Poor transferability of a distribution model for a widespread coastal marsh bird in the southeastern United States. Issue 3 (3rd March 2017)

• Record Type: Journal Article
• Authors: Roach, Nicolette S.; Hunter, Elizabeth A.; Nibbelink, Nathan P.; Barrett, Kyle
• Is Part Of: Ecosphere. Volume 8:Issue 3(2017)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2017-03-03
• Subjects: abundance -- Clapper Rail -- model validation -- Rallus crepitans -- salt marsh -- species distribution models -- wetlands; Ecology -- Periodicals; Ecology; Periodicals; 577.05
• DOI: 10.1002/ecs2.1715
• Languages: English
• ISSNs: 2150-8925

Abstract

Abstract: Species distribution models have been applied across a wide range of spatial scales to generate information for conservation planning. Understanding how well models transfer through space and time is important to promote effective species–habitat conservation. Here, we assess model transferability in coastal tidal marshes of the southeastern United States using count data of a widespread marsh bird: the Clapper Rail ( Rallus crepitans ). We developed species–habitat models at a state level in both South Carolina and Georgia, and then assessed how well top models from each state predicted abundance in the other state. Internally (locally) validated models performed well with reasonable fit (SC: R 2 = 0.35, GA: R 2 = 0.14), and high significance ( P = 0.0005); however, both models performed poorly when predicting abundance from the other state ( R 2 = 0.03 and 0.003). To assess the consequences of this lack of transferability, we applied the South Carolina‐ and Georgia‐derived parameter estimates to habitat features in South Carolina and identified the top 25% of tidal marsh habitat that each model predicted within the state. There was minimal overlap between model habitat quality predictions (<5%). Our results address the predictive power and uncertainties that arise from using habitat associations and climate models to predict species distributions or abundance in locations without training data. We discuss potential reasons model transferability was not successful and address the need for better regional datasets and the importance of intraspecific variability in response to environmental gradients.