International Environmental Modelling and Software Society (iemss) 7th Intl. Congress on Env. Modelling and Software, San Diego, CA, USA, Daniel P. Ames, Nigel W.T. Quinn and Andrea E. Rizzoli (Eds.) http://www.iemss.org/society/index.php/iemss-2014-proceedings Using Geostatistical Tools for Mapping Traffic- Related Air Pollution in Urban Areas Lubos Matejicek Charles University in Prague, Benatska 2, Prague 2, 12801, Czech Republic lubos.matejicek@natur.cuni.cz Abstract: Air pollution sources caused by increasing road traffic reduce air quality and affect people in urban areas. In order to improve living conditions in urban areas, predictions of effects on air pollution are needed for assessing exposures as part of epidemiological studies, and to inform urban air-quality policy and traffic management. A prediction system for estimation, analysis and visualization has been developed to model spatial patterns of traffic-related air pollution. In this study, several geostatistical techniques are used for prediction of NO2 and PM10. The primary data for geostatistical methods originate from sample points that are generated from a network of automatic monitoring stations and, in addition, complemented by other sample points estimated by geographically weighted regression (GWR). GWR is used to provide a local form of linear regression to explore spatially varying relationships between air pollution, as a dependent variable, and a number of explanatory variables, such as elevation, nearest distance to a major road and ratio of built-up sites in the local area (the circle with a diameter of 1 kilometre). The techniques used for spatial interpolation are based on geostatistical methods such as ordinary kriging. The attached case study is focused on the area of Prague, the capital and largest city of the Czech Republic. GWR and the prediction maps of air pollution by NO2 and PM10 show highly exposed sites that indicate the need for emergency measures in urban air-quality policy and traffic management. Keywords: Air pollution; urban traffic; geostatistics; GWR; ordinary kriging. 1. INTRODUCTION Air quality in urban areas represents a major public health burden and is a long-standing concern to citizens. Air pollutants are associated with the incidence of many diseases, symptoms and infra-clinic conditions such as difficulty in breathing or even lung cancer (Jephcote and Chen, 2013; Portnov et al., 2009). Exposure estimates of air pollutants can address individuals or large population groups, and can be based on direct methods, such as exposure monitoring, or indirect methods, such as geostatistical predictions and air dispersion modelling. Exposure monitoring can assess local air quality at locations of single stations of a monitoring network, but the mapping of air pollution over a whole area requires the use of numerical models (Borrego et al., 2006). Air-dispersion models enable estimation of the spatial and temporal distribution of air pollutants, but data quality is limited by the model complexity and input data accuracy. The input datasets are mostly based on meteorological inputs and identification of emission sources. Air quality is also monitored in European cities to implement the European Commission Directives on Ambient Air Quality, which are regularly updated. However, disparities in exposure to air pollution between and within European countries still remain due to differences in levels of development (Pascal et al., 2013). For human health, the hourly mean of NO2 must not be exceeded more than eighteen times per annum, with the limit value set at 200µg m -3. The annual mean must not be exceeded over any given area of any European member state, and is set at 40µg m -3 (Robinson et al., 2013). In Europe, annual mean PM10 should not exceed 40µg m -3 (limit value set in 2005). The WHO-AQG for PM, chosen as the lowest levels at which total, cardiopulmonary and lung cancer mortality have been shown to significantly increase in response to long-term exposure to PM, are set as an annual mean of 20µg m -3 for PM10 (World Health Organisation, 2005)..
2. GEOSTATISTICAL TOOLS FOR MAPPING AIR POLLUTION Geostatistical tools deal with processing of spatial or spatio-temporal datasets. Many geostatistical algorithms are incorporated in Geographic Information Systems (GIS). This paper is focused on using Geographically Weighted Regression (GWR) and geostatistical interpolation techniques such as kriging. GWR is a fairly recent contribution to modelling spatially heterogeneous processes (Brunsdon et al., 1996). The underlying idea of GWR is that parameters can be estimated anywhere in the area of interest given a dependent variable and a set of explanatory variables which have been measured at sites whose locations are known. GWR provides a local model for the dependent variable by fitting a regression equation to every feature in the dataset. It constructs these separate equations by incorporating the dependent and explanatory variables of features falling within the bandwidth of each target feature. The shape and size of the bandwidth is dependent on setting the kernel type, bandwidth method, distance, and number of neighbours. The geostatistical interpolation techniques are focused on Gaussian process regression, such as kriging, for which the interpolated values are estimated by a Gaussian process governed by prior covariances. The ordinary kriging is used and optimized by trend removal and anisotropy properties (Cressie, 1993). The employed spatial interpolation algorithms are implemented in ArcGIS together with exploratory spatial data analysis (Johnston et al., 2001), and are often used for prediction of air pollution in urban areas (Matejicek et al., 2006). The geostatistical tools for mapping of air pollution are implemented in the Geostatistical Analyst extension of the ArcGIS 10.x. Thus, the same projects can be applied for similar urban areas of interest in this GIS environment. Data inputs and temporary data outputs are mostly represented by vector layers such as shapefiles or geodatabase classes. The output predictions of air pollution (NO2, PM10) are based on raster layers such as ESRI grids that can be easily exported to other bitmap formats for display and visualization such as TIFF, JPEG and PNG. The input/output spatial data can be shared with other GISs such as MapWindow, Quantum GIS, or Computer-Aided Design (CAD) systems. Other possibilities for spatial data exchange are represented by GIS servers (ArcGIS server and MapServer), or by using cloud based software applications. 3. INTEGRATED RESEARCH IN THE GIS ENVIRONMENT GIS has been used in environmental studies to model pollution dispersion and its effects on residential areas for several decades. The ability to estimate environmental effects is important for decision-making processes to initiate public health intervention activities (Cromley and McLafferty, 2002). Spatio-temporal modelling in the framework of GIS occurs whenever operations attempt to emulate geographic processes in the real world, at selected points in time or over extended periods (Maguire et al., 2005). GIS is used to perform nearly all the tasks focused on spatial modelling and visualization for mapping traffic-related air pollution in urban areas. The input datasets include thematic map layers of the city such as the digital elevation model (DEM), major roads, land-use, an automatic air pollution monitoring network and a wide range of data extensions that are also focused on risk assessment and natural protection. 4. THE AREA OF INTEREST The attached case study is focused on the area of Prague, the capital and largest city of the Czech Republic, Fig. 1. As an example, mapping of air pollution is based on two pollutants, NO2 and PM10, monitored by an automatic network (http://www.premis.cz, March, 2014). The actual configuration of the monitoring network is 15 monitoring stations that measure 12 main pollutants. Data are recorded in one hour periods. The data are archived in the database of the Air Quality Information System of the Czech Republic.
Figure 1. The satellite image of the area of Prague complemented by a network of monitoring stations for air pollution and by major roads (Landsat 8, October, 2013). 5. A CASE STUDY: MAPPING TRAFIC-RELATED AIR POLLUTION The attached case study is focused on mapping air pollution (NO2 and PM10) with GWR and ordinary kriging. GWR is used to predict pollutant concentrations in a network of points that are created as a specified number of random point features. Random points are generated over the area of the city of Prague. The explanatory variables are represented by the elevation derived from DEM, by the nearest distance to a major road estimated from the raster layer and by the ratio of the built-up sites. The ratio of the built-up sites is estimated from the land-use layer in the local area of a circle with a diameter of 1 kilometre. The dependent variable is represented by the average of values recorded in the period from 8 a.m. to 8 p.m. at one hour intervals. This period covers the maximum concentration of pollutants in the morning and in the afternoon, which are characteristic for urban transport in Prague and many cities around the world. Air pollution contributed from road-transport within the selected period is often acknowledged in relation to the health outcomes across a temporal scale in risk assessment studies. The average values for a period of a month (March 2013 February 2014) are given in Fig 2 for NO2 and in Fig 3 for PM10. The attached graphs indicate a high variability of concentrations of selected pollutants. The variability in the short time periods is highly dependent on weather conditions, such as wind speed and wind direction, temperature, precipitation and humidity. The average in a year-long period represents a steady value of pollutant concentrations on the local spatial scales in comparison with the previous years. The GWR method for estimation of model coefficients and prediction of pollutant concentrations at new sites was processed with a fixed kernel type and the kernel extent determined using the Akaike Information Criterion (AICc). The results of processing after geostatistical interpolation by ordinary kriging are given for annual average values (March 2013 February 2014) in Fig 4 for NO2, and in Fig 5 for PM10. The values of the NO2 concentrations (actually measured at 11 existing stations) and PM10 concentrations (actually measured at 12 existing stations) are produced for every location in the city of Prague from the averaged point concentrations captured by automatic monitoring stations and a group of 26 other point concentrations estimated by GWR. The extended network of sampling points yields better predictions of pollutant concentrations on a local scale.
Figure 2. The average values of NO2 for a period of one month (March 2013 February 2014, the high variability in month-long periods is dependent on the weather conditions such as wind speed and wind direction, temperature, precipitation and humidity). Figure 3. The average values of PM10 for a period of one month (March 2013 February 2014, the high variability in month-long periods is dependent on the weather conditions such as wind speed and wind direction, temperature, precipitation and humidity).
Figure 4. The annual average values of NO2 concentrations (March 2013 February 2014) measured at 11 existing stations produced for every location from the averaged point concentrations measured by automatic monitoring stations and a group of 26 other point concentrations estimated by GWR. Figure 5. The annual average values of PM10 concentrations (March 2013 February 2014) measured at 12 existing stations produced for every location from the averaged point concentrations measured by automatic monitoring stations and a group of 26 other point concentrations estimated by GWR.
6. CONCLUSIONS AND RECOMMENDATIONS The presented research indicates the potential advantages of using geostatistical methods such as GWR and geostatistical interpolation in the GIS environment for estimating the environmental effects on residential areas. Considering the complexity of this research and variable weather conditions, the attached case study is focused on annual average concentrations of traffic-related air pollution. The study shows that recent GIS s can be used to perform nearly all the tasks focused on spatio-temporal air pollution modelling for analysis in the framework of decision-making processes for urban planning and risk assessment. The high variability of input datasets and variability of weather conditions in short periods require validation of predictions on the short-term scale and error assessment. 7. ACKNOWLEDGMENTS The geostatistical research was processed in the GIS Laboratory at the Faculty of Science, Charles University in Prague and was supported in the framework of FRVS project 131/2014/A/a. 8. REFERENCES Borrego, C., Tchepel, O., Costa, A.M., Martins, H., Ferreira, J., Miranda, A.I. (2006). Traffic-related particulate air pollution exposure in urban areas. Atmos. Environ. 40, 7205-7214. Brunsdon, C., Fotheringham, A.S., Charlton, M. (1996). Geographically weighted regression: a method for exploring spatial non-stationarity. Geogr. Anal. 28(4), 281-298. Cressie, N.A.C. (1993). Statistics for Spatial Data, revised ed. John Wiley & Sons, New York. Cromley, E.K., McLafferty, S.L. (2002). GIS and Public Health, first ed. The Guilford Press, New York. Jephcote, C., Chen H. (2013). Geospatial analysis of naturally occurring boundaries in road-transport emissions and children s respiratory health across a demographically diverse cityscape. Soc. Sci. Med. 82, 87-99. Johnston, K., Ver Hoef, J.M., Krivoruchko, K., Lucas, N. (2001). Using ArcGIS Geostatistical Analyst, first ed. ESRI Press, Redlands, CA. Portnov, B.A., Dubnov, J., Barchana, M. (2009). Studying the association between air pollution and lung cancer incidence in a large metropolitan area using a kernel density function. Socio. Econ. Plan. Sci. 43, 141 150. Pascal, M., Corso, M., Chanel., O, Declercq, C., Badaloni, C., Cesaroni, G., Henschel, S., Meister, K., Haluza, D., Martin-Olmedo, P., Medina, S. (2013). Assessing the public health impacts of urban air pollution in 25 European cities: Results of the Aphekom project. Sci. Total Envir. 449, 390 400 Matejicek, L., Engst, P., Janour Z. (2006). A GIS-based approach to spatio-temporal analysis of environmental pollution in urban areas: A case study of Prague s environment extended by LIDAR data. Ecol. Model. 199, 261-277. Maguire, D.J., Batty, M., Goodchild, M.F. (2005). GIS, Spatial Analysis and Modeling, first ed. ESRI Press, Redlands, CA. Robinson, D.P., Lloyd, C.D., McKinley, J.M. (2013). Increasing the accuracy of nitrogen dioxide (NO2) pollution mapping using geographically weighted regression (GWR) and geostatistics. Int. J. Appl. Earth Obs. 21, 374-383. World Health Organisation. WHO air quality guidelines for particulate matter, ozone, nitrogen dioxide and sulfur dioxide. WHO/SDE/PHE/OEH/06.02. Geneva: WHO; (2005). pp. 1-22. [28-9-2011. Ref Type: Report].