Prediction of Soil Moisture-Holding Capacity with Support Vector Machines in Dry Subhumid Tropics

Abstract

Soil moisture-holding capacity data are required in modelling agrohydrological functions of dry subhumid environments for sustainable crop yields. However, they are hardly sufficient and costly to measure. Mathematical models called pedotransfer functions (PTFs) that use soil physicochemical properties as inputs to estimate soil moisture-holding capacity are an attractive alternative but limited by specificity to pedoenvironments and regression methods. This study explored the support vector machines method in the development of PTFs (SVR-PTFs) for dry subhumid tropics. Comparison with the multiple linear regression method (MLR-PTFs) was done using a soil dataset containing 296 samples of measured moisture content and soil physicochemical properties. Developed SVR-PTFs have a tendency to underestimate moisture content with the root-mean-square error between 0.037 and 0.042 cm3·cm−3 and coefficients of determination (R2) between 56.2% and 67.9%. The SVR-PTFs were marginally better than MLR-PTFs and had better accuracy than published SVR-PTFs. It is held that the adoption of the linear kernel in the calibration process of SVR-PTFs influenced their performance.

Using Machine Learning for Prediction of Saturated Hydraulic Conductivity and Its Sensitivity to Soil Structural Perturbations

“…Several studies have used KNN type of ML to predict soil hydraulic properties (e.g., Botula et al, 2013;Nemes et al, 2006Nemes et al, , 2008; Elshorbagy et al (2010) identify KNN as an attractive modeling technique for hydrology applications. Many studies have used SVR algorithm to model soil hydraulic properties (e.g., Angelaki et al, 2018;Kaingo et al, 2018;Kotlar et al, 2019;Mady & Shein, 2018;Singh et al, 2019). Recently, some studies have found that SVR models predicted soil hydraulic properties more accurately than artificial neural network models (Khlosi et al, 2016;Twarakavi et al, 2009;Zhang et al, 2018).…”

Machine Learning Approach to Simulate Soil CO2 Fluxes under Cropping Systems

“…Although the SVR algorithm has many hyperparameters, in this study, we chose four hyperparameters (C, ε, γ and the kernel function) that were tuned using the GridSearchCV module. According to Kaingo et al [63], a successful SVR implementation depends on selecting a suitable kernel function, choice of the cost parameter C, and the "tube" insensitive variable ε. The "RBF" (radial basis function) performed better than the linear and sigmoid functions.…”

Development and Evaluation of Pedotransfer Functions to Estimate Soil Moisture Content at Field Capacity and Permanent Wilting Point for South African Soils

“…Therefore, the existing PTFs developed internationally still need to be calibrated and validated locally against measurements before they can be reliably used to estimate hydraulic properties in new sites that are located in different regions from their origins [12,[44][45][46]. However, the lack of representative information on soil hydraulic properties hinders the validation of the existing PTFs and the development of local PTFs in many countries of the world [12,24,46,47]. Consequently, the existing PTFs are often applied without any calibration or validation in many regions as the result of data constraints [13,46,48].…”