Extrapolating rock physicochemical properties from discrete sample data using computer vision - Supplementary Data

If using this data for publication please cite: Grant et al. 2026. Extrapolating rock physicochemical properties from discrete sample data using computer vision. PNAS NEXUS. Files: GeoCLR_classifications.csv- Contains the classification output of the GeoCLR model using in this study.- Each row corresponds to a training image patch where depth and arc are the location (meters) of that patch on the cyclindrical core surface. Location corresponds to the top left corner of each patch. - The groundthruth column contains the expert generated labels used for model validation. - The label column contains the 'pseudolabels' given to each patch by GeoCLR. Grant_et_al_2026_PNAS_NEXUS_Supplementary_data_tables.xlsx- Core averages: Table containing visual abundance estimates of lithological classes derived from shipboard core description during IODP Exp. 390 and post-expedition using GeoCLR- Combustion results: Table of bulk carbon data for Hole U1557D clasts and carbonates- MAD Results: Shipboard moisture and density data for Hole U1557D clasts and matrix samples generated during IODP Expedition 390.