Estimating Greenland´s present and future ice thickness using inverse modeling and high-resolution satellite data

Time period:

1 January 2025 – 31 December 2029

Project leader:

Ward van Pelt

Project member:

Regine Hock, Rickard Pettersson, Katrin Lindbäck, Ruth Mottram

Funder:

Swedish National Space Agency

Type of award:

Project grant

Total funding:

4,951,000 SEK

With this project we aim to generate new bed height and thickness maps for the Greenland ice sheet, and to simulate the evolution of the ice sheet up to 2300. In recent decades the Greenland ice sheet has contributed considerably to sea level rise and expectations are that the ice sheet will be the dominant cryosphere contributor in the remainder of the 21st century. The Greenland Ice Sheet, home to ~7 m of sea level equivalent, is expected to contribute 5-33 cm to sea level by 2100 and Greenland is likely to be ice-free within a millennium. Predictions of future ice sheet mass and volume change are commonly done with ice sheet models that describe ice flow physics and conditions at the surface, frontal and basal boundaries. A major source of uncertainty in ice sheet modelling, contributing to errors in sea level rise predictions, stems from difficulties to set initial conditions in the present-day that are needed as a starting point for forecasting runs. Knowledge of bed topography and friction is essential for accurate simulation of ice motion and thickening/thinning, but observations to constrain bed conditions are limited due to poor accessibility. To overcome this issue, inverse methods can be used that adjust input for numerical ice flow models (bed topography and friction fields) in a stepwise manner by matching simulated and satellite-observed output (ice velocity and thickening/thinning). The use of such methods has been stimulated by the recent advent of (open-access) high-resolution satellite data products of ice sheet surface conditions. In addition to inferring bed conditions, inverse methods are a fast and powerful tool to create initial conditions for prognostic simulations, circumventing the need for tens of thousands of years of model initialization ("spin-up") as in traditional methods. Nevertheless, thickness inversion and prognostic modelling with a sophisticated ice flow model remains computationally demanding, especially when done with high spatial resolution and with an ice flow model with a detailed description of flow mechanics. To considerably reduce computational time, a machine-learning-based emulator can be constructed that replaces the original ice flow model with a negligible loss in simulation accuracy. The three main aims in this project are to 1) construct a new emulator of the higher-order physics version of the Parallel Ice Sheet Model (PISM), 2) generate a present-day thickness and bed map of the Greenland Ice Sheet,  and 3) simulate the future mass loss and contribution of the Greenland Ice Sheet to sea level rise up to 2300. A novel, fast and versatile inverse approach, developed with support in a previous SNSA-funded project (dnr 189/18; 2018-2023), will be used. After recent applications on mountain glaciers across the Arctic and sub-Arctic, is now optimized for use on the larger Greenland Ice Sheet, while making optimal use of the newest high-resolution satellite data products. This ultimately allows for modelling the present and future of the ice sheet with unprecedented spatial detail and reduced model uncertainty, thereby contributing to reduced uncertainty in sea level rise predictions for a range of future emission scenarios. The high speed of the emulator enables us to perform experiments with numerous different parameter settings, which enables better calibration of the inverse method, as well as uncertainty quantification of present-day and future thickness maps.