Compositional Regression-Based Methods for SST Reconstruction

Abstract

The information in modern or fossil foraminifera assemblages is the relative abundance or percentages of species, i.e., they can be considered as compositional data. In this study we deal with CoDa and regression-based methods as tools to estimate past climatic conditions. We tested standard and robust Partial Least Squares and Principal Component Regression, applied to the log-ratio coordinates of percentage data of Atlantic Ocean and Mediterranean Sea planktonic foraminiferal assemblages. Due to the presence of groups, it was preferred to model separately high latitude and mid to low latitude assemblages. This approach implies the application of cluster analysis, MANOVA and discriminant analysis to the logratio transformed fossil assemblage's compositions. The methods were then applied on marine core assemblages to reconstruct past sea surface temperatures. The obtained results were compared with those formerly obtained by means of compositional modern analogue technique and with the information arising from other paleoclimatic proxies