A linear mixed effects model for seasonal forecasts of Arctic sea ice retreat

ABSTRACT With sea ice cover declining in recent years, access to open Arctic waters has become a growing interest to numerous stakeholders. Access requires time for planning and preparation, which creates the need for accurate seasonal forecasts of summer sea ice characteristics. One important attribute is the timing of sea ice retreat, of which current statistical and dynamic sea ice models struggle to make accurate seasonal forecasts. We develop a linear mixed effects model to provide forecast of sea ice retreat over five major regions of the Arctic – Beaufort, Chukchi, East Siberian, Laptev, and Kara Seas. In this, the fixed effect – i.e. the mean influence of the atmosphere on sea ice retreat – is modeled using predictors that directly influence the dynamics or thermodynamics of sea ice, and random effects are grouped regionally to capture the local-scale effects on sea ice. The model exhibits very good skill in forecast of sea ice retreat at lead times of up to half a year over these regions.