Prediction and assessment of edge response and abundance for desert riparian birds in southeastern Arizona

Abstract

The ability to reliably predict the impacts of habitat loss and fragmentation for many species in diverse environments is necessary to identify conservation priorities and to accelerate conservation planning. However, no general relationship has emerged that will allow a priori predictions of the expected abundance response of a given species to edges, or to reliably predict absolute abundance in novel locations. The overall objectives of my dissertation were to predict and assess edge response and abundance in novel locations to aid in conservation planning. In Chapter 1, I used a meta-analytic approach and AICc model selection to predict positive and negative edge response for birds on both sides of forest edges with ecological and life-history trait information readily available from the literature. Using Receiver and Operating Characteristic (ROC) analysis to compare observed versus predicted edge response on the same dataset in which the models were developed, I found that the percent of observations correctly classified for positive and negative edge prediction models ranged from 74-78% for the forest-open edge type and from 82-89% for the open-forest edge type. In Chapter 2, I applied the prediction models to 25 previously unstudied species in 8 sub-edge types in the San Pedro River riparian corridor to assess model robustness to novel locations, edge types, and species. The models performed well for predicting negative edge response in both forest-open and open-forest edge types (80-91% correct classification), adequately for positive edge response in the forest-open edge type (64-77% correct classification), but poorly for positive edge response in the open-forest edge type (38-44% correct classification). Using logistic regression analysis I found that classification success was robust to novel edge and habitat types, and that positive prediction models in the open-forest edge type may have failed due to a different set of mechanisms occurring on the San Pedro River compared with the meta-analysis dataset. In Chapter 3, I validated and calibrated a landscape ecological model that uses change in both habitat composition and geometry to predict change in species abundance at novel locations (Effective Area Model; EAM). For model validation, I compared the EAM to a null model in terms of its ability to accurately predict observed species abundance in 50 validation sites different from those in which the model was parameterized. The EAM outperformed the null model when considering all validation sites as well as subsets of validation sites categorized on the basis of four variables (isolation, presence of water, region, and focal habitat). For model calibration, I explored a framework to decrease prediction error given inherent temporal and spatial variability in abundance in an attempt to improve the ability of the EAM to predict absolute abundance in novel landscapes. I calibrated the EAM to new locations using a linear regression between observed and predicted EAM abundance with and without additional habitat covariates. I found that model calibration used to account for unexplained variability in time and space, as well as variability that can be explained by incorporating additional covariates, improved EAM prediction of absolute abundance in novel locations.