Mapping the global distribution of lead and its isotopes in seawater with explainable machine learning

Since the late 1800s, and especially in the last century, the natural biogeochemical cycle of lead (Pb) in the ocean has been severely perturbed by anthropogenic emissions generated by the use of leaded gasoline, waste incineration, coal combustion and non-ferrous metal smelting. Lead and its isotopes are powerful tools to study the pathways of Pb pollution from land to sea and, simultaneously, investigate biogeochemical processes in the ocean. For these reasons, the study of Pb concentrations and isotope compositions of seawater is a core part of the international marine geochemistry programme GEOTRACES. However, the scarcity and sparsity of in situ measurements of Pb concentrations and isotope compositions do not allow for a comprehensive understanding of Pb pollution pathways and marine biogeochemical cycling on a global scale.

We present here three machine learning models developed to map seawater Pb concentrations and isotope compositions leveraging the global GEOTRACES dataset together with historical data. The models are based on the non-linear regression algorithm XGBoost and use climatologies of oceanographic and atmospheric variables as features from which to predict Pb concentrations, ²⁰⁶Pb/²⁰⁷Pb, and ²⁰⁸Pb/²⁰⁷Pb. Using Shapley Additive Values (SHAP), we found that seawater temperature, atmospheric dust and black carbon, and salinity are the most important features for mapping Pb concentrations. Dissolved oxygen concentration, salinity, temperature, and atmospheric dust are the most important features for mapping ²⁰⁶Pb/²⁰⁷Pb, while atmospheric black carbon and dust, seawater temperature, and surface chlorophyll-a for ²⁰⁸Pb/²⁰⁷Pb. The output of our models shows that (i) the highest levels of pollution are found in the surface Indian Ocean, (ii) pollution from previous decades is sinking in the North Atlantic and Pacific Ocean, and (iii) waters characterised by a highly anthropogenic Pb isotope fingerprint are spreading from the Southern Ocean throughout the Southern Hemisphere at intermediate depths. The analysis of the uncertainty associated with the mapped distribution of Pb concentrations, ²⁰⁶Pb/²⁰⁷Pb, and ²⁰⁸Pb/²⁰⁷Pb suggests that the Southern Ocean is the key area to prioritise in future sampling campaigns.