FURTHER RESULTS ON PREDICTION OF SOIL PROPERTIES FROM TERRAIN ATTRIBUTES - HETEROTOPIC COKRIGING AND REGRESSION-KRIGING

Several methods involving spatial prediction of soil properties from l andform attributes are compared using carefully designed validation pr ocedures, The methods, tested against ordinary kriging and universal k riging of the target variables, include multi-linear regression, isoto pic cokriging, heterotopic cokriging and regression-kriging models A, B and C. Prediction performance by ordinary kriging and universal krig ing was comparatively poor as the methods do not use covariation of th e predictor variable with terrain attributes. Heterotopic cokriging ou tperformed isotopic cokriging because the former utilised more of the local information from the covariables. The combined regression-krigin g methods generally performed well. Both the regression-kriging model C and heterotopic cokriging performed well when soil variables were pr edicted into a relatively finer gridded digital elevation model (DEM) and when all the local information was utilised. Regression-kriging mo del C generally performed best and is, perhaps, more flexible than het erotopic cokriging. Potential for further research and developments re sts in improving the regression part of model C.