The interpolation accuracy for seven soil properties at various sampling scales on the Loess Plateau, China

Knowledge of the changes in interpolation accuracy with changing sampling scales is important when designing an appropriate sampling strategy. The objectives of this study were (1) to analyze the changes in interpolation accuracy with changing sampling scales for seven soil properties and (2) to find a suitable index that could predict the interpolation accuracy well. Nine hundred sixty-one samples were collected from a 30 x 30-m area. Seven soil properties were measured for each sample. Using a re-sampling analysis method, we grouped the samples under 16 subscales. Then, we divided the 16 subscales into two subsets, the first consisting of eight scales used as training sets and the second having the other eight scales as validation sets. Using the training sets, the interpolation accuracy and the contribution rate (CR) for the seven soil properties were compared and the relations of the interpolation accuracy to the coefficient of variation (CV), or to the ratio of sampling spacing to correlated range (S/R), or to the extent and spacing (E & S) were determined, the accuracy of prediction of which were then tested using the validation sets. The results showed that the mean interpolation accuracies varied greatly for different soil properties, with mean G values of training sets ranging from 2.4% for soil organic carbon, to 62.1% for sand content. With increasing sampling spacing or decreasing sampling extent, the interpolation accuracy decreased for all soil properties. The scales with the largest CR were not consistent with those with the highest interpolation accuracies. The interpolation accuracy was predicted better by E & S than by CV or by S/R. The measurement and analysis gave insight into the changes of interpolation accuracy and CR at various sampling scales. Predicting interpolation accuracy based on the scale parameters of sampling spacing and sampling extent was feasible, which provided a useful means by which to determine appropriate sample size and sampling strategy.