Palaeoenvironmental reconstructions through compositional data analysis: the CoDaMAT method

We present a modern analog technique (CoDaMAT) based on compositional data analysis (Aitchison, 2003). In order to reconstruct palaeoenvironmental parameters, CoDaMAT adopts the Aitchison distance (Aitchison, 2000; 2003), obtained from relative abundances expressed in log-ratios (logarithm of ratios), as a “natural” measure of similarity between modern and fossil assemblages. The number of modern analogs from which obtain the palaeoestimates was determined through cross-validation techniques taking modern assemblages as the training data set. Atypicality index was considered to detect past no-analog conditions. The average of distances and expected maximum estimate errors were taken into account to evaluate the quality of palaeoestimates. The method has been tested on Mediterranean and Atlantic planktonic foraminiferal assemblages to reconstruct past sea surface temperatures (SST). Due to the peculiar oceanographical asset of the Mediterranean, no-analog conditions may represent an important problem in attempting the reconstruction of glacial SST. Keeping in mind these limitations, we obtained for the Tyrrhenian sea, at 15 ka BP (during the Greenland stadial GS2-a), summer and winter SST estimates respectively 12°C and 6°C lower than present. These are the lowest values reached during the last 34 ka. In the coldest interval of the Greenland stadial GS1 (Younger Dryas), reconstructed summer and winter SST were respectively 11°C and 5°C lower than present. In comparison with previous foraminiferal based reconstructions (de Abreu et al., 2003), the CoDaMAT palaeoestimates of Atlantic ocean surface temperatures for the last 200 ka show a stronger coherence with the alkenones and stable isotope record (Pailler and Bard, 2002).