Multi-scale Sea Ice Code

The Arctic sea-ice cover is not a continuous stretch of pristine ice but crisscrossed by ice ridges and open cracks, known as leads. These are the results of the constant movement of the ice, driven by winds and ocean currents. While ridges can be 100 m wide and leads several kilometres wide, both can be much longer than that, sometimes spanning the entire Arctic Ocean. The resulting pattern of narrow, nearly straight lines is known as Linear Kinematic Features (LKF). They are considered a signature feature of the Arctic ice cover. Despite this, today's sea-ice models struggle to reproduce them well enough. In MuSIC, we propose a new model to simulate the ice movement, allowing us to represent LKFs better in climate models. In the first year of MuSIC, we published a paper describing the new model called the Brittle Bingham-Maxwell rheology (BBM). This paper is the first step in gaining recognition of the new model. We showed that BBM gives an even better representation of the small-scale movement and fracturing of the ice than its predecessor. Crucially, BBM also works for simulations covering multiple years, for which its predecessor gave unrealistic results. Following this successful first step, we have used BBM in MuSIC and other projects. In the other projects, we have already used BBM to gain insights into the role of leads in sea-ice formation in winter. We have also used it to understand better how we are losing the oldest Arctic sea ice to climate change. This work has provided important insights into the nature and behaviour of Arctic sea ice while also demonstrating decisively the utility of BBM as a scientific tool. In MuSIC, we have started comparing BBM results and satellite observations of the ice to create a more robust framework for comparing models and observations. This comparison is challenging, based on comparing intricate patterns of LKFs in models and observations. We have, therefore, chosen to establish our new framework on methods and algorithms from computer vision studies and machine learning with good results. The framework will help us further improve our model results and cement the case for broader use of BBM within the sea-ice modelling community.

The Arctic sea-ice cover is not a continuous stretch of pristine ice but rather crisscrossed by ridges and leads, which are the result of the constant drift of the ice. The signature of this can be seen in the large-scale drift in a phenomenon referred to as Linear Kinematic Features (LKFs). Commonly used large-scale sea-ice models struggle to capture the relevant properties of LKFs, especially when run at a resolution commonly used for climate simulations. The team behind this proposal is now ready to propose a new sea-ice dynamics model which, unlike previous attempts, can both capture LKF properties and is suitable for climate applications. The proposed model will also improve the long-term pan-Arctic drift of the ice, capturing the multi-scale nature of sea-ice dynamics. In MuSIC we will address the simulation of drift and deformation of the ice at climate scales, both spatial and temporal. This will consist of a study of the large scale drift, focusing on seasonal variations and the long term trend attributed to climate change. We will then undertake another study showing the importance of correctly simulating the flux of heat through leads at different model resolutions and how the new model makes this possible. Proper tuning and understanding of the role of different model parameters are necessary before using the model in a large-scale setting, and this need will be addressed by MuSIC. Here we will take advantage of the wealth of satellite observations available of the Arctic, as well as novel machine learning methods to constrain model parameters and improve our understanding of how to tune the model in different settings. Computational efficiency is ever important in climate modelling. In MuSIC we will spend considerable time ensuring that the new model is as efficient as possible on today’s computational architectures. In addition, we will explore how best to write an efficient version of the model for a GPU based computational architecture.