Evaluation of Parametric and Nonparametric Machine-Learning Techniques for Prediction of Saturated and Near-Saturated Hydraulic Conductivity

Autor(es)

Kotlar, Ali Mehmandoost ^[1] ; Iversen, Bo V. ^[2] ; Van Lier, Quirijn de Jong ^[1]

Afiliação do(s) autor(es)

^[1] Univ Sao Paulo, Ctr Nucl Energy Agr CENA, Caixa Postal 96, BR-13416903 Piracicaba, SP - Brazil

^[2] Aarhus Univ, Dept Agroecol, Blichers Alle 20, DK-8830 Tjele - Denmark

• Tipo de documento: Artigo Científico
• Fonte: VADOSE ZONE JOURNAL; v. 18, n. 1 FEB 21 2019.
• Citações Web of Science: 5

Resumo

Parametric and nonparametric supervised machine learning techniques were used to estimate saturated and near-saturated hydraulic conductivities (K-s and K-10, respectively) from easily measurable soil properties including the name of the pedological horizon (HOR), soil texture (sand, silt, and clay), organic matter (OM), bulk density (BD), and water contents (theta(pF1),theta(pF2),theta(pF3), and theta(pF4.2)) measured at four different matric heads (-10, -100, -1000, and -8,848 cm, respectively). Using a stepwise linear model (SWLM) and the Lasso regression as parametric methods with 316 data in training and 135 data in the testing phase, four pedotransfer functions (PTFs) were obtained in which water contents for both methods play an important role compared with other variables. The SWLM showed better performance than Lasso in the testing phase for log(K-s) and log(K-10) prediction, with RMSE values of 0.666 and 0.551 cm d(-1) and R-2 of 0.26 and 0.65. Nonparametric supervised machine learning methods trained and tested with a similar data set significantly improved the accuracy of K-s prediction, with R-2 of 0.52, 0.36, and 0.53 for Gaussian process regression (GPR), support vector machine (SVM), and ensemble (ENS) methods in the testing stage. These methods also described 74.9, 66.7, and 72.5% of the variation of log(K-10). Bootstrapping validated the strong performance of nonparametric techniques. The feature selection capability of GPR determined that instead of using a model with all predictors, HOR, silt, theta(pFl), and theta(pF3) are sufficient for the prediction of log(K-s), while HOR, silt, and OM can predict log(K-10) as accurate as the comprehensive model with all variables. (AU)

• Processo FAPESP: 16/18636-7 - Mitigação da lixiviação de Nitrato em solos tropicais usando Hidróxidos duplos lamelares
• Beneficiário: Ali Mehmandoost Kotlar
• Modalidade de apoio: Bolsas no Brasil - Doutorado