Flowline glacier bed estimation with numerical modelling and cost minimization : extending the open source model COMBINE 1D

• Main Author: Schmitt, Patrick
• Format: Master Thesis
• Language: English
• Published: 2021
• Subjects: Gletscher; Talgletscher; Inversion; Modellierung; Flowline Model; Bett; Kostenfunktion; Minimierung; Optimierung; Variationell; glacier; ice cap; modelling; bed; subglacial topography; variational; minimization; optimization; cost function; automatic differentiation; Ice cap
• Online Access: https://resolver.obvsg.at/urn:nbn:at:at-ubi:1-86926

Glacier retreat contributes to sea-level rise and leads to changes in the hydrological regime of glaciated basins. To better quantify these impacts, knowledge of glacier volume, and therefore the glacier bed is crucial. However, in contrast to satellite-based surface measurements, direct measurements of the glacier bed are sparse due to the technical difficulty to conduct such observations. The COst Minimization Bed INvErsion model (COMBINE), introduced by Gregor (2018), is a numerical method that makes use of a physical model of glacier evolution to infer glacier bed from surface observations. It iteratively searches for an optimal glacier bed profile that minimizes the misfit between the modelled and observed glacier surface. The iterative search is made possible thanks to Automatic Differentiation (AD) that computes the gradient of the cost function to minimize. This method was tested and applied on two idealized ice caps. In this thesis, COMBINE is extended with a flowline version (COMBINE 1D) to test the capability of dealing with a large number of valley glaciers with limited data availability. COMBINE 1D supports three different bed shapes along the flowline and was tested with six idealized valley glacier topographies. The use of idealized, “perfect twin” experiments allows to assess the performance of the method alone, irrespective of other uncertainties. Furthermore, two different data availability scenarios were used. The first assumes that two surface measurements (at the start and end of the forward model run) are available, avoiding the need for a spinup run. However, on a global scale, usually only one measurement of the glacier surface is available: the second scenario therefore deals with the task of estimating probable past glacier extents that led to the present day glacier surface. In both scenarios, the dynamic model was driven with a perfectly known mass-balance time series. In all test cases, the bed topography estimates could be improved with respect to the benchmark model, most of it in ...