GC33A-0480:
Challenges to large-scale simulations of permafrost freeze-thaw dynamics

Wednesday, 17 December 2014
Nathan Collier1, Gautam Bisht2 and Jitendra Kumar1, (1)Oak Ridge National Laboratory, Oak Ridge, TN, United States, (2)Lawrence Berkeley National Laboratory, Berkeley, CA, United States
Abstract:
In an effort to model the dymanics of the permafrost freeze and thaw process in the Alaskan tundra, we have implemented a finite volume method which approximates the evolution of a coupled surface/subsurface mass and energy balance within PFLOTRAN--an open source, state-of-the-art massively parallel subsurface flow and reactive transport code.
While this system is studied in the literature at one scale, we encounter many undocumented pitfalls as we exercise the model at high resolution and force using realistic datasets from the field sites. These realistic simulations for field sites near Barrow, Alaska expose the model to a wide range of moisture and thermal states that are not tested in published studies. For example, the conventional upwinding of the relative permeability used in the Darcy flux computation can yield a flow into a frozen cell. We also find that infiltration, sources, and sinks must be carefully regulated as flow into frozen portions of the domain, or out of dry or frozen regions can cause unphysical states in the simulation which cause failure. Many straight-forward solutions are not smooth which produce discontinuities in the Jacobian of the nonlinear residual. These difficulties represent a current hurdle to running large-scale permafrost dynamics simulations. We describe these challenges and present approaches to overcoming them in the pursuit of a scalable scheme.