A Bayesian Logistic Regression for Probabilistic Forecasts of the Minimum September Arctic Sea Ice Cover
Sean Horvath, Julienne Strœve, Balaji Rajagopalan, William Kleiber
Abstract
Abstract This study introduces a Bayesian logistic regression framework that is capable of providing skillful probabilistic forecasts of Arctic sea ice cover, along with quantifying the attendant uncertainties. The presence or absence of ice (absence defined as ice concentration below 15%) is modeled using a categorical regression model, with atmospheric, oceanic, and sea ice covariates at 1‐ to 7‐month lead times. The model parameters are estimated in a Bayesian framework, thus enabling the posterior predictive probabilities of the minimum sea ice cover and parametric uncertainty quantification. The model is fitted and validated to September minimum sea ice cover data from 1980 through 2018. Results show overall skillful forecasts of the minimum sea ice cover at all lead times, with higher skills at shorter lead times, along with a direct measure of forecast uncertainty to aide in assessing the reliability.