Prediction of Water Retention Curves Using Neural Networks

Thesis Type: Postgraduate

Institution Of The Thesis: Middle East Technical University, Faculty of Engineering, Department of Civil Engineering, Turkey

Approval Date: 2020

Thesis Language: English


Principal Supervisor (For Co-Supervisor Theses): Nabi Kartal Toker

Co-Supervisor: Onur Pekcan


Soil water-retention curve (WRC) relates tension in soil water to soil suction. WRC information has pivotal importance for revealing the behavior of unsaturated soils. Methods for obtaining retention curves are either too expensive or time consuming. Instigated by the demand on fast predictions, this study expressed a composition of a 88 NN designs opened up with (i) hyperparameter tuning, (ii) reexamination of expressions of WRC and GSD. The data was extracted from UNSODA database which encompasses a broad type of soils and widely varied suction ranges, without excluding or subsampling any of the textural group or suction ranges of observations as most of the existing studies did. This inclusive approach rendered the originality of the study and yet spawned a series of problems in methodology and low accuracy in predictions. Among those models, Fredlund and Xing (1994) model held the highest accuracy measure, , which varied from 0.51 to 0.85.