Abstract
Regression equations were developed to predict 3 mass components of 7 browse species important to moose (Alces gigas) on the Copper River Delta in southcentral Alaska. The accuracy of model predictions was the criterion for model selection. Model accuracy was evaluated using data splitting or jackknife procedures. Annual production of twigs and leaves and available twig mass on a stem were most accurately predicted from stem basal diameter with zero intercept models, zero intercept log-linear models, or log-log models. Twig mass eaten by moose was most accurately predicted from the diameter at the point of browsing of a twig with zero intercept or full linear models. Heteroskedasticity was significant (P < 0.05) in most of the data sets and could not be significantly reduced with log transformations or use of weighted least squares models. Heteroskedasticity appeared to have a relatively minor effect on model predictions. Most of the models gave mean predictions within +/- 20% of the actual values, particularly for the most ubiquitous species that were also the most important to moose. For each species, there were few differences (P < 0.05) in model coefficients between years and among habitat types. Differences in coefficient estimates appeared to be related to differences in stem morphology that were related to both site conditions and past browsing by moose.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have