- Open Access
A study of spatial resolution in pollution exposure modelling
© Stroh et al; licensee BioMed Central Ltd. 2007
- Received: 27 March 2007
- Accepted: 04 June 2007
- Published: 04 June 2007
This study is part of several ongoing projects concerning epidemiological research into the effects on health of exposure to air pollutants in the region of Scania, southern Sweden. The aim is to investigate the optimal spatial resolution, with respect to temporal resolution, for a pollutant database of NOx-values which will be used mainly for epidemiological studies with durations of days, weeks or longer periods. The fact that a pollutant database has a fixed spatial resolution makes the choice critical for the future use of the database.
The results from the study showed that the accuracy between the modelled concentrations of the reference grid with high spatial resolution (100 m), denoted the fine grid, and the coarser grids (200, 400, 800 and 1600 meters) improved with increasing spatial resolution. When the pollutant values were aggregated in time (from hours to days and weeks) the disagreement between the fine grid and the coarser grids were significantly reduced. The results also illustrate a considerable difference in optimal spatial resolution depending on the characteristic of the study area (rural or urban areas). To estimate the accuracy of the modelled values comparison were made with measured NOx values. The mean difference between the modelled and the measured value were 0.6 μg/m3 and the standard deviation 5.9 μg/m3 for the daily difference.
The choice of spatial resolution should not considerably deteriorate the accuracy of the modelled NOx values. Considering the comparison between modelled and measured values we estimate that an error due to coarse resolution greater than 1 μg/m3 is inadvisable if a time resolution of one day is used. Based on the study of different spatial resolutions we conclude that for urban areas a spatial resolution of 200–400 m is suitable; and for rural areas the spatial resolution could be coarser (about 1600 m). This implies that we should develop a pollutant database that allows different spatial resolution for urban and rural areas.
- Coarse Grid
- Fine Grid
- Coarse Resolution
- Environmental Impact Assessment
- Polynomial Interpolation
The air we breathe contains a vast number of chemical pollutants. Some originate from natural sources such as soil or salt particles generated from nearby fields or the sea, but most of them originate from man-made sources such as vehicles, industrial plants, combustion and heating. These chemicals not only interact with each other but also with the cells and compounds within our respiratory and cardiovascular systems. Studies have shown that exposure to air pollutants such as fine particles and nitrogen dioxide contributes to excess diseases and mortality among people suffering from cancer, respiratory and cardiovascular diseases . It has also been shown that air pollution leads to poorer health among vulnerable groups such as asthmatics, old people and children. The health effects on children are of special concern since these health problems might have repercussions later in life .
Geographic analyses have previously been used in epidemiology to study the correlation between exposure to air pollutants and health [3, 4]. Some of these analyses were conducted in three steps: (1) modelling of the pollutant field using a dispersion model and an emission source database, (2) using the pollutant field to estimate the exposure to pollutants of the study group, and (3) correlation of the exposure with health status of the study group.
This study is part of several ongoing projects involving epidemiological research in the region of Scania in southern Sweden. These projects employ population databases obtained from the Regional Office of Scania and are based on the Swedish National Population Register. These population databases are comparatively detailed regarding both the high degree of detail in the socio-economic and health data they contain, but also in their spatial resolution. All the databases contain information on individuals, including the exact location of their residence. This location is either represented as a point at the centre of the real estate where they live, or as a centroid in a 500 m-1 km grid. This detailed information regarding the population in Scania enables us to perform exposure studies with high resolution.
An emission source database including traffic, industry, and combustion was established and dispersion modelling software implemented . However, it was found to be impractical to conduct dispersion modelling for each epidemiological study. Therefore, we plan to establish a pollutant database of NOx values; this database will contain pollutant values in a fixed grid covering Scania for the time period 2000–2004. The advantage of a pollutant database, rather than an emission source database, is that it will make epidemiological studies much easier to perform for users who are not accustomed to using emission source databases and dispersion modelling. The disadvantage is that a pollutant database is not as flexible as an emission source database, mainly in that the pollutant database has a fixed resolution in time and space.
It is important to take the time factor into consideration when conducting geographical analysis in exposure studies with the aim of analysing diseases and health hazards with long latency periods, such as cancers. Samples and measurements made when a disease manifests itself do not usually represent historical exposure levels . It is therefore important to collect data in order to be able to model accumulated exposure levels from previous years. This need to store data raises the issues of storing capacity and modelling resources. Since digital data for these studies are often required as high-resolution data with a low level of spatial aggregation, the storage capacity needed and the modelling time required increase rapidly. This problem is apparent in exposure studies that take the movement of a dynamic population into account; not only do these studies demand digital data with high spatial resolution, but there is also a need for high temporal resolution.
The aim of this study was to examine the optimal spatial resolution, with respect to the temporal resolution, for our pollutant database. This was done through comparing the absolute differences and standard deviations of the differences among pollution estimates of different spatial and temporal resolutions. The choice of optimal spatial resolution was also related to the accuracy of the estimated pollutant field generated by the dispersion model and the emission database.
The impact of scale and aggregation factors in epidemiological analysis can not be ignored. As Krieger states: "in epidemiological studies completeness in geocoding does not equal success. Accuracy – and choice of geographic level – matter as much, if not more" . Cromley also concludes that, in many health studies, the aggregated level of available health data is a major limitation since data are often aggregated inappropriately for the purpose of the study . A study by João shows that scale, in terms of both detail and extent, is an important issue, which to date has been largely neglected in environmental impact assessments . This is further emphasized by Setton et al. who stated that many exposure assessments and epidemiological analyses of the impacts of air pollution on health have been undertaken on regional scales, and that only recently have researchers begun to investigate neighbourhood-level variations in pollutant levels .
Finding the optimal scale for studies concerned with factors varying in space is a complex matter of great concern. If large areas are defined they reveal more global structures, while local variations are obscured . This means that only superficial assessments may be possible, and the uncertainty will increase . However, although small neighbourhood sizes yield detailed patterns and a more detailed examination may be feasible, the understanding of the broad context may be lost [10, 11]. As a result, the choice of optimal size is important when investigating ecological associations with diseases, since these patterns might otherwise be concealed or misjudged. As Gregorio et al. concludes: "spatial analysis results may be rightfully considered conditional upon the particular geography selected for the study" .
Apart from the risk of altering the results of a study by choosing the "wrong" scale or aggregation level Ali et al. also concluded that the optimal neighbourhood size is data dependent . Their study suggests that, depending on the factors and correlations included, different neighbourhood sizes may be optimal for different variables. This is supported by findings from our previous study . Therefore, it can not always be assumed that the choice of study area for one set of parameters is optimal for another, even though they may be similar. According to Gregorio et al. nationwide efforts to promote regional health information networks/organisations that cross traditional geo-political boundaries demand a greater understanding of how aggregating health and population data may affect the analysis and interpretation of disease patterns .
Due to the four-dimensional nature of the distribution of atmospheric pollutants, the importance of scale and aggregation level applies not only to spatial but also temporal variations. According to Pénard-Morand et al., who conducted a study on schoolchildren's exposure to traffic-related air pollution in France, there is extreme variability in the level of exposure that exists from one place to another . Briggs also concludes that pollution surfaces in urban areas are often extremely complex, with steep gradients away from ground-level sources, and often highly localised hot-spots and peaks . Also, Cyrys et al. state that several studies have documented important variations in pollutant concentrations within cities, especially related to motorised traffic and location within the city – for example city centre versus suburbs . These variations are not only variations in space, but also in time mostly due to human behaviour such as rush-hour traffic and industrial activities. Results from a study by Wentz et al. suggest that there is a distinct relationship between levels of CO2 and spatial patterns of human activities in urban areas . The results of their study also show that the temporal and spatial patterns of CO2 levels correspond closely to the density of traffic, population and workplaces. Consequently, the optimal choice of resolution for a pollutant database may be mainly dependent on the spatial distribution of the concentration of air pollution in the area, but also on the temporal distribution.
João shows in her study that the chosen scale should be considered as part of the observation bias since the choice of scale, both in the meaning of the details analysed and the spatial extent of study area, can have important repercussions on the results, such as the determination of impact significance and the measurement of environmental parameters . It is therefore of fundamental importance to define the scale of observation. In other words, scale choice should be explained, justified and explicitly stated in all environmental impact assessment studies .
The overall objective of this study was to determine the optimal spatial resolution for a pollutant database for the county of Scania, southern Sweden. The database will contain concentration values of nitrogen oxides (NOx), and possible also other pollutants, over a period of five years (2000–2004).
The optimal spatial resolution is dependent on which time resolution is considered in the epidemiological study. Since the pollutant database will be used mainly for epidemiological studies with durations of days, weeks or even longer periods, we have evaluated the results using different time resolutions.
The difference between modelled and measured values, i.e. the error estimate, of air pollutant concentrations has also been evaluated since this in combination with the error estimate generated from the choice of resolution will give the total error estimate for the pollutant database.
The area studied was Scania, the southernmost county of Sweden. The county covers around 11,350 km2, which is approximately 2% of the total Swedish land area. The region is relatively densely populated, with more than 1.1 million people living in the area, which constitutes approximately 11% of the total Swedish population. In Scania, approximately 67% of the population lives close to the west coast. Most road traffic (passenger cars as well as trucks) from the European continent to Sweden and Norway passes through this area, and five motorways run through the region. There are also several harbours in the region and a considerable amount of cargo shipping and ferry transport along the coast. These factors, and the closeness to Copenhagen in Denmark, and the European continent, contribute to high concentrations of air pollutants in the region, compared with most other regions in Sweden.
Lund is the third largest city in Scania with approximately 76,000 inhabitants, and the city covers approximately 23 km2. Although the city centre consists mainly of narrow streets and pedestrian precincts, the traffic can be quite intense in some areas, and there are one motorway and two major roads along the north, south and east boundaries of the city. This creates rather steep gradients in the concentrations of traffic-generated pollutants between the city, enclosed by the motorways, and the surroundings, which consist mainly of agricultural land.
The area around Genarp consists mainly of rural and agricultural land with small clusters of buildings. A major road with a large traffic volume passes through the northern part of this study area.
The validation analysis between modelled concentrations and measured concentrations was performed using pollution data obtained from the meteorological station in Malmö (Figure 1). Malmö is the largest city in Scania with a population of about 260,000. The city is located on the west coast facing Denmark and covers an area of 66 km2. Malmö is surrounded by a ring road at which three of the five main motorways in the area converge.
1) Comparison of modelled values and observed values
Measured hourly NOX values from the meteorological station in Malmö city were collected. These measured values were then compared with hourly NOX values modelled for the same period and area using a dispersion model and an emission database. Temporal variation, mean and standard deviation was evaluated.
2) Generation of pollutant fields with a time resolution of one hour
A dispersion model was used to estimate pollutant fields. The pollutant field with the highest spatial resolution (100 metre) is below denoted the fine grid. The other four grids are denoted coarse grids (200, 400, 800 and 1600 metres). Hourly NOX values for four weeks were estimated for all grids.
3) Generation of pollutant fields with a time resolution of one day and one week
Mean values of the original hourly data were used to create new pollutant fields with time resolutions of one day and one week for the time period modelled in step 2.
4) Interpolation of the coarse pollutant fields
The hourly, daily and weekly time series of the coarse grids were interpolated to obtain the same spatial resolution as the fine grid.
5) Comparison of the fine grid and the coarse grids
The values of the fine grid and the interpolated coarse grids were plotted in time series and evaluated concerning the loss of information due to using a coarser spatial resolution in relation to the time resolution (hour, day and week). The spatial distribution of the difference between the grid values was also studied.
The results obtained from Step 1 and 5 were evaluated and the spatial resolution of the pollutant database was determined.
Comparison with measured values
The mean difference between the modelled and the measured value of NOx is 0.6 μg/m3 and the standard deviation of the differences is 12.3 μg/m3 for the hourly difference, 5.9 μg/m3 for the daily difference and 4.4 μg/m3 for the weekly difference.
Despite the temporal variations seen in Figure 3 the mean difference is low, with a value of approximately 0.6 μg/m3 NOx. However, the standard deviation of the differences reveals that the divergence for the hourly values is high. This number decreases considerably when comparing the concentrations as daily or weekly means. This indicates that the general trend predicted by the dispersion model follows the actual pattern of concentrations of air pollutants in Malmö, but that the dispersion model is too coarse to reflect finer temporal variations than daily means. If the highest and lowest quartiles are removed from the dataset the standard deviation of the means is lowered considerably, and in the case of daily means the standard deviation of the difference is reduced from 5.9 μg/m3 to 2.4 μg/m3 NOx.
The main contribution to both the modelled and measured values is traffic generated. The dispersion model uses predefined patterns of traffic flow in the area to estimate the concentration and levels of emitted air pollutants. This may be the reason for the large divergence between modelled and measured values at the beginning of the month (days 1–6). During the first week in January many people are still on holiday due to the Christmas and New Year holidays, causing the traffic flow to diverge from its normal pattern. This will probably lead to lower levels of measured traffic-generated emissions than the modelled results. However, this does not explain the large divergence between modelled and measured values during days 9–11 and 23–26, which probably originates, from errors related to emission sources or the calculation of the dispersion model.
Due to limited possibilities to measure actual levels of NOX at the two study sites (Lund and Genarp) the comparison between modelled and measured NOX levels for the two study areas should be considered with care. The difference in spatial characteristics, such as emission sources, between Malmö and the two study sites might cause the evaluation of the accuracy and performance of the dispersion model to be invalid for Lund and Genarp. Nevertheless this meteorological station was the closest one to our study areas and the accuracy of the dispersion model and emission database used have been evaluated for the whole county of Scania elsewhere  with good agreement. Subsequently we consider this comparison to be a good indication of the general agreement between the validity of the emission database and the performance of the dispersion modelled compared to actual levels of NOX.
Comparison of interpolation methods
Mean () and standard deviation () of the difference between interpolated and modelled hourly NOX values for the whole urban study site, January 2003.
The interpolated surface should preferably be as close as possible to the modelled surface of the fine grid. As can be seen in Table 1 the differences and standard deviations between the two interpolation methods are quite similar. Both the mean and the standard deviation are slightly better for the polynomial interpolation method than the bilinear.
Studying the different methods visually reveals, however, that the interpolation methods differ spatially. As can be seen from the polynomial interpolation shown in Figure 4, this method gives low values less than zero. This is caused by the fact that polynomial interpolation of degree three can model local maxima/minima between the input values where the contrast for the modelled values is high (this implies that the parameter a3 in Eq. 2 is large if the input values vary much). In Figure 4, the local minima for the polynomial interpolation are seen as black dots close to areas with high concentrations (i.e. bright areas).
Although the polynomial surface seems to follow the fine surface slightly better on average than the bilinear interpolation, we chose to use the bilinear interpolation method for this study (and recommend bilinear interpolation be used in future studies). The main reason is that polynomial interpolation can produce poor-quality output if there is high contrast in the input data.
Comparison of interpolated values and modelled values
Mean () and standard deviation (, , ) of differences in concentrations of NOX (μg/m3) between the interpolated coarse resolutions and the fine grid for the whole month of January.
Urban study site: Lund
Rural study site: Genarp
Urban study site – Lund
Rural study site – Genarp
As can be seen from Table 2 and Figure 5 there is a systematic difference between the values from the fine grid and the interpolated values from the coarse grids. The larger the cell size, the lower the NOxvalue. It should be noted that this is not due to the interpolation but the dispersion modelling. This can be illustrated by studying the original modelled values. The mean values for all modelled points (for all cells and time intervals); were computed and denoted Mean100, etc. The difference between the grids was then calculated to obtain: Mean100-Mean200 = 0.14 μg/m3, Mean100-Mean400 = 0.32 μg/m3, Mean100-Mean800 = 0.52 μg/m3, and Mean100-Mean1600 = 0.52 μg/m3. These values show a clear resemblance to the values in Table 2.
Preference of resolution
It is important to take into consideration the properties of the modelled area into consideration. Concentrations modelled in urban areas with coarser resolution than 400 metres seem to generate larger error estimates in areas with high contrasts. However, for rural areas, where the variations in air pollutants are much lower, even a grid size of 1600 metres seems to generate reasonable results. However, most individuals in Scania live in urban areas where both levels and gradients of air pollutants are high. When conducting exposure studies in these areas too low a resolution might fail to reflect these variations, thus giving incorrect exposure estimates for the inhabitants. Therefore, the results from the urban study site should have a greater impact on the choice of resolution. If data are to be stored as hourly values a resolution of, at least, 400 metres is preferable as lower resolutions will yield too high deviations in the cities and in areas with high gradients. However, this may lead to a huge increase in the need for generation and storage of data. One option could therefore be to generate a grid for the entire region of Scania with a resolution of 800–1600 metres, and in addition to this generate concentration grids with higher resolution, at least 400 metres, for the larger urban areas. This could lead to calculated concentrations in areas along major roads in the rural areas having a larger error than the rest of the region. The advantages, on the other hand, would be that the data storage and computer capacity needed would decrease considerably, while exposure estimates for the vast majority of the population in the region would be accurate.
When establishing a pollutant database certain criteria must be taken into consideration. The fact that a pollutant database has a fixed resolution in time and space makes the choice of resolution for these two parameters critical for the future use of the database. The aim of this study was to investigate the optimal spatial resolution with respect to temporal resolution for future epidemiological studies in Scania, Sweden. The pollutant database of NOx values will be used mainly for epidemiological studies with durations of days, weeks or even longer periods. Some studies might model exposure for a dynamic population (e.g. modelling being at work and at home), and therefore we need an hourly time resolution in the pollutant database. However, since the duration of the studies is at least one day the spatial resolution should be set in relation to this time resolution.
The standard deviation for the accuracy of the modelled daily mean values of air pollutants is approximately 6 μg/m3; a value that could be decreased by about 50% if the extreme values are removed. Previous studies  have given similar accuracy estimations. The choice of spatial resolution should not considerably deteriorate the accuracy of the modelled values; therefore, we believe that an error due to coarse resolution greater than 1 μg/m3 is inadvisable. Based on Table 2 we conclude that for urban areas a spatial resolution of 200–400 m is suitable; and for rural areas the spatial resolution could be coarser (about 1600 m). This implies that we should develop a pollutant database that allows different spatial resolution for urban and rural areas.
The procedure used in this study is described below and follows the study design illustrated in Figure 2.
Comparison of modelled values and observed values
NOx values were modelled using a dispersion software and an emission source database for air pollutants. This database was constructed by the GIS Centre, Lund University, supported by the Swedish National Air Pollution and Health Effects Programme (SNAP). The database includes emission data for all major sources for the years 2000–2004; the main pollutants included are nitrogen oxides (NOx) and particular matter. The emission source database covers the whole Öresund region, where Scania, Zealand (Denmark) and the sea around Scania. The purpose of extending the area is to obtain more precise description of air quality in Scania, since emissions from Zealand and shipping affect the air quality in the western part of Scania. The most important sources in the database are road traffic, shipping, aviation, rail transport, industry, power plants, small-scale heating and machinery. Information, statistics and emissions were collected from official sources. The database includes data on about 23,000 road sources, 500 point sources and nearly 100 area sources.
For modelling we used the dispersion software Enviman . Enviman estimates the pollutant values over a grid in which all emission sources are located in the centre of the cell. A box model is used to compute the contribution to the cell in which the emission sources are located. The contribution to other cells is calculated using a Gaussian dispersion method . Enviman has a default time resolution of one hour. To perform dispersion calculations a meteorological dataset is needed. In this study we used observed meteorological data, such as wind speed, wind direction, temperature and solar radiation. The meteorological observations were made on a 24 m high mast, in an open field a few kilometres from central Malmö. In the calculations, we allowed the plume from each source inside the calculation area travel 6 hours. After six hours, we assumed that the concentration was so low that it had no significant effect on the air pollution levels.
Enviman dispersion software and the emission source database were used to estimate the hourly NOx values at the location of the meteorological station in Malmö during four weeks in January 2003. Hourly NOx values for the same time period were also collected from the meteorological station. The measurements were performed by the local environment protection agency in Malmö and the measuring technique was chemiluminescence. Temporal variations, means and standard deviation between the modelled and measured concentrations were then evaluated.
Generation of pollutant fields with a time resolution of one hour
Enviman was used to produce five pollutant fields of NOx for each of the two study sites, with spatial resolutions of 100 m (denoted fine grid), 200 m, 400 m, 800 m and 1.6 km (denoted coarse grids). The time resolution for all pollutant fields was one hour and the total time modelled was four weeks in January 2003. In the modelling we used meteorological data from Malmö. It should be noted that the modelling of these pollutant fields does not include background emissions; the reason being that the aim was only to estimate the effect of using different spatial resolutions and for this estimation the background emissions has no effect.
Generation of pollutant fields with time resolutions of one day and one week
From the one-hour data we computed pollutant fields of NOX values with temporal resolutions of one day and one week. This step was simply performed by taking the mean values of the one-hour data.
Interpolation of the coarse pollutant fields
The ideal interpolation method for our application should generate a continuous surface that passes through all the modelled values (i.e., the input values for the interpolation), this and the interpolated surface should be as close as possible to the modelled surface. In this study, we tested two interpolation methods: bilinear interpolation and polynomial interpolation [19–21].
z(x p , y p ) = interpolated value at the point x p , y p (i.e. the point at which the value is to be interpolated),
z1, z2, z3, z4 = values of the four closest points (to x p , y p ) in the coarse grid,
w = (x p - x 1 )/Δx,
u = (y p - y 1 )/Δy and
Δx, Δy = the spatial resolution of the coarse grid.
Bilinear interpolation has the advantage that it is simple and creates a continuous surface that passes through all the input points (points 1–4 in Eq.1). But it is not capable of modelling extreme values between the input points.
Since only four values (along the line y = y1) are used to estimate the polynomial parameters (a i ) there is a unique solution (in reality the polynomial is written somewhat differently, (see ), but here we prefer this simple form for clarity). Equation 2 is then applied for the other three y-values; this implies that we have four estimations of z for x = x p . Finally, these values are used to interpolate the point (x p , y p ) using a polynomial in the y-direction (Figure 12).
Polynomial interpolation provides a continuous surface as long as the same input points are used. However along the boundary lines where the input points are changed the surface is discontinuous. Polynomial interpolation (of degree three) is capable of modelling extreme values between the input values. This can be both an advantage and a disadvantage. For some types of surface characteristics the extreme values will decrease the difference between a modelled surface and an interpolated surface; for other types the difference will increase.
Comparison of the fine grid and the coarse grids
Interpolation provides pollutant fields with values for the same points. For the fine grid the values are modelled and for the coarse grid the values are interpolated. Furthermore, we have pollutant fields with time resolutions of one hour, one day and one week.
i, j denotes that the cell in row i and column j in the grid,
T hour is the number of time intervals for a spatial resolution of one hour,
t denotes the time interval,
are the NOX values for time interval t and cell i, j in the fine grid, and
are the NOX values for time interval t and cell i,j in the interpolated coarse grid.
The mean value will be the same regardless of whether we study hourly, daily or weekly NOX values (since the daily and weekly data were computed as mean values from the hourly data). Behind this lies an approximation that the mean values of the discrepancies are constant throughout the day and between different days. This approximation is fairly good.
x denotes the time resolution (hour, day or week), and
where N, M are the number of rows/columns.
A program in C++ was developed to compute the daily/weekly values, perform the interpolation and compute the statistics. Input data for the program were binary files containing modelled NOX values from the dispersion software Enviman. For the implementation of the polynomial interpolation the program utilizes functions from Numerical Recipes in C . The time series were generated by a standard spreadsheet program (MS Excel) and the maps by a standard GIS program (ArcGIS from ESRI).
We would like to acknowledge EMFO (Emission research programme conducted by the Swedish Road Administration) for financial support of this project, Anna Axmon for statistical advice, Kristina Jakobsson and Petter Pilesjö for general support and Håkan Törnevik for consultation regarding the dispersion model Enviman. We would also like to acknowledge our anonymous reviewers for their helpful comments.
- WHO: Health aspects of air pollution with particulate matter, ozone and nitrogen dioxide. Report on a World Health Organisation(WHO) working group. Bonn 2003. 2003Google Scholar
- Chaix B, Gustafsson S, Jerrett M, Kristensson H, Lithman T, Boalt Å, Merlo J: Children's exposure to nitrogen dioxide in Sweden: investigating environmental injustice in an egalitarian country. Journal of Epidemiology and Community Health. 2005, 60: 234-241. 10.1136/jech.2005.038190.View ArticleGoogle Scholar
- Jerret M, Arain A, Kanaroglou P, Beckerman B, Potoglou D, Sahsuvaroglu T, Morrison J, Giovis C: A review and evaluation of intraurban air pollution exposure models. Journal of Exposure and Environmental Epidemiology. 2005, 15: 185-204. 10.1038/sj.jea.7500388.View ArticleGoogle Scholar
- Briggs D: The role of GIS: Coping with space (and time) in air pollution exposure assessment. Journal of Toxicology and Environmental Health, Part A. 2005, 68: 1243-1261. 10.1080/15287390590936094.View ArticleGoogle Scholar
- Gustafsson S: A geographical and temporal high resolution emission database for dispersion modeling of environmental NOX in southern Sweden. Lic Thesis. 2007, Lund University, Department of Physical Geographya and Ecosystemanalysis, 1: 1-14.Google Scholar
- Cromely EK: GIS and Disease. Annual Review of Public Health. 2003, 24: 7-24. 10.1146/annurev.publhealth.24.012902.141019.View ArticleGoogle Scholar
- Krieger N: Place, Space, and Health: GIS and Epidemiology. Epidemiology. 2003, 14 (4): 384-385.PubMedGoogle Scholar
- João E: How scale affects environmental impact assessment. Environmental Impact Assessment Review. 2002, 22: 289-310. 10.1016/S0195-9255(02)00016-1.View ArticleGoogle Scholar
- Setton EM, Hystad PW, Keller CK: Opportunities for using spatial property assessment data in air pollution assessments. International Journal of Health Geographics. 2005, 4: 26-10.1186/1476-072X-4-26.PubMedPubMed CentralView ArticleGoogle Scholar
- Ali M, Park J-K, Thiem VD, Canh DG, Emch M, Clemens JD: Neighborhood size and local geographic variation of health and social determinants. International Journal of Health Geographics. 2005, 4: 12-10.1186/1476-072X-4-12.PubMedPubMed CentralView ArticleGoogle Scholar
- CEAA: A reference guide for the Canadian Environmental Assessment Act: assessing environmental effects on physical and cultural heritage resources. 1996, Hull, Quebec: Canadian Environmental Assessment AgencyGoogle Scholar
- Gregorio DI, Samociuk H, DeChello L, Swede H: Effects of study area size on geographical characterization of health events: Prostate cancer incidence in Southern New England, USA, 1994–1998. International Journal of Health Geographics. 2006, 5: 8-10.1186/1476-072X-5-8.PubMedPubMed CentralView ArticleGoogle Scholar
- Stroh E, Oudin A, Gustafsson S, Pilesjö P, Harrie L, Strömberg U, Jakobsson K: Are associations between socio-economic characteristics and exposure to air pollution a question of study area size? An example from Scania, Sweden. International Journal of Health Geographics. 2005, 4: 30-10.1186/1476-072X-4-30.PubMedPubMed CentralView ArticleGoogle Scholar
- Pénard-Morand C, Schillinger C, Armengaud A, Debotte G, Chrétien E, Pellier S, Annesi-Maesano I, ISSAC-France: Assessment of schoolchildren's exposure to traffic related air pollution in the French Sic Cities Study using a dispersion model. Atmospheric Environment. 2006, 40: 2274-2287. 10.1016/j.atmosenv.2005.11.057.View ArticleGoogle Scholar
- Cyrys J, Hochadel M, Gehring U, Hoek G, Diegmann V, Brunekreef B, Heinrich J: GIS-based Estimation of Exposure to Particulate Matter and NO2 in Urban Area: Stochastic versus Dispersion modelling. Environmental Health Perspectives. 2005, 113 (8): 987-992.PubMedPubMed CentralView ArticleGoogle Scholar
- Wentz EA, Gober P, Balling RC, Day TA: Spatial Patterns and Determinants of Winter Atmospheric Carbon Dioxide Concentrations in an Urban Environment. Annals of the Association of American Geographers. 2002, 92 (1): 15-28. 10.1111/1467-8306.00277.View ArticleGoogle Scholar
- Opsis: 2006, 2006-05-26, [http://www.opsis.se/]
- USEPA: User's guide for the AMS/EPA regulatory model – Aermod. 1998, (2006-05-15), [http://www.epa.gov/scram001/7thconf/aermod/aermodug.pdf]Google Scholar
- Zhizhou W: Principles of Photogrammetry. 1990, Beijing, Press of Wuhan Technical University of Surveying and MappingGoogle Scholar
- Burrough P, McDonnel R: Principles of Geographical Information Systems. 1998, Oxford, Oxford University PressGoogle Scholar
- Press WH, Teukolsky SA, Vetterling WV, Flannery BP: Numerical Recepies in C. 1992, Cambridge, Cambridge University PressGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.