You are on page 1of 11

Science of the Total Environment 409 (2011) 2419–2429

Contents lists available at ScienceDirect

Science of the Total Environment


j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / s c i t o t e n v

STEMS-Air: A simple GIS-based air pollution dispersion model for city-wide


exposure assessment
John Gulliver ⁎, David Briggs
MRC-HPA Centre for Environment and Health, Department of Epidemiology & Biostatistics, Imperial College London, London, UK

a r t i c l e i n f o a b s t r a c t

Article history: Current methods of air pollution modelling do not readily meet the needs of air pollution mapping for short-
Received 12 July 2010 term (i.e. daily) exposure studies. The main limiting factor is that for those few models that couple with a GIS
Received in revised form 14 February 2011 there are insufficient tools for directly mapping air pollution both at high spatial resolution and over large
Accepted 1 March 2011
areas (e.g. city wide). A simple GIS-based air pollution model (STEMS-Air) has been developed for PM10 to
Available online 31 March 2011
meet these needs with the option to choose different exposure averaging periods (e.g. daily and annual).
Keywords:
STEMS-Air uses the grid-based FOCALSUM function in ArcGIS in conjunction with a fine grid of emission
Particulates sources and basic information on meteorology to implement a simple Gaussian plume model of air pollution
PM10 dispersion. STEMS-Air was developed and validated in London, UK, using data on concentrations of PM10 from
Exposure routinely available monitoring data. Results from the validation study show that STEMS-Air performs well in
Dispersion modelling predicting both daily (at four sites) and annual (at 30 sites) concentrations of PM10. For daily modelling,
GIS STEMS-Air achieved r2 values in the range 0.19–0.43 (p b 0.001) based solely on traffic-related emissions and
Air pollution r2 values in the range 0.41–0.63 (p b 0.001) when adding information on ‘background’ levels of PM10. For
annual modelling of PM10, the model returned r2 in the range 0.67–0.77 (P b 0.001) when compared with
monitored concentrations. The model can thus be used for rapid production of daily or annual city-wide air
pollution maps either as a screening process in urban air quality planning and management, or as the basis for
health risk assessment and epidemiological studies.
Crown Copyright © 2011 Published by Elsevier B.V. All rights reserved.

1. Introduction the availability of data to feed the models. Land use regression
techniques, for example, are commonly used for modelling long-term
Traffic is a major source of elevated levels of air pollution in many (i.e. annual) exposures (Hoek et al., 2008) but are not well suited to
European cities. In large urban areas, in particular, variations in traffic dealing with short-term (e.g. daily) modelling as they do not allow for
flows and speeds, meteorology, land use, and terrain lead to complex the influence of meteorology. Dispersion models on the other hand are
patterns of air pollution (Wilson and Zawar-Reza, 2006). As such, even well suited to short-term exposure modelling and can theoretically be
relatively dense networks of routine air pollution monitoring are often used to model for any averaging period if sufficient information is
not sufficient to capture the spatial and temporal patterns of air available on source emissions and meteorology. Nevertheless,
pollution exposures. Reliance on air pollution monitoring alone can although many different dispersion models have been developed,
thus result in exposure misclassification. As an alternative means, very few have the capability for detailed air pollution mapping over
modelling can be used to estimate concentrations of air pollutants at large study areas and are therefore adequate for exposure assessment
potentially any number of locations in time and space (Bellander et al., studies. One of the few models to meet these needs is ADMS-Urban
2001; Brauer et al., 2003; Briggs et al., 2000; Nyberg et al., 2000). (Carruthers et al., 2000). This borrows functionality from ESRI's ArcGIS
Indeed, air pollution models coupled with geographical information to produce high resolution maps of air pollution which can be directly
systems (GIS) are increasingly used to meet the needs of exposure used in GIS-based exposure assessment. Similar functionality has been
assessment in epidemiological studies (Morra et al., 2006, 2009; achieved with the AERMOD (Kesarkar et al., 2007), DUSTRAN (Allwine
Jensen et al., 2009). The type of model applied in these studies largely et al., 2006) and OSPM (Berkowicz et al., 2008) models. Despite these
depends on the averaging period used in the exposure assessment and developments, however, there remain three limiting factors in
applying these models to exposure studies: 1) they have hungry
data demands (e.g. detailed data on source emissions and boundary
⁎ Corresponding author at: Department of Epidemiology & Biostatistics, Imperial
layer meteorology), 2) they are often expensive (i.e. typically between
College London, St Mary's Campus, Norfolk Place, W2 1PG, UK. Tel.: +44 20 7594 5027. £3 and £10k depending on the software and type of licence), and, most
E-mail address: j.gulliver@imperial.ac.uk (J. Gulliver). importantly, and 3) they do not simultaneously deal with large

0048-9697/$ – see front matter. Crown Copyright © 2011 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.scitotenv.2011.03.004
2420 J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429

numbers of emission sources. Indeed, even if data demands can be met,


tens of model runs will usually be required to map air pollution at high
spatial resolution (i.e. b 50 m) over a large urban area.
Improved methods are therefore needed for estimation of
exposures at individual level and over large study populations,
particularly for short-term (e.g. daily) exposure periods. This paper
presents a GIS-based dispersion model (STEMS-Air: ‘Space Time
Exposure Modelling System — Air pollution’) that provides both a
means of simultaneously dealing with large numbers of emission
sources and mapping air pollution over large urban areas. The basic
framework for STEMS was first introduced by Gulliver and Briggs
(2005) which previously relied on third-party, conventional disper-
sion models for estimating air pollution concentrations. STEMS-Air is
intended to embed into the wider STEMS framework. The dispersion
model in STEMS-Air is implemented via a ‘moving-window’ approach
in ArcGIS with grid-based FOCALSUM routines. The grid-based
approach allows for very fast processing of large numbers of emission
sources and by using a fine grid (e.g. 10–20 metre grid cells) ensures Fig. 1. STEMS-Air: model components and data flow.
no significant loss of detail in the underlying emission source
geography. Although this functionality has been used in a handful of
environmental applications (Gerrard et al., 2001; Loh et al., 1996) it levels of regional and long-range (i.e. trans-boundary) particulate air
has seen remarkably little use in air pollution modelling. Collins pollution Each of the mandatory model components and related data
(1998) probably represents the first example of using FOCALSUM for is described, separately, below.
mapping air pollution. In a study in Huddersfield, UK, a distance-
weighting approach was used to sum the contributions of traffic- 2.1.1. Emissions model
related air pollution sources within a fixed diameter ‘kernel’ around The emissions model applies standard DMRB emissions factors
each receptor (i.e. grid cell). The model was validated for NO2 (DEFRA, 2007) to traffic flows (volume and speed), by type of vehicle
(r2 = 0.67) using data from routine air pollution monitoring sites. In (i.e. light or heavy), to produce emission rates (g/m/s) of a given
one of the other few published examples, Loibl and Orthofer (2001) pollutant, for each road link. To account for time-varying traffic flows,
used a similar approach to model NO2 across Austria on a 250 metre emission factor profiles can be applied by hour of the day, and days of
grid. Again models were seen to compare well with data on monitored the week, for specific road links or geographical areas. Where detailed
concentrations: 54% (72 sites) and 76% (101 sites) of predictions fell ‘local’ emissions profiles do not exist, a default set of emission factors
within ± 15 μg/m3 and ± 25 μg/m3 of monitored concentrations, are provided that can apply to the whole road network. The (vector)
respectively. To date, however, the use of FOCALSUM appears to be road geography is subsequently converted into one or more fine grids
limited to simple distance-weighted approaches. Vienneau et al. (e.g. 10 or 20 metre grid cells) of emission rates (depending on the
(2009) used FOCALSUM in ArcGIS with a fixed set of ‘distance weights’ availability of time- and space-varying emissions profiles) for each
to account for the influence of emissions around each receptor (i.e. grid pollutant.
cell receptor) to model concentrations of NO2 across Europe on a 1 km
grid. This approach did not however take account of the influence of
2.1.2. Meteorological pre-processor
meteorology on pollutant dispersion. In this paper the basic approach
Atmospheric conditions in the model are determined by applying
is extended to using a Gaussian dispersion model coupled with
Pasquill (1961) stability categories. The modified Pasquill classifica-
FOCALSUM routines. The paper describes the methods used to develop
tion (Davies and Singh, 1985), as shown in Table 1, is essentially a
the model in a GIS and related data sets, presents the results from a
function of wind speed (generally measured at about 10 m above
validation study using routinely available monitoring data, and makes
ground) and incoming solar radiation. There are a number of other
observations on potential improvements to the approach for further
schemes for assessing atmospheric conditions (Mohan and Siddiqui,
development work. The model is intended for predicting concentra-
1998) but they require measurement of temperature and/or wind
tions averaged over periods of at least 1 day and is not aimed at
speeds at two heights which are not routinely and freely available.
replicating the more advanced models that have been developed with
Incoming solar radiation is determined by the angle between the
the capability of predicting hourly concentrations of different
sun's rays and the Earth's surface, commonly referred to as the zenith
pollutants.
angle (ZN), and the amount of cloud (oktas). Zenith angles vary
for any location on the Earth's surface by time of year and time of day
2. Material and methods

2.1. Model development Table 1


Modified Pasquill stability classes.
STEMS-Air operates is two modes: short-term (daily) and long- Davies and Singh, 1985.
term (annual) to fit with the requirements of many epidemiological Wind speed Daytime incoming solar radiation Within 1 h Night-time
studies. The outputs from short-term mode can be averaged to longer (m/s) (W/m2) before sunset cloud amount
periods (e.g. weeks or months) outside the programme (e.g. in Excel). or after (oktas)
STEMS-Air comprises four model components: 1) an emissions sunrise
Strong Moderate Slight Overcast 0–3 4–7 8
model, 2) a meteorological pre-processor, 3) a GIS-based air pollution (N 600) (300–600) (b 300)
dispersion model, and 4) a GIS-based exposure assessment tool. Fig. 1 ≤ 2.0 A A–B B C D F–G F D
shows the overall structure of STEMS-Air with input and flow of data. 2.1–3.0 A–B B C C D F E D
There is also an optional component where data on ‘background’ 3.1–5.0 B B–C C C D E D D
monitored concentrations can be added from sites (e.g. rural) that are 5.1–6.0 C C–D D D D D D D
≤ ≥ 6.0 C D D D D D D D
not significantly influenced by local source emissions but represent
J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429 2421

(i.e. hour angle) and can be found by Eq. (1) (Mohan and Siddiqui, curves from Briggs (1973) by Venkatram et al. (2005) and Venkatram
1998) and Horst (2006) for application of ‘short-range’ dispersion models in
urban areas. For unstable conditions (A and B stability classes) the
cosZN = sinϕ sinδ + cosϕ cosδ cosω ð1Þ following parameters are applied:

Where δ is the solar declination angle, ϕ is the latitude, and ω is the −1 = 2


σz = 0:24xð1 + 0:001xÞ
hour angle. 0:32x ð6Þ
Declination angle (δ) is in the range ± 0.13π radians, or ± 23.4°, σy =
ð1 + 0:0004xÞ−1 = 2
and can be found by Eq. (2) (Campbell and Norman, 1998).
where x(m) is downwind distance from the source.
sinδ = 0:39785 sin½4:871 + 0:0172J + 0:03345 ð2Þ
For neutral and stable conditions (C to G classes) the following
× sinð 6:2238 + 0:0175J  parameters are applied:

−1 = 2
Where J is the calendar day (i.e. Julian day) — with J = 1 on 1st σz = 0:14xð1 + 0:0003xÞ
−1 = 2 ð7Þ
January and J = 365 on 31st December (or J = 366 in a leap year). σz = 0:16xð1 + 0:0004xÞ
The hour angle (ω) is in the range ± 180°and can be found by the
following formula (3): where x(m) is downwind distance from the source.
STEMS-Air implements the dispersions model with FOCALSUM
cosω = 0:0833πðt−tsn Þ ð3Þ GRID routines programmed in ArcGIS with Arc Macro Language (AML).
This approach has two main advantages over traditional approaches
where t is the hour of the day and tsn is the time of the solar noon. for line sources. Firstly, it overcomes the problem of having to derive
Finally, the amount of cloud cover is used to attenuate the solar an analytical expression for the concentration that accounts for the
radiation calculated under clear sky conditions. line source as a whole (Venkatram and Horst, 2006; Luhar and Patil,
The meteorological pre-processor in STEMS-Air therefore uses 1989). Secondly, grid-based routines are generally very fast GIS
data on day of the year, hour of the day, wind speed (m/s), and cloud operations and so large numbers of sources and receptors can
cover to calculate, for each time period (e.g. hour), the likely incoming simultaneously be modelled to allow rapid production of city-wide
solar radiation (Watts/m2) (N.B. The meteorological inputs required air pollution surfaces. This is of particular benefit when city-wide
are routinely monitored by national meteorological agencies (e.g. exposure assessment is needed for large numbers of individual days
Meteo-France, UK Met. Office) at a height of about 10 m above (e.g. during air pollution episodes which may last for up to 10 days and
ground). The urban sites in this study have weather sensors re-occur several times in 1 year).
positioned at about 5 m above ground. The resulting values for solar Many conventional dispersion models would require a minimum
radiation are used in conjunction with wind speed to determine the of between 25 and 100 model runs, depending on the number of
stability classes, as shown in Table 1. A ‘MET’ file is subsequently emission sources (e.g. road segments) and size of the study area, to
produced and exported to the GIS containing day of the year, hour of undertake city-wide exposure assessment. This may become very
the day, wind speed, wind direction and stability class. Users can time-consuming work, which requires expertise to tessellate the
alternatively import their own data, based on local measurement or emission sources into a manageable set of model runs. In most
other models, if appropriate. models, simply applying a regular grid (e.g. 2 km resolution) over a
city to tessellate the emission sources would result in a very large
2.1.3. Air pollution dispersion model number of model runs (c. 150–200 model runs in London).
The STEMS-Air pollution model is based on the standard solution Furthermore, many receptors (e.g. point locations of residential
of a Gaussian model of dispersion at ground level (Williams, 1990; addresses for exposure estimates) would have to be duplicated across
Venkatram and Horst, 2006). Contributions to the pollutant concen- several model runs as they may lie on the boundary between the
tration from sources along the wind direction vector are found by ‘zones’ used to tessellate the emission sources.
solving Eq. (4): ArcGIS offers a number of options in FOCALSUM that all automatically
    create summary statistics (sum, mean, median, standard deviation) for
Q 1 H 2 those cells falling within a range of different buffer sizes and window
C ðx; y; zÞ = exp − ð4Þ
πUσz 2 σz shapes (e.g. circle, annulus, rectangle, wedge) around each central
processing cell [N.B. These are, in essence, pre-programmed DOCELL
Contributions to pollutant concentrations from emission sources operations]. FOCALSUM can also deal with irregular shapes by providing
that lie at an angle (θ N 0) to the wind direction vector are found by a ‘kernel file’ with individually referenced cells.
solving Eq. (5): STEMS-Air uses the FOCALSUM routine with the ‘wedge’ option.
"   !2 # Contributions to air pollution concentrations from traffic sources are,
Q 1 H 2 y as standard, considered within a 300 metre buffer around each
C ðx; y; zÞ = exp − + ð5Þ
πUσy σz 2 σz σy receptor. In most cases the contributions from sources outside this
distance are sufficiently small to be treated as ‘background’ air
where C is the pollutant concentration at location x, y, and z, Q is the pollution within an urban area. Indeed, an analysis of the contribution
emission source rate (g/m/s), U is the wind speed (m/s), σz is the from traffic source emissions summed within different buffer sizes
standard deviation of the vertical concentration in the plume, σy is the (i.e. 100 metre concentric rings up to 1000 m) at various receptor
standard deviation of the horizontal concentration in the plume, H is sites in London showed that the total contribution beyond 300 m was
the height of the release of the emissions (e.g. 1 m above ground for b 2% of predicted concentrations. In some situations, however, traffic
ground-level sources), and Y is the distance along the normal from levels beyond 300 m are sufficiently high (e.g. motorways) that the
emission source to the plume centreline. buffer distance around a receptor needs to be extended. The user can
The stability categories, determined in the meteorological pre- achieve this by adding an additional buffer (e.g. up to 500 m from
processor, along with wind speed and wind direction, are used as the source) in the programme.
basis for calculating the vertical (σz) and horizontal (σy) dispersion Prior to model development, regression analysis was used to
parameters. These are applied using a formulation of the dispersion determine the optimum internal angle of the inner and outer wedges.
2422 J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429

A series of wedges at 5 degree intervals between 20 and 60° were no limitation to the spatial resolution of the grid but clearly processing
applied to the emissions grid using FOCALSUM, Wedges of 45° were time will significantly increase with finer spatial resolution (e.g.
seen to provide best performance in predicting monitored 10 m). Daily or annual estimates of exposure can be determined in
concentrations. point or grid mapping modes. Resulting data can be exported as
The procedure for implementing the programme in GRID is standard delimited text files for further analysis (e.g. epidemiological
illustrated in Fig. 2 as follows: analysis or health assessment).

1. For each receptor, create a FOCALSUM ‘wedge’ along the line of the
2.2. Model application
wind direction with an internal angle of 45° and a radius of 300 m.
Create two outer wedges adjacent to the inner wedge each with
Modelling was carried out in London, UK, because data on road
internal angles of 45° and of radius 40 m. The purpose of the outer
geography and traffic flow/composition are readily available from the
wedges is to account for the relatively small contributions likely to
London Atmospheric Emissions Inventory (LAEI), and also there is a
arise from neighbouring source cells that are positioned normal to
relatively dense network of air pollution monitoring sites for model
the wind direction.
validation. PM10 was chosen as the reference pollutant because of the
2. Intersect a series of circular buffers at 20 metre intervals (i.e. to
strong body of epidemiological evidence linking particulates with
match the resolution of the emissions grid and thus retain the
negative effects on health (Brunekreef and Holgate, 2002; Dockery
highest level of precision in the dispersion modelling) with the
and Pope, 2006). STEMS-Air was implemented for both short-term
wedges to create segments inside each wedge.
(i.e. daily) and long-term (i.e. annual) air pollution modelling. Short-
3. Using FOCALSUM, intersect each wedge and segment with the
term modelling was undertaken for July to December 2007 as data on
source emissions geometry (i.e. gridded emissions).
co-located hourly traffic composition and speeds, meteorological data,
4. Sum the emissions within each segment and then apply the
and air pollution monitoring were available during this period at a
dispersion calculation, using Eq. (4) for the inner wedge and Eq. (5)
number of sites. Long-term modelling was applied to 2001–2003 as
for the outer wedge. Finally sum the contribution from each wedge
this period represented the best coverage of air pollution monitoring
to provide a modelled PM10 concentration at the receptor.
sites (n = 30) with complete data records across three consecutive
years.
Ideally, separate calculations would be made between each
emission grid cell and receptor but this becomes a very protracted
process to implement when large numbers of cells are used and 2.2.1. Short-term modelling
particularly when the orientation (i.e. due to wind direction) of the Data from co-located traffic and air pollution monitoring at six
reference cells change in relation to the central processing cell. The roadside/kerbside sites (from the London Air Quality Network) were
use of FOCALSUM is therefore somewhat of a compromise, but this kindly provided by King's College London and used as the primary
approach nonetheless remains relatively fast for dealing with large basis for testing and validation of STEMS-Air in relation to short-term
numbers of emission sources. (i.e. daily) modelling. The data covered 184 days between 1st July and
31st December 2007 and comprised hourly traffic composition for
2.1.4. Exposure assessment different vehicle categories (i.e. car, motorcycles, buses, heavy goods
Exposure assessment can be performed for a series of points (e.g. etc.), counts of the number of vehicles in each category by vehicle
residential address locations) by entering co-ordinates into the speed in 10 km/h intervals, monitored concentrations of PM10, and
model, or for areas (e.g. census tracts) by their intersection with wind speed and wind direction from meteorological data co-located at
maps of air pollution from STEMS-Air. In grid mapping mode there is some of the sites. Days related to unusual events (e.g. ‘bonfire week’ in

Fig. 2. STEMS-Air dispersion modelling procedure.


J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429 2423

November) and days with unusual traffic patterns (e.g. the Christmas different pattern of emissions during the evening period on Fridays.
period from mid-December) were subsequently removed to leave a The profiles were assumed to be applicable to all other roads within
total of 149 days of data. 300 m of the traffic monitoring site as information on time-varying
Data were screened for completeness and reduced to the following factors for other roads in area surrounding each site was not available.
four sites that had good simultaneous coverage in traffic and There was no information available for time-varying vehicles speeds,
monitored concentrations of PM10: Brent 4 (BT4), Greenwhich 9 so the long-term average speed at each site was applied to all hours
(GR9), Hackney 6 (HK6), and Tower Hamlets 4 (TH4). Fig. 3 shows a for all roads surrounding each site. Emission weights were subse-
map of the locations of these sites. All sites are located alongside busy quently applied to the baseline emissions data using the profiles in
urban roads and in relatively clear suburban settings, with the Fig. 4 to provide time-varying emission rates on all source emission
exception of HK6 which is in a street canyon with multi-storey grids.
buildings. Also shown are the location and type (i.e. kerbside, roadside, Co-located hourly data on wind speed and wind direction were
urban background, suburban) of sites used in long-term modelling available with good coverage at all sites except for BT4 which was
(see below). substituted with data from TH4 — the nearest similar site type.
Data on road geography and traffic (on roads except those adjacent Data on cloud cover from Heathrow Airport, which is the nearest
to the traffic monitoring site) within a 300 metre buffer of each available site, were downloaded from the British Atmospheric Data
monitoring sites were taken from the LAEI. The LAEI contains data on Centre (www.badc.ac.uk). These data were linked to the data on wind
annual average daily traffic (AADT) and composition (number of speed and direction and entered into the meteorological pre-
light and heavy vehicles) for over 63,000 individual road links within processor to provide hourly data on atmospheric conditions in
the Greater London Area. Each link also is attributed with a single, STEMS-Air. STEMS-Air was then run for each of the four monitoring
average vehicle speed that applies to all vehicle categories. Road sites to predict daily average concentrations of PM10 (μg/m3). The
geography is based on data from Ordnance Survey Master-Map© and model was run at 1-hour time resolution to account for the effects of
has a ground precision of b 1 m and includes individual carriageways and variability in meteorology on individual days (N.B. the concentration
roundabouts. based on 24 hourly values is not the same as meteorological variables
The emissions model was run with separate emission factors for averaged over 1 day).
different vehicle speeds and categories (i.e. light/heavy) on all roads
falling in a 300 metre circular buffer around each of the four 2.2.2. Long-term modelling
monitoring sites. Emissions were summed to provide total emission Data on monitored concentrations of annual average PM10
rates of PM10 on each road link. The time-varying traffic data from between 2001 and 2003, for a total of 53 locations (including site
each monitoring site was then analysed to provide weightings (i.e. coordinates and site type), were downloaded from the website of The
‘local emission profiles’ in Fig. 1) for each hour of the day and day of London Air Quality Network (http://www.londonair.org). These years
the week. As Fig. 4 shows, the emission profiles vary both between site covered a period when relatively minor intervention took place in
in the magnitude of weights and in hour-to-hour pattern, but opening/closing of monitoring sites and thus provided a good basis for
generally the profiles naturally fall into three groups: weekdays, year-to-year comparison between sites. A reduced list of sites were
Saturdays, Sundays. The exception is at Hackney 6 which has a subsequently produced (n = 30) to include only those sites that

Fig. 3. Air pollution monitoring sites used in the validation study.


2424 J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429

Fig. 4. Time-varying emission weights for daily air pollution modelling sites.

passed the following criteria: data were available for at least 75% of shown in Fig. 3. Model performance was assessed on the basis of the
days within each year, data were denoted as fully ratified, and site validation analysis by computing adjusted r2, the root mean square
coordinates provided on the website could be verified or adjusted error (RMSE) and fractional bias (FB) — defined for each site as:
using satellite imagery from Google Earth. These monitoring sites
were used as the basis for modelling annual average PM10 from  
STEMS-Air. Cp −Co
Data on traffic flows and composition were taken from the London FB =  
0:5 Cp + Co
Atmospheric Emissions Inventory (LAEI) for each year corresponding
to the above-mentioned air pollution monitoring data. Average
emission rates were then computed in the emissions model for all
roads falling within 300 m of the 30 air pollution monitoring sites. where Co is the mean observed concentration and Cp is the mean
Data on hourly meteorological parameters (wind speed, wind predicted concentration.
direction, cloud cover) were downloaded for Heathrow Airport
from BADC for the period 1st January 2001 to 31st December 2003. 3.1. Short-term modelling
These data were then entered into the meteorological pre-processor
to obtain data on atmospheric stability for each hour of each day. Comparisons between monitored and predicted concentrations of
The resulting data were then summarised as frequency data: the daily average PM10 were made for the four sites shown in Fig. 3.
number of occasions (e.g. hours during the year) on which the same Summary statistics for model performance are shown in Table 2 and
combination of meteorological variables were recorded (e.g. number scatterplots between monitored and predicted concentrations at each
of hours during the year with frequency of wind direction of 30°, site are shown in Fig. 5. Modelling was done for nearly all of the days
wind speed of 2 m/s, and cloud cover equals 8 oktas). This method available for comparison with monitored concentrations except in the
reduces the hourly data by about 8-fold to speed up the model case of GR9 where only 54 days of monitored concentrations were
runs with no loss of information as the predictions for each line of available.
met data are averaged to provide an annual mean concentration in
long-term modelling. STEMS-Air was then run to model annual
mean concentrations of PM10 for the 30 monitoring sites shown in Table 2
Fig. 3. Summary statistics for modelled daily traffic-related PM10 (μg/m3).

Site Adjusted r2 RMSE FB N (days)


3. Results BT4 — Brent 4 0.43 17.55 − 0.84 140
GR9 — Greenwich 9 0.43 18.01 − 1.37 54
Model validation was undertaken by comparing predicted PM10 HK6 — Hackney 6 0.19 17.85 − 1.06 143
TH4 — Tower Hamlets 4 0.40 29.86 − 1.31 144
concentrations with monitored PM10 concentrations at the sites
J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429 2425

Fig. 5. Scatter-plots of modelled (local traffic) versus monitored concentrations of PM10 (μg/m3), July to December 2007.

As Table 2 shows, model performance is similar for BT4, GR9 and The results of combining modelled concentrations from the local
TH4 with adjusted r2 values of 0.43, 0.43, and 0.40 respectively model of traffic and ‘background’ concentrations are shown in Table 3
(p b 0.001), with the site at HK6 having a lower r2 of 0.19 (p b 0.001). and Fig. 6. As can be seen, model performance variably improves (r2:
The lower r2 at HK6 can perhaps be explained by this site being 0.41–0.61, p b 0.001) at all sites as a result of adding the data on rural
characterised by a continuous street canyon, the effects of which are monitored concentrations. The largest improvement in terms of r2 is at
not considered in STEMS-Air. Values of RMSE and FB are relatively HK6: r2 = 0.41. Values of RMSE and FB are, as would be expected,
high for all sites, as would be expected when only including the local, lowered at all sites. RMSE is still high at TH4 (12.75 μg/m3) compared
traffic component. Values of FB lie in the range − 0.84 to − 1.37 — a FB to the other sites (7.48 μg/m3–8.66 μg/m3) but greatly reduced with
value of − 1.00 represents under-prediction by a factor of 3. For both the addition of daily mean ‘background’ PM10 concentrations.
measures, results are weakest at TH4 with under-prediction by about Including rural concentrations of PM10 in STEMS-Air results in
a factor of 4 and RMSE of 29.86 μg/m3. significant improvement in model performance.
The results are based on comparisons made against monitored
PM10 which includes all sources (e.g. traffic, secondary particulates, 3.2. Long-term modelling
wind-blown soil and dust etc.), so in terms of the explained variation in
traffic-related PM10 the model is likely performing relatively well. Comparisons between predicted and monitored concentrations of
Indeed, examination of the ratio of PM10 concentrations between annual average PM10 were made for the 30 sites with data available
those at rural, urban background and roadside sites (by the authors) between 2001 and 2003. Annual mean concentrations of ‘background’
showed that, on average, local traffic-related emissions account for no air pollution were obtained from Harwell for each year and were
more than 10–30% of monitored concentrations at urban background added to the modelled concentrations — there were insufficient data
locations, and 30–50% at urban roadside locations: the remainder is to consider the rural site at Rochester.
attributable to regional pollution, present at the remote rural sites. In Table 4 shows the summary statistics for performance of STEMS-Air
order to consider the influence of regional background PM10 here, with the addition of concentrations of ‘background’ PM10.
therefore, daily mean concentrations of PM10 from rural monitoring As can be seen, models perform well in each year (r2: 0.67–0.77)
data were added to the modelled predictions from STEMS-Air. Rural with the best result in terms of r2, RMSE and FB for 2001 (0.77, 3.32, −
PM10 concentrations were taken as the average of data from the only 0.08, respectively). Results are weakest for 2003, especially in terms of
two rural monitoring stations in the south east of England: Harwell RMSE: 10.75 μg/m3. It should be noted, however, that concentrations
(70 km due west of central London), and Rochester (60 km east- were generally inflated in 2003 compared to the other years due to a
south-east of central London). number of air pollution episodes, with four periods of unusually high
2426 J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429

Table 3 Table 4
Summary statistics for modelled local, traffic-related and regional ‘background’ PM10 Performance statistics for predictions of annual average of PM10 (μg/m3), 2001–2003.
(μg/m3).
Year Adjusted r2 RMSE FB N (sites)
Site Adjusted r2 RMSE FB N (days)
2001 0.77 3.32 − 0.08 30
BT4 — Brent 4 0.61 7.54 0.16 140 2002 0.68 4.98 − 0.20 30
GR9 — Greenwich 9 0.58 8.66 0.22 54 2003 0.67 10.75 − 0.37 30
HK6 — Hackney 6 0.41 7.48 0.16 143
TH4 — Tower Hamlets 4 0.53 12.75 − 0.19 144

background/suburban) based on the description of site types from the


national air quality archive (www.airquality.co.uk).
PM10 levels in south east England. August 2003, for example, saw a As Table 5 indeed shows, the calibrated model reduces RMSE to
period of 10 days when daily mean concentrations of PM10 exceeded between about one third and one half of the levels found from the un-
50 μg/m3 (Stedman, 2004). calibrated model (Table 4). Also noteworthy is that the constant (i.e.
As demonstrated by the negative values of FB, the model under- background) from the regression is broadly similar to that of the average
predicts in all years, but only very slightly in 2001 (FB = − 0.08). concentration from the rural monitoring sites in 2001 (16.76 μg/m3)
Furthermore, there is over-prediction for some, mostly background sites and 2002 (18.78 μg/m3), but in 2003 the regression yields a background
in 2001. This is due to the rural concentrations being higher than levels value of 25.61 μg/m3 compared to the rural concentration of
recorded at background sites in London, which likely come from higher 20.31 μg/m3 at the rural site. Only three of the London sites had
proportions of non-anthropogenic sources at the rural site. monitored values lower than the constant from the regression for 2003.
To improve model performance, linear regression was used to The coefficients (i.e. slope) of the regression are 2.92, 3.92, and 5.79 for
calibrate the model predictions to the monitored concentrations for 2001, 2002 and 2003, respectively. Given the values of the constant in
each year. Summary statistics for the calibrated models are show in the regression, and notwithstanding that some sources are not included
Table 5. Fractional bias is not included as zero mean bias is implied by in the modelling (e.g. wind-blown soil and dust, and vehicle tyre and
a linear regression. Fig. 7 shows scatterplots of modelled versus brake wear), these values imply that STEMS-Air systematically under-
monitored concentrations for the same data with sites classified as predicts annual concentrations of traffic-related PM10. It must be borne
either roadside (i.e. roadside/kerbside) or background (i.e. urban in mind, however, that meteorological parameters were only available

Fig. 6. Scatter-plots of modelled (local traffic + rural) versus monitored concentrations of PM10 (μg/m3), July to December 2007.
J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429 2427

Table 5 by about three-fold if using wind speed from less exposed urban
Calibration models for predictions of annual average PM10 (μg/m3), 2001–2003. locations.
Year a(x) Constant SEE RMSE N (sites) Fig. 7 shows that STEMS-Air reasonably well explains the
magnitude of variation in the traffic contributions to concentrations
2001 2.92 18.12 1.85 1.90 30
2002 3.92 18.04 2.19 2.34 30 of PM10. There are some anomalies to this which may be due to the
2003 5.79 25.61 3.31 3.47 30 exclusion of other, local (i.e. domestic, industrial) sources in the
model. As can be seen in Fig. 7, one site, Marylebone (kerbside), has
consistently higher concentrations than all other sites. Clearly this site
from Heathrow Airport which is relatively exposed compared to the is influential on the performance of the model in terms of r2. To look at
locations of the air pollution monitoring sites in London. The typically the effect of this site on overall model performance, values of r2 were
higher levels of wind speed recorded at Heathrow Airport result in obtained when excluding Marylebone from the analysis. As expected,
smaller modelled concentrations from STEMS-Air. Moreover, this values of r2 are reduced but the results were still seen to be good in all
magnitude of under-prediction was not seen at any of the sites in the years (r2: 0.56–0.67, p b 0.001).
short-term modelling where meteorological parameters were more For comparative purposes, modelling was also undertaken with
representative of ‘urban’ conditions. To look at this area in more detail, ADMS-Urban (Carruthers et al., 2000), a proprietary dispersion model,
the short-term modelling was done for a second time by substituting for the same validation data described above, replicating as much as
wind speed at each of the four sites (BT4, GR9, HK6 and TH4) with wind possible the inputs (i.e. emission rates and factors, meteorology) used
speed from the meteorological site at Heathrow. Over the 6-month in STEMS-Air. ADMS-Urban was thus offered information on traffic
period from July to December 2007 average wind speeds at Heathrow emissions (as vectors) within a 300-metre circular buffer of each
were approximately three times higher than those found at the urban monitoring site, and data on background concentrations from the
sites and resulting predictions of PM10 were on average 67.5% lower same rural monitoring stations. Tables 6 and 7 show results of the
than seen when using local wind speed from the four sites. This suggests validation analysis for ADMS-Urban for short-term modelling and
that the coefficients of the regression shown in Table 5 might be reduced long-term modelling, respectively. ADMS-Urban did not produce data,

Fig. 7. Scatterplots of modelled versus monitored annual average PM10 concentrations, 2001–2003.
2428 J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429

Table 6 (Hoek et al., 2008) for modelling annual mean concentrations of


Summary statistics for modelled local, traffic-related and regional ‘background’ PM10 particulates and NO2. In general, LUR models have been seen to
(μg/m3) from ADMS-Urban.
perform well with validation studies yielding r2 values typically in the
Site Adjusted r2 RMSE FB N (days) range 0.5 to 0.7 (Hoek et al., 2008). The results for LUR models are
BT4 — Brent 4 0.46 6.94 − 0.06 138 broadly similar to those obtained here, but STEMS-Air, apart from in
GR9 — Greenwich 9 0.54 8.37 0.15 54 all likelihood still taking longer to produce modelled exposures, has
HK6 — Hackney 6 0.30 6.60 0.07 143 three distinct advantages over LUR models. Firstly, STEMS-Air is
TH4 — Tower Hamlets 4 0.34 15.73 − 0.31 142
transferable between study areas whereas LUR models, which tend to
be built with local source emissions data (e.g. traffic flows, road
geography, land use) and are calibrated to local air pollution
due to incomplete information on meteorology, for two of the days monitoring data, may not transfer well (Hoek et al., 2008). Secondly,
modelled at BT4 and TH4 (see Table 6). STEMS-Air can directly provide modelled concentrations that reflect
As shown in Table 6, in terms of r2 the performance of ADMS-Urban the spatial effects of yearly patterns in meteorology, unlike LUR
(r2: 0.30–0.54) is weaker than seen for STEMS-Air (r2: 0.41–0.61), models which have to rely on some temporally static emissions data
though nearly the same for GR9. Conversely, however, ADMS-Urban (e.g. land use) and can only be universally calibrated to monitoring
performs marginally better than STEMS-Air in terms of RMSE and FB at data for each year. Thirdly, STEMS-Air is directly based on emissions,
three out of four sites. Correlations between modelled and monitored so can be used to estimate future concentrations (e.g. under different
data and other measures of performance should, however, improve for policy scenarios), or to carry out hind-casts for past (different)
ADMS-Urban if more detailed data on meteorology (e.g. temperature, emission situations. Given these advantages, and with its simple
rainfall, and air pressure), information on street canyons, and models implementation and capacity for modelling city-wide air pollution,
of terrain are offered. For long-term modelling, as shown in Table 7, in STEMS-Air thus represents a useful alternative to LUR modelling in
terms of r2 ADMS-Urban (r2: 0.71–0.77) performs slightly better than long-term exposure assessment studies.
STEMS-Air (r2: 0.67–0.77) for two (2002 and 2003) out of the 3 years. Although STEMS-Air was seen to generally perform well, it needs
Values of RMSE and FB are broadly the same for the two models. to be recognised that it does not provide a substitute for formal
dispersion modelling where detailed data are needed for a small
4. Discussion number of locations and sources. Its role is essentially as a screening
model or mapping tool, when high resolution and time-varying
The validation results presented suggest that STEMS-Air can be estimates of air pollution are needed across large study areas or
applied to both short-term and long-term modelling of PM10. populations. A number of potential improvements could nevertheless
Attempts to make comparisons between these results and those be made to address some of the inherent weaknesses in the current
from other studies are difficult, however, simply because particle model formulation. Firstly, the current version of the model does not
validation studies are not available for many models (Holmes and consider local, non-traffic sources (i.e. domestic, industrial). In the UK,
Morawska, 2006). There are a few notable examples that have estimated annual emissions from these sources are routinely available
evaluated the performance for dispersion models in predicting daily on a 1 km grid (e.g. National Atmospheric Emissions Inventory), with
average PM10. In a study in Christchurch, New Zealand, Wilson and explicit point locations for large industrial sources (e.g. stacks). For
Zawar-Reza (2006), for example, compared predictions of daily area sources (e.g. 1 km grids) it would be possible add a ‘box model’ to
average PM10 from TAPM (The Air Pollution Model) with monitored STEMS-Air similar to those included in other dispersion models such
PM10 concentrations at 12 monitoring sites over 59 days during the as ADMS; for point sources emissions, an adaptation of the general
winter of 2003–2004. Results were reported in terms of RMSE and Gaussian dispersion model cited above could be made to account for
correlation (r). Generally higher values of RMSE were seen than those source height. A second improvement would be the addition of a
reported here, with 10 out of the 12 sites yielding values N 30 μg/m3, simple street canyon model, such as the one described by European
though the range of monitored concentrations were similar to this Environment Agency at http://www.eea.europa.eu/publications/
study. Correlations are not directly comparable but were broadly TEC11a/page014.html. A canyon model would require data on, or a
weaker (mostly in the range r = 0.4–0.5) than those seen in this study. surrogate for, building heights which have traditionally been difficult
Gokhale and Raokhande (2008) compared results from three different to obtain/derive. From 2009, however, Light Detection and Ranging
models (CALINE3, CAL3QHC, and M-GFLSM) against daily mean Data (LIDAR) (Cici et al., 2009) from which building heights can be
concentrations of PM10 and PM2.5 at a single traffic site in the city of derived will be freely available (for research purposes) in the UK for
Guwahati, India, over 4 months during the winter of 2005/2006. The large conurbations. LIDAR could theoretically be used to provide data
study reported fractional bias between ± 0.01 and ± 0.5 for the three on streets with canyons of continuous and significant height. The data
models. Again these results are broadly comparable with those could directly be added into STEMS-Air via a second grid to act as a
presented here when the ‘background’ air pollution model is applied reference to where a canyon model should be applied. The
(FB: − 0.19 to 0.22). meteorological sub-model used in STEMS-Air is also relatively simple,
Although dispersion modelling remains the only viable basis for and does not make use of the more advanced algorithms that have
short-term modelling of air pollution exposures, for long-term (i.e. become available in recent years (e.g. to describe boundary layer
annual) modelling there are a range of available techniques (Jerrett conditions). Enhancement of this component would be possible, but
et al., 2005). Of the more advanced techniques, land use regression at the expense of increasing data demands. Finally, further work
(LUR) modelling has been used widely in Europe and North America should include modelling and validation of range of pollutants which
might be required for air quality management and exposure
assessment, including key urban pollutants such as NO2 and CO.
Table 7
Performance statistics for predictions of annual average of PM10 (μg/m3) from ADMS-
Urban, 2001–2003. 5. Conclusions

Year Adjusted r2 RMSE FB N (sites)


A GIS-based air pollution dispersion model has been developed and
2001 0.77 5.04 − 0.20 30 applied in London for daily and annual modelling of PM10. In broad
2002 0.71 3.81 − 0.10 30 terms, the model was seen to produce good predictions of monitored
2003 0.73 11.06 − 0.38 30
concentrations for different time periods and in at different locations. A
J. Gulliver, D. Briggs / Science of the Total Environment 409 (2011) 2419–2429 2429

number of improvements to the model and directions for future work Davies ME, Singh S. Thorney Island: its geography and meteorology. Journal of
Hazardous Materials 1985;11:91-124.
have been cited. The result of the study is a simple dispersion model and DEFRA. Design manual for roads and bridges. UK: HMSO; 2007.
exposure assessment tool that can be implemented for screening and Dockery DW, Pope CA. Health effects of fine particulate air pollution: lines that connect.
mapping purposes by lay users. Incorporation of the ArcGIS FOCALSUM Journal of the Air & Waste Management Association 2006;56:709–42.
Gerrard R, Stine P, Church R, Gilpin M. Habitat evaluation using GIS: a case study
routine, at the heart of the model, allows relatively quick production of applied to the San Joaquin kit fox. Landscape and Urban Planning 2001;52:239–55.
high resolution, city-wide air pollution maps. The maps and exposure Gokhale S, Raokhande N. Performance evaluation of air quality models for predicting
assessment tool could be applied in epidemiological studies and health PM10 and PM2.5 concentrations at urban traffic intersection during winter period.
Science of the Total Environment 2008;394:9-24.
risk assessment looking at links between air pollution and human health Gulliver J, Briggs DJ. Time–space modelling of journey-time exposure to traffic-related
over daily or longer-term (e.g. annual) periods. air pollution using GIS. Environmental Research 2005;97(1):10–25.
Hoek G, Beelen R, de Hoogh K, Vienneau D, Gulliver J, Fischer P, Briggs D. A review of
land-use regression models to assess spatial variation of outdoor air pollution.
Acknowledgements
Atmospheric Environment 2008;42:7561–78.
Holmes NA, Morawska L. A review of dispersion modelling and its application to the
The research presented here was initially developed as part of the dispersion of particles: an overview of different dispersion models available.
EU 5th Framework Programme on Health Effects and Risks of Atmospheric Environment 2006;40:5902–28.
Jensen SS, Larson T, Deepti KC, Kaufman JD. Modeling traffic air pollution in street
Transport Systems (HEARTS), led by the World Health Organisation canyons in New York City for intra-urban exposure assessment in the US Multi-
(Rome), with further development taking place as part of the EU 7th Ethnic Study of atherosclerosis and air pollution. Atmospheric Environment
Framework Programme on GENeric European Sustainable Informa- 2009;43:4544–56.
Jerrett M, Arain A, Kanaroglou P, Beckerman B, Potoglou D, Sahsuvaroglu T, Morrison J,
tion Space for environment (GENESIS). The authors wish to thank Giovis C. A review and evaluation of intraurban air pollution exposure models.
King's College London for providing data on traffic flows and speeds Journal of Exposure Analysis and Environmental Epidemiology 2005;15:185–204.
and meteorology from six roadside air pollution monitoring sites in Kesarkar AP, Dalvi M, Kaginalkar A, Ojha A. Coupling of the weather research and
forecasting model with AERMOD for pollutant dispersion modeling. A case study
London. for PM10 dispersion over Pune, India. Atmospheric Environment 2007;41(9):
1976–88.
References Loh DK, Van Stipdonk SEP, Holtfrerich DR, Hsieh Yi-Te C. Spatially constrained
reasoning for the determination of wildlife foraging areas. Computers and
Allwine KJ, Rutz FC, Shaw WJ, Rishel JP, Fritz BG, Chapman EG, Hoopes BL, Seiple TE. Electronics in Agriculture 1996;15:323–34.
DUSTRAN 1.0 user's guide: a GIS-based atmospheric dust dispersion modelling Loibl W, Orthofer R. From national emission totals to regional ambient air quality
system. Technical report PNNL-16055. Richland, Washington: Pacific Northwest information for Austria. Advances in Environmental Research 2001;5:395–404.
National Laboratory; 2006. Luhar KA, Patil RS. A general finite line source model for vehicular pollution prediction.
Bellander T, Berglind N, Gustavsson P, Jonson T, Nyberg F, Pershagen G, Järup L. Using Atmospheric Environment 1989;23:555–62.
geographic information systems to assess individual historical exposure to air Mohan M, Siddiqui TA. Analysis of various schemes for the estimation of atmospheric
pollution from traffic and house heating in Stockholm. Environmental Health stability classification. Atmospheric Environment 1998;32(21):3775–81.
Perspectives 2001;109(6):633–9. Morra P, Bagli S, Spadoni G. The analysis of human health risk with a detailed procedure
Berkowicz R, Ketzel M, Jensen SS, Hvidberg M, Raaschou-Nielsen. Evaluation and operating in a GIS environment. Environment International 2006;32:444–54.
application of OSPM for traffic pollution assessment for a large number of street Morra P, Lisi R, Spadoni G, Maschio G. The assessment of human health impact caused
locations. Environmental Modelling & Software 2008;23:296–303. by industrial and civil activities in the Pace Valley of Messina. Science of the Total
Brauer M, Hoek G, Meliefste K, Fischer P, Gehring U, et al. Estimating long-term average Environment 2009;407:3712–20.
particulate air pollution concentrations: application of traffic indicators and Nyberg F, Gustavsson P, Järup L, Bellander T, Berglind N, Jakobsson R, et al. Urban air
geographic information systems. Epidemiology 2003;14(2):228–39. pollution and lung cancer in Stockholm. Epidemiology 2000;11:587–95.
Briggs GA. Diffusion estimation for small emissions. ERL, ARL USAEC Report ATDL-106. Pasquill F. The estimation of the dispersion of windborne material. Meteorology
Oak Ridge, TN: US Atomic Energy Commission; 1973. Magazine 1961;90(1063):33–40.
Briggs DJ, de Hoogh C, Gulliver J, Wills J, Elliott P, Kingham S, et al. A regression-based Stedman JR. The predicted number of air pollution related deaths in the UK during the
method for mapping traffic-related air pollution: application and testing in four August 2003 heatwave. Atmospheric Environment 2004;38(8):1087–90.
contrasting urban environments. Science of the Total Environment 2000;253(1–3): Venkatram A, Horst TW. Approximating dispersion from a finite line source.
151–67. Atmospheric Environment 2006;40:2401–8.
Brunekreef B, Holgate ST. Air pollution and health. The Lancet 2002;360:1233–42. Venkatram A, Isakov V, Pankratz D, Yuan J. Relating plume spread to meteorology in
Campbell GS, Norman JM. Introduction to environmental biophysics. 2nd ed. Springer- urban areas. Atmospheric Environment 2005;39:371–80.
Verlag: New York; 1998. p. 167–83. Vienneau D, de Hoogh K, Briggs D. A GIS-based method for modelling air pollution
Carruthers DJ, Edmunds HA, Lester AE, McHugh CA, Singles RA. Use and validation of exposures across Europe. Science of the Total Environment 2009;408:255–66.
ADMS-Urban in contrasting urban and industrial environments. International Williams ML. Atmospheric dispersal of pollutants and the modelling of air pollution
Journal of Environmental Pollution 2000;14:1–6. (chapter 10). In: Harrison RM, editor. Pollution, causes, effects and control. Royal
Cici A, Smith-Voysey S, Jarvis C, Tansey K. Integrating building footprints and LiDAR Society of Chemistry, 2nd editionCambridge, U.K: Thomas Graham House; 1990.
elevation data to classify roof structures and visualise buildings. Computers, p. 201–19.
Environment and Urban Systems 2009;33(4):285–92. Wilson JG, Zawar-Reza P. Intraurban-scale dispersion modelling of particulate matter
Collins, S., 1998. A GIS approach to modelling traffic related air pollution. PhD thesis. concentrations: applications for exposure estimates in cohort studies. Atmospheric
The University of Huddersfield, UK. Environment 2006;40:1053–63.

You might also like