Masters Theses

Date of Award

8-1995

Degree Type

Thesis

Degree Name

Master of Science

Major

Wildlife and Fisheries Science

Major Professor

David A. Buehler

Committee Members

Ralph Dimmick, Stuart Pimm

Abstract

Concern over apparent population declines of inland neotropical migrant birds in the United States has focused attention on the relationship between songbird habitat and forest management. To develop songbird habitat models and to assess the effects of forest management on songbirds, I surveyed breeding bird populations between 15 May and 1 July 1992 and 1993, using 20-minute, 50-m fixed-radius point counts on the Cherokee National Forest in eastern Tennessee. To assess habitat associations, I measured vegetation and physical habitat parameters at each point-count location on 0.04-ha circular plots. A sample of ~200 census points were randomly selected from the U.S. Forest Service Continuous Inventory of Stand Conditions (CISC) database for the Tellico Ranger District. Census points were stratified into 6 broad forest type classes and 3 stand condition classes.

We recorded 60 and 65 species of birds within 50 m on point counts in 1992 and 1993, respectively. Neotropical migrants comprised 73% of all species observed in 1992 and 78% of all species observed in 1993.

Optimal predictive models of habitat selection patterns by seven of the ten neotropical migrant songbird species deemed highest priority for management in the Southern Blue Ridge Mountains (acadian flycatcher (Empidonax virescens), black-throated blue warbler (Dendroica caerulescens), Canada warbler (Wilsonia canadensis), chestnut-sided warbler (Dendroica pensylvanica), hooded warbler (Wilsonia citrina), wood thrush (Hylocichla mustelina), and worm-eating warbler (Helmitheros vermivorus), were generated through stepwise logistic regression and best-subset selection techniques and evaluated using Hosmer and Lemeshow's goodness-of-fit test and Wald's chi-square test. Unbiased correct classification (jackknife) rates for the final species models varied, with chestnut-sided warbler showing the strongest model (93.5% correct classification) and hooded warbler showing the weakest model (64.5% correct classification).

The best predictive model of acadian flycatcher distribution on the Tellico Ranger District contained five habitat variables - elevation, litter depth, basal area of saplings, stand age, and 38-53 cm dbh tree size class. The best black-throated blue warbler model contained six variables - elevation, % cover by Vaccinium spp., litter depth, 53-68 cm dbh tree size class, ground cover %, and % cover by rhododendron (Rhododendron maximum). The Canada warbler model consisted of six variables - elevation, % cover by rhododendron, # of conifer trees, # tree species, % slope, and # standing snags. Chestnut-sided warbler distribution was best predicted by three habitat variables - elevation, canopy height, and litter depth. The hooded warbler stepwise model contained five variables -15-23 cm dbh tree size class, % shrub cover, elevation, % slope and forest type. The wood thrush model contained three variables - 30-38 cm dbh tree size class, 53-68 cm dbh tree size class, and canopy height. The worm-eating warbler model contained six habitat variables - elevation, slope, # tree species, forest type, # deciduous trees, and total basal area.

Overall, elevation was the most important (P ≤ 0.05 - Wald Chi-square test) variable in predicting species' distributions, occurring in six of the seven priority species models. Three of seven models contained (P ≤ 0.05 - Wald Chi-square test) slope and litter depth components.

I also used habitat parameters to develop predictive models for patterns of avian species richness and abundance. Models of species richness and abundance containing all measured and derived habitat variables (n=62 variables) for neotropical migrant and resident songbirds explained 29 to 35% of the variation in the data (R2 = 0.29 - 0.35). Patterns of avian diversity, therefore, could not be predicted with a high degree of accuracy at this scale using standard forest vegetation variables.

I also used habitat variables available in the CISC database to develop models to predict the seven priority species' distributions (logistic regression) and avian species richness and abundance (linear regression). The CISC database yielded well-fitting models for the seven priority species (P ≤ 0.05) with correct classification rates (jackknife) ranging from 63% to 92%. Elevation was important (P ≤ 0.05 - Wald Chi-square test) in six of the seven priority species models. Selection patterns in cove hardwood, northern hardwood, and oak/hickory forest types were important (P ≤ 0.05 - Wald Chi-square test) in four of the seven CISC species models. CISC models for neotropical migrant and permanent resident richness and abundance had moderate predictive power (R2 = 0.21 - 0.48). The CISC database, thus, may not be useful for modeling patterns of avian diversity at the district level, although it worked well for single-species models.

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

Share

COinS