The report presents a series of numerical experiments concerning application of Support Vector Machines for the two class spatial data classification. The main attention is paid to the variability of the results by changing hyperparameters: bandwidth of the radial basis function kernel and C parameter. Training error, testing error and number of support vectors are plotted against hyperparameters. Number of support vectors is minimal at the optimal solution. Two real case studies are considered: Cd contamination in the Leman Lake, Briansk region radionuclides soil contamination. Structural analysis (variography) is used for the description of the spatial patterns obtained and to monitor the performance of SVM.