On 20 February 2026, Science Advances published online the research article “Machine learning-based paleobathymetric reconstructions using archaeal lipid biomarkers”. The study was led by the Seafloor Microbiology and Environmental Evolution team at the School of Oceanography, Shanghai Jiao Tong University, and the National Key Laboratory for Submarine Science and Maritime Boundary Delimitation.

The first author is PhD candidate Jiaming Zhou. Associate Research Fellow Liang Dong is the corresponding author. Associate Professor Dujuan Kang and Master’s student Shijie Chen are co-authors.

Reconstructing past ocean water depth (paleobathymetry) is fundamental for understanding tectonic evolution, sea-level change, and climate dynamics. However, widely applicable quantitative proxies remain limited. Glycerol dialkyl glycerol tetraethers (GDGTs)—archaeal lipid biomarkers—have long been used to estimate past sea surface temperatures, and recent work suggests that their distributions also respond to water depth, offering an opportunity to develop a new depth proxy.

In this study, the team analyzed a global dataset of isoprenoid GDGTs (isoGDGTs) and hydroxylated GDGTs (OH-GDGTs) from marine surface sediments and applied machine-learning approaches, including random forest regression, to build quantitative predictive models for water depth. The combined isoGDGT + OH-GDGT model achieved the strongest performance (R² = 0.85; RMSE = 646 m) (Fig.1), providing a scalable framework for paleobathymetric reconstruction at the global scale.

Fig 1. FStructure of the studied GDGTs and dataset characteristics. (A) Chemical structures of isoGDGTs and OH-­ GDGTs. (B) PCA plot of the global oean surface sediment GDGT dataset, excluding GDGT-0, with data from (Varma et al.,2024). (C and D) R2 and root mean square error (RMSE) of the random forest model for bathymetry prediction.

To improve interpretability, the authors used explainable machine learning (e.g., SHAP) to identify the GDGT features that contribute most strongly to depth predictions. The analysis indicates that, as water depth increases, overall GDGT cyclization tends to decrease. In addition, the relative abundances of crenarchaeol and its isomer may reflect archaeal membrane adaptations to higher hydrostatic pressure in deep-water environments, offering a possible mechanistic link between microbial physiology and water-column depth at the molecular level.

Fig 2. Paleobathymetric reconstruction at Station U1461 over the past 6 Ma. (A) Reconstruction using the isoGDGT model, (B) reconstruction using the OH­ GDGTmodel, and (C) reconstruction using the combined isoGDGT + OH-GDGT model. Light blue shading denotes the RMSE for each model.

To demonstrate geological utility, the team applied the new model to sediments from International Ocean Discovery Program (IODP) Site U1461 on the northwest Australian shelf, reconstructing water-depth changes over approximately the past 6 million years (Fig. 2). The reconstructions capture key features consistent with foraminifera-based bathymetric indicators and further suggest that tectonically driven bathymetric evolution may have influenced Antarctic Intermediate Water intrusion and, in turn, the intensity of regional warm currents, including the Leeuwin Current.

By integrating lipid geochemistry with machine learning, this work expands the applications of archaeal lipid biomarkers beyond temperature proxies and provides a new quantitative tool for exploring seafloor dynamics and tectonics–ocean–climate interactions.

Funding

This research was supported by the National Key R&D Program of China (2022YFC2805403), the National Natural Science Foundation of China (42472370, W2511038), the Shanghai Jiao Tong University 2030 Plan (WH510244001), an independent research project of the National Key Laboratory for Submarine Science and Maritime Boundary Delimitation (SGLabZZKT2025-09), and the Ocean Negative Carbon Emissions (ONCE) project.

Article information

Title: Machine learning-based paleobathymetric reconstructions using archaeal lipid biomarkers
Journal: Science Advances
Published online: 20 February 2026
DOI: 10.1126/sciadv.adz3284

Source: Geological Microbiology Team, School of Oceanography, Shanghai Jiao Tong University