A comparison of regression and regression-kriging for soil characterization using remote sensing imagery

In precision agriculture regression has been used widely to quantify the relationship between soil attributes and other environmental variables. However, spatial correlation existing in soil samples usually makes the regression model suboptimal. In this study, a regression-kriging method was attempted in relating soil properties to the remote sensing image of a cotton field near Vance, Mississippi. The regression-kriging model was developed and tested by using 273 soil samples collected from the field. The result showed that by properly incorporating the spatial correlation information of regression residuals, the regression-kriging model generally achieved higher prediction accuracy than the stepwise multiple linear regression model. Most strikingly, a 50% increase in prediction accuracy was shown in Na. Potential usages of regression-kriging in future precision agriculture applications include real-time soil sensor development and digital soil mapping.