Whooping Crane Migrational Habitat Use Draft ‎(2)‎

• File Number: 8461.100
• Description: Adaptive Management Workgroup
• State: CO
• Basin: South Platte
• Water Division: 1
• Date: 8/6/2008
• Author: Shay Howlin, Clayton Derby, Dale Strickland, Western Ecosystems Technology, Inc.
• Title: Whooping Crane Migrational Habitat Use Draft
• Water Supply Pro - Doc Type: Report/Study

• One model was fit to evaluate the association of habitat use with characteristics measured
on the ground.
• Two models were fit to evaluate the association of habitat use with flow dependent
characteristics:
o Habitat selection in the entire study area and
o Habitat selection in the local area.
Study area models evaluated selection as compared to availability in the entire study area. Local
area models evaluated selection as compared to availability in the local area.
For each type of habitat selection model, we created a model for the systematic sample of
observations and a model for the systematic and opportunistic observations combined. For each
dataset, we conducted the model selection routine to choose a final model. We compared the
final models for each dataset by assessing the similarity of the variables in the model, and the
coefficient value, direction, and interpretation for duplicate variables. Models that resulted in the
same interpretation based on similar values for the same parameters were determined to be not
biologically different.
Models using the systematic sample of observations were weighted by the probability of
detection. The final predictive model for detectability was used to estimate the probability of
detection for each observation. The overdispersion parameter was divided by the probability of
detection for each observation, which has the effect of multiplying the contributions of the log-
likelihood function for each observation (SAS Institute).
A forward selection routine was used to develop the models. The AIC was used to determine the
entry of a variable at each step of the selection procedure. For example, each variable in the
candidate variable list was fit in a model. The model with the lowest AIC score was selected for
the first step. The variable added to this model was then removed from the candidate variable
list, along with any variables with a high correlation (greater than 0.75 or less than -0.75) with
this variable. Next, each variable in the candidate variable list was fit in a model with the
variable selected during the previous step. The model with the lowest AIC score from this set of
models was selected for the next step. The procedure was continued until there were no variables
in the candidate variable list for which their addition into the model produced a model with a
lower AIC. Since the latter steps in this model selection procedure generally continue to add
variables to the model, though no significant reduction of the AIC is occurring, a graph of AIC
through the model selection procedure was used to select a model at the point where additional
variables entering the model do not contribute substantially.
For models with quadratic relationships in the explanatory variables in which the parabolic
maximum was biologically interesting, we calculated bias-corrected percentile confidence limits
on the maximum (Manly 1997). Datasets were resampled 500 times, models were refit, and the
maximum of the quadratic curve was calculated for each replication of the bootstrap sample. The
percentiles of the bootstrap distribution were used for confidence limits.
Study area selection of land cover
Study area selection of land cover was modeled as a discrete choice in space (Manly et al. 2002).
In this model, a used location was considered a choice, and each location was considered to have
7