Predicting where rare species live by relying on their cousins

A three-phase protocol with progressive stages (in silico, minimal validation, full validation) and clear GO/NO-GO/PIVOT criteria, which limits the risk of unnecessary investment and permits methodological adjustments.

The hypothesis addresses a real and non-trivial problem: niche inference for data-poor species, where classical SDMs suffer from overfitting. The idea of using phylogenetic covariance as a prior is conceptually elegant and aligns with the family of shrinkage methods that have proven their worth in Bayesian statistics and population genomics.

The negative-control battery (identity matrix, permuted trees, spatial splines) constitutes a genuinely considered design element that most phylogenetic SDM papers omit; it at least attempts to separate phylogenetic signal from generic regularisation.

Clear addressable market: environmental consultancies (ERM, Ramboll, WSP), conservation agencies (USFWS, ONCFS, IUCN), and extractive industries subject to ecological compensation requirements must model rare species for permitting purposes. An AUC gain of 0.08–0.15 on species with <20 occurrences directly reduces regulatory uncertainty and field survey costs (estimated savings of €30–50k per project).

A falsifiable hypothesis with clear quantitative GO/NO-GO criteria (ΔAUC, p-value, negative controls, Pagel's lambda), which is highly valued by reviewers for a high-risk project.