Inverting glacier thickness in High Mountain Asia with a deep-learning-based ice flow model

Mountain glaciers provide an irreplaceable water resource in High Mountain Asia, with a significant proportion of water input to rivers coming from glacial meltwater. However, the volume of water held in these glaciers and the predicted evolution of the glaciers over the coming decades is subject to great uncertainty. 

Previous global ice thickness inversion studies have used low-order models (such as the shallow ice approximation, which is known to be locally unreliable on mountain glaciers) to describe the relationship between ice velocity and thickness. Furthermore, the reliability of the resulting thickness products in High Mountain Asia is severely limited, since at the time they were produced, only an extremely small dataset of measured thicknesses in that region was available for constraint and validation. Lastly, time-dependent ice thickness simulation runs often show an initial ‘shock’ in modelling a glacier’s evolution, due to the lack of consistency between the ice flow physics and existing thickness products, leading to unreliable results.

To construct more accurate thickness maps for selected glaciers, we use the Instructed Glacier Model, a novel deep-learning-based high-order ice flow model with the capability to invert observed glacier surface velocity for ice thickness. This inversion method, which utilises gradient-based optimization techniques, additionally allows for the inclusion of observed thicknesses to constrain the thickness field. 

A new airborne radar method for measuring ice thickness has been deployed in the Himalayas near Mount Everest, unlocking new possibilities for thickness inversion in this region, which has historically not been well covered by in-situ observations. Here, we use the data from this aerial survey to constrain the thickness inversion of the Instructed Glacier Model.

After showing that contemporary ice thickness products are generally inaccurate on High Mountain Asia glaciers, we present new inverted thickness maps for the 13 glaciers which have observations in this new dataset. We assess the accuracy of the results using a subset of the available data as validation, and demonstrate that our results show significant improvement over earlier thickness products.