Evaluation of some numerical techniques for determining paleotemperatures from planktonic foraminiferal assemblages

Abstract

This study evaluates the results of various numerical techniques intended to estimate environmental variables from the composition of associations of organisms in sediment samples. An analysis is made of various transfer functions (species regression technique, 13 variants; weighted average technique, 8 variants; principal component regression technique, 270 variants; discriminant analysis technique, one procedure) and of the modern analog technique (11 variants). The initial point is the calibration data set composed of the associations of planktonic foraminifera contained in 605 core-top samples from the North Atlantic and the Mediterranean, together with the corresponding sea surface winter temperatures. To compare the results of the various techniques, the calibration data set is randomly divided into two subsets of 450 and 155 samples, respectively. The former is used as the provisional calibration data set; this is used to estimate the winter temperatures of the subset of 155 samples (the analog set) and of another set, formed of 27 samples from the South Atlantic and the Indian Ocean (no-analog set). After evaluating the errors of the estimate of the different techniques, the best ones are selected.