| Summary: | Historical data of sea ice concentration in Canadian waters are analysed using projections methods (Principal Component Analysis, Singular Value Decomposition, Canonical Correlation Analysis and Projection on Latent Structures) to identify the main patterns of evolution in the sea ice cover. Three different areas of interest are studied: (1) the Gulf of St Lawrence, (2) the Beaufort Sea and (3) the Labrador Sea down to the east coast of Newfoundland. Forcing parameters that drive the evolution of the sea ice cover such as surface air temperature and wind field are also analysed in order to explain some of the variability observed in the sea ice field. Only qualitative correlations have been identified, essentially because of the singular nature of the sea ice concentration itself and the accuracy of available data. However, several statistical models based on identified patterns have been developed showing forecasting skills far better than those of the persistence assumption, which currently remains one of the best 'model' available. Forecasts are tested over periods of time ranging from a few days to several weeks. Some of these models constitute innovative approaches in the context of statistical sea ice forecasting. Some others models have been developed using a probabilistic approach. These models provide forecasts in terms of sea ice severity (low-medium-high), which is often accurate enough for navigation purposes for the three areas of interest. Forecasting skills of these models are also better than the persistence assumption. Finally, an existing dynamic sea-ice model has been adapted and used to predict sea ice conditions in the Gulf of St Lawrence during the Winter season 1992-1993. Simulations provided by this model are compared to the forecasts of different statistical models over the same period of time.
|
|---|