Abstract:
|
The Arctic Sea ice extent and area has been recorded since November 1978 until today by the National Snow & Ice Data Center and is available to the public as nearly daily data. There are studies looking at the data set as either time series data or as uncorrelated measurements of each day of the year. In this research, we extend this view by considering the ice extent area data as a functional data with functions estimated for each year. Of potential interest in this type of application is to be able to classify the curves based on the results from just a portion of the curve, say the first few months of the year. Two modifications of monothetic clustering, a type of clustering that creates clusters that share common characteristics, are proposed that only use information from early in each year to cluster the curves are explored. To illustrate the methods, the sea ice data from November 1978 to December 2014 are used as training data with the results from 2015 and 2016 used to test the classification performance of the methods.
|