Emissions of CFCs, HCFCs and HFCs from India

As the second most populous country and third fastest growing economy, India has emerged as a global economic power. As such, its emissions of greenhouse and ozone-depleting gases are of global significance. However, unlike neighbouring China, the Indian sub-continent is very poorly monitored by existing measurement networks. Of the greenhouse/ozonedepleting gases, India’s emissions of synthetic halocarbons (here defined as chlorofluorocarbons (CFCs), hydrochlorofluorocarbons (HCFCs) and hydrofluorocarbons (HFCs)) are not well-known. Previous measurements from the region have been 5 obtained at observatories many hundreds of miles from source regions, or at high altitudes, limiting their value for the estimation of regional emission rates. Given the projected rapid growth in demand for refrigerants in India, emission estimates of these halocarbons are urgently needed, to provide a benchmark against which future changes can be evaluated. In this study, we report the first atmospheric-measurement derived emissions of the ozone-depleting CFCs and HCFCs, and potent greenhouse gas HFCs from India. Air samples were collected at low-altitude during a 2-month aircraft campaign between June and July 1

by January 1st, 2016, followed by complete phase-out by 2040. At the 19th Meeting of the Parties in 2007, India agreed to an acceleration of this schedule. Under Stage II of the HPMP, India agreed to freeze its consumption of HCFCs at the base level (2009/10 average) by 2013, followed by a 10% reduction (relative to the base level) by 2015 and a complete phase-out by 2030.
In 2016, India adopted the Kigali Amendment, under which it will also begin to phase-down its production and consumption of HFCs. However, its developing status means it will not be required to make its first reductions until 2028, and in the meantime, 5 India's demand for HFCs is expected to rise dramatically (Purohit et al., 2016).
With a population exceeding one billion and a rapidly expanding economy, India's ODS and HFC emissions are expected to have global significance. Based on inferred consumption trends, Velders et al. (2015) estimated that India will emit 400 Tg CO 2 eq yr −1 of HFCs in 2050, a 67-fold increase over 2016 emissions. However, little else is known about India's emissions.
Estimates from bottom-up, inventory-based methods have only been made for a subset of HFCs (HFC-134a, HFC-152a and 10 HFC-23) in India and only up to 2010 (Garg et al., 2006;MoEFCC, 2012MoEFCC, , 2015. Emissions of these gases have never been estimated for India through regional 'top-down' or inverse modelling approaches that use atmospheric mole fraction measurements to infer surface fluxes. However, top-down methods have been applied elsewhere in Asia (Palmer et al., 2003;Yokouchi et al., 2005;Kim et al., 2010;Saikawa et al., 2012;Stohl et al., 2010).
Previous studies in other countries have shown that there can be large discrepancies between national inventories of ODS and 15 HFCs and those inferred from atmospheric observations (Graziosi et al., 2017;Lunt et al., 2015;Say et al., 2016). Therefore, this dual quantification approach has been highlighted by many organizations as being beneficial for accurate and transparent greenhouse gas reporting (Leip et al., 2018). In this study, we present the first top-down emissions estimates of CFCs, HCFCs and HFCs and provide a 2016 benchmark, which is critical for being able to evaluate future policy changes surrounding India's ODS and HFC emissions.  (Table 2). On nine of these flights, samples were collected over northern India at 25 altitudes ranging predominantly between 0 -1.5 km (Fig. 1). Air was drawn through a forward-facing air sampling pipe on the exterior of the aircraft and pressurised into the sample flasks using a metal bellows pump (Senior Aerospace PWSC 28823-7).
Sample flasks were evacuated to 1x10e −5 psig prior to each flight. Before sample collection, the lines within each sample case were flushed with ambient air for a minimum of one minute. Sample flasks were filled to a maximum pressure of 41 psig, giving a usable sample volume of 9 L at atmospheric pressure. Sample filling typically varied between 25 -60 seconds in 30 duration, depending on altitude (equivalent to~7 km of flight track at average cruise velocity). Canisters were filled at regular intervals during each flight (interval dependent on flight length). When not in use, flask samples were stored in a container with no air-conditioning, to eliminate the risk of sample contamination from leaking air-conditioning refrigerant. None of the gases discussed here were present on the research aircraft itself, and the laboratory at the University of Bristol does not contain an HFC filled air-conditioning unit. Apart from flasks collected over the Arabian Sea, samples were transported from India to Bristol within 1 month of collection.
Flask samples were analysed using the Medusa GCMS analytical system, with modifications to the analysis routine developed to account for the small volume and low pressure of the flask samples. In the set-up described previously (Miller et al., 5 2008;Arnold et al., 2012), atmospheric measurements were derived from 2 L samples, injected into the pre-concentration system at a flow rate of 100 cm 3 min −1 , resulting in a total injection time of 20 minutes. For this work, each measurement was derived from three 1.75 L analyses, injected into the analytical system at a flow rate of 50 cm 3 min −1 , resulting in a total injection time of 35 minutes. The analysis of each flask was bracketed by analyses of a quaternary reference gas, to account for short term drifts in detector sensitivity. ODS/HFC mole fractions are reported relative to a set of gravimetrically prepared 'primary' 10 standards, maintained at the Scripps Institute of Oceanography (SIO), via a hierarchy of compressed real-air standards held in 34 L electro-polished stainless-steel canisters (Essex Industries, Missouri, USA). The working (quaternary) standard was compared to a tertiary tank on a roughly monthly basis. System blanks were conducted monthly, to quantify possible interferences from system leaks and carrier gas impurities. For each gas, the ratio of target to qualifier ion(s) was continually monitored to ensure that co-eluting species did not interfere with the analyses. For each flask, measurement precision was estimated as the 15 standard deviation of the three replicate analyses. Average measurement precisions are shown in Table 3 and are comparable to the precisions reported previously by Miller et al. (2008).

Numerical Atmospheric Modelling Environment (NAME)
A Lagrangian particle dispersion model was used to quantify the influence of surface fluxes on each atmospheric measurement. The Met Office NAME (Numerical Atmospheric dispersion Modelling Environment) model was run in backwards mode 20 (Manning et al., 2011) to generate 30-day air histories for every minute along each flight path (each minute represents approximately 7 km of the flight track at average speed). These air histories represent the sensitivity of a measurement to fluxes from the surface (defined as 0 -40 meters above ground level). NAME was driven using meteorological output from the operational analysis of the UK Met Office Numerical Weather Prediction model, the Unified Model, with a horizontal resolution of approximately 17 km in 2016. The model domain spanned from 55 -109 • E and 6 -48 • N up to 19 kilometres altitude.

25
For each flight minute, tracer particles were released at a rate of 1000 particles min −1 from a cuboid, whose dimensions were determined by the change in latitude, longitude and altitude of the aircraft during that one-minute period. In general, samples were collected during level sections of each flight path, minimising transport errors that could arise from the result of releasing particles over a range of altitudes. At the boundaries of the domain, the three-dimensional location and time at which each particle left the domain was recorded to provide the sensitivity to boundary conditions.

30
The ability of NAME to accurately simulate transport is critical for ensuring robust emission estimates. Model simulated wind direction and speed were compared to meteorological data recorded on board the FAAM aircraft ( Fig. S1-S2). To ensure that possible transport errors arising as a result of the inaccurate simulation of meteorological parameters did not affect our posterior estimates, emissions derived using the complete set of atmospheric measurements were compared to those derived from a filtered dataset (Fig. S3), whereby observations corresponding to periods where the NAME simulated wind speed/direction differed from the measured meteorology by more than 20% were removed.

Inverse modelling using atmospheric dispersion modelling
Our inverse method is based on the trans-dimensional approach described by Lunt et al. (2016). Emissions and uncertainties were characterized using principals of hierarchical Bayesian modelling detailed in Ganesan et al. (2014). The inverse approach 5 solves for a parameter vector, x (including flux fields and boundary conditions), using measurement data, y. In a Bayesian framework, independent prior knowledge of emissions, x ap , is used in conjunction with measurements to solve for a posterior emissions distribution, x using a linear model, H (Eq. 1).
H is a Jacobian matrix of sensitivities, here describing the relationship between changes in atmospheric mole fractions and 10 changes in the parameter vector x. is uncertainty arising from the model and the measurements. In a traditional Bayesian inversion, uncertainty in x ap and the model-measurement uncertainty, , are both assigned prior to the inversion. These uncertainties are often poorly known and rely on a subjective decision by the investigator, but have been shown to significantly impact upon the derived posterior emissions (Peylin et al., 2002;Rayner et al., 1999). To minimize this impact, a hierarchical approach incorporates additional hyper-parameters, which allow for the propagation of 'uncertainties in these uncertainties' to 15 the posterior solution.
ρ(x, θ|y) ∝ ρ(y|x, θ) · ρ(x|θ) · ρ(θ) Eq. 2 is a hierarchical version of Bayes' theorem (normalizing factor not shown for brevity). In this example, the prior emissions uncertainty is governed by a hyper-parameter (θ), which has a probability density function (PDF) that is explored within the inversion. This equation can also be employed in a similar way for the model-measurement uncertainty or any 20 other unknown parameters. The hierarchical Bayesian approach was extended to a trans-dimensional framework, in which the number and configuration of the spatial grid over which emissions were estimated were also unknown parameters, prior to the inversion. Therefore, it is largely the information content of the measurements that govern these unknown aspects. This framework has been shown to result in a more robust and justifiable quantification of uncertainties in emissions than traditional approaches.

25
In general, Eq. 2 is not solvable via analytical means and was estimated using Markov Chain Monte Carlo (MCMC). In this application, we used MCMC with a Metropolis-Hastings algorithm. We sampled 320,000 variants of the parameter space with the first 120,000 discarded as 'burn-in' to ensure that the system had no knowledge of the initial state. The remaining 200,000 samples were then used to form the posterior PDFs. In our estimates, the means of these posterior PDFs are presented, with the uncertainties represented by the 5 th and 95 th percentile values. Emissions were aggregated into totals for the northern-central India (NCI) region. The majority of gases, whose sources are expected to be distributed by population, were scaled to a national total using the NCI emissions and population statistics. For HFC-23, the NCI region incorporated 4 of the 5 known manufacturing plants for . To estimate national emissions, we scaled the NCI total by the ratio of HCFC-22 produced at those four factories, to total production at all five (based on 2015 factory specific production statistics (UNEP, 2017)). Based on these statistics, over 98% of HCFC-22 was produced by 5 factories residing within the NCI. HFC-32 was not scaled to an Indian total, since results show that the distribution for this gas within India is not likely to be population-based.
While the estimates presented here represent emissions over a two-month period, they are likely to be consistent with annual emissions of gases that are not expected to have significant seasonality in India. Seasonality in global emissions of HFC-134a and HCFC-22 has been reported previously (Xiang et al., 2014), but India's tropical/subtropical climate may minimize this 10 variation. In addition, due to sampling by aircraft, our emissions estimates are likely to be representative of a regional-scale for gases that have sources that are widespread and do not vary significantly in time. These characteristics are thought to be true for most gases studied here. For gases with some episodic emissions, which we show to be the case for HFC-125, additional caution must be made in their interpretation. and was therefore assumed a priori to account for 17.7% of global CFC emissions). HCFCs: a priori total HCFC emissions over India were based on 2015 consumption data reported by India in its HPMP Stage II Road Map report (MoEFCC, 2017).

A priori emissions
Consumption is expected to be an underestimate of emissions due to the presence of banked sources such as refrigerators and foams. However, a large uncertainty was assigned to the prior.  , 2015), and extrapolated to 2016 using reported HCFC-22 production data (and assuming a constant co-production ratio) (MoEFCC, 2017).
No spatial information was available for any of the ODS/HFCs studied here. For all gases, prior emissions totals were dis- Night-lights reflect where anthropogenic activity is present but does not weight the emissions a priori according to intensity. We expected the major sources of HFCs, HCFCs and CFCs (refrigeration and air-conditioning and insulating foam) to be explicitly linked to domestic and/or commercial activities, or from industries requiring a significant work-force.
In all gases, the prior was assigned a large uncertainty (1x10 −5 times the prior), to reflect the very limited current understanding of Indian spatial ODS/HFC emissions and ensure that our flux estimates were informed overwhelmingly by the 5 atmospheric observations. To ensure that the derived results were independent of the prior used, we compared results derived using the night light-based prior with those derived with a uniform distribution (Fig. S3).

A priori boundary conditions
The footprints from NAME consider only emissions released within the model domain. Hence, a prior estimate of the mole fraction at the boundaries of model domain must be made and incorporated into the modelled mole fraction. To this end, 10 mole fraction 'curtains' of each gas were used to provide a priori information about boundary conditions (it should be noted that these boundary conditions were adjusted within the inversion). For the HFCs, mole fractions were simulated using the 3D global chemical transport model MOZART (Model for OZone and Related chemical Tracers (Emmons et al., 2010)). MOZART was driven by offline meteorological fields from MERRA (Modern Era Retrospective-Analysis for Research and Applications (Rienecker et al., 2011)). For the HCFCs and CFCs, MOZART fields were not available, and uniform curtains were assumed.

15
The mole fraction for each curtain in each month was estimated using the AGAGE 12-box model  and measurements from five baseline AGAGE observatories. For each gas, the model was used to estimate a monthly baseline mole fraction for four latitude bands. The simulated mole fraction from latitude bands 30 -90 • N, 0 -30 • N and 0 -30 • S were used to assign a priori mole fractions to the North, East and West and South curtains of the model domain respectively.
The boundary conditions associated with each NAME-simulated measurement were calculated by mapping the exit times and 20 locations of particles leaving the domain to the curtains. In the inversion, a decomposition of the a priori boundary conditions, represented as offsets to the curtains in the North, East, West and South directions, were solved for in the inversion in addition to emissions.

Global halocarbon emissions estimation
Indian ODS/HFC emissions were compared to global emission estimates calculated using the AGAGE 12-box model (Rigby 3 Results

Atmospheric Measurements
Measurements were made from whole air flask samples collected over India during June and July 2016. Fig. 1 shows the lo- Enhancements in mole fractions over the regional background form the basis for estimating regional emissions. For all species except HFC-134a (Supporting Information), the average mole fractions of samples collected over the Arabian Sea were lower than those collected directly over NCI due to the prevailing westerly winds that bring well-mixed oceanic air to the Indian subcontinent during these months. Back trajectory analysis confirmed that these samples had not interacted with any 25 other significant land mass in the 30 days prior to collection. Variability in the mole fraction of samples collected over NCI varied considerably by species. For CFC-11, CFC-12 and CFC-113, few pollution events were observed, and their signals were of similar size to the measurement precision. Similarly, only small enhancements were observed for HCFC-142b, suggesting its main use as a foam-blowing agent was not significant or was not widespread and thus could not be discerned in the aircraft samples.

30
In contrast, large enhancements in mole fraction were observed for HFC-134a and HCFC-22, suggesting that usage of these substances as a refrigerant is widespread. It is likely that these gases share a range of common sources, including use in India's largest refrigeration and air-conditioning sector, stationary air-conditioning (Purohit et al., 2016). We find a significant (R = Enhancements were also observed in HFC-32 and HFC-125, although the observed mole fractions for these species were  During two flights (B959 and B963) conducted on the 21 st and 25 th of June, a small number of samples were collected over the Arabian Sea. NAME back-trajectory analysis was used to show that these samples had not interacted with any significant landmass in the 30 days prior to collection. Except for HFC-134a, the measurements derived from these samples exhibited very little variation, and the mole fractions were amongst the lowest observed during the campaign, which was consistent with the oceanic trajectories. As such, these provided a useful constraint upon the baseline for the modelling studies. In contrast, four of 25 the six samples collected on these flights exhibited an elevated HFC-134a concentration, which did not correlate with any other species, including HCFC-22. Several possible explanations exist for these elevated measurements: 1) Flasks collected over the Arabian Sea were compromised due to long storage times (over 1 month) at temperatures exceeding 40 • C before transport back to the UK for analysis. Long-term tests on the stability of HFC-134a at these temperatures have not been conducted; 2) the enhancements were the result of ship-borne emissions from the Indian Ocean. These flights were at low-altitude (0.01 -0.8

ODS and HFC emissions estimates for NCI and India
Mean NCI and Indian emissions estimates and the relative contributions of each gas to 2016 global emissions are shown in Fig. 4 and tabulated in Table 4 (Gg yr −1 ) and Table 5 (Tg CO 2 eq yr −1 ). Uncertainties presented throughout correspond to the 5 th and 95 th percentiles of the posterior distribution.
There are no previous top-down national-scale estimates of any of these gases for India. In 2016, India's HFC emissions were approximately an order of magnitude larger than the 2016 emissions assumed by Velders et al. (2015), suggesting that future projections of India's HFC emissions could be inaccurate.  Our emission rate is comparatively large, suggesting that either there are discrepancies with inventory methodologies or that there has been substantial growth in emissions in the last decade. Regardless, these emissions are small compared to other countries, particularly China, whose emissions of HFC-152a were estimated at 16 Gg yr −1 in 2013 (Fang et al., 2016), and the USA, for which an emission rate of 51. 5 (35.5 -75.5) Gg yr −1 was estimated for 2012 (Simmonds et al., 2015).

25
HFC-32 emissions are estimated for NCI to be 0.44 (0.36 -0.54) Gg yr −1 . All the measured enhancements in HFC-32 are correlated with enhancements in dichloromethane, a feedstock in the manufacture of HFC-32 (Fig. 3). These measurements suggest that India's HFC-32 emissions originate predominantly from fugitive losses during the manufacturing process, rather than widespread use in a refrigerant blend. Our assertion is consistent with a previous study (Leedham Elvidge et al., 2015), which attributed growth in South Asian emissions of DCM to HFC-32 manufacture. Since our NCI HFC-32 estimate is at-30 tributed to production, we consider it to be decoupled from population density, and hence we have not scaled this value to a national total.

Sensitivity tests
We assessed the sensitivity of derived emissions to the a priori emissions field. For all gases, the two solutions derived using the spatially-varying and uniform priors were the same within uncertainties, showing our posterior estimates to be robust to 30 the prior used ( Fig S3). We also compared the effect of inaccurate transport modelling on derived emissions by using a second, filtered dataset, removing times where NAME wind direction and wind speed differed by more than 20% from the measured parameters. A comparison of original and 'filtered' posterior estimates is given in Fig. S3. For all 12 halocarbons, the filtered estimates were within the uncertainty of the original values, indicating that our estimates were robust with respect to small model transport errors.

Conclusions
We India is in transition between employing HCFC and HFC refrigerants. We present evidence to suggest that India is yet to adopt several common refrigerant blends, including R-410A, R-404A and R-507A, all of which are used extensively in the developed world. India's apparent lack of uptake of refrigerant blends presents as an opportunity for future climate mitigation strategies; if India can be encouraged to bypass HFCs in favour of low-GWP alternatives, substantial CO 2 eq emissions could be avoided.
We also show that following discontinuation of funding from the CDM, some or all of India's manufacturers of HCFC-22 likely 15 resumed venting of the HFC-23 by-product. As India's economy expands, projections suggest that its consumption of HFCs will grow significantly. It is important to implement long-term and continuous HFC, HCFC and CFC monitoring from this region of the world to help India evaluate its progress under the Montreal Protocol. Our 2016 estimates provide a benchmark, against which future changes to India's ODS and HFC emissions can be assessed.

Code availability 20
Hierarchical Bayesian trans-dimensional MCMC code is available on request from Anita Ganesan (Anita.Ganesan@bristol.ac.uk).    Table 2). The shading represents the model uncertainty (