Groundwater flow parameter estimation using refinement and coarsening indicators for adaptive downscaling parameterization

• Published in: Advances in water resources Vol. 100; pp. 139 - 152
• Main Authors: Hassane, Mamadou Maina F.Z., Ackerer, P.
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.02.2017; Elsevier Science Ltd; Elsevier
• Subjects: Coarsening; Coarsening indicators; Downscaling parameterization; Earth Sciences; Finite element method; Groundwater; Groundwater flow; Hydrology; Identification methods; Indicators; Inverse method; Inverse problems; Inversions; Optimization; Parameter estimation; Parameter identification; Parameter robustness; Parameterization; Parameters; Refinement indicators; Sciences of the Universe; Water flow; Refinement indicators; Coarsening indicators; Downscaling parameterization; Groundwater flow; Inverse method
• ISSN: 0309-1708; 1872-9657
• DOI: 10.1016/j.advwatres.2016.12.013

•An adaptive parameterization for parameter estimation is described.•Refinement and coarsening indicators for parameterization are extended to interpolation based parameterization.•Adaptive parameterizations are applied to some synthetic examples. In the context of parameter identification by inverse methods, an optimized adaptive downscaling parameterization is described in this work. The adaptive downscaling parameterization consists of (i) defining a parameter mesh for each parameter, independent of the flow model mesh, (ii) optimizing the parameters set related to the parameter mesh, and (iii) if the match between observed and computed heads is not accurate enough, creating a new parameter mesh via refinement (downscaling) and performing a new optimization of the parameters. Refinement and coarsening indicators are defined to optimize the parameter mesh refinement. The robustness of the refinement and coarsening indicators was tested by comparing the results of inversions using refinement without indicators, refinement with only refinement indicators and refinement with coarsening and refinement indicators. These examples showed that the indicators significantly reduce the number of degrees of freedom necessary to solve the inverse problem without a loss of accuracy. They, therefore, limit over-parameterization.