Estimation of Sea Ice Thickness through Maximum Covariance Analysis

Abstract:

Arctic sea ice is an important constituent of the global climate system and has undergone significant changes in recent years. Both a decrease in sea ice extent, especially over the summer months, as well as an overall thinning of the ice pack have been observed. Superimposed on the decreasing trend is considerable interannual variability which has proven difficult to predict, particularly when the variability deviates far from the trend. One of the challenges for model based seasonal predictions of sea ice is an accurate representation of sea ice initial conditions, particularly the distribution of sea ice thickness (SIT), for which the observational record is sparse. As a possible means of filling this gap and establishing an improved method for initializing SIT in the Canadian Seasonal to Interannual Prediction System (CanSIPS), this research aims to investigate how accurately SIT can be estimated in real time using better observed and physically relevant predictors. In this study sea ice concentration (SIC), sea level pressure (SLP), and combined SIC/SLP are used to construct a predictor-predictand model using maximum covariance analysis (MCA). The model parameters are determined over a period of 15 years prior to the initialization year, and an estimation of the SIT field is made by applying this statistical model to the anomalies of the predictor for the year of interest. Sea ice data from the Polar Science Center PIOMAS product are used to construct and test the model, and estimations of SIT over the period 1995-2012 are examined. Monte Carlo experiments are performed to test the statistical significance of the SIT prediction as a function of the predictors and how many statistical modes are retained.