Date of Award


Degree Type


Degree Name

Master of Science



Major Professor

Liem Tran

Committee Members

Henri D. Grissino-Mayer, Frank T. van Manen


The main objective was to create habitat models of three plethodontid salamander species (Desmognathus conanti, D. ocoee, and Plethodon jordani) in GSMNP. To investigate the relationships between salamanders and their habitats, I used three models—logistic regression with use-availability sampling, logistic regression with case-control sampling, and Mahalanobis distance (D2)—for each species to gain a robust view of the relationships. The secondary objective was to compare the different modeling methods within and across the three species. Elevation was the dominant variable for all three species.

D2 for D. conanti predicted low elevations, close proximity to streams, metasandstones, and previously disturbed areas. The use-availability model indicated habitat in low elevation, settled areas, pine understory, and flood overstory. The case-control model for included only elevation and undisturbed areas. Case-control and D2 predicted presences >90% correctly but absences <50% correctly. Use-availability was more balanced with 75% presences and 60.5% absences correct.

D. ocoee occurred only at the highest elevations. D2 was influenced by proximity to streams, undisturbed areas, northern hardwood overstory, and frigid sandstones. Use-availability included a positive association with increasing elevation and a negative association with spruce understory. Case-control included elevation only. Use-availability did poorly at predicting presence (48.7%). Case-control predicted presences and absences at 79.5% and 89.0%, respectively. D2 classified presences and absences at 92.3% and 78.1%, respectively.

P. jordani was also determined by elevation, but lower than D. ocoee. D2 also included proximity to streams, undisturbed areas, and sandstones. Use-availability had a negative association with spruce understory. Pine understory, northern hardwood overstory, and distance to streams were negatively associated with P. jordani occurrence in case-control. D2 predicted 97.1% of presences and 66.0% of absences correctly. Use-availability predicted 64.1% of presences and 91.8% of absences correctly. Case-control predicted 74.8% of presences and 89.1% of absences correctly.

Use-availability worked best for D. conanti, but may have been a result of the uncertain identification of the species. For the other two species, case-control had a high classification rate for both presences and absences and a more intuitive answer for what determines habitat than D2.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."