Validation for the Biodiversity Intactness Index (BII) from one
NHM source, selected via datasource:
* "Phillips" – Phillips et al. 2021 (BII-BTE, DOI 10.5519/he1eqmg1),
country-level 1970-2014.
* "DePalma" – De Palma et al. 2024 NHM v2.1.1 (DOI 10.5519/k33reyb6),
cellular 5-yearly 2000-2020, aggregated to ISO here.
De Palma et al. 2024 differs from Phillips et al. 2021 in three steps: stricter primary-vegetation baseline (minimally/lightly-used primary only, vs Phillips's all-primary including intensely-used sites), logit (not log) transform, and balanced Bray-Curtis (abundance-weighted) compositional similarity in place of Phillips's asymmetric Jaccard (species-identity only). De Palma's absolute BII is markedly lower, with a non-uniform gap that depends on land-use intensity, so the two series are NOT splice-compatible and are kept as separate model labels.
MAgPIE's own BII output (modules/44_biodiversity, coefficients from Leclère et al. 2020 "Bending the Curve" via PREDICTS) is on the Phillips scale, so it is close to Phillips et al. 2021 (~0.03 above) but sits markedly above De Palma et al. 2024. For absolute-level validation use Phillips; for harmonized (change-from-baseline) plots either series works, but mixing MAgPIE projections with De Palma historical on a non-harmonized axis is misleading. Two things drive the MAgPIE-vs-De Palma gap: the PREDICTS coefficient version (Phillips vs De Palma), and land-use allocation – MAgPIE land use is LUH3-initialised with FAO/FRA forest correction (mrlandcore::calcForestArea), vs the LUH2 underlying De Palma's v2.1.1. A 2010 decomposition shows the coefficient change dominates the gap (~0.17 of the ~0.20 total globally, and larger than the land-use term in every H12 region); the land-use/forest difference is real but secondary (~0.03 globally).