Improving generation length estimates for the IUCN Red List

Abstract

The International Union for the Conservation of Nature (IUCN) Red List classifies species according to their risk of extinction, informing local to global conservation decisions. Here we look to advance the estimation of generation length, which is used as a time-scalar in the Red List as a way of accounting for differences in species' life-histories. We calculated or predicted generation length for 86 species of antelope following the Rspan approach. We also tested the importance of both allometry (body-mass) and phylogeny (phylogenetic eigenvectors) as predictors of generation length within a Phylogenetic Eigenvector Map (PEM) framework. We then evaluated the predictive power of this PEM and two binning approaches, following a leave-one-out cross-validation routine. We showed that captive and wild longevity data are nonequivalent and that both body-mass and phylogeny are important predictors for generation length (body-mass explained 64% and phylogeny 36% of the partitioned explained variance). Plus, both the PEM, and the binning approach that included both taxonomic rank and body-mass, had good predictive power and therefore are suitable for extrapolating generation length to missing-data species. Therefore, based on our findings, we advise separating captive and wild data when estimating generation length, and considering the implications of wild and captive data more widely in life-history analyses. We also recommend that body-mass and phylogeny should be used in combination, preferably under a PEM framework (as it was less reliant on available reference species and more explicitly accounts for phylogenetic relatedness) or a binning approach if a PEM is not feasible, to extrapolate generation length to missing-data species. Overall, we provide a transparent, consistent and transferable workflow for improving the use of the Rspan method to calculate generation length for the IUCN Red List.

Fig 1. Phylogenetic tree of Bovidae (128 species), based on an updated supertree of extant mammals [12].

Those species included in our models are marked in black (86 species). Additional analyses were also performed excluding chiru (Pantholops hodgsonii) and African buffalo (Syncerus caffer) (S3 Appendix).

Fig 2. An RDI (Raw data, descriptive statistic and inferential statistic) plot of age at last reproduction in the wild (ALRw) and age at last reproduction in captivity (ALRc)—for 50 species (excluding 36 species with missing data).

The plot includes the mean as a thick black line, 95% confidence intervals as a black rectangle, a grey bean plot of data density and grey points of jittered data.

Fig 3. Observed and PEM (Phylogenetic eigenvector map) predicted values of generation length in the wild (GLw) obtained following leave-one-out cross-validation—for 54 species (excluding 32 species with missing data).

The dashed line is a 1:1 line, the solid black line is a regression line of observed values as a function of predictions, and the grey envelope represents the 95% confidence limits of the regression line. Horizontal bars are limits of the 95% confidence intervals.

Fig 4. Observed and binning approach predicted values of generation length in the wild (GLw) obtained following leave-one-out cross-validation—for 54 species (excluding 32 species with missing data).

Predictions are presented when incorporating bins of log₁₀ body-mass (A) and irrespective of body-mass (B). The dashed lines are 1:1 lines, the solid black lines are regression lines of the observed values as a function of predictions and the grey envelopes represent the 95% confidence limits of the regression lines.

Fig 5. A comparison of predicted generation length in the wild (GLw) following a Phylogenetic Eigenvector Map (PEM) approach and a combined binning approach—for 32 true missing-data species.

The plot includes the mean as a thick black line, confidence intervals as a black rectangle, a grey bean plot of data density and grey points of jittered data.