Using residual analysis in electromagnetic induction data interpretation to improve the prediction of soil properties

Abstract

The electromagnetic induction (EMI) technique has been widely used to survey soil properties at intermediate spatial scales. However, EMI data interpretation remains a challenge for more accurate and robust mapping. Residual analysis is an alternative approach that can be used to improve the EMI data mining. On a tea garden (TG) hillslope and a bamboo forest (BF) hillslope, terrain indices were used to regress the apparent electrical conductivity (ECa) in ten repeated EMI surveys using stepwise multiple linear regressions (SMLR). Residuals of ECa in these regressions, which had terrain influence removed, were then calculated. The classification and regression tree (CART) model was adopted to quantify the relative contributions of terrain indices (elevation, slope, plane curvature–PLC, profile curvature–PRC, and topographic wetness index–TWI), static soil properties (rock fragment content–RFC, depth to bedrock–DB, contents of clay, silt and sand), and dynamic soil property (volumetric soil moisture–θ) to ECa and their residuals. Results show that contributions of terrain indices to ECa are around 50%. However, contributions of terrain indices to ECa residuals are <20%, while great contributions of different soil properties to ECa residuals can be observed in some cases. On both hillslopes, better predicting accuracies were achieved when using ECa residuals as independent variables in SMLRs to predict soil properties than using only terrain indices or ECa as independent variables. Similarly, on both hillslopes, using terrain indices plus ECa residuals as independent variables also yield better prediction of θ than using only terrain indices or using terrain indices plus ECa as independent variables. Findings of this study indicate that residual analysis can be a useful technique in improving EMI data interpretation for estimating the spatial variations of soil properties. In cases that relationship between target soil properties and ECa readings are weak, this approach can probably be used to improve the mapping accuracy of the target soil properties.