SEA SURFACE-WATER TEMPERATURE AND ISOTOPIC RECONSTRUCTIONS FROM NANNOPLANKTON DATA USING ARTIFICIAL NEURAL NETWORKS

Marco Pozzi,

and

ABSTRACT

Artificial neural networks (ANN) are computer systems that differ from other pattern recognition methods in their ability to ‘learn’ one or more target variables from a set of input variables. These systems learn by self-adjusting a set of parameters to minimize the error between the desired output and network output.

To explore the potential of artificial neural networks for predictions of paleo-oceanographic parameters from relative abundances of calcareous nannoplankton species we analysed observations taken from the literature for (1) the prediction of sea surface-water temperature (SST) in samples from offshore southern California, and (2) the prediction of oxygen isotopic values in a Quaternary core from the eastern Mediterranean.

We employed a back propagation (BP) neural network to assess the ability of the network to predict SST and oxygen isotopic values. Each of the data sets was divided into five random training and test sets to assess the stability of the error rate estimates. For the California Bight samples we obtained an average Root-Mean-Square-Error of Prediction (RMSEP) in the test sets of 0.68, implying that an unknown SST can be predicted with a precision of ± 0.68°C. In the Mediterranean samples the average RMSEP in the test sets was 0.64; hence an unknown oxygen isotope value can be predicted with a precision of ± 0.64 ¹⁸O vs. PDB.