Comparison of spatial prediction methods for mapping floodplain soil pollution

Abstract

Summary The floodplain soils of the River Geul (The Netherlands) are polluted by heavy metals, i.e. lead, zinc and cadmium. Several spatial prediction methods (local trend analysis, mathematical splines, inverse distance weighting, block kriging, point kriging and point co-kriging) were used to map zinc levels in a small and intensively sampled study area. Three subsets of the set of 145 data points, containing 12, 23 and 44 data respectively, were used to estimate zinc levels at 99 test locations. Correlation coefficients of linear relations between observed and estimated zinc levels and contour maps of prediction errors indicate that weighted local averaging methods perform better than the other methods. In the study area, point co-kriging with elevation as a co-variable outperforms all other methods when only few data on zinc are available.