Contamination of groundwater by potentially toxic organic liquids is one of the more serious environmental problems facing the Nation. The release of such liquids has been widespread, originating from sources ranging from gasoline service stations to large industrial facilities. Once trapped in the subsurface these liquids serve as long- term sources of aquifer contamination. Remediation schemes include optimal liquid recovery pumping schemes, solvent and surfactant flushing, and steam injection.
Hierarchical models of groundwater flow and associated chemical transport are needed at the local, basin, and regional scales to assess waste transport and remediation schemes. Current models are limited computationally in two ways: first, models can only include a limited number of physical and chemical processes before computing runtimes become prohibitive; second, models lack adequate field data to characterize geologic and chemical factors that influence contaminant behavior. High performance computing systems are needed both to model complex transport processes and to overcome data limitations by using computationally intensive parameter estimation techniques.
The Partnership in Computational Science (PICS) consortia project funded by the DOE HPCC program is exploring new approaches to modeling groundwater flow at waste sites in extremely complex hydrogeologic settings. The consortia consists of members from Brookhaven National Lab, Oak Ridge National Lab, Rice University, SUNY-Stony Brook, Texas A&M, and the University of South Carolina. A map of the site most extensively studied (WAG 6) is given in Figure A. Researchers at ORNL have adapted an existing finite element code to run on an Intel scalable parallel distributed- memory system. By using a preconditioned conjugate gradient solver, users were able to distribute portions of the model grid to different processors and greatly reduce solution time. Using a 40,000 node grid (Figure B), they are obtaining anisotropy factors and boundary conditions that can be used to model smaller areas and test the effects of remediation schemes on models of the waste areas. One such remediation scheme is to cover areas with a water- impermeable membrane (cap) to eliminate surface recharge (Figure C). A calculation of the water table following a capping is given in Figure D. Such results point to the specific piezometric wells for which experimental indications of capping effects should be sought.
Pacific Northwest Laboratory is also exploring the use of scalable massively parallel quantum chemistry algorithms to improve methods for redesigning enzymes to better degrade pollutants, for extracting contaminants from soils, and for burning halohydrocarbons.
Researchers at the EPA's Robert S. Kerr Environmental Research Laboratory and collaborators at various universities have focused efforts on developing process understanding using complementary numerical models and laboratory studies. Implementation of numerical models has followed approaches originally developed in the petroleum industry and advanced multiprocessor techniques. A large scale physical model is currently being used to evaluate the numerical model. Release of an organic liquid is planned for the physical model, followed by testing of remedial technologies. A remediation approach involving surfactant flood, followed by vapor extraction above the water table, is being designed by bench scale laboratory studies and adaptation of a sophisticated numerical model developed by the University of Texas at Austin. Other remediation models are also under development. Joint field and modeling studies in cooperation with the USDA are underway to assess the impact of agricultural chemicals on water resources in the midwest.
Groundwater models will not eliminate the need for adequate field data but will continue to assist scientists in understanding these complex systems, thus contributing to reducing the risks to public water supplies.