Fifth Symposium on Highway and Urban Pollution, May 22-24, 1995.

Spatial Distribution of urban pollution: civilizing urban traffic

Dr. Ben Croxford, Alan Penn, Prof. Bill Hillier.
The Bartlett School of Architecture and Planning
Philips House
University College London
Gower Street, London WC1E 6BT, ENGLAND.
tel.: +44 171 380 7890
fax: +44 171 916 1887

email: b.croxford@ucl.ac.uk

Abstract


This paper presents findings from the EPSRC funded research project, "The Effects of street grid configuration on pedestrian exposure to vehicular pollution: civilising urban traffic". The project builds on work done at the Bartlett which has developed new configuration sensitive "space syntax" models that describe and quantify the pattern properties of urban space at the scale of the individual street segment. This model can provide good predictions of both pedestrian and vehicular flows throughout a city, and is currently being used commercially to predict the effect on both pedestrian and vehicular traffic flows of major urban projects. This paper will describe how the model works and how measurements of urban pollution at street level can be integrated into this model.
The main task of the project was to devise a method of measuring urban pollution at the scale of the street segment, in a reliable, accurate and cheap way. The monitoring equipment has been developed here and is set up to measure Carbon Monoxide (CO), temperature, relative humidity, light level, and wind speed at 6 minute intervals. This paper reports early results from a small number of these instruments.

Introduction


This paper presents findings from the EPSRC funded research project, "The Effects of street grid configuration on pedestrian exposure to vehicular pollution: civilising urban traffic". The project itself addresses some of the many further areas for research suggested in the first QUARG report, [QUARG 1993], which calls for more data on a street by street basis, better understanding of meteorological effects on pollutant levels, and assessing the impact of traffic management options.
The problems of research into air pollution are large, and in part reflect the immaturity of the whole field. Existing models rely mainly on estimates and extrapolations, rather than observations, of all variables including traffic flows, traffic speeds, emission rates, and the effects of climatological variables. The error on each estimate propagates through to produce a large total error on the final model. There is also a problem relating to the measuring of pollutants at enough sites, accurately enough, to validate these models.
At the start of this project there was no cheap method of accurately measuring pollutant concentrations at a large number of sites simultaneously. All existing methods are unsatisfactory in some way. Diffusion tubes provide single point measurements representing a week or a fortnight's average, bag sampling provides a cheap snapshot but is labour intensive [5], and other continuous methods are too expensive and too big to install on many street segments, so this project has developed a new method capable of measuring differences in pollutant concentrations between closely situated streets, and has validated this method against existing continuous monitors. This is a first step to providing a method of investigating spatial distributions of urban pollution, and could be used to validate the existing pollution models.
The model used as a background to this work provides a measure of how well a particular street "fits" into the whole structure of the city [1]. This spatial configuration measure can be used to provide good predictions of both pedestrian and vehicular traffic flows, validated using actual counts of both people and cars [6].
This paper has three parts, the first describes the previous work at the Bartlett which explains how to quantify the street configuration of a particular city and how this can be used to predict traffic flows. The second part describes how pollution can be measured in the street at the scale of the individual street segment, leading to the third section that ties these two parts together to investigate possible relationships between street configuration and pollution.

How to describe the street grid configuration of a city


A method of quantifying how well a particular street segment in a city is connected to the entire city's grid structure has been developed here at the Bartlett. It is based on mapping a city using longest lines of sight through that city then performing calculations on each line of sight to provide a number relating to how well "integrated"; that line of sight is within the whole system. One of the consequences of this is to explain numerically how navigation in a traditional town or city is easier than in a new housing estate. Using this method it is possible to analyse a whole city and produce a coloured map in which the relative importance of a particular route within the system, derived from the analysis, can be shown.

A Simple Description of the Space Syntax method of city mapping


By taking an accurate small scale map of the city and tracing over every street the longest line of sight possible, the following map for London can be produced, see figure 1.
Figure 1: Black and white detail of colour axial line map of London that contains over 16,000 lines of sight inside North and South circular, the river Thames can be seen as a winding white space through the map from left to right, other major white spaces are parks.

This map is then digitised into a computer program and the number of connections between streets is analysed. This enables the map to be seen as in figure 2. The streets are coloured, (in this case a grey scale), according to the number of changes of line of sight away from Edgware road. The longest possible journey would consist of 31 changes in direction of line of sight. Because Edgware road is long and straight it is connected to many other lines and is part of many possible routes through the city, thus Edgware road is considered a street that is well "integrated" into the city network.
Figure 2: Black and white version of colour axial line map of London showing number of changes of line of sight from Edgware Rd.

The next map, figure 3, is the same but coloured according to the number of changes of line of sight away from Salter Street in the Surrey Docks area, a very poorly "integrated" street.
In the computer program, Axman [7] this analysis is done for each line, then the results are all combined and normalised to produce a value of "global integration" or "integration radius n" reflecting the fact that the whole system is considered, for each line. Figure 4 shows the "global integration" map for all of London, the darkest streets are those which are the best "integrated" into the street system.

Figure 3: Axial line map of London showing number of changes of line of sight from Salter St., Surrey Docks. (Unavailable)

Other variations of the same map can be made. By colouring the lines according to the number of routes consisting of three or fewer changes in line of sight, known as "integration radius 3" the problem of edge effects can be avoided, i.e. where the proximity of a line to the arbitrarily chosen boundary can affect the results.
"Integration radius 3" is a "local integration" measure and correlates strongly with pedestrian flows (annual average daily flows), the "Integration radius 3" map picks out the "local" major shopping streets in a city, which are a focus for the small scale trips that pedestrians tend to make. When vehicular traffic is also largely "local" such as in the study area, a 1km2 area in North London, correlation coefficients of r2=0.84 are found over 116 street segments [6]. This small area of North London is largely homogenous in terms of street widths.
Figure 4: detail of colour axial line map showing "Integration radius n" of central London. Regent's park is the large white area in the upper left hand corner. Oxford Street has been "cut" as this is closed to normal traffic. (See also whole map on theSpace Syntax Laboratory home page

When larger systems are considered, usable street width is important in the vehicular analysis. Counts of vehicles over several separate and distinct areas of London are closely correlated with a combination measure of just these two variables, relativised "Integration radius 3" and relativised usable street width. The slopes of the correlations from these separate areas all fall on the same regression line, i.e. the method is valid for small areas of London and for the whole system of London as far as the North and South circular roads. Studies of more than 80 other cities world-wide have confirmed the validity of this method which can consistently predict movement with r2's of around 0.75 p<0.0001. [1], [2]
This section has shown that both vehicular and pedestrian movement are both strongly related to the street grid configuration. The question this research addresses is how, and if, the pollution mainly generated by traffic, affects the people who live, work, and walk in the city. This method of city mapping is a powerful tool to investigate possible effects of street configuration on pollution concentrations in a city.

Pollution measurements


To correlate pollution measurements with these spatial measures requires first that good pollution measurements are available for every street segment in the study area. The project initially specified the use of Carbon monoxide (CO) diffusion tubes, these however were found to be a significant source of error and would only provide, at best, one data point per street segment per week. Instead it was realised that by using Carbon monoxide fuel cells and a small datalogger, it would be possible to have near continuous data for each street segment in the study area for a similar overall cost to the diffusion tubes.
The major part of this project has been concerned with developing a method of measuring urban pollution based on these fuel cells, in a cheap, reliable and accurate way. The fuel cells themselves are relatively cheap and extremely accurate, the quoted specifications are +/-5% from 0 to 4000 ppm. These cells are accurate right down to background levels and lower, and have a detection limit of less than 0.1 ppm.
Running six of these sensors side by side gives extremely close agreement and no apparent drift even over a period of 6 months or more (picture not in original paper), the maximum error between any one sensor and the average of all six, is approximately 5%. These CO sensors have been combined with sensors measuring wind speed, light, humidity, and temperature and connected to a datalogger. All the components have been fitted into a small, (171 x 122 x 55 mm) weather proofed box powered by a battery that is kept "topped up" via a photo-voltaic panel. The result is a totally self-contained pollution monitoring instrument, the Street Box, which can be fixed anywhere.
Two of these Street Boxes have been run against a Thermo Electron type, CO monitor based on the non dispersive infra red method (NDIR). This particular instrument is sited at a background site in North London and is maintained by the South East Institute for Public Health (SEIPH). The results of this test are shown below. Despite the very low levels of CO (< 2 ppm) the correlation is excellent, with an r2 of better than 0.8, p<0.0001, for both boxes. See figure 6.
Figure 6: Comparison between the Street Box and the Islington Council government approved Thermo Electron type sensor. Good correlation with an r2 of 0.8 can be seen despite the low levels of CO present.

Wind speed is measured using an experimental sensor, it has no moving parts, and has been tested against a conventional anemometer, the results are shown in figure 7. The good correlation gives confidence in the assigning of a qualitative category to each street measured at a particular time. In this case streets can be typed into calm, breezy or windy that is important for analysis. Both temperature and humidity levels are recorded with an accuracy of +/-2% or better. Light level is measured as just a number and has yet to be calibrated to global solar radiation.
The equipment has thus been designed, built and validated, it can measure six variables simultaneously, and at present is programmed to provide 6 minute averages of one minute samples. The variables measured are :- Carbon monoxide, maximum, minimum, average, Wind speed, max., min, average, Internal and external temperature, averages, Relative humidity, maximum and minimum and Light, average.
Figure 7: Correlation between the novel wind sensor used in the Street Box and a conventional anemometer.

At this frequency of measurement the Street Box can store data for up to six weeks, the data can be offloaded at any time up to this point using a por able computer, the offloading procedure taking approximately one minute per week of data.

The datalogging board has only integer mathematic functions and uses only 8 bit numbers, (0->255), the practical consequences of this are that time is scaled from 0 to 240, so the ten's figure is the hour and the last digit expresses fractions of an hour, so 175 is 5:30 pm, 178 is 5:48 pm. The Carbon monoxide value also has to be scaled to fit into one of 255 levels, I have arbitrarily set this to be from 0 to 25.5 ppm, so the measurement is CO concentration in parts per million times 10.

Results from pollution modelling at the street segment scale.


The use of some of these street boxes in sites around University College London, has produced a complex dataset, the data from which may be presented in several ways. The next graphs, figures 8,9, show results for several Street Boxes, measuring simultaneously, for periods of one day for each graph.
Figure 8: At the start of the survey, five Street Boxes were together, then as they were installed on lamp posts they began to show the clear differences between individual streets. Peaks can be seen relating to rush hours or traffic jams.

Figure 9: Pollution by day. A bad day on Euston Road, Euston road reads much higher than the other streets for the whole day, this is perhaps due to a constant high level of slow moving traffic.

The next graphs, figures 10,11, show pollutant levels for one street for 5 days, but these graphs also show when the street was windy. Clearly when the weather is windy the CO values are lower. All the major pollution peaks occurred at times of calm or breezy weather.
Figure 10: Splitting the data into three categories, windy, calm and an in-between state called breezy, allows the following analyses to be made. Euston road with wind effects shows how increased wind reduces pollutant concentrations, and how all the pollution peaks are on calm days.

(Also Figure 11 showing Tottenham Court Road. Not in original paper)

Figure 12: Gower St with wind effects, shows the same pattern as before. Gower St is aligned North-South whereas Euston Rd is aligned West-East, the wind during this period was mainly from the West.

A novel vector based visualisation program developed at the Bartlett is being used to display results in ways that may show up unexpected patterns, this view, figure 14, shows blocks representing streets. These blocks can represent several variables at once, for example, width could represent traffic flow, block height could represent pollutant levels and colour could represent wind speed.
Figure 14: Visualisation program with blocks representing streets and the height, width and colour of the block representing variables related to that street. This visualisation program can animate an Excel spreadsheet, so each frame of the animation would show one row of 6 minute averages.

Pollution results overview


The turbulent effects of air movement mean that two measurements taken at the same junction can differ enormously from each other with only a small difference in distance. In addition the stop-start nature of traffic at junctions adds confusion to the data. I have therefore tried to measure the pollution at a point in the middle of a street segment. All the street boxes are mounted at the same height using the same criteria, that they should be at about 2.5m from the ground, near the road, away from the possibility of stationary traffic, and as far from any junctions as possible.
How representative these data are of the pollution level in a particular street and whether they can be used to correlate with spatial variables is a matter of concern and will be investigated, however it is felt that using wind speed data to select calmer periods it will be possible to find sufficient, accurate data to allow valid inter-street comparisons to be made.
The results from this initial survey show that
- spatial differences in pollutants exist and can be measured.
- pollutant levels are affected by local climatic variables applied to the street segment, as much as by traffic conditions.

Conclusion


The spatial configuration analyses described in this paper show how, by using only two variables, usable street width and a measure of how well connected a particular street is within a city system, ("Integration radius 3"), a strong prediction of vehicle movement can be made for each line in a system.
This paper has also presented data which confirms that urban pollution is not a homogeneous quantity, it varies on a street by street basis, even streets spatially close can have very different levels of pollution. A method has been outlined whereby these spatial differences can be investigated and related to the configuration of the street.
Due to the ability to select calm or windy days it is possible to compare the pollutant levels of different street segments in different climatic conditions. This enables a picture of the pollution distribution over a small area of London to be built up. Initial results using data from only 8 streets are insufficient to correlate with "integration", the next phase using 30 monitors will provide more detailed data and be used to guide the third monitoring phase which will consist of 60 monitors and be used to monitor up to 60 street segments within a small 1km2 study area.
Later work will investigate how well these variables can be used to predict other pollutants.

References


1. Hillier B., Penn A., Hanson J., Grajewski T., Xu J., (1993), Natural Movement: or configuration and attraction in urban pedestrian movement, Planning and Design: Environment and Planning B, Pion, London.
2. Penn, A.,Banister D., Hillier B., et al, (1991), The relationship between vehicular and pedestrian movement in the smaller scale urban grid: a pilot study, SERC report GR/G 23609.
3. Stonor T., Hillier B., Penn A., Karvoutzi K., (1993) The Manchester pedestrian and vehicular movement study, Internal Report.
4. QUARG group (Jan 1993), Urban Air Quality in the United Kingdom
5. L.Y.Chan, W.Y. Wu, (1993) A Study of bus commuter and pedestrian exposure to traffic air pollution in Hong KongÕ, Environment International, Vol 19, pp121-132, .
6. Penn A., Dalton N., (1994). The architecture of society, stochastic simulation of urban movement., Simulating Societies, University College London Press pp 85-126.
7. Dalton N., (1991) Axman computer program, used to generate spatial configuration models. Bartlett, University College London.
Note
Links to other world wide web pages for colour figures and further explanations, relevant pages are as follows.
Spatial Configuration or "space syntax"
Axman computer program
Back to the main Bartlett page
Or Back to the Urban Pollution home page