You are on page 1of 16

ARTICLE IN PRESS

Journal of Aerosol Science 41 (2010) 621–636

Contents lists available at ScienceDirect

Journal of Aerosol Science


journal homepage: www.elsevier.com/locate/jaerosci

Parameterizing the formation rate of new particles: The effect


of nuclei self-coagulation
Tatu Anttila a,, Veli-Matti Kerminen a, Kari E.J. Lehtinen b,c
a
Finnish Meteorological Institute, P.O. Box 503, FI-00101 Helsinki, Finland
b
University of Kuopio, Department of Physics, P.O. Box 1627, 70211 Kuopio, Finland
c
Finnish Meteorological Institute, Kuopio Unit, P.O. Box 1627, 70211 Kuopio, Finland

a r t i c l e in fo abstract

Article history: The study is based on the work of Lehtinen et al. (2007) [Lehtinen, K. E. J., Dal Maso, M.,
Received 11 June 2009 Kulmala, M., & Kerminen, V.-M. (2007). Estimating nucleation rates from apparent
Received in revised form particle formation rates and vice versa: Revised formulation of the Kerminen–Kulmala
20 April 2010
equation. Journal of Aerosol Science, 38, 988–994] who derived formulae connecting
Accepted 20 April 2010
‘‘real’’ and ‘‘apparent’’ nucleation rates. The parameterization neglected self-coagulation
of newly formed particles and clusters, however, and here we have extended the
Keywords: previous work to include the effects of the self-coagulation. Our main focus was on
Aerosol dynamics calculating the ‘‘apparent’’ nucleation rate, i.e. the rate at which particles appear at sizes
New particle formation
larger than the critical cluster size, as a function of the ‘‘real’’ nucleation rate. The
Aerosol modeling
revised parameterization was comprehensively tested against an explicit aerosol
Atmospheric aerosols
dynamic model at diverse atmospheric conditions. It was found out that nuclei self-
coagulation has importance to new particle formation when Jnuc/Q 410  2 where Jnuc is
the nucleation rate and Q is the production rate of condensable vapours. This
corresponds to the nucleation rates ranging from 410 cm  3 s  1 (free troposphere) to
4 104 cm  3 s  1 (polluted boundary layer) depending on the atmospheric conditions. In
terms of the particle number concentration, the calculations performed with the explicit
model and the predictions of revised parameterization were generally within an order
of magnitude. Several issues related to applications in large-scale models were also
discussed.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Atmospheric aerosol formation by nucleation has been observed to take place frequently in a large number of locations
in the Earth’s atmosphere (Kulmala et al., 2004; Kulmala & Kerminen, 2008, and references therein). Over regional scales,
nucleation may be responsible for a large, even dominant, fraction of the total particle number budget (e.g. Hamed et al.,
2007; Tunved, Hansson, Kerminen, Ström, & Dal Maso, 2006). Once they will grow in size, particles nucleated in the
atmosphere may further contribute to cloud condensation nuclei (CCN) populations (Laaksonen et al., 2005; Lihavainen
et al., 2003; O’Dowd et al., 1999; Wiedensohler et al., 2009) and influence climatically important cloud properties
(Kerminen, Lihavainen, Komppula, Viisanen, & Kulmala, 2005).

 Corresponding author. Tel.: + 358 9 1929 5526; fax: + 358 9 179 581.
E-mail address: tatu.anttila@fmi.fi (T. Anttila).

0021-8502/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jaerosci.2010.04.008
ARTICLE IN PRESS
622 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

Understanding and quantifying the climatic and other effects caused by atmospheric aerosol formation is probably not
possible without large-scale models that explicitly simulate nucleation processes (Ghan & Schwartz, 2007). During the
recent years, a number of global model studies related to atmospheric nucleation have been conducted (Chang, Schwartz,
McGraw, & Lewis, 2009; Lucas & Arimoto, 2006; Makkonen et al., 2009; Pierce & Adams, 2009; Spracklen et al., 2006, 2008;
Yu, Wang, Luo, & Turco, 2008; Wang & Penner, 2009). All of these models rely on ‘‘nucleation parameterizations’’, i.e.
simplified representations of the considered nucleation processes. Models simulating the dynamics of nucleated particles
need also ‘‘aerosol formation parameterizations’’, in which the nucleation rate is converted into the formation rate of
particles at the lower edge of modeled size domain, typically a few nanometers in particle diameter.
In principle, the ‘‘aerosol formation parameterization’’ should include all the processes affecting the fate and size of
nucleated particles, such as their condensational growth, their scavenging onto larger pre-existing particles, their dry
deposition, their atmospheric transportation (e.g. dilution), heterogeneous reactions and nuclei self-coagulation
(Kerminen, Lehtinen, Anttila, & Kulmala, 2004a; McMurry et al., 2005; Pierce & Adams, 2007). In practice, the existing
‘‘aerosol formation parameterizations’’ take into account only the first two of these processes and assume a steady state
(Kazil & Lovejoy, 2007; Kerminen & Kulmala, 2002; Kerminen et al., 2004a; Lehtinen, Dal Maso, Kulmala, & Kerminen,
2007). These are reasonable approximations when simulating so-called regional nucleation events that occur over the
spatial scales of tens to hundreds of kilometers (Kerminen & Kulmala, 2002). The main drawback of the existing
parameterizations is, however, that they lack nuclei self-coagulation which may become important at high nucleation
rates.
In the current paper, we will derive a new ‘‘aerosol formation parameterization’’ that includes nuclei self-coagulation.
The formulation of our parameterization follows the work by Lehtinen et al. (2007), with two additional correction factors
that are solved using an iterative approach. We will investigate the performance of the new parameterization in three
different environments: rural/continental boundary layer, polluted boundary layer, and free troposphere. In this regard, we
will quantify the uncertainties related to the new parameterization and identify the parameter space at which the original
parameterization by Lehtinen et al. (2007) should be replaced with the new one. Finally, we will discuss on various issues
related to the use of the new parameterization in large-scale modeling applications.

2. Methods

2.1. Revised parameterization for new particle formation

The parameterization described in the papers of Kerminen and Kulmala (2002) (hereinafter referred as KK2002) and
Lehtinen et al. (2007) (hereinafter referred as L2007) was extended to account for the effects of self-coagulation of freshly
formed particles. In the above referred works, the authors derived expressions for the formation rate of particles, having
initially a diameter d1, at a diameter dx, J(dx), where dx 4d1 (Eq. (7) in L2007 and Eq. (13) in KK2002). The expressions were
derived on the basis of a solution to the General Dynamic Equation for aerosols (Seinfeld & Pandis, 1998). However, similar
analytical treatment of the considered problem is extremely difficult or even impossible if the self-coagulation of the
newly formed particles is also considered. Therefore we adopted a semi-empirical approach for deriving ‘‘corrected’’
formation rate Jrev(dx) that is based on the previously derived expression for J(dx).
Self-coagulation of freshly formed particles has two competitive effects: enhanced coagulational loss and increased
particle growth rate. We first note that the magnitudes of these effects depend on the number of particles in the size range
[d1, dx], Nnuc. We will postpone the discussion on how to calculate Nnuc, and show how the discussed effects are treated
assuming that Nnuc is known.
First, let us consider the loss of particles due to self-coagulation in the size range [d1, dx]. Assuming that the size-
dependency of the coagulation kernel can be neglected, the sink due to self-coagulation of particles having a size d1,
CoagSscg(d1), can be then expressed as follows:
CoagSscg ðd1 Þ ¼ Keff Nnuc ð1Þ

where Keff (cm3 s  1) is an ‘‘effective’’ coagulation kernel. Consequently, the total coagulation sink CoagStot(d1) for particles
with a diameter d1 is given by the sum: CoagStot(d1) =CoagS(d1)+CoagSscg(d1) where the first term in the right hand side
represents loss through coagulation with background particles and is calculated as in L2007. The parameter Keff is treated
here as a free parameter of which value is adjusted to obtain optimal accuracy.
Enhanced particle growth driven by self-coagulation has a positive impact on the rate at which newly formed particles
reach larger sizes. Here we treat the growth due to self-coagulation in a manner analogous to condensation at the kinetic
regime. As in the previous approach, the particle growth rate is assumed to be size-independent. Accordingly, GRscg (in nm/h)
is proportional to Nnuc and can be thus calculated as follows (KK2002):

3  106
GRscg ¼ l c nuc Mnuc Nnuc ð2aÞ
rnuc
p
Mnuc ¼ d31 rnuc NA ð2bÞ
6
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 623

Here l is a dimensionless parameter of which value is adjusted to reach optimal accuracy as will be described in Section 4.
Also, rnuc (g m  3), cnuc (m/s), Mnuc (g/mol), and d1 (m) are the density, molecular velocity, molecular weight, and the initial
diameter of the clusters, respectively, and NA is the Avogadro’s constant (=6.0221415  1023 mol  1). Here Mnuc is calculated
under the assumption that the clusters can be treated as spherical objects with density equal to the bulk density of the
nucleating vapour. Combining (2a) and (2b) we obtain the following expression for GRscg:

GRscg ¼ 1:57  106 ld31 c nuc Nnuc NA ð3Þ


The total growth rate of the freshly nucleated particles, GRtot, is the sum of the growth rates due to condensation, GRcond,
and self-coagulation: GRtot = GRcond + GRscg where GRcond is calculated as in the original formulation of the parameterization
(KK2002).
As seen from Eqs. (1) and (3), both GRtot and CoagStot depend on Nnuc. However, Nnuc in turn depends on the growth rate
and coagulation sink of freshly nucleated particles. Therefore Nnuc is calculated iteratively as described below.
The iterative scheme is illustrated in Fig. 1. The iteration procedure is initialized by setting Nnuc = J(d1)/CoagS(d1) which is
the steady state number concentration of stable clusters/particles having a diameter d1 assuming that the only sink of the
clusters/particles is coagulation with background aerosols. In the first step, CoagSscg(d1) and GRscg are calculated using
Eqs. (1) and (3), respectively, after which CoagStot(d1) and GRtot are updated using the following formulae:
CoagStot ðd1 Þ ¼ CoagSðd1 Þ þ CoagSscg ðd1 Þ and GRtot ¼ GRcond þ GRscg ð4Þ
Next, we note that we can use the previously developed equation for calculating the formation rate of particles having
the diameter dp, d1 odp odx, with the above-presented expressions for the growth rate and coagulation sink:
 
CoagStot ðd1 Þ
Jrev ðdp Þ ¼ Jðd1 Þexp gd1 ð5aÞ
GRtot
" m þ 1 #
1 dp
g¼ 1 ð5bÞ
mþ1 d1

Eq. (5a) is based on Eq. (7) in L2007. The difference is that the terms representing the particle growth and coagulational
scavenging are replaced by the corresponding modified terms GRtot and CoagStot. The value of the parameter m is calculated
using Eq. (6) in L2007.
The reader is reminded that Jrev(dp) =(dN/ddp) GRtot by definition (see L2007). Using this expression, we obtain the
following:
Z dx
1
Nnuc ¼ Jrev ðdp Þ ddp ð6Þ
GRtot d1

where Jrev(dp) is given by (5). By expanding the integrand to Taylor series using (5), a series expansion for Nnuc can be
derived:
Jðd1 Þ X1
Zn  
Nnuc ¼ expðZd1m þ 1 Þ dxnðm þ 1Þ þ 1 dnðm
1
þ 1Þ þ 1
ð7aÞ
GRtot n¼0
½nðm þ1Þ þ1n!

Fig. 1. Schematic figure illustrating the iterative method for calculating the new particle formation rate.
ARTICLE IN PRESS
624 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

where
CoagStot ðd1 Þ
Z¼ ð7bÞ
ðm þ1ÞGRtot dm 1

The expansion is valid, e.g. for the value ranges  4/34m4 3/2 and 3/24m 4  2. Here it should be noted that the
value of m is expected to lie in the range ] 1.5,  2.0[ for accumulation mode mean diameters 80 nm 1 mm (L2007). As
will be seen later, the value of m stayed in this range also in all the simulations performed with the aerosol dynamic model,
and therefore the expansion is expected to be valid for sufficiently large range of conditions. Alternatively, Nnuc can be
calculated through a numerical integration.
After Nnuc has been calculated on the basis of (5) and (6), the iteration is started over using the updated value of Nnuc.
The iteration is repeated until the desired degree of convergence has been reached, and then Jrev(dx) is calculated using (5a)
and (5b). Furthermore, as seen from Eqs. (1), (3) and (4), GRtot = GRcond and CoagStot(d1)= CoagSscg(d1) when Nnuc =0. Hence at
the limit where the nuclei self-coagulation is negligible, Jrev(dx) is equal to J(dx) (Eq. (5)) which demonstrates that the
developed scheme is indeed an extension of the original formulation of the parameterization.

2.2. Aerosol dynamic model

A box model for simulating aerosol dynamics was adopted in order to test the performance of the developed
parameterization and to find optimal values for the free parameters l and Keff. The applied model covers the key aerosol
dynamic processes behind new particle formation (nucleation, condensation, and coagulation) and it is based on the
modeling works of Anttila, Kerminen, Kulmala, Laaksonen, and O’Dowd (2004) and Anttila and Kerminen (2008). The main
difference between the model applied here and that in the above-referred studies is that the time evolution of the particles
is simulated with the hybrid model structure (Jacobson & Turco, 1995) rather than using the moving sectional method. The
performance of the revised version was tested and the test simulations showed a good agreement with the previous
version (not illustrated here). In the simulations described here, particles were distributed into 100 sections that span the
diameter range 1 nm–1 mm and the time step was chosen to be 5 s.
The nucleation was simulated by inserting new clusters to the size bin containing the diameter d1 with a constant rate
that is determined by J(d1). The physico-chemical properties were assumed to be the same as the condensable vapour X
(see below). Some of the model variables were kept constant in order to avoid time dependency of key dynamic parameters
because the parameterizations tested here are based on assuming constant values for the parameters. Hence unaccounted
dynamical features would hamper the interpretation of the results. To this end, the parameter J(d1) was kept constant
during a simulation. Regarding condensation, background particles were not allowed to grow as a response to
condensation, and the gas-phase concentration of the condensable vapour X, CX, was set equal to the steady state value:
CX = Q/CondS. Here Q is the gas-phase production rate of X and CondS is the condensation sink of X (see e.g. KK2001). Only
loss due to condensation onto background particles was accounted for in calculating CondS, and Q did not have any time
dependency. Furthermore, the background aerosol was held fixed with respect to coagulation. This allowed for optimizing
the code by calculating the coagulation rates using a matrix that contained the coagulation sink values for all particle bin
pairs. The matrix was calculated in the beginning of a model run once the needed input parameter values were specified.
As a result of these modifications, also CX was constant during a simulation.
The modeling scheme included a semi-volatile vapour, water, and a non-volatile vapour, X. Regarding X, it was treated
as in the above-referred modeling works, and an ideal behavior was assumed in modeling the hygroscopic properties of X.
The meteorological variables in the model, temperature (T) and relative humidity (RH), were kept constant in a simulation.

3. Performed simulations

The simulations considered here consist of three sets of model runs that represent atmospheric conditions in rural/
continental background boundary layer, polluted boundary layer and free troposphere (FT). Taken together, they cover a
large portion of the atmospheric conditions where new particle formation (NPF) has been observed (Kulmala et al., 2004).
For each set, the values of input parameters in model runs were varied in Monte Carlo fashion. Each parameter of which
value varied less than an order of magnitude was assumed to be uniformly distributed, and the logarithms of those
parameters with larger value ranges were assumed to follow a uniform distribution. Furthermore, each parameter was
varied independently of each other. These assumptions are simplified as compared to the conditions in the atmosphere but
they are necessitated by the lack of data on the distribution of the parameter values at various ambient conditions.
Similarly, correlations between different parameters are poorly known (Anttila & Kerminen, 2008). Finally, the total
number of simulations performed was 750 in each scenario.
The initial size distributions consisted of log-normally distributed Aitken and accumulation modes. The distribution
parameters were varied so that they cover ranges that are typical for the considered atmospheric environment. The value
ranges are shown in Table 1 along with the references. The other varied parameters and their value ranges are shown in
Table 2. Regarding the key parameters behind new formation, Q and J(d1), the value ranges were based on the studies of
Andronache et al. (1997) and Weber et al. (1999) [free troposphere]; Dal Maso et al. (2005) and Kulmala et al. (2004) [rural/
continental background, polluted] and Kulmala et al. (2004), Hussein et al. (2008) and Murugavel and Chate (2009)
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 625

Table 1
The value rangers of the particle size distribution for the considered atmospheric scenarios. Here Ntot is the total particle number concentration in a mode,
Dg is the geometric mean diameter of a mode, and sg is the geometric standard deviation of a mode.

Aitken Accumulation Reference(s)

3
Ntot (cm ) Dg (nm) rg Ntot Dg (nm) rg
(cm  3)

Free troposphere 200–750 50–75 1.4–1.7 40–700 100–150 1.3–1.6 Schröder, Kärcher, Fiebig, and
Petzold (2002)
Background/rural 100–4000 50–100 1.3–1.7 100–1000 100–250 1.3–1.7 Tunved et al. (2003), Birmili,
Wiedensohler, Heintzenberg, &
Lehmann (2001)
Polluted 1000–50,000 50–100 1.3–1.7 1000–40,000 100–250 1.3–1.7 Wehner and Wiedensohler
(2003), Mönkkönen et al. (2005)

Table 2
Value ranges of the varied input parameters for the considered atmospheric scenarios. Here t is the duration of a simulation, DX is the gas-phase diffusion
coefficient of the vapour X, MWX is its molecular weight and rX is its density. All the other parameters are indicated in the text.

Free troposphere Background/rural Polluted

J(d1) (cm  3 s  1) 10  2–100 10  2–103 1–104


Q (cm  3 s  1) 500–2  104 104–2  105 105–5  106
t (h) 3–12 3–12 3–12
T (K) 238–288 288–308 288–308
DX (cm2 s  1) 0.8–1.2 0.8–1.2 0.8–1.2
MWX (g mol  1) 90–270 90–270 90–270
rX (g cm  3) 0.9–2.0 0.9–2.0 0.9–2.0
dx (nm) 3–5 3–5 3–5
d1 (nm) 1–2 1–2 1–2

[polluted]. It is worth noting that Q is always larger than J(d1); the reason for this is that any nucleating vapour would
condense also onto the freshly formed clusters and particles and hence the vapour concentration would not be in the
steady state if J(d1)4Q but the vapour would deplete from the gas phase.
The value of d1 was varied according to the current knowledge on the diameter of thermodynamically stable clusters
(Kulmala et al., 2005). As further seen, the value of dx was varied over a relatively narrow range that reflect the lower
detection limits of the current aerosol sizing instrumentation (McMurry, 2000). The temperature was varied as well, but
relative humidity was kept constant at 0.4. The latter was done because the main effect of varying RH in this modeling
context is to modify the condensation and coagulation sinks which is taken into account here through varying the size
distribution of background particles. Additionally, the physico-chemical properties of the non-volatile vapour X were
varied (Table 2) because of the current uncertainties regarding the identity of the vapours behind new particle formation.
The value ranges applied here reflect the current knowledge on low-volatile compounds that are formed in the gas phase
and that may contribute to NPF through gas-to-particle conversion (Kanakidou et al., ACP, 2005).

4. Results

The central model output for our purposes are the particle formation rates and the total number of particles formed, i.e.
the number of particles which have exceeded the size range 4dx by a time instant t. The latter quantity is denoted as
Npara,prev and Npara,new for the original and revised versions of the parameterization, respectively, and they were calculated
as follows:
Npara,prev ¼ Jðdx Þ Dt1 , Npara,new ¼ Jrev ðdx Þ Dt2 ð8aÞ
where
   
dx d1 dx d1
Dt1 ¼ max 0,t , Dt2 ¼ max 0,t ð8bÞ
GRcond GRtot
Here Dti is the time interval during which particles having sizes 4dx are predicted to be formed. Regarding the explicit
model, the particle formation rate, Jmodel, was calculated by tracking how many particle grow above the size range 4dx due
to coagulation and condensation over a certain time interval. Based on this, the particle formation rate can be readily
determined which was done over a time interval of 1200 s. The time interval is chosen to be relatively large compared to
the model time step (5 s) to eliminate random fluctuations in the results. The value of Nmodel is then calculated by
integrating the formation rate over time. Here it should be further noted that Nmodel is not the same quantity as the number
concentration of new particles at the time t, Ntot, because Nmodel is not influenced by coagulational loss of particles at the
ARTICLE IN PRESS
626 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

size range 4dx. However, because the parameterizations predict the number of particles formed, we based our comparison
on this quantity rather than on the simulated particle concentrations.
New particle formation (NPF) is defined to have taken place if Nmodel, Npara,prev or Npara,new exceeds 10% of the number
concentration of background particles at the end of a simulation. Based on a larger set of simulations (not shown here), the
chosen values for the free parameters in the revised parameterization, l and Keff, are 6 and 5  10  10 cm3 s  1, respectively,
unless otherwise mentioned. Unless otherwise mentioned, the calculations presented below are performed using these
values of l and Keff. Finally, Nnuc was calculated based on Eqs. (5)–(7). In order to ensure maximum numerical accuracy, 30
first terms were accounted for when calculating the series expansion in (7a) and the number of iterations was five.
We first illustrate the time development of the particle formation rates and number concentrations in individual model
simulations (Section 4.1.) after which we present comprehensive evaluation of the two considered parameterizations
(Section 4.2).

4.1. Individual model simulations

Here we focus on two modeling runs that belong to the set of simulations for the continental background/rural scenario.
NPF took place in both of the simulations. The first simulation (‘‘A’’) represents a case where self-coagulation did not
influence NPF whereas the second simulation (‘‘B’’) represents the opposite case. The time development of the new particle
formation rates and the total number of particles formed for both simulations are shown in Fig. 2 along with the values of
the key input parameters. Both of the simulations lasted around 11 h, and results for the first 11 h are displayed.
The upper panel of Fig. 2 shows the results for simulation ‘‘A’’. As seen, new particles started to appear after around 1 h
from the start of the simulation, and new particles were produced throughout the simulation such that around

Fig. 2. Time development of the total number of particles formed (left-hand side panels) and the new particle formation rates (right-hand side panels) for
two simulations, denoted as ‘‘A’’ and ‘‘B’’. The results for simulation ‘‘A’’ are on the top, and those for simulation ‘‘B’’ are on the bottom. Results for the
model and both of the tested parameterizations are shown, as indicated in the legend. The values for the key input parameters are also indicated. Please
further note that lines illustrating results for the parameterizations overlap in the case of simulation ‘‘A’’.
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 627

7  104 cm  3 particles were formed by the end of the simulation. The lines illustrating time development of Npara,prev and
Npara,new overlap, indicating that self-coagulation did not influence NPF. Furthermore, Npara,prev and Npara,new both agree well
with Nmodel throughout the simulation. As seen from the lower panel, the parameterized particle formation rates are time-
independent due to the steady-state assumption underlying the parameterizations. In contrast, the modeled new particle
formation rates exhibited rather complicated time dependence: the rate increased rapidly at the beginning of the
simulation after which it slowed down and reached maximum at around 7 h. Afterwards the new particle formation rate
decreased slightly. The behavior reflects the changing concentration of particles/clusters at the size range odx: the
concentration increases until the loss due to the combined effects of the growth to the size range 4dx and coagulational
scavenging is larger than the production rate of new clusters through nucleation.
The lower panel of Fig. 2 illustrates the results for simulation ‘‘B’’. According to the model and the revised
parameterization, NPF starts after around 1, 5, and 3 h, respectively whereas the original formulation of the
parameterization predicts onset of NPF after around 9 h. Despite the lag in the predicted onset of NPF, Nmodel and Npara,new
are within a factor of three after 4 h of the simulation and difference decreases throughout the model run. In contrast, the
original formulation fails to capture the modeled NPF and underestimates the number of new particles formed by more
than one order of magnitude by the end of the model run. The results for the new particle formation rates show that the
modeled rate increased throughout the simulation which is caused by slow accumulation of clusters/particles at the size
range odx. The accumulation is driven by relatively large nucleation rate. The parameterized formation rates differ by
around an order of magnitude which further highlights the role of self-coagulation in NPF. The question under which
conditions self-coagulation is important for NPF will be addressed in the next section, but here we would like to note that a
crucial difference between the two simulations considered here is the relative magnitudes of Jnuc and Q: in simulation ‘‘B’’,
the nucleation rate was relatively high compared to the production rate of the condensable vapour whereas the case was
opposite in simulation ‘‘A’’.

4.2. Evaluation of the parameterization

Our comparison of the performance of the revised parameterization is based on the number of particles formed during
the simulated time interval rather than on the formation rates. This is because of the following reasons. First, as illustrated
in the previous section, the simulated formation rates did often display complicated time dependent behavior.
Furthermore, the number of particles formed during a certain time interval is a more important output in many modeling
applications than the formation rate. Along with Nmodel, we also consider Ntot,model which is the sum of Nmodel and the
number of background particles, Nbgr. The corresponding quantities for the parameterizations are Npara,prev +Nbgr and
Npara,new +Nbgr. We consider these total particle concentrations because it facilitates comparison of the results in cases
where no new particles are predicted to be formed either by a parameterization or by the model.

4.2.1. Free troposphere


PF was predicted to take place in 351 simulations for the FT scenario, and the frequency of NPF events was thus 0.35
(Table 3). The corresponding frequencies were 0.26 and 0.31 for the original and revised parameterizations, respectively.
These frequencies compared favorably with the study of Venzac et al. (2008) who reported that NPF took place in 435% of
days at a high altitude site in the Himalayas. Median and mean values of Q, J(d1), and CoagS(d1) in the calculations are
consistent with empirical information on the conditions in the upper troposphere (Andronache et al., 1997; Clarke et al.,

Table 3
Statistics of the model runs with NPF (predicted either by the parameterization(s) or the explicit model) for the three considered scenarios.

J(d1) Q CoagS (d1) m GRcond GRtot Nmodel Npara,prev Npara,new


(cm  3 s  1) (cm  3 s  1) (s  1) (nm/h) (nm/h) (cm  3) (cm  3) (cm  3)

Free troposphere
Median 5.01 7.99E+ 03 2.51E  04  1.59 0.50 0.58 1.90E+ 03 6.51E + 02 1.98E + 03
Mean 18.81 8.69E+ 03 3.01E  04  1.59 0.64 0.73 4.88E+ 04 3.36E + 04 4.31E + 04
Min 0.01 5.03E+ 02 6.74E  05  1.69 0.02 0.10 0 0 0
Max 99.20 2.00E + 04 1.49E  03  1.52 3.42 3.44 1.29E+ 06 1.14E + 06 9.69E + 05

Rural/continental background
Median 1.81E+ 01 3.75E+ 04 6.37E  04  1.68 1.06 1.36 2.15E+ 04 2.09E + 03 1.02E +04
Mean 1.28E+ 02 5.78E+ 04 7.79E  04  1.68 1.86 2.09 1.19E+ 06 5.92E + 05 6.21E + 05
Min 1.01E-02 1.03E+ 03 9.44E  05  1.80 0.02 0.02 1.30E+ 01 0 0
Max 9.27E+ 02 1.98E+ 05 3.85E  03  1.56 14.90 14.90 2.27E+ 07 1.97E + 07 1.25E + 07

Polluted boundary layer


Median 3.11E+ 02 1.13E+ 06 6.17E  03  1.70 5.46 5.61 2.37E+ 04 3.03E + 04 3.70E +04
Mean 1.61E+ 03 1.62E+ 06 7.07E  03  1.70 6.92 7.29 2.51E+ 06 2.33E + 06 2.40E +06
Min 1.02E+ 00 5.09E+ 04 1.35E  03  1.83 0.49 0.49 80 0 0
Max 9.34E+ 03 4.99E+ 06 2.81E  02  1.57 28.70 28.70 1.13E+ 08 1.11E + 08 8.29E + 07
ARTICLE IN PRESS
628 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

1999). The calculated growth rates are typically around  1 nm/h which agrees with the measurements of Venzac et al.
(2008). As seen from Table 3, particle growth due to self-coagulation is not significant, GRscg making 12% contribution to
the total growth rates on average. Furthermore, the median and average values of Nmodel were 1.9  103 cm  3 and 4.9  104,
respectively. Even though the average value is higher than most of the available measurement data suggest, the median
value compares well with observations on NPF taking place at the upper or free troposphere (Lee et al., 2003; Venzac et al.,
2008; Weber et al., 1999).
Fig. 3 shows comparisons of Ntot,model with Npara,new + Nbgr (above) and Npara,prev +Nbgr (below) for the FT scenario. The
overall correlation is reasonably good with both parameterizations, the coefficient of determination being 0.75 for both of
the parameterizations. Another apparent feature is that the performance of the L2007 parameterization depends strongly
on the ratio J(d1)/Q: the L2007 underestimates Nmodel and hence also Ntot,model even by several orders of magnitude when
the values of J(d1) and Q are within a few orders of magnitude. This is explained by the fact that self-coagulation of freshly
formed particles becomes relatively more important to the particle growth as compared condensation with increasing
value of J(d1)/Q. A comparison of the calculations for the original and revised parameterization revealed that the net effect
of self-coagulation is to promote NPF rather than suppress it: Npara,new is consistently larger or equal to Npara,prev (not
illustrated here). Loss of newly formed particles due to self-coagulation has thus secondary importance as compared to
coagulational scavenging on background particles. The feature can be understood by noting that the coagulation sink due
to growing newly formed particles, CoagSscg, was consistently smaller compared to the sink due to background particles,
CoagS. In fact, it was calculated that the ratio CoagSscg/CoagStot was 0.46 at maximum and 0.10 on average the end of a

Fig. 3. Performance of the original and revised version of the parameterizations, evaluated against the aerosol dynamic model. Top: Modeled total
particle number concentrations (Ntot,model) versus corresponding calculations performed with the original formulation of the parameterization
(Npara,prev +Nbgr). Bottom: Modeled total particle number concentrations versus corresponding calculations performed with the revised formulation of the
parameterization (Npara,new + Nbgr). Results are for the free troposphere scenario and only for simulations with NPF. The lines appearing in figure are to
guide the eye.
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 629

simulation (not shown in Table 3). To conclude with, even though self-coagulation does not have large impact on NPF on
average, it becomes important for NPF when J(d1) and Q are within a few orders of magnitude.
As further seen from Fig. 3, the revised parameterization is capable of taking into account the effects of self-coagulation
such that the parameterized and simulated number of particles formed are within an order of magnitude in most of the
cases. There are still discernible number of cases among the set of simulations with J(d1)/Qo10  2 where the Npara,new + Nbgr
deviates more than an order of magnitude from Ntot,model, however. The reason for this is not clear, but it was calculated
that the simulations consist only around 6% of the total number of simulations with J(d1)/Qo10  2 and therefore they
represent a small fraction of all results. In addition, Table 3 shows that both the median and mean values of Npara,new and
Ntot,model compare well with each other and the same holds for Npara,prev. Further statistics are shown in Table 4 which
contains the average relative biases and errors (compared against of Ntot,model) for Npara,prev +Nbgr and Npara,new +Nbgr. As seen
from the table, the revised scheme slightly worse overall performance compared to the original parameterization in terms
of these metrics. The average relative error is satisfactory for both schemes, being around 100% and 120% for the original
and revised parameterizations, respectively.

4.2.2. Rural/continental background


Table 3 contains statistics of the model runs with NPF for the rural/continental background scenario. Median and mean
values of Q, J(d1), and CoagS(d1) in the simulations are consistent with the parameter values extracted from size
distribution measurements conducted at Nordic measurement stations (Kulmala et al., 2001; Dal Maso et al., 2005, 2007).
Also, median and mean values of the growth rates calculated with the parameterizations, GRcond and GRtot, agree with
empirical information presented in the above-referred studies even though the largest growth rates are somewhat larger
than measurements suggest. The parameter m, which depends largely on the mean diameter of the accumulation mode,
varied in a range typical to aerosol size distributions measured in a boreal forest (Lehtinen et al., 2007).
The NPF event frequency was 0.48 in the simulations whereas the NPF frequencies were 0.40 and 0.34 according to the
original and revised parameterizations, respectively. The calculated frequencies are somewhat higher than annual averages
derived from ambient measurements indicate (e.g. Dal Maso et al., 2007), which is probably explained by the fact the value
ranges for key input parameters such as Q and Jnuc are based on the data extracted from observed NPF events. Therefore
these parameters may have smaller values during days without NPF events which would lead larger overall variability. As
seen from Table 3, the median and average values of Nmodel were 2.2  104 cm  3 and 1.2  106 cm  3, respectively. The
median value agrees with ‘‘typical’’ number of particles formed during a NPF event, but the average value of Nmodel is
consistent with new particle formation rates of up to 100 cm  3 s  1 which is higher than the observed formation rates
(Kulmala et al., 2001). However, it should be kept in mind that Nmodel indicated the number of particles formed during a
simulation rather than the actual number concentration at the end of a model run, and cannot be thus directly compared to
ambient data. Also, high values of J(dx) are probably caused by the representation of the aerosol dynamics in the model
which neglects certain feedbacks such as decrease of Q and J(d1) due to increased condensation sink caused by NPF and
condensation onto the background particles. Because the focus of the current study is on evaluating the performance of the
two parameterizations, however, the tendency to overestimate the number of particles formed is not crucial here.
Fig. 4 shows comparisons of Ntot,model with Npara,new + Nbgr (above) and Npara,prev + Nbgr (below) for the rural/continental
background scenario. The overall correlation is good with both the parameterizations, the coefficient of determination
being 0.83 for the previous parameterization and 0.89 for the revised version. As in the FT scenario, the performance of the
L2007 parameterization is seen to depend strongly on the ratio J(d1)/Q: the L2007 underestimates Nmodel and hence also
Ntot,model up to several orders of magnitude when the values of J(d1) and Q are within a few orders of magnitude. As
explained above, this feature is due to the growth of newly formed particles through self-coagulation which enhances NPF
rather than suppresses it. For the considered scenario, the ratio CoagSscg/CoagStot was 0.53 at maximum and 0.09 on average
at the end of a simulation (not shown in Table 3). A comparison of the values of Npara,prev and Npara,new shows that they are

Table 4
Statistical errors (evaluated against Ntot,model) for Npara,prev + Nbgr and Npara,new + Nbgr for the considered scenarios. Results for simulations with J(d1)/Q below
and above 10  2 are shown separately as well as the statistics for the whole set of simulations.

Parameterization Average relative bias (%) Average relative error (%)

2 2
J(d1)/Qo 10 J(d1)/Q4 10 Overall J(d1)/Qo 10  2 J(d1)/Q410  2 Overall

Free troposphere
Original 20.3  70.0 39.7 58.9 70.0 101.5
Revised 58.9 36.5 79.6 70.0 87.0 119.7

Rural/continental background
Original  24.3  83.5  36.7 40.7 83.5 49.6
Revised  22.2  38.7  25.6 40.9 49.3 42.6

Polluted boundary layer


Original 62.6  72.8 53.4 105.4 72.8 103.2
Revised 71.5  25.0 64.9 108.5 35.8 103.5
ARTICLE IN PRESS
630 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

Fig. 4. Same as Fig. 3, but for the rural/continental background scenario.

generally within one order of magnitude when J(d1)/Qo10  2 but deviations are considerably higher for larger values of
J(d1)/Q (not illustrated here). Therefore self-coagulation plays a pronounced role in the simulations when J(d1)/Q410  2
which is consistent with the results for the FT scenario (see Fig. 3).
Overall, Fig. 4 demonstrates that the revised parameterization is also capable of taking into account the effects of self-
coagulation such that the parameterized and simulated particle number concentrations are within an order of magnitude
in most of the cases. As a further comparison, the medians of Npara,new and Ntot,model are within a factor of around two and
the same applies for the corresponding average values (Table 3). Further statistics are shown in Table 4. As can be seen
from the table, the mean relative biases and errors are smaller compared to the results for the free troposphere scenario
(see Table 4). There are also slight negative biases for both of the parameterizations implying that the parameterizations
slightly underestimate the number of new particles formed. Also, the revised scheme has better performance compared to
the original formulation especially for simulations with J(d1)/Q410  2. To conclude with, the results demonstrate that the
revised parameterization has reasonably good overall performance and it provides improved accuracy over the original
formulation of the parameterization.

4.2.3. Polluted boundary layer


NPF formed occurred in around 16% of the simulations for the polluted scenario, and the corresponding frequencies
were 13% and 14% for the original and revised parameterizations, respectively. For comparison, such frequencies are by a
factor of around three larger than the results obtained from a background urban site in Helsinki metropolitan area suggest
(Hussein et al., 2008) but are a factor of around two smaller compared to the frequency of NPF events at the urban
atmosphere in Marseille (Petäjä et al., 2007).
Key statistics summarizing the results from simulations with NPF, predicted either of the parameterizations or by the
explicit model, are displayed in Table 3 for the polluted scenario. The median and mean values of Q and CoagS reflect the
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 631

values measured in the urban atmosphere (Mönkkönen et al., 2005; Hussein et al., 2008; Murugavel and Chate, 2009). The
median and mean growth rates are predicted to be 5.6 and 7.3 nm/h, respectively, which also compare well with the
observations (Mönkkönen et al., 2005; Petäjä et al., 2007; Qian, Sakurai, & McMurry, 2007). A comparison of GRcond and
GRtot shows that the particle growth was driven by condensation and self-coagulation contributed only o10% to the
median and average growth rates. Furthermore, the median and mean values of Nmodel were 2.4  104 cm  3 and
1.2  106 cm  3, respectively. The latter value corresponds to formation rates of up to  100 cm  3 s  1 which is higher by
around an order of magnitude than the formation rates of new particles observed at the urban areas (Mönkkönen et al.,
2005; Qian et al., 2007). Such overestimation took also place in the simulations for the previous scenario (see previous
section), and the biases in the model simulations are probably caused by the same reasons. It is iterated that the primary
goal here is to compare the performance of the parameterizations against the explicit model and therefore the biases have
secondary importance.
Fig. 5 shows comparisons of Ntot,model with Npara,new + Nbgr (above) and Npara,prev +Nbgr (below) for the polluted scenario.
The overall correlation is good with both of the parameterizations, the coefficient of determination being 0.91 for the
L2007 parameterization and 0.90 for the revised version of the parameterization. In most of the cases, the results from the
model and from the parameterizations were within an order of magnitude for both of the parameterizations. Table 4
provides further statistics to illustrate the performance of the original and revised schemes compared against the model
calculations. As seen, the revised scheme has better performance in cases with J(d1)/Q410  2 whereas the opposite holds
for the rest of the simulations. The differences are relatively small though. Overall, both of the schemes have similar
performance in terms of the considered metrics with average relative error being around 100%.
Interestingly enough, no large deviations between the predictions of the two parameterizations took place even for the
cases where J(d1)/Q410  2. The small deviations between the predictions of the two parameterizations are explained by
the small role of self-coagulation played at the considered simulations: the particle growth is driven by condensation and

Fig. 5. Same as Fig. 3, but for the polluted scenario.


ARTICLE IN PRESS
632 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

the value of coagulation sink is controlled by the background particles. The latter feature was verified by calculating the
ratio of CoagSscg/CoagStot at the end of a simulation which was 0.02 on average and 0.4 at maximum (not shown in Table 3).
To conclude with, unless nucleation rates reach values 4104 cm  3 s  1 at regular basis, self-coagulation is expected to
have only secondary importance to NPF taking place at polluted areas due to large pre-existing particle surface area and
relatively strong production of condensable vapours.
We also performed a similar set of simulations with the maximum value of J(d1) increased to 105 cm  3 s  1 (not
illustrated here). The results show that large deviations between the compared parameterizations occurred when
J(d1)/Q410  1 such that the performance of the new parameterization is notably better as compared to the original
formulation. As for the previous scenarios, the total number of particles predicted by the model and the revised
parameterization are within one order of magnitude excluding a few cases. Similarly, the original parameterization may
underestimate the particle number concentrations by several orders of magnitude if J(d1)/Q410  1. These results thus are
consistent with those obtained for the previous scenarios. It is thus concluded that the importance of self-coagulation in
polluted areas depends largely also on the relative magnitudes of J(d1) and Q and ultimately on the compounds and
mechanisms responsible for nucleation and condensational growth. However, some features of the simulations, such as the
presence of cases where the particle growth rates exceeded 430 nm/h and the high average number of particles formed
(around 107 cm  3), indicate that the simulated NPF was unrealistically intense as compared to the available ambient
observations. Therefore the results for the original case where the maximum value of J(d1) was set equal to 104 cm  3 s  1,
discussed above, are expected to be atmospherically more relevant.

4.2.4. Time varying variables


The key input parameters, such as the nucleation rate, condensation sink of background particles and gas-phase
concentration of the condensable vapour, were kept constant in all of the above-discussed model simulations. Under
conditions where these quantities vary strongly with time, the revised parameterization can be applied in series by
calculating the formation rate first at e.g. 2 nm, then using the obtained prediction to calculate the formation rate at 3 nm,
for example, while updating the values of the changing parameters and so forth. The rationale behind the approach is that
conditions remain practically unchanged when sufficiently narrow time/diameter intervals are considered. This approach
works also with the revised version because of the following. First, we have successfully evaluated the performance of the
developed parameterization under steady-state conditions (Section 4). Second, the developed parameterization is an
extension of the previous parameterization as demonstrated in Section 2.1. Third, the feasibility of applying the formalism
in series has been investigated by Kerminen, Anttila, Lehtinen, and Kulmala (2004b) and Lehtinen et al. (2007) for the
original version of the parameterization. To summarize, while the procedure introduces some additional inaccuracies there
are no fundamental changes in the behavior because the parameterization can be applied in series over smaller time/
diameter intervals while accounting for changing conditions. For these reasons, detailed numerical tests addressing the
time dependence are not presented here.

5. On applying the parameterization in large scale models

5.1. Decreasing the computing time

Applying the developed parameterization involves an iterative loop which includes summing a Taylor series (Eq. (7)). In
the above-presented calculations, the number of iterations was five and 30 first terms of the Taylor series were accounted
for in computing Eq. (7). Even though computational costs of the procedure are negligible compared to those of an explicit
aerosol model (for the model applied here, the relative difference in CPU times is around 106 for 1 h of simulation with a
modern workstation class PC regardless of the value of dx), we also investigated how the computing time can be decreased
without significantly compromising the accuracy of the parameterization. These considerations are motivated by possible
large-scale modeling applications where it is desirable to minimize computational burden.
We addressed the issue through repeating calculations for the rural/continental background scenario with varying the
number of iterations in calculating Nnuc, Mit, and the number of terms included in computing Eq. (7), Mterm. We then
compared the predicted number of new particles formed (Npara,new) with the corresponding results of the base case
simulations. Results of the comparison for those cases where NPF took place are summarized in Table 5. As seen, the
deviations depends strongly on the value of J(d1)/Q: the average absolute difference stays below o200 cm  3 in all but one
case when J(d1)/Qo0.01, and the average relative differences were below o2% in all cases when J(d1)/Qo0.01. These
results can be explained by the relatively small role played by self-coagulation at this regime of the parameter space. On
the other hand, the deviations from the base case results were much larger when J(d1)/Q40.01: the average absolute
differences were of the order of 104 cm  3 in all cases and the average relative differences varied between 2% and 35%.
Choices made regarding the implementation are reflected in the results because self-coagulation was important for NPF
under these conditions (Section 4.2). Regarding the considered cases, computationally most cheapest option is to apply the
parameterization without the iteration, i.e. performing calculations with Nnuc = J(d1)/CoagS(d1). In this case, the average
relative difference stayed below o10% even when J(d1)/Q40.01, and neglecting the iteration when J(d1)/Qo0.01 is
expected to lead only to marginal errors in applying the parameterization. As further seen from Table 5, the smallest
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 633

Table 5
Average absolute and relative differences in parameterized number of particles formed for different number of iterations, Mit, and number of terms
accounted for when calculating Eq. (7), Mterm. The results are classified into two groups according to the ratio J(d1)/Q.

Mit Mterm J(d1)/Qo0.01 J(d1)/Q40.01

Average absolute Average relative Average absolute Average relative


difference in Npara,new difference in Npara,new difference in Npara,new difference in Npara,new

0 – 3.51E + 03 0.01 4.49E + 04 0.08


1 20 3.03E + 01 0.00 3.20E + 04 0.08
1 30 3.67E + 01 0.00 3.20E + 04 0.02
2 10 1.75E + 02 0.02 2.98E + 04 0.35
2 20 1.25E + 02 0.00 4.32E + 04 0.16
2 30 8.73E + 01 0.00 1.67E + 04 0.02
3 20 3.02E + 01 0.00 3.20E + 04 0.08

Table 6
Statistical errors (evaluated against Ntot,model) for Npara,prev +Nbgr and Npara,new + Nbgr for the continental background/rural scenario. The results are grouped
according to the value of dx which was fixed in each set of simulations. Here ‘‘Original’’ and ‘‘Revised’’ refer to the original formulation of the
parameterized and the revised scheme proposed here, respectively. Also, R2 is the coefficient of determination.

dx Average relative bias (%) Average relative error (%) R2

Original Revised Original Revised Original Revised

5  35.4  29.7 47.8 45.7 0.79 0.84


10 10.4 41.1 82.3 103.2 0.73 0.82
15  41.7  4.8 57.9 85.2 0.70 0.81
20  51.6 1.8 53.3 89.5 0.69 0.79
30  62.2 244.5 67.6 362.2 0.16 0.47

deviations for the simulations with J(d1)/Q4 0.01 were obtained with Mit = 2 and Mterm = 30. In this case the average relative
difference was around 2%.

5.2. Extending the particle size range covered by the parameterization

As indicated in Table 2, the diameter dx was 5 nm at maximum in the above-considered simulations. Even though this is
a reasonable choice for comparison purposes, aerosol modules implemented in large-scale models may cover only a
particle diameter range down to a few tens of nanometers (e.g. Pierce & Adams, 2009). Applying the developed
parameterization in such modeling setups would necessitate increasing dx larger than sizes of 5 nm, and therefore it is also
of interest to investigate the performance of the developed parameterization as a function of of dx. To this end, we
performed a few sets of simulations, each consisting of 750 model runs, with varying values of dx. For the sake of clarity,
the value of dx was kept fixed in each set of simulations and the other considered parameters were varied over the same
ranges as in the rural/continental background scenario except that the minimum simulation time was increased to 6 h to
allow for larger growth periods. The considered values of dx were 5, 10, 15, 20, and 30 nm. No larger values were included
since increasing the value of dx beyond 30 nm increase the temporal lag between nucleation and appearance of new
particles with sizes 4dx to such an extent that NPF would take place only in a very few simulations over the timescale
considered here. Preliminary results revealed that the performance of the revised parameterization is improved
considerably when the values of the parameters l and Keff were set to 3 and 5  10  11 cm3 s  1, respectively, when
dx 410 nm. The results presented here for dx = 15, 20, and 30 nm were obtained using these values, and it is recommended
to use these values when applying the parameterization at the size range dx 410 nm.
Some statistical features of the simulations are summarized in Table 6. The results for dx =5 nm are comparable to the
corresponding results for the base case simulations (see Table 4 and Section 4.2.2) which is expected since dx varied at the
range 3–5 nm in the base case runs. At larger sizes, both the relative biases and errors are larger but there is no coherent
trend with increasing dx. However, the degree of correlation (R2) decreases consistently with increasing dx such that R2
decreases from 0.69 to 0.16 and from 0.76 to 0.47 over the considered size range in the case of the original and revised
parameterization, respectively. Table 6 further shows that the revised parameterization has consistently better
performance in terms of the degree of correlation. However, the average relative errors are larger for the revised
scheme at the size range 45 nm. The feature is probably explained by the fact that the revised scheme predicts generally
larger formation rates of new particles as indicated by the relative biases. Also, a closer look to the predictions of the
revised scheme revealed that the results contain typically a few model runs with large, i.e. more than an order of
magnitude, overestimation of the number of particles formed (not shown here). This is reflected in the relative error and
ARTICLE IN PRESS
634 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

consequently in its average value, in particular when dx was set equal to 30 nm. In any case, the results displayed in Table 6
show that the performance of the parameterizations get notably worse when dx is increased over 20 nm.

5.3. On the errors associated with the use of the parameterization in large-scale models

As discussed in Section 4, the use of the parameterization leads errors as compared to modeling the aerosol dynamics
with an explicit model with a high size resolution. However, the current sectional aerosol modules implemented in large
scale models describe the particle size distribution using only a few tens of sections at maximum (e.g. Pierce & Adams,
2009; Spracklen et al., 2008). Consequently if the whole size range is covered with an explicit model, only a few size bins
can be allocated for the smallest particle sizes which may cause large errors to predicte new particle formation rates
(Korhonen et al., 2003). The errors can be, in fact, larger than those caused by the use of the parameterization. We
demonstrated this point by performing sensitivity calculations using the explicit aerosol model with the resolution chosen
so that the diameter range [d1, dx] was covered by 5 or 10 sections. The results for the continental background/rural
scenario showed that the average relative errors in the total particle number concentrations (as compared to the reference
model) were around 31% and 70% for the model configurations with 5 and 10 size sections in the diameter range odx,
respectively. Here the averaging was taken over those cases where NPF took place during a simulation. In comparison, the
corresponding error associated with the use of the parameterization was around 43% (Table 4). Furthermore, similar
comparison for the polluted and FT scenarios showed that the model setup with 10 sections in the size range odx yielded
better performance than the parameterization with the same margin as for the continental background/rural case. On the
other hand, the average relative errors were by a factor of around two smaller when using the parameterization as
compared to modeling the aerosol dynamics at the size range odx with 5 sections
We would like to further point out that increasing dx to sizes larger than around 5–10 nm tend to increase the temporal
lag between nucleation and appearance of new particles with sizes 4dx to such an extent that conditions may change
considerably during the lag and hence decrease the accuracy of the parameterization notably (Kerminen et al., 2004a).
Large temporal lag may also cause problems related to treating the advection of particles having sizes below dx. Finally, the
assumptions behind the parameterization start to lose their validity when dx is increased to nucleation mode sizes (L2007).
For these reasons, it is not recommended to increase dx over a few tens of nanometers in large-scale modeling applications.
Quantifying the expected level of errors arising from the use of the parameterization in large-scale models is important,
but is beyond the scope of the current study because the errors depend strongly on the model design. It can be noted,
however, that the use of the parameterization caused typically errors of 40–100% in the total particle concentrations
(Section 4.2). Because only a fraction of the newly formed particles are able to grow to CCN sizes, the associated errors in
the CCN concentrations are expected to be smaller. In comparison, the predicted tropospheric average CCN concentrations
vary up to 20% due to uncertainties in atmospheric nucleation and particle growth mechanisms (Spracklen et al., 2008;
Pierce & Adams, 2009). Also, the corresponding variability in the total particle number concentrations is notably larger,
being around a factor of three for particles at the size range 410 nm according to the study of Pierce and Adams (2009).
These results illustrate that the errors related to the use of the parameterization could be similar or even smaller compared
to variability due to uncertainties in atmospheric NPF mechanisms.

6. Summary and conclusions

The study is based on the previous works of Kerminen and Kulmala (2002) and Lehtinen et al. (2007) who derived
formulae connecting ‘‘real’’ and ‘‘apparent’’ nucleation rates. These parameterizations neglect the effects of self-
coagulation of newly formed particles, and here we have extended the previous work to include the self-coagulation. Our
main focus was on calculating the ‘‘apparent’’ nucleation rate, i.e. the rate at which particles appear at sizes larger than the
critical cluster size, as a function of the ‘‘real’’ nucleation rate. The revised parameterization developed here can be thus
applied in large scale models to calculate the new particle formation rate at the smallest diameter covered by the model.
The revised parameterization was comprehensively tested against an explicit aerosol dynamic model at diverse
atmospheric conditions. Based on the simulation results, we conclude that self-coagulation generally enhances, rather than
suppresses, new particle formation through increasing the particle growth rate. The importance of nuclei self-coagulation
to new particle formation depend largely on atmospheric conditions, e.g. on the degree of pollution. For the three scenarios
considered here, self-coagulation had largest influence on new particle formation in the scenario representing rural/
continental background conditions. It was found out that nuclei self-coagulation has importance to new particle formation
when Jnuc/Q410  2 where Jnuc is the nucleation rate and Q is the production rate of condensable vapours. According to the
performed simulations, the corresponding nucleation rates range from 410 cm  3 s  1 (in the case of the free troposphere)
to 4104 cm  3 s  1 (in the case of the polluted boundary layer). Moreover, the performed comparison showed that in terms
of the number of particles formed, the calculations performed with the explicit model and the predictions of revised
parameterization were generally within an order of magnitude.
The revised formulation of the parameterization involves an iterative scheme to calculate the ‘‘apparent’’ nucleation
rate but the iterative procedure can be skipped to minimize computational costs even though the accuracy of the
parameterization is decreased slightly. Furthermore, the original formulation of the parameterization, as well as the
ARTICLE IN PRESS
T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636 635

revised version developed here, is based on a steady state assumption. We point out that under transient conditions (e.g.
cloud outflows, plumes, coastal nucleation) this assumption may not hold, and therefore inaccuracies related to applying
the parameterization might be then higher than reported here. In particular, caution should be exercised when the
nucleation rate exceeds the gas-phase production rate of the nucleating vapour(s). Also the accuracy of the revised
parameterization decreases considerably when the parameterization is applied to cover the particle diameter range up to a
few tens of nanometers. However, it was demonstrated that the developed parameterization captures the dynamics of
newly formed particles more accurately than a sectional model with only a few size sections at this diameter range.
Therefore the scheme is a viable option for use in, e.g. large-scale models where the aerosol module covers only a size
range down to around 3–10 nm.
A plausible next step in the research is to investigate if the revised parameterization can also be used to estimate the
‘‘real’’ nucleation rates on the basis of the ‘‘apparent’’ nucleation rates, i.e. observed new particle formation rates, with an
aim to elucidate if the nucleation rates inferred from ambient measurements are sensitive to the nuclei self-coagulation.

Acknowledgements

This research was supported by the Academy of Finland Center of Excellence program (Project no. 1118615). One of the
authors (T.A.) acknowledges funding from the Emil Aaltonen Foundation.

References

Andronache, C., Chameides, W., Davis, D., Anderson, B., Pueschel, R. Bandy, A., et al. (1997). Gas-to-particle conversion of tropospheric sulfur as estimated
from observations in the western North Pacific during PEM-West B. Journal of Geophysical Research, 102, 28511–28538.
Anttila, T., & Kerminen, V.-M. (2008). Modeling study on aerosol dynamical processes regulating new particle and CCN formation at clean continental
areas. Geophysical Research Letters, 35, L07813, doi:10.1029/2008GL033371.
Anttila, T., Kerminen, V.-M., Kulmala, M., Laaksonen, A., & O’Dowd, C. D. (2004). Modelling the formation of organic particles in the atmosphere.
Atmospheric Chemistry and Physics, 4, 1071–1083.
Birmili, W., Wiedensohler, A., Heintzenberg, J., & Lehmann, K. (2001). Atmospheric particle number size distribution in central Europe: Statistical relations
to air masses and meteorology. Journal of Geophysical Research, 106, 32005–32018.
Chang, L-S., Schwartz, S. E., McGraw, R., & Lewis, E. R. (2009). Sensitivity of aerosol properties to new particle formation mechanisms and to primary
emissions in a continental-scale chemical transport model. Journal of Geophysical Research, 114, D07203, doi:10.1029/2008JD011019.
Clarke, A. D., Eisele, F., Kapustin, V. N., Moore, K., Tanner, D., Mauldin, L., et al., (1999). Nucleation in the equatorial free troposphere: Favorable
environments during PEM-Tropics. Journal of Geophysical Research, 104, 5735-5744, doi:10.1029/98JD02303.
Dal Maso, M., Kulmala, M., Riipinen, I., Wagner, R., Hussein, T. Aalto, P. P., et al. (2005). Formation and growth of fresh atmospheric aerosols: Eight years of
aerosol size distribution data from SMEAR II, Hyytiälä, Finland. Boreal Environmental Research, 10, 323–336.
Dal Maso, M., Sogacheva, L., Aalto, P. P., Riipinen, I., Komppula, M., Tunved, P., et al., (2007). Aerosol size distribution measurements at four Nordic field
stations: identification, analysis and trajectory analysis of new particle formation bursts. Tellus, 59B, 350–361.
Ghan, S. J., & Schwartz, S. E. (2007). Aerosol properties and processes: A path from field and laboratory measurements to global climate models. Bulletin of
the American Meteorological Society, 88, 1059–1083.
Hamed, A., Joutsensaari, J., Mikkonen, S., Sogacheva, L., Dal Maso, M. Kulmala, M., et al. (2007). Nucleation and growth of new particle in Po Valley, Italy.
Atmospheric Chemistry and Physics, 7, 355–376.
Hussein, T., Martikainen, J., Junninen, H., Sogacheva, L., Wagner, R. Dal Maso, M., et al. (2008). Observation of regional new particle formation in the urban
atmosphere. Tellus, 60B, 509–521.
Jacobson, M. Z., & Turco, R. P. (1995). Simulating condensational growth, evaporation, and coagulation of aerosols using a combined moving and
stationary size grid. Aerosol Science and Technology, 22, 73–92.
Kanakidou, M., Seinfeld, J. H., Pandis, S. N., Barnes, I., Dentener, F. J. Facchini, M. C., et al. (2005). Organic aerosol and global climate modelling: A review.
Atmospheric Chemistry and Physics, 5, 1053–1123.
Kazil, J., & Lovejoy, E. R. (2007). A semi-analytical method for calculating rates of new sulfate aerosol formation from the gas phase. Atmospheric Chemistry
and Physics, 7, 3447–3459.
Kerminen, V.-M., Anttila, T., Lehtinen, K. E.J., & Kulmala, M. (2004b). Parametrization for atmospheric new-particle formation: Application to a system
involving sulfuric acid and condensable water-soluble organic vapors. Aerosol Science and Technology, 38, 1001–1008.
Kerminen, V.-M., & Kulmala, M. (2002). Analytical formulae connecting the ‘‘real’’ and ‘‘apparent’’ nucleation rate and the nuclei number concentration for
atmospheric nucleation events. Journal of Aerosol Science, 33, 609–623.
Kerminen, V.-M., Lehtinen, K. E.J., Anttila, T., & Kulmala, M. (2004a). Dynamics of atmospheric nucleation processes: A timescale analysis. Tellus, 56B,
135–146.
Kerminen, V.-M., Lihavainen, H., Komppula, M., Viisanen, Y., & Kulmala, M. (2005). Direct observational evidence linking atmospheric aerosol formation
and cloud droplet activation. Geophysical Research Letters, 32, L14803, doi:10.1029/2005GL023130.
Korhonen, H., Lehtinen, K. E.J., Pirjola, L., Napari, I., Vehkamäki, H. Noppel, M., et al. (2003). Simulation of atmospheric nucleation mode: A comparison of
nucleation models and size distribution descriptions. Journal of Geophysical Research, 108(D15), 4471, doi:10.1029/2002JD003305.
Kulmala, M., Dal Maso, M., Mäkelä, J. M., Pirjola, L., Väkevä, M. Aalto, P. P., et al. (2001). On the formation, growth and composition of nucleation mode
particles. Tellus B, 53, 479–490.
Kulmala, M., & Kerminen, V.-M. (2008). On the growth of atmospheric nanoparticles. Atmospheric Research, 90, 132–150.
Kulmala, M., Petäjä, T., Mönkkönen, P., Koponen, I. K., Dal Maso, M. Aalto, P. P., et al. (2005). On the growth of nucleation mode particles: Source rates of
condensable vapor in polluted and clean environments. Atmospheric Chemistry and Physics, 5, 409–416.
Kulmala, M., Vehkamäki, H., Petäjä, T., Dal Maso, M., Lauri, A. Kerminen, V.-M., et al. (2004). Formation and growth rates of ultrafine atmospheric particles:
A review of observations. Journal of Aerosol Science, 35, 143–176.
Laaksonen, A., Hamed, A., Joutsensaari, J., Hiltunen, L., Cavalli, F. Junkermann, W., et al. (2005). Cloud condensation nucleus production from nucleation
events at a highly polluted region. Geophysical Research Letters, 32, L06812, doi:10.1029/2004GL022092.
Lee, S.-H., Reeves, J. M., Wilson, J. C., Hunton, D. E., Viggiano, A. A. Miller, T. M., et al. (2003). Particle formation by ion nucleation in the upper troposphere
and lower stratosphere. Science, 301, 1886–1889.
Lehtinen, K. E.J., Dal Maso, M., Kulmala, M., & Kerminen, V.-M. (2007). Estimating nucleation rates from apparent particle formation rates and vice versa:
Revised formulation of the Kerminen–Kulmala equation. Journal of Aerosol Science, 38, 988–994.
ARTICLE IN PRESS
636 T. Anttila et al. / Journal of Aerosol Science 41 (2010) 621–636

Lihavainen, H., Kerminen, V.-M., Komppula, M., Hatakka, J., Aaltonen, V. Kulmala, M., et al. (2003). Production of ‘‘potential’’ cloud condensation nuclei
associated with atmospheric new-particle formation in northern Finland. Journal of Geophysical Research, 108, 4782, doi:10.1029/2003JD003887.
Lucas, D. D., & Arimoto, H. (2006). Evaluating nucleation parameterizations in a global atmospheric model. Geophysical Research Letters, 33, l10808,
doi:10.1029/2006GL025672.
Makkonen, R., Asmi, A., Korhonen, H., Kokkola, H., Järvenoja, S. Räisänen, P., et al. (2009). Sensitivity of aerosol concentrations and cloud properties to
nucleation and secondary organic distribution in ECHAM5-HAM global circulation model. Atmospheric Chemistry and Physics, 9, 1747–1766.
McMurry, P. H. (2000). A review of atmospheric aerosol measurements. Atmospheric Environment, 34, 1959–1999.
McMurry, P. H., Fink, M., Sakurai, H., Stolzenburg, M. R., Mauldin, R. L., III Smith, J., et al. (2005). A criterion for new particle formation in the sulfur-rich
Atlanta atmosphere. Journal of Geophysical Research, 110, D22S02, doi:10.1029/2005JD005901.
Mönkkönen, P., Koponen, I. K., Lehtinen, K. E.J., Hämeri, K., Uma, R., & Kulmala, M. (2005). Measurements in a highly polluted Asian mega city:
Observations of aerosol number size distribution, modal parameters and nucleation events. Atmospheric Chemistry and Physics, 5, 57–66.
Murugavel, P., & Chate, D. M. (2009). Generation and growth of aerosols over Pune, India. Atmospheric Environment, 43, 820–828.
O’Dowd, C., McFiggans, G., Creasey, D. J., Pirjola, L., Hoell, C. Smith, M. H., et al. (1999). On the photochemical production of new particles in the coastal
boundary layer. Geophysical Research Letters, 26, 1707–1710.
Petäjä, T., Kerminen, V.-M., Dal Maso, M., Junninen, H., Koponen, I. K. Hussein, T., et al. (2007). Sub-micron atmospheric aerosols in the surroundings of
Marseille and Athens: Physical characterization and new particle formation. Atmospheric Chemistry and Physics, 7, 2705–2720.
Pierce, J. R., & Adams, P. J. (2007). Efficacy of cloud condensation nuclei formation from ultrafine particles. Atmospheric Chemistry and Physics, 7, 1367–1379.
Pierce, J. R., & Adams, P. J. (2009). Uncertainty in global CCN concentrations from uncertain aerosol nucleation and primary emission rates. Atmospheric
Chemistry and Physics, 9, 1339–1356.
Qian, S., Sakurai, S., & McMurry, P. H. (2007). Characteristics of regional nucleation events in urban East St. Louis. Atmospheric Environment, 41, 4119–4127.
Schröder, F., Kärcher, B., Fiebig, M., & Petzold, A. (2002). Aerosol states in the free troposphere at northern midlatitudes. Journal of Geophysical Research,
107, 8126, doi:10.1029/2000JD000194.
Seinfeld, J. H., & Pandis, S. N. (1998). Atmospheric Chemistry and Physics: From Air Pollution to Climate Change. New York: John Wiley.
Spracklen, D. V., Carslaw, K. S., Kulmala, M., Kerminen, V.-M., Mann, G. W., & Sihto, S.-L. (2006). The contribution of boundary layer nucleation events to
total particle concentrations on regional and global scales. Atmospheric Chemistry and Physics, 6, 5631–5648.
Spracklen, D. V., Carslaw, K. S., Kulmala, M., Kerminen, V.-M., Sihto, S.-L. Riipinen, I., et al. (2008). Contribution of particle formation to global cloud
condensation nuclei concentrations. Geophysical Research Letters, 35, L06808, doi:10.1029/2007GL033038.
Tunved, P., Hansson, H.-C., Kerminen, V.-M., Ström, J. Dal Maso, M., et al. (2006). High natural aerosol loading over boreal forests. Science, 312, 261–263.
Tunved, P., Hansson, H.-C., Kulmala, M., Aalto, P. P., Viisanen, Y. Karlsson, H., et al. (2003). One year boundary layer aerosol size distribution data from five
Nordic background stations. Atmospheric Chemistry and Physics, 3, 2183–2205.
Venzac, H., Sellegri, K., Laj, P., Villani, P., Bonasoni, P. Marinoni, A., et al. (2008). High frequency new particle formation in the Himalayas. Proceedings of the
National Academy of Sciences, 105, 5666–15671.
Wang, M., Penner, J. E. (2009). Aerosol indirect forcing in a global model with particle nucleation. Atmospheric Chemistry and Physics, 9, 239–260.
Weber, R., McMurry, P., Mauldin, R., III, Tanner, D., Eisele, F. Clarke, A., et al. (1999). New particle formation in the remote troposphere: A comparison of
observations at various sites. Geophysical Research Letters, 26, 307–310.
Wehner, B., & Wiedensohler, A. (2003). Long term measurements of submicrometer urban aerosols: Statistical analysis for correlations with
meteorological conditions and trace gases. Atmospheric Chemistry and Physics, 3, 867–879.
Wiedensohler, A., Cheng, Y. F., Nowak, A., Wehner, B., Achtert, P. Berghof, M., et al. (2009). Rapid aerosol particle growth and increase of cloud
condensation nucleus activity by secondary aerosol formation and condensation: A case study for regional air pollution in northeastern China. Journal
of Geophysical Research, 114, D00G08, doi:10.1029/2008JD010884.
Yu, F., Wang, Z., Luo, G., & Turco, R. (2008). Ion-mediated nucleation as an important global source of tropospheric aerosols. Atmospheric Chemistry and
Physics, 8, 2537–2554.

You might also like