Emerging technologies – deep learning
Some of the snow processes most critical to how glaciers, ice sheets and permafrost respond to climate change are currently too complex or computationally demanding to represent accurately in climate models. These gaps introduce uncertainty into projections, particularly in the polar regions.
SnowPI is exploring whether deep learning can help fill these gaps. Deep learning is a branch of artificial intelligence that allows computers to identify patterns within large datasets that may otherwise be too difficult to detect using traditional methods. Specifically, the project is training algorithms to represent three processes that are absent or poorly captured in current models: wind-blown snow sublimation and scouring; the effects of biology and dust on the albedo of snow and ice surfaces; and the formation of depth hoar within snowpacks.
To ensure that the deep learning algorithms are reliable, they are rigorously tested against established physics-based approaches. If successful, these new approaches could improve model accuracy while reducing computational costs and provide new insights into the behaviour of snow in polar environments.