REPORT SERIES IN AEROSOL SCIENCEN:o 139 (2012)

NATURAL AND ANTHROPOGENIC INFLUENCES ONTROPOSPHERIC AEROSOL VARIABILITY

ARI ASMI

Division of Atmospheric SciencesDepartment of Physics

Faculty of ScienceUniversity of Helsinki

Helsinki, Finland

Academic dissertationTo be presented, with the permission of the Faculty of Science

of the University of Helsinki, for public criticism in auditorium E204,Gustaf Hallstromin katu 2, on December 14th, 2012, at 12 o’clock noon.

Helsinki 2012

Author’s Address: Department of PhysicsP.O. Box 64FI-00014 University of Helsinkie-mail ari.asmi@iki.fi

Supervisors: Academy Professor Markku Kulmala, Ph.D.Department of PhysicsUniversity of Helsinki

Professor Veli-Matti Kerminen, Ph.D.Department of PhysicsUniversity of Helsinki

Reviewers: Associate Professor Douglas Nilsson, Ph.D.Department of Applied Environmental ScienceStockholm University, Sweden

Adjunct Professor Mikhail Sofiev, Ph.D.Air quality researchFinnish Meteorological Institute

Opponent: Senior Scientist Andreas Maßling, Ph.D.Department of Environmental ScienceFaculty of Science and TechnologyAarhus University, Denmark

ISBN 978-952-5822-70-0 (printed version)ISSN 0784-3496Helsinki 2012Unigrafia Oy

ISBN 978-952-5822-71-7 (pdf version)http://ethesis.helsinki.fi

Helsinki 2012Helsingin yliopiston verkkojulkaisut

Acknowledgements

I have worked with aerosols in one way or another since 1998. During this time, so manypeople have contributed in my work that it is completely impossible to thank everyone.I will thus only concentrate my thanks on some more recent collaborators, and eventhere just the most recently influential ones. The foremost person in this list is Acad.Prof. Markku Kulmala, who originally hired me in 1998 as a research assistant, and hasbeen more or less directly my boss and/or supervisor ever since. His has had a majorinfluence on my scientific career, from first studies with indoor aerosol modelling to projectmanagement during EUCAARI. He has also taught me not to give up easily1, but alsoto cut the losses if needed2. My second supervisor and a good friend Prof. Veli-MattiKerminen has been of extremely important help throughout my studies and scientificwork. He has also corrected several issues with my bench press technique, which I willforever be grateful. I wish naturally to thank my pre-examiners, Adj. Prof. MikhailSofiev and Prof. Douglas Nilsson for their insightful and perceptive comments on thisthesis.

For the scientific input besides my supervisors, the largest influence has been from myconstant collaborator/co-worker/room mate Dr. Risto Makkonen, who did the boringpart of the climate modelling, leaving the fun parts to me. He also had to listen myopinions on every subject related or unrelated to science. From my 60 co-authors in thisthesis, I would like to especially mention Prof. Alfred Wiedensohler (IfT), Dr. PaoloLaj (CNRS), Prof. Ari Laaksonen (FMI), Dr. John Ogren (NOAA), and Dr. MartineCollaud Coen (MeteoSwiss), for their significant contributions in this work done.

I wish to thank all the staff in the Division of Atmospheric Sciences3. There aretoo many colleagues and friends to name individually, but perhaps I should mention the”old gang“ of the old building cellar: Hanna, Ismo, Miikka, Tuukka, Pasi and Hannele.

I also want to acknowledge the funding agencies and institutions I have worked in: Majand Tor Nessling foundation, Academy of Finland, Universite Blaise Pascal – Clermont-Ferrand, Ford Motor Company, the European Commission and of course the Departmentof Physics in University of Helsinki. Their contributions has been very helpful.

There can not be many couples who argue in sauna about aerosol absorption prop-erties. By quantity (and in some cases – quality), the highest amount of constructivecriticism for this work has been from my lovely and loving wife, (Dr.) Eija Asmi. I donot think that without her I would have ever finished this work, at least in this form.

1 Which is why this thesis has two columns.2 Which is why you, the reader, are holding this book.3 Although when I started there, it was Laboratory of Aerosol and Environmental Physics, located in

the lowest cellar in the old institute building.

Ari Juhani AsmiUniversity of Helsinki, 2012

AbstractAerosol particles are everywhere in the atmosphere. They are a key factor in many important

processes in the atmosphere, including cloud formation, scattering of incoming solar radiation andair chemistry. They have also been connected to adverse effects on human health and they have astrong effect on visibility. The aerosol particles have relatively short lifetimes in lower atmosphere,typically from days to weeks, and thus they have a high spatial and temporal variability. Thisthesis concentrates on the extent and reasons of sub-micron aerosol particle variability in thelower atmosphere, using both global atmospheric models and analysis of observational data.

Aerosol number size distributions in the lower atmosphere are affected strongly by the newparticle formation from gaseous precursors, mostly organic vapours and sulphuric acid. Perhapsmore importantly, a strong influence new particle formation is also evident in the cloud conden-sation nuclei (CCN) concentrations, suggesting a major role of the new particle formation in theclimate system.

In this thesis, the sub-micron aerosol number size distributions in the European regionalbackground air were characterized for the first time from consistent, homogenized and compara-ble datasets. The European background air is highly dominated by anthropogenic influences inCentral Europe. In remote regions, such as Northern Europe, strong seasonal changes are consis-tent with a larger role of the biogenic sources suggested by earlier studies. The characterizationwork of European aerosols provides air quality and climate modellers unparalleled possibilities inmodel performance testing, and creates a basis for any regulatory efforts on sub-micron aerosolnumber concentrations.

Some recent studies have suggested that differences in aerosol emissions between weekdayscould also affect the weather via aerosol-cloud interactions. These earlier studies of this “weekend-effect” were based on aerosol mass based proxies for CCN. In this thesis, the weekday-to-weekendvariation of CCN sized aerosol number concentrations in Europe were found to be much smallerthan expected from earlier studies. This suggests that any potential large scale “weekend-effect”of European meteorology is not directly influenced by CCN-sized aerosol particles, and some otherexplanation must be proposed. A key finding was also that aerosol mass or optical properties-based measurements are poor proxies for CCN concentrations in time scales comparable toaerosol particle lifetimes in the atmosphere. This result also suggests that a lack of weekdayvariability in meteorology is not necessarily a sign of weak aerosol-cloud interactions.

An analysis of statistically significant trends in past decades of measured aerosol numberconcentrations from Europe, North America, Pacific islands and Antarctica generally show de-creases in concentrations. The analysis of these changes show that a potential explanation for thedecreasing trends is the general reduction of anthropogenic emissions, especially SO2, althougha combination of several drivers for these changes in the number concentrations are likely.

The representative emission pathways developed for the IPCC prognose radical reductionsof anthropogenic emissions in the next decades, especially of sulphur dioxide. This will mostlikely cause strong reduction in the present-day cooling effect of the atmospheric aerosols. Themodel simulations of this thesis show that effect will cause strong additional positive forcing onthe atmosphere, possibly causing further increase in the near-surface mean temperatures. Theeffect was further magnified when new particle formation in atmosphere was also considered inthe model calculations. Strong reductions in primary aerosol emissions and especially secondaryaerosol precursors should be thus considered with caution.

Keywords: Atmospheric aerosol, Emission reductions, Weekend effect, New particle forma-

tion

CONTENTS

1. General properties of atmospheric aerosols . . . . . . . . . . . . . . . . . . . . . 6

1.1 Aerosol phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Aerosol particle size and composition . . . . . . . . . . . . . . . . . . . . . 6

1.3 Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.4 Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.4.1 Climate impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.4.2 Health effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2. Small particles – global consequences . . . . . . . . . . . . . . . . . . . . . . . . 16

2.1 New particle formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.1.1 Life and death of a new particle . . . . . . . . . . . . . . . . . . . 17

2.2 Global climate model ECHAM5-HAM . . . . . . . . . . . . . . . . . . . . 17

2.3 Implications of the PI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3. Exploration of European aerosol . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.1 Measuring sub-micron atmospheric aerosols . . . . . . . . . . . . . . . . . 21

3.2 Collecting data and comparison parameters . . . . . . . . . . . . . . . . . 22

3.3 European aerosol distributions . . . . . . . . . . . . . . . . . . . . . . . . 23

3.4 The weekend effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.4.1 Statistical properties of aerosol time series . . . . . . . . . . . . . . 24

3.4.2 Wavelet analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.4.3 Lack of weekly waves or not? . . . . . . . . . . . . . . . . . . . . . 26

4. “Study the past if you would define the future” . . . . . . . . . . . . . . . . . . 29

4.1 Trends of the past . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.2 A view to the future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.2.1 Radiative forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

5. Repercussions and perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

References 42

INTRODUCTION

This thesis consists of an introductory review, followed by 5 research articles. The intro-ductory review consists of aims of the thesis, general properties of aerosols, introductionto and discussion on some of the aspects of the work described in the research articlesand references.

In the introductory part, the research articles are cited with roman numerals:

PI Makkonen, R., Asmi, A., Korhonen, H., Kokkola, H., Jarvenoja, S., Raisanen,P., Lehtinen, K. E. J., Laaksonen, A., Kerminen, V.-M., Jarvinen, H., Lohmann,U., Bennartz, R., Feichter, J., and Kulmala, M. (2009). Sensitivity of aerosol con-centrations and cloud properties to nucleation and secondary organic distributionin ECHAM5–HAM global circulation model. Atmospheric Chemistry and Physics,9(5):1747–1766.

PII Asmi, A., Wiedensohler, A., Laj, P., Fjaeraa, A.-M., Sellegri, K., Birmili, W.,Weingartner, E., Baltensperger, U., Zdimal, V., Zikova, N., Putaud, J.-P., Marinoni,A., Tunved, P., Hansson, H.-C., Fiebig, M., Kivekas, N., Lihavainen, H., Asmi,E., Ulevicius, V., Aalto, P. P., Swietlicki, E., Kristensson, A., Mihalopoulos, N.,Kalivitis, N., Kalapov, I., Kiss, G., de Leeuw, G., Henzing, B., Harrison, R. M.,Beddows, D., O’Dowd, C., Jennings, S. G., Flentje, H., Weinhold, K., Meinhardt,F., Ries, L., and Kulmala, M. (2011). Number size distributions and seasonalityof submicron particles in Europe 2008-2009. Atmospheric Chemistry and Physics,11(11):5505–5538.

PIII Asmi, A. (2012). Weakness of the weekend effect in aerosol number concentrations.Atmospheric Environment, 11(11):5505–5538.

PIV Asmi, A., Collaud Coen, M., Ogren, J. A., Andrews, E., Sheridan, P., Jefferson,A., Weingartner, E., Baltensperger, U., Bukowiecki, N., Lihavainen, H., Kivekas,N., Asmi, E., Aalto, P. P. , Kulmala, M., Wiedensohler, A., Birmili, W., Hamed, A., O’Dowd C. , Jennings, S.G., Weller, R., Flentje, H. , Fjaeraa, A. M., Fiebig, M.,Myhre, C. L., Hallar, A. G. , and Laj, P. (2012). Aerosol decadal trends (II): In-situaerosol particle number concentrations at GAW and ACTRIS stations. AtmosphericChemistry and Physics Discussions, 12, 20849-20899

PV Makkonen., R, Asmi, A., Kerminen, V.-M., Boy, M., Arneth, A., Hari, P., andKulmala, M (2012). Air pollution control and decreasing new particle formationlead to strong climate warming Atmospheric Chemistry and Physics, 12(3):1515-1524.

Aims

The original aims of this work were connected to a common project with the FinnishMeteorological Institute and the University of Kuopio4. In the mid-2000s it became clearthat the Finnish aerosol studies required a climate model to widen the perspective andget a hold on the complex feedbacks in the atmosphere, as models capable of includingdetailed aerosol microphysics started to become available. The Finnish aerosol and at-mospheric groups started a co-operation with Max Planck Institute of Meteorology inGermany, for further development of ECHAM-HAM aerosol-climate model. This workresulted in the first paper (PI), which had the aim to evaluate the sensitivity of CCNconcentrations to new particle formation in a global context.

In the work leading to this thesis, the results of PI directly lead to PII, when thequality of model comparison datasets became apparent (Fig. 1). Thus, a second aim wasto generate useful comparison datasets and metrics for large scale model sim-ulations. PII showed also a surprisingly weak weekday variation in CCN-sized aerosolparticles. This result, together with my at-time interest in more advanced time-seriesanalysis, led to a study with the main aim to characterize this weekly variation inmore detail and to provide theoretical background for the visibility of oscilla-tions over background noise.

The PV were a result of PI, in the sense that we needed to know at least the pre-industrial conditions on our model version to get any idea of the anthropogenic aerosolinfluence on the climate. This was nicely collaborated by IPCC WG1 asking for a trendanalysis on the past decades aerosol measurements. As I had some experience on aerosolnumber concentration data analysis, and it well supported the PV aims, I decided tovolunteer as (co)investigator on the subject (PIV) with the aim to evaluate the changeson aerosol number concentrations in the last decades based on experimentaldata. The modelling studies of past, present and future had the aims to characterizethe anthropogenic influence on ECHAM5-HAM with nucleation mechanismsincluded and to study the changes in the climate forcing of aerosols from IPCCemission pathways.

4 nowadays, University of Eastern Finland

PI PII PIII PIV PV

NPF parameterizations

A need of better comparison

data

Radiative forcing?

External

Studies

New Questions

End results

Need of long termtrend information

RCPemissions

EUSAARdata

NPF sensititivityModel improvements

Long termdata

Analysis methods

Measurementdataset

Comparisonparameters

Weak CCNvariation

High aerosolforcing senstitivity

to emission reductions

Decreasingtrends

SO possible2

reason

N trendsin measurements?

Long term changes?

Weaknessof weekdayvariation?

ECHAM5-HAM

The role of NPFin global context?

Fig. 1: A simplified summary of connections between the studies of this thesis.

5

1. GENERAL PROPERTIES OF ATMOSPHERIC AEROSOLS

The terrestrial atmosphere is a mixture of nitrogen, oxygen and argon with small amountof trace compounds (Lide, 2004). A good whiskey is a mixture of water and ethyl alcoholwith a small amount trace compounds (Pryde et al., 2011). In both cases, even thoughboth have major and important active compounds (oxygen and alcohol respectively), theinteresting part1 are the minor constituents. The impurities and trace elements matter,and they are as critical to the behaviour of the atmosphere, as they are for the taste ofa good single malt. In atmosphere, the impurities include components such as carbondioxide, liquid water droplets (commonly observed as clouds) and, among many others,aerosol particles.

1.1 Aerosol phases

Aerosols, in contrast to other atmosphericimpurities, are a phase mixture. By def-inition, an aerosol consists of two parts:the gas phase (air and other gasses andvapours) and the particle phase (liquid,solid or multiphase particles). The gasphase fraction of the atmospheric aerosolcan be thought to consist of essentially in-ert carrier gas and a selection of potentiallycondensing vapours, precursor gases, oxi-dants and other reactive gaseous species.The particle phase consists of a relativelylow number of particles of varying size andcomposition in a constant interaction withthe gas phase. The aerosol can be af-fected by external forces, such as radia-tion, temperature gradients and other phys-ical processes, changing the environmentin the phase mixture. These changes canthen facilitate phase changes and chemi-cal processes, affecting the properties of theaerosol.

1.2 Aerosol particle size andcomposition

Individual aerosol particles are in atmo-spheric science usually characterized mostlyby their size and composition (Seinfeld andPandis, 2006). The particle size is usu-ally determined by the representative par-ticle diameter, dp, which can vary in atmo-spheric aerosol particles from around onenanometre to approximately hundred mi-crometres. Figures 2a-b show examples ofaerosol particle sizes in comparison to wave-lengths of electromagnetic radiation andsome biological entities. The particle pop-ulation in a macroscopic volume of air iscommonly referred as aerosol size distri-bution for some aerosol property as a func-tion of particle size. Figure 2b shows someexamples of the aerosol particle numbersize distribution function, with a custom-ary dN(dp)/d log10 dp normalization2. Sev-eral wide peaks are typically visible in the

1 Admittedly depending on the person.2 This normalization is far from the only possible, normalizations with natural logarithm

dX(dp)/d loge dp and linear size range dX(dp)/dp are also used in literature. The normalization is

Fig. 2: Composite picture of (a) comparable length scales of electomagnetic radiation wave-lengths and some biological entities, (b) typical atmospheric aerosol particle number andmass size distributions, (c) typical species in continental aerosol, (d) aerosol microphys-ical processes converted to characteristic time scales, and (e) processes important toaerosol climate (e1, e3) and health effects (e2) as a function of particle dry diameter. Miecalculations of scattering efficiency are courtesy of Mr. John Backman (U. Helsinki).

7

size distributions, indicating aerosol pop-ulations, or modes. These represent per-sistent aerosol populations are often foundin the atmosphere as a result of particlesources, sinks, and mixing and growth pro-cesses. Commonly, the atmospheric submi-cron (dp < 1µm) aerosol number size dis-tributions have an Aitken mode (diametersaround 30 to 100 nm) and an accumulationmode (from 100 nm to 1 µm) separatedby the so-called Hoppel gap (Hoppel et al.,1990). This gap is more prominent in re-mote locations, and for example the datashown for Melpitz, Germany, have very lit-tle indication of such modal difference in themedian aerosol number size distribution.This gap is most likely a result of cloud pro-cessing of aerosols, and is thus indicative ofaerosol size distribution history. Sometimesthe nucleation mode (below 30 nm) is visi-ble, as are parts of the mostly super microncoarse mode. As a comparison, the meanvolume size distributions, which are indica-tive of the aerosol mass distributions, aremuch more concentrated on the larger parti-cle sizes. This shift is a natural result of thed3p scaling between the two aerosol proper-ties, and shows that the majority of aerosolnumber concentration is concentrated in thesmaller half of the submicron range; themajority of aerosol mass is concentrated inthe over 100 nm diameter particles. Typ-ical European background of aerosol parti-cle number size distributions are discussedin length in PII.

By total mass, a considerable fractionof ambient sub-micron tropospheric atmo-spheric aerosol particles are composed ofwater (Wang et al., 2008; Ervens et al.,2011). The remaining mass of the sub-micron aerosol in the continental tropo-sphere consists of sulphates, nitrates, or-ganic compounds, ammonia and black car-

bon, with contributions from mineral dustand sea salt (Figure 2c and e.g. Putaudet al., 2004). The super-micron aerosolis usually more dominated by the min-eral dust and sea salt. The spatial varia-tion in the aerosol composition illustratesthe source areas of different types of par-ticles (Figure 3). The particle composi-tion is also dependent on the particle agein the atmosphere, as more and more con-densible matter, especially sulphates andorganics, condense on the particles or areproduced by cloud processing or heteroge-neous reactions. Freshly-emitted particlesform external mixtures with pre-existingparticles, where the aerosol particles of thesame size have different compositions. Age-ing processes, such as condensation andevaporation, reduce these differences untilthe population starts to be more uniform,with different compounds internally mixedin the particles. In the real atmosphere,the aerosol is a complex mixture of bothkinds of aerosol mixing types, although thetime-scales for change e.g. from freshlyemitted hydrophobic combustion particlesto more hygroscopic internal mixtures is of-ten rather short, of the order of hours (Er-vens et al., 2010; Riemer et al., 2010).

The particles in the atmosphere havea multitude of sources. Sulphates aremostly from anthropogenic emissions, vol-canoes, marine Dimethyl Sulphide (DMS)emissions, and from subsequent photochem-istry and cloud processing. Ammonium andnitrates are generated by biological pro-cesses and fertilizers, oceans can also bea significant source of ammonium; Blackcarbon (i.e. strongly absorbing aerosol)comes from combustion sources (includingbiomass burning); organics come from thevegetation, anthropogenic sources, biomassburning and oceans; and dust and sea salt

used, as it makes the size distributions measured with differing size resolution more comparable, and ifthe size axis is given in logarithmic scale, the area under curve normalised with logarithm of diameteris proportional to the total aerosol number concentration. The particle size dp inside the logarithm isassumed to be divided by the length unit used.

8

come from deserts and oceans, respectively.

The sources of atmospheric particle aregenerally divided into two categories: pri-mary and secondary. Primary sources gen-erate aerosol particles directly to air, e.g.in the case of atmospheric dust from re-suspension over deserts, organics and seasalt particles from evaporation of sea wa-ter droplets (bubble bursting) and spuming,or carbon aggregates from diesel combus-tion. Secondary particles are formed fromgas phase compounds in the atmospherevia new particle formation. In some cases,the division is more difficult as the divi-sion between the two kinds of particles isup to the scale which the emission is char-acterized: sulphate particles from combus-tion could be either considered direct emis-sions when measured far from the source,or secondary, if measured directly from ex-haust. The division between primary andsecondary aerosol is further complicated byterminological issues. In many studies theprimary/secondary split is done for aerosolmass, not by particles per se. In thesecases, the primary aerosol (mass) is directlyemitted and secondary consist of all masscondensed or otherwise produced to aerosolparticles after emission. This separation isespecially relevant for the discussion of Sec-ondary Organic and Secondary InorganicAerosol (SOA and SIA) in papers PI andPV.

Of particular note in aerosol particlecomposition are the two main species orspecies groups discussed in PI and PV,sulphuric acid (H2SO4) and organic com-pounds. The sulphuric acid is often con-sidered to be the key species in atmo-spheric new particle formation, due its gen-eral availability as a photo-oxidation prod-uct of SO2, simplified oxidation pathwayand the extremely low value of saturationvapour pressure. These properties makethe sulphuric acid very keen on either con-densing to existing surfaces (including par-

ticles), or if no such surface is available, totake part in new particle formation. A closecorrelation between H2SO4 and new parti-cle formation in troposphere is evident fromfield data (e.g. Weber et al., 1996; Kuanget al., 2008; Nieminen et al., 2009; Paaso-nen et al., 2010), also seen in chamber ex-periments (Sipila et al., 2010; Kirkby et al.,2011). In particles, sulphuric acid reacts toform sulphates.

The second particularly interestinggroup of species are the organic com-pounds. The importance of organic com-pounds come from the fact that they area major part of aerosol particle composi-tion in the submicron range (Jimenez et al.,2009), can affect the aerosol-cloud interac-tions and aerosol optical properties (Fac-chini et al., 1999; Ramanathan et al., 2005;Andreae and Gelencser, 2006; Prisle et al.,2012), are critical on aerosol growth tocloud condensation nuclei (Kerminen et al.,2000, 2012), and that measured concentra-tions are generally poorly reproduced inthe atmospheric models (Kanakidou et al.,2005). There are figuratively innumerabledifferent organic compounds in the ambi-ent atmospheric aerosol, and thus usuallysome lumping method is used to categorizethem. The current approaches in the liter-ature are based often on chemical charac-terization methods of particles (e.g. Dece-sari et al., 2000; McFiggans et al., 2010),their saturation vapour pressure (e.g. Don-ahue et al., 2006, 2012), oxidation state(Kroll et al., 2011), or semi-empirical smogchamber partitioning properties (e.g. Odumet al., 1997). Due the high complexity oforganic chemistry and lack of suitable data,the handling of organic compounds is stillvery uncertain in aerosol models. However,the current advances in the measurementsof aerosol chemistry provide a good basisfor improved parametrizations of processesconnected to at least the most relevant or-ganic compound classes.

9

Sulphate

Organics

Black Carbon

0.5

0.125

0.125

0.251.0

1.0

0.875

0.8750.5

Dust

Sea salt All the rest

1.0

1.0 1.0

0.5 0.5

0.5

0.5

0.5

0.5

Relative contribution of mean column mass burden

MODEL:ECHAM5.5-HAM2 5 year averages (1998-2002) (Run (K) in Prisle et al.)

(including Sea Salt and Dust) (Other species)

Fig. 3: Example of relative mean contribution of different aerosol composition species to columnmass burden (ECHAM5-HAM2 simulations analyzed for Prisle et al. (2012)). The leftfigure shows the contribution of mainly coarse mode species (sea salt and dust) in com-parison to all other species, and the right figure shows the relative contribution of the restof the simulated species. ECHAM5-HAM2 lacks nitrates and ammonia, and the columnburdens are given without water in the particles. Note that scale for black carbon isdifferent from the other species’ scales.

1.3 Processes

As the particle lifetimes vary from sec-onds to a few weeks, their variability inthe atmosphere is much greater than ofmany long-lived gaseous pollutants, lead-ing to spatially and temporally inhomo-geneous aerosol concentrations (Jaenicke,2008). Variations in available radiation,clouds and oxidants transform and age theaerosol populations in atmosphere (Seinfeldand Pandis, 2006). The atmospheric aerosolpopulations are thus in a constant state ofchange.

The huge range of particle sizes (alto-gether around 5 orders of magnitude in di-ameter) is also mirrored by the variabilityof strength of different microphysical pro-cesses affecting the particles. Figure 2dshows characteristic time scales of some ofthe processes in typical boundary layer con-ditions.

The smallest particles are very sensi-tive to removal by collision to larger aerosolparticles via Brownian coagulation, with

the characteristic time scale of

τK(dp) =1∫∞

dpK(dp, dpb)Nb(dpb)ddpb

,

(1.1)where K is the Brownian coagulation coef-ficient ( cm3s−1), and Nb is the backgroundaerosol number distribution ( cm−3). Thisprocess also grows the larger particles, al-though the effect is generally relatively mi-nor due the large volume differences be-tween the particles. As the particle size in-creases, the coagulation efficiency decreasesrapidly, shown as the increase of the char-acteristic time scales in Figure 2d.

Particles can be removed also by stick-ing to existing surfaces by dry deposi-tion, which is relatively efficient for par-ticles outside of Aitken and accumulationmode ranges, but only in atmospheric lay-ers near to the ground. For a well mixedboundary layer, the characteristic dry de-position time scale τd(dp) could be definedas

τd(dp) =H

vd(dp)(1.2)

where vd is the (particle and environment

10

dependent) dry deposition velocity ( ms−1)and H ( m) is the mixed layer height. Thedry deposition characteristic time shownin Figure 2d is for 1000 m mixing layerheight and based on the vd observationsin in Hyytiala; different environments anddifferent meteorological situations can sig-nificantly change the dry deposition rates,and even the shape of the particle sizedependence (Pryor et al., 2008). Aerosolparticles can also be removed from theatmosphere by falling raindrops or snow(below-cloud scavenging), by being ac-tivated to cloud droplets and precipitating(activation scavenging) or by collidingwith cloud droplets (in-cloud scaveng-ing). The time scales of these processesare even harder to determine, as they arestrongly related to cloudiness, precipitationfrequency and rain droplet sizes. The issueis even further complicated by the fact thatmost particles activated as cloud dropletsusually evaporate when leaving the cloudand do not precipitate. The below-cloudscavenging is only efficient for small or largeparticles ( dp < 100nm or dp > 1µm, Prup-pacher and Klett (1997)), and only rela-tively large particles ( dp > 50nm) can ac-tivate as cloud droplets (see next section oncloud condensation nuclei). A rough indi-cation of below cloud wet deposition timescales τw(dp) can be estimated from

τw(dp) =t0∑

p f(p) (1− exp(−Λ(p, dp)t0)),

(1.3)where p is accumulated precipitation (m)in collection time t0, f(p) is the fraction oftime with p binned precipitation accumu-lation and Λ is the below cloud scavengingefficiency. The time scales the Figure 2dare produced using t0 of 30 minutes, pre-cipitation accumulation f(p) from Hyytialaforestry station (Months V-IX, 2000-2005)and Λ from Laakso et al. (2003).

Particle interactions with the gas phaseare crucial to aerosol population dynam-ics. Oxidation of vapours and gases createlow-volatility compounds, which can con-dense on existing particles (condensation)and in some cases even form new particles(nucleation). The rates of condensationon particles are strongly dependent on theparticle size and on concentrations of con-densible vapours. In Figure 2d, the con-densation time-scales are shown for a zerovapour pressure model compound, with aconstant concentration. The concentra-tion levels are normalized by the growthrate (GR = ddp/dt) which such concen-tration would generate for a 1 nm particle( GR|1nm). In this case, the characteristictime scale3 would then be

τC(dp) =dp

GR(dp), (1.4)

and the used values of GR|1nm are compa-rable for observed growth rates during newparticle formation in Hyytiala (Yli-Juutiet al., 2011). In the real atmosphere, theconcentrations of condensible vapours arevery rarely constant, and thus the conden-sation growth will vary in time. The small-est particles (generally diameters less than10 nm) could also be affected by the Kelvineffect in respect to organic vapour con-densation, reducing the condensation rate(Kulmala et al., 2004d), although this pro-cess is not included in the Figure 2d. Thesemi-volatile species, such as some organiccompounds, can also evaporate from theparticles (Robinson et al., 2007). Chem-istry, environmental changes, particle in-ternal partitioning and meteorology canstrongly affect the particle/gas phase pro-cesses. For example, an important globalsource of accumulation mode mass is cloudprocessing, where activated cloud dropletsscavenge pollutants from air, which afteraqueous phase chemistry and cloud evap-

3 The definition of condensation time scale here is for relative change rate of diameter and is slightlydifferent from volume change rate used in e.g. Raes et al. (2000).

11

oration remain in the particles (Yin et al.,2002). This is a major source of sulphateand organic compounds in the CCN-sizedparticles and explain a great deal of particlegrowth from Aitken to accumulation modes(Kerminen and Wexler, 1995; Ervens et al.,2011).

The lifetimes of atmospheric aerosolparticles are generally smaller thanmany other climate-impacting impurities(Jaenicke, 2008). The lifetimes are, how-ever, very strongly dependent on the par-ticle size and environmental factors. Fastcoagulation of small (dp<20 nm) particleslead to short lifetimes due to their rapidcoagulation with larger particles. Largerparticles (in forest environments, dp>200nm) are removed efficiently via wet or drydeposition. The condensation is relativelyefficient up to Aitken mode particles. Forthe larger particles the cloud processingor heterogenous reactions are more effi-cient growth processes. The lifetimes ofthe particles are generally longer in greateraltitudes of the atmosphere, but still veryshort compared to e.g. methane or carbondioxide. This means that spatial and tem-poral variability of aerosol particles is muchhigher than many other climate impactingimpurities, and the changes in emissionsare rapidly seen in the overall concentra-tion levels.

1.4 Impacts

Atmospheric aerosols affect the atmospherein many ways. These impacts are the keymotivation behind studies of atmosphericaerosols. This thesis is mostly about theaerosol effects on weather and climate, and

to less extent, on human health. Theaerosol particle size dependency of some ofthese the impacts are summarized in Figure2e.

1.4.1 Climate impacts

Aerosols are affecting the climate system inmany important ways. The most obviousof these is the scattering and absorptionof incoming solar radiation before reach-ing the surface. These direct aerosol ef-fects can either cool or warm the atmo-sphere, and change the atmospheric circula-tion. These changes in the atmosphere canalso cause complex changes in the clouds(semi-direct effects, Koch and Del Ge-nio, 2010). The absorbing aerosol gener-ally warm the atmosphere by absorbing in-coming solar radiation, dependent on theamount of absorbing material in the parti-cles (Angstrom, 1929). The extinction ofsolar radiation from scattering to aerosolparticles is related to particle surface area,and refractive index-dependent scatteringefficiency Q. Figure 2e3 shows the scat-tering efficiency of pure water particles in550 nm radiation. Although the shape ofthe scattering efficiency function dependson particle properties, generally only par-ticles larger than approximately 100 nm di-ameter have a strong contribution to directlight scattering in the atmosphere.

In this thesis the main interest is in theindirect effects of aerosols through clouds.The two most studied indirect effects4 arepresented schematically in Figure 4. Thecloud-albedo (Twomey) effect, is based onthe elevated aerosol number concentrationsincreasing the number of cloud droplets inclouds, generating whiter (higher albedo)

4 There are a number of other proposed mechanisms of aerosol indirect effects, based on aerosol affect-ing the entrainment of dry air to the clouds and thus cloud water content or albedo: Drizzle-entrainmenteffect (Lu and Seinfeld, 2005), sedimentation-entrainment effect (Ackerman et al., 2004) and evaporation-entrainment effect (Wang et al., 2003). However, even though these processes have been studied in smallscale Large Eddy Simulations (Chen et al., 2011), I am not aware of any parametrizations for global scalemodels, and thus global impacts of these processes are still very uncertain.

12

clouds and thus reflecting more solar ra-diation (Twomey, 1977). The cloud life-time (Albrecht) effect is based on smallercloud droplets having less chance to gener-ate rain, increasing cloud lifetime and thuscloudiness (Albrecht, 1989). These effectsare mostly dependent on the ability of someaerosol particles to act as a seed for thecloud droplet formation, i.e. act as CloudCondensation Nuclei (CCN).

Updraft

Cloud Condensation Nuclei (CCN)

Activation

Coalescenceefficiency

Precipitation(drizzle)

more

less

less

Activationmore

CLEAN CASE(less aerosol)

POLLUTED CASE(moreaerosol)

Cloud albedo higherlower

Updraft

moreless

Cloud dropletsmore and smallerfewer and larger

Fig. 4: Simplified schematic of cloud albedo(Twomey) effect and cloud lifetime (Al-brecht) effects, as a change of cloudproperties between clean (left) and pol-luted (right) environments.

Cloud condensation nuclei

Cloud condensation nuclei are a commonlyused term for particles on which clouddroplets can be formed in cooling air parcelsin the atmosphere. Many of the aerosol-cloud interactions are related to the num-ber of available CCN, and thus much of theresearch in this thesis is one way or anotherconnected with determining some propertyrelated to CCN. The ability of an aerosolparticle to activate as a cloud droplet de-pends on the particle properties (size, hy-groscopicity, presence of surfactants) and onthe meteorological situation (updraft veloc-ity, water vapour, temperature). Of theseproperties, the particle size is commonlyconsidered to be of greater importance thancomposition (Dusek et al., 2006; McFigganset al., 2006), and the meteorological situa-tion is typically represented only by the crit-ical water supersaturation at the moment ofcloud activation. This approach was usedin the PIII, where direct measurements ofCCN.4% (i.e. aerosol particles activated atwater saturation of 1.004) were available.However, long-term measurements of CCNare not very common. The instruments formeasuring them have only recently becamestable enough for monitoring use (Neneset al., 2001; Sihto et al., 2011), and thusother methods to derive CCN concentrationneed to be used. In PII, PIII and PIV,the CCN concentration was approximatedfrom size distribution data, using a fixedcut-off diameter to produce different CCNproxies. This simplification is based on theassumption of relatively low importance ofparticle chemistry in the CCN activation.

In many modern studies, the aerosolcomposition effects are described by a sim-plified one-parameter approach. This ap-proach, developed by Petters and Kreiden-weis (2007), describes all composition ef-fects on particle hygroscopicity, and thus onthe cloud activation, with the so-called κ

13

parameter. Low values of κ are connectedwith a low hygroscopicity, i.e. particleswhich require higher water supersaturationsto activate as cloud droplets in comparisonto high κ particles. Figure 2e1 shows theeffect of this parameter on cloud activationpotential via showing the critical water su-persaturation needed to activate particlesas cloud droplets as a function of κ and dp.As the particles age in the atmosphere, theystart to have more and more uniform κ val-ues, close κ values of 0.3±0.1 for continentaland 0.7±0.2 for marine aerosol (shown asa white shaded areas in Fig. 2e1, Andreaeand Rosenfeld (2008)). This relatively smallvariation in ambient κ values support theuse of aerosol diameter -based CCN proxies,although the cloud properties and updraftvelocity still play a strong role in potentialCCN activity.

0 500 1000 1500 2000 2500

1

2

3

4 5

6

100

200

300

400

0

7

-3N (cm )a

-3C

DN

C (

cm)

1. North Atlantic (Gultepe et al., 1996)*2. Nova Scotia (ibid)*3. North Sea (Martin et al., 1994)*4. Continental (Gultepe and Isaac, 1999)5. Lin and Leaitch, 1996 (w=.2 m/s)6. Arabian sea (Ramanathan et al., 2001)* 7. Marine (Gultepe and Isaac, 1999)

Fig. 5: Published semi-empirical relationshipsbetween “CCN-sized” aerosol numberconcentrations (Na) and cloud dropletnumber concentrations. Different stud-ies used different size ranges for theirdefinition of CCN, commonly close to70-100 nm in diameter. *) Adaptedfrom Ramanathan et al. (2001a).

The relationship between below-cloudCCN and cloud droplet number concentra-tions are not linear, and several methodsfor approximating the cloud activation havebeen proposed. The semi-empirical ap-proaches, in contrast to mechanistic meth-ods, have the advantages of having largeenough scale of measurements to be com-parable with climate model grid boxes, be-ing easy to implement, and describing anactual measured cloud activation (Gultepeand Isaac, 1999). The generalization ofsuch parametrizations to different environ-ments can be difficult. Different clouds canhave different updraft velocities, entrain-ment rates and water contents, and theaerosol properties, beyond a simple num-ber concentration over selected activationdiameter, are not taken into account. Still,the different semi-empirical parametriza-tions give relatively similar behaviour, withdecreasing response to higher CCN con-centrations, approximately related to thelogarithm or fractional power of the CCNconcentration (Figure 5). Notably, theparametrization by Lin and Leaitch (1997),taking into account both CCN (definedas particle number concentration above 70nm in diameter, N70) and model estimatedcloud updraft velocity, was used in GlobalCirculation Model (GCM) simulations ofthis thesis (PI and PV). Some models canalso use mechanistic methods for the cloudactivation (Ghan et al., 2011, for a reviewof recent parametrizations). The mecha-nistic methods give possibility to includemany additional processes into the CCN-to-CDNC processing, making possible to studyeffects of e.g. changes in aerosol compo-sition or co-condensing species (Makkonenet al., 2012; Prisle et al., 2012).

14

1.4.2 Health effects

Atmospheric aerosols (or, in air-qualityterms particulate pollution) also affect hu-man health. A classical example of thedangerousness of anthropogenic particulatepollution was observed in extreme pollu-tion events in London 1952, where a strongphoto-chemical particulate pollution cloudcovered the city and caused excess mortalityof approximately 12,000 in the metropolitanarea(Brimblecombe, 1987; Bell et al., 2004).More recently (from 1970s on) studies haveclearly connected aerosol mass concentra-tions (particulate matter, PM5) to generalhealth of the population, especially to lungcancer and cardiovascular mortality (Dock-ery et al., 1993; Dockery and Pope, 1994),thus creating a strong epidemiological basisfor the air quality standards of particulatemass (Pope, 2000).

Nanoparticles ( dp < 100 nm) have beenwidely acknowledged to have the poten-tial for adverse health effects (Donaldsonet al., 1998, 2002; Sager and Castranova,2009). The particle deposition to alveo-lar region of lungs can be especially effi-cient for particles of diameters below 50 nm(Figure 2e2 and Oberdorster et al. (2005)).This is important, as such particles do nothave any measurable effect on particle massconcentrations (Fig. 2b), and are thusnot monitored in common air quality net-works, nor regulated by legislation. How-ever, there is no consensus on which aerosolproperty (e.g. size, composition, surfacearea) has strongest effect on human health(Wittmaack, 2007, and associated onlinecorrespondence). The different behaviourof PM and nano-particles in the atmosphereare discussed in PIII.

5 The normal measure of particulate matter is PMx concentrations, which are the particle mass belowsome pre-defined size limit. In health studies and air quality monitoring the most common measures arePM10 and PM2.5, i.e. particle mass below 10 or 2.5 micrometer aerodynamic diameter.

15

2. SMALL PARTICLES – GLOBAL CONSEQUENCES

Gas phase oxidation results in generation of very low volatility compounds, which can,in absence of suitable condensation surface, create new particles in the atmosphere. Thisprocess has been detected frequently all over the continental planetary boundary layer(Kulmala et al., 2004a). Unfortunately, no clear consensus which are the actual processesdominating this phenomena has yet emerged. Another issue is how significant such processcan be in the atmosphere. The only comprehensive way to approach this problem is touse a global model capable of simulating both aerosol processes and the behaviour of thesurrounding atmosphere.

2.1 New particle formation

One of the major sources of particle numberconcentrations is the gas-to-particle conver-sion, or nucleation. The process of new par-ticle formation (NPF) in the atmosphere iscomplex, and many studies have proposedmechanisms to explain this phenomena inthe atmosphere. Some proposed mecha-nisms for the initial atmospheric nucleationare binary H2SO4-H2O (e.g. Vehkamakiet al., 2002) or ternary H2SO4-H2O-[base](e.g. Napari et al., 2002; Bonn et al., 2008;Kurten et al., 2008) processes, sometimesincluding ion-induced effects (e.g. Yu andTurco, 2000; Lovejoy et al., 2004). Thereis no reason to assume that all atmo-spheric nucleation is from the same mech-anism, and some (or none) of these pro-cesses could be active in different environ-ments. One should distinguish the actualinitial formation of the particle (nucleation)from the subsequent growth/removal com-petition, resulting in observable particle for-mation1. In this sense, the inclusion of or-

ganic vapours can be critical for the newlynucleated particles (slightly above 1 nm insize) to survive to more detectable sizes,even though they might not actually nucle-ate themselves (e.g. Anttila and Kerminen,2003; Kulmala et al., 2004b).

Different semi-empirical approacheshave also been suggested for NPF usingeither laboratory approaches (e.g. Han-son and Lovejoy, 2006) or field observa-tions (e.g. Sihto et al., 2006; Paasonenet al., 2010, 2012). In PI and PV, sim-ple parametrizations of the NPF rate wereused in the form

Jn = C[H2SO4]k, (2.1)

where Jn is the nucleation rate ( cm−3s−1),C is an experimentally determined prefac-tor ( (cm3)k−1s−1), k is experimental fit-ting parameter and the H2SO4 concentra-tion is given in molecules per cubic centime-tre (Sihto et al., 2006). In case of k = 1,Equation 2.1 describes a process called ac-tivation nucleation, and with k = 2 it iscalled kinetic nucleation. Naturally, the pa-

1 This difference is nowadays getting smaller and smaller, as detection limit of instruments is alreadyat or close the diameters particles are forming (Vanhanen et al., 2011).

rameters C and k include a lot of infor-mation on background physics and chem-istry (e.g. concentrations of condensible or-ganic vapours), and as such the overall ap-plicability of one fitting to other locationsis not certain. When PI was written, nobetter information was available, and thestudy should be considered to be a sensitiv-ity study of these processes. In PV, thisnucleation method was only used over landareas. More recently, parametrizations in-cluding a factor depending on the amountof condensible organic vapours have beendeveloped (Paasonen et al., 2010, 2012).

2.1.1 Life and death of a new particle

The nucleation mechanism is just a partof the actual new particle formation event.Right after formation, the new particles areextremely small, and thus have high po-tential to coagulate with existing particles.The only way for them to grow to the rel-ative safety of Aitken mode sizes is eitherby condensational growth or by heteroge-neous processes, helped by e.g. salt for-mation or oligomerization (Riipinen et al.,2012). This competition between coagula-tion and growth processes is extremely im-portant for the observable new particle for-mation rates, and determines often the ac-tual total and CCN number concentrationwith at least as importantly as the actualnucleation rates and mechanisms (Kulmalaet al., 2004b; McMurry et al., 2005; Pierceand Adams, 2007). Figure 6 illustrates theimportance of condensation and coagula-tion processes, as the surviving fraction ofnewly nucleated particles is a strong func-tion of both background aerosol concentra-tion and the mean growth rate of the parti-cles.

2.2 Global climate modelECHAM5-HAM

Aerosol formation, and especially the in-teraction with aerosols and atmosphere,has many complex feedbacks. The neces-sary parameters of aerosol formation andgrowth are determined by the climate sys-tem, which via aerosol-climate effects, isalso affected by the changes in aerosol fieldsthemselves. This kind of complex feedbacksrequires a specific kind of model to study: aclimate model with an aerosol microphysicsmodule.

PI and PV used the ECHAM5-HAMGeneral Circulation Model (GCM) to studyaerosol-climate interactions. ECHAM5-HAM was one of the first models to includeall of the micro-physical aerosol processes(Chapter 1) in a consistent GCM framework(Stier et al., 2005). This is significant im-provement from most other GCMs, whichstill use either prescribed aerosol fields, orsimplified aerosol mass-based mechanisms(Textor et al., 2007). The need for de-tailed aerosol processes is especially relevantfor the studies of new particle formation,as NPF is strongly influenced by changesin the aerosol size distribution and relatedchemistry.

The base ECHAM5 model is a spectralglobal general circulation model, i.e. theprimitive equations are solved in Fourierspace, which is then in each time-stepconverted back to Eulerian grid for other(including aerosol and cloud) processes(Roeckner et al., 2003). The model resolu-tion used in PI and PV was T42, whichcorrespond to approximately 2.8x2.8◦ inhorizontal resolution, and around 300 kmgrid spacing at the Equator. The verticallevels of the model are defined by 19 (PI) or31 (PV) terrain following hybrid-σ levels upto 10 hPa. As a general circulation model,it can generate a relatively realistic repre-

17

Fig. 6: “Surviving fraction”, i.e. how much of particles nucleated in 1.7 nm diameter are stillleft in larger diameters after growth and loss processes, in different idealised backgroundconditions. “High GR” describes situation with constant concentration of condensiblevapour, normalized to give 5 nm/h growth rate for 1 nm particles. “Low GR” samewith 1 nm/h growth rate for 1 nm particles. “High CS” corresponds to coagulation sinkcalculated from Melpitz and “Low CS” to Hyytiala median aerosol number distribution(see Figure 2b). The Kelvin effect is not included in the calculations. Red area shows theregion where the Kerminen and Kulmala (2002) parametrization (eq. 2.2) was used inPI and PV, blue are where the aerosol particles start to either act as CCN or efficientlyscatter solar radiation.

sentation of a climate system, including ad-vection patterns, radiation, clouds and pre-cipitation. What it can not do, however, isa realistic representation of weather. This isimportant when considering potential com-parison parameters from atmospheric mea-surements. discrepancy always remains.

The aerosol module HAM (Stier et al.,2005) is based on modal description of theaerosol size spectrum, with aerosol dynam-ics from M7 submodule (Vignati et al.,2004). For computational reasons, theaerosol mass and number size distributionsare described by a set of 7 log-normalaerosol modes (Figure 7). The differentmodes have limited set of properties andspecies, concentrating the species tracers inthe most likely modes where they could bepresent. One constrain is that the modeshave fixed standard deviations, and they are

limited in the size space, forcing the aerosolsize distribution to always have exactly 7modes in predetermined order. The M7approach also includes tracers for externalmixing, in the form of non-soluble modes inAitken and accumulation modes. Althoughcomputationally efficient2, and often quitegood representation of overall aerosol popu-lation, the modal approaches do suffer somedrawbacks, of which the most relevant toNPF studies, is the requirement of log-normality.

The nucleation processes create highconcentrations of particles in extremelysmall sizes below 3 nm diameter. The rapidnucleation and subsequent growth and co-agulation processes of sub-3nm particlesmake their modelling difficult using a singlelognormal mode, especially considering themodel chemistry time step of 30 minutes.

2 Efficient in comparison to some other size distribution descriptions. Including the aerosol tracers forHAM result in a major additional computational overhead in comparison to standard ECHAM GCM.This is one of the reasons why ECHAM-HAM is not used in long Earth System model simulations forIntergovernmental Panel for Climate Chance (IPCC) Fifth Assessment Report (AR5) long-term EarthSystem simulations.

18

10-9

10-8

10-7

10-6

10-5

0

Insoluble

Soluble

nucleation accumulation coarse

Condensation

Cloud processing

Coagulation

Aitken

SO4BC SS DUOCSO4BC SS DUOCSO4BCOCSO4(OC)

DUDUBCOC

(B)VOC

H SO2 4SO2

OCc

OH

σ =1.59gσ =1.59g σ =1.59g

σ =2.0g

d (m)p

SO4

Yield+Mixing

dN

/dlo

g(d

)10

p

aging

Fig. 7: Simplified description of M7 aerosol dynamics submodel of ECHAM5-HAM used in PIand PV (Vignati et al., 2004). The model describes the aerosol size distribution with7 lognormal models (4 soluble and 3 insoluble), and follows both number concentrationof the modes and mass concentrations of different species in the modes (OC=organiccarbon, SO4=sulphate, BC = black carbon, SS = sea salt, DU = mineral dust). Modeswhich grow (or shrink) towards model boundary will have part of their mass and num-ber transferred to neighbouring mode, thus keeping the modal structure relatively rigid.Particles in insoluble modes are transferred to soluble modes via coagulation and con-densation (“aging” in the figure). Organic tracer in nucleation mode, and possibility ofSOA forming via BVOC oxidation are PI additions to standard M7.

For this reason, the parametrization of Ker-minen and Kulmala (2002) was applied toestimate the surviving fraction from diame-ter where nucleation is assumed to happen(1 nm in PI and 1.7 nm in PV) to 3 nm:

J3 = Jn exp

[(1

3nm− 1

dpn

)CS′

GRγ

](2.2)

where J3 is the NPF rate at 3 nm diameter,Jn is the nucleation rate ( cm−3s−1), dpnis the diameter of nucleated particles ( nm),CS′ is the reduced condensation sink ( s−1),GR is the growth rate ( nm h−1) and γ isa proportionality factor. The growth ratecalculation included sulphuric acid concen-trations and in PV also part of availablegas phase organic vapours. As this processhappens during one time step in the model(i.e. newly formed particles are directly put

to 3 nm diameter), this parametrization willunder predict the time needed to grow theparticles to 3 nm size.

In the standard ECHAM5-HAM, all or-ganic aerosol was represented in the modelas primary organic aerosol (POM) withfixed mass ratio between the modes (Stieret al., 2005). This is in contrast to stud-ies suggesting that large fraction of organicmatter in the aerosol particles are origi-nally emitted in gas phase and, after ox-idation(s), condensed to existing particlesas secondary organic aerosol (SOA) (Tsi-garidis and Kanakidou, 2003; Lack et al.,2004; Fuzzi et al., 2006). Addition of SOAwas tested in PI using different oxida-tion yields of biogenic monoterpene emis-sions, mixing them in the boundary layer,

19

and condensing the resulting SOA to exist-ing particles dynamically during the samemodel time step. This approach assumesthat all of the oxidized monoterpenes willbe completely non-volatile, but has the ben-efits of both including the organic aerosolderived growth in small particles and beingcomputationally very cheap. In literature,many other approaches are used, with someusing thermodynamical equilibrium modelsfor organic condensation (O’Donnell et al.,2011) and some are based on modellingthe saturation vapour pressures over a widerange of values (Donahue et al., 2012; Tostand Pringle, 2012). The approach used inPI and PV is simple, but it produces rea-sonable growth rates for submicron parti-cles – a necessity to get realistic NPF rates.

The aerosol-cloud interactions in PIand PV are handled by activating theaerosol particles when the ECHAM5 ex-pects clouds to be formed. The model usesthe semi-empirical activation parameteriza-tion of Lin and Leaitch (1997). The methoduses model-derived updraft velocities as aninput parameter, and only uses the totalaerosol number concentration above 70 nmin diameter as the representation of CCN.The actual cloud processes, including clouddroplet derived albedo changes, dropletautoconversion and precipitation are han-dled by the two-moment cloud scheme ofLohmann et al. (2007).

2.3 Implications of the PI

PI showed that the aerosol number con-centrations are very sensitive to inclusionof nucleation. Even though the activa-tion nucleation is not necessarily the bestchoice in many conditions (over oceans, up-per atmosphere), it does, however, bring theaerosol number concentrations closer to theobserved values in continental conditions.This conclusion is consistent with studiesin the literature, done with other mod-els and other nucleation parametrizations(Spracklen et al., 2006; Merikanto et al.,2009; Fountoukis et al., 2012). However, therole of insufficiently characterized primaryaerosol number emissions remain a problemto determine the actual level of NPF to theCCN concentrations (Spracklen et al., 2010;Fountoukis et al., 2012). The sensitivity ofCCN concentrations to NPF makes the cli-mate system sensitive to changes in avail-able sulphuric acid. As a major source ofH2SO4 is anthropogenic SO2, changes inanthropogenic emissions could then signif-icantly change the aerosol climate forcing.This result was one of the motivations tostart work on the PV.

One of the key results was also the ad-hoc nature of the model/measurement com-parisons. Even though some comparisondata were available (CREATE database),there was little information or understand-ing what should be compared between themodels and measurements. This issue wasa the motivation for the work on PII andPIII.

20

3. EXPLORATION OF EUROPEAN AEROSOL

The sub-micron aerosol number concentrations have been measured since the 1890s, andnumber size distributions have been measured since the 1970s. However, different in-struments, measurement standards and ways of representing the data can cause largedifferences in the measured distributions and concentrations, making generalization ofaerosol properties from different measurements difficult. Another issue is to determinewhich properties of such distributions are important for different applications.

3.1 Measuring sub-micronatmospheric aerosols

The majority of submicron aerosol particlesare well below the wavelengths of visiblelight (Fig. 2a). Thus direct optical mea-surement of particle concentrations is notfeasible. Similarly, the mass of the sub-micron particles is very small, and espe-cially the smallest nucleation mode particleshave completely insignificant mass in com-parison to particles in diameters closer to 1µm (Fig. 2b). Direct measurements basedon either of these properties will not be veryuseful to describe the submicron aerosol.

The first quantitative measurements ofsubmicron (or to that matter, any) atmo-spheric aerosol particles were done in 1880sby J. Aitken, FRS, (Aitken, 1889b), mea-suring the aerosol number concentra-tions (N)1 by growing the nanopartiles tolarger sizes by condensation and countingthem visually (Aitken, 1889a). Later im-provements in the technology in the early

20th century made possible to do continuousmeasurements of aerosol number concentra-tions (Mohnen and Hidy, 2010), leading toCondensation Particle Counter (CPC) in-strumentation in the later half of the cen-tury (McMurry, 2000). The CPC instru-ments have their limitations. Dependingon instrument architecture and operatingfluid (commonly water or buthanol), theyhave different minimum measurable parti-cle sizes, also subject to the particle com-position. The major obvious limitation isthat CPC measures only the aerosol totalnumber concentration above some diame-ter – a bulk measure of the aerosol pop-ulation. No knowledge of particle numbersize distribution is directly gained, but inmany cases, no other long-term data of sub-micron aerosol concentrations are available(see PIV).

In PII, PIII and PIV, the instrumentsfor aerosol particle number size distribu-tion data were Differential Mobility ParticleSizer (DMPS) and Scanning Mobility Par-ticle Sizer (SMPS). They are basically the

1 Older texts prefer to use term Condensation Nuclei (CN) as the term for number concentrations mea-sured with a CPC. This term is slightly misleading, as the original intent was to claim that all particleswere counted in a CPC. Later studies showed that this was not the case in realistic CPCs (Liu and Kim,1977). Even though CN terminology is still seen in some papers, it has clearly fallen out of fashion since1980s.

same instrument, just driven with differ-ent instrument program. Both use a driedaerosol sample, get it to charge equilibriumusing ionizing radiation, use the particlemobility in an electrical field to separate dif-ferent sized particles and finally count themwith a CPC. The DMPS changes the elec-tric field voltage in steps to measure eachsize section sequentially, while the SMPSscans the measured size range with con-stantly varying voltage. Both instrumentsrequire periodic checks to run reliably. Thedifferences between individual instrumentmeasurements are minor between 20 and200 nm (Wiedensohler et al., 2012).

3.2 Collecting data andcomparison parameters

One persistent problem in using datasets,even from organized database such as EBAS(http://ebas.nilu.no), is the quality controlof time series. In most cases, the networkresponsible of the measurements have stan-dardized the file format, meta-data infor-mation and the necessary data checks be-fore submitting. However, analyses of thedata revealed considerable variation in fol-lowing these guidelines. Before using anyof the time series in this thesis, I person-ally went through all of the datasets andconsulted the data providers carefully foreach probable rupture in the datasets. Af-ter correcting the obvious ruptures, the im-proved datasets were then re-uploaded tothe database for the next user.

Just collecting the data together doesnot yet give great insights to the overallnature of atmospheric aerosol. One mustdo consistent choices how to start to re-duce the data to more usable set. In PIII started to look the problem from mod-elling point of view, especially consideringthe aerosol indirect effects. As mentionedbefore, the number concentration of CCN is

crucial for the aerosol indirect effects, andthus it was obvious that parameters relatedto CCN should be included in the compari-son. For this reason, three proxies for CCNsfrom size distribution datasets were calcu-lated, one as the total particle number from50 to 500 nm (N50), one from 100 to 500 nm(N100) and one for 250 to 500 nm (N250).

The next phase was then to se-lect usable comparison metrics for themodel/measurement comparison. Com-monly the comparison between measuredand modelled datasets have been done us-ing arithmetic averages (e.g. some of thecomparisons in PI). However, it is apparentthat most aerosol properties are not nor-mally distributed. The long tail towardshigh concentrations lead to a high sensi-tivity to the relatively rare outliers, mak-ing the comparison of arithmetic means be-tween smooth modelled aerosol concentra-tions and noisy measurement time series dif-ficult. In PII, the results suggested that theaerosol number concentrations are generallymore closely log-normally distributed. Eventhis distribution is not always a good rep-resentation, as the overall shape is some-times skewed, and thus the geometric meanis not always a good choice of comparisonparameter either. In the end, most of theresults in PII are given in percentile values,which do not assume any specific shape ofthe concentration distribution. This has theadditional advantage of giving useful hintsof the distribution shape and making themodel/measurement comparison more com-plete.

The scale differences between model re-sults and measurements also required someconsideration. Especially when comparinga GCM in free circulation mode with mea-surements, the model can not be expectedto produce directly comparable time serieswith the observations. Even GCMs nudgedto meteorological re-analysis datasets canhave difficulties in producing an exactly

22

comparable advection to measurements inshorter time scales. For these reasons, Ichose seasonal distributions as the concen-tration comparison parameters. For specialapplications, the calculated number concen-tration time-series are also available in thecomparison database.

3.3 European aerosol distributions

PII gives a quite comprehensive descriptionof findings of the European size distribu-tion datasets. However, it might be use-ful to compare the overall picture of num-ber concentration data with mass concen-trations. Figure 8 shows a simple com-parison of annual median N100 from PIIwith (arithmentic mean) PM2.5 concentra-tions from EMEP. The overall agreementbetween the levels of concentrations is rel-atively good, with clearest differences inEastern Mediterranean, where PM2.5 con-tains relatively large amount of dust, prob-ably mostly absent from N100. Another pos-sible discrepancy was in the Po valley inItaly. Overall, this agreement just confirmsthe long-term similarities between aerosolproperties, also shown by van Dingenenet al. (2004).

A surprising side result from the PIIwas the lack of weekly variation in the con-centrations. Although the test done wassomewhat questionable (see section 3.4.1),it raised some questions considering the cur-rent interpretation of aerosol weekday cy-cles in the literature.

-3N median concentration, PII, (cm )100

3P

M 2

.5 a

nn

ua

l me

an

(μ

g/m

)

>20001000-2000500-1000<500

Fig. 8: Comparison of low-land N100 con-centrations from PII with 2009 an-nual mean PM2.5 concentrations basedon EMAP/MSC-W model calculationsand EMEP observation data, adaptedfrom EMEP (2011) with permission.

3.4 The weekend effect

The weekend effect is based on observa-tions of some meteorological parametershaving an apparent difference between dif-ferent weekdays. The name comes frommost common assumption that the differ-ence is between the working days and theweekend. Some studies use the more gen-eral term weekday effect for any such dif-ference between the weekdays. One possi-ble explanation for such differences wouldbe the differences in anthropogenic emis-sions of primary aerosols or aerosol precur-sors in different days of the week. The im-portance of a weekday variation in meteo-rological parameters is apparent, as theseanthropogenic source variation would thenneed to be taken into account in weatherprediction. It would also be a strong evi-dence of aerosol-weather interactions, giveindication of the strength of anthropogenic

23

influence on clouds, and could make the 7-day variability in the atmosphere a usefultracer of anthropogenic influence.

Some studies found clear differences be-tween different weekdays temperatures orprecipitation, others found none. Many ofthese differences can be attributed to tech-nical errors (Daniel et al., 2012), but thediscussion the existence, spatial and tem-poral range, and reasons of the weekend ef-fect still continues in the literature, of whichSanchez-Lorenzo et al. (2012) gives a recentreview.

There is one thing that almost all ofthe studies however agree on: there isa weekend/weekday difference in aerosolconcentrations. However, the concentra-tions used are either aerosol optical proper-ties (in the form of aerosol optical depths)or, more commonly, aerosol mass concen-trations from air quality networks. Theaerosol mass concentrations are not neces-sarily a very good proxy for CCN concen-trations, which are the key parameter inmany aerosol indirect effects. On the fewcases, where aerosol number concentrationsor size distributions were used, the measure-ments are done very close to the emission re-gions in cities. Thus, a more directly cloud-related study of the aerosol variability inthese time scales was needed, using the re-gional background data from PII and othersources.

3.4.1 Statistical properties of aerosol timeseries

The aerosol number concentrations are notnecessarily well behaving time series in sta-tistical sense. They exhibit strong auto-correlation in time scales used in many

analyses and distributions of concentrationscan be very far from normal distribution(see PII). Many commonly used statisticaltests, such as t-test, explicitly assume nor-mal distribution and independent measure-ments. The problematic features are gen-eral problems in many of the geophysicaldatasets, and methods have been developedto take these additional complexities intoaccount.

In both PII and in PIII, the possibleproblem of non-Gaussian distributions wasavoided by using a non-parametric test (U-test, or the related generalization Kruskalland Wallis test). The papers also take thespecific features of aerosol number concen-tration time series into account by reducingthe high hourly autocorrelation using onlydaily mean values, i.e. each data point inan analysed weekday time series has oneweek time difference from the next datapoint. When I wrote the papers, I thoughtthat reducing the autocorrelation withineach weekdays’ concentrations is enough forstatistical testing. After all, this is themethodology used in majority of studies inthis field (e.g. Barmet et al., 2009).

A recent publication by Daniel et al.(2012), showed that all autocorrelation ef-fects are not removed by just daily aver-aging. Indeed, the statistical tests donein PIII are too conservative for the pvalue indicated, and are probably closerto tests done with p <0.012. Using theblock bootstrap methodology described byWilks (1997), a more realistic approxima-tion of the statistical test significance couldbe made. Using their methods, 39 of360 tests in PIII are statistically signifi-cant with p<0.05. If considering only sea-sonal tests (which are independent witheach other), the results show 28 tests out

2 In hindsight, this is is quite clear result. As the concentrations in the adjoining days are autocorre-lated, a large part of their actual difference is masked by the persistence, leading to underestimation of thesignificance of concentration change. This is just an example how important it is to be very careful withstatistical tests when there is significant autocorrelation in the datasets. Usually, some sort of bootstraptest is required, such as the one in the GLS/ARB trend analysis of PIV.

24

of 288 showing significant weekly varia-tion. This is larger fraction of stationsthan would be expected by the p valuechosen, and shows that there is some in-dication on weekday variation in Europeannumber concentrations. However, only onestation showed consistent weekday variety.Otherwise the statistically significant week-day variations were not any way concen-trated on specific size ranges, regions, orseasons. For the long term datasets, theMelpitz (MPZ) datasets showed statisti-cally significant (p<0.05) weekly cycle forthe whole year and autumn datasets for N50

and spring datasets for N100 and N250. Itshould be mentioned that previous studiesdone with the exactly same, too conserva-tive, test showed clear and strong weekdayvariability in PM concentrations – the over-all weekday signal in CCN is thus muchweaker than for PM.

3.4.2 Wavelet analysis

Dividing the time series to different week-day means and trying to extract statisti-cally significant differences is a rather lim-ited approach. The time window selectionfor such analyses is a natural limitation, andthe high seasonal and other long-wavelengthvariabilities could mask short term weekdaydifferences. Spectral methods are one wayto try to extract specific period variationsfrom the time series, and are consideredspecifically a powerful tool to study week-day variation (Bell and Rosenfeld, 2008).

Wavelet analysis is one method to sepa-rate the time series into a localized time-frequency spectra (Kumar and Foufoula-Georgiou, 1997). The time series is de-composed using a set of wavelets, with sen-sitivity to oscillations with different wave-lengths. The continuous wavelet transformof discrete time series x(t) of length N , with

sampling time of δt is defined as

W (t, s) =

N−1∑t′=t0

x(t′)Ψ(η(t′, t, s))dt′ (3.1)

where W is the wavelet transform, x(t) isthe timeseries value at time t and Ψ isthe normalized wavelet function (Torrenceand Compo, 1998). In PIII, the Morlet(6)wavelet was used,

Ψ(η) =

(δt

s

)1/2

π1/4e6iη−η2/2, (3.2)

where η is the normalized “time” parameter

η =(t′ − t)s

, (3.3)

and s is the time scale parameter (s). Fig-ure 9 gives an example of application of thecontinuous wavelet transform. The choiceof Morlet(6) wavelet is common one and itis one of the most used wavelets in dataanalysis, where orthogonality is not usuallyneeded, and it gives a good compromise be-tween frequency and time space resolution(Torrence and Compo, 1998). The com-plex nature of Morlet wavelets makes possi-ble to also get the wavelet transform phaseϕ(t, s), which is the modus(2π) of the actualphase difference between the time series andthe wavelet. The wavelet power spectrumP (t, s) is given by

P (t, s) = |W (t, s)|2. (3.4)

In wavelet analysis, the wavelets are thenscaled in frequencies by adjusting the scaleparameter s, and repeating the transformfor all the needed wavelengths (Figure 9d).The actual choice of scales in this kind ofanalysis is arbitrary, but scales below twotimes sampling period or larger than ap-proximately half of the time series lengthgive no further information.

The wavelet transform also contains thenoise in the original signal. Statistical

25

tests thus are also needed in wavelet anal-ysis to distinguish probable signals fromnoisy background. The statistical test partof PIII wavelet analysis is based on theapproach by Torrence and Compo (1998),which compares the observed wavelet pow-ers to similar powers from first order au-toregressive (AR(1)) noise:

x(t+ δt) = αx(t) + e, (3.5)

where x(t) is the noise timeseries, α is theautocorrelation coefficient and e is randomGaussian white noise. An estimate of thewavelet AR(1) power spectrum comes fromdiscrete Fourier spectrum of AR(1) noise(Gilman et al., 1963)

Pn =1− α2

1 + α2 − 2α cos (2πδt/λf ). (3.6)

The confidence intervals of such noise spec-trum are then created using the inverse χ2

2

distribution, which requires somewhat nor-mal distribution of the original data. Theused α values were from datasets whichwere de-seasoned by subtracting a 90 daymoving average. However, the trend wasnot removed from these signals, and the au-tocorrelation might have then been overesti-mated, leading to overestimation the signal-to-noise ratio in PIII. The analysis usedthus should show slightly too significant sig-nals. Note that the overall applicabilityof the analysis depends that the estimatedsignal can be modelled with a stationaryAR(1) noise. As no clear band of signalwas detected in the wavelet power spectra(Figure S1 in the supplementary materialof PIII), I am confident that the 7-day sig-nal was generally absent in the data seriesanalysed.

3.4.3 Lack of weekly waves or not?

The conclusions in PIII are somewhatchanged due the inconsistencies in the sta-tistical analysis. There are some signs of

CCN driven weekend effect in some partsof European background. The statisticallysignificant variations in some Central Eu-ropean stations are consistent with earlierfindings, but much weaker and rarer thangenerally accepted in the literature. Thevery same (too conservative) tests in ear-lier work by Barmet et al. (2009) did showhighly significant trends in PM, which is aclear indication that the weekly trends inCCN sized aerosol particles have much lessclear weekly signal than in aerosol mass pa-rameters. Thus the discussion part of PIIIis still valid.

The lack of variation in wavelet anal-ysis requires then some more discussion,due to inconsistency with statistical testsin the MPZ dataset. I re-analyzed the longdatasets of MPZ with wavelet analysis, butI could not detect a clear 7-day oscillationin the spectra. However, what is clearlyseen even in Figure 4F of PIII, is an oscil-lation with 14 day wavelength. This couldthen mean that, at least in MPZ, the oscil-lations could be more commonly bi-weeklythan weekly in nature.

Overall, the weekday signal is not soclearly seen in the CCN concentrations asit has been seen for the particle mass. Thevariability of particles dominating the CCNnumbers in weekly time scales is very dif-ferent from the variability of larger (orsmaller) particles. Thus, using either opti-cal, or mass based proxies for CCN is veryerror prone in these timescales. Generaliz-ing aerosol variation from property (or di-ameter range) to another should always bedone with some consideration.

Another issue altogether is the implicitassumption that only anthropogenic pro-cesses can create 7-day cycles. Kim et al.(2010) suggested that some natural atmo-spheric processes could also create such os-cillations, previously also hypothesised byForster and Solomon (2003). Thus differen-

26

tiating between natural and anthropogenicinfluences might not be as straightforwardas previously thought. This was the reasonwhy PIII had the phase analysis for theresults of the 7-day variation.

The aerosol mass concentrations doshow weekly variability in the literature,and recent studies have shown that the ef-fect of so-called giant CCN (in the particlediameters above 1 µm) can have a strong

effect on the rain formation in some clouds(Konwar et al., 2012). This could be an av-enue how the weekend-effect works, outsideof the measurement range of datasets usedin PIII. It might also be that due to theeffects of these giant CCN, the earlier stud-ies using PM2.5 could have co-incidentallyused a usable proxy for the actually effec-tive aerosol property for weekend effect, justnot the property they were after.

27

0 500 1000 1500 2000 2500 30000.5

1

1.5

2

2.5

3

3.5

Convolution (Wavelet transform)

Wavelet power for one λf

Wavelet power spectrum

Timeseries

0 500 1000 1500 2000 2500 3000

10-4

10-2

100

|W|2

π

-π

0

Wave

let p

hase

φLog

(P) (a

rb.)

10

-4

0

4

Time

Fourier

period

Fourier

period

Wavelet phase spectrum

a)

-3 -2 -1 0 1 2 3-1

-0.5

0

0.5

1

0(

)

(s)

Re(0)

Im(0)

Morlet (6) wavelet

b)

c)

d)

e)

Time

Time

Conce

ntr

ation (

arb

.)

Fig. 9: Schematic example of the wavelet analysis. The timeseries (a) is transformed in a con-tinuous wavelet transform (eq. 3.1) with the Morlet(6) wavelet. Resulting wavelet powerfor one scale (or Fourier wavelength λf ) is shown in (c), showing how strong the signalof this wavelength was in the sample in the neighborough of each measurement time.Repeating the wavelet transform over a range of wavelengths, a power spectrum (d) anda phase spectrum (e) can be obtained as a function of time. No considerations of datagaps or edge effects are presented.

28

4. “STUDY THE PAST IF YOU WOULD DEFINE THE FUTURE”

PI suggested that the aerosol number concentrations and CCN number concentrationsare very sensitive to new particle formation. The majority of new particle formationparametrizations and laboratory experiments are strongly controlled by the amount ofsulphur in the system, suggesting a large role of the sulphur also in the atmospheric newparticle formation. Thus the changes in available sulphur in the atmosphere should affectthe aerosol concentrations and the aerosol-climate interactions. The emissions of sulphurdioxide were (and, in some cases still are) the main reason for the acid rain, and thusfrom the 1980s and on, the emissions of SO2 have been radically cut in North Americaand in Europe. The title of this chapter is a quote from Confusius (551-479 BCE), wellcharacterizing the methodology which is used here: Is the decrease in SO2 emissionsevident in the past aerosol number concentrations? Are the future aerosol propertiesstrongly influenced by the further changes in anthropogenic emissions?

4.1 Trends of the past

The IPCC AR5 Working Group I askedthe Global Atmosphere Watch to producea study on aerosol in-situ measurementstrends. After some discussion, it was de-cided that me and Dr. Martine Col-laud Coen would do this study, splittingthe task into two: I analysed the aerosolnumber concentrations and number size dis-tribution datasets (PIV), and Dr. Col-laud Coen analysed the in-situ optical prop-erty datasets (Collaud Coen et al., 2012).The main motivation for these studies wasto find out if any trends are visible in thelong term datasets, provide trend informa-tion for decadal scale modelling efforts, andto give some indication of probable trenddrivers.

The methodologies which I adapted forthese studies (GLS/ARB and GLS/MBB)are explained in detail in PIV. Figure 10shows in a more graphical way the basic

idea behind the AR bootstrapping processused.

As it is not directly explained in PIV,it might be useful to explain shortly alsothe other method (Sen’s slope connectedto Mann-Kendall test) used in the trendfitting. The Sen’s slope estimator for thetrends is a non-parametric approach, whereeach possible pair of the measurements

[x(t(i)), x(t(j))] , j > i (4.1)

was fitted a slope, and the median of theseslopes was used as the trend estimator(Sen, 1968). In practice, this was done fordaily or two daily median values, depend-ing on data size. The significance of thesetrends were then estimated using a seasonalMann-Kendall (MK) test, which is a non-parametric technique based on rank (Hirschet al., 1982). In the Mann-Kendall trendtest, the correlation between the rank or-der of the observed values and their orderin time is considered. The most obvious

2000 2005 20101.5

2

2.5

3

3.5

Se(t)

Trend

AR(1)

Season

2000 2005 20101.5

2

2.5

3

3.5

2000 2005 20101.5

2

2.5

3

3.5

2000 2005 2010

-4

-3

-2

-1

0

1

2

3

4

α=0.46

AutoregressionNoiseTrend andseasonality

Random re-samplingwith replacement

2000 2005 2010-4

-3

-2

-1

0

1

2

3

4

Constructing a new realization of time seriesand getting a new bootstraprealization of a trend

7.29 7.3 7.31 7.32 7.33 7.34 7.35

x 105

1.5

2

2.5

3

3.5

After 1000repeats ofrandom resamplingand trend fitting

Actual Trend

Confidenceinterval

Generalized Least Squares Fitting

Timeseries

-0.035 -0.03 -0.025 -0.02 -0.015 -0.01 -0.005 0 0.0050

20

40

60

80

100

120

Bootstrap realizations ofRelative trend (1/yr)

a)

b)

c)

d)

e)

f)

5%5%

Fig. 10: More graphical version of the upper part of Figure 1 in PIVexplaining the ARB methodof trend confidence interval generation. a) Timeseries analyzed (Pallas, Finland), b)-c)GLS is used to separate the dataset to trend, season, autoregressive part (AR(1)) andnoise (Se(t)). d) Noise terms are re-sampled and using AR(1), trend and season ofthe original GLS fit, e) a new trend is fitted and saved. After 1000 repeats of d)-e)a bootrapped distribution of trend slopes is obtained and confidence interval can begenerated from 5th and 95th percentile points.

30

advantage of the MK test is that the datado not need to conform to any particulardistribution. However, as the MK test issensitive to autocorrelation in the dataset,the data had to be first pre-whitened usingthe method described by Wang and Swail(2001): (1) estimate the auto-correlation,(2) if the auto-correlation is higher than0.05, calculate the Sens slope, (3) removethe linear trend using the Sens slope, (4) re-move the auto-correlation (whitening), un-til α < 0.05 and (5) add the trend. Thepre-whitening routine is always a compro-mise between too strong removal (wherepart of the actual signal is lost) and toolittle (where some of the AR noise is leftin the dataset, possibly affecting the trendfitting). The test for either an upward ordownward trend was the two-tailed MK testat the 95% level of significance.

The trends in PIV show signs of de-creases in the number concentrations inmany locations. The lack of data does notyet allow us to conclude that this changehas been global. The analysis of trenddrivers in PIV is basic, but in my opin-ion sufficient for a paper this concentratedon the data analysis. The potential role ofSO2 can, however, direct future research.A comprehensive comparison with potentialtrend drivers should be done using GCMsimulations, more advanced methods capa-ble of taking into account several poten-tial explanatory processes into at the sametime, as well as capable in some extent toevaluate the importance of different feed-backs in the system. The results of thetrend analysis papers were recently pre-sented in a AEROCOM modelling work-shop, and there is a good chance that suchstudies will be done shortly.

The climate system is full of feedbacksrelated to aerosol particles: air temperatureis affected by the aerosol concentrationsvia aerosol-climate interactions; increase intemperature could affect the emissions of

dimethyl sulphide, creating more particlesand acting as a cooling effect (so calledCLAW hypothesis; Charlson et al. (1987));cooling feedbacks could be detected froman increase in terrestrial plant VOC emis-sions (Kulmala et al., 2004c; Arneth et al.,2009); changes in wind speeds could af-fect fluxes of sea salt and DMS (Carslawet al., 2010); increased sea surface tem-peratures could reduce the emissions fromsea spray (Martensson et al., 2003; Zaboriet al., 2012a,b); decreased sea ice could in-crease the marine sources of aerosols (Nils-son et al., 2001; Struthers et al., 2011);changes in precipitation could affect aerosolconcentrations (Iversen et al., 2010). Evenconsidering such potential feedbacks, I findthe rather strong similarity of SO2 trendswith the observed N trends, especially fromthe viewpoint of the good understanding ofrelationship between SO2 and aerosol par-ticle number concentrations, a good candi-date of number concentration trends in theUS and parts of EU.

4.2 A view to the future

PIV showed that number concentrationtrends have been decreasing, and there aresome evidence that this could also havebeen happening to CCN concentrations. Ifthe simplified analysis of PIV regarding therole of anthropogenic emissions is valid, thisraises some concern on future emission cuts.

The future anthropogenic emissionstrends are of much dependent on the poli-cies we adopt in the next decades. Asthere are a wide variety of potential choices,IPCC formed specific task force to gener-ate Representational Concentration Path-ways (RCPs). These pathways give the lat-est IPCC AR5 simulations a comparable setof alternative future concentrations. Unlikethe earlier SRES predictions (Nakicenovicet al., 2000), the RCPs did not start di-

31

rectly as emission scenarios, but instead areinternally consistent sets of projections ofthe components of radiative forcing that areused in subsequent phases of IPCC mod-elling work with concentrations as the pri-mary product. In total, a set of four harmo-nized pathways were produced that lead toradiative forcing levels of 8.5, 6, 4.5 and 2.6Wm−2 by the end of the century (van Vu-uren et al., 2011). Of these, the PV usedall except RCP6, which was not yet com-pletely characterized when the simulationswere done.

Figure 11 shows a collection of cur-rent estimates of past and future anthro-pogenic SO2 emissions. The future emis-sion estimates show a high variability inSRES emission inventories (grey area), theextreme scenarios of Cofala et al. (2007),and the current IPCC AR5 derived Repre-sentational Concentration Pathway (RCP)emission pathways. Notably, all RCPs leadto almost pre-industrial anthropogenic SO2

emissions by 2100.

The key issue with RCPs was that theyonly considered the concentration pathwaysof anthropogenic emissions, and thus thelevel of biogenic or other natural emissionsare left to be estimated by the climate mod-els. As the simulations in PV were rela-tively short (5 years) snapshots of effects ofdifferent period emissions, the long-periodchanges in e.g. ocean circulation or vegeta-tion changes were not taken into account.

RCP2.6RCP4.5RCP6RCP8.5

Estimatesof SO emissions2

in literature(Smith et al, 2004)

Estimates of future SO 2

emissions in literature(van Vuuren et al, 2011)

Smith et al., 2011

Glo

bal

anth

. SO

em

issi

ons

(Mt

S /

a)2

10

30

50

70

1850 1900 1950 2000 2050 2100

CLE

MFRCofala et al, 2007

Year

Fig. 11: Evolution of anthropogenic SO2 emis-sions. CLE and MFR correspond to“current legistlation” and “maximumfeasible reductions” by Cofala et al.(2007), used in Kloster et al. (2008,2010).

To simulate some of the possible long-term natural feedbacks, the PV includedvariation of both oceanic Dimethyl Sulphide(DMS) and biogenic VOC emissions on thescale presented in other literature. An-other important thing to notice is that thePV simulations were done for the currentday climate, i.e. by keeping the sea sur-face temperatures in current climate condi-tions and thus the overall global mean tem-perature was not allowed to drift strongly.This was necessary, as otherwise the separa-tion of aerosol-induced changes to the radia-tive forcing could have been challenging tobe distinguished from temperature drivenchanges in the atmosphere. Fixing the seasurface temperatures also suppresses tem-perature driven changes in the the sea saltemissions (e.g. Martensson et al., 2003).

4.2.1 Radiative forcing

The sensitivity of the global climate to dif-ferent changes in the driving factors is nottrivial to determine. The most obvious

32

choice would be to run the model aftereach change as long as it is needed for theclimatological temperature to reach a newstable regime. However, this stabilizationtakes quite a long time in the climate mod-els, not to mention the additional complex-ity of including a coupled complete ocean-atmosphere model to take the ocean heatcapacity and mixing into account. Typi-cal time scales for such changes are of theorder of tens of years (Hansen et al., 1981;Brasseur and Roeckner, 2005; Kloster et al.,2010), which are, considering the overalltime scales of aerosol processes and atmo-spheric variability, quite expensive compu-tationally.

There are, however, other methods toestimate the effects of aerosol concentrationperturbations on the climate system. Forthe direct aerosol effects, a change in themodel estimate of Aerosol Optical Depth(AOD) is sufficient to approximate the ef-fects. For indirect effects, no such direct pa-rameter is available, and the climate systemresponse is approached from the concept ofradiative forcing (RF), specifically radiativeflux perturbation derived radiative forcing.

In short, the radiative forcing is a termdescribing the immediate change in the at-

mospheric radiative balance from a pertur-bation, and can be connected to temper-ature by use of climate sensitivities. IPCCdefinition of RF is “the change of net (downminus up) irradiance (solar plus longwave;in Wm−2) at the tropopause after allowingfor stratospheric temperatures to re-adjustto radiative equilibrium, but with surfaceand tropospheric temperatures and stateheld fixed at the unperturbed values” (Ra-maswamy et al., 2001). This definition ofRF has been useful in comparing the im-mediate relative radiative effect of differ-ent consistuents of climate change (typicallycompared to pre-industrial levels), but ithas severe limitations for aerosol-cloud in-teractions, where the relatively fast, but notinstantaneous, changes in e.g. cloud life-time could have a strong effect (Lohmannet al., 2010). For this reason, the radiativeforcings in the PV were calculated directlyfrom radiative flux perturbations (RFPs)(i.e. changes in the mean incoming andout-coming radiation fluxes) in the simula-tions while keeping the sea surface tempera-tures at current climate values. This resultsin values comparable with traditional RF,but still allowing for finite time response ofthe aerosol-cloud interactions to play a role(Lohmann et al., 2010).

33

5. REPERCUSSIONS AND PERSPECTIVES

The tasks of the Finnish universities (research, teaching, etc.) are supposed to be donewith the viewpoint of promoting societal impacts (Yliopistolaki “University law” 2§).From this standpoint, it is necessary to not only consider scientific conclusions, but alsoto think what are the possibly wider consequences of results of this thesis.

Small particles are not good for your health(Section 1.4.2), and some actions towardslowering the particle number concentrationshave already been implemented. As an ex-ample, recent European Commission direc-tives for vehicle emissions include a limit toparticle number emission rates for particleslarger than 23 nm in diameter (EC, 2008).If these efforts limiting the aerosol particlenumber concentrations lead to air qualitydirectives, one of the first things to evaluatewould be the current background concentra-tions. The PII results give the first consis-tent analysis of the outside-of-cities concen-trations and variability in Europe.

The results in PII and to lesser ex-tent other papers of this thesis show thehigh importance of biogenic processes tothese concentrations. This complex in-teraction between anthropogenic and bio-genic emissions might lead to serious diffi-culties determining the most cost-effectivestrategies for number concentration reduc-tions. The relatively long lifetimes oflarge fraction of aerosol number concentra-tion, as evidenced in PIII, lead to obvi-ous difficulties determining the source ar-eas of any increased concentrations. Over-all, the reduction strategies would probablyget very challenging to formulate for partic-ulate number concentrations.

The PII also provided a dataset andmethodologies to do comprehensive com-parisons between the models and measure-ments. This dataset has already shownto be practical in some modelling studies(Knote et al., 2011; Reddington et al., 2011;Bergman et al., 2012; Mann et al., 2012;Fountoukis et al., 2012), and at least a fewother groups have already contacted me onthe details of using the dataset. Overall,this work in PII seem to have been usefulfor the community.

The existence or non-existence of theweekend (or weekday) effect might not be soclear indicator of aerosol-cloud interactionsas previously considered. The results ofPIII show that the CCN concentrations aremuch less sensitive to the weekday variationin the emissions of aerosol particles or par-ticle precursors. The main reasons for thislack of periodicity are different measure-ment locations, processes affecting concen-trations and especially, different lifetimes ofCCN sized particles in comparison to PM2.5

or PM10. The lack of weekday periodicityin CCN will also suggest that if there is noweekly variation in the meteorological prop-erties, it is not necessarily an indication ofa weakness of aerosol-cloud interactions.

Long-term time series of N show gener-ally decreasing trends (PIV). No clear sin-

gle reason for the decreasing trends couldbe found, but the overall trend of SO2 emis-sions do show some promise as an explana-tory driver in Northern Europe and Amer-ica. This shows that reduction in aerosolprecursors could, at least in polluted re-gions, reduce the number concentrationssignificantly. The lack of long-term datamake conclusions on the changes in CCNstill uncertain from measurement point ofview. However, the results of PV, Klosteret al. (2008, 2010) and Fountoukis et al.(2012) strongly suggest significant sensi-tivity of CCN concentrations to emissionchanges. How important this CCN changefor our current climate has been, and ifthe current models even can capture suchtrends in N or CCN, is not yet known. Com-prehensive model results could also providemuch better view on potential biogenic oroceanic aerosol feedbacks on the climatesystem and on their roles in aerosol trends.

The global number concentrations, assimulated by ECHAM-HAM simulations inPI, are sensitive to the inclusion of newparticle formation. This inclusion also im-proves model performance, and changes theCCN concentrations significantly. As thenew particle formation is dependent on sul-phuric acid concentrations, the results sug-gest that the atmosphere could be more sen-sitive to the amount of sulphur emitted thanwould be expected without NPF. This sen-sitivity is highlighted in PV, where the cur-rent day aerosol forcing is increased signif-icantly by inclusion of boundary layer newparticle formation. The PV also suggesteda strong warming effect of radical emis-sion reductions prognosed by IPCC emis-sion pathways, mainly from aerosol indirecteffects on clouds.

The importance of SO2 emissionschanges to aerosol forcing is perhaps evenbetter underlined by comparing the emis-sion reductions to the proposed climate en-gineering methods. The atmospheric sci-

ence community has been very hesitant topromote climate engineering methods, dueto potential unforeseen consequences. Cli-mate engineering methods have high inher-ent risks, could significantly make the sit-uation worse in the long run, could forceus to long-term upkeep of potentially haz-ardous operations, and could reduce themotivation to work with the actually impor-tant (but expensive) emission reductions ofgreenhouse gasses (Robock, 2008). Recentstudies have shown that the climate coolingforcing of affordable climate engineering ef-forts are close to similar magnitude as thewarming forcing from emission reductionsin PV (Shepherd et al., 2009). In the cli-mate engineering, a positive effect (cooling)is expected, but we are not sure of the ad-verse side effects. In the case of the aerosolemission decreases, a negative climatic ef-fect (warming) is suggested in PV andother studies (Brasseur and Roeckner, 2005;Andreae et al., 2005; Kloster et al., 2008,2010), although such changes will have pos-itive effects on air quality (Londahl et al.,2010).

In this context, it is unfortunate if thesociety (including the atmospheric researchcommunity) is not considering the climateeffects (including aerosol indirect effects)in impact assessment of the emission re-ductions. Such considerations are neces-sary especially considering adverse effectsof extreme weather on population health(McMichael et al., 2006). As an exam-ple, the impact assessment summary of therecent directive to reduce sulphur contentin marine fuels does not have any men-tion on negative impacts to climate (EC,2011), instead concentrating on economi-cal and health impacts. The health impactassessments are done only for changes intotal particulate mass (PM), although theincrease of SO4 might not always lead toincreased adverse long term health effects(Heyder et al., 2009). The latest EuropeanEnvironment Agency report does, however,

35

already acknowledge the potential of ad-verse side effects to climate from air qualitydirectives, with comment that current mod-els are still incapable to describe particles inthe atmosphere and the aerosol-cloud inter-actions (EEA, 2012). Perhaps in the future,also potential warming impacts will be as-sessed more in detail.

On the other hand, all studies of SO2

reduction side-effects using detailed aerosolmicrophysics (including PV) are done withversions the same GCM: ECHAM5-HAM(Kloster et al., 2008, 2010). I was recentlyinformed that there is a possibly an over-estimation of the SO2 levels in ECHAM5-HAM, leading to possible too high sen-sitivity on sulphur emissions (Dr. De-clan O’Donnell, personal communication,2012). This effect is still unknown, butcould result in smaller overall aerosol cli-mate effect in the model calculations, andthus smaller overall current day-to-futuredifferences. However, the studies donewith models without detailed aerosol mi-crophysics still give high warming poten-tial to anthropogenic aerosol emissions cuts(e.g. Brasseur and Roeckner, 2005; Andreaeet al., 2005; Lamarque et al., 2011), andthus the effect of this possible overestima-tion of SO2 is probably only a matter ofscale.

One could think that we have alreadydone a long time “accidental climate en-gineering”1 by inclusion of sulphur dioxideand primary particle emissions in the at-mosphere. Now removing this cooling ef-fect will bring us closer to the pre-industrialsituation in one aspect, but due to the in-crease in the CO2 and other long-lived cli-mate warming gases, in a completely dif-ferent end result. The rapidness of aerosolremoval in the atmosphere should be con-sidered, and the reductions in aerosol emis-sions should be perhaps concentrated to lo-

cations where their impact on air qualityis largest, and their relative climate im-pact is smallest. As the aerosol-cloud in-teraction saturates at high CCN concentra-tions, perhaps the best is to reduce the SO2

emissions close to regions with already highaerosol loading, and consider even increas-ing the emissions in regions with low aerosolloads, such as over the oceans. Naturally,the potential ecosystem damages e.g. fromincreased acidity should be taken into ac-count as well. These are just rough commonsense estimates, and actual policy decisionsshould be done using best possible knowl-edge of the trade-offs required.

This discussion of SO2 reductions shouldnot be taken as a promotion for climateengineering, or as promoting sulphur emis-sions. I consider these possible adverse ef-fects of emission reductions as a warning ex-ample of the kind of side effects climate en-gineering mechanisms can bring. The po-tential side effects of SO2 reductions arecomparable to the “termination effect” riskof the climate engineering mentioned by(Shepherd et al., 2009) and could thus bean indicative of the kinds of quagmires wemight end up if we actually start to use cli-mate engineering methods in the future.

Research needs

Model-measurement comparisons of CCNsized particles are still not satisfactory, atleast in the global scale. The lack of longtime series of number concentration datain many parts of world limit the possibil-ity to do comprehensive mode/data com-parisons. The dearth of data is even worsefor the number size distributions. Acquir-ing long-term data series require strongcommitment, not only from measurementgroups, but from the funding agencies aswell. For this thesis, the work done in the

1 “Accidental climate engineering” is slightly inaccurate terminology – most definitions require climateengineering to be intentional.

36

frameworks of GAW, EUSAAR and laterACTRIS were very valuable, and hopefullysuch networks continue to operate and evenexpand. The lack of long-term data inSouth America, Africa, Asia, Australasiaand over the oceans is crippling for a goodevaluation of the model performance. Per-haps even harder to achieve would be long-term data series from the marine bound-ary layer, and from the free troposphere.Shorter-time measurements can be a sub-stitute for some applications, but for a ro-bust comparison, statistical properties fromlong term datasets are the only really usefulcomparison parameters for GCMs.

There is almost no information on theambient number size distributions of super-micron aerosol particles. There are only afew stations reporting reliable number con-centration data from this size range, andthus the models’ abilities to predict theseconcentrations, and related aerosol-cloudinteractions, are still poorly constrained.

Using datasets from networks even withhigh standards in measurement quality cansometimes be challenging. The use of thedata sometimes needs significant extra workfrom the data user, in the form of additionalquality checking and communicating withthe data provider. Even finding the correctdatasets can be challenging, and the dataformats, measurement standards and datausage policies can widely differ in differentnetworks. There are currently several inter-national projects working on making suchenvironmental data easier to obtain2.

Going through monitoring data requiresa lot of expertise on what is expected fromthe data measurements. The data clean-ing is also very time consuming and labour

intensive. Hopefully, the current tendencytowards providing datasets peer review3,and citable persistent document identifiers(e.g. Digital Object Identifier, DOI)4, willimprove incentives and resources for dataquality checking. When these data cita-tions are also stored in scientific citationdatabases, the production of quality datacan be better taken into account whenevaluating personal achievements. The re-sources needed for the data processing andsubmission should also be explicitly re-served in the finance plans of measurementprojects.

The model/measurement inter-comparisons is further hampered by thelimited spatial and temporal resolutionsthe GCMs can work with. The differ-ences of local measurements and large spa-tial scale grid boxes of models can createmany problems for the comparisons. InGCMs many processes are already highlyparametrized due to their sub-grid nature,and such approaches could as well be usedfor the aerosol-climate interactions. Es-pecially considering the high sensitivity ofNPF to background aerosol and conden-sible vapour concentrations, the sub-gridvariability in near emission source regionscould be important and should be bettercharacterised for the global models.

Another issue with model resolutionis the availability of short-scale (diurnalor weekly) temporal variation in differentemissions. In ECHAM5-HAM, most of theemissions of aerosols and precursors areemitted using monthly-mean 2D emissionfields, i.e. no short-term temporal variabil-ity was considered. Again, considering theshort time scales of many aerosol processes,this averaging approach can lead to very dif-

2 I am currently involved in two such international projects, European ENVRI (http://envri.eu/) andEurope-US COOPEUS (http://www.coopeus.eu/)

3 e.g. Earth System Science Data (ESSD) data journal, http://www.earth-system-science-data.net/4 e.g. PANGAEA project, http://www.pangaea.de/ stores environmental datasets, provides them with

DOIs and provides search for different properties. I intend to supply the PII dataset to there within thisyear if the co-authors agree.

37

ferent results than has been seen in the na-ture.

Similarly, the emissions are usuallypoorly characterized in sub micron sizeranges, often only PM emissions are pre-scripted, and constant PM-to-number sizedistribution factors are used globally (Tex-tor et al., 2007). The model results arenecessarily very sensitive to these conver-sion factors (Spracklen et al., 2010), andmore consistent bottom-up approaches toemissions, such as recent European numbersize distribution emission inventory (Kul-mala et al., 2011), are needed. The mod-elling studies to study the importance ofprimary vs. secondary particle concentra-tions have already shown the benefits ofthese emission approaches in Europe (Red-dington et al., 2011).

There are still a lot of uncertainty fromthe feedbacks in the Earth system, whichcould easily affect the aerosol-climate in-teractions (Carslaw et al., 2010). Moreinformation on biosphere-climate interac-tions (via e.g. temperature effects on VOCemissions); ocean-climate interactions (viae.g. sea-spray or DMS emissions); oreven climate-dependent changes in anthro-pogenic emissions (e.g. changes in domes-tic combustion change in warming climate)are needed. Answers to these problems re-quire a lot of new work, not only on processscale, but also on developing Earth SystemModels with the capabilities to handle thesedynamics.

Another quite significant drawback inthe current ECHAM5-HAM is the lackof on-line oxidant chemistry. For perfor-mance reasons, the ECHAM5-HAM usesprescribed monthly OH fields, althoughwith solar radiation scaling. Such ap-proaches are very crude, especially consid-ering the influence of OH fields on SO2 oxi-dation to sulphuric acid. The oxidant fieldsare even more important when considering

more realistic approaches to gas and liquidphase organic or nitrate chemistry, espe-cially on the generation of additional oxi-dants from the organic species themselves(Mauldin III et al., 2012) or from nitratesin the soil (Su et al., 2011).

The models themselves can always beimproved process-wise. Studies in this the-sis and otherwise have shown the impor-tance of the new particle formation to theclimate system. Mechanistic approaches tothe nucleation mechanism should improvethe simulations of especially past and fu-ture conditions, where the applicability ofsemi-empirical relationships – developed incurrent atmospheric chemistry background– might not be relevant. Notably, mostof the NPF parameterisations are based onmeasurements in a limited range of differentenvironments, making even generalisationsin current climate suspect. Recent work onvery advanced laboratory measurements inCERN (CLOUD project) are very promis-ing in this respect (Kirkby et al., 2011).Even though the physico-chemical processesof the atmospheric nucleation would be de-termined, we will also need realistic knowl-edge on the concentrations and propertiesof organic and inorganic vapours responsi-ble for the particle growth.

The aerosol-cloud interactions in GCMsare necessarily very simplified. There are afew aspects of the current parametrizationsthat are known to represent the aerosolpoorly: Currently ECHAM assumes a con-stant entrainment rate for the convectiveclouds. This parameter is actually one ofthe key tuning parameters of the model, andthus considered to be a relatively free pa-rameter. An other assumption is the sim-plistic idea of adiabatic cloud systems. Al-most all GCMs use the idea of adiabat-ically rising air parcels (with highly pa-rameterised updraft velocities) as the basisof CCN activation and subsequent cloud-top CDNC. This approach does not specif-

38

ically take into account 3D complex na-ture of the clouds, and some studies haveshown that droplets can participate severalactivation-evaporation cycles before end-ing up in the cloud-top (Flossmann andWobrock, 2010). The development of aparametrization which includes such effectswill be very challenging. Similarly, the on-set of ice nucleation is a critical factor onrain formation; the level of knowledge inthis field is still far from the level needed forcomprehensive climate impact assessment(Curry and Khvorostyanov, 2012).

Overall, the GCMs will require a morecomplete and systematic evaluation of ef-fects of uncertainty to the climate pa-rameters. Such studies have been donefor global chemistry transport models (Leeet al., 2012) and for more traditional cli-mate models without interactive aerosols

(Knight et al., 2007), and they could guidethe research towards the more uncertainparts of the modelling systems.

Efforts should also be used towards us-ing and developing more advanced methodsof data mining and analysis. I chose thewavelet analysis used in PIII to find newinformation from older datasets. This ap-proach could be further developed using e.gwavelet coherency analysis to determine inwhich time scales different time-series be-have similarly. That and other such meth-ods can prove to be critical to analyse pos-sible long-term influences in the variabil-ity of climate relevant factors in the atmo-sphere. During the next decades, new andinnovative approaches to data generation,data analysis, modelling and science man-agement are needed.

39

SUMMARY OF PUBLICATIONS AND AUTHOR CONTRIBUTION

PI Makkonen, R., Asmi, A., Korhonen, H., Kokkola, H., Jarvenoja, S., Raisanen,P., Lehtinen, K. E. J., Laaksonen, A., Kerminen, V.-M., Jarvinen, H., Lohmann,U., Bennartz, R., Feichter, J., and Kulmala, M. (2009). Sensitivity of aerosol con-centrations and cloud properties to nucleation and secondary organic distributionin ECHAM5–HAM global circulation model. Atmospheric Chemistry and Physics,9(5):1747–1766. (a)

Overview: A modelling paper, where we used ECHAM5-HAM model to calcu-late the sensitivity of CDNC concentrations to different boundary layer nucleationschemes. The overall conclusions of this paper are that the boundary layer NPFcan influence the climate in significant way, although the use of single value foractivation coefficient can lead to overestimation of aerosol number concentrations.The addition of different approach to organic aerosol handling was also tested.

Author contribution: I planned together with R. Makkonen and V.-M. Kerminenthe experiments, and we also did the interpretation of the results together. I wrotethe initial discussion part of the paper (which ended up in the final version almostin verbatim), and wrote together with R. Makkonen the conclusions of the paper.I did also small part of the coding.

PII Asmi, A., Wiedensohler, A., Laj, P., Fjaeraa, A.-M., Sellegri, K., Birmili, W.,Weingartner, E., Baltensperger, U., Zdimal, V., Zikova, N., Putaud, J.-P., Marinoni,A., Tunved, P., Hansson, H.-C., Fiebig, M., Kivekas, N., Lihavainen, H., Asmi,E., Ulevicius, V., Aalto, P. P., Swietlicki, E., Kristensson, A., Mihalopoulos, N.,Kalivitis, N., Kalapov, I., Kiss, G., de Leeuw, G., Henzing, B., Harrison, R. M.,Beddows, D., O’Dowd, C., Jennings, S. G., Flentje, H., Weinhold, K., Meinhardt,F., Ries, L., and Kulmala, M. (2011). Number size distributions and seasonalityof submicron particles in Europe 2008-2009. Atmospheric Chemistry and Physics,11(11):5505–5538. (b)

Overview: A data analysis paper, which combined aerosol number size distributionmeasurements from 24 stations around Europe. This resulted in a set of metricswhich has already been used in model evaluation. The dataset also provides har-monized and reliable way to establish the current variability of regional backgroundaerosol size distributions in Europe.

Author contribution: I had the original idea of the paper (with some input fromC. O’Dowd). I used ready-made datasets from the EBAS data repository (i.e. Idid not do any measurements), programmed the analysis codes, wrote most of thearticle, decided on the chosen metrics, methods of analysis and made the figuresand tables.

PIII Asmi, A. (2012). Weakness of the weekend effect in aerosol number concentrations.Atmospheric Environment, 11(11):5505–5538. (c)

Overview: A short data analysis paper, where I used spectral and statistical meth-ods to show that there is no strong 7-day (i.e. potentially directly anthropogenic)oscillation in European regional background of CCNs or CCN-sized aerosol particlenumber concentrations. Reasons for this disparity are discussed, and I show thatthe aerosol in the smaller CCN-sizes contributing to convective cloud CDNC, arebehaving much differently from the larger aerosol measured in PM monitors.

Author contribution: I was the only scientific contributor, the data was from thedata providers, but all analyses were done by me.

PIV Asmi, A., M. Collaud Coen, J. A. Ogren, E. Andrews, P. Sheridan, A. Jefferson,E. Weingartner, U. Baltensperger, N. Bukowiecki, H. Lihavainen, N. Kivekas, E.Asmi, P. P. Aalto, M. Kulmala, A. Wiedensohler, W. Birmili, A. Hamed, C. ODowd,S.G. Jennings, R. Weller, H. Flentje, A. M. Fjaeraa, M. Fiebig, C. L. Myhre, A.G. Hallar, and P. Laj (2012). Aerosol decadal trends (II): In-situ aerosol particlenumber concentrations at GAW and ACTRIS stations. Atmospheric Chemistry andPhysics Discussions, 12, 20849-20899 (d)

Overview: Trend analysis of number concentrations. Some traits seemed to beobvious: The global trends of particle number concentrations seem to be decreasingsince the middle 1990s on; The trends of and N are relatively consistent in NorthernEurope, but no clear similarity with optical properties trends were visible in thestations where all properties were measured;

Author contribution: I coded the GLS/ARB trend fitting routine based on litera-ture algorithms, selected the data models and for a small part took a part in theMK method coding. I collected and checked the dataset, communicated with thedata providers, did the data analysis and trend fitting, selected and collected thecomparison datasets and did the related data analysis on them. I also wrote thepaper, and collated the co-author comments and revisions. Note that some of thefigures have been enlarged from the ACPD version for better visibility in the thesisformat.

PV Makkonen., R, Asmi, A., Kerminen, V.-M., Boy, M., Arneth, A., Hari, P., andKulmala, M (2012). Air pollution control and decreasing new particle formationlead to strong climate warming. Atmospheric Chemistry and Physics, 12(3):1515-1524 (d)

Overview: We tested the effect of emission changes and new particle formation onanthropogenic aerosol forcing in present-day (year 2000) and future (year 2100)conditions. The predicted reduction in SO2 emission rates will decrease the aerosolcooling effect, with potentially serious consequences, an effect even further increasedby including current nucleation parametrizations in the model. Even including twoclimate feedbacks, radical increase in marine DMS or BVOC emissions in the futureare not enough to overcome the SO2 effect.

Author contribution: I contributed to the planning of the study, data interpretationand conclusions, choosing the simulations to be done, draw some of the figures, andparticipated on the writing of the paper.

41

(a) c©Author(s) 2009. This work is distributed under the Creative Commons Attribu-tion 3.0 License.

(b) c©Author(s) 2011. This work is distributed under the Creative Commons Attribu-tion 3.0 License.

(c) c©2012 Elsevier Ltd. All rights reserved. Reproduced within the retained authorrights as described in http://www.elsevier.com/wps/find/authorsview.authors/rights: “theright to include the journal article, in full or in part, in a thesis or dissertation;”

(d) c©Author(s) 2012. This work is distributed under the Creative Commons Attribu-tion 3.0 License.

42

BIBLIOGRAPHY

Angstrom, A. (1929) On the atmospheric transmis-sion of sun radiation and on dust in the air. Ge-ografiska Annaler, 11:156–166.

Ackerman, A. S., et al. (2004) The impact of humid-ity above stratiform clouds on indirect aerosolclimate forcing. Nature, 432(7020):1014–1017.

Aitken, J. (1889a) On improvements in the appara-tus for counting the dust particles in the atmo-sphere. Nature, XXXIX.

Aitken, J. (1889b) On the number of dust parti-cles in the atmosphere of certain places in GreatBritain and on the continent, with remarks onthe relation between the amount of dust and me-teorological phenomena, part I. Nature, XLI.

Albrecht, B. A. (1989) Aerosol, cloud microphysics,and fractional cloudiness. Science, 245:1227–1230.

Andreae, M. and Rosenfeld, D. (2008) Aerosol-cloud-precipitation interactions. part 1. thenature and sources of cloud-active aerosols.Earth-Sci. Rev., 89(1-2):13 – 41. doi:DOI:10.1016/j.earscirev.2008.03.001.

Andreae, M., et al. (2005) Strong present-day cool-ing implies a hot future. Nature, 435:1187–1190.

Andreae, M. O. and Gelencser, A. (2006) Blackcarbon or brown carbon? the nature of light-absorbing carbonaceous aerosols. Atmos. Chem.Phys., 6(10):3131–3148. doi:10.5194/acp-6-3131-2006.

Anttila, T. and Kerminen, V.-M. (2003) Conden-sational growth of atmospheric nuclei by organicvapours. Aerosol Sci., 34:41–61.

Arneth, A., et al. (2009) Clean the air, heat theplanet? Science, 326(5953):672–673. doi:10.1126/science.1181568.

Barmet, P., et al. (2009) Weekly cycle in particu-late matter versus weekly cycle in precipitationover Switzerland. J. Geophys. Res., 114:D05206.doi:10.1029/2008JD011192.

Bell, M. L., et al. (2004) A retrospective assessmentof mortality from the London smog episode of1952: The role of influenza and pollution. Envi-ron. Health Perspect., 112(1):6–8.

Bell, T. L. and Rosenfeld, D. (2008) Comment on“weekly precipitation cycles? lack of evidencefrom United States surface stations”; by D. M.Schultz et al. Geophys. Res. Lett., 35(9):L09803.

Bergman, T., et al. (2012) Evaluation of the sec-tional aerosol microphysics module SALSA im-plementation in ECHAM5-HAM aerosol-climatemodel. Geosci. Model Dev., 5(3):845–868. doi:10.5194/gmd-5-845-2012.

Bonn, B., et al. (2008) How biogenic terpenes gov-ern the correlation between sulfuric acid concen-trations and new particle formation. J. Geophys.Res., 113(D12):D12209.

Brasseur, G. P. and Roeckner, E. (2005) Impact ofimproved air quality on the future evolution ofclimate. Geophys. Res. Lett., 32(23):L23704–.

Brimblecombe, P. (1987) The Big Smoke: a his-tory of air pollution in London since medievaltimes. Methuen & Co. Ltd, London, UK. ISBN0-416-90080-1.

Carslaw, K. S., et al. (2010) A review of natu-ral aerosol interactions and feedbacks within theearth system. Atmos. Chem. Phys., 10(4):1701–1737. doi:10.5194/acp-10-1701-2010.

Charlson, R. J., et al. (1987) Oceanic phytoplank-ton, atmospheric sulphur, cloud albeno and cli-mate. Nature, 326:655–661.

Chen, Y.-C., et al. (2011) A comprehensive numer-ical study of aerosol-cloud-precipitation interac-tions in marine stratocumulus. Atmos. Chem.Phys., 11(18):9749–9769. doi:10.5194/acp-11-9749-2011.

Cofala, J., et al. (2007) Scenarios of global anthro-pogenic emissions of air pollutants and methaneuntil 2030. Atmos. Env., 41:8387–8499.

Collaud Coen, M., et al. (2012) Aerosol decadaltrends – part 1: In-situ optical measurementsat GAW and IMPROVE stations. Atmos.Chem. Phys. Discuss., 12(8):20785–20848. doi:10.5194/acpd-12-20785-2012.

Curry, J. A. and Khvorostyanov, V. I. (2012) As-sessment of some parameterizations of heteroge-neous ice nucleation in cloud and climate mod-els. Atmos. Chem. Phys., 12(2):1151–1172. doi:10.5194/acp-12-1151-2012.

Daniel, J. S., et al. (2012) Identifying weekly cyclesin meteorological variables: The importance ofan appropriate statistical analysis. J. Geophys.Res., 117(D13):D13203–.

Decesari, S., et al. (2000) Characterization ofwater-soluble organic compounds in atmosphericaerosol: A new approach. J. Geophys. Res.,105(D1):1481–1489.

Dockery, D. W. and Pope, C. (1994) Acute respi-ratory effects of particulate air pollution. Annu.Rev. Public. Health, 15:107–132.

Dockery, D. W., et al. (1993) An association be-tween air polution and mortality in six U.S.cities. N. Engl. J. Med., 329:1753–1749.

Donahue, N. M., et al. (2006) Coupled partition-ing, dilution, and chemical aging of semivolatileorganics. Environ. Sci. Technol., 40:2635–2643.doi:10.1021/es052297c.

Donahue, N. M., et al. (2012) A two-dimensionalvolatility basis set part 2: Diagnostics oforganic-aerosol evolution. Atmos. Chem. Phys.,12(2):615–634. doi:10.5194/acp-12-615-2012.

Donaldson, K., et al. (1998) Ultrafine (nanome-tre) particle mediated lung injury. J. AerosolSci., 29(5-6):553 – 560. doi:DOI: 10.1016/S0021-8502(97)00464-3.

Donaldson, K., et al. (2002) The pul-monary toxicology of ultrafine particles.J. Aerosol Med., 15(2):213–220. doi:10.1089/089426802320282338.

Dusek, U., et al. (2006) Size matters more thanchemistry for cloud-nucleating ability of aerosolparticles. Science, 312:1375–1378.

EC (2008) Commission regulation (EC) no682/2008 of 18 july 2008. Off. J. Eur. Union:Legis., L199:1–136. The Commission of the Eu-ropean Communities.

EC (2011) Commission staff working paper – sum-mary of the impact assesment. accompanyingdocument to the proposal for a directive of theeuropean parliament and of the council amend-ing directive 1999/32/ec as regards the sulphurcontent of marine fuels. Tech. Rep. SEC(2011)919 final, European Commission, Brussels.

EEA (2012) Air Quality in Europe – 2012 report.No. 4/2012 in EEA Report. European Environ-ment Agency, Copenhagen, 2012.

EMEP (2011) Transboundary particulate matter inEurope, joint CCC, MSC-W, CEIP and CIAMreport 2011. Tech. rep., Convention on long-range transboundary air pollution, EMEP Co-operative programme for monitoring and evalu-ation of long-range transmission of air pollutantsin Europe., Kjeller, Norway. ISSN: 1504-6192.

Ervens, B., et al. (2010) CCN predictions using sim-plified assumptions of organic aerosol composi-tion and mixing state: a synthesis from six differ-ent locations. Atmos. Chem. Phys., 10(10):4795–4807. doi:10.5194/acp-10-4795-2010.

Ervens, B., et al. (2011) Secondary organic aerosolformation in cloud droplets and aqueous par-ticles (aqSOA): a review of laboratory, fieldand model studies. Atmos. Chem. Phys.,11(21):11069–11102. doi:10.5194/acp-11-11069-2011.

Facchini, M., et al. (1999) Cloud albedo enhance-ment by surface-active organic solutes in growingdroplets. Nature, 401:257–259.

Flossmann, A. I. and Wobrock, W. (2010) A re-view of our understanding of the aerosol–cloudinteraction from the perspective of a bin resolvedcloud scale modelling. Atmos. Res., 97:478–497.

Forster, P. M. d. F. and Solomon, S. (2003)Observations of a weekend effect in diur-nal temperature range. Proc. Natl. Acad.Sci. U. S. A., 100(20):11225–11230. doi:10.1073/pnas.2034034100.

Fountoukis, C., et al. (2012) Simulating ultrafineparticle formation in Europe using a regionalCTM: contribution of primary emissions ver-sus secondary formation to aerosol number con-centrations. Atmos. Chem. Phys., 12(18):8663–8677. doi:10.5194/acp-12-8663-2012.

Fuzzi, S., et al. (2006) Critical assesment of the cur-rent state of scientific knowledge, terminology,and research needs concerning the role of organicaerosols in the atmosphere, climate, and globalchange. Atmos. Chem. Phys., 6:2017–2038.

Ghan, S. J., et al. (2011) Droplet nucleation:Physically-based parameterizations and compar-ative evaluation. J. Adv. Model. Earth Syst.,3:M10001. doi:10.1029/2011MS000074.

Gilman, D., et al. (1963) On the power spectrumof ”red noise”. J. Atmos. Sci., 20:181–184.

Gronholm, T., et al. (2007) Measurements ofaerosol particle dry deposition velocity us-ing the relaxed eddy accumulation technique.Tellus B, 59(3):381–386. doi:10.1111/j.1600-0889.2007.00268.x.

44

Gultepe, I. and Isaac, G. A. (1999) Scale effectson averaging of cloud droplet and aerosol num-ber concentrations: Observations and models.J. Climate, 12(5):1268–1279. doi:10.1175/1520-0442(1999)012¡1268:SEOAOC¿2.0.CO;2.

Gultepe, I., et al. (1996) Parameterizationsof marine stratus microphysics based on insitu observations: Implications for GCMS.J. Climate, 9:345–357. doi:10.1175/1520-0442(1996)009¡0345:POMSMB¿2.0.CO;2.

Hand, J. L., et al. (2011) IMPROVE, spa-tial and seasonal patterns and temporalvariability of haze and its constituentsin the United States. Web report.Http://vista.cira.colostate.edu/improve/publications/Reports/2011/2011.htm, ISSN:0737–5352–87.

Hansen, J., et al. (1981) Climate impact of in-creasing atmospheric carbon dioxide. Science,213:4511.

Hanson, D. R. and Lovejoy, E. R. (2006) Measure-ment of the thermodynamics of the hydrateddimer and trimer of sulfuric acid. J. Phys.Chem., 110:9525–9528.

Heintzenberg, J., et al. (1998) Mass-related aerosolproperties over the Leipzig basin. Journal ofGeophysical Research D, 103:13125–13135.

Heyder, J., et al. (2009) Long-term responses of ca-nine lungs to acidic particles responses to acidicparticles. Inhal. Toxicol., 21(11):920–932.

Hirsch, R. M., et al. (1982) Techniques of trendanalysis for monthly water quality data. WaterResour. Res., 18(1):107–121.

Hoppel, W. A., et al. (1990) Aerosol size distribu-tions and optical properties found in the marineboundary layer in the Atlantic ocean. Journal ofGeophysical Research D, 95:3659–3686.

Iversen, T., et al. (2010) Assessing impacts ofaerosol processes on equilibrium climate sensi-tivity. In Steyn, D. G. and Rao, S. T., eds.,Air Pollution Modeling and Its Application XX,chap. Interactions between air quality and cli-mate change, pp. 493–497. Springer.

Jaenicke, R. (2008) Is atmospheric aerosolan aerosol?–a look at sources and variabil-ity. Faraday Discuss., 137:235–243. doi:10.1039/B701095H.

Jimenez, J. L., et al. (2009) Evolutionof organic aerosols in the atmosphere.Science, 326(5959):1525–1529. doi:10.1126/science.1180353.

Junker, C., et al. (2006) Aerosol light absorption inthe north atlantic: trends and seasonal charac-teristics during the period 1989 to 2003. Atmos.Chem. Phys., 6(7):1913–1925. doi:10.5194/acp-6-1913-2006.

Kanakidou, M., et al. (2005) Organic aerosol andglobal climate modelling: a review. Atmos.Chem. Phys., 5(4):1053–1123. doi:10.5194/acp-5-1053-2005.

Kerminen, V. M. and Kulmala, M. (2002) Analyti-cal formulae connecting the ”real” and the ”ap-parent” nucleation rate and the nuclei numberconcentration for atmospheric nucleation events.J. of Aerosol Science, 33:609–622.

Kerminen, V.-M. and Wexler, A. S. (1995) En-hanced formation and development of sulfateparticles due to marine boundary layer circula-tion. J. Geophys. Res., 100:23,051–23,062.

Kerminen, V. M., et al. (2000) Secondary organicsand atmospheric cloud condensation nuclei pro-duction. J. Geophys. Res., 105:9255–9264.

Kerminen, V.-M., et al. (2012) Cloud condensationnuclei production associated with atmosphericnucleation: a synthesis based on existing litera-ture and new results. Atmos. Chem. Phys. Dis-cuss., 12(8):22139–22198. doi:10.5194/acpd-12-22139-2012.

Kim, K.-Y., et al. (2010) Weekend effect: An-thropogenic or natural? Geophys. Res. Lett.,37(L09808). doi:10.1029/2010GL04323.

Kirkby, J., et al. (2011) Role of sulphuric acid, am-monia and galactic cosmic rays in atmosphericaerosol nucleation. Nature, 476(7361):429–433.

Kloster, S., et al. (2008) Influence of futureair pollution mitigation strategies on totalaerosol radiative forcing. Atmos. Chem. Phys.,8(21):6405–6437. doi:10.5194/acp-8-6405-2008.

Kloster, S., et al. (2010) A GCM study of futureclimate response to aerosol pollution reductions.Climate Dyn., 34:1177–1194. 10.1007/s00382-009-0573-0.

Knight, C. G., et al. (2007) Association of parame-ter, software, and hardware variation with large-scale behavior across 57,000 climate models.Proc. Natl. Acad. Sci. U. S. A., 104(30):12259–12264. doi:10.1073/pnas.0608144104.

Knote, C., et al. (2011) Towards an online-coupledchemistry-climate model: evaluation of tracegases and aerosols in cosmo-art. Geosci. ModelDev., 4(4):1077–1102. doi:10.5194/gmd-4-1077-2011.

45

Koch, D. and Del Genio, A. D. (2010) Black carbonsemi-direct effects on cloud cover: review andsynthesis. Atmos. Chem. Phys., 10(16):7685–7696. doi:10.5194/acp-10-7685-2010.

Konwar, M., et al. (2012) Aerosol control on depthof warm rain in convective clouds. J. Geophys.Res., 117(D13):D13204.

Kroll, J. H., et al. (2011) Carbon oxidation stateas a metric for describing the chemistry of atmo-spheric organic aerosol. Nat Chem, 3(2):133–139.

Kuang, C., et al. (2008) Dependence of nucleationrates on sulfuric acid vapor concentration in di-verse atmospheric locations. J. Geophys. Res.,113(D10):D10209–.

Kulmala, M., et al. (2004a) Formation and growthrates of ultrafine atmospheric particles: a reviewof observations. J. Aerosol Sci., 35:143–176.

Kulmala, M., et al. (2004b) Initial steps of aerosolgrowth. Atmos. Chem. Phys., 4:2553–2560.

Kulmala, M., et al. (2004c) A new feedback mech-anism linking forests, aerosols, and climate. At-mos. Chem. Phys., 4:557–562.

Kulmala, M., et al. (2004d) Organic aerosol forma-tion via sulphate cluster activation. J. Geophys.Res., 109:10.1029/2003JD003961.

Kulmala, M., et al. (2011) General overview: Eu-ropean integrated project on aerosol cloud cli-mate and air quality interactions (EUCAARI);integrating aerosol research from nano to globalscales. Atmos. Chem. Phys., 11(24):13061–13143. doi:10.5194/acp-11-13061-2011.

Kumar, P. and Foufoula-Georgiou, E. (1997)Wavelet analysis for geophysical applica-tions. Rev. Geophys., 35(4):385–412. doi:10.1029/97RG00427.

Kurten, T., et al. (2008) Amines are likely toenhance neutral and ion-induced sulfuric acid-water nucleation in the atmosphere more effec-tively than ammonia. Atmos. Chem. Phys.,8(14):4095–4103. doi:10.5194/acp-8-4095-2008.

Laakso, L., et al. (2003) Ultrafine particle scaveng-ing coefficients calculated from 6 years field mea-surements. Atmos. Environ., 37:3605–3613.

Lack, D. A., et al. (2004) Seasonal variability of sec-ondary organic aerosol: A global modeling study.J. Geophys. Res., 109:doi:10.1029/2003JD00.

Lamarque, J.-F., et al. (2011) Global and regionalevolution of short-lived radiatively-active gasesand aerosols in the representative concentrationpathways. Clim. Change, 109:191–212. doi:10.1007/s10584-011-0155-0.

Lee, L. A., et al. (2012) Mapping the uncer-tainty in global CCN using emulation. Atmos.Chem. Phys. Discuss., 12(6):14089–14114. doi:10.5194/acpd-12-14089-2012.

Leibensperger, E. M., et al. (2012) Climatic ef-fects of 1950–2050 changes in us anthropogenicaerosols – part 1: Aerosol trends and radiativeforcing. Atmos. Chem. Phys., 12(7):3333–3348.doi:10.5194/acp-12-3333-2012.

Lide, D. R. (2004) CRC Handbook of Chemistryand Physics. CRC Press LLC, 85th edn.

Lin, H. and Leaitch, W. R. (1997) Developmentof an in-cloud aerosol activation parameteriza-tion for climate modelling. In Proceedings ofthe WMO Workshop on Measurement of CloudProperties for Forecasts of Weather, Air Qual-ity and Climate, pp. 328–335. World Meteorol.Organ, Geneva.

Liu, B. Y. H. and Kim, C. S. (1977) On the countingefficiency of condensation nuclei counters. At-mos. Environ., 11:1097–1100.

Lohmann, U., et al. (2007) Cloud microphysicsand aerosol indirect effects in the global climatemodel ECHAM5-HAM. Atmos. Chem. Phys.,7(13):3425–3446.

Lohmann, U., et al. (2010) Total aerosol effect: ra-diative forcing or radiative flux perturbation?Atmos. Chem. Phys., 10(7):3235–3246. doi:10.5194/acp-10-3235-2010.

Londahl, J., et al. (2010) Aerosol exposure versusaerosol cooling of climate: what is the optimalemission reduction strategy for human health?Atmos. Chem. Phys., 10(19):9441–9449. doi:10.5194/acp-10-9441-2010.

Lovejoy, E., et al. (2004) Atmospheric ion-inducednucleation of sulfuric acid and water. J. Geo-phys. Res., 109:D08204.

Lu, M.-L. and Seinfeld, J. H. (2005) Study ofthe aerosol indirect effect by large-eddy simula-tion of marine stratocumulus. J. Atmos. Sci.,62(11):3909–3932. doi:10.1175/JAS3584.1.

Makkonen, R., et al. (2012) Brightening of theglobal cloud field by nitric acid and the as-sociated radiative forcing. Atmos. Chem.Phys., 12(16):7625–7633. doi:10.5194/acp-12-7625-2012.

46

Mann, G. W., et al. (2012) Intercomparisonof modal and sectional aerosol microphysicsrepresentations within the same 3-d globalchemical transport model. Atmos. Chem.Phys., 12(10):4449–4476. doi:10.5194/acp-12-4449-2012.

Martensson, E. M., et al. (2003) Laboratory sim-ulations and parameterization of the primarymarine aerosol production. J. Geophys. Res.,108(D9):4297–.

Martin, G. M., et al. (1994) The measure-ment and parameterization of effective radiusof droplets in warm stratocumulus clouds. J.Atmos. Sci, 51:18231842. doi:10.1175/1520-0469(1994)051¡1823:TMAPOE¿2.0.CO;2.

Mauldin III, R. L., et al. (2012) A new atmospheri-cally relevant oxidant of sulphur dioxide. Nature,488(7410):193–196.

McFiggans, G., et al. (2006) The effect of phys-ical and chemical aerosol properties on warmcloud droplet activation. Atmos. Chem. Phys.,6(9):2593–2649. doi:10.5194/acp-6-2593-2006.

McFiggans, G., et al. (2010) The sensitivity of sec-ondary organic aerosol component partitioningto the predictions of component properties part1: A systematic evaluation of some availableestimation techniques. Atmos. Chem. Phys.,10(21):10255–10272. doi:10.5194/acp-10-10255-2010.

McMichael, A. J., et al. (2006) Climate change andhuman health: present and future risks. Lancet,367(9513):859–869.

McMurry, P. (2000) A review of atmosphericaerosol measurements. Atmos. Environ.,34:1959–1999.

McMurry, P., et al. (2005) A criterion for new par-ticle formation in the sulfur-rich Atlanta atmo-sphere. J. Geophys. Res., 110:Art. No. D22S02.

Merikanto, J., et al. (2009) Impact of nucleation onglobal CCN. Atmos. Chem. Phys., 9(21):8601–8616. doi:10.5194/acp-9-8601-2009.

Mohnen, V. and Hidy, G. M. (2010) Measurementsof atmospheric nanoparticles (1875–1980). Bull.Amer. Meteor. Soc., 91(11):1525–1539. doi:10.1175/2010BAMS2929.1.

Nakicenovic, N., et al. (2000) Special Report onEmissions Scenarios (SRES). IPCC reports.Cambridge University Press, UK.

Napari, I., et al. (2002) Ternary nucleation of in-organic acids, ammonia, and water. J. Chem.Phys, 117:8418–8425.

Nenes, A., et al. (2001) A theoretical analysis ofcloud condensation nucleus (CCN) instruments.J. Geophys. Res., 106(D4):3449–3474.

Nieminen, T., et al. (2009) Connection of sulfuricacid to atmospheric nucleation in boreal forest.Environ. Sci. Technol., 43(13):4715–4721. doi:10.1021/es803152j.

Nilsson, E., et al. (2001) Turbulent aerosol fluxesover the Arctic Ocean 2. wind-driven sourcesfrom the sea. J. of Geophys Res., 106:32139–32154.

Oberdorster, G., et al. (2005) Nanotoxicology: Anemerging discipline evolving from studies of ul-trafine particles. Environ. Health Perspect.,113(7). doi:doi:10.1289/ehp.7339.

O’Donnell, D., et al. (2011) Estimating the di-rect and indirect effects of secondary organicaerosols using ECHAM5-HAM. Atmos. Chem.Phys., 11(16):8635–8659. doi:10.5194/acp-11-8635-2011.

Odum, J., et al. (1997) Aromatics, reformulatedgasoline, and atmospheric organic aerosol forma-tion. Environ. Sci. Technol., 31:1890–1897.

Paasonen, P., et al. (2010) On the roles of sulphuricacid and low-volatility organic vapours in the ini-tial steps of atmospheric new particle formation.Atmos. Chem. Phys., 10(22):11223–11242. doi:10.5194/acp-10-11223-2010.

Paasonen, P., et al. (2012) On the formationof sulphuric acid-amine clusters in varying at-mospheric conditions and its influence on at-mospheric new particle formation. Atmos.Chem. Phys. Discuss., 12(5):11485–11537. doi:10.5194/acpd-12-11485-2012.

Petters, M. D. and Kreidenweis, S. M. (2007) Asingle parameter representation of hygroscopicgrowth and cloud condensation nucleus activ-ity. Atmos. Chem. Phys., 7(8):1961–1971. doi:10.5194/acp-7-1961-2007.

Pierce, J. R. and Adams, P. J. (2007) Efficiency ofcloud condensation nuclei formation from ultra-fine particles. Atmos. Chem. Phys., 7(5):1367–1379. doi:10.5194/acp-7-1367-2007.

Pope, C. A. I. (2000) Review: Epidemiological ba-sis for particulate air pollution health standards.Aerosol Sci. Technol., 32:4–14.

Prisle, N. L., et al. (2012) Surfactant effects inglobal simulations of cloud droplet activation.Geophys. Res. Lett., 39(5):L05802–.

47

Pruppacher, H. and Klett, J. (1997) Microphysicsof Clouds and Precipitation. Kluwer AcademicPublishers, second revised and enlarged edn.

Pryde, J., et al. (2011) Sensory and chemical anal-ysis of ‘Shackletons’ Mackinlay scotch whisky. J.Inst. Brew., 117:156–165.

Pryor, S. C., et al. (2008) A review of measurementand modelling results of particle atmosphere–surface exchange. Tellus B, 60(1):42–75. doi:10.1111/j.1600-0889.2007.00298.x.

Putaud, J.-P., et al. (2004) A European aerosolphenomenology - 2: chemical characteristics ofparticulate matter at kerbside, urban, rural andbackground sites in Europe. Atmos. Environ.,38:2579–2995.

Raes, F., et al. (2000) Formation and cycling ofaerosols in the global troposhere. Atmos. Envi-ron., 34:4215–4240.

Ramanathan, V., et al. (2001a) Aerosols, climateand the hydrological cycle. Science, 294:2119–2124.

Ramanathan, V., et al. (2001b) Indian Ocean Ex-periment: An integrated analysis of the climateforcing and effects of the great Indo-Asian haze.J. Geophys. Res., 106(D22):28371–28398. doi:10.1029/2001JD900133.

Ramanathan, V., et al. (2005) Atmosphericbrown clouds: Impacts on South Asian cli-mate and hydrological cycle. Proc. Natl.Acad. Sci. U. S. A., 102(15):5326–5333. doi:10.1073/pnas.0500656102.

Ramaswamy, V., et al. (2001) Climate Change2001: The Scientific Basis. Contribution ofWorking Group I to the Third Assessment Re-port of the Intergovernmental Panel on Cli-mate Change, chap. Radiative forcing of cli-mate change, pp. 349–416. Cambridge UniversityPress, Cambridge, United Kingdom and NewYork, NY, USA.

Reddington, C. L., et al. (2011) Primary versus sec-ondary contributions to particle number concen-trations in the European boundary layer. At-mos. Chem. Phys., 11(23):12007–12036. doi:10.5194/acp-11-12007-2011.

Riemer, N., et al. (2010) Estimating black carbonaging time-scales with a particle-resolved aerosolmodel. J. Aerosol Sci., 41(1):143 – 158. doi:10.1016/j.jaerosci.2009.08.009.

Riipinen, I., et al. (2012) The contribution of organ-ics to atmospheric nanoparticle growth. NatureGeosci., 5(7):453–458.

Robinson, A. L., et al. (2007) Rethinking organicaerosols: Semivolatile emissions and photochem-ical aging. Science, 315(5816):1259–1262. doi:10.1126/science.1133061.

Robock, A. (2008) 20 reasons why geoengineeringmay be a bad idea. Bulletin of the Atomic Sci-entists, 64(14):14–18. doi:10.2968/064002006.

Roeckner, E., et al. (2003) The atmospheric generalcirculation model ECHAM5. PART I: Model de-scription. MPI-Report, 349:127 pp.

Sager, T. and Castranova, V. (2009) Surface areaof particle administered versus mass in deter-mining the pulmonary toxicity of ultrafine andfine carbon black: comparison to ultrafine tita-nium dioxide. Part. Fibre Toxicol., 6(1):15. doi:10.1186/1743-8977-6-15.

Sanchez-Lorenzo, A., et al. (2012) Assessing large-scale weekly cycles in meteorological variables: areview. Atmos. Chem. Phys., 12(13):5755–5771.doi:10.5194/acp-12-5755-2012.

Seinfeld, J. and Pandis, S. (2006) Atmosphericchemistry and physics: From air pollution to cli-mate change. Wiley Interscience, 2nd edn.

Sen, P. K. (1968) Esimates of the regression coef-ficient based on Kendall’s tau. J. Am. Statist.Assoc., 63(324):1379–1389.

Shepherd, J., et al. (2009) Geoengineering the cli-mate: science, governance and uncertainty. RSPolicy document. The Royal Society, London,UK.

Sihto, S.-L., et al. (2006) Atmospheric sulphuricacid and aerosol formation: implications fromatmospheric measurements for nucleation andearly growth mechanisms. Atmos. Chem. Phys.,6:4079–4091.

Sihto, S.-L., et al. (2011) Seasonal variation ofCCN concentrations and aerosol activation prop-erties in boreal forest. Atmos. Chem. Phys.,11(24):13269–13285. doi:10.5194/acp-11-13269-2011.

Sipila, M., et al. (2010) The role of sul-furic acid in atmospheric nucleation.Science, 327(5970):1243–1246. doi:10.1126/science.1180315.

Smith, S. J., et al. (2004) Historical sulfur diox-ide emissions 1850–2000: Methods and results.PNNL Research Report PNNL-14537, PacificNorthwest National Laboratory, Joint GlobalChange Research Institute, 8400 Baltimore Av-enue, College Park, Maryland 20740.

48

Smith, S. J., et al. (2011) Anthropogenic sulfurdioxide emissions: 1850–2005. Atmos. Chem.Phys., 11(3):1101–1116. doi:10.5194/acp-11-1101-2011.

Spracklen, D. V., et al. (2006) The contributionof boundary layer nucleation events to totalparticle concentrations on regional and globalscales. Atmos. Chem. Phys., 6(12):5631–5648.doi:10.5194/acp-6-5631-2006.

Spracklen, D. V., et al. (2010) Explaining globalsurface aerosol number concentrations in termsof primary emissions and particle formation.Atmos. Chem. Phys., 10(10):4775–4793. doi:10.5194/acp-10-4775-2010.

Stier, P., et al. (2005) The aerosol-climate modelECHAM5-HAM. Atmos. Chem. Phys., 5:1125–1156.

Struthers, H., et al. (2011) The effect of sea ice losson sea salt aerosol concentrations and the radia-tive balance in the Arctic. Atmos. Chem. Phys.,11(7):3459–3477.

Su, H., et al. (2011) Soil nitrite as a sourceof atmospheric HONO and OH radi-cals. Science, 333(6049):1616–1618. doi:10.1126/science.1207687.

Textor, C., et al. (2007) The effect of harmonizedemissions on aerosol properties in global modelsan AeroCom experiment. Atmos. Chem. Phys.,7(17):4489–4501. doi:10.5194/acp-7-4489-2007.

Torrence, C. and Compo, P. (1998) A practicalguide to wavelet analysis. Bull. Am. Meteorol.Soc., 79:61–78.

Tost, H. and Pringle, K. J. (2012) Improvementsof organic aerosol representations and their ef-fects in large-scale atmospheric models. At-mos. Chem. Phys., 12(18):8687–8709. doi:10.5194/acp-12-8687-2012.

Tsigaridis, K. and Kanakidou, M. (2003) Globalmodelling of secondary organic aerosol in thetroposphere: sensitivity analysis. Atm. Chem.Phys., 3:1849–1869.

Twomey, S. (1977) The influence of pollution onthe shortwave albedo of clouds. J. Atmos. Sci.,34:1149–1152.

van Dingenen, R., et al. (2004) A European aerosolphenomenology - I: physical characteristics ofparticulate matter at kerbside, urban, rural andbackground sites in Europe. Atmos. Environ.,38:2561–2577.

van Vuuren, D. P., et al. (2011) The representa-tive concentration pathways: an overview. Clim.Change, 109:5–31. doi:10.1007/s10584-011-0148-z.

Vanhanen, J., et al. (2011) Particle size magni-fier for nano-CN detection. Aerosol Sci. Tech.,45(4):533–542.

Vehkamaki, H., et al. (2002) An improved parame-terization for sulfuric acid/water nucleation ratesfor tropospheric and stratospheric conditions. J.Geophys. Res., 107:4622–4631.

Vignati, E., et al. (2004) M7: An efficient size-resolved aerosol microphysics module for large-scale aerosol transport models. J. Geophys. Res.,109(D22202). doi:10.1029/2003JD004485.

Virkkula, A., et al. (2011) Seasonal cycle, sizedependencies, and source analyses of aerosoloptical properties at the SMEAR II measure-ment station in Hyytiala, Finland. Atmos.Chem. Phys., 11(9):4445–4468. doi:10.5194/acp-11-4445-2011.

Wang, J., et al. (2008) Sensitivity of sulfate directclimate forcing to the hysteresis of particle phasetransitions. J. Geophys. Res., 113(D11):D11207.

Wang, S., et al. (2003) Turbulence, condensation,and liquid water transport in numerically simu-lated nonprecipitating stratocumulus clouds. J.Atmos. Sci., 60(2):262–278. doi:10.1175/1520-0469(2003)060¡0262:TCALWT¿2.0.CO;2.

Wang, X. L. and Swail, V. (2001) Changesof extreme wave heights in northern hemi-sphere oceans and related atmospheric circula-tion regimes. J. Climate, 14:2204–2221.

Weber, R. J., et al. (1996) Measured atmosphericnew particle formation rates: implications fornucleation mechanisms. Chem. Eng. Comm,151:53–64.

Wiedensohler, A., et al. (2012) Mobility particlesize spectrometers: harmonization of technicalstandards and data structure to facilitate highquality long-term observations of atmosphericparticle number size distributions. Atmos. Meas.Tech., 5(3):657–685. doi:10.5194/amt-5-657-2012.

Wilks, D. S. (1997) Resampling hypothesis tests forautocorrelated fields. J. Climate, 10:65–81.

Wittmaack, K. (2007) In search of the most rele-vant parameter for quantifying lung inflamma-tory response to nanoparticle exposure: Particlenumber, surface area, or what? Environ. HealthPerspect., 115(2). doi:doi:10.1289/ehp.8871.

49

Yin, Y., et al. (2002) Interactions of mineral dustparticles and clouds: Effects on precipitationand cloud optical properties. J. Geophys. Res.,107(D23):4724–.

Yli-Juuti, T., et al. (2011) Growth rates of nucle-ation mode particles in Hyytiala during 2003-2009: variation with particle size, season, dataanalysis method and ambient conditions. At-mos. Chem. Phys., 11(24):12865–12886. doi:10.5194/acp-11-12865-2011.

Yu, F. and Turco, R. P. (2000) Ultrafine aerosol

formation via ion-mediated nucleation. Geophys.Res. Lett., 27:883–886.

Zabori, J., et al. (2012a) Artificial primary ma-rine aerosol production: a laboratory study withvarying water temperature, salinity, and suc-cinic acid concentration. Atmos. Chem. Phys.,12(22):10709–10724.

Zabori, J., et al. (2012b) Wintertime Arctic Oceansea water properties and primary marine aerosolconcentrations. Atmos. Chem. Phys., 12:10405–10421. doi:10.5194/acp-12-10405-2012.

50