[M-IS10] Global climate change driven by the Southern Ocean and the Antarctic Ice Sheet

[MIS10-16] High-resolution record of the radiolarian Cycladophora davisiana curve in the Southern Ocean acquired using a high-throughput virtual slide scanner and deep learning

*Takuya Itaki¹, Ayumu Miyakawa¹, Minoru Ikehara³, Hiroki Matsui⁴, Yusuke Suganuma² (1.Geological Survey of Japan, AIST, 2.National Institute of Polar Research, 3.Kochi University, 4.Akita University)

Keywords:Automated microscope, deep learning, microfossils

Because sediment cores from the Southern Ocean are often poorly preserved calcareous microfossils, it is difficult to construct age models using foraminiferal oxygen isotope (δ¹⁸O) records. Recently, tuning of the dust proxy signal of a sediment core to the dust record of an ice core is a promising method to construct an age–depth model. However, dust correlations have problems, such as lack of signals and temporal discrepancies in some regions. On the other hand, the relative abundance of radiolarian Cycladophra davisiana to the total assemblage in the Southern Ocean has been used for stratigraphic correlation since CLIMAP, as it shows a variation that increases during glacial periods and decreases during interglacial periods synchronized with δ¹⁸O records. Itaki et al. (2020, Sci. Rep. 10, 21136) used automated microscopy and deep learning to obtain the C. davisiana curve, successfully automating a task that previously required a huge amount of labor and time by experts. In this study, deep learning was collaborated with a virtual slide scanner capable of 360 microscopic slides continuous processing for further high-throughput analysis, and the high-resolution C. davisiana curve was acquired from core DCR-1PC collected in the Indian Ocean sector of the Southern Ocean. The record obtained here is expected to be compared with δ¹⁸O and dust records from the same core (Matsui et al., 2022, QSR 286, 107508) to provide a high time-resolved age model that will enable correlates widely in the Southern Ocean. In this study, deep learning was collaborated with a virtual slide scanner capable of 360 microscopic slides continuous processing for further high-throughput analysis, and the high-resolution C. davisiana curve was acquired from core DCR-1PC collected in the Indian Ocean sector of the Southern Ocean.