вход по аккаунту



код для вставкиСкачать
Atmos. Sci. Let. 7: 15–20 (2006)
Published online in Wiley InterScience ( DOI: 10.1002/asl.123
Modelling and predicting urban atmospheric pollutants in
the Aosta Valley region of Italy using a site-optimised
Kim N. Dirks,1 * Alessandro Nanni2 and Vincent I. Dirks3
1 The University of Auckland, Private Bag 92019,
2 ARIANET srl, via Gilino 9–20128 Milano, Italy
3 D2 Solutions, Auckland, New Zealand
*Correspondence to:
Kim N. Dirks, Department of
Physiology, School of Medical
Sciences, The University of
Auckland, Private Bag 92019,
Auckland. New Zealand.
Auckland, New Zealand
An effective simple site-optimised urban air pollution model for predicting CO, PM10 , NO
and NO2 concentrations is developed for the topographically complex region of the Aosta
Valley in Italy. The good results suggest that such a model could be used to ‘downscale’
mesoscale-forecasted surface conditions to give real-time site-specific pollution predictions.
Copyright  2006 Royal Meteorological Society
urban; atmospheric pollutants; modelling
Received: 4 May 2005
Revised: 25 January 2006
Accepted: 26 January 2006
1. Introduction
The Regional Environmental Authority responsible for
monitoring air pollution in the Aosta Valley region
of Italy (Figure 1) implements a complete 3D air
pollution dispersion model for the region. So far,
there has been good success in predicting for most
of the pollutants, despite the complex topography of
the region (Finardi et al., 2002; Nanni et al., 2004).
However, more site-specific empirical modelling is
needed to help determine the reasons for some of
the unresolved discrepancies between observed and
predicted concentrations.
Presented here are the results of a site-optimised
semi-empirical model used to identify some of the
factors influencing vehicle-induced pollutant concentrations on a local scale. The model uses a year of
measured air pollution and surface wind data to determine the optimum parameters for the specific site,
assuming an inverse relationship with wind speed and
sorting the data by season, day of the week, time of day
and wind direction. The model was originally designed
and proved successful for a relatively flat suburban
site in Hamilton, New Zealand, and only for carbon
monoxide (CO) (Dirks et al., 2002). In the current article, the model is tested on a site in the topographically
complex environment of the Aosta Valley and applied
to particulate matter (PM10 ), and to oxides of nitrogen (NO2 and NO), in addition to CO, all of them
pollutants for which vehicles are a significant source
in urban environments. The consistency of the model
parameters from one year to the next are also verified
to ensure that model parameters optimised for one year
Copyright  2006 Royal Meteorological Society
can be used to predict pollutant concentrations the following year, as long as the emission patterns have not
changed significantly during this time.
The success of the semi-empirical model suggests
that it can be used to identify general and pollutantspecific considerations for the region to help improve
the success of the 3D pollution modelling in the Aosta
Valley environment. It also suggests that ultimately the
semi-empirical model could be used with downscaled
surface mesoscale-forecasted winds to provide realtime urban air pollution forecasts at specific sites
where air pollution measurements have been made.
2. Motivation for the study and the
semi-empirical approach
The local topographical structures found in urban
and suburban areas result in complex wind flows
that can be difficult to model using an entirely
physically based approach. Indeed, even many of the
quite detailed models for predicting concentrations at
monitoring sites from nearby sources such as roads
have been found to be unsuitable for complex urban
and suburban environments (Hoydysh and Dabberdt,
1994). Intersections in particular are difficult to model
because of the large pollutant concentration gradients
that are often present, as demonstrated by Claggett
et al., 1981.
One alternative to theoretical modelling used by
other authors in the field (Ketzel et al., 1999; Kukkonen et al., 2001; Kassomenos et al., 2004), and the
approach that is taken in this article, is to use existing air pollution data from a site of interest to
K. N. Dirks, A. Nanni and V. I. Dirks
Figure 1. Location of the study area within the European Alps. The Aosta Valley is located in the northwest corner of Italy near
the border between France and Switzerland. The dot represents the location of the Piazza Plouves site
develop a site-specific semi-empirical model for making predictions and understanding the local behaviour
of the particular site. The model presented has deliberately been kept as simple as possible and no pollutionspecific factors influencing concentrations (such as
rainfall or solar radiation) have been included.
the leeward and windward cases were separated, there
was no significant effect of wind direction on concentration, except for a slight influence in high wind
speed conditions when concentrations tend to be low
anyway. For this reason, to avoid the problems of discontinuity and for the sake of simplicity, the (sin θ )−1
term is omitted. Concentrations for leeward cases are
therefore given by
3. The semi-empirical model
C =
The semi-empirical model is based on the box model
approach where the emission rate Q (mg m−1 s−1 ) is
assumed to be constant along a road and the pollutants
mix uniformly within a two-dimensional box of height
z (Hanna et al., 1982). The horizontal wind speed,
assumed to be uniform within the layer and running
at an angle of θ to the road, removes the pollutants
through advection. At the same time, pollutants are
introduced into the box through advection of the background concentration. In the present model, a wind
speed offset (uo ) is included in order to avoid severe
over-predictions in very light wind speed conditions
(as originally suggested by Zimmerman and Thompson (1975) in the HIWAY model) and to take into
account the vehicle-induced turbulence that is likely to
dominate the dispersion process in such low ambient
wind speed conditions (Ketzel et al., 1999), especially
very close to the road and when vehicles are travelling
at high speeds (Gronskei, 1988). For the Hamilton site
tested in Dirks et al. (2002), it was found that provided
Copyright  2006 Royal Meteorological Society
+ Cl
z (u + uo )
where Cl is the background concentration for leeward
wind conditions. For the cases where the wind direction is undefined (because of a large variability in the
wind directions throughout the averaging period), the
concentration is given by
C =
+ Cu
z (u + uo )
The emissions term incorporates both emissions
from the road adjacent to the monitor and emissions
from other roads in the vicinity that have wind speed
dependencies. The terms Cl and Cu represent the
background concentrations for leeward conditions and
for conditions where the wind direction is undefined,
The optimum parameters (Ql /z , Cl , Qu /z and
Cu ) of the model were found by performing linear
regressions of carbon monoxide concentration (C )
Atmos. Sci. Let. 7: 15–20 (2006)
Site-optimised modelling of urban atmospheric pollutant concentrations
on the wind speed function 1/(u + uo ) for the leeward and undefined wind intervals separately for each
hourly average throughout the day, giving 24 regression coefficients (given by Ql /z for the leeward and
Qu /z for the undefined wind intervals separately)
and 24 intercepts (Cl for the leeward and Cu for the
undefined wind intervals separately). This procedure
was then repeated for each of the five periods for
each year (as defined in Table I), separately for weekdays and weekends. The wind speed offset (uo ) was
optimised through the minimisation of the root-meansquared error (RMSE) for each season and year. The
optimum parameters were then used to make predictions for the same period, provided wind speed and
direction were available for the particular time interval.
This procedure was repeated for the four pollutants of
interest. One of the important advantages of this model
is that no information is required about the background
concentrations of any of the pollutants, as they are
estimated empirically.
4. Study location and data
The Aosta Valley is located in the northwest of Italy
near the border with France and Switzerland, with the
main valley running approximately from east to west.
The air quality network in the region is relatively dense
with 11 monitoring stations located within a region of
about 30 km by 80 km, and mostly near the bottom of
the valley. In this study, we focus our attention on one
site called Piazza Plouves which is located within an
urban area near the base of the valley. The winds are
predominantly northerly, believed to be due to wind
channelling as a result of a smaller north–south branch
of the valley system (the Gran San Bernardo Valley)
and due to the building of structures in the immediate
vicinity of the sampling site.
The subset of the data used in this particular
study consists of two years (2000–2001), for which
hourly averaged CO, NO, NO2 and PM10 data have
been collected, as well as comprehensive hourly data
average meteorological data including wind speed
and direction, temperature, humidity, precipitation,
atmospheric pressure and solar radiation. For 2001,
the mean and maximum wind speeds at the site
were 1.2 and 6.6 ms−1 , respectively, with a calm
wind frequency (<0.5 ms−1 ) of 35.8%. The mean,
minimum and maximum temperatures were 11.9, −7.4
and 35.6 ◦ C, respectively, and the mean and maximum
radiation were 123.2 and 961 Wm−2 , respectively.
To take into account seasonal differences in pollution levels, the data have been sorted into five periods
per year consisting of approximately 10 weeks each;
two periods fall within Standard Time and three fall
within Daylight Savings Time. Data sets of about
10 weeks have been found to be ideal for optimising the model as they result in a sufficient amount of
data to produce reliable optimum model parameters,
while not spanning a sufficiently long period of time
Copyright  2006 Royal Meteorological Society
Table I. Start and end dates for the five periods
Late Winter
Early Winter
Start date
End date
Beginning of Jan
End of March
Beginning of June
Middle of Aug
End of Oct
End of March
Beginning of June
Middle of Aug
End of Oct
Beginning of Jan
for seasonal differences to be too significant within one
period (Dirks, 2001). Note that the number of weekend days in any period will always be less than the
number of weekdays, so the lower limit is set by the
number of days required to produce steady optimum
parameters for the weekend days. This has been found
to be about 20 days, corresponding to 10 weeks. The
start and end times for the five periods are presented
in Table I and labelled as early winter, late winter,
spring, summer and autumn.
Figure 2 shows the average carbon monoxide concentrations throughout the day for weekends and
weekdays for the late winter and spring seasons,
respectively, for the two years of observation. From
this figure and similar figures created for the other
periods and the other pollutants, it can be seen that
the concentrations are highest during the early and late
winter and lowest during the spring and summer. This
is consistent with the lower daily average wind speeds,
Figure 2. Average CO concentration throughout the day for
(a) late winter, and (b) spring
Atmos. Sci. Let. 7: 15–20 (2006)
K. N. Dirks, A. Nanni and V. I. Dirks
Figure 3. One week of observed and predicted concentrations for the late winter period (a) CO, (b) PM10 , (c) NO and (d) NO2
Figure 4. One week of observed and predicted concentrations for the summer period (a) CO, (b) PM10 , (c) NO and (d) NO2
Copyright  2006 Royal Meteorological Society
Atmos. Sci. Let. 7: 15–20 (2006)
Site-optimised modelling of urban atmospheric pollutant concentrations
the higher levels of emissions from home heating and
the more frequent periods of temperature inversions
conducive to high pollution episodes observed during
the winter (Finardi et al., 2001). Also observed are the
characteristic morning and evening peaks in concentrations, especially during weekdays, coinciding with
the higher traffic volumes observed during these times.
5. Basic Model Results
Figures 3 and 4 show examples of 1 week of observed
and predicted concentrations during the late winter
and summer, respectively. Table II shows model evaluation results for the ten periods of data given as
the seasonal average of the daily average statistics
within a season and the standard error of the mean
between seasonal averages. Note that the model tends
to perform well for periods of low concentration, while
showing a slight tendency to under-predict during periods of high concentration. The mean absolute error
(MAE) and RMSE are at their lowest during the summer, consistent with the lower concentrations at this
time of year. In general, there is good correspondence
between observed and predicted pollutant concentrations despite the complex topography of the region.
Given that similarly good results were obtained for the
model for flat terrain in Hamilton (Dirks et al., 2002)
and in this case in alpine terrain, it is expected that
the model could be used for any sampling site located
adjacent to a busy road for pollutants where vehicles
are the dominant source, regardless of the nature of
the topography.
The model is tested on independent data by constructing the model parameters on the basis of one
year of data and using these to make predictions for
the other year (crossed) rather than using the same data
set to make the predictions (self). Figure 5 shows an
example of 1 week of observed CO concentrations as
well as two predicted series (self and crossed). Note
that the predictions are relatively consistent. Similar
results were obtained for the other pollutants of interest. This suggests that the model could be used to make
predictions for the long term, provided on-going meteorological measurements are available and there are
no significant changes in the traffic flow patterns over
time. This means that the model is suitable for investigating the impact of legislative changes on emissions
on pollutant concentrations as it is possible to predict
what concentrations would have been expected for a
Table II. Model statistics (averaged over the ten seasons,
weekdays, and weekends, combined with standard errors of the
means) for the four pollutants for the two years of observation
CO (mg m−3 )
PM10 (µg m−3 )
NO (µg m−3 )
NO2 (µg m−3 )
Observed Predicted
1.2 ± 0.2
37 ± 3
37 ± 9
47 ± 5
1.2 ± 0.2
37 ± 3
37 ± 9
47 ± 5
given year in the absence of such changes (by using
model parameters optimised for a year prior to the
change) and comparing with the concentrations actually observed.
By coupling the site-specific model with a mesoscale
predictive model such as MM5 (Grell et al., 1994), it
should be possible to use the semi-empirical model
as an air pollution forecasting tool for specific sites
for which the optimum model parameters have been
obtained. This avoids the necessity of carrying out the
complex task of downscaling from the mesoscale to
the local scale in urban environments.
6. Conclusions
A semi-empirical model is evaluated at an urban site in
a sub-alpine region of Italy and found to be useful for
making reliable predictions of pollutant concentrations
at specific sites. Given the success of the model for a
flat urban site in Hamilton, New Zealand, it appears
that the model could be used for urban/suburban sites
of any topography and for any pollutant, provided the
site is located near a busy road and that the source of
the pollutant is predominantly vehicular traffic.
The predictions of the model are consistent from
one year to the next, suggesting that the model is
suitable for interpolating over periods of missing
data (hence providing improved long-term statistics
on air pollution datasets) and for investigating the
impact of significant legislative traffic circulation or
fleet-induced changes in emissions by comparing the
predictions that would be made with and without the
changes for given meteorological conditions. It is also
expected that the model could be used to ‘downscale’
mesoscale-forecasted surface conditions to give realtime pollution predictions at specific sites.
0.46 ± 0.07 0.32 ± 0.05
19.5 ± 1
14 ± 1
26 ± 6
16 ± 4
19 ± 3
14 ± 2
Copyright  2006 Royal Meteorological Society
Figure 5. Model predictions for 1 week of data. Self refers to
predictions based on model parameters for the current year
while cross refers to model predictions based on the previous
or the following year
The authors would like to acknowledge ARPAVdA, the Aosta
Valley Regional Environmental Authority, and in particular
Giordano Pession and Manuela Zublena, for the provision of
data and suggestions.
Atmos. Sci. Let. 7: 15–20 (2006)
Claggett M, Schrock J, Noll KE. 1981. Carbon monoxide near an
urban intersection. Atmospheric Environment 15(9): 1633–1642.
Dirks KN. 2001. PhD Thesis. Modelling the effect of wind and vehicle
flow on exhaust dispersion at two traffic intersections in Hamilton,
New Zealand. The University of Auckland, Auckland, New Zealand.
Dirks KN, Johns MD, Hay JE, Sturman AP. 2002. A simple
semi-empirical model for predicting missing carbon monoxide
concentrations. Atmospheric Environment 36: 5953–5959.
Finardi S, Carboni G, Tinarelli G. 2001. Analysis of three air
pollution episodes driven by a temperature inversion in a
sub-alpine Italian region. In Proceedings of the Workshop
on Mixing Height and Inversions in Urban Areas, European
Commission COST Action 715, Meteorology Applied to Urban
Air Pollution Problems, Piringer M, Kukkonen J (eds). Brussels,, 99–109, 3–4 October.
Finardi S, Brusasca G, Calori G, Nanni A, Tinarelli G, Agnesod G,
Pession G, Zublena M. 2002. Integrated air quality assessment of
an alpine region: evaluation of the Mont Blanc tunnel re-opening
effects, Proceedings of the 8th Conference on Harmonization with
Atmospheric Dispersion Modeling for Regulatory Purposes. Sofia,
14–17 October, 404–408.
Grell GA, Dudhia J, Stauffer DR. 1994. A description of the Fifth
Generation Pennsylvania State/NCAR Mesoscale Model (MM5).
Boulder, Colorado, National Centre for Atmospheric Research
Report No. NCAR/TN-398+STR.
Copyright  2006 Royal Meteorological Society
K. N. Dirks, A. Nanni and V. I. Dirks
Gronskei KE. 1988. The influence of car speed on dispersion of exhaust
gases. Atmospheric Environment 22(2): 273–281.
Hanna SR, Briggs GA, Hosker PH. 1982. Handbook on Atmospheric
Diffusion. Technical Information Center, U.S. Department of Energy:
Springfield, Virginia.
Hoydysh WG, Dabberdt WF. 1994. Concentrations fields at urban
intersections: fluid modeling studies. Atmospheric Environment
28(11): 1849–1860.
Kassomenos PA, Karakitsios SP, Pilidis GA. 2004. A simple semiempirical approach to modeling benzene concentrations in a street
canyon. Atmospheric Environment 38: 6073–6078.
Ketzel M, Berkowicz R, Muller W, Lohmeyer A. 1999. Dependence of street canyon concentrations on above roof wind
speed – implications for numerical modeling, 6 HARMO 99, Rouen,
11–14 October.
Kukkonen J, Harkonen J, Karppinen A, Pohjala M, Peitarila H,
Koskentalo T. 2001. A semi-empirical model for urban PM10
concentrations, and its evaluation against data from an urban
measurement network. Atmospheric Environment 35: 4433–4442.
Nanni A, Brusasca G, Calori G, Finardi S, Tinarelli G, Zublena M,
Agnesod G, Pession G. 2004. Integrated assessment of traffic impact
in an Alpine region. Science of the Total Environment 334–335:
Zimmerman JR, Thompson RS. 1975. User’s Guide for HIWAY, A
Highway Air Pollution Model, EPA-650/4-74-008. The National
Academies Press: ResearchTrianglePark, North Carolina.
Atmos. Sci. Let. 7: 15–20 (2006)
Без категории
Размер файла
533 Кб
asl, 123
Пожаловаться на содержимое документа