École Nationale Supérieure de
Techniques Avancées

Résumé

In a context of a globally warming climate, the evolution of the Earth's glaciers is a matter of great importance, because of their prominent contribution to sea level rise and also the crucial role they play on freshwater resources availability. Several types of geoscientific glacier models exist to attempt at producing accurate quantitative predictions, using different physical approximations and approaches, as well as various programming languages. The Open Global Glacier Model (OGGM) is one of them, programmed in Python, well suited to perform global simulations, and with a wide scientific community. Alternatively, ODINN (OGGM + DIfferential equation Neural Networks) is a more recent model, combining certain OGGM functionalities with physics-informed neural networks in the state-of-the-art programming language Julia. We demonstrate in this study the potential of coupling the two models in order to gain both performance and modelling possibilities, by making use of Python's and Julia's respective strengths. For that, an new Julia ice flow model was implemented into OGGM, building a bridge between both models in a new modelling paradigm. We evaluate the performance and accuracy of this new model, which is overall more time efficient than other ice flow models from OGGM according to extensive benchmarks.