Projecting 21st century GrIS surface melt using artificial neural networks

Accelerated surface melt of the Greenland ice sheet (GrIS) is currently a large contributor to sea level rise, and the primary process of GrIS mass loss. Projections of future GrIS melt are limited by the lack of explicit melt calculations within most global climate models and the high computational cost of dynamical downscaling with regional models. To translate global climate evolution to GrIS surface melt, we train artificial neural networks (ANNs) with the output of the explicit melt calculation of the Community Earth System Model 2.1 (CESM2). ANNs are well suited for this task, as they are capable of learning complex, non-linear relationships, and they are fast to run.

Our results show that the ANNs accurately project GrIS surface melt when evaluated against regional climate simulations. Further, the ANNs recognize patterns already established in litterature as important for surface melt, and use bases the projections on these patterns. Using the global climate simulations from the CMIP6 archive, the ANNs project a GrIS surface melt increase ranging from 414 Gt yr^-1 to 1,378 Gt yr^-1 by the end of the 21^st century. The main source of projection uncertainty throughout the 21^st century is due to the spread in the models’ climate sensitivity.