Exploring the Benefits of 3D City Models in the Field of Urban Particles Distribution Modelling—A Comparison of Model Results

  • Yahya GhassounEmail author
  • Marc-O. Löwner
  • Stephan Weber
Part of the Lecture Notes in Geoinformation and Cartography book series (LNGC)


We present a comparison of a particles distribution model using 3D parameters derived from a CityGML-based 3D city model with an already advanced but 2D-based Land Use Regression model. Particles, especially ultrafine particles have significant influence on the health status of the urban population. Next to emission by cars and others, its distribution is tightly coupled to the local wind field and, therefore, to urban morphology influencing this wind field. However, 3D city models, especially CityGML have been almost ignored when modelling urban particles distribution. We introduce 3D parameters derived from a CityGML-based 3D city model in an already tested Land Use Regression model and explore the benefits of 3D city models in the field of particles distribution modelling, especially, by minimizing the number of parameters entered to the model and the good results that it has shown and explore the enhancement by combining both models.


Land use regression CityGML Ultrafine particle 3D city model Geostatistical model 

1 Introduction and Problem Statement

Air quality in urban areas has a major influence on health of urban population. Health effects of ultrafine particles have been widely discussed and demonstrated in epidemiological and toxicological studies (WHO Panel 2013). The pollution effects of ultrafine particles (UFP) on the health in urban area have increased mortality and morbidity especially from cardiovascular and respiratory diseases (Brook et al. 2010). Ultrafine particles with diameter less than 0.1 microns often are the direct product of the combustion of fossil fuels by road transport (Geiser et al. 2005). In addition, Hoffmann et al. (2006) showed that long term exposure to fine particulate matter (PM2.5) may accelerate the development and progression of atherosclerosis. The UFP concentration can be estimated from measurements of the particle number size distribution. At street canyon or near-traffic sites, the number of UFP generally accounts for the majority of total particle number concentrations, i.e., greater than 80–90 % (Morawska et al. 2008; Weber et al. 2013).

Since high traffic intensity is considered as an essential source for ultrafine particles in many cities, the proximity of the residence to roads was identified as a major determinant of health effects by particulate air pollution (Jerrett 2011). Thus, accurate assessment of exposure to traffic-related air pollution is critical for both, epidemiologic studies as well as for the estimation of health impacts of current and future road networks (Gilbert et al. 2005).

Urban complexity, i.e., the road network with different traffic intensity on road sections in combination with the three-dimensional structure of buildings, which influence climate and dispersion, leads to a specific spatial distribution pattern of urban UFP concentrations. Therefore, detailed assessment of exposure by measurement and modelling of fine dust distribution is necessary in the field of urban planning, traffic management and city system modelling.

Urban morphology can be analysed concerning geometrical properties of street canyons (Vardoulakis et al. 2002), building density, alignment of streets towards the prevailing wind direction, and characteristics of crossing sections. However, the latter is only little noticed when considering the redistribution of air and exchange of fine particles between different street systems (Brand and Löwner 2014).

Land Use Regression (LUR) models were developed as an alternative to dispersion models. As a geostatistical method, they can be applied for predicting local variation in traffic pollution and to obtain accurate, small scale air pollutant concentrations without a detailed pollutant emission inventory (Briggs et al. 2000; Brauer et al. 2003). LUR predicts pollution concentrations at a given site based on surrounding spatial characteristics and other identified auxiliary data which may act as a pollution predictor (Jerrett 2011).

LUR models are multiple linear regressions that assume independent residuals and use GIS-based covariates to predict concentrations at locations without measurements (Mercer et al. 2011). They have been widely applied in cities of North America and Europe (Saraswat et al. 2013). Henderson et al. (2007) developed LUR models for UFP in the city of Amsterdam including traffic intensity of nearest road, inverse of distance squared, population density and land use. Arian et al. (2007) included wind flow effects in LUR models for predicting NO2 concentrations for health exposure studies in the greater Toronto-Hamilton region.

In comparison with meteorological data the main advantage of LUR is the incorporation of extensive data that can be measured with small errors (Jerrett et al. 2006). However, it neglects the process-oriented view of fine dust production, dilution or dispersion and accumulation or deposition. Only statistical measures are used to develop the prediction model without asking for causal linkage.

Recently, only few studies included 3D data of buildings and street canyons into LUR modelling. Tang et al. (2013) has used building height and geometry to enhance the representation of land use and the dispersion field in LUR. Eeftens et al. (2013) described an approach to calculate four indicators (the maximum aspect ratio, the mean buildings angle, median building angle, which is the angle between the horizontal street level and the line of sight to the top of surrounding buildings, and SkyView Factor) of urban canyon effect using 3D building data and evaluated the improvement on NO2 and NOx LUR models in the Netherlands. Brauer et al. (2003) added 3D parameters (sampling height, street type, and canyon) as additional variables into the regression models to determine the extent to which prediction could be improved.

Today, 3D city models are increasingly available due to the standardization processes like CityGML (Gröger et al. 2012) for semantically enriched virtual 3D city models. An advantage of CityGML is its scalability to the requirements of the user and the data available (Löwner et al. 2013b). Today, they already support a lots of application fields like urban planning and hazard assessment (Löwner et al. 2013a). If only geometrically models are needed, KML models may be sufficient.

Here, we present a comparison of a UFP model using 3D parameters derived from a CityGML-based 3D city model with an already advanced but 2D-based Land Use Regression model. A land use regression based geostatistical model (LUR) is developed to estimate UFP concentrations at given locations using 2D and 3D parameters. Conceptually, both models were built to depict particle concentrations in urban microenvironments as the result of a ‘process chain’ that can be subdivided into three parts: emission, dilution/dispersion and deposition. All these processes take place under certain boundary conditions which are variable due to anthropogenic behaviour, weather conditions and the surrounding built environment. Therefore, the models are implemented with respective input parameters that are related to the three processes of production, dilution and accumulation. Incorporating 3D attributes to further develop the aforementioned LUR model for the prediction of UFP distribution within a city quarter leads to a more precise UFP estimation.

1.1 Study Area and Measurements

The local scale variation of UFP is studied in the city of Braunschweig, a medium-sized city which is located in the south-eastern part of the federal state of Lower-Saxony, Germany. The 1.0 km2 study area is characterized by street canyons with low, medium and high traffic intensities (up to 30,000 vehicles per day), residential areas (<2,000 vehicles per day), backyards and urban park areas (Fig. 1). The area was divided into 35 grid cells each 200 × 200 m whose midpoints represented the sampling point at which mobile measurements of particulate air pollutants were conducted (Ruths et al. 2014). The sampling sites were established in six different urban settings, called ‘microenvironments’. Microenvironments are defined as micro scale areas with quasi-homogeneous concentrations which can be used to derive characteristic average concentrations (e.g. Ott et al. 2007). At each site, the average concentration of total particle number concentration, the number size distribution and black carbon were measured over 3 min (cf. Sect. 2.2). Measurements were carried out during 8 campaigns in winter (January–March 2013) and 7 campaigns in summer (June–August 2013) during calm weather conditions without rain and low wind speeds <4 ms−1.
Fig. 1

Study area in Braunschweig, Germany. It is characterized by street canyons with high traffic intensities, residential areas, backyards and urban park areas (from Ruths et al. 2014)

Monitoring was conducted with a mobile platform consisting of a bicycle connected to a 2-wheel trailer which contained the instruments in a weather protected box. Particle number size distributions were measured with a mobile SMPS the NanoScan scanning mobility particle sizer (TSI 3910, TSI Inc.) which offers a size distribution in the range 11 < Dp <365 nm with a 1 min time resolution. The size fractioned ultrafine number concentration was calculated from the number size distribution as measured by the NanoScan device. Total number concentrations were measured with an Electrical Diffusion Size Classifier (DiSCmini, MatterAerosol). Black carbon concentrations were measured with an aethalometer AE51 (MageeScientific). To ensure a high accuracy and data quality, all instruments have been synchronized before each measurement campaign.

Furthermore, each campaign was limited to 2 h to reduce a trend of particle concentration due to weather changes over the diurnal course (cf. Ruths et al. 2014 for details).

2 Modelling Fine Dust Distribution with Land Use Regression (LUR) Models

2.1 Land Use Regression Using 2D Parameters

Assuming a process related impact on pollutant concentrations, different spatial parameters were grouped into the three process categories to ensure that every process, i.e. emission, dilution and deposition is represented with at least one significant parameter to define the best model (Table 1). These are:
Table 1

List of used spatial parameters and their source next to their extraction method in ArcGIS 10.0



Buffer radii (m)



Length of different types of streets

(residential, primary, tertiary, secondary


50,100 and 200

Using model builder: generate buffer, intersect with the street, and summarize according the type of the street


Width of the streets area of the buildings

−100, 200 and 300

Using model builder: generate buffer, intersect with the buildings, and summarize according the type of the building

Position of the sites (X, Y)


Recreational are (Residential_use500

Green_village500, Commercial500, Construction50, Meadow500, Cemetery500)


Using model builder: generate buffer, intersect with the buildings, and summarize according the type of the building

  1. 1.

    Emission category: Spatial parameters belonging to this group can be considered to be characteristic for the emission rate of UFP, i.e. the length and type of a street, reflecting traffic intensity. This category is not significantly affected by seasonal changes;

  2. 2.

    Dilution category: Spatial parameters that define proxies for urban ventilation capability such as the width of streets or the distribution of built up areas; and

  3. 3.

    Deposition category: Spatial parameters that define potential deposition surfaces such as different types of recreational areas with higher fractions of vegetated surfaces.


Zones of influence, i.e. buffer radii, were specified to reflect the scale of influence appropriate for each process related variable. However, zones representing the influence of a street should be chosen narrow because traffic produced emissions are quite local (buffer radii should be small). In contrast, effects of land use stretch out over a larger area and are more complex as dilution and emission. Thus, larger zones of influence are defined.

Buffers of radii of 50, 100, 200, 300 and 500 m (cf. Table 1) were generated for each measured site using ArcGIS 10.0. Using these buffer radii, all available data (streets, buildings and recreational area) were subsequently clipped to derive the street length, buildings area, and recreational area. Buffer radii of 50, 100 and 200 m were used to derive the length of streets, whereas buffers radii of 100, 200 and 300 m were used to derive the building area. Buffer radii of 500 m were used to derive the area of recreational area.

The algorithm as proposed by Henderson et al. (2007) was used and adopted to build the LUR model for UFP according to the following procedure: all variables were ranked by the absolute strength of their correlation with the measured pollutant. The highest-ranking variable in each sub-category was identified. Afterwards the variables in each sub-category that are correlated (i.e., Pearson’s r ≥ 0.6) with the most highly ranked variables were eliminated to avoid autocorrelation. All remaining variables were implemented into robust linear regression models and variables that were not significant at a 90 % confidence level or that had a coefficient with a counterintuitive sign were rejected. Finally, the last two steps were repeated until convergence.

By following the aforementioned procedures, the most significant variables were selected and implemented in the multi regression model using the open source software R in order to build two models for UFP reflecting the summer and winter campaign, respectively.

2.2 Land Use Regression Using 3D Parameters

Here, we enhance the aforementioned LUR model for the prediction of fine dust distribution within a city quarter introducing 3D attributes. Starting from an airborne laser scan derived CityGML-based data base, first, height of the buildings adjacent to the street are selected. Second, the ratio of height and width of the street canyon and, third, volumetric density were used amongst others. However, height was already used in the reference model. The second parameter was described in Tang et al. (2013) and Eeftens et al. (2013). The newly introduced parameter of volumetric density describes the ratio of built volume and air and, therefore, refers to the dilution process. Air volume is calculated as a cylinder with the height of the highest building lying within a buffer around the point to be modelled (Fig. 2). Buffers of radii of 100 and 200 m were generated for each measured site using ArcGIS 10.0. Using these buffer radii, all the available 3D data were subsequently clipped to derive the buildings volume and to calculate the volumetric density. Then, same aforementioned LUR procedures were used to select the most significant variables and implement them in the multi regression model in order to build two 3D models for UFP, too.
Fig. 2

Calculation of volumetric density as ratio of built environment and air

3 Results

3.1 Evaluation of 2D-LUR and 3D-LUR Models

Significant 2D variables of each process category (cf. Table 1) were identified and subjected to the multi linear regression process. Then, the best model for the summer and the winter campaign was built.

3D-LUR models used two 3D parameters in each model of summer and winter campaign (volumetric density and width in summer model and volumetric density and the ratio of height and width in winter).

2D-LUR models explained 74 and 85 % of the variance of UFP with root mean square errors (RMSE) of 668 and 1,639 pt. cm−3 in summer and winter, respectively.

3D-LUR models explained 41 and 54 % of the variance of UFP with root mean square errors (RMSE) of 868 and 2,541 pt. cm−3 in summer and winter, respectively.

A standardized deviation was defined by calculating the ratio of the RMSE to the average measured UFP concentrations across the study area during the summer and winter campaign. The deviation for 2D-LUR accounts to 8 % for summer and 13 % for winter concentrations, 11 % for summer and 19 % for winter concentrations for 3D-LUR, respectively.

Figure 3 depicts that LUR models give evidence for a strong linear trend; no outliers exceeding the confidence interval during summer and winter are observable for both 2D-LUR and 3D-LUR.
Fig. 3

Plot of the residuals between the UFP and predicted UFP in a the summer and b the winter model

The standard residuals of calculated UFP concentration do not exhibit any trend for the summer and winter model (data not shown here).

Leave-one-out cross validation was applied to validate the summer and winter model estimates. The model for each season was rerun on N − 1 data points (N = 27) and then used to predict the excluded measurement. Cross-validation standard error of estimate was calculated for both models. The deviation for 2D-LUR accounts to 9 % for summer and 14 % for winter and is very close to the original model results stating satisfactory model performance.

The deviation for 3D-LUR accounts to 13 % for summer and 22 % for winter. It is evident in Fig. 4 that the different street types are important parameter in the 2D-LUR models. In both seasons the primary road has the highest influence on modelled UFP with relative contribution of 25.86 % during summer and 37.64 % during winter, respectively. This is due to the fact that road traffic acts as a major source for particles in size ranges <100 nm (Morawska et al. 2008), where in 3D-LUR models, volumetric density has the most important influence on the particulate concentration.
Fig. 4

Corresponding standardized regression coefficients for a the summer and b the winter model show the relative influence explanatory variables on the dependent variable, and their significance

Comparison between both models in each microenvironment was conducted (Fig. 5). Generally, both models represented the UFP concentrations very well in all microenvironments.
Fig. 5

The mean variation of UFP concentrations during winter and summer campaign for each microenvironment [measured (UFP), calculated by 2D-LUR and 3DLUR]

2D-LUR showed better results than 3D-LUR in BY, RO, and PA for the winter campaign and in BY, RO, and RE for the summer campaign. Both models showed close results in SL and RE for the winter campaign and in SL and SM for the summer campaign.

In spite of the low variance of UFP explained by the 3D-LUR models, the models showed good results comparison with 2D-LUR especially in summer with deviation of 27 % of RMSE between both models. Rather, 3D-LUR showed higher deviation about 38 % of RMSE in winter between both models.

3.2 Model Enhancement

The impact of 3D parameters on the model used 2D parameters has been examined by entering the significant parameters of 3D-LUR into the 2D-LUR one by one. It has been noticed that height of the building and the ratio of height and width of the street canyon do not enhance the 2D model, while the volumetric density parameter has slightly increased R2 of the model by 1 %. Furthermore, the average value of the residuals has decreased by 1 %. The new LUR models explained 75 and 86 % of the variance of UFP (Fig. 6) and, therefore, the variance of UFP slightly better than 2D-LUR model with deviation of 2 and 3 % of RMSE between both, summer and winter models. By entering both significant 3D parameters for the winter campaign (ratio of height and width, and the volumetric density) into the 2D-LUR, they have increased R2 by 1 % too. The same result was achieved by entering the two significant 3D parameters for the summer campaign (width of the streets and the volumetric density).
Fig. 6

Plot of the residuals between the UFP and predicted UFP in a the summer and b the winter model

The results indicate that 3D properties have significant impact on the spatial distribution pattern of urban UFP concentrations. We argue that more 3D parameters derived from a CityGML-based 3D city model should enhance the model represented UFP concentration and minimize the required number of 2D parameters needed to predict pollution concentrations quite well.

4 Conclusion

A new modelling approach of using 3D city models available in CityGML has been discussed in this paper to model the UFP distribution that incorporates processes of emission, dilution and deposition of UFP. First results indicate that the 3D parameter enriched model performs quite well in comparison to the 2D model. Although, exhibiting slightly greater deviations, the 3D model needs less parameters than the 2D model. Since ultrafine particle concentration is the result of a complex physical process chain, describing it using simple models with less parameters has many advantages. Fast calculation is a precondition for fast real time calculation and forecast systems.

Entering the significant 3D parameters into the 2D-LUR models has shown a slight enhancement in the final models. In our point of view, this account for the thesis that the local wind field and, therefore, the 3D built environment has a significant influence on the spatial distribution pattern of urban UFP concentrations. However, the research area and the measurement layout described in Sect. “ Introduction of Chap.  Improving the Consistency of Multi-LOD CityGML Datasets by Removing Redundancy” were initially designed to develop 2D prediction models. That means that it has great variability of properties needed to test significant 2D parameters. Further studies will designate more suitable study areas concerning 3D properties discussed here and, therefore, will yield better results when defining correlations. The research area should have variety concerning 3D properties such as building porosity as an indicator of urban ventilation capacity.

We have shown that combining parameters describing the process of emission and deposition with 3D parameters describing the dilution process and with meteorological parameters can enhance the model. However, further research will focus not only on 3D geometric attributes but on semantic attributes represented in CityGML. This especially accounts for the possibility to incorporate surface properties and window surface proportions of buildings into the model. Since these properties act like a local and temporal sink of small particles, they are of great influence of the fine particle distribution but cannot be stored in a 2D data base.

Semantic attributes available in CityGML may even support the estimation of health situation because of information about the building’s usage and the localisation of its entrance that is more accurate than its address. However, addresses are needed (in addition to pure 3D geometry) because of linking modelling results to epidemiological studies by calculating the distances between residences and major roads and assigning annual fine particulate matter concentrations, derived from the models to each address. Semantics, therefor, may contribute to urban planning.

However, CityGML-based 3D city models still act as incomplete input models for the analysis of particle concentrations. Since no dynamic attributes can be represented in the CityGML2.0 standard, traffic intensities have to be added from other sources. This accounts for the representation of results, as well. Except modelling a specific Application Domain Extension (ADE), there is no way to store different raster based value data back to a CityGML model. Even more seriously, when modelling in different heights using voxel representation model output does not fit into the concept of boundary representation (BRep). The ongoing discussion within the Open Geospatial Consortium clearly should issue the weaving of field based and BRep representation amongst dynamic representation of complex attributes.



The authors would like to thank Matthias Ruths who conducted the mobile measurements of particle and pollutant concentrations.


  1. Arain AM, Blair R, Finkelstein N, Brook RJ, Sahsuvaroglu T, Beckerman B, Zhang L, Jerrett M (2007) The use of the wind fields in a land use regression model to predict air pollution concentrations for health exposure studies. Atmos Environ 41:3453–3464CrossRefGoogle Scholar
  2. Brand L, Löwner M-O (2014) Parametrisierung und Identifikation urbaner Straßenkreuzungen im Kontext der Feinstaubmodellierung. "Parameterization and identification of street crossings in the context of fine dust modelling". Gemeinsame Jahrestagung 2014 der DGfK, der DGPF, der GfGI und des GiN (DGPF Tagungsband 23, 2014)Google Scholar
  3. Brauer M, Hoek G, van Vliet P, Meliefste K, Fischer P, Gehring U, Heinrich J, Cyrys J, Bellander T, Lewne M, Brunekreef B (2003) Estimating long-term average particulate air pollution concentrations: application of traffic indicators and geographic information systems. Epidemiology 14:228–239Google Scholar
  4. Briggs DJ, de Hough C, Gulliver J, Wills J, Elliott P, Kingham S, Smallbone K (2000) A regression-based method for mapping traffic-related air pollution: application and testing in four contrasting urban environments. Sci Total Environ 253:151–167CrossRefGoogle Scholar
  5. Brook RD, Rajagopalan S, Pope CA 3rd, Brook JR, Bhatnagar A, Diez-Roux AV, Holguin F, Hong Y, Luepker RV, Mittleman MA, Peters A, Siscovick D, Smith SC, Whitsel L, Kaufman JD (2010) American heart association council on epidemiology and prevention, council on the kidney in cardiovascular disease, and council on nutrition, physical activity and metabolism. particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American heart association. Circulation. 121:2331–2378Google Scholar
  6. Eeftens M, Beekhuien J, Beelen R, Wang M, Vermeulen R, Brunekreef B, Huss A, Hoek G (2013) Quantifying urban street configuration for improvements in air pollution models. Atmos Environ 72:1–9Google Scholar
  7. Geiser M, Rothen-Rutishauser B, Kapp N, Schürch S, Kreyling W, Schulz H, Semmler M, Im Hof V, Heyder J, Gehr P (2005) Ultrafine particles cross cellular membranes by non-phagocytic mechanisms in lungs and in cultured cells. Environ Health Perspect 113:1555–1560CrossRefGoogle Scholar
  8. Gilbert LN, Goldberg SM, Beckerman B, Brook RJ, Jerrett M (2005) Assessing spatial variability of ambient nitrogen dioxide in Montreal, Canada, with a land-use regression model. Air waste Manag Assoc 55:1059–1063CrossRefGoogle Scholar
  9. Gröger G, Kolbe TH, Nagel C, Häfele K-H (2012) OGC city geography markup language (CityGML) encoding standard, version 2.0, OGC doc no. 12-019. Open Geospatial ConsortiumGoogle Scholar
  10. Henderson S, Beckerman B, Jerrett M, Brauer M (2007) Application of land use regression to estimate long-term concentrations of traffic-related nitrogen oxides and fine particulate matter. Environ Sci Technol. 41(7):2422–8Google Scholar
  11. Hoffmann B, Moebus S, Möhlenkamp S, Stang A, Lehmann N, Dragano N, Schmermund A, Memmesheimer M, Mann K, Erbel R, Jöckel K-H (2006) Residential exposure to traffic is associated with coronary atherosclerosis. Am Heart Assoc Circ 116:489–496Google Scholar
  12. Jerrett M (2011) Spatiotemporal analysis of air pollution and mortality in California based on the American cancer society cohort. Final reportGoogle Scholar
  13. Jerrett M, Beckerman B, Brook J, Finkelstein M, Gilbert N (2006) A land use regression model for predicting ambient concentrations of nitrogen dioxide in Hamilton, Festbrennstoffe, Biokraftstoffe, BiogasGoogle Scholar
  14. Löwner M-O, Casper E, Becker T, Benner J, Gröger G, Gruber U, Häfele K-H, Kaden R, Schlüter S (2013a) CityGML 2.0—ein internationaler Standard für 3D-Stadtmodelle, Teil 2: CityGML in der Praxis. Zeitschrift für Geodäsie, Geoinformation und Landmanagement, 2, 2013, pp 131–143Google Scholar
  15. Löwner M-O, Benner J, Gröger G, Häfele K-H (2013b) New concepts for structuring 3d city models—an extended level of detail concept for CityGML buildings. In: Murgante B, Misra S, Carlini M, Torre CM, Nguyen H-Q, Taniar D, Apduhan BO, Gervasi O, (eds) ICCSA 2013, Part III, LNCS 7973, Springer, Heidelberg, pp 466–480Google Scholar
  16. Mercer LD, Szpiro AA, Sheppard L, Lindström J, Adar S, Allen R, Avol E, Oron A, Larson T, Liu LJ, Kaufman JD (2011) Comparing universal kriging and land-use regression for predicting concentrations of gaseous oxides of nitrogen (NOx) for the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air). Atmos Environ pp 4412–4420Google Scholar
  17. Morawska L, Ristovski Z, Jayaratne ER, Keogh DU, Ling X (2008) Ambient nano and ultrafine particles from motor vehicle emissions: characteristics, ambient processing and implications on human exposure. Atmos Environ 42:8113–8138CrossRefGoogle Scholar
  18. Ott WR, Steinemann AC, Wallace LA (eds) (2007) Exposure analysis. CRC Press, Taylor & Francis Group, NW, 553 ppGoogle Scholar
  19. Ruths M, von Bismarck-Osten C, Weber S (2014) Measuring and modelling the local-scale spatio-temporal variation of urban particle number size distributions and black carbon. Atmos EnvironGoogle Scholar
  20. Saraswat A, Apte SJ, Kandlikar M, Brauer M, Henderson BS, Marshall DJ (2013) Spatiotemporal land use regression models of fine, ultrafine, and black carbon particulate matter in New Delhi. Environ Sci Technol 42:12903–12911Google Scholar
  21. Tang R, Blangiardo M, Gulliver J (2013) Using building heights and street configuration to enhance intraurban PM10, NOx, and NO2 Land use regression models. Environ Sci Technol 47:11643–11650Google Scholar
  22. Vardoulakis S, Gonzalez-Flesca N, Fisher BEA (2002) Assessment of traffic-related air pollution in two street canyons in Paris: implications for exposure studies. Atmos Environ 36:1025–1039CrossRefGoogle Scholar
  23. Weber S, Kordowski K, Kuttler W (2013) Variability of particle number concentration and particle size dynamics in an urban street canyon under different meteorological conditions. Sci Total Environ 449:102–114CrossRefGoogle Scholar
  24. WHO Panel (2013) Review of evidence on health aspects of air pollution—REVIHAAP Project. WHO Regional Office for Europe, Copenhagen, Denmark. Accessed on 30, March 2014

Copyright information

© Springer International Publishing Switzerland 2015

Authors and Affiliations

  • Yahya Ghassoun
    • 1
    Email author
  • Marc-O. Löwner
    • 1
  • Stephan Weber
    • 2
  1. 1.Institute for Geodesy and PhotogrammetryTechnische Universität BranschweigBrunswickGermany
  2. 2.Institute of GeoecologyTechnische Universität BranschweigBrunswickGermany

Personalised recommendations