Reports and Papers

Model-based Decision Support
for Integrated Urban Air Quality Management

    Dr. Kurt Fedra
    Environmental Software & Services GmbH
    A-2361 Gumpoldskirchen, AUSTRIA


Urban air quality management faces new and continuing challenges, driven by new legislation and public awareness on the one hand, and the growth of urban conglomerates and increases in power consumption and traffic on the other. While air quality modeling is a well established field, the challenge is to integrate scientific tools of analysis with the environmental policy making and management process, to involve a large and diverse audience and participants in the policy and decision making processes, and to support new functions such as the information of the public. This requires to embed air quality models in a conceptual framework that includes and explicitly addresses policy relevant elements such as the control of emission sources including economic criteria, monitoring of ambient air quality and the compliance with standards, and impacts on human health and the environment.


From an information and decision support point of view, the urban air quality management problem is characterized by a number of features. They include:

  • multiple sources of information, ranging from census data compiled every few years to continuous on-line monitoring systems;
  • a dynamic and spatially distributed structure with multiple temporal and spatial scales for the complex dispersion and transformation processes that translates emissions into ambient air quality, which is the domain of air quality modeling proper;
  • distributed (and mobile) emission sources with pronounced temporal patterns that include industry, households, and traffic, or, from a different point of view, an energy sector that can be modeled as a large scale mathematical programming problem, and a traffic sector that can be modeled as a network (dynamic) equilibrium process;
  • direct regulatory and indirect economic control on emission sources, involving complex human behaviour;
  • multiple objectives and criteria at different spatial and temporal scales for the different actors and the regulatory framework.

This, obviously, defines a rather complex problem domain, which also includes a broad range of actors, stake holders and audiences in the decision and policy making process. With the shift from more or less authoritarian and technocratic to participatory decision models that characterise the political evolution of the last several decades, technical and scientific information, and the free and open access to this information, has become an important element in the political process. Consequently, information technology plays an increasingly important role where technical and scientific issues are involved, as is certainly the case in urban environmental management (Fedra [2]).

Decision Support for Air Quality Management

The management of urban air quality includes a number of closely related tasks that can broadly be grouped into monitoring, emission control, and impact assessment, together with related reporting and public information provisions (90/313/EEC). These tasks include the continuous monitoring of ambient air quality for compliance with EU (96/62/EC) and national regulations, including the appropriate responses if certain thresholds and alert levels are exceeded, as well as the regular reporting on the state of the environment. Related to the monitoring, and in particular driven by any violation of standards and thus failure to comply with existing regulations, is the formulation of general policies and strategies to reduce emissions and thus ambient concentrations (98/96/EC), or to comply with emission-related standards such as CO2 protocols.

Major emission sources are controlled by another body of regulations, (e.g., 88/609/EEC, 98/429/EEC, 89/369/EEC, COM(96)538), usually related to the commissioning of industrial plants, power plants, or waste incinerators. Mobile emission sources again are regulated by a number of strategies including general engine exhaust characteristics, vehicle inspection programs and strategies affecting fleet composition, and fuel quality requirements, e.g., in the Auto Oil Program (94/12/EC, 41/441/EEC, 96/69/EC). A third major group of regulatory tasks is related to environmental impact assessment for a number of projects and activities defined in (97/11/EC, 85/33/EEC). For a recent compilation of information resources on air quality and environmental impact assessment, see the web site of the Info 2000 project AIR-EIA:

In all these cases, the use of models provides for either descriptive or prescriptive analysis. Descriptive analysis or scenario analysis explores WHAT-IF questions, forecasting the expected behavior of the system in response to a set of changes (or the lack thereof, i.e., a business as usual scenario) projected into the future. The daily forecast of tomorrows expected ozone concentration would be one example, the forecasting of the effect of a new road construction on ambient air quality another. For the case of accidental emissions, regulated by the so-called Seveso Directive (96/82/EC), the requirements for external emergencies specifically refers to scenario analysis by defining a set of credible or most likely accident cases as the basis for safety analysis.

A specific use of descriptive modeling is in combination with monitoring, where the Air Quality Framework Directive 96/62/EC specifically addresses the use of models to supplement monitoring programs, which amounts to a mass-budget based approach to spatial interpolation.

The conceptual framework

Air quality management involves a number of basic building blocks that form a conceptual framework for the analysis and formulation of management strategies and policies:

  • Sources of emission, represented in various emission inventories for industrial, commercial, or domestic sources and the transportation system, as well as landuse related sources (biogenic emissions of VOCs, particulate matter from soils and street surfaces);
  • The monitoring system observing ambient air quality and historical trends with emphasis on the peak values that may exceed regulatory standards;
  • The dispersion and transformation processes, driven by emissions, meteorology, and local topography, that translate emissions into the ambient concentrations, represented by air quality models (e.g., Zanetti[19]);
  • Impact assessment, which translates the ambient concentrations into costs (in terms of public health and environmental damage);
  • Control strategies which basically attempt to limit emissions, relocate them, or mitigate impacts where that is possible.

It is these components and building blocks that any approach to decision support must address and refer to. The design of a comprehensive air quality strategy for an urban region must consider the spatial distribution of emissions and impacts, the distribution of cost to the economic agents, and the role of constraints imposed by environmental laws and regulations. Some decisions are discrete and of a design type (this is in particular true in the traffic control strategies based on a redesign of the network, of the setting of air quality standards), while others are amenable to a treatment via economic calculus based on comparing marginal costs.

A multi- tiered, iterative approach for the design of such strategies has been proposed (Fedra and Haurie[3]):

  1. Obtain emission scenarios for different sectors (energy, transportation) through optimization models maintaining economic efficiency and meeting sectoral objectives and constraints. This is based on rational economic behavior of realistic agents at the sectoral level. The sectoral optimization approach is based on the use of models for energy, e.g., MARKAL optimizing an energy demand-supply balance for the city including bounds on pollution emissions (Wene and Ryden [18]), and for transportation, e.g., EMME/2 simulating traffic equilibria (Hearn and Florian[6]) which again constitute an emission scenario.
  2. Get a representation of the resulting ambient air quality as a function of these emissions for different averaging periods, relating to air quality standards through solving of the dispersion equations with spatially distributed air quality models.
  3. Obtain spatially distributed measures of environmental and public health impacts based on land use information and a population distribution, resulting in a spatially distributed measure of vulnerability or damage function, for different classes of pollutants.
  4. Minimize this distributed impact function subject to economic constraints by distributing the maximum acceptable costs. Alternatively, subject to environmental quality constraints, allocate the maximum permissible emission levels to traffic and energy uses while maintaining economic efficiency. This global optimization takes emission scenarios (unmitigated or sectorally optimized) as its starting point and is based on a source-receptor matrix computed by a long-term air quality model.
  5. Use the permissible emission levels obtained as a constraint on the previous sectoral optimization. Feedback loops from the impact analysis and global optimization can help to redefine objectives and constraints of the sectoral models.
  6. Repeat the above steps to obtain a number of (sectorally) optimal scenarios.
  7. Use a discrete multi-criteria tool to find a preferred compromise solution that satisfies the objectives of all groups of actors.

    Reflecting the iterative nature of the approach, an interactive and graphical user interface supports the efficient comparative evaluation of scenarios describing alternative plans and policies, and thus the feedback loops linking the steps of the approach. The translation of emissions into ambient pollution concentrations is accomplished through an averaging of the mathematical description of a complex stochastic dynamical system (the dispersion process). Impact assessment estimates the spatially distributed impacts of the ambient air quality (the immission) resulting from pollutant emission on the environment and human health; for example, comparing census data and exposures or health risks at different locations, which is essentially a GIS application.

    The overall task necessitates a combination of energy and technology choices and traffic design policies under a spatially distributed constraint on pollution concentrations. The combination of the power of large scale optimization, spatially distributed simulation and GIS techniques is expected to help develop rational and efficient air quality management strategies.

    The technological framework

    Urban air quality management addresses tough problems: they are complex; dynamic, and involve spatially distributed 3D phenomena and models. They also involve problems of communicating difficult technical concepts and data to a largely non-technical audience, and of assisting non-technical users with complex analytical tools.

    A possible architecture to support this complexity efficiently uses an object oriented client-server approach to integrate distributed information resources, and provide an easy to use and understand user interface (Figure 1). It is based on:

    • A flexible client-server implementation for distributed and decentralized use of information resources.
    • Communication architecture based on the http protocol which is used to integrate real-time data acquisition from monitoring sites, as well as optional high-performance computing resources such as supercomputers or workstation clusters; primary consideration here is the scalability of applications over a wide range of performance requirements.
    • Multi-media user interface design to support an intuitive understanding of results.
    • Integration of GIS with data bases, monitoring results, and spatially explicit simulation modeling (Fedra[1]).
    • Embedded rule-based expert systems for logical modeling and user support.
    • Integration of a range of simulation models.
    • Optimization models that supplement the dynamic simulation models for design and decision support tasks.

    Figure 1: systems architecture

    Conceptually, the system uses a central server that coordinates the information resources and a number of display clients. The elements include a number of data bases, which may be linked to an on-line monitoring system, GIS data describing the domain and providing spatially distributed model inputs, emission inventories, and models scenarios and results. The emission data may also be linked to external models of the energy sector or of traffic, that create emission data according to their respective scenarios.

    The second major block of information resources are the component simulation models, ranging from simple steady-state screening models to dynamic, 3D photochemical models which are implemented on parallel computers for better than real-time performance. Depending on their computational requirements, ranging up to supercomputers or workstation clusters, they may be implemented locally or on a remote compute server.

    The data bases contain both spatially distributed data (topical maps) that are managed, processed, and displayed by the GIS functionality; data bases for temporal data (air quality observations and meteorological data), that are linked to on-line monitoring systems for continuous updates; and emission inventories that may be sets of point sources; polygons or regular matrices for area sources; and networks for traffic generated emissions. Point, polygon, and line sources may be converted to regular cell grids (and be displayed in this format in the GIS) as one of the input formats for some of the component models. Digital terrain data (for 3D wind field modeling), landuse (e.g., for roughness), and population data for exposure complement the data requirements.

    Sources of emission

    Emission sources are the main input to any dispersion modeling, and at the same time the major focus of any air quality management strategy. They can be grouped, from a technical and model oriented point of view into:

    • Major industrial point sources (where stack parameters like height, diameter, emission temperature and speed play an important role for the computation of virtual stack height)
    • Small point sources (e.g., from small industrial sources, commercial sources, and domestic sources such as block heating plants); depending on their size compared to the model domain, they can be treated as point sources proper or grouped into area sources;
    • Area sources such as the thousands of individual chimneys in a city, light industry districts; a special case here are airports, which can be treated as either area sources of even volume sources, depending on their size and relative importance. Another group of area sources cover non- pyrogenic emissions such as VOCs from the transportation system, industry, commerce, and households, biogenic emissions related to landuse, and entrainment areas for particulates;
    • Line sources, which are basically the arcs of the transportation network.

    Figure 2,3: Emission inventories for point and area sources

    From an administrative point of view, these sources are covered by different regulations, different permit and monitoring schemes, and will thus require different data base structures and analysis tools (Figures 2,3). In addition to the basic emission characteristics including their variability over time, technological and economic information on alternative and emission reduction technologies are required for source control optimisation.

    Monitoring data integration

    Observation data are used for several important purposes:

    • To monitor the state of the atmospheric environment and in particular, any possible violation of air quality standards (e.g., 92/86/EC and daughter Directives);
    • To analyse trends as one possible method for forecasting;
    • To provide initial and boundary conditions for simulation models;
    • For comparison with model results to establish their validity and possible us the observations for calibration purposes.

    The latter use of monitoring data, however, involves numerous complications and potential pitfalls as the spatial and temporal scales of monitoring and modeling are very different, and a naive direct comparison can be very misleading.

    Figures 4,5: Monitoring data analysis

    To use the monitoring data in a decision support system effectively, they must be integrated in a real-time manner, with direct links to the monitoring network. At the same time, functionality for time series analysis and spatial analysis is required (Figures 4,5), as well as the possibility to automatically use the observation data together with the simulation models .

    The component models

    The basic concept of the approach described above is its modular structure and flexibility; several simulation models can be integrated to provide a range of tools for different tasks. This provides the flexibility to use the most appropriate model for any task, depending on the nature of the problem (short or long-term, dynamic or steady state, planning or operational control), the pollutants of concern (conservative, simple chemistry, transformation and decay or photochemical), the terrain and meteorological conditions (simple or highly structured, sea breezes), the necessary spatial resolution, etc.

    The levels supported include:

    • A screening level with classical Gaussian steady-state models such as ISC3/AERMOD or photochemical box models or a multi-puff model for dynamic problems;
    • A forecasting level defined by the requirement of computational performance at least an order of magnitude better-than-real-time, including multi-layer dynamic Eulerian models and Lagrangian models based on diagnostic wind models;
    • A planning level including the use of fully 3D dynamic photochemical models;
    • A decision support level that requires embedding the models into an optimisation framework: this can be done directly for simple steady-state models with linear emission - immission characteristics, or will require a Monte-Carlo approach with discrete mutli-criteria optimisation Zhao et al.[20], Majchrazak[9].

    The models generating emission data include energy models such as MARKAL (Wene and Ryden[18]) and traffic simulation models such as EMME/2 (Hearn and Florian[6]) or DYNEMO Schwerdtfeger [17]).

    Application examples

    The projects ECOSIM, AIDAIR and SIMTRAP all address aspects of environmental management, the interactions between technology and its impact on the human living conditions. ECOSIM, an environmental telematics projects, is designed to support urban environmental management, integrating on-line monitoring with simulation modeling for both strategic planning and operational management questions; the environmental domains include air quality including photochemical smog (ozone), coastal water quality, and groundwater quality. ECOSIM has validation sites in Berlin, Germany; Athens, Greece; and Gdansk, Poland.:

    As any other large urban conglomerates, cities like Berlin or Athens are observing episodes of summer smog. Different models, but the same software framework and client-server architecture was used for air quality simulation in these two cities. Ozone forecasting for Berlin was attempted in two of the projects presented, and based on two different approaches: in ECOSIM ( this based on static emission matrices which are scaled to temporal patterns and using a multi-layer forecasting model, REGOZON, with the possibility of a direct comparison of the model forecasts with the on-line monitoring data from the BLUME monitoring network (Mieth et al. [9,10]. In SIMTRAP, this is based on the direct coupling of the dynamic traffic model DYNEMO with the DYMOS 3D ozone model (Schmidt and Haenisch [13]).

    An alternative approach using different models within the same software framework and client-server architecture was used for a case study in Athens. Whereas REGOZON is a combined mesoscale meteorological and dispersion model, MUSE is a dispersion model only, requiring the meteorological quantities usually computed by MEMO (Flassak and Mussiopoulos[5] ; Mussiopoulos[11,12,13] ; Mossiopoulos et al.[14]. The non-hydrostatic prognostic mesoscale model MEMO (Kunz and Moussiopoulos[7]) is a basic constituent of the European Zooming Model (EZM, previously called EUMAC Zooming Model). The EZM represents one of the most widely used European air quality model systems for urban scale applications (about 15 study cases in the last three years).

    Figures 6,7: Emission matrices for REGOZON, and comparison of results with the Berlin monitoring network data.

    MEMO solves the conservation equation for mass, momentum and several scalar quantities in terrain-influenced coordinates. Non-equidistant grid spacing is allowed in all directions. The numerical solution is based on second- order discretization applied on a staggered grid.

    AIDAIR, a EUREKA EUROENVIRON project, has similar aims, but concentrates on air quality assessment and management, within the new air quality framework Directive 96/62/EC as the guiding principle. Here the emphasis is on the integration of different sources of air pollution: industry, domestic sources, and the transportation system, and the energy system as the general framework describing them (Fedra[2]). Linking models for energy planning and optimisation with air quality simulation is one of the objectives. Case studies include Vienna, Geneva (Fedra et al[4]), and Izmir. A typical application is an environmental impact assessment for a district heating scheme in Vienna: the scenarios with and without the scheme involving both an additional block at a power plant and on the corresponding elimination of a large number of individual heating systems in several areas of the city are compared (Figure 8,9) and the changes caused by the project computed and displayed for several alternative scenarios.

    Figures 8,9: impact assessment for a distrcit heating scheme.

    Another application is the estimation of traffic generated urban scale or street level pollution, based on the results of steady- state traffic equilibrium models (Figure 10,11).

    SIMTRAP, an Esprit HPCN (High-performance Computing and Networking) project specifically addresses transportation and air quality, linking dynamic traffic simulation models with 3D, dynamic photochemical air quality models. SIMTRAP application sites include Milano, Italy; Vienna, Austria; Berlin, Germany, and Maastricht, the Netherlands.

    Figures 10,11: traffic generated pollution at city and individual street level.

    SIMTRAP ( integrates the dynamic traffic flow model DYNEMO and the 3D dynamic photochemical model system DYMOS (Schmidt et al.[16]). The traffic model DYNEMO is a simulation tool for road networks, treating individual cars (up to 100,000) as the unit of traffic flow. Car movements, however, are governed by average traffic densities on the individual links of the network. From the mean speed of the cars, and the fleet composition, their emissions are computed and form the dynamic input into the air quality model together with background emission from households, industry, and biogenic emissions.

    DYMOS consists of three meteorology/transport models and one air chemistry model for the calculation of photochemical oxidants like ozone. The meteorology/transport models include REWIMET - a hydrostatic mesoscale Eulerian model with a low vertical resolution, GESIMA - a non-hydrostatic mesoscale Eulerian model with a high vertical resolution, and one Lagrangian model. The air chemistry model is CBM-VI dealing with 34 species in 82 reaction equations for simulating the photochemical processes in the lower atmosphere.

    Air pollution simulations require extremely large amounts of computing time. In order to make the results of case studies available to users within an acceptable period of time or to enable a smog prediction to be made at all (computing time less than simulation period), the DYMOS system was parallelised (Schmidt and Haenisch[10]). A message-passing version was developed from the sequential program and implemented on various parallel computers.

    System Integration and Decision Support

    From the viewpoint of computer implementation, all the above models, their conceptual differences and range of applications notwithstanding, share basic characteristics that allows the designer to develop a common, generic interface: As model input, they use vectors of parameters, time series, or matrices and vector fields, the latter two also as time series or vertical slices of a three dimensional discretisation. As output, the models generate again time series of scalar variables (photochemical box model), spatial matrices (steady state Gauss models), time series and sets of matrices for different parameters and different vertical layers (all dynamic models). This structural similarity makes the design of a generic client-server interface, but also of a generic scenario editing and results display possible. Simulation models can therefor be integrated rather easily, depending on the specific requirements of individual applications.

    From a decision support point of view, it is important that these features are presented to the user in a problem-oriented, not in a tool-oriented way: the system must represent the administrative, regulatory, technological, and economic features of the system, including their socio-political boundary conditions, to be truly useful and usable. Besides, while technical detail and scientific precision are essential, simple and easy to understand and communicate user interfaces and intuitively understandable formats are important where direct inputs to the policy and decision making process are the ultimate goal. Providing publicly accessible environmental information over the Internet, including dynamic model and monitoring results, is one such emerging application. In summary, for an environmental decision support system, good models are a necessary, but not a sufficient condition for a successful implementation.


    [1] Fedra, K., ``Distributed Models and Embedded GIS: Strategies and Case Studies of Integration.'' In: Goodchild, M.F., Steyart, L.Y., Parks, B.O., Johnston, C., Maidment, D., Crane, M. and Glendinning, S. [Eds.], GIS and Environmental Modeling: Progress and Research Issues, pp. 413-417. GIS World Books, Fort Collins, CO. 1996.
    [2] Fedra, K. (1999) Urban Environmental Management: Monitoring, GIS and Modeling. Computer, Environment and Urban Systems (in print).
    [3] Fedra, K., Greppin, H., Haurie, A., Hussy, C., Dao, Hy, and Kanala, R. ``GENIE: An Integrated Environmental Information and Decision Support System for Geneva. Part I: Air Quality.'' pp 247-263. Arch.Sci.Gen‚ve, Vol. 49, Fasc.3, 1996.
    [4] Fedra, K and Haurie, A. (1999) A Decision Support System for Air Quality Management: Combining GIS and Optimization Techniques. Int. Journal Environment and Pollution, Vol. 12 Nos.1/2.
    [5] Flassak Th. and Moussiopoulos N., Direct solution of the Helmholtz equation using Fourier analysis on the CYBER 205, Environ. Software, 3, pp. 12-16, 1998.
    [6] Hearn, D. and Florian, M; Network equilibrium models and algorithms, Network routing, Handbooks in Operations Research and Management Science, Vol. 8, North-Holland, Amsterdam, 1996
    [7] Kunz R. and Moussiopoulos N., Simulation of the wind field in Athens using refined boundary conditions, Atmospheric Environment, 29, pp. 3575-3591, 1995.
    [8] Majchrazak, J. ``DISCRETE: an interactive decision support system for discrete alternatives multicriteria problems''. WP-88- 113, 20 pp, International Institute for Applied Systems Analysis, A-2361 Laxenburg, Austria 1988.
    [9] Mieth, P., Unger, S., and Jugel, M.L., An Environmental Simulation and Monitoring System for Urban Areas, Transactions of the Society for Computer Simulation International, Vol. 15, 3, pp. 115-121, 1998.
    [10] Mieth, P.,Unger, S. and Sydow, A , "Scenario Analysis of a Summer Smog Episode in Berlin", Proc. Second International Conference on Air Pollution, September 27 - 29, 1994, Barcelona, Spain, 1994.
    [11] Moussiopoulos N., An efficient scheme to calculate radiative transfer in mesoscale models, Environ. Software, 2, pp. 172-191, 1987.
    [12] Moussiopoulos N., ed., The EUMAC Zooming Model, EUROTRAC Special Publication, ISS, Garmisch- Partenkirchen, 1994.
    [13] Moussiopoulos N., The EUMAC Zooming Model, a tool for local-to-regional air quality studies, Meteorol. Atmos. Phys., 57, pp. 115-133, 1995.
    [14] Moussiopoulos N., Sahm P., Gikas A., Karagiannidis A., Karatzas, K. and Papalexiou S., Analysis of air pollutant transport in the Athens basin and in the Spata area with a three- dimensional dispersion model, in Air Pollution III, Vol 3: Urban Pollution (N. Moussiopoulos, H. Power and C.A. Brebbia, eds) Computational Mechanics Publications, Southampton, pp. 141-152, 1995.
    [15] Schmidt, M. and Haenisch, R., "Implementation of an air pollution transport model on parallel hardware", Proc. International Conference on Massively Parallel Processing, June 21 - 23, 1994, Delft, The Netherlands, 1994.
    [16] Schmidt, M., Sch„fer, R.-P, and N”kel, K., SIMTRAP: Simulation of Traffic-Induced Air Pollution. Transaction of the Society for Computer Simulation International Vol. 15, 3, pp. 122-132, 1998.
    [17] Schwerdtfeger, T., DYNEMO: A model for the Simulation of Traffic Flow in Motorway Networks. Transportation and Traffic Theory, VNU Science Press, pp. 65-87, 1984.
    [18] Wene, C.-O. and Ryden, B., A comprehensive energy model in the municipal energy planning process, European Journal of Operational Research, 33, pp. 212-222, 1988.
    [19] Zannetti, P. ``Air pollution modeling'' Theories, computational methods, and available software. Van Nostrand Reynolds, New York, 1990.
    [20] Zhao, Ch., L. Winkelbauer, and K. Fedra, ``Advanced Decision-oriented Software for the Management of Hazardous Substances''. Part VI. The Interactive Decision-Support Module. CP-85-50, International Institute for Applied Systems Analysis, A-2361 Laxenburg, Austria 1985

© Copyright 1995-2019 by:   ESS   Environmental Software and Services GmbH AUSTRIA | print page