engleski

Unsupervised neural networks in the analysis forest ecosystem stress using Pb soil data in Croatian karst areas

It is well known that soil pollution are directly linked to the amount of precipitation. Atmospherically introduced high lead concentrations in soils of Croatian karst occur along the sharp geomorphologic boundary along which the Mediterranean climate abruptly changes into a cold continental climate and at altitudes above 900 m. Detailed studies of Pb distribution in soil profiles showed concentrations of lead in remote regions up to 200 mgkg-1 in the upper 4 cm of the soil profiles. With the application of the Pb/Sc ratio obtained during the geochemical baseline mapping of the topsoil cover in the Croatian karst at 1088 sampling sites, the spatial risk of acid deposition in areas of high geomorphic variability was evaluated. The empirical model built on the neural networks related the amount of atmospherically introduced Pb (calculated from the Pb/Sc normalization variable which separates the anthropogenically introduced Pb from lithogenic lead), extent and type of forest cover, digital elevation model with its variations as well as the mean annual precipitation. The correlation model was very high (R>0.85). The presented model, which links soil geochemistry with precipitation and degree of forest damage, was found to be an suitable tool for evaluation of the spatial acidification risk.

Lead; soil; neural networks; karst; acid deposition; CroatiaUnsupervised neural networks; forest ecosystem stress; geochemistry; lead; soil; karst areas

nije evidentirano