Improving ANN-MLP as effective spatial modeling method for copper and lead distribution in mining affected areas

The main purpose of the conducted investigation was to develop a visualisation model of spatial distribution using linear and nonlinear mathematical methods that combine a chemical analysis and various geospatial parameters in the Bregalnica River Basin, Kriva Reka River Basin, and the area of the field Ovče Pole, in the territory of the R. N. Macedonia. Main problem that affects linear methods is that elements contents are depend only on distance, as two-dimensional dependence. The most commonly used geostatistical prediction method is ordinary kriging that use a semi-variance function. Several parameters can influence the results what can lead to the incorrect interpretation. Most common anomaly of the applied linear models are Bull’s eye contours. In other side Artificial Neural Network – Multilayer Perceptron (ANN-MLP) can reveal much realistic results. ANN-MLP was used as nonlinear model for data processing and visualization of lead and copper in the investigated area. Model obtained by ANN was tested for the lithogenic (using top soil samples, 0-5 cm layer) distribution and atmospheric (moss samples) distribution. For the copper distribution enrichments were connected to the Cu open pit and some lithological units and along the rivers (alluvial planes) – what indicative presence of river transport. Lead enrichment was connected only to Pb mining areas, some lithological units and along the rivers (alluvial planes). At the middle flow of the river Bregalnica the ANN didn’t isolated the high concentrations. This means that the sediments are trapped in the lake Kalimanci and polluted sediments accumulate in the lake. Atmospheric enrichment is also connected to the mining areas, while the high concentrations are not connected to the lithological units