Sea ice in the Canadian Arctic: inter-annual variability and predictability

• Main Author: Tivy, Adrienne
• Other Authors: Yackel, John
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: University of Calgary 2009
• Subjects: Arctic; Canada; Hudson; Hudson Bay; North Atlantic; Sea ice
• Online Access: http://hdl.handle.net/1880/104031; https://doi.org/10.11575/PRISM/3030

Bibliography: p. 189-202 Some pages are in colour. This dissertation investigates the utility of statistical methods towards the development of predictive sea ice models in support of seasonal forecasting efforts at the North American Ice Service (NAIS), the government agency responsible for relaying ice information to the public. Seasonal sea ice forecasting models are required to a) predict the dates of actual sea ice events and b) predict the general pattern of break-up. Statistical methods are employed because deterministic models do not yet adequately represent seasonal or inter-annual variability in sea ice to the level of accuracy required for forecasting. The Canadian Ice Service Digital Archive (CISDA) dataset of weekly ice charts is used as the foundation dataset. The issue of changing sensors over time in CISDA is addressed allowing the full dataset, which begins almost 20 years before the satellite era, to be used in model development. Model development is focused on the Hudson Bay region of Canada where the greatest reduction in summer sea ice cover is observed. Two forecasting models are developed. The first is an exploratory method based on multiple linear regression. Tested on forecasting the opening date of the shipping route to Churchill (OWRC), the model explains 76% of the variability in the OWRC time-series with a forecast success rate of 77%. The methodology has been automated and the regression based seasonal forecasting model (CIS-RSF) will be used operationally at NAIS for the first time in 2009. An initial evaluation of CIS-RSF on 21 time-series of sea ice events in the Arctic produced 7 forecast models; the skill of each model is greater than persistence and the current NAIS forecast technique. The second model uses Canonical Correlation Analysis (CCA). To date, it is the only sea ice forecasting model capable of predicting the spatial distribution of sea ice. The model is used to predict the spatial pattern of July ice using North Atlantic SST anomalies in the preceding fall (6 to ...