Model review 2001 - APPETISE (IST–99-11764) Air pollution...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: APPETISE (IST–99-11764) Air pollution Episodes: Modelling Tools for Improved Smog Management Deliverable D2d.1 Literature review on particulate matter modelling at a point Jaakko Kukkonen1, Mia Pohjola1, Leena Wallenius1, Ari Karppinen1, Stephen Dorling2, Tim Chatterton2, Juhani Ruuskanen3, Mikko Kolehmainen3 , Heikki Junninen3 and Anna Ruuskanen3 1. Finnish Meteorological Institute, Finland 2. University of East Anglia, United Kingdom 3. University of Kuopio, Finland Report Version: 3 Report Preparation Date: 18.09.2001 Classification: Report Contract Start Date: 01.01.2000 Duration: 2 years Project Co-ordinator: Dr. Steve Dorling, University of East Anglia (UEA), Norwich, UK Project funded by the European Community under the “Information Society Technology” Programme (1998-2002) Executive summary This deliverable reviews the modelling of concentrations of particulate matter in urban areas, including deterministic together with statistical and neural network methods. The deterministic models include urban scale dispersion models, and so-called aerosol process (or aerosol dynamics) models. The latter category of models aims to describe various transformation and deposition processes that influence aerosol concentrations. We review models belonging to each of these model categories, and their evaluation against experimental data. Although some promising progress has been made during the last few years on modelling urban particulate matter, only a few deterministic models exist at present for urban scale atmospheric dispersion and transformation of particulate matter. There is experimental data of particulate matter mass fractions (PM10 and PM2.5) for evaluating such models, e.g., in many major Western European cities. However, more detailed experimental aerosol data which could be used for evaluating these models in terms of particle size distributions (such as, e.g., number concentrations and chemical content of particles in various size fractions) is very scarce. Similarly, only a few aerosol process models have been reported that can be applied in urban areas. Such models aim to evaluate various aerosol processes, such as nucleation of particles from the gas phase, condensation and evaporation of vapours onto or from particle surfaces, chemical reactions and the coagulation of particles. Due to the large number of processes and the complexity of treating these mathematically, currently available aerosol process models are extensive programs that are not readily usable for regulatory purposes. The resuspension of particulate material from road and street surfaces is important in many Central and Northern European cities, especially in spring. Possibly the greatest problem with current estimations of PM10 from road transport is the lack of data for the resuspension of dust. However, very few methods have been developed for evaluating the resuspension of particulate matter from street surfaces, caused by traffic-induced turbulence and atmospheric turbulence. These models are semi-empirical, and cannot be applied a priori beyond the areas where the relevant measurements were conducted. The evaluation of regionally or long-range transported background aerosol is also important for evaluating particulate matter concentrations, especially for fine particulate matter (as measured, e.g., by PM2.5). For instance, within the EMEP programme (Cooperative programme for monitoring and evaluation of the long range transmission of air pollutants in Europe), methods are developed at the moment for evaluating the Europeanwide dispersion of particulate matter. However, currently the only usable European scale data for primary particle emissions comes from the inventory compiled by TNO for 1990 and 1993. This inventory is only available as total annual emissions per country and has to be spatially disaggregated on the basis of NOx emissions. The European-scale particulate matter modelling therefore suffers from the problem of poor emissions data. 2 During the last few years, neural network models have also been applied for predicting urban particulate matter concentrations for specific mass concentrations, PM10 and PM2.5. However, the evaluation of these models is based on a very limited set of data. The prediction of aerosol size fractions or the chemical composition of aerosols probably exceeds the capabilities of neural network techniques. The APPETISE project aims to build on the existing methodologies and provide new insights, particularly concerning novel statistical and neural network modelling methods. For instance, the APPETISE project will, during 2001, conduct model intercomparison and evaluation against the data from the urban air quality monitoring network in the Helsinki Metropolitan area. This comparison will also include particulate matter concentrations for PM10 and PM2.5 at several locations. We will combine (i) the Finnish Meteorological Institute (FMI) and University of East Anglia (UEA) expertise on deterministic modelling methods concerning particulate matter with (ii) that developed at UEA and the University of Kuopio (UKU) regarding statistical and neural network modelling methodologies. For instance, the meteorological insight and resources of the FMI team will be fully utilised in order that relevant meteorological data be transferred as input for the statistical and neural network models. 3 Contents Executive summary 2 List of abbreviations 5 1. Introduction 6 2. The models 2.1 Dispersion models 2.1.1 Urban Airshed Model combined with an AEROsol model (UAM-AERO) 2.1.2 The Swedish Meteorological and Hydrological Institute local PM10 model 2.1.3 The External Mixture Model 2.2. Statistical and neural network models 2.2.1 Semi-empirical model for the evaluation of urban PM10 2.2.2 Neural Network models 2.3 Aerosol dynamics models 2.3.1 The AEROFOR 2 model 2.3.2 The MULTIMONO model 2.3.3 The GATOR/MMTD Model 2.3.4 The 3-D aerosol model of Lazaridis and Melas 2.3.5 The SEQUILIB model 2.3.6 The ISORROPIA model 2.4 Fusion of data from ground-based monitoring stations, satellite-based observations and dispersion modelling 8 8 8 9 10 10 10 11 13 13 15 15 16 16 17 17 3. A summary of the models 17 4. Conclusions 19 5. References 22 4 List of abbreviations Institutes: FMI SMHI UEA UKU WHO Finnish Meteorological Institute Swedish Meteorological and Hydrological Institute University of East Anglia University of Kuopio World Health Organisation Chemical compounds and particulate matter fractions: NO NO2 NOx O2 O3 PM10 PM2.5 VOC Nitrogen monoxide Nitrogen dioxide Nitrogen oxides Oxygen Ozone Thoracic particles (aerodynamic diameter is smaller than 10 µm). Fine particles (aerodynamic diameter is smaller than 2.5 µm). Volatile Organic Compound Other abbreviations: AEROFOR BoD CAR-FMI CART EMEP Model for AEROsol FORmation and dynamics (name of a model) Burden of Disease (a measure for the effect of exposure to air pollution) Contaminants in the Air from a Road - Finnish Meteorological Institute Classification and Regression Tree Co-operative programme for monitoring and evaluation of the long range transmission of air pollutants in Europe GATOR Gas, Aerosol, Transport and Radiation air quality model (name of a model) ICAROS Integrated computational assessment via remote observation system using satellite earth observations (name of a EU research project) MATCH Mesoscale Atmospheric Transport and Chemistry (name of a model) MLP Multi-Layer Perceptron MMTD Mesoscale Meteorological and Tracer Dispersion model (name of a model) MULTIMONO MULTImodal aerosol process model based on MONOdisperse multicomponent distributions (name of a model) RPM Regional Particulate Model SCAQS Southern California Air Quality Study SEQUILIB Sectional EQUILIBrium model (name of a model) STEM-II Sulfur Transport and dEposition Model (name of a model) UK United Kingdom 5 UAM Urban Airshed Model (name of a model) 6 1. Introduction Environmental health research conducted during the 1990's has shown that urban air pollution is a substantially more important threat to public health than was previously thought. Studies of long-term exposure to air pollution, especially to particulate matter (PM), suggest an increased mortality (e.g., Woodruff et al. 1997), increased risk of chronic respiratory illness (e.g., Dockery et al., 1993), and of developing various types of cancer (e.g., Knox and Gilman, 1997). The World Health Organisation (WHO, 2000) has estimated that in Europe air pollution has caused 168.000 (range of estimate 100.000 – 400.000) excess deaths annually; in the United States the corresponding figure has been estimated to be approximately 100.000. The best estimate on the reduction in life expectancy in Central Europe is about 1 year. Künzli et al. (2000) estimated that 6 % of all deaths in Austria, France and Switzerland might be associated with exposure of the population to PM air pollution. Fine PM especially is causing a significant burden of disease (BoD) and excess deaths in Europe and North America. However, it is not known which chemical and physical characteristics of the PM are responsible for these effects, and which source categories are responsible for the most harmful exposures. A reliable assessment of human exposure and the subsequent adverse health effects is therefore crucial in terms of the promotion of public health (WHO, 2000). In order to evaluate the ambient air concentrations of particulate matter, a deterministic urban air quality model should include the modeling of turbulent diffusion, deposition, resuspension, chemical reactions and aerosol processes such as, e.g., nucleation from the gas phase, condensation of vapors onto the particles and particle coagulation. Such a model may require several input variables, e.g., size, mass, area or volume distribution of particles, chemical composition in each size fraction, and mass or number concentrations of particles in terms of particle size. Currently, the only usable European scale data for primary particle emissions comes from the inventory compiled by TNO (Nederlandse Organisatie voor ToegepastNnatuurwetenschappelijk Onderzoek/Netherlands Organisation for Applied Scientific Research) for 1990 and 1993 (TNO, 1997). However, this is only available as total annual emissions per country and has had to be spatially disaggregated on the basis of NOx emissions, taken from the EMEP (Co-operative programme for monitoring and evaluation of the long range transmission of air pollutants in Europe) emissions inventory. The necessary information concerning primary particulate matter emissions can in principle be obtained from vehicle-specific emission measurements and estimates. For example, in the the regulatory roadside dispersion model CAR-FMI (Contaminants in the Air from a Road - by the Finnish Meteorological Institute), the primary fine particle (PM2.5) emissions are modelled in terms of travel velocity, classified separately for light and heavy-duty vehicles, equipped with and without a catalytic converter. However, the 7 detailed evaluation of particulate matter emissions from traffic or stationary sources is outside the scope of this review. There are very few models for evaluating the resuspension of particulate matter from the street surfaces, caused by traffic-induced turbulence and atmospheric turbulence. The relative importance of resuspension depends on the mechanical wear of the street surfaces, on street maintenance and cleaning (also on possible winter-time sanding of streets), on traffic-induced turbulence and on the meteorological conditions (e.g., Kukkonen et al, 2001). Resuspension is thought to be an especially important source in the context of Helsinki air quality problems in connection with the transition out of the snow season. The evaluation of regionally or long-range transported background aerosol is also important for evaluating particulate matter concentrations, especially for fine particulate matter. For instance, within the EMEP programme (Co-operative programme for monitoring and evaluation of the long range transmission of air pollutants in Europe), methods are developed at the moment for evaluating the European-wide dispersion of particulate matter. EMEP is a European programme set up in 1979 to provide governments and other bodies with relevant information following the signing of the socalled Convention on Long Range Transboundary Air Pollution (LRTAP). Some regional scale air quality models contain a treatment of particulate matter: the MATCH (Mesoscale Atmospheric Transport and Chemistry) model (Bringfelt et al., 1997), the multiple-source regulatory dispersion model described by Genikhovich (1995, 1998) and and the STEM-II (Sulfur Transport and dEposition Model) system described by Carmichael et al. (1990). Little work has been done on modelling the long-range transport of primary particle matter. In the UK work has been done using two models assessing this. The UK Meteorological Office's NAME model (Maryon et al., 1991, Maryon, 1994, Ryall, 2000, Malcolm et al., 1999, Ryall and Maryon, 1996) has been used by both the Meteorological Office and UEA for modelling both primary and secondary particles (Chatterton et al., in press, Chatterton et al., 2000, Malcolm and Manning, 2001, Malcolm et al., 2000). The NAME model is a Lagrangian multiple-particle model which is capable of modelling transport of a number of pollutants across a global domain (but focussed mainly on UK and Europe, due to input data. The main limitations in using the model are the lack of nitrogen chemistry scheme (currently only sulphate chemistry can be modelled) and the absence of good quality data for primary particle emissions across Europe. Recent work has also been carried out in the UK for the Task Force in Integrated Assessment Modelling (TFIAM) under the Convention on Long Range Transboundary Air Pollution (CLRTRP). This has used a simple straight-line trajectory model to estimate annual average concentration of primary particles to receptors in Europe (ApSimon et al., 2001). This model suffers from the same problem of poor emissions data as the use of the NAME model. 8 However, the detailed discussion of such regional and larger scale models is outside of the scope of this review. This review focuses on the modelling of particulate matter in an urban environment. Once mass concentration or chemical composition data have been acquired, they can then be used to attribute pollution to various sources using a technique known as receptor modelling. Receptor modelling looks at the experienced pollution and then, often using statistical techniques such as Multiple Regression and Principal Components Analysis, differing composites of, e.g., PM10 can be linked in order to track down likely sources on the basis of finger-printing techniques. However, the detailed discussion of receptor modelling is outside the scope of this review. 2. The models In the following, we discuss consecutively (i) deterministic urban dispersion models, (ii) statistical and neural network models, and (iii) aerosol process models. An aerosol process model can be considered as a sub-model of a dispersion model. This review is not designed to be comprehensive, e.g., it certainly does not contain all the models that have been mentioned in the literature. We have selected a few models to be considered in each section; these were considered to be amongst the most promising ones in the literature. 2.1 Dispersion models We have selected three urban scale dispersion models that include a treatment of aerosol processes into this review. 2.1.1 Urban Airshed Model combined with an AEROsol model (UAM-AERO) The three-dimensional gas and particulate matter air quality model UAM-AERO (Wexler et al., 1994; Lurmann et al., 1997) has been developed to predict the size-resolved concentrations of major primary and secondary components of atmospheric particulate matter (PM). The host gas-phase air quality model is the Urban Airshed Model (UAM), which was then interfaced with an Eulerian type aerosol model. Aerosol size distribution treatment in this model is based on a fractional representation with a maximum of eight fractions; aerosol size range varies from 0.01 to 10 µm (when fog is present, the upper limit of this range is set to be 30 µm). The aerosol model is based on the assumption of internally mixed aerosol; this implies that all particles in each size fraction are assumed to have the same chemical composition. The physical processes considered are: advection, turbulent diffusion, condensation and evaporation, 9 coagulation, emissions, nucleation and deposition. The gas phase chemistry is modelled with the chemical mechanism presented by Carter (1990), which includes 60 species and more than 100 reactions, including the formation of condensable organic species from the oxidation of VOC’s. The resuspension of dust from motor vehicle exhaust and paved roads is computed by regression modeling. The effect of the presence of fog on gas and aerosol species is simulated in an empirical manner. When haze or fog exist, the model allows particles to grow to larger sizes. Particle growth and shrinkage are determined by the amount of water transferred to and from particles, based on equilibrium concentrations estimated by the SEQUILIB (Sectional EQUILIBrium) algorithm of Pilinis and Seinfeld (1987).This model predicts gas phase concentrations of 3 compounds and particle phase concentrations of 17 compounds. Dry deposition of particles in a specific size bin are calculated using the deposition velocity of Slinn and Slinn (1980). The model requires as input hourly concentrations of NOx, SO2, NH3, and PM10. The model was tested by simulating particulate matter during two summer episodes in California in the Southern California Air Quality Study (SCAQS) in 1987 (Lawson 1990). The SCAQS in the summer of 1987 consisted of data from 11 sampling days, measured at 55 monitoring sites. The organic particulate matter estimated by the model is a combination of primary and secondary organic materials. The model was reported to underestimate the concentration of fine organic matter and to produce fairly accurate estimates of the coarse organic matter (Lurmann et al., 1997). This model is not suitable for high relative humidity conditions. 2.1.2 The Swedish Meteorological and Hydrological Institute (SMHI) local PM10 model The PM10 model developed at SMHI (Bringfelt et al 1997) computes hourly concentrations of PM10. The model includes three components: meteorological preprocessing, dispersion computations and post-processing. The emission data needed is computed based on a mapping of the traffic flows in different streets and roads, and data on the emission factors of different vehicle types. Resuspension is included; its only source is assumed to be traffic-generated turbulence. The resuspension and direct emissions are combined into one emission factor for each vehicle. The resuspension term is modified according to data on daily humidity and precipitation (rainfall or snowfall). The chemistry included in the model entails reactions with sulphur oxides and oxidized and reduced nitrogen, and is almost identical to that used in the EMEP model (Iversen et al., 1998). The chemistry module contains 11 compounds. The SMHI local PM10 model has been tested against daily mean PM10 data measured at four locations in Norrköping, Sweden, during four periods during 1992 - 1994. The 10 model was reported to underestimate maximum values of PM10 at low wind speeds. The model was also reported to overestimate the source strength of resuspension. 2.1.3 The External Mixture Model The External Mixture Model (Kleeman and Cass, 1998) is a Lagrangian model, which has been developed to evaluate the evolution of the size- and chemically resolved ambient air aerosol. The model includes descriptions of emissions, transport, deposition, gas-to-particle conversion and fog chemistry. The gas phase chemistry is modelled with the chemical mechanism of Carter (1990), including 60 species and more than 100 reactions, with extensions for the formation of condensable organic species from the oxidation of VOC’s. The particulate matter emissions include those from the following sources: catalystequipped gasoline engines, non-catalyst-equipped gasoline engines, diesel engines, meat cooking, paved road dust, crustal material from sources other than paved road dust and sulfur-bearing particles from fuel burning and industrial processes. All particles interact with the same gas-phase mixture. Aged particles are treated separately from freshly emitted particles, even though they may have been emitted originally from the same source type. Dry deposition of particles is computed according to the method of Slinn and Slinn (1980), with modifications to account for the effect of atmospheric stability conditions (Kleeman et al., 1997). During periods of fog, exchange of chemical species between the gas phase and droplet phase is computed using the fog model described by Jacob et al. (1989). The modification of aqueous species by oxidation processes in fog conditions is computed by Kleeman et al. (1997). The data during the 27-28 August 1987 episode of the Southern California Air Quality Study (SCAQS) was used to test model performance, and to perform source contribution analysis. The authors report that the major features of the predicted aerosol size and composition distribution matched the observed data. However, it is clear that the above mentioned experimental data used for this evaluation was very limited. 2.2. Statistical and neural network models 2.2.1 Semi-empirical model for the evaluation of urban PM10 A semi-empirical, statistical model for the simulation of urban PM10 (Karppinen et al, 1999, Kukkonen et al, 2001) (by FMI) was developed and applied to monitoring data collected in the Helsinki Metropolitan district. The model is based on the assumption that local vehicular traffic is responsible for a substantial fraction of the street-level concentrations of both PM10 and NOx, either due to primary emissions or resuspension from street surfaces. 11 The modelling system utilises linear relationships between the measured urban PM10 data against those of NOx in various urban surroundings, based on continuously measured hourly concentration values from an air quality monitoring network in the Helsinki Metropolitan Area. The data used is from two stations in central Helsinki and one suburban station in the Helsinki Metropolitan Area during a period of three years, from 1996 to 1998. The model also includes a treatment of the regional background concentrations, and resuspended particulate matter. The model performance was evaluated against the measured PM10 data from the abovementioned three stations and from two other stations, using data that was measured in 1999. Two alternative model versions were developed, one based on separate correlation parameters (PM10 vs. NOx) for each station, and another based on parameters averaged over the stations considered. The model predicts relatively well the yearly mean concentrations of PM10: the Fractional Bias values range from – 0.05 to + 0.09. Model performance is also relatively good when predicting the yearly mean values that are classified separately for each hour of the day: the corresponding Index of Agreement values range from 0.85 to 0.96. However, model performance is substantially worse in predicting the hourly time series of the year: the Index of Agreement values using the station-specific parameters range from 0.46 to 0.65. The model can be used in order to predict long-term average particle concentrations (for example, annual average values), but it is not applicable for predicting short-term (hourly) average concentrations. The model would not be valid for an urban area in which the influence of local stationary sources on street-level NOx and PM10 concentrations would be dominant. 2.2.2 Neural Network models On the subject of neural network applications, the review by Gardner et al (1998) does not include any applications of neural networks to the modeling of particulate matter. A more recent study by Perez et al (2000) compares the predictions produced by three different methods: a multilayer neural network, linear regression and persistence methods (assigning hourly values on the next day to be equal to the equivalent values on the present day). The three-layer feed forward neural network used had a sigmoid type (nonlinear) transfer function between the input layer and the hidden layer, and between the hidden and output layers. The three methods were applied to the hourly averaged PM2.5 data for the years 1994-1995 measured at one location in the downtown area of Santiago, Chile. The data from May 1 to Sept 30 of both years was divided into the training set and a testing set. The prediction errors for the hourly PM2.5 data were found to range from 30% to 60% for the neural network, from 30% to 70% for the persistence approach, and from 30% to 60% for the linear regression, but the neural network gave overall the best results in the prediction of the hourly concentrations of PM2.5. The neural network was considered to be a reasonable method for prediction only between the hours from 11 a.m. to 7 p.m. 12 Gardner (1999) undertook a model-intercomparison using Linear Regression, MultiLayer Perceptron (MLP) Neural Network and Classification and Regression Tree (CART) (Burrows et al., 1995) approaches in application to hourly PM10 modelling in Christchurch, New Zealand (data period 1989-1992). In Christchurch, domestic solid fuel heating is a significant source of primary PM10 and the diurnal cycle of these emissions is strongly coupled to local meteorology. With its coastal location and complex local terrain, the weather in Christchurch is a function of both local, regional and synoptic scale wind systems (McKendry et al., 1986; Sturman, 1985). The predictor variables which were available to each of the three modelling techniques employed were Time of Day, Wind Speed, Wind Direction, Humidity, Temperature at two heights (1.5m and 10m), Vertical Temperature gradient between those levels, and Solar Radiation. The Wind Speed, Solar Radiation and Vertical Temperature gradient variables were considered, together, to providea measure of atmospheric stability. The time of day and temperature variables characterised the emission signal fairly well. The MLP approach was based on the so-called architecture of n:20:20:1, where n represents a varying number of input variables to test which were most significant (these form the layer with n inputs), and there are 20 weighting nodes in each of the hidden layers, and one output value. The training algorithm employed was the scaled conjugate gradient algorithm (Moller, 1993). While the output from MLP models is not easy to interpret (the ‘black box’ problem), the non-parametric CART method (Classification and Regression Tree) is more transparent. The trees produced were ‘pruned’ in order to maintain ‘generalisation’. The MLP method outperformed CART and Linear Regression across the range of performance measures employed (for example the MLP approach explained 64% of the hourly variability, while CART and Linear Regression explained 50% and 22% respectively). The most important predictor variables in the MLP approach appeared to be (with the most important first) time of day, temperature (which of course helped define the primary emissions), vertical temperature gradient and wind speed (which both helped quantify stability and dispersion potential). As with other statistical modelling approaches the techniques employed underpredicted extreme events. However, in an analysis of model performance in predicting exceedances of the 50 ug m-3 24-hour average PM10 concentration threshold during 1991, the MLP method succeeded in predicting 81% of the 31 days when the threshold was breached. However, of the exceedances predicted (41), 39 % were false alarms. Because PM10 exceedances are a feature of the winter season in Christchurch, improved performance might have been achieved through a seasonal training process. The level of performance should be expected to change, if applied in a setting where the contribution of secondary PM10 is different, or if forecasted meteorological data is used, as opposed to observed data. The multi-layer perceptron (MLP) neural network was selected as the main tool for the modelling the air quality in the APPETISE project, both by UKU and UEA. The main 13 reason for selecting MLP, instead of the self-organised map (SOM), is based on comparisons of the performance of these models against data, as presented by Gardner and Dorling (1999) and Kolehmainen et al. (2001a). The MLP is the most commonly used type of so-called feed-forward neural networks. Its structure consists of processing elements and connections (Hecht-Nielsen, 1991). However, MLP is sensitive on parameters and input variables used, the most critical parameters are the size of the net, training function, back propagation function and transfer function and also the architecture of the net. A good understanding of the modelling problem and the neural networks is therefore required in order to develop an adequate neural network model. 2.3 Aerosol dynamics models Six models are reviewed in this chapter. The first two of these have been developed at the University of Helsinki. In our opinion, these two are the most complete and most appropriate aerosol process models to apply in the context of urban areas. 2.3.1 The AEROFOR 2 model The model AEROFOR 2 (Pirjola, 1998; Pirjola 1999) is a size-fractionated model that treats four particle size fractions separately; these are nucleation, Aitken, accumulation and coarse particle fractions. The particle size distributions in each of these fractions are allowed to have a varying chemical composition. Each of the fractions is assumed to be lognormal . The AEROFOR2 model is a Lagrangian type box model, developed for the investigation of the formation and growth of particles in the atmosphere. The particles can include both soluble and insoluble material. The meteorological data needed for AEROFOR2 are provided by a separate trajectory model. AEROFOR2 includes gasphase chemistry and aerosol dynamics, and calculates the number size distribution of particles as a function of time. The chemistry mechanism is based on the EMEP mechanism (Simpson, 1992). It includes 67 compounds (inorganic and organic) with 139 chemical and photochemical reactions (Pirjola and Kulmala, 1998). The initial concentration, emission rate and deposition velocity of each compound are given as inputs. The model includes the following processes: 1) gas-phase chemistry; 2a) binary homogeneous nucleation of sulphuric acid and water (Kulmala et al., 1998a), 2b) ternary homogeneous nucleation of sulphuric acid, ammonia and water (Korhonen et al., 1999), 3) multicomponent condensation (sulphuric acid, ammonia, water and generic organic vapour) onto pre-existing particles, 4) SO2 uptake by particles, 5) parameterised cloud processes, 6) inter- and intramode coagulation of particles, and 7) deposition of particles. 14 During the growth processes (including condensation, evaporation and coagulation) particles can move from one section to another. The model can be used to estimate the nucleation rates, growth times and composition of particles as well as concentration of condensable vapours (Kulmala et al., 1998b). 2.3.2 The MULTIMONO model The model MULTIMONO (Pirjola et al., 1999, 2000b) is a multimodal model based on monodisperse multicomponent distributions. This model is actually a simplified version of the above mentioned AEROFOR2 model. The term MULTI in the above mentioned acronym refers to the fact that many size distribution sections are included, and the term MONO to the fact that each of these sections is assumed to be monodisperse. The particles are contained in four size fractions; these are the same fractions as in the AEROFOR2 model. The composition in each fraction is a mixture of water, sulphuric acid, elemental carbon, organic carbon, mineral dust, sea salt, ammonium nitrate and ammonium sulphate. The following processes are included; 1) Gas-phase chemistry; 2a) binary homogeneous nucleation of sulphuric acid and water (Kulmala et al., 1998a), 2b) ternary homogeneous nucleation of sulphuric acid, ammonia and water (Korhonen et al., 1999), 3) multicomponent condensation (sulphuric acid, ammonia, water and generic organic vapour) onto pre-existing particles, 4) SO2 uptake by particles, 5) parameterised cloud processes, 6) inter- and intramode coagulation of particles, and 7) deposition of particles. The model is reported to predict the number and mass concentration and the composition of the aerosol, such as inorganic substances (sulphate/nitrate/ammonium), sea salt, dust, elemental carbon and organic carbon and their mixtures. The results of MULTIMONO have been compared with the AEROFOR 2 results (Pirjola et al. 2000a). The input data required by MULTIMONO is complex concerning particle phase compositions, and number and mass concentrations in each size fraction. It may not always be easy to evaluate all the relevant input values for the model. The model has been developed to be a research tool, and it is not readily applicable in regulatory applications. 2.3.3 The GATOR/MMTD Model The Gas, Aerosol, TranspOrt and Radiation model (GATOR) is coupled to the Mesoscale Meteorological and Tracer Dispersion (MMTD) model (Jacobson et al., 1996 and Jacobson, 1997). The aerosol processes included in GATOR simulation are 1) coagulation, 2) gas phase and aqueous chemistry, 3) chemical equilibrization, 4) growth by condensation, 5) growth by dissolution, 6) evaporation, 7) homogeneous and heterogeneous nucleation, 8) dry and wet deposition, 9) sedimetation and 10) emissions. Also included are treatments for gas and aerosol advection, diffusion and convection, and radiative transfer. The MMTD model produces predictions of winds, diffusion, temperature, pressure and humidity, soil moisture and rainfall. 15 The model contains the computation of 18 solid species, 27 water soluble species and one species of residual containing all other compounds. To solve chemical rate equations, a sparse-matrix, vectorized Gear-type code (SMVGEAR) is used. To calculate gas dry deposition velocities, the air quality model of California Institute of Technology (CIT) was implemented (Jacobson et al., 1996). The model uses a so-called moving-center size structure in the book-keeping of the particle size distribution. In this method, the size bin edges are fixed, and the mean diameter of particles within each bin is allowed to vary. This has been included to eliminate numerical diffusion during particle growth. To test the validity of the model, results have been compared with the observations during days of 26-28 August 1987 for gas-phase pollution simulations and 27-29 August 1987 for aerosol-phase simulations, measured in the Southern California Air Quality Study (SCAQS). Emissions data were supplied by the California Air Resources Board (CARB). In addition, topographical data from the U.S. Geological Survey and data from South Coast Air Quality Management District (SCAQMD) were used. The statistics for entire period indicate that gross errors for sulfate and sodium were lowest, 1-40 %. Gross errors for elemental carbon, organic carbon, total particulate mass, and ammonium were 40-60 %. Gross errors for particulate nitrate were 65-70 %. 2.3.4 The 3-D aerosol model of Lazaridis and Melas The 3-D aerosol model of Lazaridis and Melas (1998) couples a model of aerosol dynamics with a three-dimensional aerosol dispersion model. The resultant model includes dry and wet deposition of particles, chemical reactions in the gaseous and aqueous phases simulated with the CB-IV chemical mechanism (Gery et al., 1988), nucleation, and binary condensation in the H2SO4 - H2O system. The chemistry seems to be limited to reactions in the sulphate-water system. The aerosol is described with a size distribution of 16 size bins or sections for the PM10 fraction. The model seems to be limited to treat sulfate particles only. The model was tested against experimental data collected during summer 1990 in Uniontown, Pennsylvania, USA. 2.3.5 The SEQUILIB model The Sectional EQUILIBrium model is a thermodynamic equilibrium model; this model is used, for example, in the model UAM-AERO (Pilinis and Seinfeld, 1987). This model predicts gas phase concentrations of three compounds and particle phase concentrations of 17 compounds, all of these inorganic. The aerosol size composition representation is sectional, and each section is supposed to be internally mixed (all the particles in the section have the same composition). The model was verified against 48 hours of data collected during 30-31 August 1982 in the California South Coast Air Basin. The model underpredicted slightly the nitrate 16 concentrations, overpredicted slightly the ammonia content, while the chloride prediction agreed better with the data. 2.3.6 The ISORROPIA model The ISORROPIA model is a thermodynamic equilibrium aerosol module. It calculates the equilibrium of four gas phase species, 13 liquid phase species and 9 solid phase species, this constitutes a total of 26 species (Nemes et al., 1999). ISORROPIA uses precalculated tables, whereas SEQUILIB solves the equations numerically. The aerosol particles are assumed to be internally mixed (all particles of same size have the same composition). The model was verified indirectly by comparison with the model SEQUILIB. The data used was from Central Los Angeles, between 23 and 25 June 1987. The ISORROPIA model was shown to predict aerosol ammonia and PM2.5 nitrate better, but PM10 nitrate slightly worse compared with the SEQUILIB model. 2.4 Fusion of data from ground-based monitoring stations, satellite-based observations and dispersion modelling Recently, progress has been made in constructing an interactive computational environment that allows the integration and assimilation of different data types (Sifakis et al., 1998 and Sarigiannis et al., 1998). These data types include remote sensing observations, ground-based air quality measurements and results of advanced atmospheric modelling. Such systems can be used for the minimisation of uncertainties in decision-making, regarding air pollution control and abatement in the urban environment. Methods can also be used for providing information concerning the siting of air pollution measurement stations, and the comparison of environmental quality of information. For more detailed information, the reader is referred to the web-site of the European Union 4th Framework Programme project ICAROS (ENV4970417), http://mara.jrc.it/icaros.html. 3. A summary of the models The dispersion models and aerosol process models included in this review are summarised in Tables 1 and 2. 17 Table 1. A summary of the dispersion models considered in this review. In the column 'Processes' , the processes treated by the model are listed as follows: 1. Gas -phase chemistry, 2. nucleation, 3. condensation and evaporation, 4. cloud processes, 5. coagulation, 6. deposition, 7. resuspension, 8. advection. Model acronym Aerosol processes treated UAMAERO 1 2 3 4 60 compounds 5 6 7 8 (SAPRC90) SMHI local PM10 11 compounds (similar to 7 8 EMEP) External Mixture Model Chemistry subModel 2 3 4 60 compounds (SAPRC90) 6 8 Evaluation against data Main references Limited amount of data (USA, SCAQS) In one city (Norrköping, Sweden, 1992 – 1994) Limited amount of data (USA, SCAQS) Wexler et al., 1994 Lurmann et al., 1997 Bringfelt et al 1997 Kleeman et al., 1998 18 Table 2. A summary of the aerosol process (or aerosol dynamics) models considered in this review. In the column 'Aerosol processes', the processes included are denoted as follows: 1. gas-phase chemistry, 2. nucleation, 3. condensation and evaporation, 4. cloud processes, 5. coagulation, 6. deposition Model acronym Aerosol processes Number of Evaluation compounds data against References AEROFOR 2 1,2,3,4,5,6 67 MULTIMONO 1,2,3,4,5,6 67 GATOR/MMTD 1,2,3,5,6 73 Evaluated with and without aerosol processes 3-D Model of Lazaridis and Melas SEQUILIB 1,2,3,6 < 10 Evaluation against data Pirjola et al., 1999, Pirjola and Kulmala, 2000 a,b Jacobson et al., 1996 Jacobson, 1997 a &b Lazaridis and Melas, 1998 1 20 Limited amount of data (USA; SCAQS) Pilinis et al., 1987, ISORROPIA 1 26 Indirectly evaluated by Nenes et al., comparing with 1999 SEQUILIB Sub-models evaluated separately, for each aerosol process Indirectly evaluated by comparing with AEROFOR 2 Pirjola, 1998 Pirjola, 1999 A summary of statistical and deterministic models for modelling particles is presented in Table 3. The multi-layer perceptron (MLP) neural network was selected as the main tool for the modelling particulate matter in APPETISE, both by UKU and UEA. 19 Table 3. A list of statistical and deterministic modelling techniques for particulate matter, to be considered in future work in APPETISE. Technique Partners experienced Remark Deterministic urban dispersion modelling FMI Roadside dispersion modelling FMI Combined use of the UDMFMI and CARFMI models The CAR-FMI model Street canyon dispersion modelling FMI The OSPM model Include Aerosol process modelling FMI The MULTIMONO model Semi-empirical statistical modelling FMI Semi-empirical model for PM10 Include (for evaluating the Influence of various aerosol processes) Include Multi-layer perceptron neural networks Self-organising map neural networks UEA, UKU Two models developed by UEA and UKU Two models developed by UEA and UKU UEA, UKU Recommendation for use in APPETISE Include Include Include Possibly used as a supplementary model 4. Conclusions We have reviewed the modelling of concentrations of particulate matter in urban areas, including deterministic together with statistical and neural network methods. The deterministic models include urban scale dispersion models, and so-called aerosol process (or aerosol dynamics) models. 20 Although promising progress has been made during the last few years on modelling urban particulate matter, only a few deterministic dispersion and aerosol process models exist at present for the atmospheric dispersion and transformation of particulate matter at the urban scale. More detailed experimental aerosol measurement data would be very welcome that could be used for evaluating these models in terms of particle size distributions (such as, e.g., number concentrations and chemical content of particles in various size fractions). Due to the large number of various aerosol processes and the complexity of treating these mathematically, currently available aerosol process models are extensive programs that are not readily usable for regulatory purposes. Improved models would also be needed for evaluating the resuspension of particulate matter from the street surfaces, caused by traffic-induced turbulence and atmospheric turbulence. At the start of the APPETISE project, various models have been selected preliminarily to be used in further work. The main significance and relevance to APPETISE of this literature review is that these models are presented in a wider context, in relation to corresponding previous work by others. We have compared the models selected to be used in APPETISE with those presented in the open literature. For deterministic models, such a comparison concerns both an evaluation of the physical and chemical structure of the models (e.g., which of the most important processes have been treated in each model), and the evaluation of the models against experimental data. For statistical models, this concerns the structure and main principles of the models (e.g., MLP or SOM), and the details of model training and application (e.g., the number of computational layers in a neural network model and the selection of input variables). This literature review is utilised in further work in APPETISE to give guidance on whether the models that have been preliminarily selected are appropriate and functional. The models selected should at least not to be worse than the best of the other models that have been presented in the open literature. The literature review also could give advice on whether any essential processes have been neglected in the models that have been chosen. In the following, the above mentioned matters are discussed separately for each category of models, applicable for evaluating urban particulate matter concentrations. For deterministic models, this contains the discussion of aerosol process models, and for statistical models, the discussion of semi-empirical models and models based on neural networks. Only a few aerosol process (dynamics) models have been reported in the literature that could be considered to be used in evaluating urban scale atmospheric transformation of PM, such as in the evaluations in APPETISE. These include the SEQUILIB model of Pilinis and Seinfeld (1987), the External Mixture Model of Kleemann and Cass (1998), and the MULTIMONO model developed by Pirjola and Kulmala (2000b). Other models reported do not include the treatment of relevant aerosol processes in sufficient detail (nucleation, condensation and evaporation, chemical reactions, deposition and 21 coagulation). However, all of the models mentioned are very complex computer models that have been applied only as research tools up to present. We have selected the MULTIMONO model, developed at the University of Helsinki (UH), to be utilised in further work in APPETISE. The MULTIMONO program is based on a thorough understanding and long-term research on the underlying aerosol processes, conducted at the UH. The individual modules (for instance, that for multicomponent condensation and evaporation) have been separately tested against available experimental data. Considering the statistical models that are not based on neural networks, the literature review showed that there are no such models for evaluating urban PM concentrations, except for the semi-empirical (statistical) model for evaluating urban PM10 concentrations by Kukkonen et al. (2001). Regression equations of monitored urban NOx and PM10 concentrations have been reported previously in the literature; however, for presenting a mathematical model, it also has to be specified how the contributions from various local and long-range sources are treated. We have therefore selected this model for further work in APPETISE. Only a very moderate amount of research has been conducted previously, concerning the application of neural network methodologies to evaluate urban PM pollution, before the start of APPETISE. Such studies were not reported at all in the open literature before 1998 (Gardner, 1988). The present literature review showed that most of the work in this research area has been accomplished by the APPETISE participants. During the last few years, neural network models have also been applied for predicting urban particulate matter concentrations for specific mass concentrations, PM10 and PM2.5. However, the evaluation of these models is based on a limited set of data, although there is experimental data of PM10 and PM2.5 for evaluating such models, e.g., in many major Western European cities. The literature review showed that neural networks, especially MLP, generally perform better or equally, compared with other statistical methods that include linear models. Neural network models should therefore be used as preferred statistical models in WP 2d of APPETISE in the future. The MLP model takes inherently into account the local emission, meteorological and other conditions, by using locally-derived training data. A neural network model has to be adopted to any specific city; the adopted model is not in general applicable to another city. The literature revied showed that it is therefore worthwhile to focus in developing feature selection methods; these enable the model to use optimally the local measurements, by using only a specific selected sub-set of input variables. The APPETISE project will, in 2001, conduct model intercomparisons and evaluation against the data from the urban air quality monitoring network in the Helsinki 22 Metropolitan Area. This comparison will also include particulate matter concentrations for PM10 and PM2.5 at several locations. 5. References ApSimon, H. M., Gonzalez del Campo, M. T. and Adams, H. S. (2001), Modelling longrange transport pf primary particulate material over Europe, Atmospheric Environment, 35, 343-352. Bringfelt, B., Backström, H., Kindell, S., Omstedt, G., Persson, C. and Ullerstig, A., (1997). Calculations of PM10 concentrations in Swedish cities - Modelling of inhalable particles. RMK No. 76, SMHI. Norrköping, Sweden. Burrows, W.R., Benjamin, M., Beauchamp, S., Lord, E., McCollor, D. and Thompson, B. (1995) CART decision-tree statistical analysis and prediction of summer season maximum surface ozone for the Vancouver, Montreal and Atlantic Regions of Canada. Journal of Applied Meteorology 34, 1848-1862. Carmichael, G.R., Peters, L.K., Saylor, R.D., (1990). The STEM-II regional scale acid deposition and photochemical oxidant model - I. An overview of model development and applications. Atmospheric Environment, 25A, pp. 2077-2090. Carter, W. P. L., (1990). A detailed mechanism for the gas-phase atmospheric reactions of organic compounds. Atmospheric Environment 24A, 481-518. Chatterton, T., Dorling, S., Lovett, A. and Stephenson, M. (2000), Air quality in Norwich, UK multi-scale modelling to assess the significance of city, county and regional pollution sources, Environmental Monitoring and Assessment, 65, 1-2, 425-433. Chatterton, T. J., Dorling, S. R., Lovett, A. and Stephenson, M. (in press), The Relative Influences of Primary and Secondary Particulates to Urban Air Quality in the United Kingdom, Water, Air and Soil Pollution. Dockery DW, Pope CA, Xu X, Spengler J, Ware J, Fay M, Ferris B and Speizer F. (1993) An association between air pollution and mortality in six U.S. cities. New Engl. J. Med , 329, pp. 1753-1759. Gardner, M. W., Dorling. S. R., (1998). Artificial neural networks (the multilayer perceptron) - a review of applications in the atmospheric sciences. Atmospheric Environment 32, pp. 2627- 2636. Gardner M. W., Dorling S. R., (1999). Neural network modelling and prediction of hourly NOx and NO2 concentrations in urban air in London, Atmospheric Environment, 33, 709-719. 23 Gardner, M.W. (1999) The Advantages of Artificial Neural Network and Regression Tree Based Air Quality Models. Unpublished Ph.D Thesis. University of East Anglia, UK. Genikhovich, E. L., (1995). Practical applications of Regulatory Diffusion Models in Russia. International Journal of Environment and Pollution, vol. 4-5, no. 4-6, pp. 530537. Genikhovich, E. L., (1998). Russian Regulatory Diffusion Models: Status, Results of Validation and International Intercomparisons. In: Linkov, L., Wilson, R. 'Air Pollution in the Ural Mountains. Environmental, Health and Policy aspects.' Kluwer Academic Publishers. pp. 75 - 84. Hecht-Nielsen R., (1991) Neurocomputing. Reprinted with corrections, Addison-Wesley Publishing Company, Inc. Iversen, T., Saltbones, J., Sandnes, H., Eliassen, A., Hov, Ø., (1998). Airborne Transboundary Transport of Sulphur and Nitrogen over Europe - Model Description and Calculations, EMEP/MSC-W Report 2/89, DNMI, Oslo, Norway. Jacob, D.J., Gottlieb, E.W., Prather, M.J., (1989). Chemistry of a polluted boundary layer. Journal of Geophysical Research, 94(D10), pp.12,975-13,002. Jakobson, M.Z., Lu, R., Turco, R.P., Toon, O.B., (1996). Development and application of a new air pollution modelling system. Part I: Gas-phase simulations. Atmospheric Environment, 30, pp.1939-1963. Jakobson, M.Z., (1997). Development and application of a new air pollution modelling system - Part II. Aerosol module structure and design. Atmospheric Environment, 31, pp.131-144. Jakobson, M.Z., (1997). Development and application of a new air pollution modelling system - Part III. Aerosol-phase simulations. Atmospheric Environment, 31, pp.587-608. Karppinen, A., Härkönen, J. and Kukkonen, J., (1999). A semi-empirical model for evaluating the concentrations of urban particles. Seventh International Conference “Air Pollution 99”, 27-29 July 1999, San Francisco, USA. King, A.M. and Dorling, S.R. (1997). PM10 Particulate Matter - The Significance of Ambient levels. Atmospheric Environment 31(15), 2379-2383. Kleeman, M.J. and Cass, G.R., (1998). Source contributions to the size and composition distribution of urban particulate air pollution. Atmospheric Environment, 32, pp.28032816. Knox, E. G., and Gilman, E. A., (1997). Hazard proximities of childhood cancer in Great Britain from 1953-80. J. Epidemiol. and Health, 51, 151-159. 24 Kolehmainen M., Martikainen H.., Ruuskanen J. (2001a). Neural networks and periodic components used in air quality forecasting. Atmospheric Environment, 35, 815-825. Korhonen, P., Kulmala, M., Laaksonen, A., Viisanen, Y., McGraw, R., and Seinfeld, J.H. (1999). Ternary nucleation of H2SO4, NH3, and H2O in the atmosphere. J. Geophys. Res. 104, pp. 26349-26353. Kukkonen, J., Karppinen, A., Härkönen, J., Pohjola, M.A., Pietarila, H., Koskentalo,T., (2001). A semi-empirical model for urban PM10 concentrations, and its evaluation against data from an urban measurement network. Atmospheric Environment, 11 p. (in print). Kulmala, M., Laaksonen, A., and Pirjola, L. (1998a). Parameterizations for sulfuric acid/water nucleation rates. J. Geophys. Res., 103, pp. 8301-8308. Kulmala, M., Toivonen, A., Mäkelä, J.M., and Laaksonen, A. (1998b). Analysis of the growth of nucleation mode particles observed in Boreal forest, Tellus, 50B, pp. 449-462. Künzli, N., Kalser, R., Medina, S., Studnicka, M., Chanel, O., Filliger, P., Henry, M., Horak Jr., F., Poybonnieux-Texier, V., Quenel, P., Schneider, J., Seethaler, R., R., Vergau, J.-C., and Sommer, H., (2000). Public-health impact of outdoor and traffic related air pollution: a European assessment, The Lancet, pp.795-801. Lawson, D. R., (1990). The Southern California Air Quality Study. J. Air Waste Manage. Assoc., 40, 2, (1990), pp. 156-165. Lazaridis, M. and Melas, D., 1998. New sulfate particle formation during summer pollution episodes using a 3-D aerosol model. J. Aerosol Sci., Vol 29, No8, pp. 913-927. Lurmann, F.W., Wexler, A.S., Pandis, S.N., Musarra, S., Kumar, N. and Seinfeld, J.H., (1997). Modelling urban and regional aerosols - II. Application to California’s south coast air basin. Atmospheric Environment, 31, pp. 2695-2715. Malcolm, A., R., M. and Webster, H.(1999). Development and Validation of a Pollutant Dispersion and Deposition Model for Meso- and Regional Scales, The Environment Agency, R&D Technical Report P302, Bristol, UK. Malcolm, A. L., Derwent, R. G. and Maryon, R. H. (2000). Modelling the long-range transport of secondary PM10 to the UK, Atmospheric Environment, 34, 881-894. Malcolm, A. L. and Manning, A. J. (2001). Testing the skill of a lagrangian dispersion model at estimating primary and secondary particles., Atmospheric Environment, 35, 9, 1677-1685. Maryon, R. H. (1994). Modelling the long range transport of radionuclides following a nuclear accident, Nuclear Energy, 33, 2, 119-128. 25 Maryon, R. H., Smith, F. B., Conway, B. J. and Goddard, D. M. (1991). The UK Nuclear Accident Model, Progress in Nuclear Energy, 26, 2, 85-104. McKendry, I.G., Sturman, A.P. and Owens, I.F. (1986). A Study of Interacting Multiscale Wind Systems, Canterbury Plains, New Zealand. Meteorology and Atmospheric Physics 35, 242-252. Moller, M.F. (1993). A scaled conjugate gradient algorithm for fast supervised learning. Neural Networks 6, 525-533. Nenes, A., Pandis, S.N., Pilinis, C., 1999. Continued development and testing of a new thermodynamic aerosol module for urban and regional air quality models. Atmospheric Environment, vol 33, pp. 1553-1560. Perez, P., Trier, A., Reyes, J., (2000). Prediction of PM2.5 concentrations several hours in advance using neural networks in Santiago, Chile. Atmospheric Environment 34, pp. 1189-1196. Pilinis, C., Seinfeld, J.H., (1987). Continued development and evaluation of a general equilibrium model for inorganic multicomponent atmospheric aerosols. Atmospheric Environment, 21, pp. 2453-2466. Pirjola, L. and Kulmala, M. (1998). Modelling the formation of H2SO4-H2O particles in rural, urban and marine conditions. Atmos. Res., 46, 321-347. Pirjola, L. (1998). Model studies of the formation and growth of atmospheric sulphuric acid/water particles. Report Series in Aerosol Science (Finnish Association for Aerosol Research), 38. Pirjola, L. (1999). Effects of the increased UV radiation and biogenic VOC emissions on ultrafine sulphate aerosol formation. J. Aerosol Sci., 1999, 30, 355-367. Pirjola, L., Kulmala, M., Wilck, M., Bischoff, A., Stratmann, F. and Otto, E. (1999). Formation of sulphuric acid aerosols and cloud condensation nuclei: an expression for significant nucleation and model comparison. J. Aerosol Sci., 1999, pp. 1079-1094 Pirjola, L. and Kulmala, M. (2000a). Computer codes AEROFOR2 and MULTIMONO. University of Helsinki, Department of Physics. Pirjola, L. and Kulmala, M. (2000b). Aerosol dynamical model MULTIMONO. Boreal Environment Research 5, pp. 361-374. Ryall, D. (2000). The Met Office dispersion model, NAME, NWP Gazette, December 2000, http://www.meto.gov.uk/research/nwp/publications/nwp_gazette/dec00/name.html. 26 Ryall, D. B. and Maryon, R. H.(1996). The NAME 2 dispersion model: a scientific overview, UK Meteorological Office, Turbulence and Diffusion Note 217b, UK. Sarigiannis D.A., Asimakopoulos D., Bonetti A., Huynh F., Lointier M., Schafer, Sifakis N.I., Soulakellis N., Tombrou M. (1998). ICAROS: Integrated computational assessment via remote observation system, In: Proceedings of The UK Remote Sensing Society Annual Conference, Greenwich, September 1998. Sifakis, N. I., Soulakellis, N. A. and Paronis, D. K. (1998). Quantitative mapping of air pollution density using Earth observations: A new processing method and application to an urban area. International Journal of Remote Sensing, 19, pp. 3289-3300. Simpson, D. (1992). Long-period modelling of photochemical oxidants in Europe. Model calculation for July 1885. Atmospheric Environment, 26A, 1609-1634. Slinn, S.A. and Slinn, W.G.N., (1980). Predictions of particle deposition on natural waters. Atmospheric Environment 14, pp.1013-1016. Sturman, A.P. (1985). An Examination of the Role of Local Wind Systems in the Concentration and Dispersion of Smoke Pollution in Christchurch and New Zealand. New Zealand Geographer 41, 67-76. TNO (1997). Particulate Matter Emissions in Europe in 1990 and 1993, Netherlands Organisation for Applied Scientific Research, TNO-MEP-R96/472, Netherlands. Wexler, A.S., Lurman, F.W. and Seinfeld, J.H., (1994). Modelling urban and regional aerosols: I. Model development. Atmospheric Environment, 28, 531-546. WHO (2000). Methodology for assessment of environmental burden of disease: Report on the ISEE session on environmental burden of disease, Buffalo, 22 August 2000. WHO/SDE/WSH/00.7 <http://www.who.int/environmental_information/Disburden/wsh00-7/Methodan1.htm> Woodruff TJ, Grillo J and Schoendorf KC., (1997). The Relationship between Selected Causes of Postneonatal Infant Mortality and Particulate Air Pollution in the United States. Environ Health Persp, 105, pp. 608-612. 27 ...
View Full Document

This note was uploaded on 05/06/2010 for the course PHAST 1 taught by Professor Donck during the Spring '10 term at École Normale Supérieure.

Ask a homework question - tutors are online