Comment on egusphere-2023-635

Abstract

<strong class="journal-contentHeaderColor">Abstract.</strong> Many scientific and societal questions that draw on ice sheet modelling could be best addressed by sampling a wide range of potential climatic changes and realizations of internal climate variability. For example, coastal planning literature demonstrates a demand for probabilistic sea-level projections with quantified uncertainty. Further, robust attribution of past and future ice sheet change to specific processes or forcings requires a full understanding of the space of possible ice sheet behaviors. The wide sampling required to address such questions is computationally infeasible with sophisticated numerical climate models at the resolution required to accurately force ice sheet models. Stochastic generation of climate forcing of ice sheets offers a complementary alternative. We construct a stochastic generator of Greenland Ice Sheet surface mass balance in time and space. We find that low-order autoregressive models are sufficient to accurately reproduce the interannual variability in process-model simulations of recent Greenland surface mass balance at the glacier-catchment scale. We account for spatial correlations among glacier catchments using sparse covariance techniques, and we apply an elevation-dependent downscaling to recover gridded surface mass balance fields suitable for forcing an ice sheet model while including feedbacks to ice sheet surface elevation. The efficiency gained in the stochastic method supports large ensemble simulations of ice sheet change in a new stochastic ice sheet model. We provide open source Python workflows to support use of our stochastic approach for a broad range of applications.