Understanding Drivers of Greenland Surface Melting Through Machine Learning, Game Theory and Attribution Analysis: a Test Case at the Helheim Glacier

Despite recent progress, it is still challenging to fully capture the drivers of the recently increased surface melting over the Greenland ice sheet and untangle the links between mass loss and surface (e.g., albedo, surface temperature, etc.) and atmospheric processes (e.g., blocking, persistent anticyclonic conditions). This is, among other reasons, due to the lack of current models in integrating atmospheric dynamics and circulation into surface mass balance estimates and to the feedbacks among the different system components, which are difficult if not impossible to capture through current existing physical models. In this regard, machine learning tools represent a powerful approach to overcome the issues mentioned above. Specifically, attribution analysis can be performed through the combination of models and observations and using AI tools that allow the development of experimental models (e.g. random forest, CNN., etc.). In this presentation, we report results concerning the identification of drivers of the recent extreme melting events using a Random Forest Model approach that combines model outputs (e.g., energy balance terms, atmospheric data, remote sensing observations) with estimated values of surface meltwater concentration and/or liquid water content over the Helheim glacier region combined with the so-called Shapley theory, developed within the framework of game theory combination with game theory (e.g., Shapley coefficients). Our results, obtained for different melting years and months, show that this approach can be used to evaluate the surface melting drivers in a robust way, paving the way to better understand the relative role of the changes in both surface and atmosphere over the recent decades on Greenland's mass loss.