28th International Hydrology and Water Resources Symposium: About Water; Symposium Proceedings, p.2.307-2.314
Estimating soil-hydraulic parameters by the inverse solution of the Richards equation represents a cumbersome numerical procedure. This paper analyses the alternative methodology RIAN, which uses both the Richards equation and an Artificial Neural Network for solving the problem in two steps. Firstly, Radial Basis Functions (RBF) networks are trained for a confined domain of soilhydraulic parameters by simulating multi-step outflow experiments. In the second step, the trained RBF networks are applied to idetifying soil-hydraulic parameters from transient flow patterns which were not included in the training process. These comprehensive verification runs were performed with high accuracy and demonstrated a convincing reliability of the new methodology for the straightforward prediction of soil-hydraulic parameters. In a comparative analysis with the Levenberg- Marquardt optimization, RIAN produces significantly better results without depending upon parameter estimates for starting off the optimization and is shown to be unconditionally stable.
Subsurface drainage ; Computer network architectures ; Neural networks (Neurobiology) ; Inverse problems (Differential equations)--Numerical solutions ; Software
Informit (RMIT Publishing)
View record in Informit