Scholarly record

ESTIMATING SOIL HYDRAULIC PROPERTIES FOR UNSATURATED ZONE USING GRADIENT AND METAHEURISTIC METHODS

Roxana-Gabriela Dobre

Abstract

For modelling the water fluxes through the vadose zone, it is necessary to know the soil hydraulic properties. Because many of the soil parameters cannot be measured directly in the field, inverse modeling methods are used to find the most appropriate solution. For a limited number of parameters, the minimization approach provided by Marquardt-Levenberg (ML) has become a usual technique in nonlinear least-squares fitting in soil and hydrological studies. Classical gradient-based methods find the solution in the neighborhood of a starting point. For a larger number of parameters, local search methodologies cannot provide a unique solution and may fail to locate the global minimum, so these cases require the use of global optimization methods. Nature-inspired metaheuristics represents a modern approach for estimation of soil hydraulic properties. The metaheuristic methods have the major advantage that do not require gradients and an initial solution, so they can search large ranges of candidate solutions to find the optimal one. For obtaining accurate estimations of the soil hydraulic functions given by Mualem -van Genuchten (MVG) expressions in the unsaturated zone, we investigate the performance of the global and local optimization methods. We illustrate these developments with two case studies based on laboratory and field measurements. Using the laboratory provided data, the study compared the potential of a gradient based method on generalized reduced gradient algorithm (GRG) implemented in Microsoft Excel Solver and a Genetic Algorithm (GA) using R package GA. GAs are modified and adapted for solving the inverse problem in the unsaturated zone. The function related to soil water retention curves was implemented in R package GA in order to estimate MVG unsaturated hydraulic soil parameters. Also, using time domain reflectometry (TDR) measured water contents and soil textural class information obtained in laboratory from an experimental field plot in Padina area, Romania, inverse estimation of soil MVG parameters was necessary for an accurate simulation of infiltration from snowmelt. Rosetta provides the input parameters, using pedotransfer functions based on sand, silt and clay percentages obtained from laboratory analysis, for ML inverse optimization code implemented in HYDRUS-1D. The results outline that GA is able to solve the inverse problem of estimating the soil hydraulic properties for the unsaturated zone and have good convergence capabilities in estimating global best solutions.

Publication Impact Profile

• Citations
• CrossRef - Citation Indexes: 1
• Scopus - Citation Indexes: 1

• Captures
• Mendeley - Readers: 3

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2018/3.2/s13.022

Publication details

Back to publication list