A Machine Learning Model of Arctic Sea Ice Motions

Abstract:

Sea ice motion plays an important role in the polar climate system by transporting pollutants, heat, water and salt as well as influencing the sea ice concentration and thickness. Numerous dynamical models have been constructed to solve the coupled momentum equation and constitutive law for sea ice in physics-based models. In this study, we propose a new data-driven deep-learning approach that utilizes a convolutional neural network (CNN) to model how Arctic sea ice moves in response to the present surface winds given its initial ice velocity and concentration fields. Such an approach can take the place of a dynamical equation-based scheme, and by-pass debates about the best constitutive law to describe the sea ice rheology. Results show that CNN predicts the sea ice motion with a correlation of 0.82 on average with respect to reality, which surpasses a set of pixel-based predictions, such as persistence (PS), linear regression (LR), random forest (RF), multiple layer perceptrons (MLP) and a widely-used dynamical model. The superior predictive skill of CNN suggests the important role played by the connective patterns of the predictors to describe the sea ice rheology.