Ecological Archives E096-149-A4
Joseph A. LaManna, Amy B. Hemenway, Vanna Boccadori, and Thomas E. Martin. 2015. Bird species turnover is related to changing predation risk along a vegetation gradient. Ecology 96:1670–1680. http://dx.doi.org/10.1890/14-1333.1
Appendix D. Detailed description of all methods used in this study.
Vegetation gradient.– To classify the gradient from deciduous to coniferous shrubs and trees, vegetation was surveyed by sampling 5 m radius plots nested within 11.3 m radius plots placed systematically along a 35 m × 35 m grid within each forest stand, following the BBIRD protocol (Martin et al. 1997). Forest stands were defined as the area of a minimum convex polygon encompassing aspen forest bounded on all sides by any other vegetation type (Tewksbury et al. 1998). All understory woody stems, including shrubs and young trees (>50 cm in height and <8 cm diameter at 10 cm height), were classified to species and one of two size classes (<2.5 cm, or 2.5–8 cm diameter at 10 cm height) in each 5 m radius plot. Larger trees (>8 cm diameter at breast height, or DBH) were classified to species and one of three size classes (8–23 cm DBH, 23–38 cm DBH, or >38 cm DBH) in the 11.3 m radius plots. Understory stem and tree basal areas were calculated using the mid-point DBH for each stem or tree size class.
Measurements from all systematic vegetation plots within each stand were used in these analyses (total across stands = 860 vegetation plots; mean = 45 vegetation plots/stand; range = 13–102 vegetation plots/stand). Transition from deciduous to coniferous vegetation characterizes this system (Romme et al. 1995), and aspen and Douglas fir were the dominant deciduous and conifer trees; accounting for 83.6% and 77.8% of total deciduous and conifer tree basal area respectively. Therefore, four vegetation variables (conifer tree basal area, understory conifer stem basal area, deciduous tree basal area, and understory deciduous stem basal area) were used in a PCA to determine the major axes of change in vegetation structure and composition. The first two principal component (PC) axes from this analysis (PCconifer and PCdeciduous) accounted for 68.6% of the variation among the 860 systematic vegetation plots and were used in all analyses below.
Local habitat use.–Nests of all bird species breeding at our study sites were searched for in all study forest stands. We searched for nests of all breeding bird species in this system, but 16 species were selected for this study because we had collected sufficient sample sizes of nests to estimate nest predation rates for these species only (Appendices B, C). To quantify variation in nest-site use, we recorded vegetation properties from plots centered at each nest (identical to the systematic vegetation plots above). We used these data to calculate scores for each nest along the two PC axes. Territory-scale habitat use was quantified by averaging PC scores across the nest-site vegetation plot and all (2–4) systematic vegetation plots within 50 m of the nest. This distance approximates the mean core radius of a nesting songbird territory at our sites. Thus, there were a total of three to five vegetation plots per territory with one plot centered at the nest for territory-scale analyses.
Bird and nest predator population densities.–Numbers of territories for each bird species were censused using a spot-map technique; each stand was visited every two days during the breeding season, allowing for careful assessment of all singing males and their territories (Svensson et al. 1970). Bird density (territories/ha) was calculated for each species, and linear models were fit to describe densities along PC axes. Year and site were random factors in all models to account for temporal and spatial autocorrelation. Models were selected based on a forward stepwise procedure, using likelihood ratio tests to test for a significant contribution of a given PC axis to the model compared to a null model. Log-link functions were used instead of identity-link functions if bird density among stands was not normally distributed.
Red squirrel (Tamiasciurus hudsonicus), chipmunk (Tamias spp.), gray jay (Perisoreus canadensis), Steller's jay (Cyanocitta stelleri), and common raven (Corvus corax) were potential nest predators in our study system. All nests were filmed using hi-8 camcorders once during the early incubation stage and at least once during the nestling stage for 6 hours beginning at sunrise to capture predation events and determine our suite of nest predators. We then measured nest predator density in each study stand using a standardized spot-mapping technique because territories of most predators can be readily detected by alarm and territorial calls as well as middens of squirrels (Tewksbury et al. 1998). We calculated combined nest predator density within each stand and fit linear models to describe differences in nest predator density along the PC axes as described above for bird species.
Nest predation.–Nest predation rates were determined by intensive nest monitoring following standard protocols (Martin and Geupel 1993). Every attempt was made to locate and monitor nests within all territories identified by spot-mapping. Nests were classified as either depredated or not based on absence of nest contents when too young to have fledged, egg shells in the nest, disturbed nest lining, or video recordings of predation events. These data were used to estimate daily nest predation rates with logistic exposure methods (Shaffer 2004). Differences in nest predation rates with vegetation were calculated using data from 1067 nests (Appendix C).
Nest predation was modeled as a function of the PC axes at three spatial scales (forest stand, territory, and nest site). Three spatial scales were tested because nest predation may be more strongly influenced by habitat characteristics at the nest or in the nesting territory than in the forest stand at large. For each PC axis, stand-scale variables were calculated by averaging across all systematic vegetation plots in a forest stand, and territory and nest-site variables were calculated as described above (see Local habitat use). All possible combinations of two PC axes with three spatial scales yielded six single-variable models describing differences in nest predation with vegetation for each species. Year and site were used as random factors in all nest predation models to account for temporal and spatial autocorrelation, and territory- and nest-site-scale models used forest stand as a random factor to account for sampling of territories and nests within stands. For each model, likelihood ratio tests were conducted against a null model of constant nest predation, and the best-fitting model was determined by these tests.
Nest predation and density.– We examined the relationship between density and nest predation both among and within bird species. Among species, standardized effect sizes (standardized regression coefficients; Schielzeth 2010) describing the relative strength of vegetation's effect on density and nest predation were calculated for each species. In most cases, a single PC axis explained variation in bird densities. However, if the density of a species was associated with both PC axes, then the axis that also explained nest predation was selected for this analysis. These standardized effect sizes were compared across species using a weighted least squares regression to assess if the relative strength of vegetation's effect on nest predation was inversely related to the relative strength of its effect on bird density. For this weighted regression, error weights were calculated as the combined inverse variance in each species' estimates for vegetation effects on density and nest predation (i.e., 1/(SEdensity × SEpredation)) to account for variable error in these estimates across species.
We next examined within-species relationships between density and nest predation to determine if bird abundances decreased as predation rates increased, suggesting preferences for low-risk habitat. Nest predation rates were calculated for each study stand and regressed against density for each bird species. Variables were square-root transformed and log-link functions were used if data were not normally distributed.
Experimental conifer removal.– We conducted a landscape-scale experimental removal of conifer trees and shrubs from aspen stands to test vegetation as the cause of observed effects on bird abundances and nest predation rates. All conifer trees and most understory conifer shrubs were mechanically removed with chainsaws and small tractor-fellers from within and 33 m around three forest stands at the Mount Haggin field site following the 2010 breeding season (see Appendix A for experimental stand details). Five other stands were selected a priori to serve as controls, allowing a before-after-control-impact (BACI) experimental design. Because this experimental treatment removed all coniferous vegetation, we expected it to also increase the abundance and density of smaller deciduous shrubs and trees that were early successional species (willows and aspens). Therefore, we expected our landscape-scale habitat experiment to manipulate both PC axes and monitored effects of this manipulation on these axes as well as densities and nest predation rates of birds and nest predator density.
To test for a significant change following conifer removal in treatment stands corrected for any changes in control stands, a model with an interaction term between time (i.e., before/after removal) and treatment was tested against a null model with no interaction term using a likelihood ratio test. Effects of the experiment on bird and nest predator densities (measured by partial η²) were compared to relative changes in densities from observational analyses using a weighted regression to assess whether the relative change in density among species observed along the gradient could predict responses to the experiment. For this weighted regression, error weights were calculated as the combined inverse variance in each species' estimates for the observed and experimental change in density (i.e., 1/(SEobserved change in density × SEexperiment change in density)) to account for variable error across species. All statistical analyses were conducted with R version 3.0.3 and the 'lmtest,' 'lme4,' 'lmerTest,' and 'repeated' packages (Lindsey 2001, Zeileis and Hothorn 2002, Bates et al. 2014, Kuznetsova et al. 2014, R Core Team 2014).
Literature Cited
Bates, D., M. Maechler, B. Bolker, and S. Walker. 2014. lme4: Linear mixed-effects models using Eigen and S4. R package version 1.1-5.
Kuznetsova, A., P. B. Brockhoff, and R. H. B. Christensen. 2014. lmerTest: Tests for random and fixed effects for linear mixed effect models (lmer objects of lme4 package). R package version 2.0-6.
Lindsey, J. 2001. repeated: Non-normal repeated measurements models. R package version 1.0.
Martin, T. E., and G. R. Geupel. 1993. Nest-Monitoring Plots: Methods for Locating Nests and Monitoring Success. Journal of Field Ornithology 64:507–519.
Martin, T. E., C. Paine, C. J. Conway, W. M. Hochachka, P. Allen, and W. Jenkins. 1997. BBIRD (Breeding biology research and monitoring database) field protocol. Montana Cooperative Wildlife Research Unit, University of Montana, Missoula, Montana, USA.
R Core Team. 2014. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.
Romme, W. H., M. G. Turner, L. L. Wallace, and J. S. Walker. 1995. Aspen, elk, and fire in northern Yellowstone Park. Ecology 76:2097–2106.
Schielzeth, H. 2010. Simple means to improve the interpretability of regression coefficients. Methods in Ecology and Evolution 1:103–113.
Shaffer, T. L. 2004. A unified approach to analyzing nest success. Auk 121:526–540.
Svensson, S., K. Williamson, C. Ferry, A. Joensen, D. Lea, H. Oelke, and C. Robbins. 1970. An international standard for a mapping method in bird census work recommended by the International Bird Census committee. Audubon Field Notes 24:722–726.
Tewksbury, J. J., S. J. Hejl, and T. E. Martin. 1998. Breeding productivity does not decline with increasing fragmentation in a western landscape. Ecology 79:2890–2903.
Zeileis, A., and T. Hothorn. 2002. Diagnostic checking in regression relationships. R News 2(3):7–10.