You are on page 1of 16

Science of the Total Environment 610-611 (2017) 175–190

Contents lists available at ScienceDirect

Science of the Total Environment


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

A GIS-based atmospheric dispersion model for pollutants emitted by


complex source areas
Sergio Teggi a, * , Sofia Costanzini a , Grazia Ghermandi a , Carlotta Malagoli b , Marco Vinceti b
a
Department of Engineering Enzo Ferrari, University of Modena and Reggio Emilia, Via P. Vivarelli 10, Modena 41125, Italy
b
Environmental, Genetic, and Nutritional Epidemiology Research Center – CREAGEN, Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia,
Modena, Italy

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Air dispersion model for pollutants


emitted by very complex source areas
• Thousands of receptors and source
areas with hundreds of vertices and
holes
• Ground level concentration and dry
deposition flux in the shapefile format
• CAREA can be used as a tool in a GIS
environment.
• Good agreement with AERMOD simu-
lations

A R T I C L E I N F O A B S T R A C T

Article history: Gaussian dispersion models are widely used to simulate the concentrations and deposition fluxes of pollu-
Received 11 May 2017 tants emitted by source areas. Very often, the calculation time limits the number of sources and receptors
Received in revised form 5 July 2017 and the geometry of the sources must be simple and without holes. This paper presents CAREA, a new GIS-
Accepted 22 July 2017 based Gaussian model for complex source areas. CAREA was coded in the Python language, and is largely
Available online xxxx
based on a simplified formulation of the very popular and recognized AERMOD model. The model allows
users to define in a GIS environment thousands of gridded or scattered receptors and thousands of complex
Keywords:
sources with hundreds of vertices and holes. CAREA computes ground level, or near ground level, concentra-
Air dispersion models
tions and dry deposition fluxes of pollutants. The input/output and the runs of the model can be completely
Population exposure
AERMOD managed in GIS environment (e.g. inside a GIS project). The paper presents the CAREA formulation and its
Atmospheric emission from complex applications to very complex test cases. The tests shows that the processing time are satisfactory and that
source areas the definition of sources and receptors and the output retrieval are quite easy in a GIS environment. CAREA
GIS and AERMOD are compared using simple and reproducible test cases. The comparison shows that CAREA
Pesticide satisfactorily reproduces AERMOD simulations and is considerably faster than AERMOD.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction emitted from a variety of sources. These models have been widely
used in many environmental fields. Typical examples on a local scale
Atmospheric dispersion models are designed to predict ground-
are studies of the pollutants emitted from industrial plants (Mazur et
level concentrations and surface deposition fluxes of pollutants
al., 2009; Llanos et al., 2010; Ghermandi et al., 2015, 2013; Giua et al.,
2014; Bowler et al., 2016), incinerators (Morra et al., 2009; Ghermandi
et al., 2012; Candela et al., 2015, 2013), mining, open pits and quarries
* Corresponding author.
E-mail address: sergio.teggi@unimore.it (S. Teggi). (Huertas et al., 2014; Robins et al., 2012; Tartakovsky et al., 2016a,b)

http://dx.doi.org/10.1016/j.scitotenv.2017.07.196
0048-9697/© 2017 Elsevier B.V. All rights reserved.
176 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Fox, 2005). Among the Eulerian models, the most the widely used are
List of symbols the steady-state Gaussian models.
The combined use of air dispersion models and GIS (Geographic
C Concentration at the receptor Information System) techniques is very common in research focusing
Cs Concentration at the receptor due to a source strip on the effects produced by aerial pollutants on a regional-local scale
Dg1 First buffer distance from progressive grid (e.g. Owen et al., 1999; Beyea, 1999; Canepa et al., 2007; Zhang et
Dg2 Second buffer distance from progressive grid al., 2008; Maantay et al., 2009; Gulliver and Briggs, 2011; Heckel
dp Particle diameter and Lemasters, 2011; Danuso et al., 2015). However, there are often
FA2 Factor 2 index serious problems due to the formats often adopted by GIS (vector
FB Fractional bias index objects) and the formats adopted by atmospheric dispersion models
Fd Dry deposition flux (tabulated text data) for the input/output data describing sources,
fp Fraction of particles with diameter less than dp receptors, concentrations and deposition fluxes.
Fq Source depletion factor due to deposition The authors were recently involved in an epidemiological investi-
Fy Lateral dispersion term gation aiming to assess the risk to populations exposed to pesticides
Fys Lateral dispersion term for a strip applied by airblasts from farming processes (Malagoli et al., 2016).
GMB Geometric mean bias Considering that in the European Community aerial (aircraft, heli-
GV Geometric variance copters) spraying of pesticides is limited to very special cases
IA Index of agreement (Directive 2009/128/EC of The European Parliament), airblast appli-
L Monin-Obukhov length cations represent a very frequent source of population exposure to
mg1 First multiplication factor for progressive grids pesticides. Population exposure was evaluated both on a target-source
mg2 Second multiplication factor for progressive grids proximity analysis (results shown in Malagoli et al., 2016) and on the
Q Source (point) emission rate atmospheric dispersion of the pollutant emitted by agricultural fields
qs Emission rate of source area to the exposed receptors (results not yet published). This is a typical
R Pearson’s linear correlation coefficient “air dispersion from source areas” problem and there are numerous
ra Aerodynamic resistance models that can be applied to solve it (Bilanin et al., 1989; Theobald
rb Quasilaminar resistance for bulk surface et al., 2012; Nimmatoori and Kumar, 2013a; Ghenai and Lin, 2006),
rc Bulk surface resistance including the very popular AERMOD Gaussian model (Cimorelli et
rp Quasilaminar sublayer resistance al., 2005, 2004). However, application of AERMOD, and other similar
u Wind velocity models, was very difficult because of the following:
u* Friction velocity
w* Convective velocity scale - A very large number (hundreds) of source areas (polygons)
vd Deposition velocity had to be considered. AERMOD requires a description section
Vt Vertical dispersion term f (geometry, emission rate) for each polygon in the input file (U.S.
Vts Vertical dispersion term for a strip Environmental Protection Agency, 2004b);
z0 Surface roughness length - The polygons representing the sources were complex: tens of
zi Mixing height vertices, including holes, convex or concave. In AERMOD the
Dg Grid spacing geometry of a polygonal source can be described using only a few
Ds Width of the source strips (up to twenty) vertices (U.S. Environmental Protection Agency,
sv Root-mean-square lateral turbulence velocity 2004b);
s vc Convective portion of s v - Sources, receptors and outputs (concentrations and deposition
s vm Mechanical portion of s v fluxes) are defined and managed in a GIS project, consequently
sy Lateral dispersion coefficient the transformation of GIS to AERMOD format and vice versa was
sz Vertical dispersion coefficient required;
s ze Elevated contribution to s z - The high number of sources and receptors resulted in very high
s zg Ground contribution to s z computation time (days);
- The epidemiological study required ground concentrations and
deposition fluxes, representing “surface level” solution of the
more complex 3D outputs of AERMOD, which resulted in a rising
of computation times.
chemical accidents (Till et al., 2014; Brzozowska, 2014), vehicular
traffic (Clarke et al., 2004; Pospisil et al., 2004; Gulliver and Briggs, An AERMOD simulation of this type can be achieved by simplifying
2011; Ghermandi et al., 2014; Tadano et al., 2014), aircraft traffic the polygonal sources, considering sub sets of sources and running
(Pecorari et al., 2016; Penn et al., 2015), and farming processes (Beyea, AERMOD for each sub set. However, there remains the GIS to AERMOD
1999; Gil and Sinfort, 2005; Canepa et al., 2007; Bisignanesi and Borgas, format conversion task. This approach was tested but proved to be
2007; Theobald et al., 2012; Nimmatoori and Kumar, 2013a,b; Danuso very problematic.
et al., 2015; Pivato et al., 2015). The authors solved this problem by implementing CAREA, a new,
Air dispersion formulation can vary widely between models. The GIS-based Gaussian dispersion model. CAREA was devised to operate
most widely used methods are the Lagrangian and the Eulerian in the following simplified scenario: gaseous or particulate pollu-
approaches (Collett and Oduyemi, 1997; Holmes and Morawska, tants emitted at ground-level from one or more complex source
2006). Lagrangian models describe atmospheric transport and dis- areas; receptors located close to the ground (e.g. human receptors),
persion (turbulence) more completely than Eulerian models, but they far from buildings, on horizontal or near-horizontal terrain. Though
require a complex meteorological data set (e.g. turbulence preproces- simple, this scenario is actually common, for example, in agricultural
sors), an in depth knowledge of the atmospheric processes and long processes (Bilanin et al., 1989; Teske et al., 2002; Batiha et al., 2014,
computation times. For these reasons Eulerian models are more fre- 2010; Nimmatoori and Kumar, 2013a,b; Pivato et al., 2015; van den
quently adopted than Lagrangian models in numerous fields like long Berg et al., 2016), emission from water bodies (Martinez et al., 2015),
term population exposure and regulatory purposes (see for example contaminated soils and accidents (Bauer, 2013; Llanos et al., 2011),
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 177

mining and quarries (Tartakovsky et al., 2016a,b). This aspect makes the stable boundary layer (SBL), meandering and random (pancake
CAREA potentially useful for many studies requiring the simulation plume) dispersion in low wind conditions.
of the concentrations and of the deposition of substances emitted by Each plume formulation also considers terrain and mixed-layer-
areal sources. top reflections, asymmetry of the vertical distribution dispersion
The main characteristics of CAREA are as follows: coefficient (modelled with a bi-Gaussian distribution) and, option-
ally, plume depletion due to wet and dry deposition. Building effects
- The Gaussian dispersion formulation is mainly based on a sim- (e.g. plume downwash) and plume rise due to buoyancy effects are
plification of the AERMOD formulation; also considered. Deposition and surface flux of gaseous or particulate
- The model simulates mean concentrations (over the simulation pollutants are modelled by considering dry and wet deposition.
period) and total dry deposition fluxes due to emissions from As introduced, CAREA was devised to operate in the follow-
source areas over flat or undulating zones; ing simplified scenario: gaseous or particulate pollutants emitted
- A large number (dependent only on hardware capabilities) of at ground-level from one or more complex source areas; recep-
polygons and receptors can be considered; tors located close to the ground (e.g. human receptors), far from
- Source polygons can be very complex, with hundreds of vertices buildings, on horizontal or near-horizontal terrain.
and holes; The use of AERMOD in this scenario can encounter some difficul-
- Most of the simulation steps including the definition of source ties regarding source definition. With AERMOD a source is defined as
polygons, source emission rates, receptor locations, and pre- a polygon of up to 20 sides (U.S. Environmental Protection Agency,
dicted ground-level concentrations and deposition fluxes are 2004b). In many cases, for example when polygons are extracted
carried out in the GIS environment; from digital maps, the polygonal sources are very complex, with
- Source decomposition and receptor rotation, required by the holes and a large number of sides. In such cases users have to sim-
Gaussian formulation, are achieved using object (vector shapes) plify the geometry of the polygons, extract the vertices, and include
oriented modules; the vertex coordinates and emission factors (optional) in the input
- The model is considerably faster than AERMOD; text file (a line for each vertex). The receptors have to be treated in
- The model was coded in the Python language, can directly a similar way: in the case of a very large number of sparse receptors
interacts with the open-source QGIS and with ArcMAP® and (e.g. hundreds of human subjects exposed to the pollutant) the user
also with other GIS software (not tested); must obtain the coordinates for each receptor and include them in
- CAREA is available on request. the input file (a text line for each receptor). Moreover, the AERMOD
outputs need post-processing routines in order to be inserted into a
digital map. In CAREA the sources (polygons) and receptors are vec-
Other software designed to couple AERMOD and GIS have been tor objects that can be directly defined (input) and managed (output)
developed (e.g. Maantay et al., 2009), but, as far as we know, they in the GIS environment, and the emission rate of each source can be
offer only a loose-coupling solution to the problem, being mainly specified as an attribute of the object.
dedicated to managing the input/output flows between AERMOD In the CAREA scenario, computing time can be reduced by simpli-
and GIS. CAREA is a dispersion model designed for easy interfacing fying the AERMOD formulation, which is achieved by CAREA on the
with GIS software. For a specific GIS software, coupling with CAREA basis of the following assumptions:
could be achieved by implementing a specific (GIS dependent) GUI
(Graphical User Interface). Using such a GUI, CAREA could be consid- i Plume buoyancy is only a key aspect for emissions from stacks
ered as an air dispersion tool for the GIS package. We made some test and so plume rise due to buoyancy effects is not considered;
with the open-source QGIS and with ArcMap® but no GUIs ready to ii The pollutants are emitted from the ground, therefore plume
be distributed have been implemented to date. penetration and plume entrainment in the stable boundary
The general structure of CAREA and its links to the AERMOD layer are not relevant and so not considered;
are presented in Section 2. In Section 3 the formulation of CAREA iii Plume meandering and pancake plume contribution (low
and its implementation in Python is described. Finally, in Section 4 wind conditions) are not considered;
CAREA and AERMOD are compared (concentration, deposition flux iv Variation in the vertical concentration distribution due to
and processing time) using a reproducible experiment (the data sets deposition is not considered;
to reproduce the experiment are available on request), moreover, the v Vertical inhomogeneity of dispersion coefficients is not
CAREA processing time is evaluated in more complex scenarios using considered;
also a real case study. vi Vertical variation in wind velocity is considered only to
calculate ground wind velocity values from measurements
taken at different heights (Section 3.7);
vii Wet deposition is not considered.
2. CAREA and AERMOD
The main steps in a generic CAREA simulation within a GIS project
The CAREA model is largely based on the air dispersion and can be summarized as follows (Fig. 1):
deposition formulation used by AERMOD.
AERMOD is among the models preferred/recommended by the 1. The user defines the sources (complex polygons) and their
U.S. Environmental Protection Agency (EPA) for air quality regulatory emission rates (optional) in the GIS environment and then
purposes (Fox, 2005). It is widely used and well known; extensive exports them in vector format;
literature describing its formulation and algorithms, applications, 2. The user defines the receptor locations (optional) in the GIS
comparisons and validations is available. A detailed discussion and environment and then exports them in vector format (optional);
description of AERMOD is not among the aims of this paper, and so 3. The user prepares the meteorological data set;
this section provides the details strictly pertinent to CAREA. 4. The user compiles an experiment input text file containing
AERMOD evaluates the concentrations emitted from point, line, pollutant information, computing options, and input/output
area and volume sources as a weighted sum of Gaussian plume con- directives (this can be done in GIS environment using a GUI
tributions, each one representing a dispersion phenomenon scaled implemented for the specific GIS software);
by a deposition factor. The various plumes define direct contribu- 5. The model is run (this can be done in GIS environment using a
tion, elevated terrain contribution, entrainment and penetration in GUI implemented for the specific GIS software);
178 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Eq. (1) is specific for a point source. Area sources are first decom-
posed into elementary strips perpendicular to the wind direction
(Section 3.3) and then the total concentration is computed as the sum
of the contribution of each strip:


Ns Ns 
 
q
C(xr , yr , zr ) = Cs (xr , yr ) = Fy,s • Vt,s (2)
u
i=1 i=1

where the subscript s indicates an elementary strip, Ns is the num-


ber of strips, q is the surface emission rate (g s −1 m −2 ), Fy,s is the
lateral dispersion term of the strip (m −1 ) (Section 3.4.2) and Vt,s is
the vertical term of the strip (m −1 ) (Sections 3.4.1 and 3.5).
In the general case of multiple area sources the total concentration
is obtained by the contributions of each area given by Eq. (2) for all
the sources, which is the basic equation used in CAREA.
Eq. (2) describes the concentration for each time step (usually 1 h)
of the simulation. The average concentration and the deposition flux
at the surface (Section 3.5) for the period of simulation are output by
the model.

3.2. Sources

Fig. 1. GIS and CAREA workflow. Polygons representing the source areas are input to CAREA as
vector objects in shapefile format (ESRI, 1988). Shapefile is a format
widely used in GIS applications and most of the principal GIS
6. The user obtains the concentrations and dry deposition (if software are able to manage it.
requested) of the pollutant at the receptor locations in vector The user can either build the polygonal sources in the GIS
format. The output is uploaded to the GIS project for further environment or use predefined polygons from an external data set
analysis. (e.g. WCS servers, WMS servers). The vector format permits elabora-
tion of a large number of very complex polygons that can be convex
3. CAREA formulation and implementation in Python or concave, with hundreds of vertices and holes. This is an important
advantage over AERMOD, in which a polygon can be only defined
This section presents the formulation of CAREA and the most with a maximum of 20 vertices and holes are not permitted (U.S.
important differences compared with AERMOD, together with some Environmental Protection Agency, 2004b).
details about the coding of the algorithm in the Python language. The user can define a specific emission rate q (g s −1 m −2 ) for each
AERMOD computes concentrations and deposition fluxes as a polygon as an attribute of the object, or set a single value for all the
weighted sums of plumes representing different dispersion phenom- sources in the case specification text file (Section 3.10).
ena: on the basis of assumptions i, ii and iii (Section 2) only the direct Fig. 2 shows an example of source polygons defined for an area
contribution is considered in CAREA. of the Po Valley (Italy). The wider polygons, over the Po river, were
In order to take into account the vertical inhomogeneity of wind drawn by the user using a GIS editing tool. The other 20 polygons are
velocity, total vertical turbulence, total lateral turbulence (u, s vT , s wT ) branches of minor rivers/channels (20 m–30 m wide) extracted from
and various other dispersion coefficients AERMOD transforms these a public shapefile of the National River database. The emission rate of
parameters into effective parameters (mean values in the dispersion the polygons is variable and is set by the user in the object attribute
layer, ũ, s̃vT , s̃wT ) (Cimorelli et al., 2004). To simplify the computation table.
(based on assumptions v and vi in Section 2) CAREA applies ground- Fig. 3 shows another example; 113 orchards were extracted from
level values and hereafter the tilde symbol is not used in notation a land use database in shapefile format and elected as sources of
regarding both the models. pesticides. The detail in the lower left of the figure highlights the
complexity of the polygons.
3.1. Basic equations
3.3. Source decomposition and rotation
Based on the CAREA assumptions and the CAREA scenario, the
general relation of the concentration C at a receptor r resulting from The application of Eq. (2) requires that the source polygons are
a point source (direct plume only) used by AERMOD reduces to decomposed into strips perpendicular to the wind vector. In CAREA
each polygon is first rotated toward the downwind direction and
Q then decomposed into elementary strips of constant width set by the
C(xr , yr , zr ) = Fy • Vt (1)
u user (Ds , Fig. 4). Polygon rotations and decompositions are carried
out using the Shapely Python library (Gillies et al., 2015) which
where xr , yr and zr are the horizontal coordinates of r in the reference
permits the decomposition of very complex polygons.
system centered on the source (ground level) and with the x axis
oriented in the downwind direction, Q is the source emission rate
(g s −1 ), u is the effective wind speed (m s −1 ), Fy is the lateral 3.4. Concentration without dry deposition
dispersion term (dimensionless) and Vt is the vertical term (m −1 ).
If dry deposition is considered, the vertical term also includes the 3.4.1. Vertical term, Vt,s
effect of plume depletion due to deposition, leading to two different The equation used by CAREA for the computation of the vertical
formulations (with and without dry deposition) of Vt . term Vt,s of each strip is obtained from Eq. (1-50) in the ISC3 model
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 179

algorithm documentation (U.S. Environmental Protection Agency, reflections (Ni , set by the user) is infinity, in practice it is observed
1995) for emissions at ground level (z = 0): that after a few reflections the additional contributions are very
small.
⎧ When the hypothesis and constraints reported in Section 2 are
√  
2 ⎨ assumed the algorithm used by AERMOD for vertical term calculation
Ni
z2  (zr − 2 • zi )2
Vt,s =√ exp − r 2 + exp − reduces to Eq. (3).
psz ⎩ 2sz 2sz2
j=1
 ⎫
(zr + 2 • zi )2 ⎬
Ni

+ exp − (3) 3.4.2. Lateral term, Fy,s
2sz2 ⎭
j=1 The general formulation adopted to describe the lateral term
in Gaussian dispersion models for a point source, Fy , is (U.S.
Environmental Protection Agency, 1995)
where zi is the mixed layer height (Section 3.7), s z the vertical
dispersion coefficient computed at the distance xr from the strip. 
Each term of the sum accounts for a plume multiple reflection by 1 y2
Fy (xr , yr ) = √ exp − r 2 (4)
the surface or from the mixed layer top. In theory, the number of 2psy 2sy

Fig. 2. Top: Example of source definition. The source is composed by 21 polygons (red) with variable emission rate in an area in the Po Valley (Italy). The wider polygon has been
drawn by the user using a GIS editing tool. The other polygons are branches of rivers/channels extracted from the National Rivers database (shapefile) and with variable width
(about 20 m–30 m). Top: input source (polygon). Bottom: Progressive grid of receptors (50 m, 100 m, 200 m) built around the polygons above. (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)
180 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Fig. 3. Main figure: 113 polygons (pink) corresponding to orchards in Catania (Italy) considered as pollutant (pesticide) sources. The circles indicate the receptor locations. Source
polygons and point receptors were provided in shapefile format by the user (extracted from land use database and health registry database). Upper left: detail of receptors with
ground concentration (labels) computed by CAREA. Lower left: detail of source polygons, colors are used to differentiate the polygons. (For interpretation of the references to color
in this figure legend, the reader is referred to the web version of this article.)

Since xr and yr represent the distance of the receptor from the width is set narrow enough (the default value is 20 m) to assume that
source, the lateral term for a strip is obtained with a double integra- the lateral term Fy (xr , yr ) does not change significantly in the x direc-
tion of the source position (x0 , y0 ): tion (Smith, 1995). The value at the center of the strip is assumed
as representative for the whole strip. In the y direction CAREA uses
x2,s y2,s the same analytical solution as AERMOD and ISC3 (Eq. (1-66) in U.S.
Fy,s (xr , yr ) = Fy (xr , yr )dy0 dx0 (5) Environmental Protection Agency, 1995).
x1,s y1,s
Based on these assumptions, the CAREA equation used for the
lateral term is in practice in the Smith (1995) formulation
where x1,s , x2,s , y1,s and y2,s are the end coordinates of the strip.
In AERMOD the integration for the x direction is approximated     
using the numerical method adopted by ISC3 (Eq. (1-67) in U.S. Envi- Ds yr − y1,s yr − y2,s
Fy,s (xr , yr ) = erf √ − erf √ (6)
ronmental Protection Agency, 1995). In the CAREA model the strips 2 2sy 2sy

3.4.3. Vertical dispersion coefficients, s y and s z


AERMOD computes the dispersion coefficients s y and s z as a com-
bination (squared sum) of the dispersion coefficients resulting from
ambient turbulence, and the dispersion coefficient that describes
turbulence induced by plume buoyancy (Cimorelli et al., 2004). AER-
MOD also applies additional algorithms to take building induced
turbulence into account. Considering the CAREA assumptions and the
CAREA scenario (Section 2) only dispersion due to ambient turbu-
lence is considered relevant. For the computation of these coefficients
CAREA applies the AERMOD formulation with some simplifications.

3.4.3.1. s z in the convective boudary layer (CBL, L < 0). In AERMOD


the CBL s z is given by the squared sum of an elevated contribution
(s ze ) and a surface contribution (s zg ) (Eq. (85) in Cimorelli et al.,
2004):

Fig. 4. Source decompositions. Left: input source (polygon). Right: source rotation
toward the downwind direction (u) and decomposition into strips. sz2 = sze
2 2
+ szg (7)
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 181

Considering emissions and receptors at ground level, in CAREA the and (1-57b) in U.S. Environmental Protection Agency, 1995) is not
AERMOD formulation (Eqs. (34), (35), (36), (86), (87) in Cimorelli et considered.
al., 2004) is simplified into the equations When dry deposition is considered, the source factor (qs ) is
multiplied by a source depletion factor (Fq ):
u∗ (xr − xs )  x 
sze = 0.78 (8) vd
u Fq = exp − Vt (x )dx (15)
0 u

  where x is the source to receptor distance and vd is the pollutant


1 u∗ 2 (xr − xs )2
szg = (9) deposition velocity (m s −1 ) (Section 3.6).
2 u |L|
To speed up computation the integral in Eq. (15) is computed
with a linear interpolation method. Fixing an elementary strip, Fq
where u* is the friction velocity and L is the Monin-Obukov length is computed at a set of fixed distances {xj }, then the Fq value for
(see Section 3.7). all the receptors exposed to the strip emissions are computed by
3.4.3.2. s z in the stable boundary layer (SBL, L > 0). In AERMOD s z in linear interpolation based on their distance from the strip. The {xj }
the SBL is obtained as the weighted sum of a near-surface component distances are among the environmental parameters of CAREA and
and an elevated component (Eqs. (79) and (84) in Cimorelli et al., can be set by the user. The defaults are {1, 50, 100, 250, 500, 750,
2004). In CAREA the elevated component is not considered and s z is 1000, 1250, 1500, 2000, 3000, 4000, 5000, 10,000}.
given by For each time step (i) of the simulation, the surface flux Fd,i
(mg m −2 s −1 ) is given by the equation
   − 1
2 u∗ (xr − xs ) (xr − xs ) 3
sz = 1 + 0.7 (10)
p u L Fd,i (xr , yr ) = vdi Ci (xr , yr ) (16)

3.4.3.3. s y in the CBL. The lateral coefficient s y in CBL is computed


following the AERMOD scheme (Eqs. (38), (40), (41), (77) in Cimorelli The total deposition flux, Fd (usually lg m −2 ), in model output is
et al., 2004) and considering ground emission and ground receptors. the sum over all the time steps:
Under these assumptions the simplified formulation is

sv (xr − xs ) Fd (xr , yr ) = Fd,i (xr , yr ) (17)
sy =  0.3 (11) i
s x −x
u 1 + 78 v (u rz s )
i

3.6. Deposition velocity vd


where s v is the root-mean-square lateral turbulence velocity:
The deposition velocity (vd , m s −1 ) is computed following the
scheme used by AERMOD (U.S. Environmental Protection Agency,
sv2 = svm
2 2
+ svc (12)
2004a; Wesely et al., 2002).
For dry deposition of gases vd is given by
s vm is the mechanical portion of s v computed near the surface
(z  0): vd = (ra + rb + rc )−1 (18)

2
where ra (s m −1 ) is the aerodynamic resistance, rb (s m −1 ) is the
svm = 3.6u2∗ (13) quasi-laminar bulk surface resistance, and rc (s m −1 ) is the bulk
surface resistance.
s vc is the convective portion of s v computed near the surface For dry deposition of particles vd is computed using “Method 2”
(z  0): in U.S. Environmental Protection Agency (2004a):

1 1
2
vd = fp + (1 − f p ) (19)
svc = 0.35w2∗ (14) ra + r p ra + rp + 0.002ra rp

where rp is the quasi-laminar sublayer resistance (s m −1 ) and fp is


w* is the convective velocity scale (Section 3.7). the fraction of particulate with aerodynamic diameter less than dp
3.4.3.4. s Y in the SBL. The lateral coefficient s y is computed using (set by the user).
the same scheme as the CBL and assuming s vc = 0 (Cimorelli et al., Considering the number of equations involved in this scheme, the
2004). readers are referred to the documents U.S. Environmental Protection
Agency (2004a) and Wesely et al. (2002) for details.

3.5. Concentration and deposition flux with dry deposition


3.7. Meteorology

The scheme used in CAREA to compute the concentration and dry


The CAREA formulation requires several meteorological quan-
deposition flux at the surface follows the source depletion model
tities describing the planetary boundary layer. The meteorological
used by AERMOD (Chamberlain, 1953; U.S. Environmental Protec-
data set is input to the model in the AERMET format (hourly data
tion Agency, 1995, 2004a), with the difference that the modification
records) (Cimorelli et al., 2005, 2004). The AERMET format was
of the vertical concentration profile due to deposition (Eqs. (1-54)
preferred because it is well known, widely used, and adopted by
182 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

AERMOD. However, CAREA can easily be modified to accept other in practice a limitation is imposed by the hardware and software
formats, like for example, the CALMET format (Scire et al., 2000). Not capacities. An example of this kind of receptor grid is shown in Fig. 2
all the meteorological parameters and variables provided by AERMET (bottom).
are required by CAREA, Table 1 lists the parameters and the variables
required by CAREA for each simulation mode. 3.9. Output

3.7.1. Calm wind conditions CAREA output consists of a shapefile of point objects (receptors)
The modelling of atmospheric dispersion in wind calm conditions ready for use in a GIS project. In this format, the concentrations and
represents a very difficult task for steady-state models like AERMOD the deposition fluxes at the surface can be easily uploaded in the GIS
and CAREA (Cirillo and Poli, 1992; Ghermandi et al., 2012). In these environment and mapped (continuous 2D distribution) using some
cases CAREA, AERMOD and ISC3 adopt the same EPA calm policy (Bai- common spatial interpolator. An example is shown in Fig. 5. This
ley, 2000): in wind calm conditions, i.e. for u < 1 m s −1 , the hour is figure reports a particular (three orchards fields) of the Epidemiology
not processed and ignored for the concentration statistics. Case Study considered in Section 4. Atmospheric dispersion generated
a main plume in the North-West direction and a less evident sec-
3.8. Receptors ondary plume in the South-West direction (the spatial discontinuity
of this plume is an artifact of the spatial interpolator). For each point
In CAREA the receptors can be defined according to three modal- the pollutant ground concentration and deposition flux are stored
ities: sparse, regular grid, progressive grid. as attributes. By contrast, AERMOD output requires post-processing
In the sparse mode, the user defines the locations of the recep- before being loaded into GIS projects. Nevertheless, AERMOD outputs
tors in the GIS environment, building a shapefile of point objects. An are more complete than those of CAREA: in addition to mean con-
example of sparse receptors is shown in Fig. 3. centrations and dry deposition, the user can obtain other statistics,
In the regular grid mode CAREA builds an equally spaced grid like maximums and percentiles, which are often used for regulatory
of receptors with step (Dg ) defined by the user. The grid covers a purposes. This shortcoming of CAREA will be the focus for future
rectangular region including all the sources with an overlap set by improvements to the model.
the user.
In the progressive grid mode (not available in AERMOD) an 3.10. Experiment setup
irregularly space grid is defined as follows:
The model is set up for a specific case by compiling the
1. The polygons (sources) are surrounded (buffered) by a regular “Experiment setup” text file (Fig. 1). This file contains all the vari-
grid with spacing Dg up to a distance of Dg1 ; ables and information characterizing a specific experiment. The
2. The area surrounding the polygons from a distance of Dg1 to default values for other general CAREA parameters (environmental
Dg2 is covered by a regular grid with spacing mg2 • Dg ; parameters) can be changed by editing the “Parameters” text file.
3. The area surrounding the polygons beyond Dg2 is covered by a
regular grid with spacing mg3 • Dg . 4. Comparing CAREA and AERMOD simulations

Dg , Dg1 , Dg2 , mg2 , and mg3 are set by the user; their default values The performances and the validation of AERMOD were widely
are 50 m, 200 m, 500 m, 2, and 4 respectively. addressed in literature (e.g. Rood, 2014; Jittra and Thepanondh,
This procedure is mainly based on the Python Shapely library. In 2015), and it is not a subject of this study. As stated above, CAREA
theory there is no maximum limit to the number of receptors, but represents an alternative to AERMOD, or other similar models, for

Table 1
AERMET meteorological data set and quantities required by CAREA. NODEP: simulation without dry deposition. PM: simulation with dry particle
deposition. GAS: simulation with gaseous dry deposition. N: parameter/variable not required. Y: parameter/variable required.

Parameter, variable Units NODEP PM GAS

Year – Y Y Y
Month – Y Y Y
Day – Y Y Y
Julian day – N N N
Hour – Y Y Y
Sensible heat flux W m −2 N N Y
Surface friction velocity m s −1 Y Y Y
Convective velocity scale m s −1 Y Y Y
Vertical potential temperature gradient above PBL K m −1 N N N
Height of the convectively-generated boundary layer m Y Y Y
Height of the mechanically-generated boundary layer m Y Y Y
Monin-Obukhow length m Y Y Y
Surface roughness length m N Y Y
Bowen ratio – N N Y
Surface albedo – N N Y
Wind speed m s −1 Y Y Y
Wind direction Deg Y Y Y
Height at which the wind was measured m Y Y Y
Air temperature K N N Y
Height at which the air temperature was measured m N N N
Precipitation code – N N N
Precipitation mm h −1 N N Y
Relative humidity – N N Y
Pressure hPa N N Y
Cloud cover – N N N
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 183

Fig. 5. Dispersion map of the pesticide emitted by agricultural fields obtained by the spatial interpolation (kriging) of the CAREA results at the receptors (regular grid) (this figure
is part of the results of the epidemiological experimentation considered in Section 4 and not yet published).

complex source areas sources. For this reason the CAREA ability of wind events were excluded (621 runs). In July there were
reproducing the same output of AERMOD was tested. 495 unstable hours and 249 stable hours, 114 wind calms
In order to compare CAREA and AERMOD simulations the follow- events were excluded (630 runs).
ing simple and reproducible experiment was set up (the data sets
and the setup of the experiment are available on request):
Considering the three different pollutants and the two periods,
- Domain of the simulation: The dispersion occurred in a six simulations, and therefore six comparisons, were conducted.
rural area (far away from buildings), flat, at sea level with In addition to visual comparison several numeric and graphic
dimensions of 5 km × 5 km. indicators widely used in literature for atmospheric dispersion
- Sources: Three agricultural fields were assumed as source model comparisons (e.g. Cirillo and Poli, 1992; Rood, 2014; Kadiyala
areas for pollutants (polygons in Fig. 6). The area and num- and Kumar, 2012; Kukkonen et al., 2001; Mohan and Siddiqui, 1997;
ber of polygon vertices were 6 ha , 5 ha, 4 ha and 8, 6, 7 Moreira et al., 2005; Nimmatoori and Kumar, 2013a,b; Anfossi and
respectively. Trini Castelli, 2014) were considered:
- Receptors: A regular grid of 10,000 receptors, with steps of
50 m, was defined around the sources.
- Scatter and box plots;
- Pollutants: Three different kinds of pollutants were consid-
- Normalized mean square error index:
ered: a generic pollutant for which deposition is negligible;
fine particulate matter (PM2.5) subject to dry deposition;
and benzene as representative of gaseous pollutants and (X 0 − X 1 )2
subject to dry deposition. NMSE = (20)
X0 ¯X1
- Emission rates: The emission rates of the three sources
were set to the arbitrary value of 100 mg s −1 m −2 (may
be unrealistic) for all the pollutants in order to facilitate where X indicates ground concentration or deposition flux
numerical and graphical (plots) comparisons. and 0 and 1 refer to AERMOD, taken as reference, and CAREA
- Meteorology: Considering that the experiment was not respectively. NMSE describes CAREA performance in terms
aimed at simulating a real case but it must be easy to of the scatter in an entire data set of receptors, the ideal
reproduce by other users, the AERMET meteorological value is 0;
file SALEM_86-90.SFC, available among the test cases in - Fractional Bias index
the AERMOD distribution (U.S. Environmental Protection
Agency, 2016) was used. X0 − X1
- Periods: the simulations were run for February 1990 and FB = 2 (21)
X0 + X1
July 1990. These months were chosen because they exhib-
ited the lowest and highest instability/stability events in
1990, respectively. In February there were 261 unstable gives the overall tendency of CAREA to overestimate or
(L < 0) hours and 411 stable (L > 0) hours, 51 calm underestimate AERMOD, the optimal value is 0;
184 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Fig. 6. Mean concentration (lg m −3 ) and dry deposition flux (lg m −2 ) contour maps obtained for the PM2.5 case in the studied periods. Contour lines were obtained after kriging
spatial interpolation (Isaaks and Srivastava, 1989) of the values at the receptors (grid) computed by the models in GIS environment (Johnston et al., 2001). Black lines: AERMOD.
Blue lines: CAREA. Upper plots: concentrations. Lower plots: deposition flux. Left plots: February. Right plots: July. Red polygons: source areas. (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

- Factor 2 index (FA2), corresponding to the number of is a measure of correlation between the two data sets of
receptors that satisfy the condition estimates, the ideal value is 1;
- Mean Geometric Bias
X0
0.5 ≤ ≤2 (22)
X1
 
- Index of agreement MGB = exp ln X0 − ln X1 (24)

 2
X0 − X 1
IA = 1 −    2 (23)
    is a more representative bias than FB for log-normal dis-
X0 − X1  + X1 − X1 
tributed samples, the ideal value is 1;
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 185

- Geometric variance - Deposition flux: CAREA overestimated the AERMOD upper


  extreme (95th percentile) values. The overestimation ranges
GV = exp (ln X0 − ln X1 )
2
(25) from +5% to +15%. Lower and medium values (5th, 25th, 50th
and 75th percentiles) of the CAREA and AERMOD simulations
were very close.
is a more representative variance estimator index than sim-
ple variance for log-normal distributed samples, the ideal The July period was characterized by a greater number of unsta-
value is 1; ble hours and wind calms than the February period. This could be
- Pearson’s linear correlation coefficient (R) between X0 and linked to the systematic ground concentration underestimation of
X1 (well known). CAREA in the July cases and could indicate that the simplifications of
the AERMOD formulation adopted in CAREA are more important
MGB and FB are measures of mean bias and indicate only system- (in the negative sense) for the CBL formulation and in low wind
atic errors which lead to always underestimate or overestimate the conditions.
measured values. NMSE and GV are measures of scatter and reflect The CAREA overestimation of the upper extreme values of the
both systematic and unsystematic (random). IA and R are measures deposition fluxes is more difficult to explain. This phenomenon was
of correlation between the data sets. observed both in the February and July cases, therefore it seems
Fig. 6 shows the mean concentration and dry deposition flux con- not linked to CBL or SBL formulations or to low wind conditions.
tour maps obtained for the PM2.5 case in the studied periods. Con- Moreover, if the overestimation were linked to deposition velocity
tour lines were obtained after kriging spatial interpolation (Isaaks formulation it would be noticeable also for low and medium values.
and Srivastava, 1989) of the values at the receptors (grid) computed These issues will be addressed in future improvements of the
by the models in the GIS environment. The western and eastern parts model.
of the computational domain were eliminated because the concen- CAREA is based on a simpler formulation than AERMOD, so lower
trations and deposition fluxes were negligible. The corresponding CAREA processing times are expected. In addition, CAREA simula-
contour maps for the No Deposition and Gas cases were very similar tions are automatically divided into sub-simulations (up to 4), which
to those reported and are thus not shown. In all cases the geometries are then executed in parallel using the multiprocessing package
(contour lines) of the spatial distribution of concentration and depo- (part of the basic Python distribution), further reducing the process-
sition flux obtained from the two models are in agreement. It can ing times. The model can assign to each sub-process a sub-set of
be observed that CAREA slightly overestimates AERMOD. This effect the simulation period (hours) or a sub-set of the receptors. In the
is more noticeable in the July scenario and is discussed in the next tests presented here the first modality was chosen. The model pro-
paragraphs. cessing times for July-No Deposition, July-Gas, 1990-No Deposition
Most of the areas in the computational domain exhibit very low (whole of 1990, 7529 h) and 1990-Gas were recorded. Two addi-
or null values (Fig. 6). The inclusion of the receptors located in these tional cases, quoted as Orchards and Epidemiological Study, were
areas in quantitative comparisons would lead to misinterpretations. also considered only for CAREA since they are to much complex for
For this reason, only the 2187 receptors located at a distance less AERMOD.
than 1000 m from the sources, hereafter named test receptors, are Orchards Case: 300 polygons, extracted from the orchards geo-
considered in the comparison. database over the Po Valley, Italy, were selected as sources; 17,024
Figs. 7 and 8 report the scatter plots of the CAREA and AERMOD receptors were distributed over a regular grid with steps of 75 m ×
evaluations of ground concentrations and deposition fluxes. All the 75 m; the meteorological data set was the same as that used for the
points, in all the cases studied, lie close to the identity line. This other tests.
agreement was moderately worse for the ground concentrations in Epidemiology Case: in the epidemiological study of Malagoli et al.
the July scenario, where CAREA showed systematic underestimations (2016) the exposure of 1475 people (targets) to the pesticide used on
of AERMOD. 1410 agricultural fields, distributed in the Po Valley (Italy) on an area
Figs. 9 and 10 report the box plots of the CAREA and AERMOD approximatively 100 km × 80 km wide, was analysed. CAREA was
evaluations. The lower and the upper whiskers indicate the 5th and used to compute the exposure using an at-the-target concentration
95th percentiles. The boxes indicate the 25th, the 50th, and the 75th approach. The results are still under investigation and therefore they
percentiles. Fig. 9 shows that the frequency distribution of the data were not presented in cited paper. The CAREA processing time for
sets from the two models do not differ widely in February cases. In this real case are analysed. The period of simulation was March–July
the July scenario CAREA ground concentrations are about 10% lower 2014 and deposition was not considered.
that AERMOD ground concentrations. Fig. 10 shows that CAREA In these tests CAREA was run both with multiprocessing (default)
deposition fluxes were slightly greater than AERMOD deposition and forced to single processing. A 32-bit version of Python was used
fluxes in all cases. and the models were run on a PC with an Intel(R)Core(TM)i7-4712-
The values of the indexes used for quantitative comparison are 2.3 GHz CPU and Windows 10 OS. The results, reported in Table 3,
reported in Table 2. NMSE, GV, IA, F2 and R are very close to the show that CAREA is considerably faster than AERMOD. In particular,
ideal values in all cases. FB and MGB show some significant deviation in the default multiprocessing mode the time processing ratio CAREA:
from the ideal for the ground concentrations computed in July. In AERMOD ranges approximately from 1:20 (July) to 1:40 (1990) for
these cases, the indexes indicate that CAREA systematically underes- the No-Deposition tests and from 1:70 (July) to 1:100 (1990) for the
timates the ground concentrations. The worst evaluation was −16%, Gas tests. The results also show that running CAREA in the multipro-
corresponding to MGB = 1.19, obtained in the Particulate Matter - cessing mode reduces processing times by 70% compared with single
July case. processing mode. The Orchards and the Epidemiology tests demon-
In summary, the graphic and quantitative comparisons of the case strated that CAREA can be used to handle large numbers of complex
studies demonstrate that sources and large numbers of receptors in a reasonable time of pro-
cessing. In the Orchards case the processing times (multiprocessing
- Ground concentration: the CAREA simulations were very close mode) were less than 2 h for the July scenario and less than one day
to the AERMOD simulations in the February cases. In the for the 1990 scenario. In the Epidemiology case the processing times
July cases CAREA systematically underestimated AERMOD. The (multiprocessing mode) was less than 12 h. In both cases, the benefit
underestimation ranges from −10% to −15%. produced by a multiprocessing code is evident.
186 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

5. Conclusions - averylargenumber(hundreds)ofsourceareascanbeconsidered;
- the source areas can be described using complex polygons with
CAREA is a new GIS-based Gaussian air dispersion model for a large number (hundreds) of vertices and holes;
complex source areas. The CAREA formulation of Gaussian dispersion, - sources and receptors are vector objects defined in the shapefile
CBL and SBL parameterization and dry deposition is based on the AER- format, which is widely used in the GIS environment;
MOD formulation. This model represents an alternative to AERMOD, - emission rates can be defined (input) as a GIS object attribute
and other similar models, in specific scenarios. of each source area;
CAREA is dedicated to predicting ground concentrations and sur- - output concentrations and dry deposition fluxes are provided
face dry deposition fluxes emitted by complex source areas, and its as an attribute of the receptor GIS objects;
characterizing features are as follows: - most of the model input/output can be managed in the GIS
environment. The only exception is the preparation of the
- it is considerably faster than AERMOD: this allows the simula- meteorological data set;
tion for long period of time and for a large number of sources - the model could easily be integrated as a tool into the open-
and receptors that would be unreasonable using AERMOD; source QGIS and ArcMAP® software and in many other GIS

Fig. 7. Comparisons between AERMOD and CAREA concentrations (C) at the 2187 test receptors and for the six cases studied.
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 187

Fig. 8. Comparisons between AERMOD and CAREA dry deposition flux (F) at 2187 the receptors inside the area of comparison for the four cases studied.

(not tested) used by the environmental science and technology not compute concentration statistics (e.g. max, min, percentiles); and
community. wet deposition is not considered.
- the model is coded in the Python language and is available on CAREA and AERMOD were applied to the same theoretical exper-
request; iments and the results compared. In the six cases studied, emissions
from source areas of a generic pollutant without dry deposition,
Compared with AERMOD, CAREA is less complete, in particular it particulate matter with dry deposition, and a gaseous pollutant (ben-
only provides ground values; is not applicable to complex terrains; zene) with dry deposition were considered. The simulations were
does not simulate emissions from stacks, volumes, or open pits; does conducted using a meteorological data set for two months, February

Fig. 9. Comparisons (box-whisker plots) between AERMOD and CAREA concentrations at the 2187 test receptors and for the six cases studied. The lower and the upper whiskers
show the 5th and 95th percentiles. The boxes show the 25th, the 50th, and the 75th percentiles.
188 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Fig. 10. Comparisons (box-whisker plots) between AERMOD and CAREA dry deposition fluxes at the 2187 test receptors and for the four cases studied. The lower and the upper
whiskers show the 5th and 95th percentiles. The boxes show the 25th, the 50th, and the 75th percentiles.

and July respectively, characterized by low and high occurrences of systematically underestimate by −10% to −15% the AERMOD
atmospheric instability respectively. Different comparison methods simulations. In the other studied cases, the ground concentra-
from among the most cited in the literature were applied. In summary, tions simulated by the two models showed low differences. This
these tests indicate that effect seems to indicate that the simplifications of the AER-
MOD formulation adopted in CAREA are more important (in
- The ground concentration simulated by CAREA in periods the negative sense) for the CBL formulation and in low wind
characterized by frequent low wind and unstable conditions conditions.

Table 2
Indexes of comparison obtained for the 2187 test receptors and for the considered test cases: N: no deposition, P: particulate matter with dry deposition, G: gas with dry
deposition, F: February, J: July. CA and CC are the ground-level concentrations (lg m −3 ) simulated by AERMOD and CAREA respectively. FA and FC are the deposition fluxes
(lg m −2 ) simulated by AERMOD and CAREA respectively. Overline and tilde indicate mean and median values respectively. FB: fractional bias. NMSE: normalized mean
square error. MGB: mean geometric bias. GV: geometric variance. IA: index of agreement. F2: factor 2 index. R Pearson’s linear correlation coefficient.

Ground concentration

Case CA CC C˜A C˜C FB NMSE MGB GV IA F2 R

NF 580 557 323 300 0.04 0.01 1.07 1.02 1.00 0.99 0.999
NJ 676 592 372 315 0.13 0.03 1.18 1.03 1.00 1.00 0.999
PF 537 515 290 263 0.04 0.01 1.10 1.02 1.00 0.99 0.998
PJ 621 547 331 277 0.13 0.03 1.19 1.03 1.00 1.00 0.998
GF 580 557 323 300 0.04 0.01 1.07 1.02 1.00 0.99 0.999
GJ 675 592 371 314 0.13 0.03 1.18 1.03 1.00 1.00 0.999

Deposition flux

Case FA FC F˜A F˜C FB NMSE MGB GV IA F2 R

PF 1714 1830 889 887 −0.07 0.08 1.00 1.01 0.99 1.00 0.997
PJ 2422 2526 1144 1081 −0.04 0.09 1.04 1.01 0.99 1.00 0.995
GF 3 4 2 2 −0.05 0.04 0.98 1.01 1.00 0.99 0.998
GJ 21 23 10 11 −0.11 0.09 0.96 1.00 0.99 1.00 0.998

Table 3
Processing times for different test cases: July = July period; 1990 = whole of 1990; ND = no deposition, G = gas with dry deposition; O = orchards case; E = epidemiology case.

Case Num. of sources Num. of receptors Num. of runs Proc. time Processing time

AERMOD CAREA

Single Proc. Multi Proc.

July-ND 3 10,000 630 4 min 14 s 23 s 12 s


July-G 3 10,000 630 19 min 7 s 34 s 15 s
1990-ND 3 10,000 7529 50 min 11 s 4 min 2 s 1 min 18 s
1990-G 3 10,000 7529 3 h 45 min 52 s 6 min 29 s 2 min 1 s
O-July-ND 300 17,024 630 – 4 h 48 min 40 s 1 h 26 min 18 s
O-July-G 300 17,024 630 – 5 h 8 min 5 s 1 h 32 min 59 s
O-1990-ND 300 17,024 7529 – 57 h 11 min 20 s 17 h 15 min 14 s
O-1990-G 300 17,024 7529 – 61 h 41 min 45 s 18 h 30 min 32 s
E-ND 1410 1475 4713 – 39 h 40 min 4 s 11 h 7 min 41 s
S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190 189

- The deposition fluxes simulated by CAREA overestimated the Cimorelli, A.J., Perry, S.G., Venkatram, A., Weil, J.C., Paine, R.J., Peters, W.D., Brode, R.W.,
Paumier, J.O., 2004. AERMOD: description of model formulation. Technical Report
AERMOD upper extreme (95th percentile) values. The overesti-
EPA-454/R-03-004. U.S. Environmental Protection Agency.
mation ranges from +5% to +15%. Lower and medium values Cirillo, M.C., Poli, A.P., 1992. An intercomparison of semiempirical diffusion models
(5th, 25th, 50th and 75th percentiles) of the CAREA and AERMOD under low wind speed, stable conditions. Atmos. Environ. Part A Gen. Top. 26,
765–774.
simulations were very close.
Clarke, A., Robertson, L., Hamilton, R., Gorbunov, B., 2004. A Lagrangian model of the
evolution of the particulate size distribution of vehicular emissions. Sci. Total
These issues will be addressed in future improvements of the Environ. 334–335, 197–206.
Collett, R.S., Oduyemi, K., 1997. Air quality modelling: a technical review of
model.
mathematical approaches. Meteorol. Appl. 4, 235–246.
The CAREA and AERMOD processing times were also compared. Danuso, F., Rocca, A., Ceccon, P., Ginaldi, F., 2015. A software application for mapping
CAREA is considerably faster than AERMOD. Using a PC with an livestock waste odour dispersion. Environ. Model. Softw. 69, 175–186.
ESRI, 1988. ESRI Shapefile Technical Description. An ESRI White Paper, July 1998. last
Intel(R)Core(TM)i7-4712-2.3 GHz CPU the CAREA: AERMOD time
visited 12 Apr. 2017.
processing ratio ranges from 1:20 to 1:40 when dry deposition is Fox, T., 2005. Revision to the guideline on air quality models: adoption of a preferred
not considered and from 1:70 to 1:100 when dry deposition is general purpose (flat and complex terrain) dispersion model and other revi-
considered. sions. Technical Report. Final rule. Tech. Rep. 40 CFR Part 51. U.S. Environmental
Protection Agency.,
CAREA was also tested on two cases characterized by a thousands Ghenai, C., Lin, C., 2006. Dispersion modeling of PM10 released during decontamina-
of complex sources and receptors and thousands of time steps tion activities. J. Hazard. Mater. 132, 58–67.
(hours). These tests demonstrated that CAREA can be used for the Ghermandi, G., Fabbi, S., Zaccanti, M., Bigi, A., Teggi, S., 2014. Urban micro-scale
investigation of NOx and CO emissions from vehicular traffic and comparison on
simulation of such complex scenarios. with air quality data. . pp. 380–385.
CAREA is coded in the Python language, which facilitates further Ghermandi, G., Teggi, S., Fabbi, S., Bigi, A., Cecchi, R., 2012. Model comparison in sim-
improvements and integrations of the code by users. Some aspects ulating the atmospheric dispersion of a pollutant plume in low wind conditions.
Int. J. Environ. Pollut. 48, 69–77.
that could be addressed to enhance the model include the possibil- Ghermandi, G., Teggi, S., Fabbi, S., Bigi, A., Zaccanti, M., 2015. Tri-generation power
ity of considering complex terrain, wet deposition, integration into plant and conventional boilers: pollutant flow rate and atmospheric impact of
GIS software using GUIs, the computation of statistical parameters stack emissions. Int. J. Environ. Sci. Technol. 12, 693–704.
Ghermandi, G., Teggi, S., Fabbi, S., Bigi, A., Zaccanti, M.M., 2013. From local-scale
(e.g. max values, percentiles) as output, and the validation with
to micro-scale assessment of the atmospheric impact of the pollutant plume
ground concentration measurements. emitted from a power-plant stack. Proceedings of 15th International Confer-
ence on Harmonisation within Atmospheric Dispersion Modelling for Regulatory
Purposes. pp. 6–9.
Gil, Y., Sinfort, C., 2005. Emission of pesticides to the air during sprayer application: a
References bibliographic review. Atmos. Environ. 39, 5183–5193.
Gillies, S., Bierbaum, A., Lautaportti, K., Tonnhofer, O., 2015. Shapely 1.5.15, geo-
metric objects, predicates, and operations. https://pypi.python.org/pypi/Shapely.
Anfossi, D., Trini Castelli, S., 2014. Atmospheric tracer experiment uncertainties
Accessed: 2016/04/01.
related to model evaluation. Environ. Model. Softw. 51, 166–172.
Giua, R., Morabito, A., Tanzarella, A., Spagnolo, S., Pastore, T., Bevere, M., Valentini,
Bailey, D.T., 2000. Meteorological Monitoring Guidance for Regulatory Modeling
E., La Ghezza, V., De Gennaro, G., Brusasca, G., Tinarelli, G., Assennato, G., 2014.
Applications. EPA-454/R-99-005. U.S. Environmental Protection Agency.
Application of a Lagrangian particle model to the source apportionment for
Batiha, M., Al-Makhadmeh, L., Batiha, M., Ramadan, A., Kadhum, A., 2014. General-
primary macropollutants in Taranto area (South Italy). Int. J. Environ. Pollut. 55,
ization of the MAFRAM methodology for semi-volatile organic agro-chemicals.
31–40. http://dx.doi.org/10.1504/IJEP.2014.065902.
Water Air Soil Pollut. 225,
Gulliver, J., Briggs, D., 2011. Stems-air: a simple GIS-based air pollution
Batiha, M.A., Kadhum, A.A.H., Batiha, M.M., Takriff, M.S., Mohamad, A.B., 2010.
dispersion model for city-wide exposure assessment. Sci. Total Environ. 409,
MAFRAM-a new fate and risk assessment methodology for non-volatile organic
2419–2429.
chemicals. J. Hazard. Mater. 181, 1080–1087.
Heckel, P., Lemasters, G., 2011. The use of AERMOD air pollution dispersion models to
Bauer, T.J., 2013. Comparison of chlorine and ammonia concentration field trial data
estimate residential ambient concentrations of elemental mercury. Water Air Soil
with calculated results from a Gaussian atmospheric transport and dispersion
Pollut. 219, 377–388.
model. J. Hazard. Mater. 254–255, 325–335.
Holmes, N.S., Morawska, L., 2006. A review of dispersion modelling and its applica-
van den Berg, F., Jacobs, C.M.J., Butler Ellis, M.C., Spanoghe, P., Doan Ngoc, K.,
tion to the dispersion of particles: an overview of different dispersion models
Fragkoulis, G., 2016. Modelling exposure of workers, residents and bystanders to
available. Atmos. Environ. 40, 5902–5928.
vapour of plant protection products after application to crops. Sci. Total Environ.
Huertas, J., Huertas, M., Cervantes, G., Diaz, J., 2014. Assessment of the natural sources
573, 1010–1020.
of particulate matter on the opencast mines air quality. Sci. Total Environ. 493,
Beyea, J., 1999. Geographic exposure modeling: a valuable extension of geographic
1047–1055.
information systems for use in environmental epidemiology. Environ. Health
Perspect. 107 Suppl, 181–190. Isaaks, E.H., Srivastava, R.M., 1989. An Introduction to Applied Geostatistics. Oxford
Bilanin, A.J., Teske, M.E., Barry, J.W., Ekblad, R.B., 1989. AGDISP: the aircraft spray dis- University Press.
persion model, code development and experimental validation. Trans. Am. Soc. Jittra, N.b., Thepanondh, S.b., 2015. Performance evaluation of AERMOD air dispersion
Agric. Eng. 32, 327–334. model in Maptaphut industrial area, Thailand. , pp. 225–228.
Bisignanesi, V., Borgas, M., 2007. Models for integrated pest management with chem- Johnston, K., Ver Hoef, J.M., Krivoruchko, K., Lucas, N., 2001. Using ArcGIS geostatistical
icals in atmospheric surface layers. Ecol. Model. 201, 2–10. analyst. 380. Esri Redlands.
Bowler, R., Beseler, C., Gocheva, V., Colledge, M., Kornblith, E., Julian, J., Kim, Y., Bollweg, Kadiyala, A., Kumar, A., 2012. Evaluation of indoor air quality models with the ranked
G., Lobdell, D., 2016. Environmental exposure to manganese in air: associations statistical performance measures using available software. Environ. Prog. Sustain.
with tremor and motor function. Sci. Total Environ. 541, 646–654. Energy 31, 170–175.
Brzozowska, L., 2014. Modelling the propagation of smoke from a tanker fire in a Kukkonen, J., Harkonen, J., Walden, J., Karppinen, A., Lusa, K., 2001. Validation of
built-up area. Sci. Total Environ. 472, 901–911. the dispersion model CAR-FMI against measurements near a major road. Atmos.
Candela, S., Bonvicini, L., Ranzi, A., Baldacchini, F., Broccoli, S., Cordioli, M., Carretta, E., Environ. 35, 949–960.
Luberto, F., Angelini, P., Evangelista, A., Marzaroli, P., Giorgi Rossi, P., Forastiere, Llanos, W., Higueras, P., Oyarzun, R., Esbrí, J., López-Berdonces, M., García-Noguero,
F., 2015. Exposure to emissions from municipal solid waste incinerators and mis- E., Martínez-Coronado, A., 2010. The MERSADE (European Union) project: testing
carriages: a multisite study of the MONITER project. Environ. Int. 78, 51–60. procedures and environmental impact for the safe storage of liquid mercury in
Candela, S., Ranzi, A., Bonvicini, L., Baldacchini, F., Marzaroli, P., Evangelista, A., Luberto, the Almadén district, Spain. Sci. Total Environ. 408, 4901–4905.
F., Carretta, E., Angelini, P., Sterrantino, A., Broccoli, S., Cordioli, M., Ancona, C., Llanos, W., Kocman, D., Higueras, P., Horvat, M., 2011. Mercury emission and
Forastiere, F., 2013. Air pollution from incinerators and reproductive outcomes: a dispersion models from soils contaminated by cinnabar mining and metallurgy.
multisite study. Epidemiology 24, 863–870. J. Environ. Monit. 13, 3460–3468.
Canepa, E., D’Alberti, F., D’Amati, F., Triacchini, G., 2007. The GIS-based SafeAirView Maantay, J., Tu, J., Maroko, A., 2009. Loose-coupling an air dispersion model and a
software for the concentration assessment of radioactive pollutants after an geographic information system (GIS) for studying air pollution and asthma in the
accidental release. Sci. Total Environ. 373, 32–42. Bronx, New York City. Int. J. Environ. Health Res. 19, 59–79.
Chamberlain, A., 1953. Aspects of travel and deposition of aerosol and vapour Malagoli, C., Costanzini, S., Heck, J., Malavolti, M., De Girolamo, G., Oleari, P., Palazzi,
clouds. Technical Report. Atomic Energy Research Establishment, Harwell, Berks, G., Teggi, S., Vinceti, M., 2016. Passive exposure to agricultural pesticides and risk
England. of childhood leukemia in an Italian community. Int. J. Hyg. Environ. Health 219,
Cimorelli, A., Perry, S., Venkatram, A., Weil, J., Paine, R., Wilson, R., Lee, R., Peters, W., 742–748.
Brode, R., 2005. AERMOD: a dispersion model for industrial source applications. Martinez, A., Spak, S., Petrich, N., Hu, D., Carmichael, G., Hornbuckle, K., 2015.
part i: General model formulation and boundary layer characterization. J. Appl. Atmospheric dispersion of PCB from a contaminated Lake Michigan harbor.
Meteorol. 44, 682–693. Atmos. Environ. 122, 791–798.
190 S. Teggi et al. / Science of the Total Environment 610-611 (2017) 175–190

Mazur, M., Mintz, R., Lapalme, M., Wiens, B., 2009. Ambient air total gaseous mercury Scire, J.S., Robe, F.R., Fernau, M.E., Yamartino, R.J., 2000. A user’s guide for the CALMET
concentrations in the vicinity of coal-fired power plants in Alberta, Canada. Sci. meteorological model. Earth Tech, Inc., Concord, MA.
Total Environ. 408, 373–381. Smith, R., 1995. A Gaussian model for estimating odour emissions from area sources.
Mohan, M., Siddiqui, T.a., 1997. An evaluation of dispersion coefficients for use in air. Math. Comput. Model. 21, 23–29.
Bound.-Layer Meteorol. 177–206. Tadano, Y., Borillo, G., Godoi, A., Cichon, A., Silva, T., Valebona, F., Errera, M., Penteado
Moreira, D.M., Tirabassi, T., Carvalho, J.C., 2005. Plume dispersion simulation in low Neto, R., Rempel, D., Martin, L., Yamamoto, C., Godoi, R., 2014. Gaseous emissions
wind conditions in stable and convective boundary layers. Atmos. Environ. 39, from a heavy-duty engine equipped with SCR aftertreatment system and fuelled
3643–3650. with diesel and biodiesel: assessment of pollutant dispersion and health risk. Sci.
Morra, P., Lisi, R., Spadoni, G., Maschio, G., 2009. The assessment of human health Total Environ. 500–501, 64–71.
impact caused by industrial and civil activities in the pace valley of Messina. Tartakovsky, D., Stern, E., Broday, D., 2016a. Dispersion of TSP and PM10 emissions
Sci. Total Environ. 407, 3712–3720. http://dx.doi.org/10.1016/j.scitotenv.2009. from quarries in complex terrain. Sci. Total Environ. 542, 946–954.
03.005. Tartakovsky, D., Stern, E., Broday, D., 2016b. Indirect estimation of emission factors for
Nimmatoori, P., Kumar, A., 2013a. Development and evaluation of a ground-level area phosphate surface mining using air dispersion modeling. Sci. Total Environ. 556,
source analytical dispersion model to predict particulate matter concentration for 179–188.
different particle sizes. J. Aerosol Sci. 66, 139–149. Teske, M.E., Bird, S.L., Esterly, D.M., Curbishley, T.B., Ray, S.L., Perry, S.G., 2002.
Nimmatoori, P., Kumar, A., 2013b. Evaluation of area source models to predict near AgDRIFT: a model for estimating near-field spray drift from aerial applications..
ground level concentrations due to emissions released during agricultural appli- Environ. Toxicol. Chem./SETAC 21, 659–671.
cations. J. Hazard. Mater. 246–247, 44–51. Theobald, M.R., Løfstrøm, P., Walker, J., Andersen, H.V., Pedersen, P., Vallejo, A., Sutton,
Owen, B., Edmunds, H., Carruthers, D., Raper, D., 1999. Use of a new generation urban M.A., 2012. An intercomparison of models used to simulate the short-range atmo-
scale dispersion model to estimate the concentration of oxides of nitrogen and spheric dispersion of agricultural ammonia emissions. Environ. Model. Softw. 37,
sulphur dioxide in a large urban area. Sci. Total Environ. 235, 277–291. 90–102.
Pecorari, E., Mantovani, A., Franceschini, C., Bassano, D., Palmeri, L., Rampazzo, G., Till, J., Rood, A., Garzon, C., Lagdon, R., Jr., 2014. Comparison of the MACCS2 atmo-
2016. Analysis of the effects of meteorology on aircraft exhaust dispersion and spheric transport model with Lagrangian puff models as applied to deterministic
deposition using a Lagrangian particle model. Sci. Total Environ. 541, 839–856. and probabilistic safety analysis. Health Phys. 107, 213–230. http://dx.doi.org/
Penn, S., Arunachalam, S., Tripodis, Y., Heiger-Bernays, W., Levy, J., 2015. A comparison 10.1097/HP.0000000000000102.
between monitoring and dispersion modeling approaches to assess the impact Environmental Protection Agency, U.S., 1995. User’s Guide for the Industrial Source
of aviation on concentrations of black carbon and nitrogen oxides at Los Angeles Complex (ISC3) Dispersion Model (revised). Volume II - Description of Model
international airport. Sci. Total Environ. 527–528, 47–55. Algorithms. EPA-454/b-95-0036.
Pivato, A., Barausse, A., Zecchinato, F., Palmeri, L., Raga, R., Lavagnolo, M., Cossu, R., Environmental Protection Agency, U.S., 2004a. AERMOD deposition algorithms – sci-
2015. An integrated model-based approach to the risk assessment of pesticide ence document (revised draft). Technical Report. U.S. Environmental Protection
drift from vineyards. Atmos. Environ. 111, 136–150. Agency.,
Pospisil, J., Katolicky, J., Jicha, M., 2004. A comparison of measurements and cfd Environmental Protection Agency, U.S., 2004b. User’s Guide for the AMS/EPA Regula-
model predictions for pollutant dispersion in cities. Sci. Total Environ. 334–335, tory Model – AERMOD. Technical Report EPA-454/B-03-001. U.S. Environmental
185–195. Protection Agency.,
Robins, N., Hagan, N., Halabi, S., Hsu-Kim, H., Gonzales, R., Morris, M., Woodall, G., Environmental Protection Agency, U.S., 2016. Technology Transfer Network Support
Richter, D., Heine, P., Zhang, T., Bacon, A., Vandenberg, J., 2012. Estimations of Center for Regulatory Atmospheric Modeling - Preferred/Recommended models.
historical atmospheric mercury concentrations from mercury refining and pre- last visited 27 Apr. 2017.
sent-day soil concentrations of total mercury in Huancavelica, Peru. Sci. Total Wesely, M., Doskey, P., Shannon, J., 2002. Deposition parameterizations for the indus-
Environ. 426, 146–154. trial source complex (ISC3) model. Draft ANL report ANL. Technical Report.
Rood, A., 2014. Performance evaluation of AERMOD, CALPUFF, and legacy air ER/TRB01/003. Argonne National Laboratory, Argonne, Illinois 60439.
dispersion models using the winter validation tracer study dataset. Atmos. Zhang, Q., Wei, Y., Tian, W., Yang, K., 2008. GIS-based emission inventories of urban
Environ. 89, 707–720. scale: a case study of Hangzhou, China. Atmos. Environ. 42, 5150–5165.

You might also like