High frequency, in situ measurements of HFC-152a were made by gas chromatography-mass spectrometry (GC-Agilent 6890) coupled with quadrupole mass selective detection (MSD-Agilent 5973/5975). Measurements commenced at Mace Head, Ireland in 1994 and Cape Grim, Tasmania in 1998, using a custom-built automated pre-concentration system (adsorption desorption system – ADS) to selectively and quantitatively retain halogenated compounds from 2 L air samples. Based on a Peltier–cooled pre-concentration microtrap cooled to -50 ∘C during the adsorption phase, the ADS provided on-site calibrated air samples every 4 h, i.e., six per day (Simmonds et al., 1995). In 2004, the ADS-GC-MS was replaced with a more advanced custom-built pre-concentration system (Medusa) with enhanced cooling to ∼-180 ∘C and the relatively mild adsorbent HayeSep D (Miller et al., 2008; Arnold et al., 2012). Agilent 5973 MSDs (mass selective detector) were also upgraded to the Agilent 5975 MSDs over the course of the Medusa observations. Analysis of each 2 L sample of ambient air was alternated with analysis of a 2 L reference gas (designated as a working standard) to correct for short-term instrumental drift, resulting in 12 (Medusa) individually calibrated air measurements per day. Working standards were prepared for each station by compressing ambient air into 34 L electropolished stainless steel canisters (Essex Industries, Inc., Missouri) using modified oil-free compressors (SA-6, RIX, California). Exceptions to this were the Cape Grim and Zeppelin stations, where the working standards were filled using a cryogenic filling technique. Research-grade helium, which was used as a carrier gas in the Medusa systems, was further purified by passage through a heated “getter” type purifier (Valco Instruments, Houston, TX). The carrier gas was analyzed for blanks on a regular basis and blank levels of HFC-152a were below the limit of detection at all field stations.

Table 1 lists the geographical location and the time when routine ambient measurements of HFC-152a began at each monitoring station. Stations with the longest observational records that deployed both ADS and Medusa GC-MS instruments include Mace Head (MHD), Jungfraujoch (JFJ), Ny-Ålesund (ZEP) and Cape Grim (CGO). Medusa GC-MS instruments were installed at five other AGAGE stations Trinidad Head (THD), Gosan (GSN), Ragged Point, (RPB), Shangdianzi (SDZ), and Cape Matatula (SMO) between 2003 and 2010. In addition two AGAGE affiliated stations Monte Cimone (CMN) and Hateruma (HAT), which use comparable GC-MS instruments, but a different pre-concentration design for sample enrichment, commenced HFC-152a measurements in 2001 and 2004, respectively. Importantly, all 11 stations listed in Table 1 report HFC-152a measurements relative to the Scripps Institution of Oceanography (SIO-05) calibration scale (as dry gas mole fractions in pmol mol-1).

The estimated accuracy of the calibration scale for HFC-152a is 4 %: a more detailed discussion of the measurement technique and calibration procedure is reported elsewhere (Miller et al., 2008; O'Doherty et al., 2009; Mühle et al., 2010). HFC-152a was determined using the MS in selected ion monitoring mode (SIM) with a target ion CH3CF2+ (m/z 65) and qualifier ion CH3CF+ (m/z 46). To ensure that potential interferences from co-eluting species did not compromise the analysis, the ratio of the target to qualifier ion was continuously monitored. Measurement precision was calculated as the daily standard deviation (1σ) of the ratios of each standard response to the average of the closest-in-time preceding and subsequent standard responses. Typical daily precisions vary from station to station with a range of 0.1–0.4 ppt. Individual station precisions were used to estimate the precision of each in situ measurement.

In order to extend the HFC-152a data record back before the commencement of high-frequency measurements, analyses of Northern Hemisphere (NH) and Southern Hemisphere (SH) archived air samples dating back to 1978, were carried out using three similar Medusa GC-MS instruments at the Scripps Institution of Oceanography (SIO), La Jolla, California, the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Aspendale, Australia and the Cape Grim Baseline Air Pollution Station, Tasmania. The SH samples are part of the Cape Grim air archive (CGAA) described in Langenfelds et al. (1996), and Krummel et al. (2007). The NH samples analyzed for this paper were filled during background conditions mostly at Trinidad Head, but also at La Jolla, California; Cape Meares, Oregon; Ny Ålesund, Svalbad and Point Barrow, Alaska (some samples are courtesy of the National Oceanic and Atmospheric Administration (NOAA).

In addition, eight SH samples were measured at SIO and compared with SH samples of similar age measured at CSIRO (February 1995, July 1995, November 1995, June 1998, July 2004, February 2006, August 2008, and December 2010, Δx=0.01–0.07 ppt Δt=1–33 days) and three NH samples were measured at CSIRO and compared with NH samples of the same age measured at SIO (May 1989 and April 1999, Δx=0.02–0.06 ppt, Δt=1–11 days). The good agreement between SIO and CSIRO archived air stored in different types of tanks (stainless steel tanks, Essex Industries, Inc and Silcosteel treated tanks, Restek Corporation) serves both as proof of the good consistency of the individual Medusa GC-MS instruments and the integrity of the tanks used. Samples were analyzed in replicate typically 3–6 times each and several NH tanks were re-measured over a number of years.

Baseline in situ monthly mean HFC-152a mole fractions were calculated by excluding values enhanced by local and regional pollution influences, as identified by the iterative AGAGE pollution identification algorithm, (see Appendix in O'Doherty et al., 2001). Briefly, baseline measurements are assumed to have a Gaussian distribution around the local baseline value, and an iterative process is used to filter out the points that do not conform to this distribution. A second-order polynomial is fitted to the subset of daily minima in any 121-day period to provide a first estimate of the baseline and seasonal cycle. After subtracting this polynomial from all the observations a standard deviation and median are calculated for the residual values over the 121-day period. Values exceeding 3 standard deviations above the baseline are thus identified as non-baseline (polluted) and removed from further consideration. The process is repeated iteratively to identify and remove additional non-baseline values until the new and previous calculated median values agree within 0.1 %. For the core AGAGE stations, in situ baseline data and archive air data, extending the record to periods prior to the in situ measurement period, are then combined for each hemisphere, and outliers are rejected by an iterative filter.

To estimate global-average mole fractions and derive growth rates, a two-dimensional model of atmospheric chemistry and transport was employed. The AGAGE 12-box model simulates trace gas transport in four equal mass latitudinal sections (divisions at 30–90∘ N, 0–30∘ N, 30–0∘ S and 90–30∘ S) and at three heights (vertical divisions at 200, 500 and 1000 hPa). The model was originally developed by Cunnold et al. (1983) (nine-box version), with subsequent improvements by Cunnold et al. (1994) and Rigby et al. (2013, 2014). Emissions were estimated between 1989 and 2014 using a Bayesian method in which an a priori constraint (EDGAR v4.2) on the emissions growth rate was adjusted using the baseline-filtered AGAGE observations (Rigby et al., 2011a, 2014). Global emissions were derived that included estimates of the uncertainties due to the observations, the prior and the lifetime of HFC-152a, as detailed in the supplementary material in Rigby et al. (2014). Note that historically and here the 12-box model only uses observations from the core AGAGE sites, Mace Head, Trinidad Head, Ragged Point, Cape Matatula, and Cape Grim.

We used the methodology outlined in Lunt et al. (2015) and Rigby et al. (2011b) to derive emissions of HFC-152a from continental regions. The high-resolution, regional UK Met Office Numerical Atmospheric-dispersion Modelling Environment (NAME), Manning et al. (2011) was used to simulate atmospheric HFC transport close to a subset of AGAGE monitoring sites, which were strongly influenced by regional HFC sources (domains shown by red boxes in Fig. 1). Simultaneously, the influence of changes to the global emissions field on all measurement stations was simulated using the global Model for OZone and Related Tracers, MOZART (Emmons et al., 2010). We estimated annual emissions for the period 2007–2012 and aggregated the derived emissions fields into continental regions, separating countries that either do (“Annex-1”), or do not (“non-Annex-1”) report detailed, annual emissions to the UNFCCC. Emissions were estimated using a hierarchical Bayesian inverse method (Ganesan et al., 2014; Lunt et al., 2015) and all high-frequency observations from 10 of the 11 sites listed in Table 1, excluding Shangdianzi due to the short time series. The hierarchical Bayesian method includes uncertainty parameters (e.g., model “mismatch” errors and a priori uncertainties) in the estimation scheme, reducing the influence of subjective choices on the outcome of the inversion.

A method for estimating emissions from observations and atmospheric transport modeling with NAME referred to as InTEM, “Inversion Technique for Emission Modelling” (Manning et al., 2011), uses a simulated annealing method (Press et al., 1992) to search for the emission distribution that produces a modeled times series that has the best statistical match to the observations from certain AGAGE stations (e.g., Mace Head, Cape Grim). NAME was driven with output from the operational analysis of the UK Met Office Numerical Weather Prediction model, the Unified Model, at global horizontal resolution of 17–40 km (year dependent). InTEM estimates the spatial distribution of emissions across a defined geographical area, and can either start from a random emission distribution or be constrained by an inventory-defined distribution. Emission totals from specific geographical areas are calculated by summing the derived emissions from each grid (non-uniform) in that region.

The uncertainty estimation used within InTEM is described in detail elsewhere (Manning et al., 2011). Briefly, the uncertainty space was explored by (a) solving the inversion multiple times with a range of baseline mole fractions within the baseline uncertainty estimated during the baseline fitting process and (b) by altering the 3-year inversion time window by 1 month throughout the data period thereby solving over a particular 1-year period many times using different observations. In total for each annual estimate, up to 111 inversions were performed; the median and 5th and 95th percentiles were used as the final total and spread. For the Australian estimates data between 2002 and 2011 were used, for the NW European estimates data between November 1994 and December 2013 were used.

A regional Bayesian inversion system using backward simulations of a Lagrangian particle dispersion model FLEXPART (Stohl et al., 2005) was applied to the HFC-152a observations from Mace Head, Jungfraujoch and Mt. Cimone for the period 2006 to 2014. The inversion technique follows the description by Stohl et al. (2009) and was previously applied to regional halocarbon emissions from Europe (Keller et al., 2012; Maione et al., 2014) and China (Vollmer et al., 2009). For these emission estimates, the background was determined by applying the Robust Extraction of Baseline Signal (REBS) filter described in detail by Ruckstuhl et al. (2012). The transport model FLEXPART was driven with output from the operational analysis of the Integrated Forecast System (IFS) of the European Centre for Medium Range Weather Forecast (ECMWF) using a spatial resolution of 0.2∘ × 0.2∘ for a nested domain covering the larger area of the European Alps and a spatial resolution of 1∘ × 1∘ elsewhere.

The FLEXPART model was applied to the HFC-152a observations from Mace Head, Jungfraujoch and Mt. Cimone for the period 2006 to 2014. Prior to 2006, the model resolution of Integrated Forecast System (IFS) was not sufficiently fine to realistically simulate the transport to the two high altitude sites Jungfraujoch and Mt. Cimone. Therefore, no attempt was made here to apply the inversion system to years before 2006. As prior information of the HFC-152a emissions we used country totals as submitted to UNFCCC. These were spatially disaggregated following the HFC-152a distribution given in EDGAR (v4.2). For countries not reporting HFC-152a emissions to UNFCCC we used the values given in EDGAR. The EDGAR inventory was only available up to the year 2008 beyond this year the EDGAR 2008 distribution was used. The uncertainty of the prior emissions was set so that the region total uncertainty equalled 20 % of the region total emissions. The regional inversion grid covered a region similar to that shown in Fig. 1.

We also present regional emissions estimates using inter-species correlation (ISC) methods (Yokouchi et al., 2005). Emissions of a number of trace gases from the Melbourne/Port Phillip region (CFCs, HCFCs, HFCs, carbon tetrachloride: Dunse et al., 2001, 2002, 2005; O'Doherty et al., 2009; Fraser et al., 2014a, b), including HFC-152a (Greally et al., 2007), have been estimated utilizing in situ high frequency measurements from Cape Grim and ISC with co-incident carbon monoxide (CO) measurements.

ISC works best for co-located sources – however extensive modeling has shown that by the time the Melbourne/Port Phillip plume reaches Cape Grim (300 km from the source) it is well mixed and the likely inhomogeneity of the source regions (for CO and HFC-152a in this case) does not have a significant influence on the derived emissions. It should be noted that in order to obtain a significant sampling of Port Phillip pollution episodes at Cape Grim, data from 3 years (for example 2011–2013) are used to derive annual emissions (for 2012). (InTEM also uses data from 3 years to derive annual emissions.) The ISC uncertainties given in the paper include (1) the uncertainties in the estimates of CO emissions from Melbourne/Port Phillip (2) the uncertainties in the overall correlation between CO and HCFC-152a as seen in pollution episodes at Cape Grim (3) the uncertainties in the geographic extent of the HFC-152a and CO source regions impacting on Cape Grim and their entrained population.

Using HCFC-22 as the reference tracer, Li et al. (2011) reported that China is the dominant emitter of halocarbons in East Asia. North American HFC-152a emissions have been estimated from atmospheric data using interspecies correlation based techniques with CO (Millet et al., 2009; Barletta et al., 2011) and fossil fuel CO2 (Miller et al., 2012) as the reference emissions.

The time series of HFC-152a in situ observations recorded at selected AGAGE and affiliated monitoring stations are shown in Fig. 2a–c. Data have been filtered into baseline (black) and above baseline (red) using the AGAGE pollution algorithm, as discussed in Sect. 2.3. Figure 2a shows the mole fractions in ppt for the four stations that deployed both ADS and Medusa GC-MS instruments (Mace Head, Zeppelin, Jungfraujoch, and Cape Grim). Most notable are the substantial above baseline events at Mace Head and Jungfraujoch that are influenced primarily by emissions from European sources. Conversely, the Zeppelin Arctic station and the SH station at Cape Grim have relatively small above baseline events implying smaller emissions from local or regional sources.

Figure 2b shows measurements at the five other AGAGE stations (Trinidad Head, Gosan, Ragged Point, Shangdianzi, and Cape Matatula), which used only Medusa GC-MS instruments. The North American site at Trinidad Head and the Asian sites at Shangdianzi and Gosan are the most strongly influenced by regional emissions. The tropical sites at Ragged Point, Barbados, and Cape Matatula, American Samoa show very few enhancements above the baseline and these are due mostly to local emissions occurring under nighttime inversion conditions and occasional influences from regional emission sources (note the different y axis scales). Although the Shangdianzi station was operational for only a short period, the enhancements above baseline are significant due to the sensitivity of this site to Chinese emissions, and comparable in magnitude to those at Gosan.

Figure 2c illustrates the time series from the two affiliated AGAGE stations (Monte Cimone and Haturuma) that used comparable GC-MS instruments but with different methods of pre-concentration. Monte Cimone, like the Jungfraujoch, is also influenced by substantial emissions from sources in continental Europe. Hateruma is influenced by sources in China, Korea, Taiwan, and Japan (Yokouchi et al., 2006).

Figure 3 shows the in situ measurements of HFC-152a, as baseline monthly means (excluding pollution events), obtained from the two AGAGE stations Mace Head and Cape Grim with the longest time series that deployed both ADS and Medusa GC-MS instruments. Superimposed in Fig. 3 are the NH and SH archived flask data extending back to 1978. Annual average mole fractions at Mace Head increased from 1.2 ppt in 1994 to 10.2 ppt by 2014, Cape Grim annual average mole fractions increased from 0.84 ppt in 1998 when in situ measurements first began to 4.5 ppt in 2014. However, in the last few years the rates of growth at both sites have slowed to almost zero.

The NH archived samples are more variable than the SH archived samples. The SH archive is collected only under strict baseline conditions (Southern Ocean air) and is far removed from the major sources of HFC-152a. Conversely in the NH, where most major sources of emissions are located, sampling under strict baseline conditions is more difficult to achieve.

Figure 4a illustrates HFC-152a baseline monthly means obtained from the five other AGAGE observing sites (Ragged Point, Gosan, Cape Matatula, Trinidad Head, and Shangdianzi using only the more advanced Medusa GC-MS. There is a large seasonal cycle at Gosan with a very deep minimum due to summertime transport from the Southern Hemisphere (Li et al., 2011). Barbados can also be influenced by Southern Hemispheric air during the hurricane season (Archibald et al., 2015).

Figure 4b shows the baseline monthly mean mole fractions for the three mountain stations. Ny-Ålesund and Jungfraujoch, using combined ADS and Medusa GC-MS measurements and Monte Cimone, which used a commercial pre-concentrator GC-MS. In most years Monte Cimone exhibits enhanced mole fractions during the NH spring months (March–May).

The HFC-152a seasonal cycles at Mace Head and Cape Grim shown in Fig. 5a and b, are broadly representative of the Northern Hemisphere and Southern Hemisphere, respectively. The seasonal cycle at Mace Head shows a NH spring maximum (April–May) and late summer minimum (August–October), while the SH seasonal cycle at Cape Grim exhibits a broad austral spring maximum (July–November) and a late summer minimum (January–April). The summer minimum at both locations is attributed to enhanced summertime loss (OH) with possibly a contribution from seasonally varying emissions in the NH that may be out-of-phase with the NH sink. At Cape Grim an additional source of seasonality is due to seasonally varying transport between the NH and SH, which is generally in phase with the sink induced seasonal cycle. This competition between OH summertime loss and seasonally varying transport has been observed at many other AGAGE locations (Prinn et al., 1992; Greally et al., 2007; O'Doherty et al., 2009, 2014; Li et al., 2011).

Figure 6 shows the mole fractions output from the AGAGE global 12-box model, along with the monthly mean semi-hemispheric average observations used in the inversion. The figure also shows the running mean growth rate, smoothed using a Kolmogorov–Zurbenko filter with a window of approximately 12 months (Rigby et al., 2014). Most notable is the positive growth rate from 1995 reaching a maximum of ∼ 0.84 ppt yr-1 in 2006, followed by a steady decline in the growth rate with a minimum during the economic recession in 2008–2009. The positive growth rate then resumes increasing to ∼ 0.4 ppt yr-1 in 2010 followed by a subsequent decrease with an annual average growth rate in 2013–2014 of minus ∼ 0.06 ppt yr-1.

The strong inter-hemispheric gradient demonstrates that emissions are predominantly in the NH, as has been illustrated for many other purely anthropogenic trace gases (Prinn et al., 2000). The globally averaged mole fraction in the lower troposphere in 2014 is estimated to be 6.84 ± 0.23 ppt and the annual rate of increase is -0.06 ± 0.05 ppt yr-1. As reported by Rigby et al. (2014) the major long lived synthetic greenhouse gases (SGHGs) which include CFCs, HCFCs, HFCs, and perfluorocarbons (SF6 and NF3), as well as CH3CCl3 and CCl4 were responsible for 350 ± 10 mW m-2 of direct radiative forcing in 2012. The radiative forcing of HFC-152a, determined from the AGAGE 12-box model in this study, was 0.61 ± 0.02 mW m-2 in 2014, which represents only a tiny fraction (∼ 0.2 %) of the global radiative forcing of the SGHG.

Estimated global emissions of HFC-152a using the 12-box model and the reported UNFCCC and EDGAR emission inventories are shown in Fig. 7 and Table 2. The blue solid line represents our model-derived emissions, with the 1σ error band shown by the shaded areas. Model derived emissions grew steadily from 1995 to 2007 with a non-statistically significant decrease in emissions in 2009 to 48 ± 16.4 Gg yr-1, during the economic downturn in 2008–2009. The mean emission reached a maximum of 54.4 ± 17.1 Gg yr-1 in 2011, followed by a period of relatively stable emissions, the mean showing a slight decline to 52.5 ± 20.1 Gg yr-1 (7.2 ± 2.8 Tg-CO2 eq yr-1) in 2014.

The data shown in column 3 of Table 2 are the totals of submissions by the national governments to the UNFCCC (Rio Convention) as reported in Table 2(II) s1 in the Common Reporting Format (CRF), available on the UNFCCC website (http://unfccc.int/national reports/annex ighg inventories/national inventories submissions/items/8108.php). The values were taken from the 2014 database and cover years 1995 (the baseline year for submissions) to 2012. In addition to reporting calculated emissions of HFCs 23, 32, 125, 134a, 143a, 152a, 227ea, 236fa, 245ca, and 43-10mee individually, many countries also included “unspecified” emissions in this database (as the sum of their CO2 equivalents). Where the unspecified component was small in relation to the national specified emissions, it was disaggregated by assuming that it had the same fractional contribution of each HFC as reported in the specified components (adjusted for their CO2 equivalence). However, in the US, although values of emissions of several HFCs are calculated specifically for the individual substances, HFCs 152a, 227ea, 245ca and 43-10mee are shown in the database as “commercially confidential” and their emissions apparently constitute the substantial aggregated “unspecified” emissions reported. Hence, for the US, these unspecified annual emissions were divided only between HFCs 152a, 227ea, 245ca and 43-10mee, assuming the same ratio as their reported global emissions, all expressed as CO2 equivalents. The values shown in column 4 of Table 2 are the global totals of HFC-152a after adjusting in these ways for the quantities included in “unspecified” emissions.

The additional component of US emissions makes a substantial contribution to the very large difference between the UNFCCC data as reported and the adjusted values. This is partly due to the low global warming potential of HFC-152a (a factor of 10 lower than other HFCs) which magnifies its mass component in the 8200 Gg CO2 equivalent of US “unspecified” emissions.

The AGAGE observation based global emissions are substantially higher than the emissions calculated from the UNFCCC GHG reports (2014 submission). It is not unreasonable that UNFCCC-reported emissions are lower than the AGAGE global emission estimates, since countries and regions in Asia (e.g., China, Indonesia, Korea, Malaysia, the Philippines, Taiwan, Vietnam), the Indian sub-continent (e.g., India, Pakistan), the Middle East, South Africa, and Latin America do not report to the UNFCCC. Where we include the HFC-152a component of unspecified emissions (green line in Fig. 7) results are consistent within the error bars until approximately 2003 to 2005 when they start to diverge (UNFCCC + “unspecified” lower). From 1996 to 2002, estimated emissions from EDGAR (v4.2) are generally consistent with AGAGE emissions, but then begin to diverge with EDGAR emissions 22 Gg below 2008 AGAGE emissions, the last year for which EDGAR reports emissions.

Lunt et al. (2015) have reported global and regional emissions estimates for the most abundant HFCs, based on inversions of atmospheric mole fraction data, aggregated into two categories; those from Annex 1 countries and those from non-Annex 1 countries. The inversion methodology used the NAME model to simulate atmospheric transport close to the monitoring sites, and the Model for Ozone and Related chemical Tracers (MOZART, Emmons et al., 2010) to simultaneously calculate the effect of changes to the global emissions field on each measurement site. The model sensitivities were combined with a prior estimate of emissions (based on EDGAR) and the atmospheric measurements, in a hierarchical Bayesian inversion (Ganesan et al., 2014), to infer emissions.

Using this method we infer emissions estimates for the entire world, Europe, North America, and East Asia. Table 3 lists our estimated regional emissions in Gg yr-1 averaged across two time periods: 2007–2009 and 2010–2012, together with our global emission estimates averaged over the same time periods from the 12-box model. It is apparent that North American average annual emissions (∼ 30 Gg) are the major contributor to the global total with Europe contributing annual average emissions from about 5–6 Gg yr-1. East Asia and Europe contribute emissions of ∼ 7 and ∼ 6 Gg yr-1, respectively to the global total. The 2007–2009 North American emission estimate of 28 Gg yr-1 agrees within the uncertainties of HFC-152a emission estimates reported in Barletta et al. (2011) and Simmonds et al. (2015). The North American estimate indicates one reason why the UNFCCC reported amount appears to be so low; more than half the global emissions appear to come from this continental region, yet the UNFCCC reports do not include specific HFC-152a emissions from the US.

The entire ALE/GAGE/AGAGE data base comprising every calibrated measurement including pollution events is archived on the Carbon Dioxide Information and Analysis Center (CDIAC) at the US Department of Energy, Oak Ridge National Laboratory.