Leveraging Massively Parallel Computing to Evaluate the Effect of Sorption and Heterogeneous Permeability On Natural Attenuation of Uranium at the Hanford 300 Area.

Massively parallel computation is becoming more accessible as the race to the exascale begins.  This talk demonstrates the use of HPC (high performance computing) applied to modeling a uranium plume at the Hanford 300 Area which borders the Columbia River.  Particularly perplexing are the longevity of the plume and the factors controlling its mobility at the site.  Model simulations in the early 1990's had predicted that the should dissipate is approximately ten years.  However, uranium concentrations still exceed the maximum contaminant level at the site.  Calculations were carried out on the Cray XT4/5 Jaguar at ORNL with up to 40,000 processor cores using the computer code PFLOTRAN developed under a DOE SciDAC-2 project. The calculations demonstrated that hexavalent U(VI) is discharged to the river at a highly fluctuating rate in a ratchet-like behavior as the Columbia River stage rises and falls.  The high frequency fluctuations must be resolved in the model to calculate the flux of U(VI) at the river boundary.  By time averaging the instantaneous flux to average out noise superimposed on the river stage fluctuations, the cumulative U(VI) flux to the river is found to increase approximately linearly with time.  The model simulations compared the effects of sorption based on equilibrium and multirate formulations with the case of no sorption and investigated the role of heterogeneity on the flux of uranium into the Columbia River. Key to matching the observed uranium flux of 50 kg/y was incorporation of nonlabile uranium leaching at a slow rate of release. In addition, it was found important to account for the reduced conductance in the hyporheic zone to reasonably match the piezometric head obtained from well data.