PRELIMINARY RESULTS REGARDING THE APPLICATION OF STATISTICAL METHODS FOR SPATIAL PREDICTION OF SOIL PARAMETERS IN DOBROV BASIN (CENTRAL MOLDAVIAN PLATEAU)

Our study tests the usefulness of two statistical methods (linear and logistic regression), for spatial prediction of soil quantitative and qualitative parameters in uncharted areas, using the available data from neighbouring areas. Linear regression is used to map quantitative variables, such as soil reaction and occurrence depth of calcium carbonates, while logistic regression is tested for prediction of qualitative variables, such as the spatial occurrence probability of soil types (Chernozems and Preluvosols) and A mollic horizon. The explained variance for soil reaction and depth of calcium carbonates is 56% and 51% respectively, the regression models including the digital elevation model, wetness index, surface ratio and Y coordinate as predictors. The spatial occurrence probability of Chernozems, Preluvosols and A mollic horizon is computed on the basis of quantitative (digital elevation model, slope, mean annual temperature and precipitations) and qualitative (land use) predictors, the overall accuracy ranging from 69 to 90%. These preliminary results show that such methods are potentially useful for soil scientists, as they provide a first approximation of the spatial distribution of soil variables and may constitute an initial stage prior to traditional soil mapping.