The predictive accuracy of population viability analysis: a test using data from two small mammal species in a fragmented landscape

Abstract

This study examines the predictive accuracy of the population viability analysis package, ALEX (Analysis of the Likelihood of EXtinction). ALEX was used to predict the probability of patch occupancy for two species of small native Australian mammals (Antechinus agilis and Rattus fuscipes) among 13 patches of suitable habitat in a matrix of plantation pines (Pinus radiata). The study was retrospective, running each simulation from 1900 until 1997, and the model parameterised without knowledge of the 1997 observed field data of patch occupancy. Predictions were made over eight scenarios for each species, allowing for variation in the amount of dispersal between patches, level of environmental stochasticity, and size of habitat patches. Predicted occupancies were compared to the 1997 field data of patch occupancy using logistic regression, testing H random, that there was no relationship between observed and predicted occupancy, and H perfect, that there was a perfect, 1:1 relationship between observed and predicted occupancies. Rejection of H random and failure to reject H perfect was taken as a good match between observed and predicted occupancies. Such a match was found for one scenario with R. fuscipes, and no scenarios with A. agilis. In general, patch occupancy was underestimated, with field surveys finding that 9 of the 13 patches were occupied by R. fuscipes and 10 by A. agilis. Nonetheless, PVA predictions were in the 'right direction', whereby patches predicted to have a high probability of occupancy were generally occupied, and vice versa. A post hoc search over additional scenarios found few scenarios with a better match than the original eight. The results of this study support the notion that PVA is best thought of as a relative, rather than absolute predictor of the consequences of management actions in threatened populations.