Coherency-Based Clustering of Global Sea Surface Temperature Time Series for the Detection of Large-Scale Teleconnections

• Activity Number: 50
• Type: Invited
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: The International Environmetrics Society
• Author(s): Francesco Finazzi and Marian Scott*+
• Companies: University of Bergamo and University of Glasgow
• Keywords: sea surface temperature ; temporal coherence ; clustering ; state-space modelling ; expectation maximization ; large datasets

Abstract

Essential Climate Variables (ECVs) are climate variables which are technically and economically feasible to be systematically observed. Ocean sea surface temperature (SST) is one such ECV as it directly affects the behavior of the Earth's atmosphere and is easily measured by satellite or buoy. The study of SST requires the analysis of millions of time series. Clustering of SST time series with respect to their temporal coherence simplifies the task by identifying groups of time series that behave similarly. In this work we present a novel coherency-based clustering technique that can handle millions of time series with missing data, and hence is suitable for SST analysis. The technique is based on a classic state space model, the parameters of which are estimated using a modified version of the EM algorithm in such a way that the cluster membership can be read from the model observation matrix. When applied to SST time series, the technique identifies clusters with a different spatial structure. The analysis of the temporal and the spatial patterns of the clusters will contribute to better understanding of the role of SST in global climate and the variations of the SST itself.