Evaluation of Water Retention Curves by Regression and Machine Learning Methods

Abstract

The paper presents two methods that simplify the estimation of the water retention curves. The case study is evaluated for the soils of Záhorská lowland in the paper. These methods are based on the supposed dependence of the soil water content on the percentage content of the 1st, 2nd, 3rd and 4th Kopecký grain categories, and the dry bulk density. The representative set of the drying branch of water retention curves was measured using soil samples from the Záhorská lowland region in a laboratory. Particle size distribution and dry bulk density were also determined. In this paper support vector machines and multiple linear regression is compared to estimate the pedotransfer functions that can be used for the prediction of the drying branch of the water retention curve. Both methods were verified on other data set of measured water retention curves than the one which was used for building the models with a close agreement to measured results.