A reduced-order representation of the West Antarctic Ice Sheet

In a warming climate, the potential collapse of the West-Antarctic Ice Sheet is a threat for coastal livelihood as it implies a multimeter sea level rise with unprecedented rate. It is a high-dimensional process, since it involves a number of spatio-temporal variables that interact with each other (e.g. the ocean temperature, the ice thickness and the bedrock elevation). To obtain a better understanding of what initiates a collapse, we propose to apply model reduction techniques to simulations of the Antarctic Ice Sheet under warming scenarios and rely therefore on modern machine learning concepts. Based on this reduced-order representation, we propose an early warning signal that predicts the onset of the marine ice sheet instability in the Thwaites region with a lead time of several decades. This is of key importance to develop meaningful adaptation strategies, which can save lives and critical infrastructure. Most importantly, the early warning signal can be projected back onto the full dimensionality of the problem, thus giving insights into the physics underlying the collapse. This implies that the present work can even serve for developing efficient monitoring systems and mitigation strategies.