Avian Conservation and Ecology (Dec 2019)
Model transferability and implications for woodland management: a case study of Pinyon Jay nesting habitat
Abstract
Understanding the transferability of ecological models from one area to another is important for the effective conservation of species of management concern, particularly when the ability to sample across multiple areas is limited. Two measures of transferability, area under the receiver operating curve (AUC) and threshold-specific classification accuracy (sensitivity and specificity), are often employed in assessing models of bird occurrence or resource selection. Although informative, these measures may have limited practical utility in guiding on-the-ground habitat management like forest thinning, which often relies on simple metrics such as tree diameter. We addressed this challenge in studying the occurrence of Pinyon Jay (Gymnorhinus cyanocephalus) nest locations in four piñon-juniper woodland sites in New Mexico, USA. Using generalized linear mixed models, we employed covariates describing woodland structure at several scales to discriminate nest from non-nest plots. We found fair transferability of model predictions between sites via AUC (mean = 0.71), fair threshold-specific specificity (mean = 67%), and poor threshold-specific sensitivity (mean = 42%). Under a hypothetical scenario of forest thinning, we employed a covariate predictive in models of each site, nest (or non-nest plot center) tree root crown diameter (RCD), to assess a management-practical measure of transferability. Using critical RCD values at which 75% of nest trees were retained, we found fair transferability between sites (MEAN = 70%), though we observed retention of nest trees as low as 21%. Average retention increased to 75% (minimum = 60%) when information from multiple sites was combined. This example illustrates that the application of information from one area to the management of another may be effective but may also result in loss of important habitat or other resources. We found that management of Pinyon Jay nesting habitat should proceed on a single-site basis when local nesting information is available. Caution should be exercised when prescribing management in locations where information on nesting is lacking, and information from as many nearby sites as possible should be employed.