Comparison of kriging with external drift and simple linear regression for predicting soil horizon thickness with different sample densities

This study examines two mapping methodʹs sensitivity to the sampling density of the variable of interest, which is the thickness of a silty-clay–loam (TSCL) horizon. The two methods are simple linear regression (SLR) and universal kriging with external drift (UKEXD). As slope gradient (β) derived from a DEM, is available for the whole study area and linearly related to TSCL horizon, was used for TSCL prediction by SLR and by UKEXD. The accuracy performance of TSCL prediction using these methods was assessed by comparison with another group of 69 sample points (validation sample) where the TSCL is actually measured. In the validation procedure for the two methods, two indices were calculated from the validation sample (measured values) and predicted values. These two indices are the mean error (ME) and the root mean square error (RMSE). The results showed that UKEXD was more accurate than the SLR. The improvement of the accuracy of the prediction from SLR to UKEXD was about 38%. To examine the effect of sampling density of TSCL (variable of interest) on the performance of both mapping methods, five subsets of 40, 50, 75, 100 and 125 observation sites of TSCL were randomly selected from the 150 sites of the prediction sample. For each subset, a prediction of TSCL was performed over the study area by: (i) SLR; (ii) UKEXD. The validation sample was used to compare the performance of the two methods according to the sample size of the variable of interest. The results show that whatever the sample size may be, UKEXD performs on average more accurate predictions than SLR. Moreover, the results indicate that UKEXD performed better when the sample size of the variable of interest increases. On the contrary, the performances with the linear regression remain stable whatever the sample size may be.