Soil depth prediction supported by primary terrain attributes: a comparison of methods

Abstract

The objective of this study was to investigate the benefits of methods that incorporate terrain attributes as covariates into the prediction of soil depth. Three primary terrain attributes - elevation, slope and aspect - were tested to improve the depth prediction from conventional soil survey dataset. Different methods were compared: 1) ordinary kriging (OK), 2) co-kriging (COK), 3) regression-kriging (REK), and 4) linear regression (RE). The evaluation of predicted results was based on comparison with real validation data. With respect to means, OK and COK provided the best prediction (both 110 cm), RE and REK gave the worst results, their means were significantly lower (79 and 108 cm, respectively) than the mean of real data (111 cm). F-test showed that COK with slope as covariate gave the best result with respect to variances. COK also reproduced best the range of values. The use of auxiliary terrain data improved the prediction of soil depth. However, the improvement was relatively small due to the low correlation of the primary variable with used terrain attributes.

Keywords

• soil depth
• geostatistics
• terrain
• ordinary kriging
• co-kriging
• regression-kriging