Baltic Sea 1 arc-minute TOC/MAR machine-learning features, labels

Processed input data for regional machine-learning maps of total organic carbon (TOC) and mass accumulation rate (MAR) in Baltic Sea surface sediments at 1 arc-minute resolution (Parameswaran et al., Baltic TOC/MAR study). The archive contains:(1) Compiled TOC and MAR label tables and source measurements;(2) Preprocessed training feature matrices and normalisation statistics for MAR and TOC pipelines (Baltic_MAR_men, Baltic_TOC_men);(3) Baltic-wide prediction grid stack (BalticGridFeatures: grid_features.npy, lon/lat, sea/land masks);(4) Mud-substrate masks for mud-restricted MAR experiments;(5) 74 NetCDF predictor layers (78 feature entries) regridded to the Baltic 1′ grid, including HELCOM BALANCE substrate, bathymetry, CMEMS physics/BGC/waves, and derived seabed-energy and 5 km neighbourhood fields. Companion code repository: https://github.com/paramnav/baltic-toc-mar Not included: raw multi-terabyte CMEMS subset downloads, trained model weights, and prediction maps (separate Zenodo record). After download, run install_zenodo_data.sh to link this bundle into a baltic-toc-mar clone. See README.md in the archive.