Oceanic lead concentrations and isotopes mapped using explainable machine learning

Science Highlights

24 September 2025

Using three machine learning models, Olivelli and her colleagues (2025, see reference below) generated global climatologies of lead (Pb) concentrations, 206Pb/207Pb, and 208Pb/207Pb ratios using the non-linear regression algorithm XGBoost. The models were trained and tested on high-quality Pb data collected as part of the GEOTRACES programme, to which the authors added historical data, and use climatologies of oceanographic and atmospheric variables as features from which to predict Pb concentrations and isotope compositions.

They examined which factors most influenced the model’s predictions (using the SHAP framework). Interestingly, the most important predictors of Pb concentrations are slightly different from those of Pb isotope ratios. While all include seawater temperature, atmospheric dust, and salinity, 206Pb/207Pb ratios are also sensitive to dissolved oxygen concentration, whereas Pb concentrations and 208Pb/207Pb ratios are more sensitive to atmospheric black carbon and surface chlorophyll a.

The model predictions match observations well. They confirm that the Indian Ocean is the most impacted by lead pollution, while the latter is decreasing in the Atlantic and Pacific Oceans. Intermediate waters of Southern origin are spreading and distributing highly anthropogenic Pb isotope signatures to the remaining oceans, making the Southern Ocean a key area to prioritise in future sampling campaigns.

Figure: Global maps of Pb concentrations (a), 206Pb/207Pb (b), and 208Pb/207Pb (c) reconstructed using the three machine learning models developed by Olivelli et al. (2025). The four panels represent different depth levels (10, 1000, 2500, and 4000 m), with the white patches corresponding to the seafloor. The filled circles with black edges represent true observational values for comparison with the modelled values in the background.

Reference:

Olivelli, A., Arcucci, R., Rehkämper, M., & van de Flierdt, T. (2025). Mapping the global distribution of lead and its isotopes in seawater with explainable machine learning. Earth System Science Data, 17, 3679–3699. Access the paper:10.5194/essd-17-3679-2025

Latest highlights

An original approach to assess the particulate trace metal concentrations in seawater

Sohrin and co-workers propose defining particulate trace metal as the difference between total dissolvable and dissolved metals after a long storage of filtered and unfiltered acidified seawater.

Constraining aerosol deposition over the global ocean by the cosmogenic beryllium-7

He and co-workers propose a global estimate of aerosol deposition onto the ocean using the cosmogenic radionuclide beryllium-7

Dissolved nickel sources: transformation and sinks in the Arabian Sea

Malla and co-authors present an extensive study of the distribution of dissolved nickel in the Arabian Sea.

Linking cadmium cycling to phosphate dynamics in the Indian Ocean: Evidence from GEOTRACES transects

Mishra and Singh determined cadmium and phosphate concentrations along 34 complete vertical profiles in the Indian Ocean.

Rechercher