A statistical model for seasonal forecasting of spring break-up in Hudson Bay

A linear regression model is developed to predict spring ice conditions in Hudson Bay 3 months in advance. The model has 8 atmospheric, oceanic and sea ice predictors that together explain 90% of the variability in the predicted time series with a MAE of 0.26. The categorical forecast accuracy ranges from 73% to 77% A series of diagnostics including Monte Carlo simulations are used to ensure that the model is statistically significant and is not adversely affected by serial correlation or multicolinearity. The results suggest that for long-lead forecasts, sea ice severity may be forecasted using a simple statistical model that does not require any information regarding the future climate