Exploring the future expansion of perennial firn aquifers in Antarctica using a random forest emulator

Perennial firn aquifers (PFAs) are year-round bodies of liquid water within firn. In Antarctica, they can cause hydrofracturing of ice shelves, leading to accelerated ice-sheet mass loss. PFAs were only recently discovered in the Antarctic Peninsula, at locations with high melt and snow accumulation rates. Likely, PFAs will expand in the future as both snowfall and melt increase. So far, this has not been considered when assessing the future vulnerability of Antarctic ice shelves. One could use a firn model to predict future Antarctic PFA evolution, but to do so for a wide range of climatic forcings is computationally expensive. Therefore, we set up a random forest emulator. The emulator represents both firn and surface climate processes, so that only 2 metre temperature and precipitation are required as input. To train the emulator, we use simulations of three scenarios (SSP1-2.6, SSP2-4.5 and SSP5-8.5) from firn densification model IMAU-FDM forced by the regional climate model RACMO2, which was driven by CESM2 in turn. The emulator successfully explains 98% of the PFAs variation, therefore we use its versatility and speed to predict Antarctic PFA evolution for 15 additional RCM/GCM model forcings. We find a range of solutions, highlighting the usefulness of the emulator. Until 2100, PFA occurrence remains restricted to ice shelves in the Antarctic Peninsula for SSP1-2.6 and SSP2-4.5. For SSP5-8.5, PFAs expand to the Bellingshausen Sea region in West-Antarctica, and to Enderby Land in East-Antarctica. The meteorological conditions in Enderby Land exhibit similarities to those observed in the Antarctic Peninsula. In contrast, Getz ice shelf, which experiences high snow accumulation rates, remains insensitive to PFA formation, because it is rather cold. Our results highlight the sensitivity of relatively warm, high-accumulation ice shelves to future PFA formation and subsequent hydrofracturing.