 |
 |
 |
 |
|
The Culebra Dolomite at the Waste Isolation Pilot Plant (WIPP) is a potential migration pathway for radionuclides that may leave the repository. Due this role as a potential pathway, the Culebra has been the focus of a large amount of groundwater flow and transport modeling. In particular, the Culebra has served as a test bed for a number of different stochastic inverse approaches to estimating transmissivity fields conditioned to both transmissivity and head measurements. In the majority of these stochastic inverse approaches, the non-uniqueness of the transmissivity parameter estimates has been acknowledged, but it has not been explicitly accounted for in the groundwater travel time predictions made on the resulting transmissivity fields.
This work is aimed at developing and applying stochastic inverse modeling techniques to the generation of T fields that can explicitly account for parameter non-uniqueness in the resulting groundwater transport predictions. One technique for doing this is called "predictive estimation," PE, and has been developed in the PEST parameter estimation software. The PE technique allows for maintaining a calibration to measured head values within a user specified tolerance while simultaneously minimizing the predicted groundwater travel time. The potential of this technique for optimizing 100 transmissivity fields while minimizing travel time has been explored on a test data set (McKenna et al., in review). See animation of groundwater streamlines through a calibrated T field.
The computational expense of calibrating 100 different T fields using 10's of adjustable parameters is daunting. We custom built a heterogeneous parallel cluster of 17 Desktop Personal Computers(PCs) running Linux to solve this large-scale optimization problem.
The cluster is, shown above, with its builders, Lane Yarrington and David Hart. This cluster was built for several hundred dollars by recovering retired PCs from the Sandia reapplication yard, scavenging as much memory as possible from other retired PCs and using level I routers to connect the PCs. The majority of the PCs are PentiumII models running at a clock speed of 200 MHz.
We continue to explore the power of inverse modeling techniques for creation of transmissivity fields and the use of low-cost super computers with application to performance assessment calculations at the WIPP site. In 2002, we have constructed a new PC cluster with 16, 1.9Ghz processors
Papers:
- McKenna, S.A., J. Doherty and D.B. Hart, (in press) Non-Unique Transmissivity Field Calibration and Predictive Transport Modeling, accepted for publication in: Journal of Hydrology, August, 2001.
|
|
|
