engleski

Spatial prediction of heavy metal soil contents in continental Croatia comparing machine learning and spatial interpolation methods

Soil contamination caused by heavy metals presents a potential long-term issue to human health and biodiversity due to the bioaccumulation effect. Previous research at the micro level in Croatia detected soil contamination caused by heavy metals above maximum permitted values, which also implied the necessity of their current spatial representation at the macro level in Croatia. The aim of this study was to provide a spatial prediction of six heavy metals considered as contaminants of soils in continental Croatia using two approaches: a conventional approach based on interpolation and a machine learning approach. The prediction was performed on the most recent available data on cadmium (Cd), chromium (Cr), copper (Cu), nickel (Ni), lead (Pb) and zinc (Zn) concentrations in soils, from the Ministry of environment and energy. The conventional prediction approach consisted of the interpolation using the ordinary kriging (OK) in case of input data normality and stationarity, alongside the inverse distance weighting (IDW) method. For the machine learning approach, random forest (RF) and support vector machine (SVM) methods were used. IDW outperformed RF and SVM prediction results for all soil heavy metals contents, primarily due to sparse soil sampling. Soil Cr contents were predicted above the maximum allowed limit, while elevated soil contamination levels in some parts of the study area were detected for Ni and Zn. The highest soil contamination levels were observed in the urban areas of generalized land cover classes, indicating a necessity for its monitoring and the adjustment of land-use management plans.

soil contamination ; random forest ; support vector machine ; ordinary kriging ; inverse distance weighting ; land cover

nije evidentirano