| Summary: | Présentation de congrès Palaeoceanographic studies often rely on microfossil species abundance changes, with little consideration for traits like size that could also relate to environmental changes. We hypothesize that whole-assemblage and/or species-specific planktonic foraminiferal test size could be good predictors of environmental variables, and we test this using a tropical Indian Ocean core-top dataset. We use an automated imaging and sorting system (MiSo) and a convolutional neural network model (CNN) to identify species, analyze morphology, and quantify fragmentation using machine learning techniques. A total of 311380 images were acquired at an average of 3797 images per sample. Machine model accuracy is confirmed by comparison with human classifiers (98% accuracy achieved). Data for 32 environmental parameters are extracted from modern databases and, through Exploratory Factor Analysis and regression models, we investigate the potential of using planktonic foraminiferal size to reconstruct oceanographic parameters. The size frequency distribution of most planktonic foraminifera species is unimodal and larger species show polymodal distributions. Within our tropical dataset, we find that intraspecies size response to environmental parameters is species-specific with carbonate ion concentration, temperature, and salinity identified as primary drivers. At the assemblage level, our analyses suggest that internal biogenic processes (primary) and temperature (secondary) are key drivers of morphometric changes in planktonic foraminifera. Our assessment of the potential to utilize assemblage size in reconstructing sea surface temperature in the tropical Indian Ocean showed that the reconstructed SST of the test MD90-0963 downcore site, relatively followed the delta O18 signals from previous works for the same site.
|
|---|