| Abstrakt: |
This study aimed to compare the performance of modern data mining (DM) techniques such as random forest (RF), support vector machine (SVM), and artificial neural networks (ANN) for estimating the two key points of the soil moisture curve including field capacity (FC) and permanent wilting point (PWP). Furthermore, their performance was compared to traditional methods proposed by Saxton and Rawls (2006) and Tyler and Wheatcraft (1990). To this end, 1960 soil samples were collected from various locations throughout Iran. The results showed that the Tyler and Wheatcraft (1990) model provided more accurate estimates of soil moisture at low suction (FC), whereas the RF method was more efficient at high suction (PWP). Root-mean-square error (RMSE), normalized root-mean-square error (NRMSE), mean bias error (MBE), and Willmot's index of agreement (d) were estimated at 3.32, 0.13, − 0.59, and 0.96, respectively, for estimating FC (w w−1%). In addition, the RMSE, NRMSE, MBE, and d values for estimating PWP (w w−1%) were 2.46, 0.18, − 0.24, and 0.9, respectively. The MBE value in this study was negative, indicating that there was no overestimation problem. The results demonstrated that the RF method was more efficient than other methods for large datasets. In general, the DM technique was more accurate and efficient than other models, and it can be confidently used in high suction conditions. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink