Automatic recognition of Cenozoic calcareous nannofossils: a reliable tool for paleoceanographic studies

Calcareous nannofossils are microscopic-scaled plates (mainly coccoliths) produced by marine phytoplankton. They are abundant in marine sediment, diversified through geologic time and sensitive to climatic/environmental changes. That's why they are often used in biostratigraphy and (paleo-)oceanographic/climatic reconstructions. We propose recent improvements on an automated system called SYRACO which is based essentially on artificial neural networks. Originally able to recognized around ten species from the Upper Pleistocene, this new system is now trained to recognize thousand of species from the Upper Eocene to actual species. It is now based on the use of several neural networks on a large database combined with a morphometric approach from statistical classifications. This new system is able to keep at least 90% of the nannofossils in a sample and filter out around 84% of the non-nannofossils. Moreover we added a special neural network coupled with macro for Florisphaera profunda, an important species in paleoceanography, which is difficult to indentify by neural network. As examples, we present applications nannofossil assemblage composition in core-tops of the Indian Ocean compared with human counts and in a core retrieved in the Gulf of Papua. These results show a very good correlation between human and automatic counts and a precise paleoceanogrphic dynamics.