Machine learning-based paleobathymetric reconstructions using archaeal lipid biomarkers.

Area of Science:

• Geochemistry
• Paleoceanography
• Biogeoscience

Background:

• Reconstructing past ocean depths is crucial for understanding climate and tectonic interactions.
• Current paleobathymetry methods have significant limitations.

Purpose of the Study:

• To evaluate glycerol dialkyl glycerol tetraethers (GDGTs) as a quantitative proxy for paleobathymetry.
• To develop and validate machine learning models for paleodepth reconstruction.

Main Methods:

• Analysis of a global dataset of marine surface sediments.
• Application of random forest machine learning models trained on GDGT profiles (isoGDGT and OH-GDGT).
• Testing the model on a 6-million-year sedimentary record from the northwest Australian Shelf.

Main Results:

• GDGT distributions show systematic changes with water depth.
• Machine learning models achieved high predictive accuracy (R² = 0.85, RMSE = 646 m).
• Reconstructions align with foraminifera data and reveal tectonic influences on bathymetry and the Leeuwin Current.

Conclusions:

• GDGT-based machine learning models offer a reliable method for reconstructing paleo-ocean depths.
• This approach effectively probes links between oceanography, tectonics, and climate change.
• The study highlights the potential of GDGTs for detailed paleoceanographic and tectonic analyses.