Climate induced changes in benthic macrofauna – a non-linear model approach

Abstract:

The non-linear methods "optimally pruned extreme learning machine" (OPELM) and
"optimally pruned k-nearest neighbours" (OPKNN) are applied to relate various climate indices to time
series of biomass, abundance and species number of benthic macrofauna communities in the southern
North Sea for the period 1978-2005. The results of these methods show that the performance in
forecasting macrofauna communities is as poor as linear statistical downscaling if only one climate
index is used as a predictor. If a multivariate predictor is used, OPKNN shows a good forecast for
biomass and species number, but not for abundance. The improvement of the forecast is of major
relevance especially in the presence of biological and climate regime shifts which occurred in the
considered period.

Bio:

Dusan Sovilj got his B.Sc. at University of Novi Sad (Serbia) in 2006, and M.Sc. in Technology at HUT in 2009. He is doing Ph.D. in EIML group at ICS department on machine learning approaches to modeling Baltic Sea related measurements. His research interests are time series prediction, variable/feature selection for regression and environmental modeling.