Soil depth plays an important role in plant cultivation. Assessing soil depth using digital terrain analysis is advantageous in comparison with conventional field observation both in terms of reduced time and labor costs and in terms of avoiding solum destruction. The objective of the study is to improve the accuracy of soil depth prediction in digital terrain analysis based on a dataset on parent materials and a nonlinear partial least squares regression. Soil depth modeling is performed and compared using simple partial least squares regression (SPLSR), PLSR with parent materials (PLSRP), and nonlinear PLSR with parent materials (NPLSRP), simultaneously. Using PLSRP and NPLSRP, various models corresponding to parent materials within the study area are constructed. The models’ fits are assessed using the coefficient of determination in calibration (R{₂₀₋}^2), the coefficient of determination in validation (R{ₕ₀₋}^2), root mean square error of calibration (RMSEC), and root mean square error of prediction (RMSEP). The results of PLSRP outperform SPLSR by 0. 08 in R{ₕ₀₋}^2 and by 6. 2 in RMSEP. NPLSRP, compared to the PLSRP, shows an increase in R{ₕ₀₋}^2 by 0. 31 and a decrease in RMSEP by 17. 1. The results indicate that NPLSRP can achieve a significant improvement in the accuracy of soil depth predictions.
Tak et al. (Mon,) studied this question.