Air pollution exposure monitoring and estimating Part I. Integrated air quality monitoring system† Jocelyne Clench-Aas,* Alena Bartonova, Trond Bøhler, Knut E. Grønskei, Bjarne Sivertsen and Steinar Larssen Norwegian Institute for Air Research, PO Box 100, N-2027, Kjeller, Norway. E-mail; jocelyne.clench.aas@nilu.no; Fax:+47 6389 8050; Tel:+47 6389 8000 Received 7th April 1999, Accepted 25th June 1999 This paper presents an integrated exposure monitoring system, based on an expansion of existing air quality monitoring systems using dispersion modelling.The system allows: (1) identifying geographical areas whose inhabitants are most exposed to ambient pollution; (2) identifying how many people in an area are exposed to concentrations of pollution exceeding air quality guidelines; (3) describing the exposure of population subgroups (e.g.children); (4) planning pollution abatement measures and quantifying their eVects; (5) establishing risk assessment and management programs, and (6) investigating the short- and long-term eVects of both pollutants and pollution sources on public health. The eVect of pollution is rarely very large and in order to discover it, exposure estimation must provide data that reflects both spatial and temporal variations.Estimates of pollution exposure are obtained using an integrated approach that combines results of measurements from monitoring programs with dispersion calculations. These values can serve as estimates for individual short-term or long-term exposure. The grouped data allows the expression of ambient pollution concentrations as the spatial distribution of estimates such as the mean or 98th percentile of such compounds as SO2, O3, NO2, PM10 and PM2.5. This integrated approach has been combined into a single software package, AirQUIS.ments from monitoring networks can be improved by sup- Introduction plementing measurements with information obtained by The possible role of ambient air pollution on the development dispersion modelling.Should a network of measuring stations or aggravation of chronic diseases such as asthma needs to be be chosen as a basis for exposure estimates without including clarified. However, in the majority of studies, exposure has dispersion models, a very dense network of stations would be been determined without using data on spatial and temporal needed to specify the location of pollution gradients such as variation in pollution concentrations.Improved estimates of those close to roads in urban areas. exposure are needed to attempt to causally relate individual Many large cities have pollution monitoring systems to pollutant exposure to development and/or aggravation of monitor the population’s exposure, that in particular are used chronic disease.to initiate pollution reduction measures and to follow develop- There is a need for information on the distribution of ment of the pollution situation in the area. Adding an intepollution in geographical areas. Potential stakeholders include: grated air quality monitoring system, using results of dispersion (1) public authorities at diVerent levels (municipal, regional, modelling as supplementary information allows: (1) identifying national, international ); (2) industrial users; (3) schools, uni- geographical areas whose inhabitants are most exposed to versities and the scientific community; (4) various organis- ambient pollution; (2) identifying how many people in an area ations; (5) the public and media.are exposed to concentrations of pollution exceeding air quality This paper describes a method developed over 15 years, the guidelines; (3) describing the exposure of population subintegrated air quality monitoring system, that allows a detailed groups (e.g. children); (4) planning pollution abatement estimate of pollution exposure both for populations and for measures and quantifying their eVects; (5) establishing risk individuals using as a base, concentration measurements and assessment and management programs, and (6) investigating data on emissions, wind and dispersion to specify spatial the short- and long-term eVects of either pollutants or pollution variations.In an urban area, emissions from home heating, sources on public health.roads and often industry cause a complex distribution of pollution, with concentration gradients between polluted and less polluted areas. Data on wind conditions and spatial The integrated air quality monitoring method distribution of emissions give information on pollution gradi- The integrated air quality monitoring method combines ents within the city, important when population and individual information from measurement monitoring systems with a exposure is being assessed.dispersion model, using a digital map of a geographic area. Pollution concentrations vary substantially with time, due The method can also be coupled to information concerning to variations in source strength (i.e. traYc flow) and dispersion the population distribution, input data concerning the geo- (e.g.wind velocity). These features can be accounted for by graphical placement of all homes and workplaces using a dispersion calculations. The quality of the continuous measure- Geographical Information System (GIS) system, and also be coupled to global positioning system (GPS) sensors to follow individuals in time. †Presented at AIRMON ’99, Geilo, Norway, February 10–14, 1999.J. Environ. Monit., 1999, 1, 313–319 313The central feature of the integrated air quality monitoring To use wind observations for interpolation in a complex terrain, the terrain influenced wind speed model Mathew is system is the air quality measurements. Dispersion modelling completes the description of the spatial variation of pollution included, as a first approximation.This model is fast and can on an hourly basis estimate inhomogeneous wind fields based concentrations. The dispersion model combines geographical and topographical information with emissions and meteor- on wind observations as input to the dispersion models for concentration calculations. ology. This results in estimates that reflect the pollution situation both spatially and temporally.Empirical data indicate that the vertical gradient of temperature (inversion intensity) and urban scale convergence of The dispersion model horizontal wind are important parameters for describing urban scale dispersion, in particular during pollution episodes. The dispersion model used, EPISODE, combines a finite diVerence model, with a point source and line model, to Emission inventory.To describe spatial concentration distri- account for both stationary and mobile sources. Dispersion butions, including maxima and minima between monitoring modelling can be done combining information from diVerent stations, emissions must be quantified on the same time frame. scales. The local scale involves street segments, area surround- The emission inventory is based on the following features ing chimneys, or street canyons.An urban scale model is also for the appropriate compounds: TraYc—standard emission important when describing air pollution episodes. In Oslo the factors for diVerent types of vehicles, both diesel and gasoline model was run for the urban scale.1,2 However, in Grenland,3 driven, for diVerent speeds and gradients of the roads.Home given in another example, two towns were combined in a heating—standard temperature-dependent emission factors regional model. A regional scale model may also be important based on consumption of wood and heating oil, distributed describing the pollution coming into the area of calculation spatially according to number and type of heating units.by long-range transport or accumulation within the airshed Shipping—based on average number of ships in the harbour. covering a larger area. Industry—location specific emission inventory based on ques- Currently, the model has been run for CO, NOx, NO2, SO2, tionnaires to industry, that includes direct emissions through Cl2, PM2.5, and PM10. It is planned to expand the model to stacks and indirect through leakage from roofs and in loading, include photochemistry, O3, volatile organic compounds transfer or other known processes.(VOCs) and aerosol properties. Emissions are collected either on an annual basis, or patterns In some cases, for example for O3 and NO2, the model of emissions that reflect hourly emissions are provided. For includes a photochemical transformation module that adjusts example, average daily traYc on unmeasured roads is distrib- the concentrations for the presence of other chemically active uted over 24 h using typical time trends obtained by traYc agents in the presence of sunlight.counts. The PM10 model accounts for road dust as a function of As part of the model evaluation, the diVerent elements of road wetness.This is especially important in the Scandinavian the calculation procedure are controlled and evaluated. countries where studded tyres are in use during the winter. Dispersion calculations. Dispersion modelling accounts for Geographical and topographical information. The dispersion background pollution from for example long-range transport, model estimates the concentration of each compound in each from point sources such as industrial emissions, emissions grid square.The dimensions of the grid square are chosen to from domestic heating, tunnel outlets, and from line sources reflect the uses of the estimates and it is usually 1 km2 in the such as roads. Dispersion modelling allows for high spatial centre of cities and possibly 2×2 km2 in more rural areas.resolution. For the spatial model, the topographical contours of the The model computes on an hourly basis concentrations area are given using x–y–z co-ordinates. This information is based on the emission inventory and the meteorological para- an important contributor to spatially distributing pollution in, meters on the same time frame. This allows, especially in areas for example, valleys.with industrial sources, the separation of compounds that are A building register is included that contains building emitted by geographically distinct sources. Fig. 1 shows the co-ordinates, height of roof, number of floors and number of concentration distribution for the same hour of four com- residents. Building height is the basis for stack emissions pounds emitted in an industrial region by geographically connected to home heating.The building centres are defined distinct sources in the Grenland area of southern Norway. as receptor points. Industrial point sources are depicted using Fig. 2 shows how changing wind directions in the area can x–y co-ordinates, and include extra information as to stack influence the geographic distribution of NOx concentrations. height, and height of roofs.At 0600 and 0800 the wind was blowing from the south-west Information for the line model concerning traYc is also up the valley. At 1000 the wind direction began to change. necessary in the dispersion model. Roads are divided into From 1200 the wind was primarily from the north leading to segments delineated by nodes.The parameters needed for each lower NOx values in the northern versus the southern area. segment are: (1) the height of the building associated to the Dispersion models compute pollution concentrations based roads; (2) the distance from the road centre to the buildings; on a short time interval, for example hour by hour. These (3) the width of the road; (4) the slope of the road; (5) the values can be used as is, for individual short-term estimates.presence of important crossings; (6) the presence of crossing However, the data can be aggregated to provide long-term lights and other traYc impediments; (7) the orientation of the exposure estimates. The aggregation time is dependent on the street segment; (8) number of lanes; (9) permitted speed on use of the estimate.For comparison to air quality guidelines, the segment; (10) direction of traYc flow. yearly and daily averages are the most usual. However, for epidemiological studies, the aggregating time may be the time Meteorological data. Standard meteorological information is needed for key points in the geographic area. It is necessary course of the study or a fixed time previous to the study.The grouped data allows the expression of ambient pollution to know temperature and temperature gradients, wind speed, wind direction and stability conditions in the time frame being concentrations as the spatial distribution of estimates such as the mean, the 98th percentile and maximum. The models can estimated. More detailed information on temperature diVerences at diVerent heights, including those obtained using such be used to provide estimates for geographic regions in a grid, or provide estimates for specific receptor points (for example equipment as the SODAR, is advantageous in order to map inversion intensity, pollution flow and dilution. the home address of the participant). 314 J. Environ. Monit., 1999, 1, 313–319mg m_3 Fig. 1 Dispersion of O3, fine fraction of particulate matter (PM2.5), NO2 and SO2, March 9, 1988 at 0100. mg m_3 Fig. 2 Estimated values for NOx on March 9, 1988 in Skien and Porsgrunn, at diVerent times of day. Errors in pollution calculations. A typical root mean square the gradients well when data for the traYc intensity local wind and dispersion is known. (2) The description of dispersion error of hourly values in specified points is comparable to the long-term average value.The error in estimating long term from high chimneys causes a high stochastic error in zones where the plumes touch the ground.6 For describing pollution average values is typically 10–20%.4,5 For many applications such errors are not acceptable. distributions in urban areas, the point source model is important in discriminating between plumes characterised by high However many small-scale structures and characteristics of the pollution situation in urban areas become evident from and by low concentrations in the respective maximum zones.The influence of many high chimneys is of minor importance dispersion calculations using data on emissions and dispersion i.e. (1) Considering pollution coming from car traYc, the for the pollution distribution within the city.However a few chimneys may cause high concentrations within limited areas gradients close to roads are important for people living close to roads. A line source sub-model describes the location of depending on wind and dispersion conditions. These areas are J. Environ. Monit., 1999, 1, 313–319 315specified by the dispersion calculations.(3) An area source is composed of a high number of small sources located on e.g. the roof of each house. The error in calculating the concentration contribution from area sources shows diVerent characteristics than errors in calculating the contribution from line and point sources. The concentration contribution from area sources is high during meteorological situations characterised by low wind and inversion (poor vertical mixing in the atmosphere).Observations of pollution concentrations and results of tracer experiments show that it is diYcult to describe vertical exchange of pollution in these situations and to avoid systematic errors.7,8 The error describing the contribution from area sources does not change from one hour to the next as in the case of the contribution from point sources. When measurements are used in addition to results of dispersion calculations to estimate exposure, the results are improved at least close to the measuring stations.To improve Fig. 3 The AirQUIS system. the method of assimilation of measurements to the results of dispersion calculations the type of errors has to be considered.Further work on the assimilation procedure is expected to survey, background information and other relevant data improve the method for exposure calculations. directly or aggregated for diVerent types of users. Results of concentration measurements are also needed to The air quality data are usually linked to geographical sites. account for pollution concentration from sources outside the In particular when monitoring data are supported and supplied urban area or from an accumulation in an airshed larger than by model estimates of spatial concentration distributions and the urban area.4 impacts, it is suggested that the presentation of the results Calculated concentrations are adjusted using measured data would involve the use of maps or digitalized geographical by using simple kriging for interpolation of diVerences between information systems (GIS).measured and calculated values at and in the neighbourhood of measuring stations. Measurements from stations close to Uses of the integrated air quality monitoring system roads characterized by high traYc intensity are not used for kriging.Depending on the stakeholder, the needs for information vary The accuracy of model results, both spatially and temporally from simple concentrations in diVerent areas, to forecasting have been examined by comparing estimated and measured concentrations before and after pollution abatement measures. values at key sites.5 The common need, however, is for pollution exposure information.The AirQUIS system Estimation of exposure The dispersion model is being incorporated into an interactive air quality management system called AirQUIS9,10,11 that will Pollution exposure can be expressed on a population or individual basis. It is desirable to identify how many people include: (1) a manual and automatic data entering application; (2) an on-line monitoring system; (3) a measurement data in an area are exposed to concentrations of pollution exceeding air quality guidelines, the population estimate. On the other base for meteorology and air quality; (4) a modern consumption/ emission inventory database with emission models; hand, we need to know the exposure for each participant in an epidemiological study to develop exposure/eVect relation- (5) numerical models for transport and dispersion in air of pollutants; (6) an eVect module for population exposure; ships, the individual estimate.(7) statistical treatment and graphical presentation of measurements and modelling results; (8) import/export wizards for The population exposure estimate. The geographical distribution of pollution combined with the geographical distri- import of data and dissemination of results.These elements are integrated in a map and menu oriented bution of the population allows estimation of population exposure. Air pollution impact on health can be estimated by user friendly interface with a direct link to the databases for measurements and emissions and presentation tools. Advanced combining calculated concentrations, either in grid or receptor points such as addresses, and the population distribution. import/export wizards allows the user to easily transfer data to and from the AirQUIS system (see Fig. 3). Exposure estimates can be used to describe how many people that are exposed to air pollution above air quality guidelines The development of an associated database or metadata is important to all air quality monitoring and information sys- and for how long.These data are often used as input to local air quality assessment.5 tems. The database system may consist of several databases which serve as main storage platforms for: (1) on-line collected Population exposure can be calculated in two ways. The exposure of the number of people living in each km2 for air quality data; (2) source oriented emission data and emission modelling procedures; (3) calculated fields of emissions, con- example can be related to hour by hour concentrations in the same km2.12 However, using a GIS system, point estimates of centrations and exposure; (4) historical data with trends, background information such as land used, population distri- air pollution exposure for all homes and buildings in a geographic area can also be made.butions; (5) regulations, guideline values and information for the support and decision making process. The following exposure parameters are calculated:5 Exposure hours—number of hours a number of people is The databases contain information that enables an evaluation of the actual state of the environment and that include exposed to pollution over a selected value.Person dose— accumulated exposure of pollution over a selected value per data for establishing trend analyses, warnings and the undertaking of counter measures in case of episodic high pollution. person. Population load—accumulated exposure of pollution over a selected value for all persons within a grid square. An important part of an integrated system is to present measurements, statistical and modelling results, emission A risk assessment and management program can use popu- 316 J.Environ. Monit., 1999, 1, 313–319lation exposure estimates to assess exceedances of air quality air quality guidelines. The choice of air quality indicators can be expanded. It is not certain that the simple averages or indicators of pollution concentrations. Fig. 4 provides an example that shows an assessment of the number of individuals 98th percentile of pollution concentrations reflect how pollution influences health. It is possible that the time structure in the Oslo area that exceeded air quality guidelines, either during the hour where pollution concentrations were highest, underlying pollution concentrations is an important factor in damaging health.The dispersion model method allows the or during the hour when the greatest number of individuals exceeded the guidelines. identification of other air quality indicators (AQI) that may more correctly reflect health damage.14 Population exposure can also be estimated for the current situation, and forecast those concentrations that would occur after the implementation of pollution abatement measures.Individual pollution exposure estimate The model can quantify the relative importance of various In quantifying the health eVects of air pollution exposure, it sources of pollution in the area for the exposure values.4 This is desirable to both quantify the size of the eVect and relate information has been used to develop abatement strategies.13 this quantity to the eVects of other individual sociodemographic factors such as smoking or passive smoking, Air quality indicators income, sex, age, etc.The eVect of pollution is rarely very large15,16 and in order to discover it, exposure estimation must There is a need, especially for public authorities, to examine pollution indicators and compare them to accepted or existing reflect a natural variability as much as possible.Fig. 4 Estimation of population exposure, where the hour either with the highest calculated value of NO2 is portrayed, or the hour where the greatest number of people exposed to values exceeding the Norwegian air quality criteria value is presented. Table 1 Diagram of the dynamic exposure assessment model (DINEX) for each time unit Dispersion Model Emissions Meteorological conditions Dispersion calculation of a concentration field Adjustment of results to reflect background concentrations and measurements Exposure estimate Localization in the area Accounting for traYc Accounting for indoor environment (1) NOx and O3 chemistry (1) Ventilation—window open/closed (2) Extra suspended particles (2) Season (3) Smoking (own/passive) Estimate of personal exposure J.Environ. Monit., 1999, 1, 313–319 317Fig. 5 Relationship between pollution estimated at each child’s home using the STINEX method and the average exposure estimated for each child in the panel using the DINEX method. Several methods of measuring/estimating individual expo- Individual pollution exposure estimate—short-term.In studying the short-term eVects, a more detailed, dynamic approach sure have been used. The method most often applied to measure exposure to pollution has been to use concentrations is necessary, DINEX (Dynamic INdividual EXposure estimate). A method has been developed to estimate exposure to measured at one or several fixed stationary pollution measuring stations in pertinent locations in the area.Measurements at a air pollutants that is called the diary method.44 In this method an individual fills out the diary with information about station at a fixed site generally do not represent the pollutant concentrations people are in reality exposed to. Since it is location, time and health which is recorded chronologically.The location information from the diary is associated to the costly to do measurements, it is usually not possible to measure at a suYcient number of stations to get a more complete spatial and temporal distribution of air pollutants as described by using results of dispersion calculations.1,2 Estimated concen- picture of population/individual exposure. Other methods of measuring exposure include both passive trations are controlled by measurements and may be rejected when the deviation is large.Table 1 briefly summarises the and active portable monitoring equipment. Active portable monitoring equipment is costly and leads to altered behaviour elements in the suggested computer pollution exposure model. The DINEX method can also be used to check the estimates while individuals must carry relatively little robust and expensive equipment.It is diYcult for children to carry such equip- of exposure by population subgroups based on the concentrations calculated hour by hour as a function of where the ment. Passive equipment is more robust, but gives a cumulative concentration that does not allow measuring peak values or individuals actually are.Fig. 5 compares NO2 and PM2.5 exposure estimates for children living in Oslo either as a static other exposure over a short time period. Therefore, several methods of estimating exposure have estimate outside their home or as an aggregated estimate of dynamically estimated exposure using a diary over a 6 week been used. Generally, these methods provide pollution exposure based on statistical information for concentrations in period.The figure shows considerable variability between the two methods. This comparison provides valuable information diVerent microenvironments combined with general activity patterns in specialised population groups.17–24 These methods on the inherent variability of the more traditional exposure methods.2 In the future, other air quality indicators can also seldom allow individual pollution estimates, rarely allow continuous estimation of pollution exposure, or forecasting or be estimated based on the dynamic exposure of the individuals that will provide more detailed information on the temporal testing of eVects of pollution abatement measures.Although used only sporadically previously,25–42 dispersion pattern of exposure of the population to pollution.For some areas there is a positive correlation between agglomeration of modelling is becoming a supplementary tool in combination with concentration measurements. Seldom, however, has this people and pollution concentration. method been used to generate exposure estimates on an hour by hour basis as described in this paper.Conclusion Individual pollution estimating methods have been used in Individual pollution exposure estimate—long-term. Longterm health eVects can be related to air quality indicators, Norway, both for cross-sectional studies and for panel studies. In cross-sectional studies, health status of a selection of the calculated for the home. This can be called the STatic INdividual EXposure estimate (STINEX).The choice of air population is assessed together with a set of socio-demographic parameters, whereof pollution is one. The pollution estimate quality indicators can diVer for diVerent health end-points. Air quality indicators include peak, average, 98th percentiles used in cross-sectional studies is usually a yearly average. In panel studies, however, one follows the individual and com- of compounds such as SO2, O3, NO2, PM10 and PM2.5 .Dispersion modelling can be used to provide estimates of pares the health of the individual when exposed to pollution to the same individual when not exposed to pollution. Here ambient exposure outside the home, the work place or school, or both. These estimates can reflect the time frame of interest, the pollution estimate used is the hourly or daily concentration. These studies have been done both in areas with industrial such as yearly, seasonal or monthly averages.1 These estimates can also be used to estimate cumulative exposure over for pollution sources3 and with traYc pollution as the primary source.1,2,45 The methods have also been used in the Czech example a life time or several years, by estimating concentrations in the geographic units, based on known or estimated Republic46 and China.47 In order to estimate the proportion of the population changes in emissions of diVerent compounds.43 Since the estimate is calculated for each individual in a exposed to unsatisfactory air quality, and to quantify in exposure–eVect relationships the eVects of air pollution expo- survey, it can be used to estimate exposure in population subgroups such as children or the elderly.sure, refined methods of exposure estimating are necessary. 318 J. Environ. Monit., 1999, 1, 313–31919 M. L. Williams, Sci. Total Environ., 1995, 168(2), 169. Methods that account for the spatial and temporal aspects of 20 N. Duan and D. T.Mage, J. Expo. Anal. Environ. Epidemiol., pollution dispersion are currently commercially available. In 1997, 7(4), 439. the future, methods to accurately estimate indoor concen- 21 M. E. Korc, J. Air Waste Manag. Assoc., 1996, 46(6), 547. trations need to be incorporated into these models. 22 T. McCurdy, J. Expo. Anal. Environ. Epidemiol., 1995, 5(4), 533. 23 T. R. Johnson, J. Expo.Anal. Environ. Epidemiol., 1995, 5(4), 551. 24 D. L. MacIntosh, J. Xue, H. Ozkaynak, J. D. Spengler and P. B. References Ryan, J. Expo. Anal. Environ. Epidemiol., 1995, 5(4), 375. 25 P. J. Van den Hazel and C. H. Waegemaekers, Public Health Rev., 1 A. Bartonova, J. Clench-Aas, F. Gram, K. E. Grønskei, C. 1991–92, 19(1–4), 251. Guerreiro, S. Larssen, D. Tønnesen and S. E. Walker, J.Environ. 26 S. Larssen, D. Tønnesen, J. Clench-Aas, M. J. Aarnes, K. Monit., 1999, 1, 337. Arnesen, Sci. Total Environ., 1993, 134(1–3), 51. 2 J. Clench-Aas, A. Bartonova, K. E. Grønskei and S. E. Walker, 27 D. Zmirou, A. Deloraine, P. Saviuc, C. Tillier, A. Boucharlat and J. Environ. Monit., 1999, 1, 333. N. Maury, Arch. Environ. Health, 1994, 49(4), 228. 3 J. Clench-Aas, A.Bartonova, K. E. Grønskei, L. O. Hagen, O. A. 28 P. G. Georgopoulos, V. Purushothaman and R. Chiou, J. Expo. Braathen and S. E. Walker, J. Environ. Monit., 1999, 1, 341. Anal. Environ. Epidemiol., 1997, 7(2), 191. 4 K. E. Grønskei, S. E.Walker and F. Gram, Atmos. Environ., 1993, 29 C. Sacre, M. Chiron and J. P. Flori, Sci. Total Environ., 1995, 27B (1), 105. 169(1–3), 63. 5 S.E. Walker, L. H. Slørdal, C. Guerreiro, F. Gram and K. E. 30 J. E. Till, Health Phys., 1988, 55(2), 331. Grønskei, J. Environ. Monit., 1999, 1, 321. 31 M. C. Hatch, J. Beyea, J. W. Nieves and M. Susser, Am. 6 H. R. Olesen., Int. J. Environ. Pollution, 1995, 5(4–6), 776. J. Epidemiol., 1990, 132(3), 397. 7 K. E. Grønskei, in Proceedings of the third International Clean Air 32 A. Q.Eschenroeder and E. J. Faeder, Risk Anal., 1988, 8(2), 291. Congress, VDI-verlag, Dusseldorf, 1973. 33 A. Bouvill and A. Despres, Rev. Epidemiol. Sante Publique, 1982, 8 K. E. Grønskei, in Volume of the Ninth Symposium on Turbulence 30(2), 205. and DiVusion, Roskilde, Denmark. American Meteorological 34 C. E. Bostrom, J. Almen, B. Steen and R. Westerholm, Environ. Society, Boston, MA, 1990.Health Perspect., 1994, 102, Suppl 4, 39. 9 T. Bøhler and B. Sivertsen, A modern Air Quality Management 35 P. M. Cavalini, Arch. Environ. Health, 1994, 49(5), 344. system used in Norway, Norwegian Institute for Air Research 36 S. Wing, D. Richardson, D. Armstrong and D. Crawford-Brown, (NILU F 4/98), Kjeller, Norway, 1998. Environ. Health Perspect., 1997, 105(1), 52. 10 T.Bøhler, Environmental surveillance and information system, 37 C. B. Thompson and R. D. McArthur, Health Phys., 1996, 71(4), presented at the Air Pollution 95 Conference, Porto Carras, 470. September 26–29, 1995, Norwegian Institute for Air Research 38 L. L. Philipson, J. M. Hudson and A. M. See, Toxicology, 1996, (NILU F 13/95), Lillestrøm, Norway, 1995. 111(1–3), 239. 11 B. Sivertsen, Presentation for the International Emergency 39 N.A. Esmen and G. M. Marsh, J. Expo. Anal. Environ. Management and Engineering Conference, Florida, April 18–21, Epidemiol., 1996, 6(3), 339. 1994, Norwegian Institute for Air Research (NILU F 7/94), 40 D. H. Kraig, Health Phys., 1997, 73(4), 620. Lillestrøm, Norway, 1994. 41 M. P. Janssen, R. O. Blaauboer and M. J. Pruppers, Health Phys., 12 L. H. Slørdal, Calculation of exposure to NO2, PM10 and PM2.5 for 1998, 74(6), 677. Oslo, Drammen, Bergen and Trondheim, Norwegian Institute for 42 M. M. Ihrig, S. L. Shalat and C. Baynes, Epidemiology, 1998, Air Research (NILU OR 38/98) (in Norwegian) Kjeller, Norway, 9(3), 290. 1998. 43 J. C. Caldwell, T. J. WoodruV, R. Morello-Frsoch and D. A. 13 K. E. Grønskei and F. Gram, in: Proceedings of the 8th World Axelrad, Toxicol. Ind. Health, 1998, 14(3), 429. Clean Air Congress, 1989, ed. L. J. Brusser and W. C. Muldur, 44 N. Duan, Environ. Internat., 1982, 8, 305. Elsevier Science Publishers, Amsterdam, 1989. 45 J. Clench-Aas, S. Larssen, A. Bartonova, M. J. Aarnes, K. Myhre, 14 C. Guerreiro, J. Clench-Aas and A. Bartonova, J. Environ Monit., C. C. Christensen, I. L. Neslein, Y. Thomassen and F. Levy, The 1999, 1, 327. health eVects of traYc pollution as measured in Va°lerenga area of 15 Asthma and outdoor air pollution. Committee on the Medical EVects Oslo, Norwegian Institute for Air Research (NILU OR 7/91), of Air Pollutants, ed. S. T. Holgate and H. R. Anderson, Dept. of Lillestrøm, Norway, 1991. Health, HMSO, London, 1995. 46 K. E. Grønskei, A. Bartonova, J. Brechler, S. E. Walker and 16 Non-biological particles and health. Committee on the Medical S. Larssen, J. Environ Monit., submitted. EVects of Air Pollutants, ed. S. T. Holgate and R.Waller, Dept. of 47 J. Clench-Aas, S. Larssen, L. Zhiqin, K. Aunan, unpublished Health, HMSO, London, 1995. work. 17 A. C. Taylor, J. Expo. Anal. Environ. Epidemiol., 1993, 3(3), 285. 18 R. E. Keenan, B. L. Finley and P. S. Price, Risk Anal., 1994, 14(3), 225. Paper 9/02775K J. Environ. Monit., 1999, 1, 313–319 319