GlacierGan: Visualizing the Alps during the Last Ice Age

In this work, we develop a deep-learning generative model to offer a new visualization for the Alps and its glaciation over the last 120’000 years as if a satellite had passed over and taken high resolution images from above. This visualization utilizes a recently coupled climate-glacier evolution model, which uses the latest paleo-climate and ice thickness field reconstructions (Jouvet et al., 2023). The ultimate goal of this project is to use it in the “IceAgeCam”, a joint SNSF (Swiss National Science Foundaton) project developed by ZHDK, UZH and UNIL that aims to better inform the public about the cause of climate change in a long-term climatic context.

To obtain such a visualization, we use an image-2-image translation model called Pix2PixHD (Wang et al., 2018). Similar to how one can use an image-2-image translation model to map images of winter to summer, or zebras to horses, we will map relevant fields of multi-band climatic images into artificial satellite images. Each multi-band image is composed of physical predictors such as ice thickness, ice velocity, precipitation, surface temperature, etc. In the end, the model produces semantically meaningful results that allow one to visualize the last glacial cycle. Moreover, although the motivation is rooted in the aforementioned ''IceAgeCam'' and seeks to visualize the last 120'000 years, it is a well-generalizable model, and as such, can be applied to visualize future simulations as well as terrain outside the Alps, given that the user has access to the same predictors. Finally, we aim to include this into IGM (the Instructed Glacial Model), a community-led glacier modeling software, such that it is user-friendly and easily accessible. Overall, we hope to advance efforts in the domain of remote sensing in relation to the cryosphere by providing a new way to visualize scientific results and foster community outreach.