CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Multi-resolution clustering of time dependentfunctional data with applications to climatereconstruction Authors:  Johan Strandberg - Umea University (Sweden) [presenting]
Konrad Abramowicz - Umea University (Sweden)
Lina Schelin - Umea University (Sweden)
Sara Sjostedt-de Luna - Umea University (Sweden)
Abstract: A multi-resolution approach used to cluster dependent functional data is presented. Given a lattice of (time) points, a function is observed at each grid point. We assume that there are latent (unobservable) groups that vary slowly over time. We consider the case when at different time scales (resolutions) different groupings arise, with groups being characterised by distinct frequencies of the observed functions. We propose and discuss a non-parametric double clustering based method, which identifies latent groups at different scales. We present an application of the introduced methodology to varved lake sediment data, aiming at reconstructing winter climatic regimes in northern Sweden at different resolutions during the last sixthousand years.