Sie sind auf Seite 1von 6

Digital Soil Assessments and Beyond Minasny, Malone & McBratney (eds) 2012 Taylor & Francis Group,

p, London, ISBN 978-0-415-62155-7

Spatial modeling of human exposure to soil contaminationan example of digital soil assessment
J. Caudeville
INERIS (National Institute for industrial Environment and Risks), France

P. Goovaerts
BioMedware, Inc., US

F. Carr & R. Bonnard


INERIS (National Institute for industrial Environment and Risks), France

D. Arrouays
INRA (National Institute of Agronomic Research), France

A. Cicolella
INERIS (National Institute for industrial Environment and Risks), France

C.C. Jolivet & N.P.A. Saby


INRA (National Institute of Agronomic Research), France

C. Boudet
INERIS (National Institute for industrial Environment and Risks), France

ABSTRACT: Contamination of soil can have strong impacts on population health. Modeling these potential impacts is then an important issue of Digital Soil Assessment. Indeed, the quality of the health risk assessment strongly depends on the quality of digital soil contamination mapping. Thus, communicating about the risks should not be done independently of the overall uncertainties. The purpose of this study is to spatially assess human exposure to soil contaminants (Cadmium presented here) as a second step of digital soil contamination mapping. To this aim, a GIS-based raster platform is developed at 1 km2 resolution in order to incorporate soil contaminants and environmental pathways of these contaminants into human exposure. Contaminant soil concentrations are used to estimate soil ingestion pathway and transfer to vegetation and animal product. Soil concentrations are estimated using a kriging method that integrates data from surface and point spatial supports. Losses and inputs of chemical substances by several mechanisms, including leaching, runoff and deposition are also taken into account. After modeling the soil contaminant concentrations, the multimedia exposure model is applied for getting population exposure risk assessment. Results show two highly exposed areas, associated with ingestion of locally grown food. These areas with intermediate DSM uncertainty, correspond to a former industrial site and the suburb of the Lille city agglomeration. The platform, called PLAINE, allows the detection of hot-spot areas with significantly elevated exposure indicator values and for the design of further environmental sampling campaigns. 1 INTRODUCTION home-grown produced. This contamination and its associated risks are spatial processes which can be mapped. Modeling potential health impacts is then a good candidate for Digital Soil Assessment (Carr et al., 2007). The aim of the study is to map at fine resolution human exposure to soil contaminants in the French Nord Pas de Calais Region. Legacy data coming from environmental monitoring networks are used for characterize exposure pathways. Since these data rarely fit the

Trace metals are widely distributed in the environment both as natural components of soils, and as anthropogenically derived pollutants. Prolonged exposure to these pollutants can cause deleterious health effects in humans (Reilly 1991). They may enter the human body through inhalation and ingestion of soil, vegetation, meat and animal product both commercial and

19

final map specifications, they should be transformed and their uncertainty analysed using digital soil mapping procedures. Gay & Korre (2006) proposed a combination of quantitative probabilistic human health risk assessment and geostatistics to map human exposure to contaminants. However, this method does not take into account the pollutant dynamics such as multiple concomitant transfers, losses and accumulations of contaminants into the different media (air, soil and water). The methodology we propose integrates the dynamic of the pollutant and combines probabilistic human health risk assessment with spatial statistical methods. The paper describes first the multimedia model and the legacy data we used. The digital soil mapping and the modeling of uncertainties related to the risk exposure are then presented, followed by a discussion on how to interpret and communicate the results according to the confidence level of prediction.

Population is extremely dense with about four million inhabitants over an area of 12,414 km2. 2.1 The health risk model

We used a stochastic multimedia exposure model developed by Bonnard (2003) to assess the transfer of contaminants from the environment (air, soil, water) to individual exposure through the local food chain. It combines assumptions and concentration measurements of environmental agents in air, water, soil and food with data on inhalation and ingestion rates to produce an estimate of intake of the environmental agent and associated chronic health risk. Non-cancer risk is characterized using a hazard quotient (HQ). For ingestion pathway, this is the ratio of the average daily dose (ADD; milligrams per kilogram per dayalso called intake) of a chemical to a reference dose (RfD, milligrams per kilogram per day). The RfD is defined as the maximum tolerable daily intake of a specific pollutant that does not result in any deleterious health effects. 2.2 Legacy data

MATERIALS AND METHODS

The French region Nord-Pas-de-Calais is recognized as a former highly industrialized zone.


Table 1.

The legacy data, which are described in Table 1, were required for running the health risk model.

Legacy data used for spatially assess the human exposure risk to Cadmium. Support and resolution Centrod (point) of 0.5 0.5 Grid Source Aggregation of data from the Eulerian atmospheric dispersion model Chimere (INERIS) using European Monitoring and Evaluation Program (EMEP) emission register New collect of trace metal surface soil database from the INRA & ADEME Program: concentration of trace metals in surface soil layers) (Duigou & Baize 2010) Robust geostatistical interpolation of the Trace Element content in soil (French Soil Quality Monitoring Network) using parental material information (GIS SOL) (Marchant et al. 2010). Aggregation, using the Siseeaux database, the administrative boundary map of France and distribution unit serve map Aggregation of the Global Resource Information Database (GRID) Combination of Meteo France data with the Eulerian atmospheric dispersion model Chimere (INERIS) Building of an Indicator based on national autoproduction study and spatialization, using Land Cover (CORINE Land Cover) dataset and census data Data aggregation from administrative areas, using the CIBLEX database

Parameter Atmospheric deposition of cadmium and concentration Cadmium concentration in soil

Point: sample; surface: commune

Cadmium background concentration in soil

1 1 km grid

Cadmium concentration in water Evapotranspiration Precipitation

Georeference: commune

0.5 0.5 Raster Centrod of 0.5 0.5 Grid Georeference: census data (district); Areas (CLC)

Ingestion scenario

Food consumption data

Georeference: region

20

2.3

Digital Soil Mapping (DSM)

The Digital Soil Mapping procedure consists in spatially predicting soil cadmium concentration using data from the newly INRA/ADEME collected trace metals in soil surface in France (BD ETM,, Duigou and Baize, 2010). This dataset includes a total of 25,000 sample analysis for cadmium in the Nord-Pas-de-Calais region. Observations are provided on two different geographical supports: data points and data aggregated at district levels (10 km2). Usually, for mapping purposes, these data are processed independently using different sets of methods. However, incorporating areal data could be especially useful to improve prediction accuracy in regions without any point data and vice versa. Goovaerts (2011) recently introduced a geostatistical approach to combine areal and point data in the mapping of soil properties. Areal data are first disaggregated to the nodes of the interpolation grid using Areato-point kriging, resulting in an isopleths map of local means. Differences between point data and local mean are then interpolated by using simple kriging, and the kriged residuals are added to the local means to derive the final estimates. Information on land contamination is usually not complete. The absence or unavailability of important information on contamination often leads to an underestimation of associated risk. The bivariate local Moran statistics (Anselin, 1986) was used to characterize the magnitude and

types of uncertainties related to the two DSM kriging procedures, based on areal and point data, respectively. The bivariate Morans I measures the degree of linear association (positive or negative) between the value of one variable at a given location and the mean value of another variable at neighbouring locations. It suggests two classes of positive spatial correlation and two classes of negative spatial correlation. The significance of Morans I was computed on the basis of a permutation approach, whereby a reference distribution is created by calculating the Morans I for spatially random layouts. In this paper, we applied the bivariate Morans I to the kriging variances of the two interpolation procedures, one for the areal data and the other for the point data. The resulting map shows area where prediction is based on sufficient spatial information. 2.4 Digital Soil Assessment (DSA)

The global flowchart of the DSA phase is presented in Figure 1. Environmental and population data are integrated and processed in the PLAINE (environmental inequalities analysis) GIS-based platform developed by Caudeville et al. (2011) in order to run the multimedia exposure model. Soil concentrations from the DSM phase are used to estimate soil ingestion pathway and transfer to vegetation and animal product in the model. We consider this model to be appropriate for taking

Figure 1.

The Digital Soil Assessment (DSA) procedure.

21

into account transfers and losses of substances by several mechanisms, including leaching and runoff. Deposition is used to take into account for soil pollutant inputs. For soil, the concentrations are estimated by combining fallouts from emission source (transport and industry) with losses by leaching and runoff. Due to sources and past contamination, the total soil concentration is calculated by integrating the background soil concentration. The background is defined as the ambient representative concentration in substances or compounds in a specific media. It takes in account natural concentrations (local pedo-geochemical background content) and those due to diffuse anthropic sources. Sensitivity analysis showed that the most sensitive parameters in the model are transfer factors of heavy metal uptake by plants from soil (Bonnard, 2005). To achieve a better assessment of the transfer of contaminants, a large database on heavy metal uptake by plants was constructed to build probability distributions. Monte Carlo simulations were then used to propagate uncertainty through the model. The simulation results are exposure dose and hazard quotient expressed as percentile values. 3 RESULTS AND DISCUSSION

The modeling is performed on a 1 km spacing grid covering an area of 12,784 km2. The result of soil cadmium concentration interpolation is displayed in Figure 2a. Two main areas are highlighted where soil cadmium concentrations are above 5 mg . kg1. Figure 2b shows the result of the bivariate local Moran test. The map informs on the location and sampling density of areal and point data used for soil cadmium prediction.

Areas with significant correlation (p-value below 0.05) are allocated into one of four classes of spatial autocorrelation produced in the Moran analysis and displayed in different colors (non significant locations are colored gray). Variance is elevated in sparsely sampled areas. High-high category means cluster with high kriging variances for both point and areal data kriging. In those areas, predictions were based on few point and areal neighboring observations. Low-low category means cluster with low kriging variances. Predictions were based on many point and areal data, and should be reliable. Low-high and High-low categories correspond to negative spatial correlation between kriging variances: low variance for one and high variance for the other. Intake and HQ were estimated from 1,000 simulations. In the modeling, the receptor population is exposed over a 70-years period. Because for each pollutant the 27 years old age group is the highly exposed group, this one is used for the analysis. The outputs of the model include mean, median inter-quartile and complete probability density distribution. The analysis takes into account intakes from soil contamination and from other exposure pathways and environmental compartments. Figure 3 shows the histograms of the relative contribution of exposure pathways to total intake for cadmium. Relative contributions are calculated by integrating intake by exposure media for each grid to total intake. For cadmium, most exposure is mainly due to vegetation ingestion (average 67%). Excepted for water ingestion, other contributions are very low. Local and commercial product contributions were separated. Because the spatial variability of cadmium in drinking water is very low in this region, the spatial pattern of the HQ cadmium

Figure 2.

Maps of predicted cadmium soil concentration (a) and associated uncertainty (b).

22

Figure 3. Histograms of the relative contribution of exposure pathways to total intake for cadmium and P50 (separated according to their origin). Results are obtained by averaging all grids.

Figure 4.

Cadmium HQ maps for 27 years old age group in the study area. (a) Median and (b) 90th percentile.

map reflects those of soil concentration maps (Figure 4). Since parameter uncertainty is incorporated in the calculations, the risk manager can work with, e.g. median (P50) numbers shown in map (a) or consider a worst case scenario e.g. using P90 numbers shown in map (b). Results in Figure 4 indicate very large differences between P50 and P90 (factor 2 to 3 with a maximum value of 7.7 for P90). The map based on P50 presents two areas where HQ is over 1.0. A HQ > 1.0 suggests that a person may experience adverse health effects during his or her lifetime, nevertheless the HQ is a highly conservative index and relates

to very minor biologic responses (Teuschler et al., 1999). The largest risk values correspond to an ancient industrial site (Metaleurop) and the city of Lille city. In those areas, soil surface cadmium concentrations are very high (7 mg/kg for the maximum value) and could present a risk by ingestion of local vegetation pathway. A large amount of information is contained in the outputs of the model which can be described at different spatial scales, and mapped to provide a powerful visual tool for risk managers, facilitating the implementation of efficient measures for reducing exposure risk in different locations.

23

This type of result must be interpreted jointly with the spatial uncertainty map to account for the density of information used to build these indicators. For instance, the two highlighted areas in Figure 4a correspond to non significant risk-locations according to the bivariate local Morans I, with medium uncertainty both for point and areal data kriging (gray areas in Figure 3b). A third area with elevated Cd concentration belongs to the Low-high category and does not appear as an area at risk in the HQ maps. Low-high category corresponds to areas described by only well known areal data in the initial soil concentration dataset and suggests high potential concentration and risk underestimations around an industrial point source. Those maps allow identifying where additional data or sampling campaigns are needed. It can also highlight exposure hotspots where exposure risk reduction measure should be taken, particularly in areas with sufficient level of confidence (sufficient amount of information and low uncertainties). 4 CONCLUSION

Current research is exploring the use of secondary information, such as proximity to industrial facilities, to identify further sampling areas, in particular areas where point data are lacking and contaminated sites are suspected. REFERENCES
Anselin, L. 1986. MicroQAP: a microcomputer implementation of generalized measures of spatial association: University of California, Santa Barbara, Department of Geography. Bonnard, R. 2003. Evaluation de limpact sur la sant des rejets atmosphriques des tranches charbon dune grande installation de combustionPartie 2: Exposition par voies indirectes. Rapport INERIS 03/45956. Bonnard, R. 2005. Impact des incertitudes lies aux coefficients de transfert dans les valuations de risque sanitaire, DRC/ERSA/RBn. Rapport INERIS 67645/204. Carr, F., McBratney, A.B., Mayr, T. & Montanarella, L. 2007. Digital Soil Assessment: Beyond DSM. Geoderma 143, 6979. Caudeville, J., Boudet, C., Denys, S., Bonnard, R., Govaert, G. & Cicolella, A. 2011. Caractrisation des ingalits environnementales en Picardie fonde sur lutilisation couple dun modle multimdia et dun systme dinformation gographique. Environnement, risques et sant 10(6), 239242. Duigou, N. & Baize, D. 2010. Nouvelle collecte nationale danalyses dlments en traces dans les sols (horizons de surface)(Cd, Cr, Cu, Hg, Ni, Pb, Se, Zn). Rapport final. Contrat ADEME 0875C0036, 284 pp. Gay, J.R. & Korre, A. 2006. A spatially-evaluated methodology for assessing risk to a population from contaminated land. Environmental Pollution 142(2):227234. Goovaerts, P. 2011. A coherent geostatistical approach for combining choropleth map and field data in the spatial interpolation of soil properties. European Journal of Soil Sciences, 62(3):371380.1. Marchant, B.P., Saby, N.P.A., Lark, M., Bellamy, P.H., Jolivet, C.C. & Arrouays, D. 2010. Robust analysis of soil properties at the national scale: cadmium content of French soils. European Journal of Soil Science 61:144152. Reilly, C. 1991. Metal Contamination of Food. 2nd ed. London: Elsevier Applied Science. Teuschler, L.K., Dourson, M.L., Stiteler, W.M., McClure, P. & Tully, H. 1999. Health risk above the reference dose for multiple chemicals. Regul Toxicol Pharm 30:S19S26.

The purpose of this study was to spatially assess human exposure to soil contaminants as a second step of digital soil contamination mapping. The before map of soil contaminant levels was produced from areal and point-level data using geostatistical interpolation. Two different types of uncertainties were characterized in order to assess the spatial distribution of risks and provide a measure of uncertainty in the assessment. Results highlighted two hot-spot areas with elevated risk values. The case study illustrates the benefits, in terms of scientific knowledge, of combining exposure assessment methodology with spatial analysis techniques to increase the effectiveness of risk maps and better guide stakeholders to take decisions for safeguarding citizen health. At the same time, this study highlights some of the conceptual and practical difficulties that arise in assessing and interpreting exposure dose and the necessity to adapt methodology to specific data.

Das könnte Ihnen auch gefallen