Sea surface-water temperature and isotopic reconstructions from
nannoplankton data using artificial neural networks

Marco Pozzi, Björn A. Malmgren, and Simonetta Monechi

Artificial neural networks are computer programs that are able to 'learn' to recognize patterns in data. They do this by being able to be 'trained' to produce a target answer value from a set of input values. This article explores the potential of artificial neural networks to predict (1) sea surface temperatures and (2) the abundances of oxygen isotope values from datasets composed of abundance values of a series of modern marine plankton species from the Mediterranean and the California Bight. Sea surface temperatures and the concentrations of various isotopes in marine waters are important indicators of past climates. This test is possible because the actual sea surface temperatures and isotopic values of the sea water in which these plankton lived is known. Results of this test suggest that the artificial neural network approach can predict temperatures to an accuracy of ± 0.68°C and isotopic compositions to an accuracy of ± 0.64 parts per thousand. Based on these results it is concluded that artificial neural networks have great potential for being able to determine sea surface temperatures and isotopic compositions of ancient seawaters based on the abundances of animals and plants that lived in those waters, but are now part of the fossil record.