Integrating auxiliary data and geophysical techniques for the estimation of soil clay content using CHAID algorithm

Abstract

This study was conducted to estimate soil clay content in two depths using geophysical techniques (Ground Penetration Radar—GPR and Electromagnetic Induction—EMI) and ancillary variables (remote sensing and topographic data) in an arid region of the southeastern Iran. GPR measurements were performed throughout ten transects of 100m length with the line spacing of 10m, and the EMI measurements were done every 10m on the same transect in six sites. Ten soil cores were sampled randomly in each site and soil samples were taken from the depth of 0–20 and 20–40cm, and then the clay fraction of each of sixty soil samples was measured in the laboratory. Clay content was predicted using three different sets of properties including geophysical data, ancillary data, and a combination of both as inputs to multiple linear regressions (MLR) and decision tree-based algorithm of Chi-Squared Automatic Interaction Detection (CHAID) models. The results of the CHAID and MLR models with all combined data showed that geophysical data were the most important variables for the prediction of clay content in two depths in the study area. The proposed MLR model, using the combined data, could explain only 0.44 and 0.31% of the total variability of clay content in 0–20 and 20–40cm depths, respectively. Also, the coefficient of determination (R2) values for the clay content prediction, using the constructed CHAID model with the combined data, was 0.82 and 0.76 in 0–20 and 20–40cm depths, respectively. CHAID models, therefore, showed a greater potential in predicting soil clay content from geophysical and ancillary data, while traditional regression methods (i.e. the MLR models) did not perform as well. Overall, the results may encourage researchers in using georeferenced GPR and EMI data as ancillary variables and CHAID algorithm to improve the estimation of soil clay content.