Global maps overstate the reliability of blue-carbon credits

Abstract

This archive contains the derived data, harmonised labels, frozen evaluation splits, and complete numerical results supporting the study "Global maps overstate the reliability of blue-carbon credits." It provides everything required to reproduce every figure and number in the paper, and a reusable benchmark for testing the geographic transferability of blue-carbon stock models. Contents. data_processed/ harmonised mangrove soil organic carbon labels (2,489 cores integrated to a standardised 0–100 cm stock, Mg C ha⁻¹, from the Coastal Carbon Network Data Library); aboveground biomass carbon points (from the Simard et al. 2019 30 m product, converted with a 0.47 carbon fraction); feature tables joining these labels to a covariate stack (CHELSA bioclim 1–19, GSOCmap, SoilGrids, Copernicus land cover, distance-to-coast and distance-to-river, and EOT20 tidal range/form factor); the frozen leave-one-delta-out fold definition for eight named core deltas (delta_registry.csv) and the data-driven 29-region leave-one-region-out folds used for the scale/robustness analysis (region_registry.csv, region_folds.csv); and one JSON file per experiment recording all reported results (three-tier validation, area-of-applicability, conformal coverage, region-scale test, aboveground-biomass and total-carbon estimates, published-map evaluation, terrestrial comparison, covariate- and concept-shift diagnostics, GSOC ablation, AOA threshold sensitivity, few-shot calibration, and the carbon-crediting stock-span calculation). manifests/ per-source provenance (source, DOI, file names, sizes, checksums, URLs) for the public input datasets (CCN, Global Mangrove Watch v3, Simard 2019, EOT20). config/ delta and region bounding-box definitions and the full source catalogue. README.md and SHA256SUMS.txt — file-by-file documentation and integrity checksums. Scope and limitations. This deposit contains derived data and results only. It does not include the large public raster inputs (CHELSA, SoilGrids, GSOCmap, Global Mangrove Watch, Simard 2019, EOT20, WoSIS), which are not redistributed here; their exact sources are documented in manifests/ and config/sources.yaml, and the full acquisition-to-analysis pipeline that regenerates this archive from those public sources is provided in the code repository. Derived data are released under CC-BY-4.0; the underlying source datasets retain their original licences and should be cited alongside this archive. Code. The complete reproducible pipeline, pinned software environment, and single-command runner are available at https://github.com/rakibhhridoy/mdbc-blue-carbon