Information Fusion for Computational Assessment of ... - IngentaConnect

04-906.qxd 1/2/04 11:15 AM Page 235

Information Fusion for Computational Assessment of Air Quality and Health Effects Dimosthenis A. Sarigiannis, Nikolaos A. Soulakellis, and Nicolas I. Sifakis

Abstract The functional layout, the operational features, and the characteristic results of the information fusion method ICAROS NET are presented. This is an innovative technique for the assessment of air quality and the related potential health effects at urban and regional scales. It is based on the multilayer fusion of environmental and epidemiological data and models aiming at reducing the error inherent in environmental measurements and their statistical interpretation. ICAROS NET exploits to the fullest the information potential of Earth observation data, atmospheric chemical and transport models, and ground-based measurements. The assimilation of information from all three data sources into an optimized computational model allows the estimation of tropospheric particulate loading at very high precision and very high spatial resolution.

Introduction Recent studies worldwide have revealed the relation between urban air pollution—particularly fine aerosols—and human health (Pope et al., 1991; Schwartz et al., 1996; Goldsmith and Kobzik, 1999; Roemer et al., 2000; Medina et al., 2001; Brunekreef and Holgate, 2002). This, in turn, has created a pressing request from both environmental scientists and decisionmakers for spatial, timely, and comparable information on air pollution and associated indicators (WHO, 2000). The exploitation of high spatial resolution (HSR) Earth Observation (EO) satellite data acquired over urban areas during air pollution episodes could provide us with spatial, timely, and reliable information on air pollution (Kaufman et al., 2002) and associated indicators such as aerosol optical thickness (AOT) (Sifakis et al., 1998). Various algorithms for urban air pollution assessment by means of HSR satellite imagery have already been developed and applied to urban sites in Europe such as Athens, Greece and Brescia, Italy (e.g., Sarigiannis et al., 1998; Sifakis and Soulakellis, 2000; Sarigiannis et al., 2002). High spatial resolution (HSR) sensors that provide information for local detection are only slightly dynamic regarding the geographic dispersion of pollutants because the satellite pictures are taken at an interval of a few days, ranging typically from 1 to 2 days to 2 weeks. However, this relatively D.A. Sarigiannis is with the Joint Research Centre—European Commission, 8, Square de Meeûs, B-1050 Brussels, Belgium ([email protected]).

“static” spatial measurement can complement and link the two tools most widely used for the study of air pollution transport to date, namely, dispersion/transport modeling and analytical observations. Dispersion/transport modeling provides dynamic spatial information but it is heavily based on initial model assumptions. Analytical observations (mainly ground-based) may produce an extensive time series of point data but are “space-deficient” unless taken from an extremely large number of costly stations. Furthermore, due to cost considerations, the physical deployment of ground-based monitoring stations does not keep up with the pace of land-use change in most urban settings. Attempts to overcome these problems with advanced spatial statistics introduce computational error, which increases exponentially with the increasing demands for spatial precision. To date, the state of the art in air quality information systems comprises only information and data processing tools using data from ground-based measurements (produced in the context of established monitoring networks or ad hoc campaigns) and atmospheric modeling (i.e., models of meteorological parameters, transport, and chemical transformation of pollutants in the atmosphere). The main weaknesses of this approach are (1) the need to introduce linearity assumptions in the spatial correlation of information from neighboring data points (for spatial interpolation of ground-based data) and (2) the fact that numerical modeling is based on operational assumptions concerning the initial composition of ambient air and the actual emissions inventory. Current-day experience shows that, in most cities and urban conglomerates worldwide, maintaining an up-to-date and comprehensive emissions inventory is a difficult and costly exercise. Furthermore, the generation of high spatial resolution results from numerical models covering large geographic areas is computationally cumbersome and poses extreme demands on the information technology infrastructure usually available to regulatory bodies. Satellite-derived data can be used to bridge the gap between models simulating the transport and chemical transformation of ambient air pollutants on the one hand, and analytical observations on the other. Data obtained by HSR satellites at the time of their daily overpass offer a unique possibility to benchmark and calibrate the state-of-the-art simulation models. These can be assisted by the use of inference engines to identify emerging meta-knowledge regarding the dynamics of the interacting socio-economic and environmental systems, which determine urban air quality (Sarigiannis, 1999). Integration of

N.A. Soulakellis is with the Department of Geography, University of the Aegean, University Hill, GR-81100 Mytilene, Greece ([email protected]).

Photogrammetric Engineering & Remote Sensing Vol. 70, No. 2, February 2004, pp. 235–245.

N.I. Sifakis is with the National Observatory of Athens, Institute for Remote Sensing and Space Applications, Penteli, Greece ([email protected]).

0099-1112/04/7002–0235/$3.00/0 © 2004 American Society for Photogrammetry and Remote Sensing

P H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G

February 2004

235

04-906.qxd 1/2/04 11:15 AM Page 236

the AOT values measured by satellite sensors with other types of data stemming from ground-based air quality measurements and pollution dispersion models provides a coherent depiction of air pollution distribution over a large geographic area at a spatial resolution of 500 m. Thus, in order to validate the AOT values, derived by Landsat 5 TM and SPOT XS satellite systems, measurement campaigns have been scheduled and performed during the time of satellite image acquisition in the framework of the ICAROS project (Schäfer et al., 2002). The results obtained were inviting (Sarigiannis et al., 2002; Sarigiannis et al., 2003a), and they supported the idea of developing an object-oriented computational platform for integrating the variety of information classes affecting decision making on environmental problems and related health concerns (Sarigiannis, 2002). This platform is mainly addressed to environmental and transport policy and decision makers at the urban and regional levels.

Methodology The purpose of the work presented herein was the development and implementation of the ICAROS NET networked interactive computational environment that allows the integration of multisource and multiscale datasets for monitoring air pollution and its effect on health, paying particular attention to the spatial variation in deriving exposure-response relationships. ICAROS NET is based on

• • • •

Integration and assimilation of different environmental data types, including data coming from satellite-based remote sensing, ground-based air quality measurements, and advanced atmospheric modeling; Emergence of meta-knowledge through the use of inference techniques for systemic data validation and forecasting; Sharing of the knowledge acquired through environmental data fusion among urban policy makers and concerned stakeholders using current and future telematics infrastructure; and Decision support for air quality management at the urban and trans-boundary scales based on multicriteria analysis and interactive optimization.

The optical atmospheric effects of pollution on HSR Earth Observation (EO) data are more pronounced in certain spectral bands than in others; this permits a first delineation of polluted areas and localization of emission sources, through computer assisted photointerpretation of satellite imagery. The AOT of aerosol scattering in the green visible spectrum (i.e., at 550 nm) is used as a surrogate for pollution loading. The method applied for the evaluation of aerosol optical thickness combines two approaches that consider physically independent optical effects in the atmosphere (Sifakis et al., 1998): (1) contrast reduction by scattering efficient airborne particles in short wavelengths (i.e., visible spectrum), and (2) radiation attenuation that particles engender in longer wavelengths (i.e., thermal infrared spectrum). The profiles of aerosol optical thickness () derived from Earth Observation refer to the total atmospheric column. It is reasonable to assume, however, that the majority of the pollutants of interest for air quality assessment remain within an atmospheric layer that spans from the ground up to the mixing height in the atmosphere. Mixing height is calculated from meteorological data, based either on observation, or on meteorological models. By dividing the value of calculated from the satellite signal with the mixing layer height, is corrected to reflect the relative importance of lower atmospheric aerosol, i.e., the one of greatest relevance vis-à-vis adverse health effects. In this way, the scattering coefficient of fine particles gathered in the lower troposphere, i.e., within the boundary layer, is calculated according to the equation bsp hmix. 236

February 2004

(1)

Ground-based air quality measurements (coming from the fixed monitoring network and/or from ad hoc experimental campaigns) are stored in an air quality database. These data serve as input to a chemical model used for the transformation of primary pollutants such as NOx and SO2 into secondary aerosol. The model predicts the quantity and composition of secondary atmospheric aerosols containing sulfate, nitrate, and ammonium salts. A thermodynamic approach is adopted to predict the chemical composition of multiphase aerosols. Through this process and measurements of primary PM10 and PM2.5, the amount of atmospheric aerosol and the chemical species comprising it are calculated. The scattering coefficient of the suspended fine particles is then statistically correlated to the mass of the chemical species comprising the tropospheric aerosol to produce aerosol maps of the area of interest. Fusion of the air quality information stemming from the fusion of EO and ground data and the output of the atmospheric models gives corrected maps of surface concentration of pollutants over the whole terrain of interest. The fusion function proposed here is a weighted average of the respective values of fine particle concentration (at ground level) as calculated by the EO-derived map (wEO) on the one hand, and by the corresponding chemical transport model (wm) on the other hand. This function is applied in the domain of interest on a cell-by-cell basis. The weights used vary from 0 to 1 depending on whether the EO-derived or the model-based calculation is more reliable at the cell in question. A typical reason for assigning a value higher than 0.5 to the weight of the atmospheric model output can be abrupt changes in surface albedo (e.g., in the case of permanent snowsheds or at the coastline of water bodies). Inversely, the value of the weight of the particulate concentration calculated from the fusion of EO with ground-based measurements (wEO) becomes 1 at the loci of the ground stations and decreases to 1 wm as the distance from them increases. Obviously, the following relationship must hold at each cell in the domain of analysis: wEO wm 1.

(2)

Based on the corrected pollutant maps, the emission factors used to reckon the initial emission inventory could be calibrated as appropriate. Introducing the final aerosol concentration values in exposure-response relations obtained from epidemiological investigations, and maps of health indicators, such as expected morbidity from particulate loading, can be calculated from the aerosol maps. Fusing these maps with density maps of vulnerable population, health vulnerability and risk maps are produced (Sarigiannis et al., 2003b). Figure 1 shows the information flow in the ICAROS NET fusion algorithm.

System Architecture The information fusion processes at the heart of ICAROS NET exploit the capabilities offered by geographic information systems (GIS) and object-oriented databases for the assimilation of the three data sources (ground-based analytical measurements, Earth observation, and atmospheric modeling). Georeferenced ancillary information such as land use and land cover, road and utility networks, siting of key vulnerable pollution receptors, population density, and urban and regional planning maps are also imported into the computational environment. Environmental and health-related information sources are integrated into a single environmental information processing tool, which can be used for pollution monitoring, extreme incident forecasting, and strategic environmental assessment in the urban environment. ICAROS NET is a multimodular software, developed after long consultation with the identified end users in order to meet P H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G

04-906.qxd 1/2/04 11:15 AM Page 237

their real needs and requirements for advanced air quality monitoring and public health impact assessment. The entire computational environment is object-oriented in order to ensure efficient management of multivariate information classes and easy upgrading of the software in subsequent releases (in order to readily address the evolving needs of end-users). The architecture of the ICAROS NET system is depicted in Figure 2. It is a three-tier structure articulated as follows: 1. The ICAROS NET applications, each of which serves the needs of different users by implementing the functional characteristics required for the specific role of each end-user community (research institutes, decision makers, public, etc), through suitable user interfaces that are custom-made to fit the needs of and be understandable to each end-user community. All these applications access the same underlying georeferenced database through an application server. The applications will run on any workstation with Internet access. 2. The “ICAROS-NET” server, which provides basic services to all ICAROS-NET applications in the form of Application Programming Interfaces (APIs) and web services. The application server has mainly three tasks:

Figure 1. Flowchart of the ICAROS NET information fusion algorithm.

•

To handle user access to the system and manage access rights regarding management, data extraction, and updating of the georeferenced database.

Figure 2. ICAROS NET tiered system architecture (color version at www.asprs.org).


February 2004

237

04-906.qxd 1/2/04 11:15 AM Page 238

• •

To implement the functionalities required in common by the ICAROS-NET applications, so that this common set is centrally managed and maintained. To manage all request traffic for database access to and from the various applications.

3. The third layer consists of the Central Geographical Database, which stores and manages all classes of underlying georeferenced information (pollutant concentrations, atmospheric modeling results, 3D terrain models, satellite images, temporal information, statistical data, population density, etc.).

The six application modules composing Layer 1 of the system are the following:

ICAROS-NET

• •

•

•

•

•

238

Module 1. ICAROS-NET Database Management: enables the creation, updating and validation of the database required for handling all air quality, health-related, and ancillary information in both geographical and tabular formats. Module 2. Aerosol Optical Thickness Calculation: enables the calculation of the Aerosol Optical Thickness values by means of high- and very-high-resolution satellite images (e.g., data derived from sensors on board Landsat, SPOT, IRS, MODIS, and CHRIS satellites). Module 3. Atmospheric Modeling: enables the calculation of the horizontal and vertical distribution of atmospheric pollutants based on the emissions inventory and the wind profile as calculated by the meteorological profile sub-module (including wind rose, and other meteorological parameters as appropriate). This module encompasses publicly available Gaussian plume models such as CTDM used widely by regulatory agencies. Other atmospheric modeling software, which may be available at the end-user site, such as CALPUFF, the Urban Air shed Model (UAM) (for estimation of gaseous pollutant concentrations), and REMSAD (for estimation of tropospheric aerosol), can be readily incorporated into the ICAROS NET platform through an open-ended I/O architecture. Module 4. Data Fusion: enables the integration and fusion of the available information to estimate air pollution concentrations based on Aerosol Optical Thickness—values reckoned from EO data/images (first layer of information fusion). It further enables fusion between the Earth observation-derived concentrations of aerosol with pollutant profiles calculated using atmospheric models (second layer of information fusion). Module 5. Health Impact Assessment: enables the further fusion of ancillary health-related information (like georeferenced population density and exposure data and doseresponse functions) in order to allow the estimation of health risk associated with the spatially resolved exposure to airborne chemicals in fine and ultra-fine particle form (third layer of information fusion). The output of this module is health risk and health-related air quality maps, which can be used either for strategic planning for sustainable urban and conurban development, or for assessing and forecasting extreme atmospheric pollution incidents. Module 6. Network Optimisation: enables the optimization of the configuration (both in terms of spatial distribution and typology) of ground-based air quality monitoring networks taking advantage of the information derived from the Earth observation data and its interpretation (Saisana et al., 2001). This is a multicriteria decision-support system, which takes into consideration the different needs of various stakeholders associated with or affected by air quality management at the urban and/or regional levels. It is empowered with easy visualization interfaces, which allow the user-friendly use of the software demonstrating visually the possible trade-offs among different types of monitoring networks in order to allow the final decision maker (user) to take into account the optimal set of solutions rather than a single optimal one. Finally, the ICAROS-NET platform offers two interactive visualization tools: (a) An Internal Visualization Tool, for visualizing all the available information; and (b) An Internet Map Server, for making this information available to the public through the Internet.

February 2004

System Functionality Object-Oriented Geo-Database Design and Development This database is the key pillar of the ICAROS-NET platform because it enables the effective management of the information required for the operation of all the remaining modules in a user-friendly manner. The main functional difficulty in the ICAROS NET system is the requirement to use simultaneously a multitude of data classes with and without georeferencing and time dependence. The spatial resolution and accuracy of the data may differ as well, while all output calculations need to be set in the same type of georeferenced grid at the highest possible spatial resolution. Finally, the origin of the data and the respective nominal accuracy of the source is a fundamental piece of meta-information for efficient optimization of the algorithm and minimization of the overall error. Object-oriented database technology provides additional advantages over conventional relational databases, because it allows for clustering of different data types, identification of meta-information (e.g., identification of the source of the information), and grouping of model output with input data for each application type. Several software tools, i.e., Visio and Unified Model Language (UML), were used to build the object-oriented geographical database aiming at serving all the data handling needs of the ICAROS NET computational modules. The database needs to be adapted to the specific needs and data availability of each application site. The predefined database scheme is being installed on the ICAROS-NET platform server during software set-up. Via a user-friendly interface (Figure 3), the platform offers to the user the following functionalities: (i)

Import: Enables importation into the ICAROS-NET Databases: a. air quality measurements from ground-based monitoring stations in a predefined ASCII format, b. atmospheric modeling results in a predefined ASCII format, c. satellite images in raw format, and d. geographical data (i.e., road and rail networks, digital elevation models (DEM), etc.) in GIS format (coverages, shapefiles, or grids).

Figure 3. A user-friendly interface assists the end-user to readily update the object-oriented databases of the ICAROS-NET platform.


04-906.qxd 1/2/04 11:15 AM Page 239

During the import process data values are being validated according to the predefined database scheme. (ii) Compact: Enables compaction and re-indexing the ICAROSNET database. This process is taking place after importing a new dataset, without end-user intervention, and it optimizes use of memory space and data accessibility. (iii) Database status reporting: Creates reports on all the available information related to the ICAROS-NET databases, i.e., directories, file names, file sizes, metadata, time and stored measurement information, etc. These reports can be provided to the user upon request or automatically after every import procedure, according to the specific system requirements.

During software development, special attention has been paid to enhancing the interoperability and communication between ICAROS-NET and other common data sources, which provide information about air quality in a variety of formats, i.e., ground-based monitoring networks, to provide information on pollutant concentrations in DBF, ASCII, or XLS format; atmospheric modeling software, to provide information on both meteorological and air pollutant concentration in ASCII format; and Earth Observation, to provide satellite data mainly in raw format. In some cases, however, satellite images can be provided to the end-user in a more common format such as IMG or GEOTIFF. During data import, the ICAROS-NET platform also performs the geometric correction of the satellite data based on a predefined set of control points. This minimal control point set needs to be predefined and is characteristic of each application site. Geographical information such as digital elevation models, land cover, road network, population density, etc., are mainly provided in vector or raster GIS format, i.e., point features, line features, polygon features, and grids. Earth Observation Data Processing This module enables the calculation of Aerosol Optical Thickness (AOT) by processing high spatial resolution satellite images, i.e., Landsat, SPOT, and IRS/LISS. This information is crucial because it provides the EO-derived environmental information, which feeds into the overall fusion processes of ICAROS NET. It has been developed based on the existing DTA, SMA (Sifakis and Deschamps, 1992), and SIPHA codes (Sifakis and Soulakellis, 2000). This module is fully automated. An ad hoc software agent periodically searches the EO-data input directory. When it identifies a new datafile, it opens and characterizes it by reading the header information, and calls the appropriate image processing algorithm. The results are automatically stored in the EO output directory and a flag message appears on the user interface. This module (Figure 4) offers to the user the following functionalities:

• • •

Radiometric Correction: Performs the radiometric correction of the satellite data by reading all the required parameters, i.e., gain and offset values, sun elevation angle, day of the year, etc., from the leader files of the images. AOT Calculation: Calculates by applying the appropriate algorithm to the radiometrically corrected satellite images. This process is taking place after the radiometric correction of a new image, without the interaction of the end-user. AOT-Database reporting: Creates a report of the available Aerosol Optical Thickness, i.e., directories, file names, file sizes, metadata, time, etc. This report is provided to the user: (1) upon a request for the full report or (2) automatically after every new calculation.

Atmospheric Modeling The atmospheric modeling module allows the user to perform the calculations necessary for the estimation of primary and secondary pollutants concentrations in the domain of interest, both horizontally and vertically. It also performs meteorological calculations, such as estimating wind speed and direction, relative and absolute humidity, ambient air temperature, and


Figure 4. A user-friendly interface allows the user to calculate Aerosol Optical Thickness from satellite-derived information.

atmospheric mixing height. The necessary input for these calculations is taken from the georeferenced ICAROS NET database. Such input includes meteorological data (including horizontal and vertical wind profiles), digital elevation models of the modeling terrain, emissions inventory (in grid format), and initial concentrations of primary pollutants. The output is a set of estimated georeferenced atmospheric data, including the height of the well-mixed lower layer of the troposphere (mixing height), and the horizontal (and vertical if needed) concentrations of primary and secondary pollutants. Special emphasis is given in the calculation of human-made tropospheric aerosol. The software provided in the ICAROS NET platform is based on the CTDM regulatory model (used by the U.S. Environmental Protection Agency, EPA) and its enhancements. The model has been modified to run in a fully user-friendly and WYSIWYG (“what you see is what you get”) environment in order to facilitate its use. The ICAROS NET modified software allows the user to incorporate existing DEMs or to create his/her own by using the graphical capabilities of Arc/GIS. Equally, the user interface developed in the realm of ICAROS NET eliminates the cumbersome transfer of data between different software structures and formats by integrating all input/output (I/O) operations into a unique common GIS-based environment for the graphic visualization or the statistical treatment of information. In addition, the CTDM implementation in ICAROS NET allows the full exploitation of GIS capabilities in treating metadata, and not only final pollutant concentrations or deposition rates. Finally, the atmospheric model implemented in the ICAROS NET platform allows for the direct integration with other relevant environmental features and gives the users the possibility to easily visualize 3D and dynamic pollution and air quality management scenarios. The atmospheric modeling module of the ICAROS NET software platform is geared with a user-friendly, “wizard”-like interface, which guides the user through the necessary steps for running the CTDM model. Most of the file specifications are already hardwired in the stand-alone version of the platform for each application site. This renders the whole operation of

February 2004

239

04-906.qxd 1/2/04 11:15 AM Page 240

the model very user-friendly. The I/O files are all put in ASCII format to facilitate communication with the other platform modules and the easy integration into the GIS environment. Information Fusion The most important module of the ICAROS-NET platform enables the integration and fusion of the information derived from ground measurements, atmospheric modeling, and satellite images in order to provide (1) estimates of the particle scattering coefficient; (2) pollutant concentrations based on the values derived from the satellite images, and exposure maps based on epidemiological exposure research; and (3) health vulnerability maps based on population density data and pollutant concentrations. This module offers the following functionalities:

•

•

•

•

AOT Normalization: Performs the normalization of values derived by Module-2 (AOT calculator), taking into consideration the mixing height derived by atmospheric modeling or radiosoundings. In this way, the particulate scattering coefficient is calculated in each cell of the domain of interest. Pollutant Estimation Based on AOT Values: Performs the calculation of pollutant concentrations, i.e., particulate matter in the form of PM2.5, PM10, etc., based on the scattering coefficient values derived through the process above. This process is taking place automatically after AOT normalization. Fused Pollutant Maps: Performs the generation of fused pollutant maps based on (1) the maps derived at the previous stage and (2) the equivalent information derived by atmospheric modeling. This process takes into consideration intelligence regarding the main features of the landscape in the domain of interest in order to modulate the decision functions used for fusing the estimates of tropospheric particulate loading from Earth observation, ground measurements, and chemical transport models. Fusion-Database Reporting: Creates a report of the available scattering coefficient maps, pollutant maps, and fused pollutant maps. These report is provided to the user (1) upon a request for the full report or (2) automatically after every new calculation.

Health Impact Assessment This module enables the integration of the information derived from (1) chemical transport models and (2) fusion of Earth observation with ground-based data with population density and dose-response functions (derived via epidemiological studies and stored in the ICAROS NET database) in order to provide health vulnerability maps. Most of the exposure-response relationships are related to a certain risk group, such as children, elderly, or asthmatics. To quantify effects within a certain risk group, the share of the risk group in the total population needs to be known. Exposureresponse functions for children are applied to the population that is less than 15 years of age. Note that the mean European estimates for the prevalence of asthma among adults is 7 percent. The population-weighted increment of the concentration is defined as n ci pi Cpop p i1

total

(3)

where ci is the concentration in grid cell i, pi is the vulnerable population in grid cell i, and ptotal is the total population within the reference environment. This index is used as an indicator of vulnerability of the local population to key effects of particulate pollution. The impact assessment procedure adopted is based on exposureresponse functions derived from a large number of epidemiological studies (Goldsmith and Kobzik, 1999; Schwartz and Neas, 2000; Medina et al., 2001). For ease of implementation, the key relationships were linearized and annualized, assum240

February 2004

ing independence of background. The exposure-morbidity relationship (Schwartz et al., 1996) used here is given by

0.124(low) 0.187(mid) Ipop [PM10] 0.251(high)

(4)

where Ipop is the change in hospital admissions for respiratory infection per 105 of population and [PM10] is the change in annual concentration of PM10 (in g/m3). Anticipated health impact maps based on exposureresponse estimates represent the cumulative societal risk from fine particle pollution in the ambient air. They describe the anticipated impact of human population taking into account both population density and individual exposure patterns, clustered around population groups. Although they are valid for the estimation of health risk from the air quality management point of view, they cannot be used as indicators of individual risk due to exposure to polluted ambient air. In order to fill this gap, the U.S. Environmental Protection Agency guideline for the calculation of a health-related air quality index (AQI) (EPA, 1999) has been implemented in the ICAROS NET platform. We have based the AQI calculation on the reckoned ground-level concentrations of PM10 and PM2.5 according to the following formula: AQI AQIlo AQIhi AQIlo C BPlo BPhi BPlo

(5)

where AQI is the air quality index based on the pollutant of concentration C, C is the rounded concentration of the pollutant of interest, BPhi is the break point that is higher than or equal to C, BPlo is the break point that is lower than or equal to C, AQIhi is the value of the air quality index corresponding to BPhi, and AQIlo is the value of the air quality index corresponding to BPlo. The values of the break points for the AQI function above (Equation 5) are given in Table 1. Solving Equation 5 for AQI, the value of the air quality index can be calculated for every 900-m2 cell of the computation grid in the domain of interest as follows: AQIhi AQIlo AQI (C BPlo) AQIlo. BPhi BPlo

(6)

The values of AQI are then classified according to the EPA categorization given in Table 1 and color-coded to produce maps of the air quality index. These maps can be used as indicators of individual risk, taking into account worst-case and homogeneous exposure scenarios for the whole domain of interest. Graphical User Interface and Result Visualization The effective visualization of information coming from either the ground-based monitoring network or the satellite images is critical for a better understanding of the phenomenon and TABLE 1.

VALUES OF THE BREAK POINTS FOR THE CALCULATION OF THE AQI ACCORDING TO THE EPA CATEGORIZATION OF AIR QUALITY

PM10 (g/m3)

PM2.5 (g/m3)

AQI

Category

0–54 55–154 155–254

0–15.4 15.5–40.4 40.5–65.4

0–50 51–100 101–150

255–354 355–424 425–504 505–604

65.5–150.4 150.5–250.4 250.5–350.4 350.5–500.4

151–200 201–300 301–400 401–500

Good Moderate Unhealthy for sensitive groups Unhealthy Very unhealthy Hazardous


04-906.qxd 1/2/04 11:15 AM Page 241

Figure 5. Internal visualization tool allows the end-user to visualize interactively all the available information deriving from the different ICAROS-NET modules.

consequently for betted management decisions. An interactive visualization tool has been developed for (1) querying the database and selecting data for mapping and (2) interactive cartographic visualization of the selected items. The ICAROSNET platform is equipped with ad hoc tools for interactive visualization in a stand-alone configuration (Figure 5) and through the Internet (Figure 6). These visualization tools offer the following functionalities to the users:

• • • • • •

Zoom in/out to the desirable scale, permits the user to chose the appropriate scale for the representation of the ICAROS-NET datasets; Layer selection for selective visualization, permits the user to select the information desirable for visualization; Layer activation and symbol changing, permits the user to activate a thematic layer and modify the cartographic symbols used; Layer activation and information retrieval, permits the user to activate a thematic layer and interact with the map in order to retrieve the exact value for a desirable point of the surface; Qualitative data visualization, permits the user to apply standard color palettes for visualizing specific geographic information (road network, land cover, FCC-satellite images, etc.); and Quantitative data visualization, permits the user to apply standard classification schemes (natural breaks, quantiles, equal-interval, standard deviation, etc.) to quantitative data (i.e., pollutant concentrations) and then select the most appropriate cartographic symbols (i.e., graduated symbols, changing colors, dot density, etc.).

Results and Discussion The capital of Greece, Athens, is well known for its frequent air pollution episodes. For this reason it has been selected as a pilot area for the first application of the ICAROS-NET system. This application was based on existing satellite data as well as ground measurements provided by the local monitoring network. The main objective of the application of the ICAROS-NET platform to Athens was to evaluate and test the developed platform. In addition, this generated for the first time satelliteP H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G

derived maps of ground concentrations of fine particles at a spatial resolution of 30 m, covering the whole area of the Athens basin (Figure 7). The relationship between aerosol optical thickness and concentrations of particulate matter is given by the following evidence-based Equation 7, which has resulted from the treatment of historical data series on urban air pollution concentrations and their correlation to the values of optical thickness of the tropospheric aerosol as estimated by the Earth observation data processing algorithms: 0.21 ln[ANO3] 0.0015 [AH2O] 0.76 [ASO4] 0.083 ln[ASO4];

(7)

R 0.9680; 0.19 2

The model variables are given in Table 2. All variable values were calculated at the time of the satellite passage. The negative regression coefficient for ln[ASO4] can be explained by the fact that ASO4 takes values between 0 and 1; thus, and ASO4 are positively correlated. Forward stepwise multiple regression produced a parsimonious model that maximizes accuracy with an optimally reduced number of input variables. The dataset for the domain of interest was first checked for the theoretical assumptions of

TABLE 2. Symbol [ANO3] [ASO4] [ANH4] [AH2O] Aerosol

VARIABLES FOR THE AOT-PM CONCENTRATION MULTIPLE REGRESSION MODEL Parameter Description

Units

OUTPUT Aerosol optical thickness value INPUT Concentration of NO3 in the aerosol phase Concentration of SO4 in the aerosol phase Concentration of NH4 in the aerosol phase Concentration of H2O in the aerosol phase Total aerosol concentration

g/m3 g/m3 g/m3 g/m3 g/m3

February 2004

241

04-906.qxd 1/2/04 11:15 AM Page 242

Figure 6. Internet Map Server allows the end-users to serve information to the public.

Figure 7. Map of ambient air concentration of fine particulate matter (in g/m3 of PM10) in Athens on 08 April 1999 (color version at www.asprs.org).

242

February 2004


04-906.qxd 1/2/04 11:15 AM Page 243

regression analysis using the normal probability plot of residuals (observed minus predicted values), the plots of residuals versus the predicted values, and the plots of the predicted variable versus each of the input variables. These tests show that the dataset used satisfies the major assumptions and validity criteria of regression analysis. This correlation was done on a cell-by-cell basis and was based on pollution profiles characteristic of both winterspring and summer-early fall conditions (both in terms of climatic differences and of chemical composition and intensity of particulate and gaseous precursor emissions). The logarithmic form of the model reflects well the expected theoretical relation between optical thickness () and the concentration of the optically active compounds that constitute the secondary aerosol mass in the lower atmosphere. The scattering coefficient of the aerosol, as calculated from multiple linear regression of the same dataset (forced to have a zero intercept), is given by Equation 8: i.e., bsp 7.26(m2gr)CPM10(gm3).

(8)

The fraction of the variance determined by the square of the regression coefficient (R2) was 0.90. The slope of the regression line is higher than reported in other studies in areas characterized by urban aerosol similar to the one in Athens, such as the San Joaquin Valley in California (Richards et al., 1999), probably due to a higher percentage of non-organic aerosol in the ambient air of Athens than in the San Joaquin Valley. An estimation of the total error of the approach followed here for the description of air pollution using optical thickness maps can be given from the following:

• •

The algorithms calculating the aerosol optical thickness from the visible and near-infrared channels of the sensors on board SPOT and Landsat satellites have a standard error on the order of 5 percent (S1 0.05); The chemical model for calculation of formation of sulfate and nitrate salts from SO2 and NO2 has an error of 2 percent (S2 0.02);

• • •

The ARES model for the estimation of aerosol concentrations is a thermodynamic model; therefore, no verification results are available; Spatial interpolation using kriging estimates the temperature and absolute humidity with a standard error on the order of 5 percent (S3 0.05); and The statistical models correlating optical thickness values with the aerosol formed by the measured concentrations of gaseous pollutants has an error on the order of 4 percent (S4 0.04).

Thus, the overall error of the algorithm can be estimated from the following equation: Overall Standard Error 1 {(1 (S1 S2 S3)) (1 S4)} 15.52%.

(9)

In the results shown in this study, historical data on particulate air pollution in Athens were analyzed. Two seasons, characterized by high and low risk, respectively, were selected and the corresponding spatial variation of health risk intensity was assessed. The two seasons in question were spring and summer of 1999, respectively. According to a statistical analysis of air pollution loading time series in Athens, the period of March-April seems to be characterized by increased values of key pollutants such as particulate matter. This is mainly due to the combination of two key factors: good climatic conditions with high sun coverage during the day and still moderate temperatures, which induce continued operation of diesel-based central heating. The net result is both increased production of primary aerosol and increased photochemical activity generating secondary aerosol from gaseous primary pollutants such as NOx, SO2, and VOCs. At the same time, March and April are months with intensive economic activity and Athens seems to reach its population capacity. This feature intensifies the need for urban transport, which, in turn, enhances the related environmental pressure. In August, and in particular around mid-August, the situation is dramatically different, because most of the city residents

Figure 8. Spatially resolved health vulnerability and risk indicators in Athens (health risks attributed to fine particulate pollution in the greater Athens area) (color version at www.asprs.org).


February 2004

243

04-906.qxd 1/2/04 11:15 AM Page 244

Figure 9. Map of the air quality index for the greater Athens area on 08 April 1999 (color version at www.asprs.org).

are on vacation and the intensity of economic activity (including power generation and transport) is at its minimum. The results of the analysis performed using the ICAROS NET information fusion platform corroborate the qualitative description above. Indeed, the health risk estimate in the spring is much more pronounced than the one in mid-summer (Figure 8). Here, color-coding moves from light (very low risk) to dark gray (relatively high risk). Furthermore, the spatial spread of health risk is significantly reduced in August, with only the main downtown area (which maintains a non-negligible level of economic activity and services) being under some pressure. Spatial analysis of the data, however, indicates the presence of hot spots, characteristic of heavy industry sites outside of the urban area but within the metropolitan domain of Athens, and of polluting but continuously intensive activity such as the old site of the Athens international airport. In Figure 9, a map of the air quality index (AQI) is given in the same area of analysis. Note the difference in the spatial pattern between the PM10 concentrations at ground level (Figure 7), the morbidity index shown in Figure 8, and the AQI map. The comparison reveals the qualitative difference in the environmental and health impact information depicted in the various datasets—the more information pertaining to the estimation of the final impact of air pollution on health is put into the information fusion algorithm, the more pertinent to risk evaluation becomes the outcome of the calculation. At the same time, the different spatial patterns demonstrate the need to differentiate between cumulative societal risk and individual health risk in producing environmental health impact estimates. This differentiation is facilitated largely by the integration of spatially resolved and Earth observation-derived information. The brief overview included herein highlights the significance of georeferenced information and spatial analysis in investigating the link between air quality and human health, as well as in devising “intelligent” strategies for reduction of environmental risk to public health.

Conclusions ICAROS NET technology shows great potential for integrated air

quality and health management. It brings together the state of the art in air quality monitoring and assessment, namely, stateof-the-art atmospheric models and analytical in situ measurements. It further encompasses advanced Earth observation data 244

February 2004

processing tools for optimizing the monitoring and assessment capabilities. The information fusion methodology implemented herein functionally integrates all three information sources (Earth observation, modeling, and analytical measurements) and reduces the overall error of the method to about 15 percent. This is a significant improvement over the best atmospheric models and pollutant concentration maps produced by spatial interpolation of measurements from the ground. Based on the increased accuracy of the result of information fusion in terms of air quality characterization and taking advantage of its specially designed geo-database, ICAROS NET provides an equally powerful tool for calculating the relative risk to human health posed by airborne chemicals in fine particle form. Future work will be geared towards checking the interoperability of the platform in different urban and regional settings and the investigation of the spatial differentiation of exposure-response functions, an issue that is key for the detailed characterization of the environmental health risk attributed to fine particle atmospheric pollution. Finally, the integration of datasets from advanced multi- and hyperspectral sensors on board satellite vectors with very high, high, and moderate spatial resolutions will allow the fusion platform to take advantage of the chemical speciation information, which can be provided by such systems. This will permit a synoptic, yet detailed, profiling of the lower atmosphere vis-à-vis anthropogenic particulate matter and its precursor chemicals.

References Brunekreef, B., and S.T. Holgate, 2002. Air pollution and health, Lancet, 360:1233–1242. Goldsmith, C.A., and L. Kobzik, 1999. Particulate air pollution and asthma: A review of epidemiological and biological studies, Review of Environmental Health, 14:121–134. EPA, 1999. Guideline for Reporting of Daily Air Quality—Air Quality Index (AQI), EPA-454/R-99-010, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, 25 p. Kaufman, Y.J., D. Tanré, and O. Boucher, 2002. A satellite view of aerosols in the climate system, Nature, 419:215–223. Medina, S., A. Plasència, L. Artazcoz, P. Quénel, K. Katsouyanni, H.G. Mücke, E. De Saeger, M. Krzyzanowsky, and J. Schwartz, 2001. APHEIS: Monitoring the Effects of Air Pollution on Public Health in Europe, Scientific Report, 1999–2000, Institut de Veille Sanitaire, Saint-Maurice, France, 136 p.


04-906.qxd 1/2/04 11:15 AM Page 245

Pope, C.A., III, D.W. Dockery, J.D. Spengler, and M.E. Raizenne, 1991. Respiratory health and PM10 pollution: A daily time series analysis, American Journal of Respiratory and Critical Care, 144:668–674. Richards, L.W., S.H. Alcorn, C. McDade, T. Couture, E. Lowenthal, J.C. Chow, and J.G. Watson, 1999. Optical properties of the San Joaquin Valley aerosol collected during the 1995 integrated monitoring study, Atmospheric Environment, 33:4787– 4795. Roemer, W., G. Hoek, and B. Brunekreef, 2000. Pollution effects on asthmatic children in Europe, the PEACE study, Journal of Clinical Exposure and Allergy, 30:1067–1075. Saisana, M., D. Sarigiannis, D.A. Chaloulakou, and N. Spyrellis, 2001. Air quality monitoring design: Optimisation of PM2.5 networks using satellite observations, Proceedings, 7th Global Nest Conference on Environmental Science and Technology, 03–06 September, Syros, Greece, pp. 747–754. Sarigiannis, D.A., 1999. Data fusion and inference systems for environmental decision support, Proceedings, 11th American Association of Artificial Intelligence Conference, 21–26 July, Orlando, Florida. ———, D., 2002. Integrated environment and health protection: implementation of the precautionary principle, Fresenius Environmental Bulletin, 11:750–756. Sarigiannis, D.A., D. Assimakopoulos, A. Bonetti, F. Huynh, M. Lointier, K. Schäfer, N.I. Sifakis, N. Soulakelis, and M. Tombrou, 1998. ICAROS—Integrated computational assessment via remote observation system, Proceedings, Annual Conference of the UK Remote Sensing Society, 08–12 September, Chatham Maritime, Kent, United Kingdom. Sarigiannis, D., N. Soulakellis, K. Schäfer, M. Tombrou, N. Sifakis, D. Assimakopoulos, M. Lointier, E. Bossioli, A. Dantou, and M. Saisana, 2002. ICAROS: An integrated computational environment for the assimilation of environmental data and models for urban and regional air quality, Water, Air and Soil Pollution: Focus, 2(5):641–654. Sarigiannis, D., N. Soulakellis, N.I. Sifakis, M. Tombrou, and K. Schäfer, 2003a. Multidisciplinary data and model fusion:


A key to integrated air quality assessment, Proceedings, Remote Sensing of Clouds and the Atmosphere VII (Klaus Schäfer, Olga Lado-Bordowsky, Adolfo Comerón, and Richard H. Picard, editors), 21–27 September 2002, Aghia Pelagia, Greece (SPIE, Bellingham, Washington), 4882:520–531. Sarigiannis, D., N. Sifakis, N. Soulakellis, K. Schäfer, and M. Tombrou, 2003b. A new approach to environmental data fusion for integrated assessment of particulate matter loading and its effect on health in the urban environment, Proc. International Geoscience and Remote Sensing Symposium 2003, 21–25 July, Toulouse, France (IEEE), 7:4579–4581. Schäfer, K., G. Fömmel, H. Hoffmann, S. Briz, W. Junkermann, S. Emeis, C. Jahn, S. Leipold, A. Sedlmaier, N. Soulakellis, N. Sifakis, and D. Sarigiannis, 2002. Three-dimensional groundbased measurements of urban air quality to evaluate satellite derived interpretations for urban air pollution, Water, Air and Soil Pollution: Focus, 2(5):91–102. Schwartz, J., D.W. Dockery, and L.M. Neas, 1996. Is daily mortality associated specifically with fine particles? Journal of Air and Waste Management Association, 46:927–939. Schwartz, J., and L.M. Neas, 2000. Fine particles are more strongly associated than coarse particles with acute respiratory health effects in schoolchildren, Epidemiology, 11:6–10. Sifakis, N., and P.Y. Deschamps, 1992. Mapping of air pollution using SPOT satellite data, Photogrammetric Engineering & Remote Sensing, 58:1433–1437. Sifakis, N., N. Soulakellis, and D. Paronis, 1998. Quantitative mapping of air pollution density using Earth observations: A new processing method and application on an urban area, International Journal of Remote Sensing, 19:3289–3300. Sifakis, N., and N.A. Soulakellis, 2000. Satellite Image Processing for Haze and Aerosol Mapping (SIPHA): Code description and presentation of results, Proceedings, IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 2000, 24–28 July, Honolulu, Hawaii, pp. 222–224. WHO, 2000. Air Quality Guidelines for Europe, Second Edition, WHO Regional Publications Europe Series 91, World Health Organization, Geneva, Switzerland, 287 p.

February 2004

245