Investigating the SOA formation and volatility using ToF-CIMS measurements

REPORT SERIES IN AEROSOL SCIENCE N:o 240 (2021)

Investigating the SOA formation and volatility using ToF-CIMS measurements

ARTTU YLISIRNIÖ

Department of Applied Physics Faculty of Science and Forestry University of Eastern Finland

Kuopio, Finland

Academic dissertation

To be presented, with the permission of the Faculty of Science and Forestry of the University of Eastern Finland, for public criticism in auditorium MS301,

Yliopistonranta 1, Kuopio, on Friday 8.10.2021 at 12 o'clock.

Kuopio 2021

Author’s Address: Department of Applied Physics P.O.Box 1627

FI-70211 University of Eastern Finland arttu.ylisirnio@uef.fi

Supervisors: Professor Annele Virtanen, Ph.D.

Department of Applied Physics University of Eastern Finland

Associate Professor Siegfried Schobesberger, Ph.D.

Department of Applied Physics University of Eastern Finland

Reviewers: Dr. Hilkka Timonen, Ph.D.

Finnish Meteorological Institute Helsinki, Finland

Professor Miikka Dal Maso, Ph.D.

Tampere University Tampere, Finland

Opponent: Professor, Mattias Hallquist, Ph.D.

Department of Chemistry & Molecular Biology University of Gothenburg, Sweden

ISBN 978-952-7276-63-1 (printed version) ISSN 0784-3496

Helsinki 2021 Unigrafia Oy

ISBN 978-952-7276-64-8 (pdf version) http://www.atm.helsinki.fi/FAAR/

Helsinki 2021

Acknowledgements

This thesis was conducted in the Department of Applied Physics of the University of Eastern Finland during years 2016-2021. I wish to thank the Head of the Department, Kari Lehtinen for providing me with the working facilities for this thesis.

I thank Prof Miikka Dal Maso and Dr. Hilkka Timonen for reviewing this thesis and Prof.

Mattias Hallquist for agreeing to be my opponent.

I wholeheartedly thank my supervisors Prof. Annele Virtanen and Associate Prof. Siegfried Schobesberger for their guidance and supervision during this PhD work. Annele has allowed me to grow as a scientist on my own time. She has also looked through her fingers at my projects which, while often benefiting the research group, have not truly advanced my PhD work. Sigi has always had time to discuss any research or instrumental details, regardless of how insignificant they had been.

The invaluable contribution of all my co-authors is gratefully acknowledged, without their help the studies shown in this thesis could not have been completed. Special thanks are sent to Iida Pullinen, Angela Buchholz, Pasi Miettinen, Eetu Kari, Zijun Li, Aki Pajunoja and Olli Väisänen for their practical help with measurements and studies over the years.

My gratitude also goes to our wonderful work community in the Aerosol Physics research group. When I first came to this group as a summer worker more than 8 years ago, I have always felt myself welcomed and appreciated. I truly value our group’s habit to gather for common coffee breaks to discuss work issues or daily gossip, laugh at not-always-so-good jokes or just stare at the floor in silence. Laughter has often echoed around the office corridors and I have heard some slightly astonished comments from neighbouring research groups such as “Do those guys ever work?” I do think that our common coffee breaks have a major role in both helping with keeping up the working spirit and inspiring new research ideas. I especially want to thank my office mates Arto Heitto, Tuuli Miinalainen and Kimmo Korhonen for their peer support and great attitude.

I also thank my friends Simo Ojanen, Petteri Ronkainen, Matti Niskanen, Timo Ikonen and Pekka Kuoppa for the fun moments and discussions over the years that helped me to peek out from the academic bubble. I recite Pekka’s wise words “Deadline makes an M.Sc.”, as it also seems to make a PhD.

I want to thank my parents and brothers for encouraging me to apply to university and supporting me during my PhD work. Especially when answering the question “so, when will you actually graduate?” was so often “probably next summer”. Fortunately, that summer has finally arrived.

My especially deep gratitude belongs to my wife Kiira, who has supported me when the days have been long, and stress has been a burden during both my university studies and now this PhD work. Additionally, my deep gratitude belongs to my small ray of light Linnea, whose joy of learning new things never ceases to delight me.

Investigating the SOA formation and volatility using ToF-CIMS measurements Arttu Sakari Ylisirniö

University of Eastern Finland, 2021

Abstract

Aerosol particles can be found everywhere in the air, indoors or outdoors, and influence both directly and indirectly the health of humans and other organisms and the environment in which they live. These tiny, floating liquid or solid particles have many properties; they can penetrate deep into human lungs, impact the quality of the air we breathe, modify the formation and chemistry of clouds, and influence the climate of the whole planet.

The volatility of aerosol particles plays a major role on their lifetime and size-evolution and these properties affect how they influence the climate. In recent years, special attention has been paid to Secondary Organic Aerosols (SOA) as they have been demonstrated to constitute a large mass fraction of the particulate matter in the ambient air. In the atmosphere the SOA are usually produced from oxidation products of Volatile Organic Compounds (VOCs) released into the air from biogenic and anthropogenic sources. It is evident that clarification of the volatility and chemical composition of the organic vapours forming the SOA would make it easier to incorporate the evolution of SOA particles into climate models and assess the impact of specific air pollutants.

SOA volatility and chemical composition can be studied with FIGAERO-CIMS. In this thesis, my first aim was to improve the volatility calibration of this instrument and to develop methods to ease the interpretation of the volatility data. The second aim was to clarify how different oxidative conditions and different precursor VOCs affect the volatility of the formed SOA and finally the third aim was to investigate if SOA from single VOC precursors is comparable to SOA from real-life plant emissions.

As a result of this thesis, we were able to significantly improve the accuracy of the FIGAERO-CIMS volatility calibration and successfully apply PMF in the interpretation of the complex composition and volatility data sets, giving additional information about the physicochemical properties of the SOA particles. Furthermore, the results of this thesis show that the volatility of SOA particles decreases as the average oxidative state of the SOA particles increases, as was to be expected. However, the structure of the precursor SOA was observed to have a major influence on both the volatility and average oxidative state of the SOA. Especially, the oxidation of acyclic sesquiterpenes was shown to produce more highly oxidized products with very low volatilities than the oxidation of α-pinene in similar oxidative conditions. Interestingly, the SOA mass yield of these sesquiterpenes was however lower than the SOA mass yield of α-pinene. Nonetheless, the most highly oxidized and least volatile SOAs were produced from complex real-life plant emissions. We also detected differences in oxidation products produced from the oxidation of healthy and stressed Scots pine emissions: the oxidation products formed from the emissions of stressed

trees were more highly oxidized and displayed a higher volatility than oxidation products from the emissions of healthy trees.

As a summary, this thesis improves the accuracy of volatility measurements of FIGAERO- CIMS and introduces new tools to gain more information about these novel volatility measurements. It also produces novel information of the formation and volatility of SOA particles produced from different VOC sources. Findings emerging from this thesis reveal that the whole emission profile of the biogenic VOCs from vegetation, rather than just single species, should be considered if one wishes to estimate accurately the physicochemical properties of SOA systems.

Keywords: aerosol, organics, volatility, FIGAERO-CIMS, SOA

Contents

1 Introduction ... 11

2 Methods ... 17

2.1 ToF-CIMS ... 17

2.1.1 Used ionization schemes ... 19

2.1.2 FIGAERO inlet and volatility calibration ... 20

2.1.3 Determining volatility distributions from CIMS data ... 23

2.2 SOA production in an Oxidative Flow Reactor and the Melania chamber ... 24

2.3 Residence Time Chamber ... 25

2.4 Other used instruments ... 26

2.5 Positive Matrix Factorization and its application to FIGAERO-CIMS measurements ... 27

3 Results ... 30

3.1 FIGAERO-CIMS particle phase calibration for extracting volatility information 30 3.2 Using PMF in the FIGAERO measurements ... 33

3.3 Factors affecting SOA formation and volatility ... 35

3.3.1 Effect of oxidative conditions ... 35

3.3.2 Effect of precursor VOC structure on formed SOA ... 38

3.3.3 SOA formed from oxidation of real-life plant emissions ... 40

4 Review of papers and the author’s contribution ... 44

5 Conclusions ... 45

6 Outlook and future aspects ... 49

7 References ... 50

List of publications

This thesis consists of an introductory review, followed by five research articles. In the introductory part, the papers are cited according to their Roman numerals. Papers I-IV are reproduced under the Creative Commons Attribution 4.0 License and Paper V is reprinted with permission from the publisher.

I Ylisirniö A., Barreira L., Pullinen I., Buchholz A., Jayne J., Krechmer J. E., Worsnop D. R., Virtanen A. and Schobesberger S.: On the calibration of FIGAERO- ToF-CIMS: importance and impact of calibrant delivery for the particle-phase calibration, Atmospheric Measurement Techniques, 14, 355-367, 2021. URL:

https://amt.copernicus.org/articles/14/355/2021/amt-14-355-2021.html

II Buchholz A., Ylisirniö A., Huang W., Mohr C., Canagaratna M., Worsnop D. R., Schobesberger S. and Virtanen A., Deconvolution of FIGAERO-CIMS thermal desorption profiles using positive matrix factorisation to identify chemical and physical processes during particle evaporation, Atmospheric Chemistry and Physics, 20, 7693-7716, 2020. URL: https://acp.copernicus.org/articles/20/7693/2020/acp- 20-7693-2020.html

III Buchholz A., Lambe A., Ylisirniö A., Li Z., Tikkanen O-P., Faiola C., Kari E., Hao L., Luoma O., Huang W., Mohr W., Worsnop D. R., Nizkorodov S. A., Yli-Juuti T., Schobesberger S. and Virtanen A., Insights into the O : C- dependent mechanisms controlling the evaporation of α-pinene secondary organic aerosol particles, Atmospheric Chemistry and Physics, 19, 4061-4073, 2019. URL:

https://acp.copernicus.org/articles/19/4061/2019/

IV Ylisirniö A., Buchholz A., Mohr C., Li Z., Barreira L., Lambe A., Faiola C., Kari E., Yli-Juuti T., Nizkorodov S. A., Worsnop D. R., Virtanen A., and Schobesberger S.: Composition and volatility of secondary organic aerosol (SOA) formed from oxidation of real tree emissions compared to simplified volatile organic compound (VOC) systems, Atmospheric Chemistry and Physics, 20, 5629-5644, 2020. URL:

https://acp.copernicus.org/articles/20/5629/2020/

V Faiola C. L, Pullinen I., Buchholz A., Khalaj F., Ylisirniö A., Kari E., Miettinen P., Holopainen J. K., Kivimäenpää M., Schobesberger S., Yli-Juuti T. and Virtanen A., Secondary Organic Aerosol Formation from Healthy and Aphid-Stressed Scots Pine Emissions, ACS Earth and Space Chemistry, 3, 1756-1772, 2020. URL:

https://pubs.acs.org/doi/10.1021/acsearthspacechem.9b00118

List of abbreviations

ACN Acetonitrile

AMS Aerosol Mass Spectrometer

APi-TOF Atmospheric Pressure interface Time of Flight Mass Spectrometer BSQ Big Segmented Quadrupole

CCN Cloud Condensation Nuclei

CIMS Chemical Ionization Mass Spectrometer CNerr Constant Error

CPC Condensation Particle Counter DMA Differential Mobility Analyzer

ELVOC Extremely Low Volatile Organic Compounds FIGAERO Filter Inlet for Gases and AEROsols

TD-GC-MS Thermal Desorption Gas Chromatograph Mass Spectrometry HOM Highly Oxygenated Organic Molecules

IMR Ion Molecular Reaction

IVOC Intermediate Volatile Organic Compound LVOC Low-Volatile Organic Compound

MCP Microchannel plate O:C -ratio Oxygen-to-Carbon ratio OFR Oxidative Flow Reactor PEG Polyethylene glycol

PB Primary Beam

PLerr Poisson-like error

PMF Positive Matrix Factorization PTFE Polytetrafluoroethylene

PTR-MS Proton Transfer Reaction Mass Spectrometer

RH Relative humidity

RTC Residence Time Chamber

SEM Scanning Electron Microscope SMPS Scanning Mobility Particle Sizer SOA Secondary Organic Aerosol(s)

SSQ Small Segmented Quadrupole SVOC Semi-Volatile Organic Compound

ToF Time-of-Flight

VBS Volatility Basis Set

VOC Volatile Organic Compound

ULVOC Ultra-Low Volatile Organic Compound

11

1 Introduction

The effect of aerosol particles on the atmosphere has been studied from the 1920s, but their role and importance as a part of the atmosphere has been only recently appreciated (Boucher et al., 2013; Seinfeld and Pandis, 2016). One important driver for aerosol research has been the need to understand its contribution to the ongoing climate change. The direct and indirect aerosol and cloud effects on Earth’s radiation balance are still significant sources of uncertainty when trying to model and predict changes in the climate (Boucher et al., 2013;

Neelin, 2010).

By definition, an aerosol is a mixture of liquid and/or solid particles suspended in the surrounding gaseous medium; usually this medium is air which is mainly composed of oxygen and nitrogen molecules. The word aerosol also sometimes refers only to the particles whereas the term “aerosol particles” is used to emphasize the particle phase (Hinds, 1999).

Aerosol particles range from sizes of 1 [nm] up to 100 [μm]. The size range can be further divided into nucleation mode (1-10 [nm]), Aitken mode (10-100 [nm]), accumulation mode (100-1000 [nm]) and coarse mode (1-100 [μm]). Particles smaller than 1 [nm] are considered as molecules or molecular clusters, even though the distinction between cluster of molecules and small particles is somewhat hazy. The majority of the atmospheric particles number concentration reside in the accumulation mode, as particle removal mechanisms tend to remove nucleation, Aitken and coarse mode particles more effectively than those in the accumulation mode (Seinfeld and Pandis, 2016).

Atmospheric aerosols have different effects on the environment in which they reside.

Aerosols close to the surface impact on the biosphere e.g. exerting a variety of often negative, health effects which depend on both the composition and the number concentration of the particles (Seinfeld and Pandis, 2016; Shiraiwa et al., 2017). At higher altitudes in the troposphere and stratosphere, aerosol particles influence the climate both directly and indirectly (Mahowald et al., 2011). They also affect the atmospheric chemistry as they are involved in heterogeneous reactions, which are especially important for stratospheric ozone removal (Bednarz et al., 2016). The direct aerosol effect is due to the absorption and scattering of solar radiation, which can either heat or cool the climate depending on the properties of the aerosol. For example, black carbon particles released from almost all combustion processes absorb solar radiation very effectively (Bond et al., 2013). Another example is salt particles formed from sea sprays that effectively scatter and reflect light. The indirect effect (historically also known as a Twomey effect, Twomey, (1974)) is attributable to the ability of aerosol particles to act as cloud condensation nuclei (CCN). This kind of process of course happens also close to surface in the form of fog formation. Through CCN activity, atmospheric aerosols play a vital role in Earth’s water cycle as without CCNs, it would require as high a RH as 400% for water molecules to homogeneously nucleate into cloud droplets (Wyslouzil and Wölk, 2016). Thus, without atmospheric particles, the water cycle and possibly life itself on our planet would be very different.

12

As the CCNs form cloud droplets in the atmosphere, their number concentration influences the radiative and precipitative properties of the forming clouds. A high CCN concentration produces a more condensable surface for the water molecules and as such, the formation of smaller cloud droplets with a high number concentration. From a macro perspective, clouds with a high amount of small cloud droplets tend to be whiter (or have a higher albedo), than clouds with larger less numerous cloud droplets and hence they reflect more solar radiation back to space (Paasonen et al., 2013). Small cloud droplets also do not evolve into rain droplets as easily as larger droplets, which affects the amounts of precipitation and therefore the lifetime of the clouds.

Aerosol particles are often divided into primary and secondary particles based on how they have formed. As the name indicates, primary particles are released into the atmosphere as such. Sand dust, sea spray particles and combustion soot particles are examples of primary particles. Secondary particles are particles that are formed in the atmosphere from the condensation of inorganic or organic vapours. These vapours form new particles through nucleation, provided that their concentration is high enough and existing condensation sink is low, or condense onto existing particles. The division into primary and secondary particles is however somewhat vague, as atmospheric particles can go through several evaporation and condensation cycles, after which the differentiation becomes hazy (Donahue et al., 2009).

Although aerosol particles are key contributors to cloud droplet formation, all particles do not act as CCN under all conditions. The chemical and physical properties of the atmospheric particles have a significant influence on their ability to act as CCNs (Väisänen et al., 2016). Examples of these properties are hygroscopicity (ability to take up water), light absorption, particle size and volatility. Hygroscopicity directly effects on how easily a single particle can take up water and become activated into a cloud droplet (Seinfeld and Pandis, 2016; Tang et al., 2019). Light absorption properties are important during daytime as highly absorptive particles such as black carbon particles heat up easily, which effectively evaporates condensed water molecules (also known as semi-direct aerosol effect, Koch and Del Genio, (2010)). The size of the aerosol particles exerts a major influence on their CCN activity, in such a way that larger particles provide a more condensable surface, and are more prone to become activated into cloud droplets (Dusek et al., 2006). The volatility describes the tendency of the chemical compound in the aerosol particle and in the surrounding air to either condense onto or evaporate from the particle, affecting the particle size and hence their CCN activity (Hallquist et al., 2009).

One particular type of secondary particles that have attracted considerable interest during the past decades are called Secondary Organic Aerosols (SOA). These particles are formed from condensation of oxidized Volatile Organic Compounds (VOCs) released from biogenic and anthropogenic sources. These particles have major impacts on Earth’s climate as VOCs are being emitted into the atmosphere from vegetation. It has been estimated that in local conditions, SOA can comprise more than half of the total aerosol mass (Hallquist et al., 2009; Jimenez et al., 2009). Evaporation of biogenic VOCs from vegetation is strongest in the tropical regions and declines when moving to colder regions. However, for example

13

boreal forests are still major emitters of biogenic VOCs due to the vast land areas in which they grow, covering about 17 percent of the Earth’s land surface area (Juday, Encyclopedia Britannica)

VOCs are also emitted from anthropogenic sources, such as motor vehicles. While the SOA formed from anthropogenic sources can have a great impact on the local air quality and can be major source of air pollution in highly populated areas (Huang et al., 2014), globally the majority of SOA is of biogenic origins. In this thesis, I will focus on the properties and formation of SOA formed from biogenic VOC sources.

The transformation of VOCs into SOA particles is a complicated chemical process where the precursor VOC compounds undergo several oxidation reactions in the gas and particle phase (Bianchi et al., 2019). The physical and chemical properties of the SOA particles are dictated by both the structural complexity of the initial VOC compound and the oxidative conditions they experience (Glasius and Goldstein, 2016). Typically, the oxidation of precursor VOC compounds, such as terpenes, produces a wide variety of oxidation products which in turn can continue to be further oxidised. Many plants emit terpenes such as isoprene, monoterpenes and sesquiterpenes as part of their physiology. The key properties determining whether these oxidation products are adsorbed and stay in the particle phase is their saturation vapour pressure (Psat [Pa]) and effective saturation mass concentration (C*

[μg/m³]) (later called just saturation mass concentration) in the air (Donahue et al., 2011;

Pankow, 1994). C* is related to Psat through equation 𝐶^∗ =^{𝑃𝑠𝑎𝑡𝑀𝑤}

𝑅𝑇 , (1)

where Mw is the molecular mass of the compound, R is universal gas constant and T is temperature expressed in [K]. The relation of Psat to C* is again discussed in Sect. 2.1.3.

Briefly, the Psat expresses the partial pressure (and with C* the mass concentration) of what certain compound needs to have in the air, to allow it to condense onto an existing surface.

The Psat of the compound is determined by its molar mass, structure and functional group composition (Capouet and Müller, 2006; Pankow and Asher, 2008).

The number of organic compounds found in the atmosphere is staggering, with estimates ranging from thousands to tens of thousands of different compounds (Hallquist et al., 2009).

The C* values of these compounds range by over 30 orders of magnitude, from 10¹⁰ to 10^-

20 [μg/m³]. To ease the discussion of such a large range of compounds, the volatility range was divided into sections by Donahue et al., (2009). Originally the sections were termed in descending order as Volatile Organic Compounds (VOC, C* > 10⁶ [μg/m³]), Intermediate Volatile Organic Compounds (IVOC, C* 10³-10⁶ [μg/m³]), Semi Volatile Organic Compounds (SVOC, C* 1-100 [μg/m³]), Low Volatile Organic Compounds (LVOC, C* 10^-

3-0.1[μg/m³]) and Extra Low Volatile Organic Compounds (ELVOC, C* < 10^-3 [μg/m³]).

The most recent modification to the list has seen the subdivision of the lowest ELVOC range values into Ultra Low Volatile Organic Compounds (ULVOC, C* < 3 ⅹ 10^-9 [μg/m³]) (Schervish and Donahue, 2020). It should be noted that even though C* has units in [μg/m³], it only refers to mass concentration of gaseous molecules in the air and should not be

14

confused with the particulate mass concentration such as PM2.5 that is also expressed in [μg/m³].

In ambient conditions, VOC and IVOC class compounds are almost only found in the gas phase. SVOC class compounds can partition into both the gas phase and particle phase, depending on ambient conditions. LVOC class compounds are mostly found in the particle phase and ELVOC/ULVOC class compounds can be considered effectively non-volatile in ambient conditions.

A group of chemical compounds worth noting are called highly oxygenated organic molecules (HOMs), identified by Ehn et al., (2014); these have attracted substantial interest in recent years (Bianchi et al., 2019). These compounds are determined as organic molecules containing at least six oxygen molecules and fall within a wide range of volatilities from SVOC to ULVOCs. These compounds have been shown to effectively contribute to nucleation and the growth of new particles (Schervish and Donahue, 2020).

As incorporating volatility information of every single compound for example in a climate model would be too computationally expensive, the contribution of each compound to the aerosol can be summed into logarithmically spaced bins and presented as a so-called Volatility Basis Set (VBS, Donahue et al., (2009)). An example of the VBS distribution can be seen in Figure 1. In the figure, the y-axis shows the relative contribution or measured concentration of each bin.

Figure 1. VBS distribution determined from α-pinene photo-oxidation SOA. The data in the figure is further discussed in Sect. 3.3.

If the number of gaseous compounds is enormous, so also is the number of compounds in the particle phase, ranging from hundreds to thousands of compounds even in SOA formed from the oxidation of a single monoterpene (Paper IV). Very little information about these

15

compounds is available in chemistry databases, as many of them might not have been previously analysed, which was the case with HOMs that were discovered from ambient air by Ehn et al., (2014). Thermodynamic data in the literature does exist for some of compounds, but it is often available only for compounds in the upper part of the volatility range. Unfortunately, volatility measurements of lower volatility compounds have proven to be difficult and saturation vapor pressure measurements can vary by as much as 4 orders of magnitude for the same compounds (Bilde et al., 2015).

Nonetheless, as volatility still plays an important role in understanding the evolution of atmospheric aerosols, several different methods have been developed to estimate the volatility of organic molecules in the SOA particles. These methods encompass both theoretical approaches and direct experimental methods or a hybrid of both. An example of a purely theoretical method is calculating the SOA properties with quantum chemical modelling such as with COSMO-RS (Kurtén et al., 2018). Several other methods take advantage of the advances in modern mass spectrometric methods which allow an estimation of the volatility of the SOA based on chemical information of the SOA constituents by using semi-empirical parametrizations (Donahue et al., 2011; Li et al., 2016;

Simon et al., 2020). Some of these methods use the number of different atoms or ratios of atoms in the molecules (Peräkylä et al., 2020), or assume some certain structure for the compounds (e.g. SIMPOL by Pankow and Asher, (2008)).

Other methods employ measurements of particle size reduction over time and back-calculate the initial volatility (for example with Volatility Tandem Differential Mobility Analyzer, Burtscher et al., (2001)) or use more elaborate modelling in the determination of volatility (Yli-Juuti et al., 2017; Tikkanen et al., 2019). Similar modelling has also been used to determine the volatility distribution based on the growth of the particles while the composition of surrounding gas phase is known (Mohr et al., 2019). As particle growth and evaporation is strongly interlinked with the concentration of surrounding gaseous compounds, these kinds of evaporation-based methods can often produce only an estimation of effective C*, rather than true saturation concentration (C⁰) of the molecules. The C⁰ is defined as saturation concentration of vapor over a pure sub-cooled liquid (Donahue et al., 2011).

A group of more experimental methods relies on thermal desorption with subsequent mass spectral analysis of chemical compounds from the SOA particles. These are methods such as Thermal Denuder Chemical Ionization Mass Spectrometer, (TD-CIMS, Smith et al., (2004), Thermal Denuder Aerosol Mass Spectrometer (TD-AMS, Huffman et al., (2008)), Thermal Denuder Gas Chromatograph Mass Spectrometry (TD-GC-MS, Presto et al., (2012)), Micro-Orifice Volatilization Impactor Coupled to a Chemical Ionization Mass Spectrometer (MOVI-CIMS, Yatavelli and Thornton, (2010)), the Chemical Analysis of Aerosols Online – Proton Transfer Reaction Mass Spectrometer (CHARON – PTR-MS, Eichler et al., (2015) and Filter Inlet for Gases and AEROsols Chemical Ionization Mass Spectrometer (FIGAERO-CIMS Lopez-Hilfiker et al., (2014).

16

Especially, FIGAERO-CIMS has gained popularity in the past years (Thornton et al., 2020) and it was the main tool used in this thesis for acquiring volatility information of SOA particles. It enables a simultaneous measurement of both bulk concentration of chemical constituents in the aerosol particle, as well as their volatility. Even though several dozens of research articles have been published utilizing FIGAERO-CIMS data, only a handful of them have reported volatility information (Lopez-Hilfiker et al., 2014; Stark et al., 2017;

Bannan et al., 2019; Joo et al., 2019; Nah et al., 2019; Ye et al., 2019; Wang et al., 2020).

This is likely due to the fact that the volatility measurement of SOA with FIGAERO-CIMS relies on an accurate calibration of the instrument, and even a small shift in calibration parameters can lead to several orders of magnitude changes in the estimated saturation vapor pressures. The published calibration lines also differ from each other substantially. In this thesis, I will investigate the reason for this discrepancy and aim to improve the existing calibration methods.

As reliable volatility information is vital for understanding the evolution of SOA particles in the atmosphere, this thesis aims to provide tools to achieve accurate volatility measurements of particle phase SOA constituents. It also strives to increase the information on how the structure of the initial precursor VOC compound affects the composition and volatility of its oxidation products and further of the resulting SOA particles, and examines how the SOA from single precursor compounds contrast to the SOA formed from real-life plant VOC emissions. The main aims of this thesis are:

i. To improve and develop methods to gain more reliable volatility information from FIGAERO-CIMS measurements (Papers I and II)

ii. To clarify how different oxidative conditions and precursor VOCs affect the volatility of the formed SOA (Papers III and IV)

iii. To investigate if SOA from single VOC precursors is comparable to SOA from real- life plant emissions (Papers IV and V)

17

2 Methods

In this section I will introduce the methods used in this thesis, starting with introducing the measurement devices used and continuing with methodological aspects. A complete list of used instruments in this thesis is shown in Table 1.

2.1 ToF-CIMS

The Time-of-Flight Chemical Ionization Mass Spectrometer (Aerodyne Research Inc. and Tofwerk) has been used to study ambient gas phase and particle phase constituents for a decade (Bertram et al., 2011; Junninen et al., 2010; Yatavelli et al., 2012). The instrument consists of an Atmospheric Pressure interface Time of Flight Mass Spectrometer (APi-ToF) coupled with an Ion Molecule Reaction (IMR) chamber for chemical ionization. A schematic picture of the instrument is shown in Figure 2. The instrument has been depicted in detail elsewhere (Junninen et al., 2010), so only a brief description is given here. The sample molecules are typically introduced into the IMR region of the instrument through 2 [sL/min] pin hole, where it mixes with flow carrying the reagent ions. In a typical configuration, the reagent ion is first charged, in our case using a ²¹⁰Po α-radiation source.

The ionization methods used in thesis are further discussed in Sect. 2.1.1. The pressure of the IMR region is pumped to 100 mbar with a dedicated scroll vacuum pump. The ionized sample molecules then enter the Small Segmented Quadrupole (SSQ) segment of the instrument through another flow restricting nozzle. The pressure of the SSQ is held at 2 [mbar] with a second dedicated scroll vacuum pump. The front and back part of the SSQ can be held at different voltages, which means that charged ions can collide with any remaining neutral air molecules to cause collisional dissociation; for example, this is feasible when using the acetate ionization method, which will be discussed later. The sample ions then enter the Big Segmented Quadrupole region through a skimmer hole. The BSQ is used as a pressure drop region from high mbar range pressure to low pressures of the ToF region. The BSQ operates at ~1e-2 [mbar] and is used to collimate the sample ion beam.

Both SSQ and BSQ work as radio-frequency-only quadrupoles, which enables them to act as both ion guides and mass band filters. The frequency and amplitude settings of the SSQ and BSQ dictate the transmission function of the instrument, which is assumed to be in the form of a modified Gaussian function (Heinritzi et al., 2016). From BSQ, the sample ions travel through one more region called the Primary Beam (PB) which is a lens stack used to further focus the ion beam while pressure is reduced to ~1e-5 [mbar]. From the PB area, the ions finally enter the ToF region which is held at ~1e-6 [mbar] pressure. The BSQ, PB and ToF regions are all simultaneously pumped with a single segmented turbo-molecular pump.

Directly at the entrance of the ToF region, sample ions are sent on their flight path using a high voltage extractor. The ions travel on their flight path to the other side of the ToF chamber where a high voltage reflector bounces them back towards the detector. The ions finally hit a Microchannel Plate (MCP), which acts as the detector. The MCP is an array of

18

electron multipliers which amplifies the signal from a single ion to a value which can be detected and recorded with a data acquisition card.

Figure 2. Schematic picture of the CIMS and how ions travel inside it.

Recorded data is then presented as a mass spectrum, with the x-axis showing the mass-to- charge ratio of the measured ions [m/z] and y-axis as their measured signal (counts per second, [cps]). Depending on the ionization method, measured molecules can carry multiple charges. However, with the ionization methods used in this thesis, multicharging is highly improbable and easily distinguishable, and therefore all of the measured ions can be assumed to be singly charged. The measured molecular mass of an ion is expressed in Daltons [Da].

The settings and physical properties of the ToF-CIMS determine the measured mass range, sensitivity and mass resolving power of the instrument. The majority of molecular compounds observed in the ambient air fall in a range from 10 to 600 [Da], therefore a mass range of 1-1000 [Da] is often used. In some special cases, where larger molecules can be expected to form, the mass range can be extended up to several thousands of Daltons.

However, an increasing mass range decreases the sensitivity of the instrument. The sensitivity of the ToF-CIMS is related to its ability to measure chemical substances and is usually expressed as [cps/ppb] (parts-per-billion) or in best cases, even in the range of [cps/ppt] (parts-per-trillion). Apart from sensitivity, one of the most important properties of any mass spectrometer is its mass resolving power, expressed as a unitless number, which indicates the instrument’s ability to separate different massed compounds from each other.

Knowing the exact mass of the measured ions is important as this can be used to back- calculate the chemical formula of the ion. However, the ion has to be assumed to be formed from certain elements and no new information about the structure of the ion is gained. The mass resolving power of the ToF-CIMS used in this thesis was approx. 4000-5000. This

19

resolution is more than sufficient to resolve compounds from each other up to 300 [Da] or so, after which assigning chemical formula to the measured signal becomes uncertain.

The measured signal is transferred from units of cps for example to ppt by using predetermined sensitivity calibration factors of known compounds. However, these factors are often compound specific. As thousands of compounds are routinely measured from ambient air, acquiring calibration factors for all of them is next to impossible. Therefore, measured concentrations are often reported with considerable error bars, or with some mention that the measurements are semi-quantitative. All of the ToF-CIMS measurements shown in this thesis are reported as the measured signal or as ratios of measured signals.

The mass spectral analysis shown in this thesis was performed using computer software tofTools (Junninen et al., 2010) running on MATLAB (Mathworks Inc.) and Tofware (Aerodyne Research Inc.) running on Igor Pro 7 (Wavemetrics Inc.).

2.1.1 Used ionization schemes

In this thesis, the instrument was operated with two different chemical ionization methods, namely iodide ionization methods (Paper I-IV) and acetate ionization (Paper V).

In iodide ionization method (Lee et al., 2014; Iyer et al., 2016; Hyttinen et al., 2018), methyl iodide (CH3I) permeation tube is kept at regulated 1-2 [sL/min] N2 flow, which is fed through an α-radiation ²¹⁰Po source into the IMR. The α-radiation breaks down the methyl iodide molecule leading to the formation of negatively charged I^- anions. The analyte molecules M are ionized mainly through ligand exchange and adduct formation with compounds containing hydroxy, hydroperoxyl, carboxyl and peroxycarboxyl groups in their structures. The main reaction while water vapour is present in the sample flow is

I(H₂O)⁻+ M → H₂O + [M + I]⁻, (R1)

through ligand exchange. In the absence of water molecules, the reaction happens through adduct formation with the help of neutral third body collision

I + M + X → X + [M + I]⁻, (R2)

where X is for example O2 or N2 molecules, which absorb part of the collision energy. Some analytes, such as peroxyacids might dehydroxylate instead of clustering with I(H2O)^- as

R(O)OOH + I(H₂O)⁻ → HOI + H₂O + R(O)O⁻, (R3) and as such are seen without the iodide atom. Some compounds, such as nitric acid, are also often observed in their deprotonated form in the mass spectrum in minor amounts.

The acetate ionization method (Veres et al., 2008) works through proton abstraction. In certain instrument conditions for certain analytes, proton abstraction can also be seen in the iodide ionization scheme. In the acetate ionization method, a small (e.g. 50 [smL/min]) flow

20

of N2 is passed through a vial of acetate anhydride which can then be mixed with a 1.5 [sL/m] N2 flow that is led through the α-radiation ²¹⁰Po source before entering the IMR where it becomes mixed with the sample flow. In a typical reaction pathway, the acetate anhydride forms acetic acid anions, which will abstract a proton from sample molecules as

M + CH₃COO⁻ → [M − H]⁻+ CH₃COOH. (R4)

The sample molecules M are in this case mostly acids that have a higher gas phase acidity than acetic acid. However, acetic acid might also cluster with other compounds as well as acids, like levoglucosan (Aljawhary et al., 2013), and the analyte is then observed as [M+CH3COO]^-. However, this kind of adduct formation is generally inhibited with instrument settings in the acetate ionization scheme for easing up the analyte identification.

2.1.2 FIGAERO inlet and volatility calibration

Filter Inlet for Gases and AEROsols (FIGAERO) coupled with ToF-CIMS was the key instrument in my work and therefore I will describe it in more detail. The FIGAERO-inlet enables ToF-CIMS to measure both gas phase and particle phase chemical constituents of ambient aerosols. The inlet was initially introduced in Lopez-Hilfiker et al., (2014) and was later commercialized by Aerodyne Research Inc. The inlet has two ports, one of which is always closed while the other is open. A schematic picture of the Aerodyne FIGAERO inlet is shown in Figure 3. While the gas phase inlet is open, aerosol particles are collected into a polytetrafluoroethylene (PTFE) membrane filter (Zefluor, poresize 1 or 2 [μm], Pall Corp.). After the collection period is over, the tray of the inlet moves so that the particles’

phase pinhole becomes exposed and PTFE filter is situated on top of it while the gas phase pin hole is blocked. The collected particles are then evaporated into the instrument with ultrapure nitrogen flow that is gradually heated up to 200 [˚C]. The temperature of the nitrogen flow is then kept at this temperature for an additional time to evaporate any residual low volatile compounds. The heating times of the nitrogen flow can be set to suit the needs of the study. For example, in Paper I we used a heating ramp time of 15 minutes which corresponds to an 11.6 [K/min] heating rate.

We also manufactured a custom FIGAERO inlet with identical working principles as the commercial version, deviating from the commercial version only in the choice of the motor moving the tray, the N2 heat source and specific dimensions. This custom-made version enabled a more flexible customisation of the inlet to suit our laboratory. The custom inlet was partly used in Papers I and IV, and the commercial version was used in all papers except Paper V.

21

Figure 3. Visual description of the Aerodyne FIGAERO inlet assembly. Figure reprinted from Bannan et al., (2019), under CC BY 4.0 licence.

When signal of the ToF-CIMS is plotted against the temperature of the heated nitrogen flow, the resulting graph is called a thermogram. An example of a thermogram is shown in Figure 4a.

In an ideal case, the thermograms have an almost Gaussian shape with a clear ascending part, followed by a peak value and a descending part. The thermogram can be also thought as an evaporation graph, where the bulk amount of material has evaporated at the peak of the thermogram. The corresponding temperature value of the peak is called Tmax, which is specific to each compound and is determined by the compound’s volatility, which is expressed as the saturation vapour pressure Psat [Pa] at that temperature. In reality, the descending part often has a prolonged “tailing”, meaning that the shape of the thermogram is commonly asymmetric. This tailing is possibly caused by evaporation compounds that more tightly bound with the PTFE-filter or evaporate from the surfaces of the FIGAERO inlet or IMR. The tailing effect can be decreased with additional heating of the IMR.

The value of Psat is highly temperature sensitive, so that it increases with increasing temperature. This relationship is however well described with the Clausius-Clapeyron relationship. Therefore, Psat values obtained at different temperatures can be translated to match values at other temperatures. Quite often, a temperature of 298 K/25 [˚C] is used as a reference value.

The Tmax vs. Psat relationship can be utilized to construct a calibration line when compounds with known saturation vapour pressures are deposited onto the PTFE filter. Figure 4a shows measured Tmax values of polyethylene glycols (PEGs) with polymer lengths from 5 to 8.

PEGs are synthetic polymers used in many fields of industry, which have been shown to have very predictable Psat properties when the polymer length increases (Krieger et al., 2018). Figure 4b shows the log-linear relationship of Tmax with ln(Psat, 298K), where Psat, 298K

has been determined at 298 [K] (Krieger et al., 2018). When a linear fit is applied to the data as

22

𝑙𝑛(𝑃_{𝑠𝑎𝑡,298𝐾}) = 𝑎 𝑇_𝑚𝑎𝑥 + 𝑏. (2)

The fit coefficients a and b can be used to estimate the Psat, 298 K values of chemical constituents in aerosol particles. The Psat, 298 K values can then be further converted to saturation concentration C* [μg/m³] values. This conversion and the volatility determination expressed in C* values are further discussed in the next section. As all Psat and C* values shown in thesis are determined for 298 [K], the notation is omitted from all values shown later.

As all measured variables, both Tmax and Psat inherently contain uncertainty, and this needs to be addressed in the fitting process. The fit shown in the Figure 4b was made using bivariate least squares fit (York et al., 2004; Pitkänen et al., 2016), which takes into account the uncertainty in both fitting variables.

Figure 4. Panel a) shows thermograms of four PEG’s used for FIGAERO-CIMS volatility calibrations.

Dashed lines and text above show the Tmax value of each compound. Panel b) shows linear fit of acquired Tmax values and corresponding Psat values in logarithmic space.

The way in which the calibration compounds are delivered onto the FIGAERO-CIMS has a major effect on the acquired calibration coefficients. Paper I revealed how by applying the previously widely used micro syringe deposition method for calibrant delivery can easily cause errors of several orders of magnitude in the estimated saturation concentration results.

The paper also presents a new method for calibration that more accurately captures the evaporation behaviour of chemical constituents from aerosol particles utilizing the atomization of the calibration compounds in a solution. This is followed with the collection of the atomized particles onto the FIGAERO filter in a similar fashion as sample particles

23

would normally be collected onto the filter. An illustration of the two calibrant deposition methods is shown in Figure 5.

With both methods, all of the used calibration compounds were dissolved in acetonitrile (ACN), while varying the concentration of the solution. With the syringe method, the used solution concentrations were 0.003, 0.01 and 0.1 [g/L] per compound. The solution concentration of the atomized solution was ~ 0.5 [g/L] per compound, which then changed slowly over time as the solvent evaporated in the atomization process. The produced aerosol population was monitored with a Scanning Mobility Particle Sizer (SMPS) system, which also provided an estimation of the collected aerosol mass.

Figure 5. Illustration of different deposition methods, a) syringe deposition method, b) atomizer method with using polydisperse aerosols and c) atomizer method using monodisperse aerosols.

2.1.3 Determining volatility distributions from CIMS data

As discussed in Sect. 1, C* values can be derived from ToF-CIMS data either by using information from identified chemical compositions or, if the FIGAERO inlet was used, by utilizing measured Tmax values. In Paper V, we used the parametrization described by Li et al., (2016), to estimate saturation concentration values for measured gas phase constituents.

The parametrization estimates log10(C*) values from the numbers of oxygen, carbon, nitrate and sulphur atoms in the molecule as

𝑙𝑜𝑔₁₀(𝐶^∗) = (𝑛_𝐶⁰ − 𝑛_𝐶)𝑏_𝐶− 𝑛_𝑂𝑏_𝑂− 2 𝑛_𝐶𝑛_𝑂

𝑛_𝐶+ 𝑛_𝑂𝑏_𝐶𝑂– 𝑛_𝑁𝑏_𝑁– 𝑛_𝑆𝑏_𝑆, (3)

24

where nC is the number of carbon atoms, nO is the number of oxygen atoms, nN refers to the number of nitrogen atoms and ns is the number of sulphur atoms. The bx values are constants determining the contribution of each atom to the log10(C*) value, and bCO is the carbon–

oxygen nonideality. Since in Paper V we only studied organic molecules consisting of C, H and O, the nN, nS, bN and bS values can be neglected. The 𝑛_𝐶⁰ is the so-called reference value, which depends on the composition of the molecule of interest.

When CIMS is operated with the FIGAERO inlet, the C* values can be estimated from the measured Tmax values of each ion thermogram, as was done in Papers I and IV. The Psat is calculated using equation

𝑃_𝑠𝑎𝑡 = 𝑒𝑥𝑝(𝑎𝑇_𝑚𝑎𝑥+ 𝑏), (4)

where a and b are fit constants determined in Equation (2) from calibration experiments conducted with the same measurement settings. The Psat value can then be translated to C*

with the equation (1), which is again presented here for easier reading:

𝐶^∗ =^{𝑃𝑠𝑎𝑡𝑀𝑤}

𝑅𝑇 , (5)

2.2 SOA production in an Oxidative Flow Reactor and the Melania chamber

The SOA studied in Papers II, III and IV was produced with a commercial oxidative flow reactor (OFR) (Potential Aerosol Mass (PAM), Aerodyne Research Inc. Kang et al., 2007;

Lambe et al., 2011) without seed particles. The reactor works as a continuous flow system and can be used for both dark ozonolysis and photo-oxidation of VOCs to form SOA under different oxidative conditions. The precursor VOCs are fed into the carrier gas (typically air) which is led into the inlet of the reactor. The precursor VOCs are then oxidized, and the resulting oxidation products exit the OFR from the rear of the reactor. Ozone is mixed into the inlet flow just upstream of the entrance of the OFR. The OH-radicals are then produced from water molecules with UV-photolysis of ozone molecules, which produces excited oxygen [O(¹D)] atoms. The radical O(¹D) atoms then react with water vapor molecules, producing OH-radicals that react with the precursor VOCs. The precursor VOCs often also react with the externally formed ozone mixed into the inlet flow. The physicochemical properties of the resulting SOA depend on the oxidative conditions in the OFR, which can be adjusted by varying different parameters e.g., the residence time in the OFR, the amount of externally fed ozone and the intensity of the UV lighting.

In Paper V, the studied SOA particles were generated in a 9 [m³] Teflon environmental reaction chamber located in a temperature-controlled room in the University of Eastern Finland (Melania building, Kuopio, Finland, later called The Melania chamber). The Melania chamber was used as a batch mode chamber, so that known amounts of precursor VOCs, RH and oxidants were fed into the chamber at the start of each experiment and the

25

reaction products were sampled with a suite of instruments. No external dilution air was supplied. As the air is being sampled with the instruments, the chamber slowly collapses with the help of counterweights so that the pressure inside the chamber remains constant at the ambient pressure. The oxidants used were generated either with an external ozone generator or through the photolysis of nitrous acid (HONO) with UV-lights.

The main difference between the two introduced SOA production methods is the time scale in which SOA is produced and the mass concentration of the formed SOA. A typical SOA experiment in the Melania chamber in the single oxidative condition often takes about 8-12 hours while in the same time frame, the OFR can be used with several different oxidation conditions. The relatively small reaction volume of the OFR also enables it to produce a high mass concentration of SOA, typically in the order of 100s of [μg/m³], while the mass concentration of SOA in the Melania chamber is typically around 10 [μg/m³]due to its higher volume. Therefore, the adoption of OFR enables shorter sampling times that are more sensitive to particulate mass, such as FIGAERO and Aerosol Mass Spectrometer (AMS). It is also easier to further oxidize the precursor VOCs to higher oxidative states in the OFR than in the Melania chamber. However, it is not possible to monitor the oxidation steps in the OFR while they are more visible in the Melania chamber.

Certain precursor VOC’s ability to form SOA in the oxidation process can be described with SOA mass yield, which is defined as

𝑴𝒂𝒔𝒔 𝒚𝒊𝒆𝒍𝒅 = 𝑪_𝑶𝑨

∆𝑽𝑶𝑪, (6)

where COA [μg/m³] is the amount of condensed particulate mass and ΔVOC [μg/m³] is amount of consumed VOC’s. Mass yield is then often presented against the COA. The SOA yield can also be expressed with equation defined by Odum Jay et al., (1996):

𝑴𝒂𝒔𝒔 𝒚𝒊𝒆𝒍𝒅 = 𝑪_𝑶𝑨 ∑ 𝛂_𝒊𝑲_𝒊 𝟏 + 𝑲_𝒊𝑪_𝑶𝑨

𝒊

, (7)

where αi is proportionality constant relating the amount of reacted precursor VOC to the total concentration of species i and Ki is the partitioning coefficient for species i. Parameters αi and Ki are usually estimated with fit to Mass yield vs. COA measurements.

2.3 Residence Time Chamber

In Papers II and III, we investigated the particle evaporation characteristics to define the effective vapour pressure of SOA particles. For this purpose, we used a 100 [L] cylindrical stainless-steel Residence Time Chamber (RTC), which played a central role in these investigations. The stainless-steel walls of the RTC were assumed to act as a perfect sink for evaporating chemical compounds, effectively diluting the gas-phase around the

26

evaporating particles. The studied SOA particles were generated using OFR. The particle evaporation was investigated by size-selecting quasi-monodisperse particles with Differential Mobility Analyzer (DMA), after which the particles were fed into the RTC and their size change was monitored periodically with the SMPS system. The chemical composition of the particles was probed with AMS at the same time as SMPS was sampled.

The volatility and more complex chemical composition of the SOA were measured with FIGAERO-CIMS at the beginning of the evaporation experiment and finally at the end of the experiment.

2.4 Other used instruments

Other instruments used in the studies of this thesis were Aerosol Mass Spectrometer (AMS, Aerodyne Research Inc., DeCarlo et al., (2006)), Thermal Desorption Gas Chromatograph Mass Spectrometer (TD-GC-MS, TD - PerkinElmer, ATD 400, USA; GC-MS - Hewlett Packard, GC 6890, MSD 5973, USA), Scanning Mobility Particle Sizer (SMPS, TSI Inc.), Proton Transfer Reaction Mass Spectrometer (PTR-MS, Model PTR-TOF 8000, Ionicon Analytik) and ozone monitor (Thermo Fischer Scientific, model 49i). Even though they were important in understanding the studied SOA-systems, they did not play major roles in this thesis and are therefore only briefly discussed. A full list of used instruments in different studies can be seen in Table 1. In addition to the instruments mentioned above, we also took Scanning Electron Microscope (SEM, Sigma HD variable pressure field emission gun - SEM, Carl Zeiss NTS) pictures in Paper I, to visualize the differences between the two calibration methods.

The AMS was used to monitor the bulk composition of aerosol particles in Papers III-V.

The AMS enables an estimation of the bulk O:C ratio of measured aerosols and provides estimates of the amount of sulphate and nitrate containing compounds. However, AMS utilizes flash vaporizing of sampled particles at temperatures of ~600 [˚C] with a subsequent electron impact ionization, which effectively fragments the sample molecules. It is therefore not possible to conduct a direct chemical characterization of sampled molecules.

The TD-GC-MS was used to determine the composition and structure of VOCs used for SOA generations in Papers IV and V. The VOCs were sampled into cartridge sample collection tubes (Markes International Inc.) and trapped VOCs were desorbed into the retention column of the GC unit. The TD-GC-MS is typically used for measuring compounds of the VOC and IVOC volatility classes while compounds with lower vapour pressure are usually stuck to the retention time column. The TD-GC-MS also uses an electron ionization method, which produces a distinctive fragmentation pattern for each sampled compound. With the help of the retention time measurement and online fragmentation tables, TD-GC-MS can achieve an identification of isomers with the same composition, such as monoterpenes and sesquiterpenes. TD-GC-MS had key importance in Papers IV and V for understanding the volatility and composition differences between different SOA.

27

Particle size distribution measurements and particle mass concentration estimations were done with an SMPS system in all studies. The measured size-range was selected differently in each study, ranging from 10 [nm] to 500 [nm]. The SMPS system was used in all papers except Paper II.

In Papers III-IV, the precursor VOC concentrations used for SOA generation were measured with PTR-MS. The PTR-MS we used is sensitive to VOC and some IVOC and SVOC volatility class compounds. The PTR-MS however mostly used to monitor the precursor VOC concentrations used in the SOA production. Some first-generation oxidation products and other VOC compounds were also monitored in Paper V. The proton transfer reaction mass spectrometry can be considered as a soft-ionization method, where H3O+

molecules are used as the reagent ion. However, in the method, hard collisions are induced in the instrument to break apart clusters, which tend to fragment into larger molecules such as monoterpenes and sesquiterpenes. This process can be accounted for and PTR-MS measurements are often used for example in SOA mass yield calculations, which were also done in Papers IV and V.

Table 1. List of instruments used in the thesis.

Instrument Paper I Paper II Paper III Paper IV Paper V

ToF-CIMS Iodide FIGAERO

Iodide FIGAERO

Iodide FIGAERO

Iodide FIGAERO

Acetate gas phase

AMS No No Yes Yes Yes

SMPS Yes No Yes Yes Yes

PTR-MS No No Yes Yes Yes

TD-GC-MS No No No Yes Yes

O3 monitor No No Yes Yes Yes

2.5 Positive Matrix Factorization and its application to FIGAERO- CIMS measurements

Paper II utilizes Positive Matrix Factorization (PMF, Paatero and Tapper, 1994; Ulbrich et al., 2009) to extract volatility information from FIGAERO-CIMS thermogram data. PMF can be used to reduce the complexity of mass spectrometer data, which often consists of several hundred compounds and their time-series. PMF is a bilinear statistical model that can be represented as

28

𝑿 = 𝑮𝑭 + 𝑬, (8)

where solutions are constrained into non-zero and non-negative space. When applied to mass spectral data, the X is a m ⅹ n sized matrix containing the measured observations, G is a m x p sized matrix that contains the factor time series as columns, F is p x n sized matrix that contains the factor mass spectra as rows and E is m x n sized matrix that accounts for residuals between the observation matrix and fitted values. The PMF does not need any a priori information for G and F or for the number of factors (p), instead the user must decide which number of factors best represents the data.

For acquiring final output for the PMF, equation (8) is solved iteratively in least squares sense by minimizing an object function

𝑄 = ∑ ∑ (𝐸_𝑖𝑗 𝑆_𝑖𝑗)

𝑛 𝑖 = 1 𝑚

𝑗=1

2

(9) where Sij, is the uncertainty (or error) of each observed data point. This minimization process is continued until values of Q converge to some close-to stable value.

In an application of PMF, the user must choose the number of factors and can further aid the convergence of the solution by utilizing so-called “seed” values, which are pseudorandom starting points for the iteration. The use of seed-values might be beneficial when the solution space contains several local minima. The seed values might help the algorithm to converge to additional solutions. Additional solutions can also be found by using so-called “rotations”, which enable a further exploration of the solution parameter space. However, a more detailed discussion about effects of seed and rotations to PMF results is beyond the scope of this thesis. Further information can be found from Ulbrich et al., (2009) and references therein.

The values of S in Equation (9) act as weighting values for the observation matrix, so that observation points with higher error values are weighted less and vice versa. Therefore, correctly choosing the error matrix plays a critical role for obtaining reliable information from PMF. Typically, when applied to long time-series data, the values of S follow the actual data. For example in Yan et al., (2016), Sij was defined as a Poisson-type distribution:

𝑺_𝑖𝑗 = 𝛼 ⋅ √𝐗_𝑖𝑗

𝑡_𝑠 + σ_{𝑛𝑜𝑖𝑠𝑒,𝑖},

(10)

where Xij, is the signal intensity of ion i, ts is the sampling or averaging interval of the data, σnoise,i is the electrical noise of the ion i and α is experimental constant. In Chen et al., (2020) the error was determined simply via Poisson-counting statistics:

𝑺_𝑖𝑗 = √𝑿𝒊𝒋. (11)

29

These type of error modulations (later termed Poisson-like errors, PLerror) work well with long datasets where rapid changes in the data often indicate outliers or instrument malfunctioning. However, this kind of weighting is problematic when applied to quickly changing FIGAERO-CIMS thermogram data as shown in Figure 4a. The PLerror gives too much weight to the stable start and end parts of the thermograms and less weight to the actual peak region of the data, which contains most of the volatility information. We therefore introduced a new error scheme where individual Sij values are constant (later constant noise, CNerror) as

𝑺_𝑖𝑗 = σ_{𝑛𝑜𝑖𝑠𝑒,𝑖}. (12)

In both Equations (10) and (12), σnoise,i was determined by first fitting a line to the stable part of each thermogram, usually at the end of the thermogram and calculating the residual (resij) between the data and the line fit. The error was then determined as

σ_{𝑛𝑜𝑖𝑠𝑒,𝑖} = {𝑚𝑒𝑑𝑖𝑎𝑛(𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠)) | 𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠_𝑖𝑗) ≤ 𝑚𝑒𝑑𝑖𝑎𝑛(𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠))

𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠_𝑖𝑗) | 𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠_𝑖𝑗) > 𝑚𝑒𝑑𝑖𝑎𝑛(𝑠𝑡𝑑𝑒𝑣(𝑟𝑒𝑠)). (13) In other words, the value of σnoise,i was at least the median value of all stdev(res) values.

This method was chosen since converging the solution to a stable value would have taken an excessively long time, if the values of Sij were too small.

For the correct interpretation of PMF results, it is critical to choose the correct solution from all possible solutions with different number of factors and seed values. Normally, a certain indication about the correct solution can be found by investigating the ratio Q/Qexp, where Qexp is equal to the degree of freedom of the model solution. In the ideal case, this ratio should approach unity. However, when using the CNerror-scheme, the Q/Qexp-ratio is often larger as some part of the signal is bound to be left unexplained without overfitting the data with too many factors. Therefore, we calculated the fraction of the explained absolute variance (Ratioexp) for determining the “best solution” as:

𝑎𝑏𝑠𝑉𝑎𝑟_total= ∑ |𝑋_{𝑖𝑗 𝑖𝑗}− 𝑋̅ |_𝑖 , (14)

𝑎𝑏𝑠𝑉𝑎𝑟_explained = ∑ |R_{𝑖𝑗 𝑖𝑗}− 𝑋̅ |_𝑖 , (15) and

𝑅𝑎𝑡𝑖𝑜_𝑒𝑥𝑝 =absVar_explained

absVar_total , (16)

where Rij is the value of each ion in the reconstructed data matrix (R = GF), 𝑋̅_𝑖is the average of the measured ion value, absVartotal and absVarexplained are the total and explained variance.

It should be noted that even with help of Ratioexp in choosing the criteria, the ultimate choice of “best” solution is still left to the end user and therefore must be considered as somewhat subjective.