Testing spatial PVA models of Australian treecreepers (Aves: Climacteridae) in fragmented forest

Abstract

Population viability analysis (PVA) and other stochastic population models are frequently built and often used, but rarely tested. Stochastic metapopulation models of the White-throated Treecreeper (Cormobates leucophaea) and the Red-browed Treecreeper (Climacteris erythrops) were developed in a system of 39 remnant patches of eucalypt forest in southeastern New South Wales, Australia. Parameters of the model were estimated using data obtained outside the fragmented system. Field surveys of the patches were conducted to test the predicted probabilities of patch occupancy, which is one of few instances where stochastic population models have been tested with empirical data. The initial models underestimated the occupancy of the patches, and the models were modified using the results of the tests in conjunction with further information on the biology of the species. A number of different modifications were made to determine changes that produced results that matched the observations. The best of these modifications made reasonable predictions, although this is not equivalent to a test with independent data because the data were known prior to the modifications. The best-fitting modified models were tested by comparing the observed number of extinction and colonization events to the predicted number. The models underestimated the observed number of events although imperfect survey methods may have contributed to these differences. The tests of the stochastic models contributed to their development by highlighting the nature of the predictive error. The modified models predicted that the White-throated Treecreeper would be likely to persist over the next 100 years in most of the 39 patches. In contrast, the Red-browed Treecreeper was predicted to become extinct in most patches within about 50 years of fragmentation. This study illustrates how spatial patterns can be used to test the predictions of population dynamic models, although we note that the tests are limited by survey error and spatial correlation in occupancy data. Show more