Impact of time-dependent data assimilation on ice flow model initialization and projections: a case study of Kjer Glacier, Greenland

Abstract

Ice sheet models are often initialized with data assimilation of present-day conditions, in which unknown model parameters are estimated using the inverse method. While assimilation of snapshot observations has been widely used for regional- and large-scale ice sheet simulations, data assimilation based on time-dependent data has recently started to emerge to constrain model parameters while capturing the transient evolution of the system. However, this method has been applied only to a few glaciers with fixed ice front positions, using spatially and temporally limited observations, and has not been applied to marine-terminating glaciers of the Greenland Ice Sheet that have been retreating over the last 30 years. In this study, we assimilate time series of surface velocity into a model of Kjer Glacier in West Greenland to better capture the observed acceleration over the past 3 decades. We compare snapshot and transient inverse methods and investigate the impact of initialization procedures on the parameters inferred, as well as model projections. We find that transient-calibrated simulations better capture past trends and better reproduce changes after the calibration period, even when a short period of observations is used. The results show the feasibility and clear benefits of a time-dependent data assimilation for initializing ice sheet models. This approach is now possible with the development of longer observational records, though it remains computationally challenging.