A DIATOM-BASED ARTIFICIAL NEURAL NETWORK FOR NORTH ATLANTIC MARINE QUATERNARY PALEOTEMPERATURE ESTIMATES

A DIATOM-BASED ARTIFICIAL NEURAL NETWORK FOR NORTH ATLANTIC MARINE QUATERNARY PALEOTEMPERATURE ESTIMATES B.A. Malmgren (1), E. Witon (1), H. Schrader (2), H. Jiang (3) (1) Dept. Earth Sciences, Göteborg University, Sweden (2) Dept. Geology, University of Bergen, Norway, (3) Dept. Earth Sciences, University of Aarhus, Denmark ([email protected]/Fax. +46-31-773 4903)

Artificial neural networks (ANNs) have been recently applied for estimating past summer and winter sea-surface water temperatures (SST) from planktonic foraminifer relative-abundance data from the Atlantic and Indian oceans. ANNs are computer systems that have the ability to "learn" the relationship between a set of input vectors (faunal data) and one or several output vectors (SST data). This "learning" is accomplished through an algorithm that gradually adjusts the structure of the network in order to minimize the error between the target vector and network output. An attempt has now been made to apply the ANN technique for paleotemperature predictions from diatom relative-abundance data of 61 species from the Atlantic Ocean, applicable to the 0-10 m water-depth interval. Two-thirds of the available samples (195 samples) were used for training of the neural networks, and the remaining one-third of the samples was employed for testing their performance (prediction error in terms of root-mean squared errors of prediction, RMSEPs). Six independent runs were made for each of the summer and winter SSTs to assess the stability of the RMSEPs using different training- and test-set partitions. The average RMSEP is 1.29 degrees C for summer SST and 1.60 degrees C for winter SST (the average correlation between actual and predicted SSTs is 0.988 for summer SST and 0.986 for winter SST), suggesting that this technique holds much promise for estimates of past SSTs. Of particular significance is the ability of the diatom-based ANNs to well predict SSTs below 5 and above 25 degrees C.

1