Final Report on EPSRC GR/J50613

Effects of street grid configuration on pedestrian exposure to vehicular pollution: civilising urban traffic

Alan Penn, David Banister, Pat O'Sullivan

Researcher: Dr. Ben Croxford

Period: 24 Months

Start Date: 4/1/94: End Date: 3/1/96

1. Background and original objectives

This project builds on research from recent projects SERC GR/G23609 & H48422 which found that grid configuration explains over 80% of variance in both pedestrian and vehicular flow rates in urban areas1,2 [see Box 1 in Appendix A]. It is known that the primary source of many urban air pollutants of health concern is vehicular traffic3, and also that dispersion of these pollutants from the car exhaust by wind in urban areas depends on the configuration of surrounding buildings4,5,6. Since the pedestrian consumer, the vehicular producer and wind dispersion are all related to the spatial configuration of the built complex, we hypothesised that spatial variations in pollutant concentrations at head height might also be found that would lead to differential exposure of the pedestrian population as they moved through the city. If we could understand how the configuration of the grid gave rise to the distribution of pollutant concentrations it was hoped that ultimately one could begin to use the design of urban space and the management of traffic flows through the network to reduce pedestrian exposure. This was seen to be a pressing concern in the light of public worries about both amenity and health in city centres7.

The problem is that there is a shortage of kerbside pollution data at the micro-scale in urban areas8. Since accurate monitoring equipment is expensive, most effort has been concentrated on gathering background pollution data, and where more than one kerbside site is available these have been dispersed rather than clustered. Other techniques such as the national NO2 survey have also tended to distribute sites as widely as possible in order to gain a good picture of overall distributions for geographic areas rather than to look in detail at differences from street to street within a neighbourhood. Where detailed kerbside data have been gathered (eg. Hickman et al9,10) often only a few sites have been monitored at a time with high temporal resolution in order to investigate variations of pollution concentration with traffic flow and speed. The assumption behind most studies, models and policy formulations has been either that pollution concentrations vary smoothly or that variations on the micro scale would be so masked by 'noise' that they were not worth collecting. A current study by Margaret Bell's group at NUTRG backs up the last contention11. By developing a pollution monitor for use in association with traffic light control systems in the Instrumented City at Leicester they are investigating whether traffic light timing and sequencing can be used to minimise ambient pollution concentrations. However their equipment is constrained by power supply and data communications to be located at traffic light control boxes, and so necessarily to measure main route locations at junctions which are spatially complex, where traffic flows are high and where queuing, stopping and starting may have significant effects on the data. Little in the way of consistent spatial variation has been reported from this work, and the relationships reported between dynamic traffic flows and monitored kerbside pollution concentrations suggest that the data are subject to a great deal of noise. It seems likely that new methods of handling these data need to be developed if spatial variations are to be addressed.

The study we proposed therefore aimed to gather fine scale ambient carbon monoxide (CO) data concentrated within a small urban neighbourhood so that a number of scales of route and different configurations of space could be monitored at the same time in order to see whether or not we could identify regular variations in pollution concentration from location to location. If we could identify regular spatial variations then we proposed to move on to investigate whether configurational modelling techniques could be used to predict those variations, and whether we could identify ways of estimating pollution exposure of the pedestrian population. Finally we wanted to develop the modelling and monitoring protocols needed to evaluate traffic management and calming measures at both design and implementation stages for their effects on pedestrian exposure to pollutants. The following detailed objectives were therfore proposed:

2. Progress of the research, management and use of resources

It was originally envisaged that the 2 year project would be divided into three main tasks: setting up and validation of the adsorption tube methodology (3 months), data gathering and extension of the space use database (18 months) and analysis of the data and estimation of pedestrian exposure patterns (3 months). The project appointed Dr Ben Croxford who had previous experience of the use of the B&K photo-acoustic gas analyser as the main research fellow, and Carolyn Kirby, an environmental monitoring expert with Cambridge City Council with extensive experience of NO2 tube surveys, to assist part time for two months during the setting up period. During this phase of the project it became clear that the proposed adsorption tube methodology was problematic. While the tubes and and analyser could be used to measure high concentrations such as those found in vulnerable work environments to give 8 hour data, at the low concentrations we expected to find in urban back streets it was clear that the level of accuracy achievable would not be adequate for the needs of the study. This required flexible management and led to a major change in programme. In order to acquire the data needed to fulfil the main objectives of the research a new monitoring methodology had to be designed, developed and validated, all within the constraints of programme and budget. The impact of this was mainly on the time course of the project. It would no longer be possible to devote 18 months to data gathering, and so it was clear that the new methodology must be designed to allow for finer resolution temporal sampling (<8hrs) and continuous monitoring so that the survey time could be compressed. The main requirements for the monitor which were defined during the early stage work were as follows:

a) that it should be cheap enough to allow simultaneous monitoring at a number of sites.

b) that it should have a fine temporal resolution to allow data to be acquired for extremes as well as averages.

c) that it should be able to log all variables continuously so that the survey programme could be compressed;

d) that it should be able to be sited independent of power supply or other wiring constraints, according to the needs of the monitoring programme.

e) that it should be able to monitor key meteorological variables at the monitoring site (wind speed, light, humidity, and temperature) at the same time as pollutant concentration.

The monitoring method was based on the use of a carbon monoxide electro-chemical fuel cell (City Technology Ltd.) a relatively cheap and extremely accurate sensor with a detection limit of less than 0.1ppm and an accuracy of ±5%. Laboratory tests verified the manufacturer's accuracy claims using standard gases, and identified the temperature and humidity sensitivity of the sensor. An algorithm for temperature correction was developed based on these laboratory data. The algorithm may have commercial value and is not described here. Variations from sensor to sensor and drifts over time were investigated through long term experimental set ups (> 6 months). The sensor was established to be highly accurate and robust in all respects. The Street Box monitor itself was developed to include the CO sensor as well as sensors for temperature (within and outside the box), relative humidity and incident light (to detect direct sun, daylight or night). In addition a novel, solid state, wind speed sensor was developed, and validated against conventional mechanical anemometers in the wind tunnel and outdoors. This provides qualitative (windy, breezy, calm) measures of local wind speed at the monitor. All sensors were mounted within a small box together with a small data logging computer, (from Orion Instruments). This setup allows the user to specify logging intervals from 3 seconds upwards for all sensors. A program with a simple user interface was written for downloading data on site into a Psion Organiser, and also for data processing directly into a statistics package on a desktop computer. With sampling at 6 minute intervals the datalogger can hold up to 6 weeks of data between downloading (which takes about 1 minute per week of data). A battery power supply with a top up solar cell was developed and the entire instrument tested under laboratory conditions in the environmental chamber and out on the street. Early prototype designs were subject to failure for various reasons including water penetration, inadequate power supply and internal overheating under direct sun. The final design corrected these faults and was tested against both Westminster and Islington Council's London Air Quality Network sites (GFC NDIR carbon monoxide monitors). An extremely close match (r2=.89) was found (hourly averages over a period of 3 weeks) and 24 monitor units were built [see Box 2].

The Task 1.2 objective 'to assess the relative concentrations of CO to volatile hydro-carbon pollutants including benzene, by means of gas chromatography analysis of duplicate sample tubes' was replaced by the compilation and use of existing national monitoring point data from the Russell Square site, and bringing it together with meteorological data from the closest Met. Office station. This allowed an investigation to be carried out of the use of CO as a proxy for a wide range of other pollutants including particulates (PM10) and benzene, whilst the new monitoring method was being developed.

A large quantity of fine, temporal (six minute average) data were collected at over 20 monitoring sites in 12 streets within a small central London area. Paired sites were located as nearly as possible on opposite sides of each street strapped to lamp posts at 1.8-2m high. In addition Street Boxes were located at roof top level (above the 10th floor) alongside the Bartlett weather station at Torrington Place and within 1m of the sampling point at the Westminster City Council urban network station at Baker Street. Whilst these data were being gathered new pedestrian and vehicular observations were gathered at each site as well as at other locations in the area. All of these data were brought together in statistical databases for analysis. New methods had then to be developed to handle the massive datasets that resulted, to characterise particular street locations in terms of their pollution characteristics and to visualise the way that concentrations vary through time.

A particularly important achievement was the development of a simple way of comparing fine temporal data from a number different sites. The problem is that kerbside pollution concentrations vary enormously from moment to moment as the wind gusts and as traffic passes. This results in a signal to noise ratio so low that direct time series comparison from location to location, or between pollution concentration and passing traffic flows is almost meaningless9. In order to overcome this and since we were interested in 'systematic' spatial variations (those that exhibit regularity over time) rather than temporal variations per se, we developed the use of frequency and percentile plots to characterise a particular location and in effect to give it a 'fingerprint' that reflected not just the average concentrations but the extreme peaks and troughs and all concentrations in between. Given a large number of samples, most factors other than spatial location are randomised. Where there is a consistent bias (say a particular direction of prevailing wind) the bias is in effect a characteristic outcome for that spatial location [Box 3]. It is also possible to split the sample by wind direction, and then to investigate the effect of different wind directions in constructing the frequency distribution of pollution concentrations at that location. In this way the method can be used as a flexible means of investigating the main criteria which construct the pollution fingerprint for a particular location [Box 4].

Once the data had been analysed in this way it became a straightforward matter to investigate the statistical relationships between observed vehicular flows, configurational variables and pollution concentrations characterised in terms of percentile concentrations at various percentile ranges [Box 5]. Very powerful and significant results were found for both, suggesting strongly that the spatial configuration of the built-up area can be used as a single interrelating factor connecting location to pollution concentration at head height, both at mean and extreme values. This allowed us to prototype a novel modelling method to allow prediction of pollution concentrations on the basis of the configuration of urban space [Box 6], it also allowed us to make estimates of likely pollution exposure of the pedestrian population, and to identify the major needs for further research [Box 7].

3. Achievements of the research

In view of the international lack of knowledge, empirical data and methodology in the field of fine spatial resolution pollution in built-up urban areas, and the current public concern over effects of traffic in urban environments, we believe that the research achievements of this project are highly significant. We have succeeded in piloting a new monitoring methodology, gathered a greater quantity of high quality fine resolution data than has been gathered before and developed a means for analysing this that overcomes problems of 'noise' that have confounded previous research using kerbside data. We have then developed and gone some way towards validating a novel modelling technique for predicting ambient pollutant concentrations at head height and kerbside in the fine structure of the street grid. Specifically the research achievements relate to the original objectives as follows:

- The project has evaluated the use of adsorption tube methods (Objective 2) for collecting fine resolution CO data using existing datasets in collaboration with Cambridge CC, and through detailed analysis of the performance of the B&K photo-acoustic gas analyser. Although equipment was cheap, lab analysis time was expensive and the methods were found not to have the accuracy or temporal resolution needed to address questions of spatial variation at ambient urban concentrations.

- The project has therefore developed a monitor for measuring urban CO pollution at head height and at kerbside in built up areas (new Objective). This monitor is cheap (<£500 per Street Box) and accurate (comparable to NDIR monitors). This has allowed data to be collected at finer spatial and temporal resolutions than has been achieved anywhere else in the world to the best of our knowledge. The Street Box has aroused considerable interest in both the research community and amongst Local Authorities and is currently being requested for use in three research projects and by two LA's. We believe that the development of this methodology will prove to be highly significant in the field.

- By monitoring a number of locations in close proximity in the street grid during the same time period the project has developed knowledge about effects of monitor location, and has developed a protocol for obtaining meaningful results in high spatial resolution studies of urban head height pollution concentrations (Objective 1). Methodological recommendations are that monitoring sites be paired on opposite sides of street canyons in order to account for vortex effects and that sites should ideally be mid-block, well away from road junctions so as to achieve more reliable wind conditions, and be more representatvie ofthe street as a whole. These conjectures come from analysis of empirical data, but have also been investigated using CFD modelling of the study area in association with Ni Riain & Littler at the Research in Buildings Group at the University of Westminster1 (new Objective).

- By developing a solid state wind speed sensor and logging other meteorological variables using the Street Box monitor, the project has collected data that allows an understanding to be gained of the effects of meteorological conditions on pollutant concentrations at head height monitoring sites. This is the first time that data on pollution concentration and meteorology have been collected simultaneously at this spatial resolution (Objectives 1 & 3) and the results of this survey have helped define a methodology that is potentially highly significant.

- By constructing a pollution profile or 'fingerprint' of a monitoring location, based on the frequency distribution of pollutant concentrations, a method has been developed for dealing with massive datasets so that many monitoring locations can be compared under different meteorological conditions, local building configurations and traffic management strategies (Objectives 1, 3 & 4). The 'fingerprint' methodology is highly significant in that it overcomes the low signal to noise ratio which has been assumed to make fine spatial scale monitoring unviable. It also makes it possible to unpack the various factors that contribute to the pollution fingerprint of a particular site. We believe that by comparing profiles at particular locations before and after traffic management and calming implementations, as well as at similar 'control' sites, it will be possible to directly visualise and quantify the effects of different management measures on ambient kerbside pollution concentrations as these accrue through time (Objective 5). This is now under test by the TRL who have asked to use the Street Box and associated methodology in their DOT funded study of environmental impact traffic management measures12. If this methodology stands up to the test it will provide a cheap and simple method of environmental appraisal and verification of projected impact for traffic management schemes.

- A method of visualising these huge datasets in both space and time has been developed, using 3-D animations (new Objective) in the Pangea visualisation programme developed under the Intelligent Architecture project (EPSRC GR/J33609). See sample animation.

- Analysis of these data has found that current modelling methods based on background measurements that assume pollution concentrations to vary smoothly from location to location are not valid at head height in built up areas (Objective 1). This is an important finding since these methods are currently in widespread use for both policy formation and scheme implementation. Analysis of our dataset has found ten fold variations between monitoring locations around a street corner from each other. These variations are not random, but are systematic, and the 'fingerprint' can be used to study the impact of rooftop wind velocity and traffic flows on head height concentrations.

- Analysis of national network data confirmed previous findings3,13 that CO can be used as proxy for a number of pollutants including Benzene and PM10 (task 1.2). By bringing that dataset together with meteorological data, this study has verified the relationship between CO and other pollutants and provided further support for the use of CO as a primary indicator pollutant in urban air quality studies concerned with vehicular emissions.

¥ Based on these findings, a novel modelling method has been developed which is tailored to complex urban environments. This uses the pedestrian/vehicular traffic prediction model developed under GR/H48422 which allows detailed street segment level predictions of traffic flows based on the physical configuration of the urban street network14. The new pollution prediction model has been validated in the single case study area for which monitored data were collected, and assessments have been made of pollution exposure of the pedestrian population (Objective 4) based on knowledge of the way that pedestrian numbers vary with urban configuration, and that of how pollution concentrations at head height vary. At present this method constitutes the only practical method for predicting kerbside/head height pollution levels in the fine structure of urban areas. Should the predictive capability of these modelling methods be replicated over a wider range of built form and traffic conditions (see Future Research) they will provide a simple and economic method of general application to many aspects of design and environmental impact assessment.

4. Research Outputs

1. The Street Box, an economic prototype monitoring instrument for measuring urban pollution at multiple sites, together with data downloading software and analysis protocols.

2. The monitoring protocol, including suggested siting standards and analysis procedures to compensate for fine scale 'noise' in data.

3. A simple prototype configurationally based model for fine scale pollution prediction.

4. Papers in refereed journals, articles in books, and other publications, media coverage, reports and web site (see attached list and sample papers).

5. A definition of future research needs [see Box 7].

5. Collaboration and links with other institutions

Early in the research programme a 'peer review group' was established consisting of experts in the various domains the project covered, including pollution monitoring, environmental health, traffic management, meteorology, dispersion modelling and end users from local authorities and engineering consultancy. This group met from time to time to review and comment on progress, and to advise on future direction. This has helped greatly in the management of the project. The group has also proven highly valuable in providing exploitation opportunities for the results of the research. We would like to thank both Islington and Westminster Councils for their help during the project, City Technology, Orion Instruments, and the EPSRC.

Following on from the results of this project four proposals have been tendered to the Cities and sustainability programme of the EPSRC. These proposals have included contributions from other institutions including, Dr Steve Sharples at Sheffield University, Kent County Council, Camden Council, Dr. Paul Wilkinson at the London School of Hygiene and Tropical Medicine, and Dr. Marcel Bottema at Nantes University. The work done during this project has also led to the involvement in two Technology foresight challenge applications, both of which have been accepted into the last round.

Following a presentation at the Highway and Urban Pollution '95 symposium in Copenhagen an informal group of pollution researchers has been set up which meets regularly, this group includes representatives from 6 universities in London, 3 councils, and 2 other institutions. Links to other countries were made at this conference and also at Air Pollution '95 in Greece, regular contact is kept with researchers in Italy, France, Mexico, Spain ,Portugal, Sweden and Denmark.

6. Commercial exploitation

A great deal of interest has been demonstrated in both the monitor itself and in the modelling methods by Local Authorities. LA's are now required to implement policies and schemes in line with PPG13, but have difficulty in assessing the likely impact of proposals particularly in terms of ambient air pollution. The methodology developed in this project is seen as providing an important tool to them in their work. Discussions are already under way with Camden, Islington, Westminster and Kent CC over possible applications of the new techniques. The Transport Research Laboratory is also interested in using Street Boxes to assess the environmental impact of traffic management schemes 12. A license agreement is due to be signed in the next few months with a small company in Berkshire. It is hoped that other license agreements for an indoor and a personal montior will also be arranged soon. The commercialisation is proceeding to plan with an estimated date of Autumn 1996 for commercial availability. Development is continuing and the next version will include the possibility of measuring other gases, and also including a radio link for easier downloading of data.

Appendix A

Explanatory Boxes 1-7

Box 1

The correlation between a fitted variable incorporating effective road width and Radius 3 integration, and normalised all day average hourly traffic flows (r=.914, n=405, p<.0001)

Radius 3 integration is a measure of the degree to which a linear element of 'street space' is accessible within the urban street network. It provides a simple metric of demand for both vehicular and pedestrian flows in urban areas. Effective road width is the main supply metric for vehicular flows but is not related to pedestrian flows. These two variables alone account for a high proportion in the variance of observed flows between different streets in the fine structure of the urban street grid (see Penn et al.1,2).

Box 2

The graph on the left shows the extremely close correlation (r2=0.89, p<0.0001) between the Street Box and the non-dispersive infra-red (NDIR) based monitor used by Westminster Council. The graph shows hourly averages for the period August 25 to August 27, 1995.

Box 3

The graph on the left shows an example of how the pollutant concentrations can vary between closely situated streets. The highest peak of CO is seen on Euston Rd and reaches 12 ppm, the next highest polluted road is Gower St, but there is an isolated peak on Tottenham Court Rd, just after 5:30 pm. The other streets only register near background levels of CO.

Time series data of this sort is highly variable, however by plotting the frequency with which different concentrations are measured over a long sampling period a profile can be created. These profiles vary from site to site giving a 'fingerprint' of the pollution characteristics for each location. The profiles are systematic, and serve to remove temporal 'noise' from the spatially determined characteristics of the location

Box 4

The graphs above show how the pollutant concentrations differ with wind direction. The graph on the left shows a comparison of various percentile values for 3 roads the windward axis is when the monitoring point was on the windward side of the road the other axis represents all other wind directions. This shows systematically higher concentrations when the sensor is on the windward side of the road. The graph on the right shows 6 minute average CO levels for two parallel street canyons with monitoring sites on opposite sides of the road. The roads have similar traffic flows and height to width ratios. Again there is a clear difference in pollutant concentration with wind direction. The values for the monitoring point on the SW side of Tottenham court road are higher with a SouthWest or West wind and lower with a NE or an Easterly wind those for the monitoring point on the NE side of Upper Woburn place show the opposite relationship. These findings confirm that the street geometry determines the flow regime as described in Oke[Oke 1987] in most of the streets measured in this project concentrations were affected by simple vortices 'skimming flow' the main exception was Gordon Square where the sensor was on one side of a block sized park and experienced no apparent vortices exhibiting 'isolated roughness flow'.

The wind direction above the rooftops determines if vortices form in the street canyons. The shape of these street canyons, the average building height and the street width, determines, the shape and scale of these vortices, and this determines whether "clean" rooftop air or "dirty" air from passing traffic passes the monitoring site.

Box 5

The following graphs present the main findings of the project, they show how pollution measurements have been linked to both traffic flows and to variables derived from the spatial configuration of the monitored area. Using simple regression models for the 50th %ile (from the graph of Integration radius3 against 50% ile) and for people density (from2), a mean pollution value for the area of Greater London within the North and South circulars can be calculated.

An unweighted mean of the fitted 50th %ile value of carbon monoxide is calculated at 1.041 ppm, whilst a mean weighted by people density of 1.116, a difference of 7%, (based on the fact that larger numbers of people are exposed to higher levels of pollution on the main roads than on the secondary route structure).

This graph shows the relationship found between measured traffic counts and monitored pollutant concentrations.

A similar relationship to that above is shown for carbon monoxide with the spatial variable combined from the street width and the street grid configuration.

For the 50%ile vs rel r3 +width (r2 = 0.76 p=0.0002) and for the mean, a slightly worse correlation is found (r2=0.72 p=0.0002)

Using the regression values from the graph below allows fitted pollution values to be calculated for each of over 17,000 street segments in the whole of London within the North and South circular roads. The 50th %ile value for carbon monoxide = Integration(3)*0.36 +.17).

Box 6

The fine scale pollution model is based on a representation of the configuration of urban space represented in terms of the main linear elements of the street grid. Relational properties lead to patterns of vehicular flow, and these have been linked to monitored kerbside pollution concentrations [Box 5]. A simple regression model is then used to predict likely concentrations on the basis of the main configurational determinants of monitored pollution concentration - radius 3 Integration and street width.

Fine scale observations have also been linked to 3-D visualisations that allow monitored data on CO concentration to be viewed as colour changes at the same time as wind direction and speed, as these vary through time. A sample animation is available on the world wide web.

Box 7

Future research

There are several directions in which we believe that research should now move.

References

1 Penn A., Hillier, B., Banister, D., Jianming Xu, (1996) Configurational modelling of urban movement networks, 1996, Environment and Planning B, in press.

2 Penn, Hillier, Banister & Xu (1994) Final Report on EPSRC GR/H48422 The relationship between urban structure and movement at the neighbourhood level: civilising urban traffic, The Bartlett, UCL;

3 Quality of Urban Air Review Group (QUARG) (Jan 1993), Urban Air Quality in the United Kingdom.

4 Oke, T.R., (1987), Boundary Layer Climates, Methuen London.

5 Givoni, B. (1989), Urban wind profile as a factor affecting urban ventilation and vehicular air pollution concentration at street level, in Proc. Controlling summer heat islands, UCal, CA, USA.

6 Berkowicz, R., Hertel, O., Modelling Air pollution from Traffic in Urban Areas, IMA Conference on Flow and dispersion through groups of obstacles, March, 1994, University of Cambridge, UK.

7 Royal Commission on Environmental Pollution, (1994), Transport and the Environment.

8 QUARG, op cit p87, 5.8 Research Recommendations.

9 Hickman, Bevan and Colwill (1976) Atmospheric pollution from vehicle emission: Measurement from four sites in Coventry, 1973; TRRL Report LR695

10 Hickman & Lunn (1981), Atmospheric pollution from vehicle emissions: Measurements near the North Circular Road; TRRL Report SR660.

11 Bell, M., (1996) ITEMMS - Integration of Traffic and Environmental Monitoring and Management Systems, Final report to the EPSRC.

12 Abbott, P.G., Baughan, C.J., Chinn, L., Cloke, J., Hickman, A.J., Layfield, R.E., Nelson, P.M., (1996) Environmental Assessment of Traffic Management Schemes: Draft Programme of Research., Transport Research Laboratory, (TRL), PR/SE/141/96, Project record UG93.

13 Ekberg, Lars, (1995), Concentrations of NO2 and other traffic related contaminants in office buildings located in urban environments, Building and Environment, Vol. 30, No. 2, pp 293-298.

14 Penn, A.& Hillier, B. (1995) Analysis of Configuration: knowledge of context and precedent in architectural design in IEE Computing and Control Division Colloquium on "Knowledge Based Approaches to Automation in Construction", June 1995, IEE Digest No: 1995/129, 7/1-9, IEE, London.

15 Dockery, D.W., Schwartz, J., Spengler, J.D., Air pollution and daily mortality: Associations with particulates and acid aerosols, Environmental Epidemiology Program, Harvard School of Public Health, Environmental Research (1992)

16 Morris, R.D., Naumova, E.N., Munasinghe, R.L., Ambient air pollution and hospitalization for congestive heart failure among elderly people in seven large US cities, American Journal of Public Health, 1995, Vol. 85, No. 10, pp 1361-1365

17 Chan,L.Y., Wu, Helen W.Y. (1993) A Study of bus commuter and pedestrian exposure to traffic air pollution in Hong Kong', Environment International, Vol 19, pp121-132.

18 Pikhart. H., (1995), Association between air pollution and respiratory health in Prague (1993-1994), Msc Thesis, School of Hygiene and Tropical Medicine, University of London.

19 Wilkinson, P., personal communication.

Back to urban pollution home page