Abyssal Claims: A FAIR-aligned integration platform for deep-sea and terrestrial mining transparency

Abyssal Claims is an independently developed, non-commercial web platform (https://something-rare.com) that integrates more than fifty publicly funded and open-licensed geospatial datasets onto a single interactive three-dimensional globe. It is designed to make the environmental and governance context of deep-sea mineral extraction and terrestrial industrial mining legible to researchers, journalists, policy analysts and the general public. Version 1.7 (September 2026) — a corrections release. Version 1.6 stated that the source code was not deposited; since 12 September 2026 it is, as a separate Zenodo record (10.5281/zenodo.22728476, AGPL-3.0-or-later; repository github.com/ice13ball/something-rare-engine). Release v1.1.0 of the code (10.5281/zenodo.22944144) contains every correction described in this version. This version corrects statements in version 1.6 that were false or had become stale — among them the soil organic carbon attribution (ISRIC SoilGrids, CC BY 4.0, not FAO GSOCmap), the Reports V2 species radius (10 km, not 55 km), the tailings compilation (WAPHA, Hudson-Edwards et al. 2023) and the dams reference (Global Dam Watch v1.0) — and gives layer, record and station counts from a single production snapshot of 23 September 2026. A new companion document, Corrections and changes since version 1.6, lists every correction individually with its source, including the 61 user-visible data errors found and fixed between 29 July and 23 September 2026. Layers: Vent Fluid Chemistry (MARHYS 4.0, Diehl and Bach 2024) is added. The Key Biodiversity Areas and WDPA layers were withdrawn on 3 September 2026, and the data drawn from the Global Tailings Portal on 23 September 2026, because their terms do not permit redistribution through an interactive map without the rights holders' permission; the data are retained but not served. The platform now serves 43 sea-domain and 13 land-domain layers. Method: a sample date is shown at exactly the precision the source gave it, and a synchronisation date is never allowed to stand in for the age of the data; sources are mirrored unmodified and suspected source errors are reported upstream. The modeled layers now open with a statement of status and intended use: they are static snapshots that pool records and predictors from different periods, have not been peer-reviewed or validated against independent observations, and are not suitable for management, conservation or regulatory decisions. Figures refer to the coral suitability fit of 14 September 2026 (AUC 0.872, Boyce 0.453; present-day corrosive fraction 0.637, newly corrosive fraction 0.107). Version 1.6 (29 July 2026) — first platform-authored modeled products, and a data-integrity pass. Four further sea-domain layers are added, taking the total to 42 sea layers and 15 land layers. Three of them are platform-derived modeled products rather than ingested third-party datasets, and each is framed as modeled — not observed — in the legend, the click panel and the exported provenance. (a) VME Coral Suitability (vme-suitability) — a MaxEnt species-distribution model fitted by the platform over NOAA DSCRTP presences of two reef-forming scleractinians, Desmophyllum pertusum and Solenosmilia variabilis, with target-group background sampling, spatial thinning and spatial-block cross-validation. Published model reef_scleractinia_v1 (trained 16 July 2026): n = 166 thinned occurrences, AUC 0.866, Boyce index 0.528, 3,837 predicted equal-area hex cells each carrying a per-cell uncertainty and an extrapolation flag. Predictors, in order of permutation importance: temperature (0.300), depth (0.073), substrate (0.073), oxygen (0.048), aragonite saturation (0.030), salinity (0.024) — drawn from GEBCO 2024 bathymetry and slope, seabed substrate (Dutkiewicz 2015, CC-BY-NC), WOA23, ISAS20 and GLODAPv2.2016b with aragonite saturation via PyCO2SYS. Coverage is predictor-limited rather than global: a cell is dropped where any seafloor predictor is unavailable, which is why 14,667 source occurrences reduce to 166 usable ones and the surface covers 3,837 of the 33,391 grid cells. The output is relative habitat suitability, not probability of presence. (b) Ocean Acidification (ocean-acidification) — aragonite and calcite saturation state (ΩA, ΩC) read verbatim from GLODAPv2.2016b's own published OmegaA / OmegaC fields, never recomputed, plus one explicitly platform-derived quantity: the aragonite saturation-horizon depth, and a preindustrial→present horizon shift reconstructed with PyCO2SYS (opt_k_carbonic = 10, Lueker 2000 — the constant set GLODAP itself used) from GLODAP's TCO2 and PI_TCO2. Both epochs pass through one identical solver configuration so that any solver bias cancels to first order in the difference; the published Ω field is never differenced against a reconstruction. QC: reconstructed present-day ΩA against GLODAP's published ΩA over 3,000 random valid cells gives a median difference of +0.00167, IQR [−0.0087, +0.0088]. (c) Coral Acidification Exposure (coral-acid-exposure) — a co-location analysis crossing the VME suitability surface with the reconstructed horizon shift and GEBCO seafloor depth, classifying every cell as newly corrosive, corrosive at preindustrial, supersaturated, or no data, and reporting a suitability-weighted exposure fraction beside a 0.3 / 0.5 / 0.7 threshold-sensitivity table. Across the 3,837 modeled cells (103 newly corrosive, 3,352 corrosive already at preindustrial, 382 supersaturated, 0 without data), two distinct suitability-weighted fractions are reported: the present-day corrosive fraction (habitat below today's horizon, including cells already below it before industrialisation) is 0.800, and the newly corrosive fraction (only cells that crossed between the epochs) is 0.059 — only the second is attributable to the epoch change; at suitability cut-offs of 0.3 / 0.5 / 0.7 the corresponding exposed fractions are 80.6% / 76.3% / 71.6% over 469 / 304 / 201 cells, with newly-corrosive fractions of 6.6% / 5.9% / 6.0%. The reference epoch is the GLODAPv2.2016b mapped reference year, approximately 2002. This is a double-modeled quantity and is presented as such: exposure denotes corrosive water, not coral loss — cold-water corals are documented living below the horizon at metabolic cost — and a two-epoch snapshot supports no rate or trend claim. Method, QC distribution, weighting formula and every disclosed assumption are in the deposited methods note. (d) Cumulative Human Impact (cumulative-human-impact) — the fourth new sea layer, and the only ingested one. It carries the present-state cumulative-impact grid of Halpern et al. 2025 (Science, doi:10.1126/science.adv2906), published by the National Center for Ecological Analysis and Synthesis on the KNB repository (doi:10.5063/F18K77KZ), CC0 1.0 Public Domain — the least restrictive licence in the catalogue. It is a dimensionless index summing roughly ten anthropogenic stressors across six categories, each weighted by the vulnerability of the ecosystem it acts upon: a modelled quantity, not a measurement. Values are not rescaled to zero-to-one — the present-state distribution has a median of 0.20, a 99th percentile of 0.78 and a maximum of 1.82, and the colour ramp is anchored on that observed range. The 3,617 × 1,814 Mollweide source raster is reprojected once onto a regular 0.1° EPSG:4326 grid and served as on-demand PNG tiles with a pickable hexagon view for exact values. Present state only: the package's SSP2-4.5 and SSP5-8.5 mid-century projections are deliberately excluded from this version rather than blended into one surface. The layer is carried as context, not accusation — seabed mining is only a minor component of the total, ISA claim areas sit in comparatively low-impact abyssal zones, and a concession polygon is not a causal source of the pressure measured around it. Marine Carbon co-location expansion. The unified hexagon readout gains acidification and seabed-substrate field groups and a nearest-measurements service (/v1/carbon/nearest-obs) returning the closest actual observations from eight ocean-chemistry sources — Argo, GEOTRACES, MEMENTO, WOD oxygen profiles, CASCADE sediment, methane seeps, hydrothermal vents and ISA DeepData stations — beside every modeled value, so a reader can always see what was measured near what was modeled. Values from different sources are reported side by side and are never blended into a fused metric. Data-integrity corrections. (i) MEMENTO: the source declares CH4 and N2O in both a dissolved and an atmospheric sphere, separated by a parameterType that its CSV export does not emit; both spheres were therefore landing in one column. The platform now derives per-gas atmospheric flags (source label plus a per-cruise-median, per-value rule), recomputes water-only surface rollups, and badges flagged values rather than hiding them — 7,461 CH4 and 73,433 N2O value-bearing samples are flagged, while genuine seep signals are preserved. The rule is platform-derived and disclosed as such. (ii) ONC: the ADCP layer was rebuilt against ONC's asynchronous dataProductDelivery depth-binned product (RADCPTS / NTS) after the previous scalardata path was found to expose no backscatter at all; the never-populated acoustic-events table was retired rather than left in place. (iii) OBIS: corals are now selected by taxonomic rank and exempted from the 50 m depth floor (2,480,586 coral records across 6,046 species; Scleractinia 1,195,551 — the complete set OBIS reports). Separately, an eight-million-row duplication in the same table was found and corrected — disclosed here because it changes the size of the platform's largest table and therefore the result of any query run against it. OBIS regenerates a record's identifier on republication, so the insert guard never recognised a republished record and orphaned the generation before it: 40,774,532 rows stored against the 32,724,724 the snapshot yields, about a fifth of the table. T