The assessment of variability of the concentration of chromium in soils with the application of neural networks

Various ways of approaching the horizontal distribution trend (tendency) of Chromium (Cr) in soil, where pollution by this element is high, were analyzed. Interpolation algorithms: triangular irregular network (TIN), kriging, regularized spline with tension (RST), and artificial neural networks; radial basis function network (RBF), probabilistic neural network (PNN), generalized regression neural network (GRNN) and mixture density network (MDN) were applied. Data from field experiments, carried out in the area of the chemical plant in Alwernia, were used. The soil pollution spatial distribution examinations lead to the conclusion that in the first place was the information precision determination, and also the limit of error, through the pollution evaluation acceptance, whereas in the second place was the indication or standing out the regularity connected with the emission effect mechanism. It seems that the chromium concentration in soils variation, noticed even on short distances, makes the acceptance of interpolation method difficult, as a method of contamination distribution evaluation. On the other hand the considerable nonlinearity makes difficult the acceptance of regression model. In these circumstances, the possibility which is worth consideration is the modelling with the application of neuron networks, that is also hybrid solution application (for instance MDN), which gives the possibility of Cr concentration in soil variation deeper analysis (e. g. calculation local probability distribution, local variance, etc.).