Skip to main content
U.S. flag

An official website of the United States government

Publication Date
19 April 2018

Evaluating Ice Sheet Model Initializations

Design and results of the ice sheet model initialization experiments initMIP-Greenland: an ISMIP6 intercomparison.
Print / PDF
Powerpoint Slide
RMSE of initial modeled ice thickness compared to observations for all participating models. Interpretation of the diagnostics requires distinction between the different initialization techniques, but in general, there is a large impact from the initial state on projected sea level rise of by the model. The CISM model used a spin up technique and BISICLES used a data assimilation of velocity.

An international community project to compare, evaluate, and improve the initialization techniques used in ice sheet modeling.


Provides the first community benchmark to asses new ice sheet model initialization methods, which strongly influence the results of model projections, and identified the initialization methods which will be needed for the 100-200 year projections planned for CMIP6. 


Large-scale Greenland ice sheet projections of sea level, such as those run during the ice2sea and SeaRISE initiatives, have shown that the initial state of the Greenland ice sheet used in the model has a large effect on the projected sea level rise, leading to large uncertainties. The initMIP-Greenland intercomparison exercises allow ice sheet modelers to compare, evaluate and improve their initialization techniques and estimate their associated uncertainties. Two experiments were performed by each modeling group: (1) produce an initial present-day state of the Greenland ice sheet and (2) simulate the response of this initial state in two idealized forward experiments. The forward experiments allow modelers to evaluate the model drift and response to a large perturbation. Though this effort, the initialization methods, and datasets have been identified that will be needed for the 100-200 year projections planned for CMIP6. A freely available benchmark dataset has also been generated to help modelers evaluate and improve future initialization methods.

Point of Contact
Joseph H. Kennedy
Oak Ridge National Laboratory (ORNL)
Funding Program Area(s)