| Summary: | Information about subglacial topography is essential for modelling ice flow and estimating the potential contribution of glaciers to sea-level rise. In-situ measurements of glacier bed elevation are costly, cumbersome and only sparsely available for mountain glaciers and ice caps. Here, we present the Cost Minimization Bed Inversion model (COMBINE), which estimates bed topography from surface topography and glacier outlines alone. A distributed shallow ice model (forward model) provides an estimate of the ice surface topography as a function of the unknown bed topography. A coarse first guess of the bed elevation is then iteratively optimized by minimizing a cost function of surface misfit. The gradient of the cost function obtained by Automatic Differentiation (AD) is used by the minimization algorithm to efficiently converge to a (local) cost minimum. To test the method, two synthetic ice caps were created using the forward model driven by a prescribed climate and realistic topography. In this surrogate world where everything is known a priori, the performance of COMBINE can be assessed in various use case scenarios. With perfectly known surface elevation, errors in the reconstructed bed are found to be small and comparable to uncertainties obtained by GPR measurements. Repeating these experiments with varying first guess gives similar results, indicating that the method is robust to this arbitrary choice. Further experiments with noise imposed on the provided surface elevation show decreasing performance of the bed inversion with increasing noise. However, COMBINE is able to partly compensate for this noise by imposing physical constraints on unrealistic inputs. COMBINE can also be extended to use ice thickness observations to better constrain the bed estimation ("data assimilation"). This leads to an improvement of the bed reconstruction but depends on the location and extent of the observations. COMBINE and its variational principle can be used on real mountain by Philipp Gregor Abweichender Titel laut ...
|
|---|