NEWS: PISM and PETSc 3.7
|Modeling the evolution of the Juneau Icefield between 1971 and 2100 using the Parallel Ice Sheet Model (PISM)|
|investigators:||F. Ziemen, R. Hock, A. Aschwanden, C. Khroulev, C. Kienholz, A. Melkonian, and J. Zhang|
|journal:||Journal of Glaciology|
The large icefields of North America are geometrically-complex. They are exposed to heterogeneous climatic conditions. They are intermediate in size between the individual mountain glaciers and ice sheets which are common modeling targets. And thus they are a good test of the capability and effectiveness of ice dynamics models.
This paper studies the 4000 km^2 Juneau ice field straddling the USA/Canada border using PISM. Perhaps it is no surprise that the modeled outcome of a future warming scenario is (1) loss of much of the glacier volume, and (2) strong dependence of the outcome on the precipitation and surface mass balance inputs. Climatic observations are too sparse and unrepresentative to allow use of interpolated values for climate inputs, though this was attempted. Instead, atmospheric climate model (20 km Weather Research and Forecasting Model) output was used. Simulated and observed surface mass balance gave good agreement only after precipitation adjustments to account for unresolved orographic effects. Under a RCP6.0 emission scenario, the PISM results then project a decrease in ice volume by 58–68% by 2099 compared with 2010. If the modeled 2070–99 climate is held constant beyond 2099, the icefield is eliminated by 2200. With constant 1971–2010 climate, the icefield stabilizes at 86% of its present-day volume.
PETSc 3.7 was released on April 25, 2016. We are currently working on making PISM compatible with PETSc 3.7 and will announce it here as soon as possible.
In the meantime, please install petsc 3.6.4 from here. PISM version 0.7 (
stable0.7 branch) works with any PETSc 3.5.X and higher.
The paper is based on PISM simulations with grid resolution down to 600 m over the entire Greenland ice sheet. To start, each of an initial ensemble of 14 lower-resolution (1500 m) experiments has a single ice-sheet-wide value for all parameters. The best of these, in an ice-sheet-wide measure, is re-run at the 600 m resolution and various coarser resolutions. The quality of this flow model for 29 outlet glaciers is assessed; each outlet glacier sees the same physics. The main result is that the majority of the outlet glaciers show strong correlation between modeled and present-day-observed velocity, when it is compared along cross-flow and near-ocean profiles.
Before this paper one might suppose, based on the most prominent literature on the subject, that a detailed, measurably-accurate, outlet-glacier-resolving model of the present-day velocity of an entire ice sheet was dependent both on removing shallow assumptions from the stress balance and on tuning a very large number of basal parameters. Both of these “required” properties would be very bad news for the prospect of using ice sheet simulations to do science! On the one hand, Stokes models are computationally-expensive, while on the other hand only present-day, and not past or future, data are available to set all these basal parameters through inversion.
Such a pessimistic view turns out to be substantially false. Aschwanden et al. (2016) show that four things do matter: (i) an accurate map of bedrock topography, (ii) a stress regime in which viscous membrane stresses are part of the balance with basal sliding resistance, (iii) an energy-conservation-driven basal stress model derived (conceptually) from a model of a wet, pressurized, deformable basal layer, and (iv) high model resolution over all areas of the ice sheet where sliding is possible and/or steep/rough basal topography exists.
NASA IceBridge missions, and the mass-conserving-bed technology of Morlighem et al (2014), are shown by this paper to represent major progress on item (i). Items (ii) and (iii) are properties of the PISM continuum model, and item (iv) of its implementation as parallel-scalable software. Certainly all of these “things that matter” are improvable. More-complete stress balances and the use of inversion of present-day velocities will both be essential to improvements. The main idea remains, however: if the modeled flowing ice has the right bottom geometry, and if the dynamical model has certain key features, then the resulting dynamics are already inside the ballpark!
PISM is jointly developed at the University of Alaska, Fairbanks (UAF) and the Potsdam Institute for Climate Impact Research (PIK). For more about the team see the UAF Developers and PIK Developers pages.
UAF developers, who are in the Glaciers Group at the GI, are supported by NASA's Modeling, Analysis, and Prediction and Cryospheric Sciences Programs (grants NAG5-11371, NNX09AJ38C, NNX13AM16G, NNX13AK27G) and by the Arctic Region Supercomputing Center.