual variability was of the order of ±5%. Loading of both CHBr 3 and 760 CH 2 Br 2 around the tropopause over the East Pacific is strongly coupled to ENSO activity but no strong correlation to ENSO or sea surface temperature was found when averaged across the wider tropical domain.
The location of the surface sites is summarised in Table 3. Model output based on CHBr3_L tracer (i.e. using aseasonal emissions inventory of Liang et al. (2010)