Application of artificial neural networks (ANN) to high-latitude dinocyst assemblages for the reconstruction of past sea-surface conditions in Arctic and sub-Arctic seas

The artificial neural network (ANN) method was applied to dinoflagellate cyst (dinocyst) assemblages to estimate palaeoceanographical conditions. The ANN method was adapted to three distinct data bases covering the northern North Atlantic (N = 371), plus the Arctic seas (N = 540) and the Bering Sea (N = 646). The relative abundance of 23 dinocyst taxa was calibrated against hydrographic variables (sea-surface temperature, salinity and density in February and August, and seasonal extent of sea-ice cover) using ANNs. The estimation of hydrographical parameters based on an ANN yields high coefficients of correlation between observations and reconstructions for each variable selected. The validation tests performed on the different data bases suggest more accurate calibration at the scale of the North Atlantic and Arctic (N = 540) than on a multibasin scale, i.e. when including the subpolar North Pacific (N = 646). The ANN calibrations and the modern analogue technique (MAT) have been applied to two sequences from the northwest North Atlantic spanning the past 25 000 yr for the purpose of comparison. Both approaches yielded similar results, generally within the range of their respective uncertainties, demonstrating their suitability. The main discrepancies generally correspond to assemblages with poor modern analogues for which we have to admit a higher degree of uncertainties in the reconstruction, whatever the approach used. Copyright  2001 John Wiley & Sons, Ltd. Journal of Quaternary Science