Mark Hebblewhite, Evelyn Merrill, and Greg McDermid. 2008. A multi-scale test of the forage maturation hypothesis in a partially migratory ungulate population. Ecological Monographs 78:141–166.

Appendix A. Modeling peak of growing-season availability of forage-biomass components for elk.

Herein are provided a brief description of the methods used to model forage biomass, as well as in tabular form the top statistical models predicting forb, graminoid, and shrub biomass at the peak of the growing season, as well as average biomass availability in the main landcover types found in the study area.

Forage biomass (g/m²) at the peak of the growing season was statistically modeled within a landcover map derived from a supervised classification of LANDSAT Thematic Mapper (TM) and topographic derivatives from a digital elevation model (DEM) at a spatial resolution of 30 m² (Franklin et al. 2001). Landcover types included: closed conifer, moderate conifer, open conifer, shrublands, upland herbaceous, mixed forest, deciduous, water, and rock/ice (see McDermid et al. [2006] for details). This landcover classification was expanded to include three burned vegetation types (forest, grassland, and shrub) within 4 age classes (0–1, 2–4, 5–15, >14 years) for 12 burn types. Alpine meadows and shrublands were delineated using an elevation cut of 2200 m (Holland and Coen 1983) because of their importance to elk (Morgantini and Hudson 1989). Cutblocks and salvage-logged burns were important in the eastern part of the study area, and a cutblock cover type and two age classes of salvage-logged burns were therefore included (Hebblewhite et al. 2007). Thus, we modeled biomass within 26 landcover types.

Peak forage biomass was sampled randomly from 983 sites following a proportional allocation design (Krebs 1989, Thompson 1992) within strata designated by landcover type, the four fire age classes, two slope (> or <15 degrees), and three aspect classes (as defined in main article). Sampling occurred between 2 July and 28 August of 2001–2004, with an average sampling date of 29 July (Julian day (JD) = 211, SD = 12.8); there was no effect of sampling date on biomass (Hebblewhite 2006). At each site, plant cover was estimated within three 0.1-m² (2001) or five 0.25-m² quadrats (2002–2004) systematically placed along a 30-m transect and clipped total (green + standing dead) herbaceous biomass in three quadrats, which were averaged for one biomass estimate/site. Wet mass of forage biomass was weighed in the field; wet mass were converted to dry mass based on conversion ratios for each plant class when oven drying was not possible (Hebblewhite 2006). Total shrub biomass was estimated during 2002–2004 using the same approach as described in Methods: forgae maturation and biomass-quality relationships: forage growth section based on basal diameter–biomass relationships (Visscher et al. 2006). Shrub biomass was converted to biomass of only forage species and leaf-forage biomass (g/m²) using mean conversions for both within each landcover types (Hebblewhite 2006).

We developed predictive statistical models for forb, graminoid, and total shrub biomass (g/m²) at the peak of the growing season as a function of spatial covariates (sensu Frair et al. 2005). Spatial covariates measured at a 30-m² resolution included: landcover class, year, three aspect classes, hillshade (indexing xeric sites with high solar incidence), a soil-drainage index (indexing the area draining into a pixel) (Gessler et al. 1995), slope (in degrees), elevation (m), a greenness vegetation index derived from an August 1998 LANDSAT image (Stevens 2002), and distance to the continental divide in km (see Hebblewhite 2006). Covariates were screened for collinearity using a criteria of r < 0.5 and variance inflation scores <1 (McCullough and Nelder 1989). Forb and graminoid biomass were modeled using generalized linear models (GLM) with the log-link, and shrub biomass was modeled using the identity link (McCullough and Nelder 1989). Backward-stepwise model selection (McCullough and Nelder 1989) was used instead of an information-theoretic approach because of the difficulty in selecting meaningful a priori models and because prediction was the goal (Stephens et al. 2005). For forb and graminoid models, 20% of sites were randomly withheld for out-of-sample cross-validation by comparing observed to predicted biomass using Pearsons r. Because of reduced sample size for shrubs, model validation was only conducted within sample, and predicted and observed shrub biomass was compared using Pearson’s r.

TABLE A1. Top forage biomass component statistical models predicting forb, graminoid, and total shrub biomass at the peak of the growing seasons, 2001–2004, eastern slopes of BNP, Alberta.

Notes: Bolded coefficients are significant at P = 0.05. Blanks cells did not significantly differ from the reference category, which was closed conifer for all 3 models with the exception of burned habitats. Burned habitats were dummy coded for GLM models such that the statistical comparison was with the unburned reference habitat of that burn type. For example, burned forest 5–14 years old for shrubs was not different than closed conifer.  See text for details.

† Forb biomass was ln-transformed.

‡ Graminoid biomass was ln-transformed.

†† Total shrub biomass (leaf and twig) was untransformed.

TABLE A2. Mean total herbaceous and forage spp. shrub (leaf and twig) biomass at the peak of the growing season (Aug 4) for the 14-landcover types used in the study, from 2001–2004, with total number of plots sampled and standard deviation.

Notes: * indicates open habitat used in forage modeling.

LITERATURE CITED

Frair, J. L., E. H. Merrill, H. L. Beyer, J. M. Morales, D. R. Visscher, and D. Fortin. 2005. Scales of movement by elk (Cervus elaphus) in response to heterogeneity in forage resources and predation risk. Landscape Ecology 20:273–287.

Franklin, S. E., G. B. Stenhouse, M. J. Hansen, C. C. Popplewell, J. A. Dechka, and D. R. Peddle. 2001. An Integrated Decision Tree Approach (IDTA) to Mapping Landcover Using Satellite Remote Sensing in Support of Grizzly Bear Habitat Analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing 27:579–592.

Gessler, P. E., I. D. Moore, N. J. McKenzie, and P. J. Ryan. 1995. Soil-landscape modeling and spatial prediction of soil attributes. International Journal of GIS 9:421–432.

Hebblewhite, M. 2006. Linking predation risk and forage to ungulate population dynamics. University of Alberta, Edmonton, Alberta, Canada.

Hebblewhite, M., R. M. Munro, and E. H. Merrill. 2007. Effects of post-fire logging on elk habitat during the first 3-years post burn. Conservation Biology, In Review.

Holland, W. D., and G. M. Coen. 1983. Ecological (biophysical) land classification of Banff and Jasper National Parks. Alberta Institute of Pedology, Edmonton, Alberta, Canada.

Krebs, C. J. 1989. Ecological Methodology. Harper Collins, New York, New York, USA.

Morgantini, L. E., and R. J. Hudson. 1989. Nutritional significance of wapiti (Cervus elaphus) migrations to alpine ranges in western Alberta, Canada. Arctic and Alpine Research 21:288–295.

Thompson, S. K. 1992. Sampling. John Wiley and Sons, New York, New York, USA.

Visscher, D. R., E. H. Merrill, D. Fortin, and J. L. Frair. 2006. Estimating woody browse availability for ungulates at increasing snow depths. Forest ecology and mangement. 222:348–354.