. . . . "2026-05-03"^^ . . "HEALPix family validated as common DGGS substrate for biodiversity × Copernicus EO × Destination Earth integration" . "The HEALPix family is validated as fit for purpose as the common DGGS substrate for integrating biodiversity occurrence data with high-resolution Copernicus EO products and Destination Earth climate-model output. The validation rests on six fitness criteria, evaluated end-to-end across nine grid systems on synthetic uniform sphere data and on real GBIF Quercus suber occurrences. (1) Equal-area correctness: HEALPix-on-sphere has a small but systematic ~0.7% area bias at boreal latitudes vs WGS84; HEALPix-on-WGS84 via authalic-sphere mapping (healpix-geo / GRID4EARTH) and rHEALPix resolve it. (2) Cell-shape preservation: all DGGS family members preserve compact cells across latitudes; the projection family does not. (3) Hierarchical refinement: HEALPix NESTED bit-shift makes parent-child operations O(1) integer arithmetic, faster than alternative lookup-based hierarchies. (4) Ellipsoidal correctness: healpix-geo and rHEALPix provide it; HEALPix-on-sphere does not. (5) Iso-latitude pixelization: HEALPix-only among the family; makes zonal climate-zone analyses essentially free. (6) ML-ecosystem compatibility: HEALPix is the substrate for spherical-CNN architectures (DeepSphere), scattering networks (foscat / FIESTA stack), and native sphere-harmonic transforms (healpy.map2alm); alternative DGGS lack this depth. " . . "Notebook 01: 1,000,000 uniform random points on the sphere binned on 5° lat-lon vs HEALPix nside=16 NESTED produce a 23× count bias at the equator vs 85°N from cell-area geometry alone, matching the analytical prediction. Notebook 02 + 07: 20,100 Quercus suber GBIF occurrences aggregated onto eight equal-area grids (HEALPix-sphere, HEALPix-geo-WGS84, H3 res 3, rHEALPix res 4, Mollweide 100 km, EEA reference grid 100 km, ISEA3H res 8) plus the lat-lon cautionary baseline produce visually identical density patterns over the species' Mediterranean range, while lat-lon distorts. Notebook 03: aspect-ratio chart from 0° to 70°N for nine grids — DGGS family stays near aspect 1.0–1.3, projection family diverges (Behrmann 5.0 at 65°N). Notebook 04: 3×3 ML kernel at 65°N, 15°E shows Behrmann aspect 5.0 (long N–S strip), Mollweide 1.3, lat-lon 1.2, EEA reference grid 1.0 (azimuthal preserves shape near 52°N centre), HEALPix-sphere 1.3, HEALPix-geo 1.3, rHEALPix 1.2, H3 res 3 1.0, ISEA3H res 8 1.2. Notebook 06: HEALPix NESTED parent–child relation verified as integer bit-shift (parent = pix >> 2, children = pix << 2 | k) against healpy.pix2ang/ang2pix. Notebook 08 Section A: pyproj.Geod-exact ellipsoidal area vs HEALPix-on-sphere area shows +0.45% at the equator, −0.65% at 65°N, −0.88% at 85°N — total 1.3 percentage-point swing across populated latitudes. Notebook 08 Section C: 12 pixels in a HEALPix nside=16 ring confirmed at identical colatitude to 1e-10 precision, vs H3 hex tessellation which breaks iso-latitude alignment.\n\nRepository: https://github.com/annefou/dggs-biodiversity-bias\n" . "The validation is a multi-criteria fitness-for-purpose evaluation, not a single-metric benchmark; the relative importance of the six criteria depends on the use case. For biodiversity counts in isolation at coarse resolution, any equal-area DGGS satisfies the count-correctness criterion; the HEALPix-specific advantages (NESTED bit-shift, iso-latitude, ML-ecosystem compatibility, ellipsoidal-correctness path) become decisive in the integrated regime where biodiversity is stacked at high resolution with Copernicus EO and Destination Earth climate-model output, less so in standalone biodiversity workflows. The ~0.7% sphere-vs-WGS84 area bias is small per-cell; the case that it compounds materially across millions of cells × decades into climate-attribution errors is sketched (notebook 08 Section A) but not fully quantified by a stacked-product end-to-end demonstration. The real-data demonstration uses a single species (Q. suber, 20,100 occurrences); broader cross-taxon and global-coverage tests are out of scope. The ML-ecosystem advantage is qualitative (HEALPix has DeepSphere / foscat / healpy.map2alm; alternatives do not) rather than a head-to-head benchmark on a specific ML task." . . . . . "Anne Fouilloux" . "2026-05-04T21:19:00.429Z"^^ . . . . . "NP created using Declaring a replication study outcome according to FORRT" . . "RSA" . "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" . "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" . . .