Presenter/Author Information

Carlos Osorio
Matthew Over
Daniel P. Ames
Yoram Rubin

Keywords

inverse modeling, hydrologic information systems, r, bayesian inversion

Start Date

1-7-2012 12:00 AM

Abstract

Inverse modeling can be a useful – though non-trivial to execute – technique for estimating unknown spatial parameter fields (e.g. hydraulic conductivity) necessary for understanding and modeling groundwater, subsurface contaminant movement, and related quantities. This paper presents the design, implementation, and a test case of an extensible software architecture intended to simplify application of inverse modeling techniques by integrating the Method of Anchored Distributions (MAD), the CUAHSI Hydrologic Information System (HIS) HydroDesktop tool, and the R statistical software package. The test case presents the inversion of model parameters related to the log-transformed hydraulic conductivity, conditional on steady-state pressure head measurements from wells in the domain. The domain is fully saturated and of unit thickness, and is bounded by no flow conditions on two sides and constant head conditions otherwise. HydroDesktop’s GIS-based plugin architecture and inherent access to a large database of hydrologic data serves as the core software framework with a MAD and R integrated custom plugin. MAD is a Bayesian inversion technique for conditioning computational model parameters on relevant field observations yielding probabilistic distributions of the model parameters, related to the spatial random variable of interest, by assimilating multi-type and multi-scale data. The implementation of a desktop software tool for using the MAD technique is expected to significantly lower the barrier to using inverse modeling in education, research, and resource management. The HydroDesktop MAD plugin is being developed following a community-based, open-source approach that will help both its adoption and long term sustainability as a user tool. This presentation will briefly introduce MAD, HydroDesktop, and the MAD plugin and software development effort.

Share

COinS
 
Jul 1st, 12:00 AM

An Extensible Inverse Modeling Software Architecture for Parameter Distribution Estimation

Inverse modeling can be a useful – though non-trivial to execute – technique for estimating unknown spatial parameter fields (e.g. hydraulic conductivity) necessary for understanding and modeling groundwater, subsurface contaminant movement, and related quantities. This paper presents the design, implementation, and a test case of an extensible software architecture intended to simplify application of inverse modeling techniques by integrating the Method of Anchored Distributions (MAD), the CUAHSI Hydrologic Information System (HIS) HydroDesktop tool, and the R statistical software package. The test case presents the inversion of model parameters related to the log-transformed hydraulic conductivity, conditional on steady-state pressure head measurements from wells in the domain. The domain is fully saturated and of unit thickness, and is bounded by no flow conditions on two sides and constant head conditions otherwise. HydroDesktop’s GIS-based plugin architecture and inherent access to a large database of hydrologic data serves as the core software framework with a MAD and R integrated custom plugin. MAD is a Bayesian inversion technique for conditioning computational model parameters on relevant field observations yielding probabilistic distributions of the model parameters, related to the spatial random variable of interest, by assimilating multi-type and multi-scale data. The implementation of a desktop software tool for using the MAD technique is expected to significantly lower the barrier to using inverse modeling in education, research, and resource management. The HydroDesktop MAD plugin is being developed following a community-based, open-source approach that will help both its adoption and long term sustainability as a user tool. This presentation will briefly introduce MAD, HydroDesktop, and the MAD plugin and software development effort.