Experimental performance of spatial interpolators for groundwater salinity

Mapping groundwater qualities requires either sampling on a fine regular grid or spatial interpolation. The latter is usually used because the cost of the former is prohibitive. Experimental performance of five spatial interpolators for groundwater salinity was investigated using cross-validation. The methods included ordinary kriging (OK), lognormal kriging, inverse distance, inverse squared distance and inverse cubed distance. The results show that OK outperformed other interpolators in terms of bias. Interpolation accuracy based on mean absolute difference criterion is relatively high for all interpolators with small difference among them. While the three-dimensional surfaces produced by all inverse distance based procedures are dominated by isolated peaks and pits, surfaces produced by kriging are free from localized pits and peaks, and show areas of low groundwater salinity as elongated basins and areas of high salinity as ridges, which make regional trends easy to identify. Considering all criteria, OK was judged to be the most suitable spatial interpolator for groundwater salinity in this study