# [Bil17] How to Take into Account Uncertainty in Species Extinction Probabilities for Phylogenetic Conservation Prioritization

**Revue Internationale avec comité de lecture : **
*Journal Environmental Modeling and Assessment*,

vol. 22(6),

pp. 535-548,
2017, (

doi:

10.1007/s10666-017-9561-7)

**Mots clés: ** Biodiversity conservation, phylogenetic diversity, protected taxa, extinction risk, robustness, optimization, integer programming, experiments.

**Résumé: **
In this article we are concerned with the general problem of choosing from a set of taxa T a subset S to protect in order to try to contribute to halting biodiversity loss as efficiently as possible given limited resources. The protection of a taxon decreases its extinction probability and the impact of protecting the taxa of S is measured by the resulting expected phylogenetic diversity (ePD) of the set T. The primary challenge posed by this approach lies in determining the extinction probability of a taxon (protected or unprotected). To deal with this difficulty, the uncertainty about the extinction probabilities can be described through a set of possible scenarios, each corresponding to different extinction probabilities values for each taxon. We show how to determine an “optimal robust set” of taxa to protect, defined as the set of taxa that minimizes the maximum “regret”, i.e. the maximum relative gap, over all the scenarios, between 1) the ePD of T obtained by protecting the taxa of this set and 2) the ePD of T which would be produced by protecting the subset of taxa optimal for the considered scenario. In our experimental conditions covering one hundred cases, this gap is almost always less than 1%. Consequently, the ePD of T obtained by protecting the taxa of the optimal robust set is not far from the maximum ePD of T that could have been obtained if we had known the true scenario. In other words, a way of escaping (in large part, at least) from the uncertainty related to the extinction probabilities of the taxa consists of choosing to protect those belonging to the optimal robust set. We also compare the optimal robust set to other relevant subsets of T.