An Evaluation of some Meteororological and Topographical Parameters Factors Influencing Levels of Airborne Lead in Soils Near a Point Source Using Using Artificial Neural Networks (ANNs)

The purpose of this work is to use both observational and non-parametric statistical modeling data of atmospherically deposited Pb in soils around an historical industrial air emission point source to evaluate some important meteororological and topographical factors influencing levels of lead in soils. The application site is the major area of Lavrio, located in the southeast Attiki peninsula, about 60 km from Athens, Greece. The Ag bearing structures around Lavrio were known and exploited since the 6th century B.C.A. The intensive mining and smelting activities that took place till the early 90’s resulted to soil contamination by heavy metals. Previous studies in the area include both exposure and risk assessment studies; and also several epidemiological studies.

Our work presented in this paper focuses on the area in the vicinity of the lead production plant operated from 1868 to 1988. A total number of 66 soil samples were collected from an area of about 98 km2 located around the plant. The soil samples were taken from a depth of 10 cm. Pb concentrations were determined in the extracted solutions by using flame- and graphite-furnace atomic absorption spectrometry (AAS) with a detection limit of 100 ppb and a precision of 1% RSD. Artificial Neural Network models (ANNs) are employed to quantitatively evaluate the spatial patterns of Pb in the soils around the emission source. NNs have recently become the focus of much attention, largely due to their wide capability to efficiently model environmental data that are known to be complex and often non-linear. For the modeling purposes, for each of the sampling points the following variables were taken into account: the altitude; the orientation; the distance from the plant; and the mean annual percentage of the wind direction from the stack to the sampling point. In addition, the geological formation was considered because the natural Pb content in soils is strongly related to the composition of the bedrock.

Our preliminary results show that the various individual network models employed here have values of R2 ranged between 0.35 and 0.85. Based on these results it is concluded that neural networks may be used as attractive alternative tools to numerical models in various cases where source-receptor relationships are of concern; and to evaluate parameters influencing airborne pollutant levels in multimedia environmental settings.