Non-linear partial least squares regression model for improving the prediction of soil profile thickness according to the parent rock database and topographic characteristics
JOO-HYUN TAK, YOUNG-SEONG JUNG, IL-BEOM JUNG, MYUNG-HAK JUNG, HYUN-U KIM
Kim Il Sung University, Pyongyang, DPRK
Keywords: soil depth, digital terrain analysis, non-linear regression, GIS, digital soil map, digital topographic model
Abstract
Soil depth plays an important role in plant growth. Evaluation of soil depth using digital terrain analysis is able to be conducted not only with less time and labors, but also without solum destruction, compared to conventional field observation. This research aims to improve the accuracy of soil depth prediction in digital terrain analysis using parent material dataset and non-linear partial least squares regression. Modeling of soil depth was performed and compared using simple partial least squares regression (SPLSR), PLSR with parent materials (PLSRP), and non-linear PLSR with parent materials (NPLSRP), simultaneously. While using the PLSRP and NPLSRP, different models were built, corresponding to parent materials within the study area. Models fit was evaluated with coefficient of determination of calibration (R2cal), coefficient of determination of validation (R2val), root mean square error of calibration (RMSEC), and root mean square error of prediction (RMSEP). The use of the PLSRP improved the performance of the prediction by 0,08 for the R2val and by 6,2 for the RMSEP, compared to the SPLSR. The NPLSRP increased the R2val by 0,31 and decreased the RMSEP by 17,1, compared to the PLSRP. The results indicated that the use of the NPLSRP would be able to improve accuracy of soil depth prediction, significantly.
|
