-
Technical Note: A new SIze REsolved Aerosol Model
(SIREAM)
E. Debry, K. Fahey, K. Sartelet, B. Sportisse, M. Tombette
To cite this version:
E. Debry, K. Fahey, K. Sartelet, B. Sportisse, M. Tombette.
Technical Note: A new SIze RE-solved Aerosol Model (SIREAM).
Atmospheric Chemistry and Physics, European GeosciencesUnion, 2007,
7 (6), pp.1547.
HAL Id: hal-00328488
https://hal.archives-ouvertes.fr/hal-00328488
Submitted on 21 Mar 2007
HAL is a multi-disciplinary open accessarchive for the deposit
and dissemination of sci-entific research documents, whether they
are pub-lished or not. The documents may come fromteaching and
research institutions in France orabroad, or from public or private
research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au
dépôt et à la diffusion de documentsscientifiques de niveau
recherche, publiés ou non,émanant des établissements
d’enseignement et derecherche français ou étrangers, des
laboratoirespublics ou privés.
https://hal.archives-ouvertes.frhttps://hal.archives-ouvertes.fr/hal-00328488
-
Atmos. Chem. Phys., 7, 1537–1547,
2007www.atmos-chem-phys.net/7/1537/2007/© Author(s) 2007. This work
is licensedunder a Creative Commons License.
AtmosphericChemistry
and Physics
Technical Note: A new SIze REsolved Aerosol Model (SIREAM)
E. Debry, K. Fahey, K. Sartelet, B. Sportisse, and M.
Tombette
CEREA, Joint Research Laboratory,École Nationale des Ponts et
Chaussées / EDF R&D, France
Received: 9 November 2006 – Published in Atmos. Chem. Phys.
Discuss.: 22 November 2006Revised: 26 February 2007 – Accepted: 2
March 2007 – Published: 21 March 2007
Abstract. We briefly present in this short paper a newSIze
REsolved Aerosol Model (SIREAM) which simu-lates the evolution of
atmospheric aerosol by solving theGeneral Dynamic Equation (GDE).
SIREAM segregatesthe aerosol size distribution into sections and
solves theGDE by splitting coagulation and
condensation/evaporation-nucleation. A quasi-stationary sectional
approach is usedto describe the size distribution change due to
condensa-tion/evaporation, and a hybrid equilibrium/dynamical
mass-transfer method has been developed to lower the computa-tional
burden. SIREAM uses the same physical parameter-izations as those
used in the Modal Aerosol Model, MAMSartelet et al. (2006). It is
hosted in the modeling systemPOLYPHEMUS Mallet et al., 2007, but
can be linked to anyother three-dimensional Chemistry-Transport
Model.
1 Introduction
Atmospheric particulate matter (PM) has been negativelylinked to
a number of undesirable phenomena ranging fromvisibility reduction
to adverse health effects. It also has astrong influence on the
earth’s energy balance Seinfeld andPandis (1998). As a result, many
governing bodies, es-pecially in North America and Europe, have
imposed in-creasingly stringent standards for PM. As an exemple
the1999/30/CE European Directive has imposed a daily
PM10concentration limit of 50µg/m3 since January 2005.
Atmospheric aerosol is a complex mixture of inorganicand organic
components, with composition varying over thesize range of a few
nanometers to several micrometers. Theseparticles can be emitted
directly from various anthropogenicand biogenic sources or can be
formed in the atmosphere byorganic or inorganic precursor
gases.
Given the complexity of PM, its negative effects, andthe desire
to control atmospheric PM concentrations, mod-els that accurately
describe the important processes thataffect the aerosol
size/composition distribution are there-fore crucial.
Three-dimensional Chemistry-Transport Mod-
Correspondence to: B.
Sportisse([email protected])
els (CTMs) provide the necessary tools to develop not onlya
better understanding of the formation and the distributionof PM but
also sound strategies to control it. Historically,CTMs focused on
ozone formation or acid deposition and didnot include a detailed
treatment of aerosol. Several modelshave been developed that
include a very thorough treatmentof aerosol processes Gong et al.
(2003); Adams and Sein-feld (2002); Spracklen et al. (2005), but
there are still manylimitations Seigneur (2001).
In detailed models that seek to describe the time and spa-tial
evolution of atmospheric PM, it is necessary to includethose
processes described in the General Dynamic Equa-tion for aerosols
(condensation/evaporation, coagulation, nu-cleation, inorganic and
organic thermodynamics). Theseand additional processes like
heterogeneous reactions at theaerosol surface, mass transfer
between aerosol and clouddroplets, and aqueous-phase chemistry
inside cloud dropletsrepresent some of the most important
mechanisms for alter-ing the aerosol size/composition
distribution.
Among the aerosol models, one usually distinguishes
be-tweenmodal modelsWhitby and McMurry (1997) andsizeresolved or
sectional models Gelbard et al. (1980). We referfor instance to the
modal model of Binkowski and Roselle(2003) and the sectional model
of Zhang et al. (2004) fora description of state-of-the-science
aerosol models, hostedby the Chemistry-Transport Model, CMAQ Byun
and Schere(2004).
Here we present the development of a new SIze RE-solved Aerosol
Model (SIREAM). SIREAM is the size-resolved alternative to the
modal model, MAM ModalAerosol Model, Sartelet et al. (2006). Both
models use thesame physical parameterizations through the library
for at-mospheric physics and chemistry ATMODATA Mallet andSportisse
(2005). Both have a modular approach and relyon different model
configurations. They are hosted in themodeling system POLYPHEMUS
Mallet et al., 2007 and usedin several global, regional and local
eulerian applications. Adetailed description of SIREAM and MAM can
be found inSportisse et al. (2006) (available at
http://www.enpc.fr/cerea/polyphemus). A key feature of SIREAM is
its modular de-sign, as opposed to an all-in-one model. SIREAM can
be
Published by Copernicus GmbH on behalf of the European
Geosciences Union.
http://www.enpc.fr/cerea/polyphemushttp://www.enpc.fr/cerea/polyphemus
-
1538 E. Debry et al.: A new SIze REsolved Aerosol Model
used in many configurations and is intended for ensemblemodeling
(similar to Mallet and Sportisse (2006)).
This paper is structured as follows. The model formula-tion and
main parameterizations included in SIREAM aredescribed in Sect. 2.
The numerical algorithms used for solv-ing the GDE are given in
Sect. 3. A specific focus is devotedto condensation/evaporation,
which is by far the most chal-lenging issue.
2 Model formulation
In this section we focus on aerosol dynamics, i.e. on
thenucleation, condensation/evaporation, and coagulation
pro-cesses. In addition, we briefly describe some processes thatare
strongly related to aerosols (heterogeneous reactions atthe aerosol
surface, mass transfer between the aerosols andthe cloud droplets
and aqueous-phase chemistry in clouddroplets). We also include the
parameterizations for Semi-Volatile Organic Compounds (SVOCs).
In order to deal with different parameterizations and toavoid
the development of an “all-in-one” model, the parame-terizations
have been implemented as functions of the libraryATMODATA (Mallet
and Sportisse (2005)), a package for at-mospheric physics. As such,
they can be used by other mod-els.
2.1 Composition
The particles are assumed to beinternally mixed, i.e., thatthere
is a unique chemical composition for a given size. Eachaerosol may
be composed of the following components :
– liquid water;
– inert species : mineral dust, elemental carbon and, insome
applications, heavy trace metals (lead, cadmium)or radionuclides
bound to aerosols;
– inorganic species : Na+, SO2−4 , NH4+, NO−3 and Cl
−;
– organic species : one species forPrimary OrganicAerosol (POA),
8 species for Secondary OrganicAerosol (see below for more
details).
A typical version of the model (trace metals or radionuclidesare
not included) tracks the evolution of 17 chemical speciesfor a
given size bin (1+2+5+1+8). These species (externalspecies) should
be distinguished from the species that are ac-tually inside one
aerosol in different forms (ionic, dissolved,solid). Letne be the
number of external species.
The internal composition for inorganic species is deter-mined by
thermodynamic equilibrium, solved by ISOR-ROPIA V.1.7 Nenes et al.
(1998). Water is assumedto quickly reach equilibrium between the
gas and aerosolphases. Its concentration is given by the
thermodynamicmodel (through the Zdanovskii-Stokes-Robinson
relation).
The organic composition is given by the SORGAM modelSchell et
al. (2001b) which we detail in Sect. 2.2.5.
Hereafter, the particle massm refers to thedry mass. Inorder to
reduce the wide range of magnitude over the particlesize
distribution and to better represent small particles, theparticle
distribution is described with respect to the logarith-mic massx =
ln m Wexler et al. (1994); Meng et al. (1998);Gaydos et al.
(2003).
The particles are described by a number distribution,n(x, t) (in
m−3), and by the mass distributions for species Xi ,{qi(x,
t)}i=1,ne (in µg.m−3). The mass distributions satisfy∑i=ne
i=1 qi = m n. We also define the massmi(x, t) =qi
(x,t)n(x,t)
of species Xi in the particle of logarithmic massx. It
satisfies∑i=ne
i=1 mi(x, t) = ex .
2.2 Processes and parameterizations for the GDE
2.2.1 Nucleation
The formation of the smallest particles is given by the
ag-gregation of gaseous molecules leading to thermodynami-cally
stableclusters. The mechanism is poorly known andmost models assume
homogeneous binary nucleation of sul-fate and water to be the major
mechanism in the formationof new particles. Binary schemes tend to
underpredict nucle-ation rates in comparison with observed values.
Korhonenet al. (2003) has indicated that for the conditions typical
inthe lower troposphere ternary nucleation of sulfate, ammo-nium
and water may be the only relevant mechanism.
SIREAM offers two options for nucleation: the H2O-H2O4 binary
nucleation scheme ofVehkamki et al. (2002)and the H2O-H2O4-NH3
ternary nucleation scheme of Na-pari et al. (2002).
The output is a nucleation rate,J0, a nucleation diameter,and
chemical composition for the nucleated particles. Thenew particles
are added to the smallest bin.
2.2.2 Coagulation
Atmospheric particles may collide with one another due totheir
Brownian motion or due to other forces (e.g., hydro-dynamic,
electrical or gravitational). SIREAM includes adescription of
Brownian coagulation, the dominant mecha-nism in the atmosphere.
There may be a limited effect onthe particle mass distribution and
this process is usually ne-glected Zhang et al. (2004). However
coagulation may havesubstantial impact on the number size
distribution for ultra-fine particles.
The coagulation kernelK(x, y) (in unit of volume per unitof
time) describes the rate of coagulation between two par-ticles of
dry logarithmic massesx andy. K has differentexpressions depending
on the relevant regime Seinfeld andPandis (1998).
Atmos. Chem. Phys., 7, 1537–1547, 2007
www.atmos-chem-phys.net/7/1537/2007/
-
E. Debry et al.: A new SIze REsolved Aerosol Model 1539
2.2.3 Condensation/evaporation
Some gas-phase species with a low saturation vapor pressuremay
condense on existing particles while some species in theparticle
phase may evaporate. The mass transfer is governedby the gradient
between the gas-phase concentration and theconcentration at the
surface of the particle. The mass flux forvolatile species Xi
between the gas phase and one particle oflogarithmic massx is
computed by:
dmi
dt= Ii = 2πDgi dpfFS(Kni , αi)
(
cgi − c
si (x, t)
)
(1)
dp is the particle wet diameter (see Sect. 2.2.6 for the
relationto mass).Dgi andc
gi are the molecular diffusivity in the air
and the gas-phase concentration of species Xi , respectively.The
concentrationcsi at the particle surface is assumed to beat local
thermodynamic equilibrium with the particle compo-sition:
csi (x, t) = η(dp) ceqi (q1(x, t), . . . , qne (x, t), RH, T )
(2)
T is the temperature andRH is the relative humidity.
η(dp) = exp(
4σvpRT dp
)
is a correction for the Kelvin effect,
with σ the surface tension,R the gas constant andvp theparticle
molar volume. In practice,ceqi is computed by there-verse mode of a
thermodynamics package like ISORROPIANenes et al. (1998) in the
case of kinetic mass transfer.
The Fuchs-Sutugin function,fFS , describes the non-continuous
effects (Dahneke (1983)). It depends on theKnudsen number of
species Xi , Kni =
2λidp
(with λi the airmean free path), and on the accommodation
coefficientαi(default value is 0.5):
fFS(Kni , αi) =1 + Kni
1 + 2Kni (1 + Kni )/αi(3)
When particles are in a liquid state, the condensation of
anacidic component may free hydrogen ions and the conden-sation of
a basic component may consume hydrogen ions.Thus the
condensation/evaporation (c/e hereafter) processmay have an effect
on the particlepH . The hydrogen ionflux induced by mass transfer
is:
JH+ = 2JH2O4 + JHCl + JHNO3 − JNH3 (4)
with Ji the molar flux in species Xi . ThepH evolution dueto c/e
can be very stiff and cause instabilities, due to the verysmall
quantitynH+ of hydrogen ions inside the particle. Thehydrogen ion
flux is then limited to a given fractionA of thehydrogen ion
concentration following Pilinis et al. (2000) :|JH+ | ≤ AnH+ ,
whereA is usually chosen arbitrarily be-tween 0.01 and 0.1. A is a
numerical parameter that has nophysical meaning and does not
influence the final state ofmass transfer. It just modifies the
numerical path to reachthis state. We refer to Pilinis et al.
(2000) for a deeper under-standing.
2.2.4 Inorganic thermodynamics
There are a range of packages available to solve thermody-namics
for inorganic species Zhang et al. (1998). ISOR-ROPIA Nenes et al.
(1998) was shown to be a computation-ally efficient model that is
also numerically accurate and sta-ble and provides both aclosed
mode (for global equilibrium,a.k.a. forward mode) andopen mode (for
local equilibriumand kinetic mass transfer, a.k.a.reverse mode).
Particles canbe solid, liquid, both or in a metastable state, where
particlesare always in aqueous solution.
Moreover, the inclusion of sea salt (NaCl) in the computa-tion
of thermodynamics is also an option in SIREAM.
When the particles are solid, fluxes of inorganic speciesare
governed by gas/solid reactions at the particle surface.In this
case, thermodynamic models are not able to computegas equilibrium
concentrations. For solid particle, SIREAMcalculations are based on
the solutions proposed in Piliniset al. (2000).
2.2.5 Secondary Organic Aerosols
The oxidation of VOCs leads to species (SVOCs) that
haveincreasingly complicated chemical functions, high
polariza-tions, and lower saturation vapor pressure.
There are many uncertainties surrounding the formation
ofsecondary organic aerosol. Due to the lack of knowledge andthe
sheer number and complexity of organic species, mostchemical
reaction schemes for organics are very crude repre-sentations of
the “true” mechanism. These typically includethe lumping of
“representative” organic species and highlysimplified reaction
mechanisms.
The default gas-phase chemical mechanism for SIREAMis RACM
Stockwell et al. (1997). Notice that the gas-phasemechanism and the
related SVOCs are parameterized andcan be easily modified.
The low volatility SOA precursors and the partitioning be-tween
the gas and particle phases are based on the empiri-cal SORGAM
model (Schell et al. (2001a); Schell (2000)).Eight SOA classes are
taken into account (4 anthropogenicand 4 biogenic). Anthropogenic
species include two fromaromatic precursors (ARO 1 and ARO 2), one
from higheralkanes (OLE 1) and one from higher alkenes (ALK 1).The
biogenic species represent two classes fromα-pinene(API 1 and API
2) and two from limonene (LIM 1 andLIM 2) degradation. Some
oxidation reactions of the formVOC + Ox → P where Ox is OH, O3, or
NO3 have beenmodified to VOC+ Ox → P+ α1 P1 + α2 P2 with P1 andP2
representing SVOCs among the eight classes. Updatedvalues of these
parameters have also been defined in otherversions of the mechanism
(not reported here).
The partitioning between the gas phase and the particlephase is
performed in the following way. LetnOM be thenumber of organic
species in the particle mixture (this in-
www.atmos-chem-phys.net/7/1537/2007/ Atmos. Chem. Phys., 7,
1537–1547, 2007
-
1540 E. Debry et al.: A new SIze REsolved Aerosol Model
cludes primary and secondary species) which are assumed
toconstitute anideal mixture:
(qi)g = γi (xi)a qsati (5)
For species Xi , qsati is the saturation mass concentration in
apure mixture,(xi)a is the molar fraction in the organic mix-ture
andγi is the activity coefficient in the organic mixture(a default
value of 1 is assumed).(xi)a is computed through:
(xi)a =
(qi)a
MiqOM
MOM
=
(qi)a
Mij=nOM∑
j=1
(qj )a
Mj+
(qPOA)a
MPOA
(6)
qOM is the total concentration of organic matter (primary
andsecondary) in the particle phase. The molar massMi of
com-ponenti is expressed inµg/mol (in the same unit as the
massconcentrationsqi); MOM is the average molar mass for or-ganic
matter inµg/mol. POA stands for the primary organicmatter, assumed
not to evaporate.
qsati is computed from the saturation vapor pressure with
qsati =MiRT
psati . A similar way to proceed is to define the
partitioning coefficientKi = (qi )aqOM (qi )g (in m3/µg). Ki
can
be computed from the thermodynamic conditions and the
sat-uration vapor pressure through:
Ki =RT
psati γi(MOM)(7)
The saturation vapor pressurepsati (T ) is given by
theClausius-Clapeyron law:
psati (T ) = psati (298K) exp
(
−1Hvap
R(
1
T−
1
298)
)
(8)
with 1Hvap the vaporization enthalpy (in the default version,a
constant value 156 kJ/mol).
The mass concentration of a gas at local equilibrium withthe
particle mixture is given by Eq. (5). The global equi-librium
between a gas and the particle mixture is given byEq. (5) and mass
conservation for speciesXi :
(qi)a + (qi)g = (qi)tot (9)
with (qj )tot representing the total mass concentration (forboth
phases) to be partitioned. This with Eq. (6) leads toa system ofnOM
algebraic equations of second degree:
− ai ((qi)a)2 + bi(qi)a + ci = 0 (10)
where the coefficients depend on concentrations{(qj )a}j
6=ithroughai = 1Mi , bi =
qsatiMi
− 6i , ci = qsati 6i and
6i =j=nOM∑
j=1,j 6=i
(qj )a
Mj+
(qPOA)a
MPOA.
The resulting system is solved by an iterative approachwith a
fixed point algorithm. Each second degree equationis solved in an
exact way: the only positive root is computedfor each equation of
type (10).
2.2.6 Wet diameter
Parameterizations of coagulation, condensation/evaporation,dry
deposition and wet scavenging depend on the particle“wet”
diameterdp. Two methods have been implementedin SIREAM to compute
it, one based on thermodynamics,another on the Gerber’s
Formula.
The thermodynamic method consists in using the particleinternal
composition{mi} provided by the thermodynamicmodel ISORROPIA. Many
of aerosol models use a constantspecific particle massρp Wexler et
al. (1994); Pilinis and
Seinfeld (1988) supposed to satisfyρpπd3p
6 =∑ne
i=1 mi . InSIREAM, following Jacobson (2002), the particle
volume is
split into a solid part and a liquid part:πd3p
6 = Vliq + Vsol. Aseach solid represents one single phase, the
total solid particlevolume is the sum of each solid volume :Vsol
=
∑
is
misρ∗is
,
with ρ∗is the specific mass of pure componentXis . The
liquidparticle phase is a concentrated mixing of inorganic
species,whose volume is a non linear function of its inorganic
in-ternal composition :Vliq =
∑
ilVilnil whereVil is the par-
tial molar volume of ionic or dissolved speciesXil andnil isthe
molar quantity inXil . Due to some molecular processeswithin the
mixture (e.g. volume exclusion), the partial molarvolume is a
function of the internal composition. However,we assume thatVil
≃
Miρ∗il
whereMi andρ∗il are the molar
mass of Xi and the specific mass of a pure liquid solutionof X i
, respectively. This method is well suited for
condensa-tion/evaporation for which thermodynamic computation
can-not be avoided.
For other processes (coagulation, dry deposition and
scav-enging) the particle “wet” diameter is computed through
afaster method, the Gerber’s Formula (Gerber (1985)). Thisone is a
parameterization of the “wet” radius as a function ofthe dry one
:
rw =[
C1(rd)C2
C3(rd)C4 − logRH+ (rd)3
] 13
(11)
whererw andrd are respectively the wet and dry particle ra-dius
in centimeters,RH is the atmospheric relative humiditywithin [0,
1]. Coefficients(Ci)i=1,4 depend on the particletype (urban, rural
or marine). TheC3 coefficient is tempera-ture dependent (T) through
the Kelvin effect:
C3(T ) = C3[1 + C5(298− T )] (12)
We have actually modified the coefficients given by
Gerberthrough a minimization method so that the Gerber’s
Formulagive results as close as possible to the “wet” diameters
givenby the thermodynamic method Sportisse et al. (2006):
C1 = 0.4989, C2 = 3.0262, C3=0.5372 10−12
C4 = −1.3711, C5=0.3942 10−02 (13)
The choice of which method to use (thermodynamics orGerber’s
Formula) is up to the user.
Atmos. Chem. Phys., 7, 1537–1547, 2007
www.atmos-chem-phys.net/7/1537/2007/
-
E. Debry et al.: A new SIze REsolved Aerosol Model 1541
2.2.7 Logarithmic formulation for the GDE
On the basis of the parameterizations described above,
theevolution of the number and mass distributions is governedby the
GDE :
∂n
∂t(x, t) =
∫ x̃
x0
K(y, z)n(y, t)n(z, t) dy
−n(x, t)∫ ∞
x0
K(x, y)n(y, t) dy
−∂(H0n)
∂x(14)
∂qi
∂t(x, t) =
∫ x̃
x0
K(y, z)[qi(y, t)n(z, t) + n(y, t)qi(z, t)] dy
−qi(x, t)∫ ∞
x0
K(x, y)n(y, t) dy
−∂(H0qi)
∂x+ (Iin)(x, t) (15)
H0 = I0m (in s−1) is the logarithmic growth rate. The nucle-
ation threshold isx0= ln m0. Moreover,x̃ = ln(ex −ex0) andz =
ln(ex − ey) in the above formula.
At the nucleation threshold, the nucleation rate determinesthe
boundary condition :
(H0n)(x0, t) = J0(t) , (H0qi)(x0, t) = mi(x0, t)J0(t) (16)
The evolution of the gaseous concentration for the semi-volatile
species Xi is given by:
dcgi
dt(t) = −mi(x0, t)J0(t) −
∫ ∞
x0
(Iin)(x, t) dx (17)
or by mass conservation :cgi (t) +∫ ∞x0
qi(x, t) dx = Ki .
2.3 Other processes related to aerosols
The following processes are not directly part of SIREAM.As such,
the core of SIREAM (the parameterizations and thealgorithms for the
GDE) is independent. As for SOA, otherparameterizations can be
used. For the sake of completeness,we have chosen to include a
brief description of the defaultcurrent parameterizations.
2.3.1 Mass transfer and aqueous-phase chemistry for
clouddroplets
For cells with a liquid water content exceeding a criticalvalue
(the default value is 0.05 g/m3), the grid cell is as-sumed to
contain a cloud and the aqueous-phase module iscalled instead of
the SIREAM module. A part of the parti-cle distribution is
activated for particles that exceed a criticaldry diameter the
default value isdactiv = 0.7µm Straderet al. (1998). The
microphysical processes that govern the
evolution of cloud droplets are parameterized and not
explic-itly described. Cloud droplets form on activated particles
andevaporate instantaneously (during one numerical timestep)
inorder to take into account the impact of aqueous-phase chem-istry
for the activated part of the particle distribution Fahey(2003);
Fahey and Pandis (2001).
The activated particle fraction is then incorporated into
thecloud droplet distribution. The VSRM model can simulatea
size-resolved droplet distribution, but we use only a bulkapproach
in order to decrease the computational. In this casethe average
droplet diameter is fixed at 20µm. The chemicalcomposition of the
cloud droplet is then given by the acti-vated particle
fraction.
Aqueous-phase chemistry and mass transfer between thegaseous
phase and the cloud droplets (bulk solution) are thensolved. The
aqueous-phase model is based on the chem-ical mechanism developed
at Carnegie Mellon UniversityStrader et al. (1998). It contains 18
gas-phase species and28 aqueous-phase species. Aqueous-phase
chemistry is mod-eled by a chemical mechanism of 99 chemical
reactions and17 equilibria (for ionic dissociation).
Mass transfer is solved dynamically only for “slow”1
species, while “fast” species are assumed to be described
byHenry’s equilibrium, we refer to Sportisse et al. (2006) for
adetailed list of species and their status.
The radical chemistry is not taken into account. The
com-putation of H+ is made with the electroneutrality
relationwritten asfelectroneutrality(H+) = 0. This nonlinear
algebraicequation is solved with the bisection method. If no
conver-gence occurs, we take a default valuepH=4.5 as
observeddroplet pH often ranges between 4.0 en 5.0 Seinfeld and
Pan-dis (1998); Pruppacher and Klett (1998).
After one timestep, the new mass generated from aqueouschemistry
is redistributed onto aerosol bins that were acti-vated. To do so
the initial aerosol distribution is assumedto have a bimodal shape
(log-normal distributions) that givesweighting factors for each
aerosol bin. Median diameter andvariance for each mode are
respectively 0.4µm and 1.8 forfirst one, 2.5µm and 2.15 for second
one. The tests in Fahey(2003) illustrate the low impact of the
choice made for thisassumption.
We use a splitting method, the gas-phase chemistry beingsolved
elsewhere (in the gas-phase module of the Chemistry-Transport
Model). Aqueous-phase chemistry and mass trans-fer are solved with
VODE Brown et al. (1989).
2.3.2 Heterogeneous reactions
The heterogeneous reactions at the surface of condensed mat-ter
(particles and cloud or fog droplets) may significantly im-pact
gas-phase photochemistry and particles. This process issolved
together with gas-phase chemistry. Following Jacob
1“slow” and “fast” refer to the time for given species to
reachequilibrium
www.atmos-chem-phys.net/7/1537/2007/ Atmos. Chem. Phys., 7,
1537–1547, 2007
-
1542 E. Debry et al.: A new SIze REsolved Aerosol Model
(2000), these processes are described by the first-order
reac-tions:
HO2PM→ 0.5 H2O2
NO2PM→ 0.5 HONO+ 0.5 HNO3
NO3PM→ HNO3
N2O5PM,clouds−→ 2 HNO3
The heterogeneous reactions for HO2, NO2 and NO3 at thesurface
of cloud droplets are assumed to be taken into ac-count in the
aqueous-phase model and are considered sepa-rately.
The first-order kinetic rate is computed for gas-phase
species Xi with ki =(
a
Dgi
+ 4c̄gi γ
)−1Sa wherea is the par-
ticle radius,c̄gi the thermal velocity in the air,γ the
reactionprobability andSa the available surface for condensed
matterper air volume.
γ strongly depends on the chemical composition andon the
particle size. We have decided to keep the varia-tion ranges from
Jacob (2000) for these parameters in or-der to evaluate the
resulting uncertainties:γHO2 ∈ [0.1-1], γNO2 ∈ [10−6-10−3], γNO3 ∈
[2.10−4-10−2] andγN2O5 ∈ [0.01-1]. The default values are the
lowest val-ues. For numerical stability requirements, these
reactions arecoupled to the gas-phase mechanism.
3 Numerical simulation
3.1 Numerical strategy
On the basis of a comprehensive benchmark of algorithmsDebry
(2004), the numerical strategy relies on methods thatensure
stability with a low CPU cost. First, we use a split-ting approach
for coagulation and condensation/evaporation.Second, the
discretization is performed with sectional meth-ods which remain
stable even with a few discretizationpoints, contrary to spectral
methods Sandu and Borden(2003); Debry and Sportisse (2005b). Third,
condensa-tion/evaporation is solved with a Lagrangian method,
thequasistationary method of Jacobson Jacobson (1997) is em-ployed
to reduce the numerical diffusion associated with Eu-lerian schemes
in the case of a small number of discretizationpoints (typically
the case in 3D models).
The splitting sequence goes from the slowest pro-cess to the
fastest one (first coagulation and
thencondensation/evaporation-nucleation). The nucleation pro-cess
is not a numerical issue and is solved simultaneouslywith
condensation/evaporation. In the following, we presentthe numerical
algorithm used for each process.
The particle mass distribution is discretized intonb bins[xj ,
xj+1]. We define the integrated quantities over the bin
j for the number distribution and the mass distributions
forspecies Xi :
N j (t) =∫ xj+1
xjn(x, t) dx , Q
ji =
∫ xj+1
xjqi(x, t) dx (18)
m̃ji =
Qji
Njis the average mass per particle inside binj for
species Xi .We use a Method of Lines by first performing size
dis-
cretization and then time integration. After discretization,
theresulting system of Ordinary Differential Equations (ODEs)has
the generic form:
dc
dt= f (c, t) (19)
where the state vectorc is specific for each process.cn is
thenumerical approximation ofc(tn) at timetn, with a timestep1tn =
tn+1 − tn. A second-order solver is specified foreach case with a
first-order approximationc̃n+1. The vari-able timestep1tn is
adjusted by:
1tn+1 = 1tn
√
εr‖cn+1‖2‖c̃n+1 − cn+1‖2
(20)
whereεr is a user parameter, usually between 0.01 and 0.5.The
higherεr is, the faster1tn increases.‖.‖2 is the Eu-clidean
norm.
3.2 Size discretization
3.2.1 Coagulation
Coagulation is solved by the so-calledsize binning
method(Jacobson et al. (1994)). Equations (14) and (15) are
inte-grated over each bin, which gives:
dNk
dt(t) =
1
2
k∑
j1=1
k∑
j2=1f kj1j2Kj1j2N
j1N j2 − Nknb∑
j=1KkjN
j
dQki
dt(t) =
k∑
j1=1
k∑
j2=1f kj1j2Kj1j2Q
j1i N
j2 − Qkinb∑
j=1KkjN
j(21)
Kj1j2 is an approximation of the coagulation kernel
betweenbinsj1 andj2.
The key point is to compute the partition coefficientsf
kj1j2that represent the fraction of particle combinations
betweenbins j1 andj2 falling into bin k. As these coefficients
onlydepend on the chosen discretization, they can be computedin a
preprocessed step. The computation depends on theassumed shape of
continuous densities inside each bin forthe closure scheme, see
Debry and Sportisse (2005a). InSIREAM, we use a closure scheme
similar to Fernndez-Dazet al. (2000).
Atmos. Chem. Phys., 7, 1537–1547, 2007
www.atmos-chem-phys.net/7/1537/2007/
-
E. Debry et al.: A new SIze REsolved Aerosol Model 1543
3.2.2 Condensation/evaporation-nucleation
Lagrangian formulation Let x̄j (t) be the logarithmic massof one
particle at timet whose initial value corresponds topoint xj of the
fixed discretization. The time evolution ofx̄j (t) is given by the
equation of the characteristic curve :
dx̄j
dt(t) = H0(x̄j , t) , x̄j (0) = xj (22)
One crucial issue is to ensure that the characteristic curvesdo
not cross themselves. If this happens the Lagrangian for-mulation
is no longer valid. In real cases we have no proofthat this does
not happen, even though we have not seen sucha situation up to
now.
Provided that the characteristic curves do not cross, wecan
define integrated quantitiesN j and Qji for each La-
grangian bin[x̄j , x̄j+1] : N j (t) =∫ x̄j+1
x̄jn(x, t) dx and
Qji =
∫ x̄j+1
x̄jqi(x, t) dx.
Mass conservation can be easily written in the form :cgi (t)
+
∑nbj=1 Q
ji (t) = Ki .
The time derivation of integrated quantities leads to
theequations:
dN j
dt= 0 ,
dQji
dt= N j Ĩ ji (23)
Ĩji is an approximation of the mass transfer rate for
species
Xi in bin j :
Ĩji = 2πDid
jpf (K
jni , αi)
︸ ︷︷ ︸
aji
(
Ki −nb∑
k=1Qki − η
j (ceqi )
j
)
(24)
with ηj = e4σvp
RT djp . (ceqi )
j is computed at̃mji .For the nucleation process, the first
boundx1 is assumed
to correspond to the nucleation threshold, so that the
La-grangian bound̄x1 does not satisfy (22) but:
dx̄1
dt= j (t) , x̄1(0) = x1 (25)
where j (t) is the growth law of the first bound due
tonucleation and given by the nucleation parameterization.
Theequations for the first Lagrangian bin therefore are written
as:
dN1
dt= J0(t) ,
dQ1i
dt= N1Ĩ1i + mi(x1, t)J0(t) (26)
where[m1(x1, t), . . . , mne (x1, t)] is the chemical
composi-tion of the nucleated particles, also given by the
nucleationprocess.
The Lagrangian formulation consists in solving Eqs. (22),(23)
and (26). In the next section we detail the variousnumerical
strategies to perform the time integration , whichis by far the
most challenging point in particle simulation.
Interpolation of Lagrangian boundaries One has to solvethe
equations for the characteristic curves in order to knowthe
boundaries of each bin. Notice that the c/e equations forboundaries
are similar to those for integrated quantities. In-deed, forj = 1,
. . . , nb and x̃j = ln(m̃j ), one gets fromEq. (23):
dx̃j
dt= H̃ j0 , H̃
j
0 =Ĩ
j
0
m̃j, (27)
In practice, in order to reduce the computational burden,
onetries to avoid solving boundary equations. An alternative isto
interpolate the bin boundaries from integrated quantities.
First method Koo et al. (2003) consists of utilizing the
ge-ometric mean of two adjacent bin :
for j = 2, . . . , nb , m̄j (t) =√
m̃j−1(t)m̃j (t) (28)
This algorithm would have a physical meaning if Eqs. (22)and
(27) were conserving formula (28), which is not the case.We have
therefore developed another algorithm.
Equations (22) and (27) are similar and thereforex̃j andx̄j
evolve in the same proportion given byλj (t) (j ≥ 2):
λj (t) =x̄j (t) − x̃j−1(t)x̃j (t) − x̃j−1(t)
(29)
λj (0) is known becausēxj (0) = xj . The time
integrationover[0, t] of Eqs. (22) and (27) gives forj ≥ 1:
x̄j (t) = xj + 1x̄j , 1x̄j =∫ t
0H
j
0 (t′) dt ′
x̃j (t) = x̃j (0) + 1x̃j , 1x̃j =∫ t
0H̃
j
0 (t′) dt ′ (30)
The variation of each boundarȳxj is then computed fromthat of
its two adjacent bins̃xj−1 andx̃j :
1x̄j ≃ (1 − λj (0))1x̃j−1 + λj (0)1x̃j (31)
where one assumes thatλj remains constant.
Redistribution on a fixed size gridUsing a Lagrangian ap-proach
for condensation/evaporation requires the redistribu-tion or
projection of number and mass concentrations ontothe fixed size
grid required by a 3D model or for coagula-tion.
Let N and(Qi)nei=1 be the integrated quantities of one La-
grangian bin after condensation/evaporation. We assume thatthis
Lagrangian bin is covered by two adjacent fixed bins la-belledj
andj + 1.
The redistribution algorithm must be conservative for thenumber
and mass distribution of speciesXi :
N = N j + N j+1 , Qi = Qji + Qj+1i (32)
Two algorithms have been developed:
www.atmos-chem-phys.net/7/1537/2007/ Atmos. Chem. Phys., 7,
1537–1547, 2007
-
1544 E. Debry et al.: A new SIze REsolved Aerosol Model
1. If x̄lo andx̄hi are the boundaries of the Lagrangian binafter
condensation/evaporation, the redistribution is per-formed as
follows for the number distribution and themass distribution of
speciesXi :
N j =x̄
jhi − x̄lo
x̄hi − x̄loN , Q
ji =
x̄jhi − x̄lo
x̄hi − x̄loQi
N j+1 =x̄hi − x̄j+1lox̄hi − x̄lo
N , Qj+1i =
x̄hi − x̄j+1lox̄hi − x̄lo
Qi (33)
The number and aerosol mass are redistributed in
equalproportions, depending upon the part of each fixed binscovered
by the Lagrangian one. Nevertheless this im-plies that the average
masses of fixed bins be equal tothe Lagrangian bin one (Q/N ) and
may fall out of fixedbin boundaries.
2. Another approach consists in conserving the averagemass.
Letm̃ = Q/N be the averaged Lagrangian bin,m̃j andm̃j+1 be
respectively the centered mass of binsj andj+1. The number
distribution and the mass distri-bution of speciesXi is
redistributed ensuring conserva-tion of relationsQj = m̃jN j
andQj+1 = m̃j+1N j+1,which together with (32) lead to the
algorithm:
N j =1 − m̃
m̃j+1
1 − m̃jm̃j+1
N , Qji =
m̃j+1m̃
− 1m̃j+1
m̃j− 1
Qi
N j+1 =1 − m̃
m̃j
1 − m̃j+1m̃j
N , Qj+1i =
1 − m̃jm̃
1 − m̃jm̃j+1
Qi (34)
This algorithm comes to the fitting method developedby Jacobson
in Jacobson (1997).
The first method takes advantage of the more
sophisticatedcomputation of bound sections but does not conserve
aver-age mass. The second method conserves average mass butmay
increase numerical diffusivity due to the lack of boundsection
information.
Both schemes are available in SIREAM.
3.3 Time integration
3.3.1 Coagulation
As coagulation is not a stiff process, we solve it by the
secondorder explicit scheme ETR (Explicit Trapezoidal Rule) withthe
sequence:
c̃n+1 = cn + 1tf (cn, tn)
cn+1 = cn +1t
2
(
f (cn, tn) + f (c̃n+1, tn+1))
(35)
with c = (N1, . . . , Nnb , Q11, . . . ,Qnb1 , . . . ,Q
1ne
, . . . ,Qnbne ).
3.3.2 Condensation/evaporation
Here,c = (Q11, . . . ,Q1ne , . . . ,Qnb1 , . . . ,Q
nbne )
T . nc = ne ×nb is the dimension ofc.
SIREAM offers three methods for solving
conden-sation/evaporation: a fully dynamic method that
treatsdynamic mass transfer for each bin, a bulk
equilibriumapproach, and a hybrid approach that combines the
twoprevious approaches.
Fully dynamic method Due to the wide range of timescalesrelated
to mass transfer, the system is stiff and implicit al-gorithms have
to be used. The second-order Rosenbrockscheme Verwer et al. (1999);
Djouad et al. (2002), ROS2,is applied for the time integration
:
cn+1 = cn +1tn
2(3k1 + k2)
[I − γ1tnJ (f )]k1 = f (cn, tn)[I − γ1tnJ (f )]k2 = f (c̃n+1,
tn+1) − 2k1 (36)
wherec̃n+1 = cn + 1tnk1 andγ = 1 + 1√2.This scheme requires the
computation of the Jacobian ma-
trix of f (a matrixnc ×nc) defined by[J (f )]kl = ∂fk
∂cl. f k is
thekth component of functionf andcl is thelth componentof c.
Let us writek = (i − 1)nb + j andl = (i′ − 1)nb + j ′wherei
andi′ label the semi-volatile species whilej andj ′
label the bins. The(kl)th element of the Jacobian matrix maythen
be written as
∂f k
∂cl=
∂Iji
∂Qj ′
i′
(37)
The derivation off k may be split into one linear part, dueto
mass conservation, and one non-linear part related to
thecoefficientaji , to the Kelvin effectη
j , and to the gas equi-librium concentration(ceqi )
j . The linear part is analyticallyderived :(
∂f k
∂cl
)
lin= −aji N
j ′ (38)
The non-linear part has to be differentiated by
numericalmethods, like the finite difference method :(
∂f k
∂cl
)
non−lin=
f k(. . . , cl(1 + εjac), . . .) − f k(. . . , cl, . .
.)clεjac
(39)
whereεjac is generally close to 10−8. During the numeri-cal
computation, the linear part is arbitrarily kept constant toavoid
deriving it twice.
A default option, advocated for 3D applications, is
toapproximate the Jacobian matrix by its diagonal. Themotivation
here is to reduce the CPU time.
Atmos. Chem. Phys., 7, 1537–1547, 2007
www.atmos-chem-phys.net/7/1537/2007/
-
E. Debry et al.: A new SIze REsolved Aerosol Model 1545
Hybrid resolution Solving the c/e system, even with animplicit
scheme, can be computationally inefficient. In or-der to lower the
stiffness, hybrid methods for condensa-tion/evaporation have been
developed Capaldo et al. (2000).The method consists in partitioning
the state vectorc into itsfast components (cf ) and its slow
components (cs) respec-tively:
dcs
dt= f s(cs, cf , t) , f f (cs, cf , t) = 0 (40)
The algebraic equation states that the fast part is a functionof
the slow part,cf (t)=g(cs(t), t). The time evolution of theslow
part is now governed by :
dcs
dt=f s
(
cs, g(cs(t), t), t
)
(41)
As cs gathers particle species and sizes which have a slowc/e
characteristic time, stiffness is substantially reduced.
The issue is now to determine whether particle sizes andspecies
are “slow” or “fast”. The spectral study of the c/e sys-tem Debry
and Sportisse (2006) indicates how to compute acutting diameterdc
between “slow” and “fast” species/sizes,such that the partitioning
consists of cutting the particle dis-tribution as follows: the
smallest bins are at equilibriumwhile the coarsest ones are
governed by kinetic mass trans-fer. The cutting diameter can be
computed by QSSA criteria,defined by :
QSSAji =
cgi − η
ji (c
eqi )
j
cgi + η
ji (c
eqi )
j(42)
for a given chemical speciesXi and one particle sizej .
Thecloser this ratio to unity, the closer the species and the
sizeare to equilibrium.
In practice all binsj for which(QSSAji )nei=1 is greater
than
one, the user parameterεQSSA (close to unity) will be
con-sidered fast and solved by an equilibrium equation. In
thefollowing we writejc as the bin corresponding to the
cuttingdiameter. Binjc is the largest fast bin and binjc + 1 is
thesmallest slow bin.
In SIREAM (to be used in 3D modeling), the defaultoption is a
fixed cutting diameter (1.25 or 2.5µm).
The thermodynamic equilibrium between the gas phaseand the fast
particle bins is now written for speciesXi as:
Kfi −
jc∑
j=1Q
ji − η
ki c
eqi (Q
k1, . . . ,Q
kne
) = 0 (43)
with Kfi =Ki −∑nb
j=jc+1 Qji the total mass of speciesXi for
fast bins.There are two approaches for solving this equilibrium:
the
bulk equilibrium approach and the size-resolved particle
ap-proach. For the size-resolved approach, we refer to Jacobson
et al. (1996) (with the use of the fixed point algorithm) and
toDebry and Sportisse (2006) (with a minimization procedure).
In SIREAM, the bulk equilibrium has been implemented(Pandis et
al. (1993)). It consists in merging all fast binsj ≤ jc into one
bin, referred as the “bulk” aerosol phase :
1 ≤ i ≤ ne , Bi =jc∑
j=1Q
ji (44)
The thermodynamic model ISORROPIA is then applied tothe “bulk”
aerosol phase(Bi)
nei=1 and one gets equilibrium
“bulk” concentrations(Beqi )nei=1 with theforward mode of
the
thermodynamics solver (global equilibrium).The variation from
initial to final “bulk” concentrations is
then redistributed among fast bins 1≤ k ≤ jc for speciesXiPandis
et al. (1993):
(Qki )eq = Qki + b
ki (B
eqi − Bi) , b
ki =
aki Nk
∑jcj=1 a
ji N
j(45)
This redistribution scheme is exact provided that the
particlecomposition is uniform over fast bins and that the
variationof the particle diameter can be neglected for fast bins
Debryand Sportisse (2006).
Bulk approach It is a special case of the hybrid approachwith
the cutting diameterjc = 1 (all bins are at equilibrium).
4 Implementation
The SIREAM module is written entirely in fortran 77.Its external
dependencies are the thermodynamic moduleISORROPIA (version 1.7
currently used)Nenes et al.(1998) and the VODE solver from the
ODEPACK ordinarydifferential equation package (“double precision”
version re-quired). ISORROPIA is not distributed with SIREAM andhas
to be retrieved by the user on the ISORROPIA web-site
(http://nenes.eas.gatech.edu/ISORROPIA/). The VODEsolver can be
retrieved from http://www.llnl.gov/CASC/software.html, but as this
solver is in the public domain wealso freely ship it together with
SIREAM.
5 Conclusions
We have summarized the main features of the aerosol modelSIREAM
(SIze REsolved Aerosol Model). SIREAM sim-ulates the GDE for
atmospheric particles and can be easilylinked to a
three-dimensional Chemistry-Transport-Model.Moreover, the physical
parameterizations used by SIREAMcan be easily modified. They are
currently hosted by thelibrary ATMODATA and shared by another
aerosol modelMAM, Sartelet et al. (2006).
The next development steps are related to the improvementof the
modeling of Secondary Organic Aerosol. The cur-rent
parameterization of SOA is limited because it does not
www.atmos-chem-phys.net/7/1537/2007/ Atmos. Chem. Phys., 7,
1537–1547, 2007
http://nenes.eas.gatech.edu/ISORROPIA/http://www.llnl.gov/CASC/software.htmlhttp://www.llnl.gov/CASC/software.html
-
1546 E. Debry et al.: A new SIze REsolved Aerosol Model
take into account the hydrophilic behavior of organic
speciesGriffin et al. (2002b,a); Pun et al. (2002). Furthermore
newgas precursors such as isoprene and sesquiterpene should
beadded.
The modularity of SIREAM will be also strengthened byadding new
alternative parameterizations (such as other ther-modynamics models
or simplified aqueous-phase chemicalmechanisms) and new numerical
algorithms (especially fortime integration of
condensation/evaporation).
A further step is also the extension toexternally
mixedaerosol.
Acknowledgements. Part of this project has been funded by
theFrench Research Program, Primequal-Predit, in the framework
ofthe PAM Project (Multiphase Air Pollution). Some of the
authors(K. Fahey, K. Sartelet and M. Tombette) have been partially
fundedby the Region Ile de France.
Edited by: R. MacKenzie
References
Adams, P. and Seinfeld, J.: Predicting global aerosol size
distribu-tions in general circulation models, J. Geophys. Res.,
107, 4370,2002.
Binkowski, F. and Roselle, S.: Models-3 Community Multiscale
AirQuality (CMAQ) model aerosol component. 1. Model descrip-tion,
J. Geophys. Res., 108, doi:10.1029/2001JD001 409, 2003.
Brown, P., Byrne, G., and Hindmarsh, A.: VODE: A Variable
Coef-ficient ODE Solver, SIAM J. on Sci. and Stat. Comp., 10,
1038–1051, 1989.
Byun, P. and Schere, K.: EPA’s Third Generation Air Quality
Mod-eling System: Description of the Models-3 Community, J.
Mech.Review, 2004.
Capaldo, K., Pilinis, C., and Pandis, S.: A computationally
efficienthybrid approach for dynamic gas/aerosol transfer in air
qualitymodels, Atmos. Environ., 34, 3617–3627, 2000.
Dahneke, B.: Theory of Dispersed Multiphase Flow, Academicpress,
New York, 1983.
Debry, E.: Numerical simulation of an atmospheric aerosol
distri-bution, Ph.D. thesis, ENPC, CEREA, in French, 2004.
Debry, E. and Sportisse, B.: Solving aerosol coagulation with
size-binning methods, Appl. Numer. Math., accepted, 2005a.
Debry, E. and Sportisse, B.: Numerical simulation of the
Gen-eral Dynamics Equation (GDE) for aerosols with two
collocationmethods, Appl. Numer. Math., accepted, 2005b.
Debry, E. and Sportisse, B.: Reduction of the
condensa-tion/evaporation dynamics for atmospheric aerosols:
theoreticaland numerical investigation of hybrid methods, J.
Aerosol Sci.,37, 950–966, 2006.
Djouad, R., Sportisse, B., and Audiffren, N.: Numerical
simulationof aqueous-phase atmospheric models : use of a
non-autonomousRosenbrock method, Atmos.Environ., 36, 873–879,
2002.
Fahey, K.: Cloud and fog processing of aerosols: modeling the
evo-lution of atmospheric species in the aqueous phase, Ph.D.
thesis,Carnegie Mellon University, 2003.
Fahey, K. and Pandis, S.: Optimizing model performance:
variablesize resolution in cloud chemistry modeling, Atmos.
Environ.,35, 4471–4478, 2001.
Fernndez-Daz, J., Gonzlez-Pola Muiz, C., Rodrguez Braa, M.,
Ar-ganza Garca, B., and Garca Nieto, P.: A modified
semi-implicitmethod to obtain the evolution of an aerosol by
coagulation, At-mos. Environ., 34, 4301–4314, 2000.
Gaydos, T., Koo, B., Pandis, S., and Chock, D.: Development
andapplication of an efficient moving sectional approach for the
so-lution of the atmospheric aerosol condensation/evaporation
equa-tions., Atmos. Environ., 37, 3303–3316, 2003.
Gelbard, F., Tambour, Y., and Seinfeld, J.: Sectional
Representa-tions for Simulating Aerosol Dynamics, Journal of collod
andInterface Science, 76, 541–556, 1980.
Gerber, H.: Relative-humidity parameterization of the Navy
aerosolmodel (NAM), Tech. Rep. 8956, Natl. Res. Lab.
WashingtonD.C., 1985.
Gong, S., Barrie, L.A., and Blanchet, J.-P., et al.: Canadian
AerosolModule: a size-segregated simulation of atmospheric
aerosolprocesses for climate and air quality models. I Module
devel-opment, J. Geophys. Res., 108, 2003.
Griffin, R., Dabdub, D., Kleeman, M., Fraser, M., Cass, G.,
andSeinfeld, J.: Secondary organic aerosol 3. Urban/regional
scalemodel of size- and composition-resolved aerosols, J.
Geophys.Res., 107, 2002a.
Griffin, R., Dabdub, D., and Seinfeld, J.: Secondary organic
aerosol1. Atmospheric chemical mechanism for production of
molecularconstituents, J. Geophys. Res., 107, 2002b.
Jacob, D.: Heterogeneous chemistry and tropospheric ozone,
At-mos. Environ, 34, 2131–2159, 2000.
Jacobson, M.: Development and application of a new air
pollutionmodeling system - PART II . aerosol module structure and
de-sign, Atmospheric Environment, 31, 131–144, 1997.
Jacobson, M.: Analysis of aerosol interactions with numerical
tech-niques for solving coagulation, nucleation, condensation,
disso-lution, and reversible chemistry among multiple size
distribu-tions, J. Geophys. Res., 107, 2002.
Jacobson, M., Turco, R., Jensen, E., and Toon, O.: Modeling
coagu-lation among particles of different composition and size,
Atmos.Environ., 28, 1327–1338, 1994.
Jacobson, M., Tabazadeh, A., and Turco, R.: Simulating
equi-librium within aerosols and nonequilibrium between gases
andaerosols, J. Geophys. Res., 101, 9079–9091, American
Geophys-ical Union, 1996.
Koo, B., Gaydos, T., and Pandis, S.: Evaluation of the
Equilibrium,Dynamic, and Hybrid aerosol modeling approaches,
Aerosol Sci-ence Technology, 37, 53–64, 2003.
Korhonen, H., Lehtinen, K., Pirjola, L., Napari, I., Vehka-maki,
H., Noppel, M., and Kulmala, M.: Simulation of at-mospheric
nucleation mode: a comparison of nucleation mod-els and size
distribution representations, J. Geophys. Res.,
108,doi:10.1029/2002JD003 305, 2003.
Mallet, V. and Sportisse, B.: Data processing and
parameterizationsin atmospheric chemistry and physics: the AtmoData
library,Tech. Rep. 2005-12, ENPC/CEREA, 2005.
Mallet, V. and Sportisse, B.: Toward ensemble-based
air-qualityforecasts, J. Geophys. Res., accepted, 2006.
Mallet, V., Qúelo, D., Sportisse, B., Debry,É., Korsakissok,
I., Wu,L., Roustan, Y., Sartelet, K., Tombette, M., de Biasi, M.
A., and
Atmos. Chem. Phys., 7, 1537–1547, 2007
www.atmos-chem-phys.net/7/1537/2007/
-
E. Debry et al.: A new SIze REsolved Aerosol Model 1547
Foudhil, H.: Technical note: The air quality modeling
systemPolyphemus, Tech. Rep. 8, CEREA, 2007.
Meng, Z., Dabdub, D., and Seinfeld, J.: Size-resolved and
chemi-cally resolved model of atmospheric aerosol dynamics, J.
Geo-phys. Res., 103, 3419–3435, 1998.
Napari, I., Noppel, M., Vehkamki, H., and Kulmala,
M.:Parametrization of ternary nucleation rates for
H2O4-NH3-H2Ovapors, J. Geophys. Res., 107, 2002.
Nenes, A., Pandis, S., and Pilinis, C.: ISORROPIA : A new
Ther-modynamic Equilibrium Model for Multicomponent
InorganicAerosols, Aquatic geochemistry, 4, 123–152, 1998.
Pandis, S., Wexler, A., and Seinfeld, J.: Secondary organic
aerosolformation and transport -II. Predicting the ambient
secondaryorganic aerosol size distribution, Atmos. Environ., 27A,
2403–2416, 1993.
Pilinis, C. and Seinfeld, J.: Development and evaluation of an
eu-lerian photochemical gas-aerosol model, Atmos. Environ.,
22,1985–2001, 1988.
Pilinis, C., Capaldo, K., Nenes, A., and Pandis, S.: MADM - a
newmulti-component Aerosol Dynamic Model, Aerosol Science
andTechnology, 32, 482–502, 2000.
Pruppacher, H. and Klett, J.: Microphysics of Clouds and
Precipi-tation, Kluwer Academic Publishers, 1998.
Pun, B., Griffin, R., Seigneur, C., and Seinfeld, J.: Secondary
or-ganic aerosol 2. Thermodynamic model for gas/particle
partition-ing of molecular constituents, J. Geophys. Res., 107,
2002.
Sandu, A. and Borden, C.: A framework for the numerical
treatmentof aerosol dynamics, Appl. Numer. Math., 45, 475–497,
2003.
Sartelet, K. N., Hayami, H., Albriet, B., and Sportisse, B.:
Develop-ment and preliminary validation of a Modal Aerosol Model
fortropospheric chemistry: MAM, Aerosol Sci. and Technol.,
40,118–127, 2006.
Schell, B.: Die Behandlung sekundrer organischer Aerosole
ineinem komplexen Chemie-Transport-Modell, Ph.D. thesis, Univ.Kln,
2000.
Schell, B., Ackermann, I., Hass, H., Binkowski, F., and Ebel,
A.:Modeling the formation of secondary organic aerosol within
acomprehensive air quality model system, J. Geophys. Res., 106,28
275, 2001a.
Schell, B., Ackermann, I., Hass, H., Binkowski, F., and Ebel,
A.:Modeling the formation of secondary organic aerosol within
acomprehensive air quality model system, J. Geophys. Res., 106,28
275–28 293, 2001b.
Seigneur, C.: Current status of air quality modeling for
particulatematter, J. Air Waste Manage. Assoc., 51, 1508–1521,
2001.
Seinfeld, J. and Pandis, S.: Atmospheric chemistry and
Physics,Wiley-interscience, 1998.
Sportisse, B., Debry, E., Fahey, K., Roustan, Y., Sartelet, K.,
andTombette, M.: PAM project (Multiphase Air Pollution):
descrip-tion of the aerosol models SIREAM and MAM, Tech. Rep.
2006-08, CEREA, available at
http://www.enpc.fr/cerea/polyphemus,2006.
Spracklen, D., Pringle, K., Carslaw, K., Chipperfield, M.,
andMann, G.: A global off-line model of size-resolved
aerosolmicrophysics: I. Model development and prediction of
aerosolproperties, Atmos. Environ., 5, 2227–2252, 2005.
Stockwell, W., Kirchner, F., Kuhn, M., and Seefeld, S.: A
newmechanism for regional atmospheric chemistry modeling.,
J.Geophys. Res., 95, 16 343–16 367, 1997.
Strader, R., Gurciullo, C., Pandis, S., Kumar, N., and Lurmann,
F.:Development of gas-phase chemistry, secondary organic aerosoland
aqueous-phase chemistry modules for PM modeling, Tech.rep., STI,
1998.
Vehkamki, H., Kulmala, M., Napari, I., K.E.J., L., Timmreck,
C.,Noppel, M., and Laaksonen, A.: An improved parameteriza-tion for
sulfuric acid-water nucleation rates for tropospheric
andstratospheric conditions, J. Geophys. Res., 107, 4622, 2002.
Verwer, J., Spee, E., Blom, J., and Hundsdorfer, W.: A second
orderRosenbrock method applied to photochemical dispersion
prob-lem, SIAM J. SCI. COMPUT., 20, 1456–1480, 1999.
Wexler, A., Lurmann, W., and Seinfeld, J.: Modelling urban
andregional aerosols -I. model development, Atmos. Environ.,
28,531–546, 1994.
Whitby, E. and McMurry, P.: Modal Aerosol Dynamics
Modeling,Aerosol Science and Technology, 27, 673–688, 1997.
Zhang, Y., Seinfeld, J., Jacobson, M., Clegg, S., and Binkowski,
F.:A comparative review of inorganic aerosol thermodynamic
equi-librium modules: similarities, differences and their likely
causes,Atmos. Environ., 1998.
Zhang, Y., Pun, B., Vijayaraghavan, K., Wu, S., Seigneur, C.,
Pan-dis, S., Jacobson, M., Nenes, A., and Seinfeld, J.:
Developmentand application of the Model of Aerosol Dynamics,
Reaction,Ionization and Dissolution (MADRID), J. Geophys. Res.,
109,2004.
www.atmos-chem-phys.net/7/1537/2007/ Atmos. Chem. Phys., 7,
1537–1547, 2007
http://www.enpc.fr/cerea/polyphemus