Machine learning-based paleobathymetric reconstructions using archaeal lipid biomarkers

Accurate reconstruction of paleo-ocean depths is essential for understanding the interplay between tectonic evolution and global climate change, yet existing methods face substantial limitations. Here, we assess the potential of glycerol dialkyl glycerol tetraethers (GDGTs)-archaeal lipid biomarkers-as a quantitative proxy for paleobathymetry. Analyzing a global dataset of marine surface sediments, we find that GDGT distributions exhibit systematic shifts with water depth. Using random forest machine learning models trained on combined isoGDGT and OH-GDGT profiles, we achieve high predictive performance (R² = 0.85, RMSE = 646 m). Applying this approach to a 6-million-year sedimentary record from the northwest Australian Shelf reproduces key features captured by foraminifera-based reconstructions and reveals tectonically driven bathymetric evolution, including potential influences on the development of the Leeuwin Current. These results demonstrate the GDGT-based machine learning models offer a robust and effective framework for reconstructing paleo-ocean depths and probing the links between oceanography and tectonics.