EVALUATION OF VARIOUS INTERPOLATION TECHNIQUES FOR ESTIMATION OF SELECTED SOIL PROPERTIES

Abstract

Efficient soil management practices depend on the spatial distribution of soil properties which
varies significantly even within the same field. Considering that it is impossible for any monitoring technique to
provide spatially continuous data, spatial interpolation plays an indispensable role in estimating the missing
values where no actual value was measured. The objective of this study was to evaluate various interpolation
techniques for the estimation of selected soil chemical properties in a study area located about 85km to the north
west of Cairo, Egypt. The studied soil properties included soil salinity, available phosphorus and nitrogen. The
interpolation techniques included two commonly used techniques namely, Ordinary Kriging (OK) and Inverse
Distance Weighting (IDW).The Artificial Neural Networks (ANN) method, which is considered a somewhat
new approach was also evaluated. Soil samples were collected at approximately 200×200 m grids at 0-25 cm
depth. The cross-validation method was used for evaluating the selected methods utilizing root mean square
error (RMSE) and mean relative error (MRE). This study revealed that ANN had the highest accuracy followed
by OK then IDW in terms of both RMSE and MRE when interpolating the studied soil properties. Nevertheless,
these results are dependent on the accuracy of the designed network which must have an overall accuracy of
coefficient of correlation (R) more than 0.80 between the predicted and the actual data. It also revealed that
the best IDW with the highest accuracy must have a power of 2 for salinity and nitrogen and a power of 3 for
phosphorus.