Meeting Documents

A Reanalysis of the Greenland Ice Sheet 1980-2020

Abstract

Numerical ice sheet models can be useful tools to estimate future mass loss from Greenland and Antarctica if the models are well validated. The predictive skill of a model depends not only on model physics but also on the fidelity of the numerical implementation, the quality of data available for validation and the initial conditions and time-dependent boundary forcing. An ever-growing body of observations is providing data for validation, initial and boundary conditions, and targeted observations, resulting in a better understanding of processes within the ice sheet and its interfaces with the atmosphere, the ocean, and the subglacial environment.

In spite of these improvements, ice sheet models are still struggling with even getting the sign of the mass change right in hindcasts. Predicting the evolution of ice sheets is not unlike trying to predict the weather. To produce reliable weather predictions, weather forecasting models are combined with observations using an objective methodology to provide an estimate of the most likely state and its uncertainty, an approach called 'data assimilation'. Data assimilation adds value to observations (by filling in the spatio-temporal gaps in observations) and the model (by constraining it with the observations). Besides improving operational weather forecasts, data assimilation is enabling estimates of the states of the atmosphere (ERA5 MERRA/MERRA-2) and the ocean (ECCO).

Here we present a physically consistent state-estimate of the Greenland Ice Sheet (GrIS) from 1980 to 2020 by combining the ice sheet model PISM and a rich set of observations. To account for uncertainties in parameters and boundary conditions, we generated a large ensemble of hindcasts that assimilate observations of ice front positions and surface speeds. Given observations of mass change, we then filtered the ensemble using Bayesian calibration (particle filtering) to produce the reanalysis.