Identifying the signature of sea-surface properties in dinocyst assemblages: Implications for quantitative palaeoceanographical reconstructions by transfer functions and analogue techniques

Abstract

Abstract Dinoflagellate cyst (=dinocyst) assemblages are widely used for the reconstruction of multiple oceanographic variables through the application of transfer functions. There is evidence that the number and kind of variables driving compositional changes in dinocyst assemblages vary regionally and that the selection of driving factors and the evaluation of transfer function performances are method-specific and complicated by spatial autocorrelation. Here, we used two new modern datasets from the Northern Hemisphere Atlantic-Arctic and the Northern Hemisphere Pacific oceans to re-evaluate the impact of sea-surface properties in dinocyst assemblages. We determined the dimensionality of the dinocyst ecological response and identified the main drivers for both regions. We calibrated and evaluated transfer function methods for the prediction of these variables and estimated their performances considering spatial autocorrelation. In both datasets, multiple environmental variables mutually and independently affect assemblage compositions, but the number and kind of these variables differ between datasets. We detected spatial autocorrelation, which was often due to environmental similarity, but some variables appeared to reflect geographical closeness, implying that spatial independence between sample sites depends upon the variables. We identified different primary drivers in both areas, highlighting the merit of regional calibrations and the necessity to carry out variable selection for each region separately. The multiple gradients identified imply that potentially multiple parameters could be reconstructed from the same fossil dinocyst assemblages. However, as the multiple gradients reflect geographical structuring, we propose that regional calibrations, even at the expense of generalisation, could improve the reliability and interpretation of transfer function reconstructions.