Estimating the risk of Antarctic ice shelf collapse using Bayesian nonparametric statistical modelling.

Ice shelves form the floating extensions of the East, West and Antarctic Peninsula ice sheets and play a crucial role in regulating ice sheet flow and global sea-level rise. The presence of an ice shelf provides resistive (back or buttressing) forces, which partly compensate the driving forces of inland ice flowing to the sea1. Since the mid-20th century and during the satellite 'big data' observational era, several ice shelves in the Antarctic Peninsula have substantially retreated or even catastrophically collapsed2. This has resulted in acceleration of inland ice flow3 by a factor of up to eight4, with some basins still adjusting to pre-collapse velocities some twenty years after disintegration5. Acceleration of inland ice following shelf collapse has resulted in significant contributions to sea-level rise from this region5,6, with future contributions expected to be heavily dependent on the state and fate of the remaining shelves in the peninsula and elsewhere in Antarctica7. Forecasts of future sea-level rise require ice sheet models incorporating realistic predictions of the timing of future ice shelf collapses. Risk estimation of Antarctic ice shelf collapse thus remains a important goal of the cryospheric sciences.

This project will utilise satellite and climate model 'big data' to construct a statistical model of ice shelf collapse risk. Despite many satellite observations and proxy reconstructions of previous collapse episodes, the complexity of governing processes occuring within ice shelves so far precludes the use of physically-based forecast models. However, the emerging and substantial observational 'big data' record of ice shelf properties, and surveys of more than half a century of ice shelf collapse episodes2, lend themselves well to combination within a statistical model framework. Bayesian nonparametrics provide a class of data-led statistical models that adapt their complexity to the data itself. This approach incorporates existing (perhaps imperfect or incomplete) observations to model a phenomenon, yet is flexible enough to allow future inclusion of new datasets. This quality is essential to modelling ice shelf collapse risk, where new observation and information are often made available. The project will make use of ice shelf physical properties, environmental conditions and collapse timing histories to estimate the risk of future collapse events. In particular, we will seek to assign probabilities to major collapse events at individual ice shelves over the course of the next 100 years. These probabilities can then be used in physically-based ice sheet models to improve forecasts of the Antarctic ice sheet contribution to sea-level rise.