Bottom-up and Top-down Emission Estimates of Hydrofluorocarbons (HFCs) in Different Parts of the World

(1)

MSc Chemistry

Science for Energy and Sustainability

Literature Thesis

Bottom-up and Top-down Emission Estimates of

Hydrofluorocarbons (HFCs) in Different Parts of the World

by

Hannah Flerlage (hannah.flerlage@posteo.de)

UvA student ID: 12370355 VU student ID: 2655594 Amsterdam, May 2020 12 EC Research period: October 14, 2019 – December 24, 2020 Examiner:

Prof. Dr. Jacob de Boer Environment & Health,

VU Amsterdam

Second reviewer:

Dr. Jan P. Dekker, Associate Professor Biophysics, Department of Physics and

Astronomy, VU Amsterdam

Supervisor:

Prof. Dr. Guus Velders (g.j.m.velders@uu.nl)

Institute for Marine and Atmospheric research Utrecht (IMAU), Utrecht University

(2)

1

Abstract

Hydrofluorocarbons (HFCs) are widespread alternatives for the ozone depleting substances (ODS) chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs) and used today in a variety of applications, mainly as refrigerants for cooling and air conditioning or as foam-blowing agents. HFCs do not deplete the ozone layer, but they are very potent greenhouse gases, already now contributing to global warming. Since HFCs are now regulated under the Kigali Amendment to the Montreal Protocol, reliable emission estimations are necessary to enforce regulations. Quantification of emissions is performed with two methods: bottom-up from product inventories or data on chemical sales; or top-down, inferred from atmospheric measurements by inverse modelling or interspecies correlation. Here, the methods are described and results from different parts of the world are reviewed. Emission estimates reported by the different methods have been found to vary considerably. HFC emissions of developed countries (Annex I) were reported to the United Nations Framework Convention on Climate Change (UNFCCC), following the Kyoto protocol. These bottom-up estimates add up to only half of global emissions estimated from atmospheric data. Several studies have shown with regional top-down estimates that this gap is not owed to large-scale underreporting of emissions from developed countries, but mostly due to emissions from developing countries (Non-Annex I). China accounts for a large amount of emissions leading to the gap, but not entirely. Bottom-up and top-down estimations of emissions from other developing countries to identify other large emitters are largely unavailable. Especially South America, West-, Central- and East-Africa, India, the Arabian Peninsula and Northern Australia are not well covered with measurement stations providing atmospheric data for top-down emission estimates.

(3)

2

Summary for non-scientist readers

Hydrofluorocarbons (HFCs) are a class of chemicals used in refrigerators and air conditioners, as foam blowing agent (for example for building insulation), as aerosol propellant, as fire extinguishers and as flame retardants in different products. From all these applications, HFCs are released into the atmosphere. The concentration of HFC-134a, the most abundant one in the atmosphere, is 90 parts per one trillion parts of air.⁠

1_{This seemingly small amount of chemicals has a big effect – HFCs are very potent greenhouse}

gases. Over 100 years, an emission of HFC-134a contributes 1 360 times as much to global warming and climate change as the same emitted amount of CO2 would. For other commonly used HFCs, this number

is even higher.

HFCs were introduced as replacements for the chemically similar compounds CFCs and HCFCs. CFCs and HCFCs destroy the ozone layer and are also greenhouse gases. The production and use of CFCs were stopped by the Montreal protocol, a global treaty that all countries agreed on in 1987 to save the ozone layer. The production and use of HCFCs is phased down with a global stop effectively in 2030.⁠

2_{The protocol}

was successful and today CFCs and HCFCs are used much less and are not produced anymore in most countries. The ozone layer can now slowly recover. HFCs do not destroy the ozone layer, but because they have a large effect of global warming, they are not a final solution either.

In 2016, the countries have decided to include HFCs in the Montreal protocol because of their negative effect on the climate. This was agreed in the Kigali Amendment. The Kigali Amendment is an extension of the Montreal protocol and should phase down the production and use of HFCs. Many application sectors of HFCs are projected to grow very fast in the coming decade, especially the demand for refrigerators and air conditioning will likely increase a lot. Without the Kigali Amendment, HFC emissions were projected to rise fast and make up a big part of total greenhouse gas emissions. To make sure that all countries follow the rules of the Kigali Amendment, reliable estimations of HFC emissions are needed.

Emissions can be estimated in two ways. In the bottom-up approach, information about production, import, export, sales, and more are collected from a country to estimate the amount of HFCs used and how much of that is released in the atmosphere. For the top-down approach, concentrations of HFCs are measured in the atmosphere. From these measurements and information on how chemicals move in the atmosphere, a model can estimate the emissions from a country.

The results from both methods, top-down and bottom-up, often don’t match. For many parts of the world, not enough data is available to make bottom-up estimations or there are no measurement stations close enough to provide atmospheric data for top-down estimations. Especially in South America, West-, Central- and East-Africa, India, the Arabian Peninsula and Northern Australia measurement stations are lacking. For many regions, more and more studies with both top-down and bottom-up estimations of HFC emissions are published.

It is important to have reliable information about HFC emissions and their sources to identify large emitters. This is the basis to enforce the Kigali Amendment and prevent additional global warming caused by HFCs.

(5)

4

Abbreviations and Terms

AGAGE Advanced Global Atmospheric Gases Experiment

Annex I country Countries listed in Annex I of the Kyoto protocol, which have committed to mitigation of climate change; members of the OECD and former Soviet countries. Non-Article 5 countries.

Article 5 country Countries listed in Article 5 of the Montreal protocol. These countries are eligible for financial support from the Multilateral Fund.

Non-Annex I countries.

CCAC Climate and Clean Air Coalition

CFC Chlorofluorocarbon

CO2eq. Carbon dioxide equivalent

ECMWF-IFS European Centre for Medium Range Weather

Forecasts - Integrated Forecast System

EDGAR Emissions Database for Global Atmospheric

Research

EEA European Environment Agency

EGD European Geographical Domain

EMPA Inversion model using FLEXPART and an extended

Kalman Filter

EMPA2 Inversion model using FLEXPART and a Bayesian

Framework

EPA Environmental Protection Agency

EU European Union

F-gas Fluorinated greenhouse gases, including CFCs,

HCFCs, HFCs

FLEXPART FLEXible PARTicle dispersion model

FYROM Former Yugoslav Republic of Macedonia

GAINS Greenhouse Gas and Air Pollution Interactions

and Synergies (model)

GDP Gross Domestic Product

Gg yr-1 _{Giga gram per year}

GWP Global Warming Potential

HCFC Hydrochlorofluorocarbon

HFC Hydrofluorocarbon

HFO Hydrofluoroolefines

IPCC International Panel on Climate Change

ISC Interspecies correlation

JRC European Commission Joint Research Centre

LAC Latin America and the Caribbean

LPDM Lagrangian particle dispersion model

MAC Mobile air conditioning

(6)

5

NAME Numerical Atmospheric dispersion Modeling

Environment

NIES Japan’s National Institute for Environmental

Studies

NILU Inversion model using FLEXPART and a Bayesian

Framework

NOAA US National Oceanic and Atmospheric

Administration

ODS Ozone Depleting Substances

OECD Organization of Economic Co-operation and

Development

PBL Netherlands Environmental Assessment Agency

posteriori Here, his refers to the data obtained as a result of

the modelling

ppt Parts per trillion, commonly used unit in the field

of atmospheric chemistry

priori Here, this refers to data input into the model

RAC Refrigeration and Air conditioning

SOGE System for Observation of halogenated

Greenhouse gases in Europe

t Metric ton

TFA Trifluoroacetic acid

Tier 1 Category in bottom-up emission estimation, less

detailed

Tier 2 Category in bottom-up emission estimation, less

aggregated and more detailed

UN United Nations

UNDP United Nations Development Program

UNEP United Nations Environment Program

UNFCCC United Nations Framework Convention on

Climate Change

UNFCCC United Nations Framework Convention on

Climate Change

UNIDO United Nations Industrial Development

Organization

(7)

6

Figures

Figure 1: Geographical distribution of sampling sites from the Advanced Global Atmospheric Gases Experiment (AGAGE), including stations from Japan’s National Institute for Environmental Studies (NIES),37_{and the US National Oceanic and Atmospheric Administration (NOAA) HATS flask sampling}

program.38_{Some further measurement sites have been used to measure HFCs as described in this work,}

in Cape Point, South Africa19_{, K-Pustza, Hungary}39_{and Finokalia, Greece}40_{. ... 12}

Figure 2: Scheme of top-down emission estimation using inverse modelling. ... 13 Figure 3: Emissions of HFC-410A from the RAC sector in China. The uncertainty assigned to the bottom-up estimated values by Monte-Carlo simulation is highlighted in pink. The dotted line from the study of Wang et al. (2016) displayed for comparison represents projected emissions. From Liu et al. (2019).44_{.. 19}

Figure 4: Differences between emissions inferred from atmospheric measurements and emissions reported to UNFCCC for HFC-134a globally. Shading represents uncertainty. The light blue line represents UNFCCC reported emission estimates where US and EU values are substituted with top-down values from Hu et al. (2017) and Graziosi et al. (2017). East Asian emissions are from Stohl et al. (2010), Chinese emissions from Lunt et al ., 2015, (a-1, red ), Fortems-Cheiney et al. (2015) (a-2, pink and shading), Su et al. (2015) (I-2, light blue triangles pointing down) and Fang et al. (2016) (I-1, dark blue triangles, pointing up). From: Montzka, Velders et al. (2018).⁠1 ... 23 Figure 5: Consumption values for HFC-134a were found for some non-Annex I countries. The Climate and Clean Air coalition published bottom-up studies of HFC consumption in Ghana68_{, Bangladesh}69_,

Vietnam70_{and Indonesia}71_{. The United Nations Development Program released a report for HFC}

consumption from Moldova72_{. The United Nations Industrial Development Organization released reports}

on HFC consumption in Jordan73_{and South Africa}65_{. Further, HFC-134a consumption from some}

countries were reported by government agencies, such as the Ministry of Environment in Chile74_{, the}

National Ozone Unit of Colombia (UTO)75_{and the Environmental Protection Agency of Liberia}76_{. ... 25}

Figure 6: Coverage of measurement stations used to monitor HFC emissions in Europe. ... 28 Figure 7: Emissions of HFC-32, HFC-125, HFC-143a, HFC-152a and HFC-134a from China between 2005 and 2017 according to different studies. ... 31 Figure 8: Different studies on HFC emissions in Europe estimated emissions for different geographical domains. Graziosi et al. (2017) (left, green) grouped emissions in the regions FR (France); UK (United Kingdom); ES-PT (Spain, Portugal); IT (Italy); DE (Germany); NEE (Poland, Czech Republic, Slovakia, Lithuania, Latvia, Estonia, Hungary, Romania, Bulgaria); SCA (Norway, Sweden, Finland, Denmark); SEE (Slovenia, Croatia, Serbia, Bosnia-Herzegovina, Montenegro, Albania, Greece); BE-NE-LU (Belgium, The Netherlands, Luxembourg), IE (Ireland); AT (Austria); CH (Switzerland). Schoenenberger et al. (2018) (center, blue) separated Europe in the regions Turkey (Turkey, Cyprus), Balkans (Serbia, Montenegro, Kosovo, Albania, Bosnia and Herzegovina, Croatia, Slovenia, FYROM), Eastern (Ukraine, Romania, Moldova, Bulgaria), Middle East (Jordan, Lebanon, Syria, Palestine, Israel), Maghreb (Morocco, Algeria, Tunisia, Libya, not colored), Central E (Poland, Slovakia, Czech-Republic, Hungary), Central W

(Switzerland, Liechtenstein, Germany, Austria, Denmark), Western (France, Luxembourg, Netherlands, Belgium), Iberian Peninsula (Spain, Portugal) and British Isles (Ireland, UK), Turkey (Turkey, Cyprus) and Greece, Egypt and Italy as national states; and Keller et al. (2012) (right, red) considered the following groups: central west (Belgium, France, and Luxembourg), central north (Denmark, Germany, and The Netherlands), northwest (Ireland and the United Kingdom), central south (Austria, Italy, and

Switzerland), southeast (Albania, Bulgaria, parts of Greece, Hungary, Romania, and former Yugoslavia), northeast (Czech Republic, Poland, and Slovakia), east (Belarus, Latvia, Lithuania, Moldova, and the western part of the Ukraine), and southwest (Portugal and Spain). ... 34 Figure 9: Emissions of HFC-32, HFC-125, HFC-143a, HFC-152a and HFC-134a from Europe between 2005 and 2017 according to different studies. The studies used different geographical boundaries. Graziosi et

(8)

7

al. did not give uncertainty intervalls for their emissions estimates for the total European Geographical Domain in the supplementary data tables, therefore no error bars are displayed. Uncertainty for regional estimates varied between 15% and 83% for 32, 16% and 80% for 125, 20% and 82% for HFC-143a, 19% and 80% for HFC-152a and 12% and 80% for 134a. Values from Schoenenberger et al. for Europe were obtained by subtracting emissions estimated for subregions Maghreb, Turkey, Egypt and Middle East from the emission estimate for the total domain of the study (compare Figure 6) for better comparison. ... 35 Figure 10:Emissions of HFC-32, HFC-125, HFC-143a, HFC-152a and HFC-134a from the USA between 2005 and 2017 according to different studies. In the ISC study by Simmonds et al. (2015) no values with R2_>0.5

were obtained for HFC-32 for 2007-2010, thus they are not included here. ... 37 Figure 11 Overview of publications of top-down regional emission estimations analyzed in this study per world region of measurements and year of publication. Countries and regions for which measurement-based emission estimations were made are divided in the following world regions: Northern America (USA, North America Annex I), Latin America and the Caribbean (LAC) (Mexico), Africa (South Africa, Morocco, Mahgreb, Egypt), Europe (European Geographical Domain, Europe, EU, Balkans), West Asia (Turkey, Middle East), East Asia (Japan, Taiwan, South Korea, North Korea, East Asia), China, India and Australia... 39 Figure 12: Overview of gases for which top-down emission estimates were made per world region of measurement and year of publication. Every point represents a country or regional emission estimation of one gas, which is indicated by color. In areas of high overlap different shapes show that estimates are from different publications. ... 40 Figure 13: Bottom-up studies providing values for HFC consumption reviewed in this study. Displayed for world regions with predominantly Non-Annex I countries by year of publication. ... 40

Tables

Table 1: Formulas, GWPs, lifetimes and main applications of common HFCs regulated under the Kigali amendment. ... 10 Table 2: Regional and national consumption estimates. ... 21 Table 3: Regional and national emission estimates derived by top-down and bottom-up methodologies. ... 21 Table 4: Central recommendations and focus areas for further research. ... 46

(9)

8

1 Introduction

Hydrofluorocarbons (HFCs) are widespread alternatives for the ozone depleting substances (ODS) chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs) and used today in a variety of applications, mainly as refrigerants for cooling and air conditioning or as foam-blowing agents.⁠

1_{Other uses}

are in metered dose inhalers, sprays, fire protection systems, and solvents.⁠

3,

⁠

4_{ODS are phased out by the}

Montreal protocol, which is often described as a unique example of a global agreement that successfully averted an environmental crisis.⁠

5_{It was formulated as a response to the global threat of chemical}

emissions of halogenated compounds destroying the earth’s protective stratospheric ozone layer. The protocol was agreed upon in 1987 and is still the only UN treaty ratified by all 197 member states.⁠

6_The

ozone layer is now on its way to recovery after countries made concerted efforts of control, regulation and substitution.⁠

7_{Parties are required to report production and trade data of the regulated chemicals to the}

UNEP ozone secretariat on an annual basis.⁠

2

As ozone depleting substances (ODS) are also very potent greenhouse gases, the protocol has been contributing to climate change mitigation as well. The climate benefit of the Montreal protocol was estimated to 10 Gt CO2eq. annual emissions avoided by 2010.⁠

8_{The corresponding radiative forcing avoided}

by regulating ODS in the Montreal protocol amounts to about 35% of radiative forcing from CO2 in 2010.⁠

4

Following the protocol, the phase-out of ozone depleting CFCs was finalized in 2010 and the phase-out of the less, but still ozone depleting HCFCs will be virtually completed in 2020 for non-Article 5 (“developed”) countries and 2030 for Article 5 (“developing”) countries, with only small amounts allowed afterwards for servicing.⁠

2_{Article 5 countries under the Montreal are those parties eligible to receive financial support}

from the Multilateral Fund, while non-Article 5 countries contribute to financial support, following the principles of responsibility and ability to act.

Atmospheric mole fractions of HFCs are growing rapidly, with an average rate of 1.6 ppt per year between 2012 and 2016. The annual growth rates between 2012 and 2016 are higher than between 2008 and 2012 for HFC-32, HFC-125, HFC-134a and HFC-143a, the most abundant HFCs in the atmosphere.⁠

9,

⁠

1_{An overview}

of common HFCs, their relevant properties and applications is given in Table 1.

HFC-23 is not considered here since it is emitted predominantly as a byproduct of HCFC-22 production. It is barely used intentionally, and the patterns of emission and necessary regulations are very distinct from the ones of other HFCs.10

HFCs do not deplete the ozone layer like chlorine- or bromine-containing analogues do.11_{Recent findings}

show an indirect depletion potential due to radiative forcing increasing tropospheric and stratospheric temperatures, which alters atmospheric circulation and accelerates catalytic ozone destruction cycles.12

While this effect has limited impact, HFCs being halocarbons, are potent greenhouse gases.13_Global

warming potentials (GWP) express the effect of a substance on global warming relative to CO2, based on

the mass of the substance emitted. HFCs have GWPs of up to 5 000 and thus significantly contribute to global radiative forcing.14_{Millet et al. (2009) estimated that halocarbons make up 9% of US total}

GHG emissions and 32% of Mexican total CO2eq GHG emissions.15 Currently, the amount of HFC emissions

is equal to about 1.5% of total emissions in CO2-equivalents from all long-lived greenhouse gases such as

CO2 and N2O, despite the comparatively low mole fractions of HFCs in the atmosphere.⁠

1

The fundamental property determining a chemical species radiative efficiency is the absorption and emission behavior at thermal IR wavelengths. Absorption takes place by transitions between the vibrational-rotational energy levels of the molecule. If a molecule strongly absorbs in the IR region of the

(10)

9

spectrum, it can contribute to the greenhouse effect once it is present in the atmosphere. The thermal IR region (4-200 µm or 2 500 to 50 cm-1_{) is relevant for the radiative balance of the earth, as the planet is}

emitting in this wavelength area and the atmosphere does not block radiation in this wavelength window, permitting emission to space. The absorption behavior of a compound is characterized by the spacing of the vibrational energy levels. With increasing complexity of the molecular structure, the number of possible vibrational modes increases and this results in more complex IR absorption spectra.16_{The C-F}

bonds in HFC molecules absorb in the atmospherically relevant region of the IR spectrum, causing the high contribution to radiative forcing of many HFCs in the atmosphere.16,17

Due to the hydrogen atom(s) constituting part of HFCs, their atmospheric reactivity is increased. The hydrogen atom provides a reactive part of the molecule which can be attacked by hydroxyl radicals in the troposphere, leading to degradation.18,19_{Despite this reactivity, many HFCs have relatively long lifetimes,}

for example HFC-125 with 30 years.⁠

1_{Longer lifetimes and stronger IR absorption result in higher GWPs, as}

GWP is the integrated radiative forcing over a period of reference, mostly 100 years.

As a group of potent greenhouse gases, HFCs are subject to the Kyoto protocol and Annex I countries (countries committed to mitigate climate change, according to responsibility and ability to act; OECD and former Soviet countries) must report annual emission data to the UNFCCC. However, there are no specific regulations for the treatment of HFCs in the Kyoto Protocol. In 2016 the phase-down of production and consumption of HFCs was added to the Montreal protocol in the Kigali Amendment.⁠

1_{Under the Kyoto}

protocol, countries report emissions to the UNFCCC, parties to the Kigali Amendment of the Montreal protocol however, will report production and consumption data instead of emissions to UNEP.⁠

2

Article 5 parties to the Montreal protocol establish the baseline for the stepwise phase-down of HFCs in the years 2020-2022 with an additional 65% of the HCFC consumption baseline. Exempt from that is group 2 of the Article 5 countries: Bahrain, India, Iran, Iraq, Kuwait, Oman, Pakistan, Qatar, Saudi Arabia, and the United Arab Emirates will use 2024-2026 as a baseline (+65% of the HCFC consumption baseline). For non-Article 5 countries the baseline is formed by the years 2011-2013 (+15% of the HCFC baseline) and a 10% reduction is set to the beginning of 2019.⁠

2

Without the Kigali amendment or any other kind of regulation of HFC production and use, severe impacts on the global climate were projected. Velders et al. (2015) found that HFC production and consumption would have rapidly increased over the next decades, causing significant increases in HFC mole fractions in the atmosphere. The resulting radiative forcing from unregulated HFCs would have reached 0.22 - 0.25 W m-2_{in 2050 according to the baseline scenario.}20_{Global warming of up to 0.5°C by 2100 was foreseen to}

be caused by unregulated production and use of high GWP HFCs alone.21

The success of the Kigali Amendment is vital for protecting the climate. Without global measures, it is expected that the consumption of HFCs will increase strongly within the next years, driven by a rising demand for air conditioning and refrigeration.⁠

1_{Hence, to limit their climate impact, it is important to}

control and regulate HFC use and to limit emissions to the atmosphere through a working global agreement.

Reliable emission estimates of HFCs are necessary to ensure that this agreement can be enforced. Two approaches exist to estimating emissions: top-down, from atmospheric measurements, or bottom-up, from production and sales data. As Nisbet and Weiss (2010) point out, bottom-up emission estimates are generally more prone to errors and manipulation. Top-down estimates rely on meteorological data like

(11)

10

wind speeds and directions to project atmospheric transportation and obtain emission values by inverse modelling.22_{The spatial resolution of emission estimates is relevant to be able to compare top-down}

emission estimates to bottom-up reports at a national scale. A country-level resolution of emission data furthermore enables the control of emission legislation and the evaluation of policy tools.23

The 2018 scientific assessment of ozone depletion states that bottom-up emission estimates reported by the Annex I countries are accounting for less than half of top-down emission estimates on a global scale.⁠

1

This gap in emissions could in principle be due to underreporting of Annex I countries, systematic differences in the estimation methods or large emissions from non-Annex I countries.

Under the Kigali Amendment, all parties to the Montreal Protocol must now or soon report to UNEP. Thus, there is a growing number of studies on HFC use and emissions in different parts of the world, recently also many from East Asia. China and India are especially interesting in this context because of their large size and the growth potential of the air conditioning and refrigeration markets.24

This review seeks to give an overview of current state of information on HFC emissions in different parts of the world and compare the methods used to derive them.

Table 1: Formulas, GWPs, lifetimes and main applications of common HFCs regulated under the Kigali amendment.

Name Formula GWP (100 year

time horizon)⁠ 1 Lifetime (year)⁠ 1 _{Main Applications} HFC-134a CH2FCF3 1,360 14 Refrigeration, AC, Mobile AC, Insulating foams24_, Aerosols25 HFC-32 CH2F2 705 5 .4 Refrigeration, AC24

HFC-125 CHF2CF3 3,450 30 Refrigeration, AC_{Fire suppression}2524

HFC-143a CH3CF3 5,080 51 Refrigeration, AC24 HFC-152a CH3CHF2 148 1.6 Plastic foams, Aerosols24 HFC-227ea CF3CHFCF3 3,140 36 Extinguishing agent⁠ 9_{, Fire} suppression25

HFC-236fa CF3CH2CF3 7,680 213 Extinguishing agent⁠

9

HFC-365mfc CH3CF2CH2CF3 810 8.9 Insulating foams 24

HFC-245fa CHF2CH2CF3 880 7.9 Insulating foams24

(12)

11

2 Top-down emission estimation

The top-down method estimates emissions from atmospheric measurement data. This approach is especially suitable for HFCs as they are entirely synthetic compounds. Thus, there is no background of naturally occurring fluxes, as it is the case for example for carbon dioxide, where large natural sinks and sources makes the estimation of anthropogenic emissions from atmospheric data more complex.22_Every

mole fraction of HFC detected in the atmosphere clearly originates from human activity. The greater challenge is to determine from where exactly the emissions arise.

On a global scale, emissions can be estimated with simple box models or three-dimensional models, even from individual measurement sites.27_{Emission estimates on a regional scale considered here are derived}

by inverse modelling or interspecies correlation. Back trajectory methods have been employed to allocate emissions of various atmospheric pollutants, but inverse modelling has generally proven to yield more accurate results.28_{Inverse modelling uses chemical transportation models and inversion algorithms to}

retrace emissions on a spatial grid.29_{Using interspecies correlation, ratios between measured HFC}

concentrations and concentrations of a substance of known emission flux, such as carbon monoxide, can also be employed to obtain emission estimates and deduce geographical origins of emissions.27

2.1 Atmospheric measurements

Basis for both inverse modelling and interspecies correlation are atmospheric measurements. There are several networks of in-situ measurement stations around the world, for example the Advanced Global Atmospheric Gases Experiment (AGAGE), it’s subnetwork, the System for Observation of halogenated Greenhouse gases in Europe (SOGE) and the network of the National Institute for Environmental Science (NIES) in Japan. An overview of measurement stations is presented in Figure 1. All networks provide high frequency measurements of HFCs. To quantify HFCs in the atmosphere, automated low-temperature preconcentration and re-focusing are employed before the gases are measured with an automated gas chromatograph coupled to a mass spectrometer.23,29–31

NOAA’s Global Greenhouse Gas Reference Network maintains intensive wide-range air flask sampling across the USA and central analysis of the samples in one of two GC-MS instruments.32_{Fortems-Cheiney}

et al. (2015) include a list of global atmospheric measurement stations that quantify HFC-134a.33_{Next to}

the permanently installed stations, measurements are performed on air sampled on aircraft campaigns.34

Atmospheric HFC concentrations can be subject to seasonal variability.35_{Kuyper et al. (2017) found this}

effect for HFC-152a at Cape Point and reasoned that the seasonal variability is likely due to the comparably short lifetime of HFC-152a and the minimum winter OH radical concentration in the troposphere.19

Because of their sufficiently long lifetimes, HFCs are relatively homogenously distributed in the troposphere and global emissions can thus be estimated with a few measurement sites distributed across the globe. For regional or country-level emission estimates on the other hand, a network of many atmospheric measurement stations in necessary, with different sensitivities to emissions from regional sources.36

(13)

12

Figure 1: Geographical distribution of sampling sites from the Advanced Global Atmospheric Gases Experiment (AGAGE), including

stations from Japan’s National Institute for Environmental Studies (NIES),37_{and the US National Oceanic and Atmospheric}

Administration (NOAA) HATS flask sampling program.38_{Some further measurement sites have been used to measure HFCs as}

described in this work, in Cape Point, South Africa19_{, K-Pustza, Hungary}39_{and Finokalia, Greece}40_.

2.2 Inverse modeling

Inverse modelling enables the estimation of emissions from atmospheric data based on source-receptor relationships. This approach requires chemical transport models to trace back the origin of increases in HFC mole fractions determined from atmospheric data from geographically distributed measurements.31

The process of top-down emission estimation by inverse modelling is demonstrated in Figure 2.

To show chemical transport in the atmosphere, Lagrangian particle dispersion models (LPDMs) are used. Two commonly employed LPDMs are FLEXPART (FLEXible PARTicle dispersion model), which uses three-hourly analysis and forecasts sourced from the European Centre for Medium Range Weather Forecasts - Integrated Forecast System (ECMWF-IFS) and NAME, which is based on analyses of the UK Met Office's Numerical Weather Prediction model.36_{LPDMs simulate atmospheric transport, backwards in time}

operated in a receptor oriented mode, meaning that virtual particles are released at the measurement station and followed backwards in time.36,41

The time frame used for the backwards simulation of the LPDMs vary. Say et al. (2019) employed the NAME model with a 30-day backwards modulation of surface fluxes, when estimating Indian HFC emissions. This is a 10-day longer backwards modeling time than employed by Stohl et al. (2010), who used FLEXPART for their inversion model.34_{Kuyper et al. (2019) used NAME to evaluate emissions from}

(14)

13

Lunt et al. (2015) used the models NAME and MOZART (Model for Ozone and Related Tracers) to infer HFC emissions from Annex I and Non-Annex I countries. MOZART simulates the global changes in emissions, whereas the NAME model is providing a higher spatial resolution for the area closer to the measurement site.31

Emission sensitivity maps, called footprints, are the result of the LPDMs. For the time the simulations run backwards, the sensitivity of the observation at the measurement site to surface emissions is measured and together these values form the footprint.36_{With the LPDM, measured atmospheric concentrations of}

HFCs are connected to emissions on the ground, e.g. from HFC-producing industrial sites. An inversion algorithm then adapts the emissions in the model to optimally fit the resulting modelled concentrations with the observed concentrations.29_{A scheme illustrating the process of inverse modelling is shown in}

Figure 2.

The sensitivity maps are reduced in resolution to lower the complexity of the inversion. With the reduced resolution the grid of the sensitivity map is coarser, meaning that larger geographical areas are contained in one grid cell. One emission value is then derived per cell and thus for all the area contained within it. Cell distribution in the inversion grid can follow country outlines to prevent that one grid cell comprises areas of different countries. Shared grid cells make it more difficult to estimate emissions on a country level, as emissions of one cell need to be ascribed to the countries therein, for example according to fraction of area or relative population share.36

To obtain regional fluxes, an inversion algorithm is fed with the output of the LPDM, measurement data from different sites and bottom-up emission data as a priori information and their respective uncertainties.27

A Bayesian framework is employed to better account for uncertainties in the a priori information. Within the Bayesian framework, the inversion algorithm is taking advantage of a priori data considering uncertainties contained therein and gives the most likely solution regarding both a priori emissions data and measurement data.29

To obtain prior data for Article 5 countries for their model, Schoenenberger et. al. (2018) subtracted all reported emissions from Annex I countries to the UNFCCC from global emission estimates and then divided the rest to the non-Annex I countries according to population density. For Annex I countries, a priori emission estimates were taken from reports to the UNFCCC.40_{In another study, no specific, spatially}

resolved emission inventory was available for China, hence global total emissions of HFCs in 2011 were used.⁠

9_{These were likewise disaggregated resembling approximate population densities. A spatial}

(15)

14

distribution of emissions oriented at data of night light distribution from NOAA, serving as a proxy for population density, was used as prior in several studies.19,34

The posteriori emissions are obtained after the model run several iterations, optimizing the model output to fit atmospheric observations. The difference between posterior to prior emissions indicates a growth in emissions or stems from methodological discrepancies between bottom-up (prior) and top-down estimates.19

Error sources and uncertainty of inverse modelling

Atmospheric measurements of HFC mole fractions are relatively precise with only small errors from the GC-MS analysis of the air samples in the order of about 2%. The determination of background level atmospheric mole fractions poses a relevant, but mostly not dominant source of uncertainty.42,42

Therefore, the error values of top-down emission estimates are dominated by model errors, for example from the chemical transport model.32

Inaccurate model simulations are causing relatively high uncertainties of emission values derived by inverse modelling.⁠

9_{Statistical information about the uncertainty included in the emission estimates}

generated by inverse modelling can be obtained by producing several results, by varying input and model parameters, to analyze variance and obtain uncertainties as spread of standard deviations. Multiple emission estimates from uncorrelated inversions were generated for example by Say et al. (2016) when estimating UK's HFC-134a emissions from atmospheric measurements and can be used to ascribe uncertainties to the derived annual emission values.43

The emissions of HFCs with short atmospheric lifetimes are systematically underestimated by inverse modeling methods from atmospheric data. As the particle dispersion is modeled backwards in time, some of the HFC can be lost during that transportation time by natural atmospheric degradation. The amount of HFC detected by the measuring station will then be less than the amount emitted at the source. For example, for HFC-152a (comparatively low lifetime, see Table 1) in a 20 day backward simulation this error can lead to a result up to 3.5% below the actual value.29

Motivated by understanding systematic uncertainties in models based on inversion, Brunner et al. (2017) tested four such models that were run with the same data from atmospheric measurements from three measurement stations in Europe and used the same a priori emission values.36_{Several general insights}

into possible error sources of inverse modelling of HFC emissions were gained. Factors that were varied in their investigation were the type of Lagrangian transport model (either FLEXPART or NAME), the inversion method (Bayesian or extended Kalman filter), the approach of baseline mole fractions, the spatial grids and the a priori uncertainty.36

The four models were EMPA (FLEXPART, extended Kalman filter), used previously by Brunner et al. (2012)44_{, EMPA 2 (FLEXPART, Bayesian framework), used by Stohl et al. (2009)}29_{and Vollmer et al. (2009)}45_,

NILU (FLEXPART, Bayesian framework), used by Thompson and Stohl (2014)46_{and UKMO (NAME, Bayesian}

framework), used by Manning et al. (2011)47_{. Atmospheric observations were measured at the sites Mace}

Head (coast, Ireland), Monte Simone (mountain, Italy) and Jungfraujoch (mountain, Switzerland). The models differed in the used background concentrations. NILU background concentrations were lower than those of the other models, compared after one year at Mace Head. At Jungfraujoch and Monte Simone more frequent pollution events were observed which led to more uncertainty in the background level definition.36

(16)

15

FLEXPART based models were found to yield better results than the NAME-based UKMO system. This better performance was evident especially at Jungfraujoch, where differences in levels were bigger than at Mace Head. The different dispersion models, underlying meteorological models and the model setup like the particle release height for the backwards simulation are all factors that could possibly contribute to the performance difference.36_{The models are not reproducing observations from all measuring stations}

equally well. Mace Head station observations were better fitted by the priori model values in time than observations from the two mountain stations Jungfraujoch and Monte Simone. The posteriori results show good fits for Mace Head and also for Jungfraujoch but the observations from Monte Simone are somewhat less well reproduced.36_{For some areas, for example Germany, different models show very different spatial}

patterns of posterior-prior emissions.36

Aggregation of emissions to country levels poses a significant error source. Emissions from grid cells containing several countries need to be distributed. This distribution varies in the models and follows either the area fractions covered by the respective countries (EMPA, NILU) or the relative share of population within the emission grid cell area (EMPA 2). The fourth model considered uses grid cells oriented at national borders and thus no distribution is necessary. For emission grid cells covering land and sea, all emissions were attributed to the land in all models, assuming no relevant emissions over the sea. The coarse grids led to total country emission errors of 1-6%.36_{The use of a proxy like the night light}

distribution for population density may contribute to uncertainty added by aggregating or disaggregating emission values. Bottom-up inventories are generated from country level data, which suggests that there could be stronger error correlations within country borders than across in the prior information. However, this approach was not used in any of the models examined by Brunner et al. (2017).36

The emission estimates produced by the four different models varied significantly, mainly for Spain, Portugal and Italy but also for Germany, France and the UK. For the individual country emission estimates the uncertainty was higher for countries further from measurement stations, like Spain and Portugal. The uncertainty (1σ standard deviation) varied from 5-22% for the UK, 16-28% for France, 38-43% for Germany, 23-63% for Italy and 42-51% for the Iberian Peninsula (Spain and Portugal). The difference between the estimations of the models was up to a factor of 2 and for some countries, where overall uncertainty was also higher up to a factor of 3.36

Why the four models produced different results could not be fully explained. The results of the four different models do not fall within the uncertainty intervals of each other. This mismatch shows that the uncertainty given by the models is smaller than the real uncertainty. The real uncertainties are consequently found to be subject to parametric and structural uncertainties, which are not accounted for in the given uncertainty intervals. Several factors playing a role in creating the differences not covered by the analytical uncertainties were identified. Amongst are the subsampling of observations used in the model, the treatment of the background and whether the correlation structures, especially the spatial distribution, of the prior uncertainties are considered and with what magnitude.36

Typical uncertainty values of top-down studies reviewed here are around 25%, ranging from below 10% up to 90%. The high uncertainty values of above 80% are mostly associated with relatively small emission values.

(17)

16

2.3 Interspecies correlation

To estimate emissions via the interspecies correlation technique, the correlation of the HFC emissions (target compound) with those of a tracer compound is exploited. The target emissions can be derived from the emissions of the tracer compound by scaling it with the ratio of target/tracer compound according to the correlation. For both the target and the tracer molecules, baselines can be established from the measured atmospheric concentration data. The relative concentration enhancements of the observed mole fractions over the baseline (Δ) are then multiplied by respective molecular weights (𝑀) of the compounds and divided to give the ratio of tracer (𝐸𝑟𝑒𝑓 ) and target ( 𝐸𝐻𝐹𝐶−𝑋 ) emissions, from which the

emissions of the respective target HFC (𝐸𝐻𝐹𝐶−𝑋) are calculated, according to the following equation.48–50

𝐸𝐻𝐹𝐶−𝑋= 𝐸𝑟𝑒𝑓

𝑀𝐻𝐹𝐶−𝑋Δ𝐻𝐹𝐶−𝑋

𝑀𝑟𝑒𝑓Δ𝑟𝑒𝑓

If the slope of the regression curve is used for interspecies correlation (ISC) emission estimates, the baseline concentration does not need to be subtracted from the measurement data. In that case, instead of considering enhancements, the total mixing ratios can be used.51

By using ISC, emissions on a national scale can be calculated from atmospheric measurements if the emissions of a suitable reference compound on the same scale are known. In a first step, a tracer molecule with sufficiently significant correlation with the target molecule is selected. The emission values for the tracer compounds are taken from different sources, calculated by a bottom-up approach or by inverse modelling.

A prerequisite for the unbiased application of the interspecies correlation method is that the sources of tracer and target compound have effectively the same origin, both in time and space.48_{Often, HFC}

emissions originate from common sources, in the dominant application as refrigerant blends, for example as R-410A, a 50%/50% mix of HFC-32 and HFC-125 and are thus correlated with each other.14_{HFCs also}

correlate statistically significant with HCFCs, the production of which is often co-located with HFC production and whose application is also similar to the HFCs which replace them. Significant correlation is also found with CO, which is a general proxy for human and industrial activity, emitted from fossil fuel burning in transportation or heating.

Simmonds et al. (2017) derived US emissions of several HFCs using HFC-125 as a reference and basing their estimate on HFC-125 reported emissions to the UNFCCC.52_{Kim et al. (2010) found significant correlation}

between HCFC-22 and HFC-143a and derived the emission value of the HCFC-22 tracer by inverse modelling using the FLEXPART transport model.48_{Fang et al. (2012) compared CO and HCFC-22 as tracer}

molecules and found comparable results in their estimation of Chinese HFC-134a emissions.53_{Millet et al.}

(2009) derived halocarbon emissions from the US and Mexico from aircraft measurement data by using CO as a tracer molecule.15

(18)

17

3 Bottom-up emission estimation

Bottom-up emission estimates are based on statistical data.22_{Data sources can be direct surveys of HFC}

producing industry,54,55_{data on imports and exports,}25_{servicing companies and waste processors,}55

assumptions and estimations about HFC containing or related products43_{and surveys of end users.}25

Information for national inventory estimates can be (bulk) chemical sales (sometimes referred to as top-down data) or market data such as equipment and product sales (sometimes called bottom-up data).56

Before, the terms top-down and bottom-up have also been used to describe the estimation of F-gas demand from either the Gross National Product (GNP) (top-down) or by extrapolating historic demand (bottom-up).57_{Within the context of this work, all these approaches are different forms of bottom-up}

emission estimates, as they are not based on atmospheric measurements. From the sales data, HFC production and consumption of a country can be estimated. The amount of chemical production minus the amount exported plus the amount imported is the country's consumption of a chemical, as defined in the Montreal Protocol.14

According to IPCC guidelines for the estimation of HFC emissions, prompt emissions originate from uses such as metered dose inhalers or open-cell foams, from which HFCs are emitted within the first two years.56_{For this group of applications, chemical consumption (bulk) can be assumed to be emitted within}

this time, making estimations less complex, as bank effects do not need to be considered.

Banks are defined as the difference between consumption of a chemical and the amount already emitted.56,58_{Banks build up due to a time delay in emission from applications such as air conditioning,}

refrigeration, fire protection and closed cell foams, where emissions can occur many years after manufacturing (consumption of bulk chemical). Emissions arise during the use-phase through leakages or servicing, from recycling at the end of life or from landfills thereafter. For many applications it can be assumed that emissions occur before the end of life and the bank is estimated by the total number of devices in use multiplied with their chemical charge.56

For mobile air conditioning (MAC) the amount of banked HFCs can be estimated from automobile statistics, which sometimes even contain details on the fraction of cars equipped with air conditioning. Estimates not considering time delay of emissions are referred to as potential emissions, as they are assuming the release of the total amount of chemical consumed equals the annual emissions. This leads to overestimation, as large quantities of the chemical consumed (e.g. filled into devices) will be banked in equipment for years, being released slowly during use or at the end of life. If countries are reducing the consumption of HFCs, this approach would lead to underestimation of annual emissions, as the emissions from banks could surpass the amounts consumed e.g. assumed emitted that year.56

The IPCC guidelines present two approaches to HFC emission estimation.56_{One is the mass-balance}

approach, which can be used if equipment is refilled annually and the market is static in number of devices and sub-application composition. Then, emissions equal annual chemical consumption levels and potential emissions are actual emissions.

The other approach is the emission-factor approach, necessary for appropriate estimation in most situations. In this approach, emissions are inferred from consumption data and emission factors. Both approaches can be performed in an aggregated (IPCC method category Tier 1) or disaggregated (IPCC method category Tier 2) manner, where information at least about sub-application distribution if not more detailed is considered.56_{The emission factors depend on the sectors, as different uses imply different}

(19)

18

factors differ for every stage of a life cycle of a product containing HFCs. Usually, three stages are differentiated: production, use-phase including emissions from servicing and leakages, and disposal or recycling, the last of which was found to have the largest contribution.55_{Because applications of HFCs vary}

widely, it is rewarding to examine sub-applications separately, following a Tier 2 method criterion of the IPCC. Aggregated information (Tier 1) is, however, for many sectors the only form data is available in.56

Parties listed in Annex I of the Kyoto protocol are reporting national emissions of greenhouse gases, including F-gases (HFCs, perfluorocarbons, SF6 and NF3) to the UNFCCC, based on inventory data.23

Reported data to the UNFCCC used in the present study was accessed at the UNFCCC GHG data interface.59,60_{HFCs are included in sections 2E and 2F, which treat the production and consumption of}

halocarbons and SF6.43 The consumption can be estimated from several factors, such as the number of

refrigeration devices in a country. Emission functions give national emission values.61

Using a model for refrigeration and air conditioning (RAC) applications, bottom up estimates of HFC emissions from that sector can be derived. RAC is a major use sector of HFCs. In the UK for example, RAC applications comprise 70% of HFC-134a emissions, of which a big contribution comes from MAC in cars.43

The RAC model used for national inventories considers 13 sub-applications accounted for separately. Emissions are calculated from activity data, emission factors (differentiated according to life cycle stage) and assumptions about the market.43

By varying the input factors of the RAC model such as refill, unit lifetime, market penetration (% of cars with AC) and life-cycle emission rates, Say et al. (2016) tried to test the sensitivity of the model and bring the bottom-up estimations closer to the emissions they derived from atmospheric measurements. For example, the default assumption of annual refill does not agree with industry experts who state a refill every 2 years as a typical interval for the UK. By reducing the refills to half, the bottom-up estimate reduced by 1 Gg in 2010. Market penetration and refill were the factors that the results showed to be most sensitive for. Combining a lower refill and penetration rate can significantly decrease the MAC model emission estimate. This lowered value is, however, still higher than the estimates inferred from atmospheric measurements.43

The Greenhouse Gas and Air Pollution Interactions and Synergies (GAINS) model from the International Institute for Applied Systems Analysis uses emission factors adapted to the country, for example with specific information about maintenance levels or transportation fleet composition.62

Wherever information is available, emission factors are adjusted for specific HFC emission sources. For each sector, a common GWP is calculated from the shares of the used HFCs and their respective GWPs.63

Emissions were distributed in time (monthly) and space on a 0.5° x 0.5° geographical grid.63

The Emissions Database for Global Atmospheric Research (EDGAR) from the European Commission Joint Research Centre (JRC) and the Netherlands Environmental Assessment Agency (PBL) contains bottom-up estimates of HFCs, among other greenhouse gases, on a country level for most countries globally up to 2010. The method used is based on sector and technology specific emission factors, which are adopted to the country. Abatement measures are accounted for in the calculation. Emissions are calculated for 0.1° x 0.1° grid cells. The distribution of national inventory-based sector-specific emission estimates into these grid cells follows spatial proxies like manufacturing sites or population density. The data input for EDGAR are international statistics.64

In the 2017 UNIDO report on the South African HFC inventory, trade data on HFC import and export was collected from state agencies, official trade statistics and major importers and suppliers and verifications from further stakeholders. This data was used for an bottom-up estimation of South African HFC

(20)

19

emissions.65_{Liu et al. (2019) collected production and consumption data from major room air conditioning}

manufacturers and air conditioning recycling companies to estimate Chinese annual emissions of R-410a, a blend of HFC-125 (50%w) and HFC-32 (50%w) from 2006 to 2017.55

Error sources and uncertainty of bottom-up estimations

The uncertainty of bottom-up estimates is often relatively high and difficult to quantify due to the many assumptions made. Uncertainty in the mass-balance approach is mainly determined by quality and completeness of the information underlying the consumption values: data about import and export and chemical sales, or market data about product sales.

In the generally used emission-factor method, additional sources of uncertainty lay within the specific emission factors for sub-applications or blends and further assumptions necessary for the calculation. Especially in the disaggregated (Tier 2) form, when emissions are estimated at the sub-application level or even more detailed, the selection of appropriate emission factors is relevant.56

One of the major obstacles in the collection of data for detailed national inventories (Tier2) are confidentiality issues preventing companies from disclosing unaggregated data.56

Activity data and leakage rates of cooling equipment contain high uncertainty, and this translates into high uncertainty of bottom-up emission estimations.63_{The uncertainty of emission estimates from different}

sectors is highest in stationary air conditioning. Commercial refrigeration and mobile air conditioning emission estimates also contain high uncertainty.66

While emission factors seem to be most important for the accuracy of historical emission estimates, uncertainty in activity data is the largest source of the total uncertainty in emission projections.63

The efficiency of air conditioners, refrigerators and other devices is gradually improving over time and alternatives to HFCs are increasingly used. This might lead to out-of-date assumptions for the bottom-up estimates of emissions from numbers of devices and could be a factor contributing to the overestimation of emissions in bottom-up estimates for some HFCs compared to top-down estimates found in some parts of the world.63

Figure 3: Emissions of HFC-410A from the RAC sector in China. The uncertainty assigned to the bottom-up estimated values by Monte-Carlo simulation is highlighted in pink. The dotted line from the study of Wang et al. (2016) displayed for comparison represents projected emissions.

(21)

20

In the study of R-410A emissions by Liu et al. (2019), uncertainties of for the bottom-up estimations were analyzed by Monte-Carlo simulation (see Figure 3).55_{The results of this study, furthermore, show that}

almost all emissions occur at the end of life stage of room air conditions. This highlights the great need for proper disposal and handling of recycling processes and the requirement to properly account for these often very unregulated sources in bottom-up estimates.

The methods used by each country to derive national inventories of HFCs are different, which makes a comparison between countries difficult. Incompleteness and the use of non-adapted emission factors further hinder comparability of the emission estimates from different countries. These findings for Latin America and the Caribbean can be generalized to the global scale, as harmonization efforts beyond the IPCC guidelines are scarce. Sometimes inventories are incomplete, e.g. not considering one or several sectors of HFC use.63

Most bottom-up studies reviewed here do not give any estimate of the uncertainty contained in the given values. Uncertainties are neither included in bottom-up data reported to the UNFCCC.

(22)

21

4 Overview of HFC consumption and emission data

The collected data can be accessed with the following links: Table 2: Regional and national consumption estimates.

https://drive.google.com/file/d/1TUsGE1t6V33TtG_7AT8v5ocpEgZHFxbr/view?usp=sharing

Table 3: Regional and national emission estimates derived by top-down and bottom-up methodologies.

(23)

22

5 Results and Discussion

Firstly, the results from the two emission estimation methods are compared in a general way and then specifically for each of the nine most common HFCs. The geographic coverage of measurement stations is discussed afterwards. Furthermore, the situation in different parts of the world is described, finalizing with an overview of the geographical distribution of the emission and consumption estimation studies collected.

5.1 Comparison of top-down and bottom-up estimations of HFC emissions

A few comparative studies have been conducted on bottom-up and top-down emission estimates of HFCs. For other greenhouse gases too, these two approaches have been compared, for example for methane emission estimations.67

Many studies that conducted atmospheric measurements of HFCs and estimated emissions from a top-down approach using an inversion model or interspecies correlation compared their findings with bottom-up estimates, either from EDGAR or the UNFCCC reports of Annex I countries. A common finding is that for aggregated emissions of different HFCs together on a larger geographical scale the methods agree; bottom-up estimates match values derived from atmospheric measurements. Regarding specific compounds however, quite large discrepancies in both directions are found, which cancel out when aggregating the emissions.

Graziosi et al. (2017) found a difference of only 13% between their inversion-based results and reported emissions to the UNFCCC when looking at total HFC emissions in CO2eq. from Europe. For individual

compounds, the different emission estimation methods resulted in larger discrepancies.23_{The EDGAR}

inventory is generally overestimating HFC emissions in Europe. For most HFCs (except HFC-365mfc and HFC-32, where large underestimation is found), EDGAR is emissions are higher than top-down derived emissions. On average, EDGAR estimates for aggregated HFC emissions from the European geographic domain were 35% higher than values derived from inverse modelling. The inversion study did not show any significant trend while both UNFCCC and EDGAR show an increasing trend from 2003 to the respective year data was available, 2014 for UNFCCC reported values and 2010 for EDGAR data.23_{The median of}

emissions estimated of European countries by all four models compared by Brunner et al. (2017) is 24% higher than the total HFC emissions reported to UNFCCC.36_{Say et al. (2016) find in their study on UK HFC}

emissions a large emission gap between bottom-up and top-down estimates. Values reported to the UNFCCC were much higher than their results inferred from atmospheric measurements by inverse modelling. The growth rates of the reported data from bottom-up estimates was significantly larger for many years between 1995 and 2012 than the values derived from air sample measurement data. One major reason for the difference were wrong assumptions in the bottom-up model, suggesting higher than actual consumption.43

On a global scale, emission estimates reported to the UNFCCC by Annex I parties make up for only half of the emissions derived by atmospheric measurements.14

Lundt et al. (2015) studied five different HFCs and found that differences between reported and modelled estimates observed for single compounds from one same country or region cancelled out when considering aggregated HFC emissions. For Annex I countries, emissions derived with their inverse modelling approach matched very well with emissions reported to the UNFCCC by these countries when considering the aggregated CO2eq HFC emissions. This agreement of emission values suggests that the

(24)

23

previously found mismatches in reported and modeled emissions are largely due to non-Annex I country emissions which are required to be reported only from this year on.31

For total HFC emissions, it is estimated that China contributes only about 35% to emissions of HFCs from the non-Annex I group of countries, indicating that other big non-Annex I emitters exists and share responsibility for the gap.⁠

1,10

Emissions of HFCs reported to the UNFCCC and measured by NOAA are increasing. However, the gap of reported emissions and measured emissions is increasing as well, as demonstrated in Figure 4 for HFC-134a. A small difference could be explained by emissions from banks, when they are not accounted for accurately in bottom-up estimations.14

Figure 4: Differences between emissions inferred from atmospheric measurements and emissions reported to UNFCCC for HFC-134a globally. Shading represents uncertainty. The light blue line represents UNFCCC reported emission estimates where US and EU values are substituted with top-down values from Hu et al. (2017) and Graziosi et al. (2017). East Asian emissions are from Stohl et al. (2010), Chinese emissions from Lunt et al ., 2015, (a-1, red ), Fortems-Cheiney et al. (2015) (a-2, pink and shading), Su et al. (2015) (I-2, light blue triangles pointing down) and Fang et al. (2016) (I-1, dark blue triangles, pointing up). From: Montzka,

(25)

24

HFC-134a

Global HFC-134a emissions amounted to 223 ± 22 Gg yr-1_{in 2016.}

⁠

1_{The global annual mean atmospheric}

mole fraction of HFC-134a was 89.5 ppt in 2016 and grew about 7% per year between 2012-2016 with an increasing rate.⁠

1_{The emissions of HFC-134a are the fastest growing HFC emissions.}

HFC-134a was the most emitted HFC in Europe in the period from 2003 to 2014. The emissions derived from atmospheric measurements (20.1 ±6.3 Gg yr-1_{) are 25% lower than the emissions derived from data}

reported to the UNFCCC. This mismatch likely originates from too high emission factors used in the bottom-up calculation. EDGAR shows a yet higher emission estimate.23_{Keller et al. (2012) found much}

higher emissions of HFC-134a by inversion than reported from Romania in 2009.39_{Say et al. (2016) used}

the NAME model and the inversion technique inTEM from the UK Meterological Office to estimate HFC-134a emissions from the UK from atmospheric measurement data and compared it to UNFCCC reported emissions. They find that reported emissions estimated with the bottom-up RAC model used for UNFCCC reporting are almost twice the magnitude of top-down derived emissions.43

Hu et al. (2015) used a Bayesian approach to the inversion problem of estimating HFC-134a emissions in the US. Emission estimates of HFC-134 in the US derived from inverse modelling agree with reported emission values from the US EPA.32,42

Emissions reported to the UNFCCC by Japan in 2007 were confirmed by inverse modelling, which returned emissions of 3.1 Gg yr-1_{for 2008. Further top-down studies agreed with reported values for previous years}

which show a decreasing trend of emission.27_{For 2010, emissions of HFC-134a from Japan derived by}

inverse modelling by Fortems-Cheiney et al. (2015) were with 12 Gg yr-1_{much higher than reported to the}

UNFCCC and also much higher than other top-down (inversion) estimates for 2008 would indicate. Bottom-up estimates of Japanese emissions from EDGAR, which were used as a priori information, are even higher than the posterior results from Fortems-Cheiney et al. (2015). The authors suggest the choice of the prior as major influence explaining the different results of the top-down studies.33

Comparing atmospheric measurements with UNFCCC reports for HFC-134a on a global scale exposes a big gap. The UNFCCC reported emissions from Annex I countries account for less than half of the emissions derived from measurements globally.⁠

1_{This gap in HFC-134a emissions constitutes the majority of the}

global gap between reported and measured emissions of all different HFCs together.

Comparing global HFC-134a emissions measured in the atmosphere with regional estimates both reported to the UNFCCC and from atmospheric measurements and inverse modelling, shows that a significant amount of emissions originates from non-Annex I countries other than China. Chinese emissions of HFC-134a in 2016 were estimated to 30 (24-36) Gg yr-1_{by inverse modelling.}

⁠

9_{As much as 30% of global}

HFC-134a emissions are left to stem from the group of non-reporting (non-Annex I) countries excluding China.⁠

1

While previous bottom up estimates for HFCs in India were 1.1 Gg yr-1_{in 2005 for HFC-134a and 0 Gg yr}-1

were reported to the UNFCCC in 2010, the first top-down estimation published by Say et al. (2019) found significant emissions of HFCs from India in the year 2016 (8.2 (6.1–10.7) Gg yr-1_{for 134a). This large}

mismatch indicates a very rapid growth in HFC emissions since 2005 and (or) a relevant difference between top-down and bottom-up estimations.34_{However, this measured value of HFC emissions from India can}

only explain a small fraction of the 30% gap, which was about 67 Gg yr-1_{in 2016, in emissions left to stem}

(26)

25

Consumption values can give some indication about possible emissions and are especially interesting from non-Annex I countries, which are not reporting emission values to the UNFCCC. Consumption values estimated from various reports for non-Annex I countries are compared to consumption values for China in Figure 5. Consumption of HFC in all listed countries are much lower than Chinese consumption, which leads to the conclusion that emissions from those countries are not likely to be very significant in explaining the gap.

Figure 5: Consumption values for HFC-134a were found for some non-Annex I countries. The Climate and Clean Air coalition

published bottom-up studies of HFC consumption in Ghana68_{, Bangladesh}69_{, Vietnam}70_{and Indonesia}71_{. The United Nations}

Development Program released a report for HFC consumption from Moldova72_{. The United Nations Industrial Development}

Organization released reports on HFC consumption in Jordan73_{and South Africa}65_{. Further, HFC-134a consumption from some}

countries were reported by government agencies, such as the Ministry of Environment in Chile74_{, the National Ozone Unit of}

Colombia (UTO)75_{and the Environmental Protection Agency of Liberia}76_.

HFC-32

Global HFC-32 emissions in 2016 are estimated at 35 ± 4 Gg yr-1_{and the global mean atmospheric mole}

fraction was 11.9 (11.2–12.6) ppt.⁠

1

Global emissions of HFC-32 derived from atmospheric measurements were two times higher than UNFCCC reported emissions from Annex I countries. Emission estimates from China can explain the majority of the emission gap of HFC-32.⁠

1_{HFC-32 emissions are not considered to scale with population density, because}

they are expected to originate from production sites rather than from products. Total Indian emissions are therefore not available but only emissions from northern central India where the measurements were performed and the model shows the highest sensitivity.34

The emissions of HFC-32 from the EU and USA, the biggest Annex I emitters, were higher in the UNFCCC reported estimates than in estimates derived from atmospheric measurements. For HFC-32, emissions of 2.3 ± 0.8 Gg yr-1_{in Europe were estimated. While this emission value agrees with UNFCCC data, the annual}

0.35 0.96 0.02 0.02 1.07 0.72 0.06 0.48 0.64 3.08 64.00 0 10 20 30 40 50 60 70 0 1 1 2 2 3 3 4 G g/y r G g/y r