SPATIAL PREDICTION OF SOIL-SALINITY USING ELECTROMAGNETIC INDUCTION TECHNIQUES .1. STATISTICAL PREDICTION MODELS - A COMPARISON OF MULTIPLELINEAR-REGRESSION AND COKRIGING

We describe a regression-based statistical methodology suitable for pr edicting field scale spatial salinity (EC(e)) conditions from rapidly acquired electromagnetic induction (EC(a)) data. This technique uses m ultiple linear regression (MLR) models to estimate soil salinity from EC(a) survey data. The MLR models incorporate multiple EC(a) measureme nts and trend surface parameters to increase the prediction accuracy a nd can be fitted from limited amounts of EC(e) calibration data. This estimation technique is compared to some commonly recommended cokrigin g techniques, with respect to statistical modeling assumptions, calibr ation sample size requirements, and prediction capabilities. We show t hat MLR models are theoretically equivalent to, and cost-effective rel ative to cokriging for estimating a spatially distributed random varia ble when the residuals from the regression model are spatially uncorre lated. MLR modeling and prediction techniques are demonstrated with da ta from three salinity surveys.