An open smelter-resolved spatial prior for CF4 (PFC-14) emissions

An open spatial prior for CF4 (PFC-14) emissions from primary-aluminium smelting, gridded by smelter location, with the open 94-smelter registry behind it (95 rows incl. one idle plant; coordinates, nameplate capacity, 2020 operating status, per-row source notes; the technology column ships unpopulated). Three weighting variants (capacity, presence-only, and production-rescaled to USGS 2020 national output — the recommended global variant), at 0.1° Europe and 1° global resolutions, plus files formatted on the grid of the egusphere-2025-5656 deposited emission fields. Values are relative spatial weights (units "1"): renormalize and rescale to your sectoral total before use; never absolute emissions; exactly zero outside smelter cells — do not use as a standalone total-CF4 prior (blend with a smooth background per your sectoral split). Evaluation (Europe, ICOS PARIS 2020 flat-prior 6-member CF4 posterior ensemble, exact-2020 fields, vs EDGAR v8.0): the smelter field correlates with the posterior better than the population/built-up proxy in Iceland (0.25 vs 0.00; significant under fixed-tile block-bootstrap and shift-permutation tests in both covering members — both RHIME, i.e. one inversion system) and directionally in 5 of 6 members over the pooled smelter countries (0.051 vs 0.022; no member individually significant). Norway, Spain and the UK are EDGAR-favoured; the only individually significant French results favour EDGAR (system-dependent disagreement). On a common per-cell-mass basis with midrank ties, the posterior places the 22 in-domain smelter cells at the 72nd percentile vs EDGAR's 59th, and EDGAR assigns exactly zero CF4 to 15 of the 22 (including all three Icelandic smelter cells); the construction-independent posterior/prior ratio puts the smelter cells at the 70th percentile (Alcoa Fjarðaál 99th). Recommended use: a citable aluminium-sector location layer and a prior-sensitivity member alongside the standard prior. The aluminium sector is ~60–80% of global CF4; combine with an electronics/other component. Version 3 (2026-06-12), year-selection correction: v2 was published while the audit was still propagating one finding; this version completes it. Version 2 changelog (superseded numbers corrected here): the gridded fields and registry are UNCHANGED from v1 (all data values verified to reproduce exactly from the published code, and the correlations verified insensitive to the density-vs-mass basis choice). Two corrections to the evaluation and documentation: (1) a year-selection bug was found and fixed — the harness selected the posterior year by nearest-neighbour time matching, which silently returned the 2019 field for the four ELRIS/InTEM ensemble members (their mid-year stamps are equidistant from 2020-01-01), so v1's ensemble-wide numbers mixed 2019 and 2020 fields; all numbers are recomputed from exact-2020 fields (Iceland, covered only by the two RHIME members, was unaffected; the pooled directional count is now 5 of 6 members). (2) The enrichment metric was rebuilt on a consistent per-cell-mass basis with midrank tie handling (the v1 "60th vs 26th percentile" framing mixed bases and floored tied zeros — corrected to 72nd vs 59th, with the flat-prior column and the posterior/prior ratio now reported); the EDGAR zero-cell count corrected from 14 to 15 of 22; China's registry-capacity share corrected from 28.0% to 27.5%; a France correlation quoted in the v1 description was withdrawn (the only individually significant French results favour EDGAR and are reported symmetrically); Iceland's significance now always carries the two-members-one-system qualifier; the Kim et al. (2021) precedent is described accurately. Method note Rev 3 reflects all of the above. Code (MIT): https://github.com/zachdissington/climate-solutions-research-kit (priors/pfc-aluminium-prior/). Author ORCID: 0009-0007-0196-1371.