[ESAC Student Seminar] Björn Lütjens

“MeltwaterBench: Deep learning for downscaling surface meltwater”

The Greenland ice sheet (GrIS) is likely melting at the fastest rate since the past 12.000 years and is one of the main contributors to contemporary sea level rise. The future contribution to sea level rise by 2100 varies largely in part because melting processes and their relationship with climate change still contain many uncertainties. Maps of surface meltwater are essential to understanding melting processes, for example, by indicating extreme melt events that can be linked to climatological phenomena. But, existing surface meltwater maps have a resolution that is either too coarse in space or time to study some melting processes, such as the effect of rapid melt events in coastal regions. This talk will demonstrate a new deep learning-based downscaling method that creates a daily high-resolution (100m) surface meltwater product to fill this data gap. We focus our study on the Helheim Glacier in Eastern Greenland and process data streams from 2017-2023 from synthetic aperture radar, passive microwave (PMW), elevation, and a regional climate model (MAR) at multiple spatial and temporal resolutions. We show that MAR- and PMW-derived estimates of surface meltwater overestimate early season melting at the Helheim glacier and how the deep learning-based UNet model can improve on those biases. We are formatting the dataset as a new benchmark, MeltwaterBench, that will enable transparent comparison of novel machine learning methods.

ESAC Student Seminar Series —

A forum for students and postdocs to share recent research, hone presentation skills, and build community among peers, sponsored by the EAPS Student Advisory Committee. Open to current EAPS graduate and undergraduate students and postdocs. Typically hosted on Thursdays during the semester, including pizza lunch.

Contact: esac.officers@gmail.com