D7 Computational Modeling and Simulation

Topic

Computational modeling of the cryosphere translates the physics of glaciological processes into systems of differential equations describing how those processes change across space and time; it then solves these equations numerically on a grid that divides the ice, snow, or terrain into a vast number of small, three-dimensional cells.

An ice-flow model solves the equations governing a viscous fluid whose internal resistance to flow is not constant but varies according to the rate of deformation; by solving these equations on the glacier's three-dimensional grid, the model calculates the distribution of flow velocity, temperature, and internal stress within the ice.

In contrast, a snowpack model determines how heat and mass are transferred through a system of stacked snow layers—each with distinct properties regarding density, temperature, and water content—and uses this inter-layer transfer to predict the snowpack's evolution, layer by layer.

Data assimilation combines actual field observations with computational model outputs using an optimal estimation method that adjusts the model's initial state and internal parameters until the model is better constrained by those observations; this method may employ a variational approach—adjusting the entire model state simultaneously—or an approach based on an ensemble of parallel simulations that correct one another as new observations arrive.

A tipping-point simulation, meanwhile, uses bifurcation analysis combined with a large number of parallel simulations to map which parameter combinations might drive the cryospheric system to transition irreversibly into a new equilibrium state, distinct from the original one. The expected behavior of computational cryosphere modeling, when integrating all these techniques, is that a calibrated model reproduces historical observations with acceptable root-mean-square error and statistical correlation, and that future projections from different models converge on the same direction of change—even if they differ regarding the exact magnitude of that change, reflecting the uncertainty that still exists in both the internal parameters of each model and their mathematical structures. Furthermore, dynamical downscaling allows these projections to be translated to a spatial resolution of between one and ten kilometers—a resolution fine enough to be useful for the management of specific river basins.

ID:2664