Quantifying multi-factor contributions to rare earth element enrichment in marine sediments using explainable machine learning

To elucidate the enrichment mechanisms and controlling factors of rare earth elements (REEs) in deep-sea sediments and offer theoretical guidance for deep-sea REEs resource exploration, this study compiles major, trace and rare earth element datasets from 7476 sediment samples across global marine regions. Nine key geological environmental proxies were selected, and four machine learning algorithms (CatBoost, LightGBM, Random Forest, and XGBoost) were integrated with SHapley Additive exPlanations (SHAP) to quantitatively characterize the driving mechanisms of REEs enrichment. All models showed robust predictability, with XGBoost performing optimally (MAE = 41.48, MAPE = 12.21, RMSE = 81.98, R 2 = 0.97). SHAP interpretation and multi-model validation revealed distinct factor contribution hierarchies: phosphate content is the dominant REEs enrichment driver (40.29%–62.49%), with enrichment facilitated when P 2 O 5 > 0.61 wt%. Sediment accumulation rate is the secondary factor (16.36%–18.8%), favoring REEs at <0.01 cm/yr. Spatially, terrigenous/eolian inputs dominate continental margins, while productivity and redox conditions control open-ocean basins. Ferromanganese nodules and hydrothermal activity have limited effects. This study highlights interpretable machine learning’s efficacy in multi-factor quantification, advancing understanding of sedimentary REEs geochemistry and supporting deep-sea REEs exploitation and marine biogeochemical research.