Emulating the expansion of Antarctic perennial firn aquifers in the 21st century

Perennial firn aquifers (PFAs) are year-round bodies of liquid water within firns, which modulate meltwater runoff to crevasses, potentially impacting ice-shelf and ice-sheet stability. Recently identified in the Antarctic Peninsula, PFAs form in regions with both high surface melt and snow accumulation rates and are expected to expand due to the anticipated increase in surface melt and snowfall. Using a firn model to predict future Antarctic PFAs for multiple climatic forcings is relatively computationally expensive. To address this, we developed an XGBoost perennial firn aquifer emulator, a fast machine learning model. It was trained, using a scenario and spatial blocking evaluation approach, on PFA output of simulations from the firn densification model IMAU-FDM, which was forced by the combined regional climate model RACMO2.3p2 and the global climate model CESM2 for three emission scenarios (SSP1-2.6, SSP2-4.5 and SSP5-8.5). The trained emulator was applied on nine additional forcings (2015-2100) from the regional climate models MAR and HIRHAM in combination with five global climate models. We show that the emulator is robust, explaining at least 89 % of the variance in PFA presence and meltwater storage. Our results indicate that, for the SSP1-2.6 and SSP2-4.5 scenarios, PFAs remain mostly restricted to the Antarctic Peninsula. For SSP5-8.5, PFAs expand to Ellsworth Land in six out of the seven simulations and to Enderby Land in East Antarctica in five out of the seven simulations. Furthermore, the emulator predicts PFAs for similar surface melt and accumulation conditions when forced with MAR or RACMO data. For HIRHAM these conditions are slightly different, due to the different relationship between temperature, accumulation and melt in HIRHAM compared with RACMO. Overall, our findings show that PFAs are likely to expand in a warmer Antarctica, irrespective of the emission scenario, increasing the risk that an ice shelf collapses due to hydrofracturing.