Machine-learning-based global mapping of marine surface sediment physical properties

Marine surface sediments form a key interface between the water column, the seafloor, and the shallow sediment column, and their physical properties regulate sediment transport, pore water exchange, organic matter preservation, and the fate of particle-associated substances. However, globally continuous and variable-consistent datasets of marine surface sediment physical properties remain limited. Here, we present a machine-learning-based global dataset of marine surface sediment physical properties, including mean grain size, sand/silt/clay content, porosity, and specific surface area. By integrating observations from published literature and existing sediment databases with spatially continuous environmental covariates, we generated a 10′ × 10′ global gridded products within a unified modelling framework. To reduce the influence of uneven sample distribution and single-model structural bias, we introduced a spatial density stratified sampling strategy and a two-layer stacking ensemble-learning framework. Prediction uncertainty was quantified from repeated spatial density stratified modelling runs. The resulting products reproduce the global transition from relatively coarse-grained sediments on continental shelves and in marginal seas to fine-grained sediments in open-ocean and deep-sea environment. Regional assessments indicate that the predictions are broadly consistent with observational samples and known sedimentological patterns. Compared with existing sediment type or single variable products, this dataset improves spatial continuity, cross-variable consistency, and uncertainty representation. We further derived a permeability potential index to characterize relative pore water exchange potential. This dataset provides a unified spatial basis for sedimentary-environment analysis, organic carbon burial studies, contaminant fate assessment, and Earth system model parameterization. The data package contains six gridded GeoTIFF files. The GeoTIFF files provide global 10′ × 10′ gridded products of marine surface sediment physical properties. Mean grain size (Φ).tif represents the predicted mean grain size of marine surface sediments in phi units. Sand content (%).tif, Silt content (%).tif, and Clay content (%).tif provide the predicted sand, silt, and clay fractions, respectively, expressed as percentages. These three grain-size fraction layers satisfy the compositional closure constraint. Porosity (%).tif provides the predicted surface sediment porosity expressed as a percentage. Specific surface area_1500m (m^2_g^-1).tif provides the predicted specific surface area of marine surface sediments within the 0-1500 m water-depth range, expressed in square metres per gram.