Parallel finite-element implementation for higher-order ice-sheet models

Higher-order models represent a computationally less expensive alternative to the Stokes model for ice-sheet modeling. In this work, we develop linear and quadratic finite-element methods, implemented on parallel architectures, for the three-dimensional first-order model of Dukowicz and others (2010) that is based on the Blatter-Pattyn model, and for the depth-integrated model of Schoof and Hindmarsh (2010). We then apply our computational models to three of the ISMIP-HOM benchmark test cases (Pattyn and others, 2008). We compare results obtained from our models with those obtained using a reliable Stokes computational model, showing that our first-order model implementation produces reliable and accurate solutions for almost all characteristic length scales of the test geometries considered. Good agreement with the reference Stokes solution is also obtained by our depth-integrated model implementation in fast-sliding regimes and for medium to large length scales. We also provide a comprehensive comparison between results obtained from our first-order model implementation and implementations developed by ISMIP-HOM participants; this study shows that our implementation is at least as good as the previous ones. Finally, a comparison between linear and quadratic finite-element approximations is carried out, showing, as expected, the better accuracy of the quadratic finite-element method.