Automated calibration methodology to avoid convergence issues during inverse identification of soil hydraulic properties

• Published in: Advances in engineering software (1992) Vol. 173; p. 103278
• Main Authors: Kuraz, Michal, Jačka, Lukáš, Ruth Blöcher, Johanna, Lepš, Matěj
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2022
• Subjects: Automatic calibration; Computational issues in geosciences; Convergence issues; Inverse modeling; Richards equation; Single ring experiment; Soil hydraulic properties; Soil hydraulic properties; Computational issues in geosciences; Automatic calibration; Convergence issues; Single ring experiment; Richards equation; Inverse modeling
• ISSN: 0965-9978
• DOI: 10.1016/j.advengsoft.2022.103278

Inverse modeling of in-situ experiments is already a standardized approach for identifying various types of material parameters. In this contribution we are focused on the single ring (hereafter SR) infiltration experiment, which is a standard and robust dynamic field experiment. The steady state part of this experiment is traditionally used for the identification of saturated hydraulic conductivity. We explore here the possibility of extending the applicability of this experiment for evaluating the hydraulic parameters for unsaturated conditions from an unsteady part of this experiment for the top soil layer using inverse analyses of the governing flow motion equation. The problem of SR infiltration is governed by the quasilinear Richards equation. We present a new scanning methodology to avoid convergence issues with the nonlinear operator, originating from difficult combinations of input parameters, which can be hard to avoid when automatically analyzing a broad parameter space. We validated our methodology with virtual infiltration problems for clay and sand, and applied it on real-world SR infiltration data. To evaluate non-uniqueness, local optima were identified and mapped using a modified genetic algorithm with niching. Our results show the existence of multimodality in, both, the benchmark problems and the real-world problem. This is an important finding as local optima can be identified, which are not necessarily physical and also for systems that do not exhibit multimodal grain size distributions. The identified local optima were distinct and showed different retention and hydraulic conductivity curves. The most physical set of SHP could be identified with the knowledge of the saturated water content.