Error sources in basal yield stress inversions for Jakobshavn Isbræ, Greenland, derived from residual patterns of misfit to observations

• Published in: Journal of Glaciology
• Main Authors: MARIJKE HABERMANN, MARTIN TRUFFER, DAVID MAXWELL
• Format: Article in Journal/Newspaper
• Language: English
• Published: Cambridge University Press 2017
• Subjects: glacier modeling; ice-sheet modeling; subglacial processes; Environmental sciences; GE1-350; Meteorology. Climatology; QC851-999; Greenland; Jakobshavn Isbræ; glacier; Ice Sheet; Jakobshavn; Jakobshavn isbræ; Journal of Glaciology
• Online Access: https://doi.org/10.1017/jog.2017.61; https://doaj.org/article/a535d76128074176b5af29f0ffc7f7c1

The basal interface of glaciers is generally not directly observable. Geophysical inverse methods are therefore used to infer basal parameters from surface observations. Such methods can also provide information about potential inadequacies of the forward model. Ideally an inverse problem can be regularized so that the differences between modeled and observed surface velocities reflect observational errors. However, deficiencies in the forward model usually result in additional errors. Here we use the spatial pattern of velocity residuals to discuss the main error sources for basal stress inversions for Jakobshavn Isbræ, Greenland. Synthetic tests with prescribed patterns of basal yield stress with varying length scales are then used to investigate different weighting functions for the data-model misfit and for the ability of the inversion to resolve details in basal yield stress. We also test real-data inversions for their sensitivities to prior estimate, forward model parameters, data gaps, and temperature fields. We find that velocity errors are not sufficient to explain the residual patterns of real-data inversions. Conversely, ice-geometry errors and especially simulated errors in model simplifications are capable of reproducing similar error patterns and magnitudes. We suggest that residual patterns can provide useful guidance for forward model improvements.