PNNL-13501 - Pacific Northwest National Laboratory

Study Control Number: PN99076/1404
Vadose Zone and Pore-Scale Process Modeling
Mark D. White, Jim A. Fort, Matt Rosing
Numerical simulation on massively parallel computers provides scientists and engineers with the analytical tools
necessary to assess risks and devise effective remediation schemes for subsurface contamination. Computational tools are
being developed on this project that allow researchers to accurately represent geologic and engineered subsurface
features, while maintaining the efficiencies of parallel computing.
Project Description
The objective of this project is to develop computational
capabilities for massively parallel computers for modeling
multifluid flow and transport in the subsurface on
physical domains that contain complex geologic features
and engineered structures. Of particular interest are
preferential pathways, such as clastic dikes in the deep
vadose zone at the Hanford Site, which potentially
provide conduits for fluid and contaminate migration to
the groundwater. Conventional methods for modeling
complex geometries in the subsurface involve computing
on unstructured grids using finite-element discretization
of the mathematical equations that describe multifluid
flow and transport through geologic media. In an attempt
to maintain scalability and parallel computing efficiency,
this project is directed at solving subsurface flow and
transport problems using multiple grid blocks, where each
grid block (sub-grid) is a curvilinear orthogonal structured
grid. The critical issues are to develop parallel computing
schemes that balance the computational load and
minimize communications across the processors on a
massively parallel computer. We have implemented
capabilities for computing on curvilinear coordinate
systems into a suite of subsurface flow and transport
simulators and developed FORTRAN preprocessor
(Rosing 1999) directives and interpreters for multi-block
grids of structured blocks. When applied to
environmental restoration and stewardship, the resulting
software will provide a stronger scientific rationale for
management decisions concerning the vadose zone,
particularly at the Hanford Site, where preferential
pathways and subsurface structures are critical issues.
Introduction
The vadose zone is defined as the hydrologic region that
extends from the soil surface to the water table.
Preferential pathways within the vadose zone include
natural geologic structures (clastic dikes), engineered
structures (monitoring wells and boreholes), and unstable
multifluid systems (high density brines migrating through
freshwater aquifers). Heterogeneous subsurface
environments of particular concern to this project include
geologic discontinuities (faults and embedded structures)
and engineered structures (underground storage tanks).
Accurate representations of the geologic structure,
stratigraphy, and buried engineered structures are critical
to numerically modeling subsurface flow and transport
behavior. Scientists and engineers interested in assessing
risks and devising remediation strategies for contaminants
migrating through the vadose zone toward the
groundwater need high performance analytical tools with
capabilities for modeling subsurface complexities.
Numerical models of subsurface flow and transport
operate on tellesations of the physical domain, where
hydrologic and thermodynamic properties are computed
at discrete points (grid cells), as an approximation to the
continuum of physical space. Analyses of contaminant
migration through three-dimensional, field-scale domains
requires the computational capacity of massively parallel
computers. For example, to numerically model a cluster
of six underground storage tanks, (typical of those on the
Hanford Site, which are known to have leaked radioactive
contaminants into the vadose zone) at 1-m resolution in
three-dimensions, requires approximately 1 million grid
cells. Depending on the assumptions taken to develop the
conceptual model for the simulation, each grid cell can
have between one and five unknowns which must be
solved implicitly to resolve the system nonlinearities.
Memory and execution-speed limitations make executing
this scale of problem prohibitive on sequential computers.
However, these types of problems have been successfully
solved on a distributed-memory 512-processor IBM-SP
parallel computer using 64 processors. Incorporating
subsurface complexities into the computational domain by
increasing the grid resolution within the region of the
complexity (clastic dike, underground storage tank) can
increase computational requirements beyond the range of
Earth System Science 259