MODELING NORTHERN GOSHAWK (ACCIPITER GENTILIS) NESTING HABITAT ON THE LEWIS AND CLARK NATIONAL FOREST USING EIGENVECTOR FILTERS TO ACCOUNT FOR SPATIAL AUTOCORRELATION
Habitat suitability modeling has become a valuable tool for wildlife managers to identify areas of suitable habitats for management and conservation needs. The Northern goshawk (Accipiter gentilis) has been the focus of many modeling efforts, however, the current models guiding goshawk management on...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | unknown |
| Published: |
University of Montana
2014
|
| Subjects: | |
| Online Access: | https://scholarworks.umt.edu/etd/4355 https://scholarworks.umt.edu/context/etd/article/5381/viewcontent/Lehr_Morganne_FinalThesis.pdf |
| Summary: | Habitat suitability modeling has become a valuable tool for wildlife managers to identify areas of suitable habitats for management and conservation needs. The Northern goshawk (Accipiter gentilis) has been the focus of many modeling efforts, however, the current models guiding goshawk management on the Lewis and Clark National Forest may not fully capture the unique habitat characteristics that the goshawk is actually selecting for nesting habitat. Therefore, the first objective of this study was to explore the use of Maxent for modeling suitable goshawk nesting habitat on the Lewis and Clark National Forest in central Montana. However, goshawk territoriality and their use of alternate nest locations creates, spatial autocorrelation between the nest locations (nest locations that occur close to one another are not independent) and can complicate the development of a habitat suitability model. Spatial autocorrelation can have drastic effects on model prediction and can lead to false conclusions about ecological relationships, but when accounted for can lead to insights that may have been otherwise overlooked. As a result, this study also explored the use of eigenvector filters as additional explanatory variables to assist in “filtering” out the effects of spatial autocorrelation from the modeling effort. Furthermore, this study evaluated the difference in model outputs using different resampling methods (bootstrap and cross-validation) and number of variables to determine the differences between models. The results of the study showed that the use of eigenvector filters not only improved model performance and reduced commission error, but created more precise predictions of suitable habitat. Furthermore, this study also found that using bootstrap methods and all biologically relevant environmental variables (with the additional of eigenvector filters) provided the best overall model. However, wildlife managers should closely review the methods and results provided in this study and choose the model that best ... |
|---|