Transcript Breeding and Non-breeding Survival of Lesser Prairie

MODELING VERTEBRATE
USE OF
TERRESTRIAL RESOURCES
Lyman L. McDonald, Wallace P. Erickson,
Mark S. Boyce, J. Richard Alldredge
Introduction
It is essential that wildlife management studies identify habitat
selection (i.e., vegetation types and foods used) by animals in
comparison to those resources in a study area.
The availability and use of the environmental components that are
necessary for life impact abundance of animals and distribution of
their populations in space and time.
Biologist must collect site- and time-specific information on
patterns of vegetation types and food use.
But how is such information obtained? What should be
considered when a study is designed to identify vegetation types
or food use? This chapter provides an outline of the major
techniques used to study these issues and some problems likely to
be encountered.
Resource Selection Definitions
►
►
►
►
►
►
Use: indicates association or consumption
Selective: if components are exploited disproportional to their availability
Resource Availability: the quantity accessible to an animal
Abundance: the quantity of the resource in a study area
Preference: as selection independent of availability
Enclosure Experiments: provide habitat categories in equal abundance
Cafeteria Experiments: captive animals are presented a variety of foods
and allowed to choose among them
►
Study Area: When choosing areas, one must consider the distribution of
resource units, scale of selection studied, what is truly available to the
animals, and manpower and budget constraints for collection of data.
►
Resource Units: indicate either habitat units, points in the habitat, or
food items.
Modeling Occupancy
►
A primary application of the methods presented is in
monitoring populations by estimation and mapping of the
relative probability that units (grid cells, pixels, etc.) are
occupied by a species “as measured by the sampling design.”
►
Study protocols requiring multiple independent surveys of units
have been developed to obtain “patch occupancy models”,
giving “clean” estimates of the probability of occupancy.
►
One should consider use of more expensive and timeconsuming survey methods when faced with monitoring
species distribution where probability of detection varies
significantly among vegetation types.
Levels of Selection and Effects of
Scale
►
Habitat selection can occur at a variety of levels or scales with animals
selecting habitats according to a hierarchical scheme. These scales include
the biogeographic (e.g., the eastern deciduous forest), home range, activity
points, such as a den, nest, or roost site within a home range, and selection
of particular foods at an activity point. Factors that influence selection at
each of these scales also vary. For example, climatic extremes may affect
the geographic range of a species, whereas vegetation structure may
influence home range size and shape, and competition with conspecifics or
predation risk may influence territory placement within a home range. The
distribution of food and cover is probably most influential in affecting local
movements within a home range.
►
The choice of an appropriate spatial and temporal scale of measurement,
and consideration of spatial pattern will directly influence results and their
interpretation
Management Implications of
Resource Selection
►
The limits of data should be recognized when applications of
research conclusions to manipulations of habitats or populations are
considered. Analyses can be helpful in identifying patterns of
habitat or food selection. However, biologists should not
necessarily conclude biological need from such patterns. Although
selection may have been demonstrated, one has not shown how
fitness (e.g., survival or reproductive success) of an animal varies
with different amounts of the selected habitat. We cannot make a
"biological leap of faith'' One cannot assume that if one increases the
amount of the selected habitat (or food), one will have more
animals.
► Therefore, if the objective is to evaluate the biological importance of
a particular habitat, one should consider some type of manipulation
experiment in which the amounts of the selected habitat (or food)
are varied and fitness is monitored.
Sampling Protocols and Study
Designs
►
The researcher must identify the scale of selection to study
consisting of resolution (grain) and extent (size).
►
The biology of the animal is important (e.g., if the animal being
studied is territorial then selection is commonly studied on a
different scale from that used for a non-territorial animal).
►
As a general rule, resource selection studies should consider
selection at more than 1 scale.
►
Resource selection may be detected and measured by
comparing any 2 of the 3 possible sets of resource units (used
units, unused units, or units in the study area).
Sampling Protocols for Resource Units
►
Sampling protocol (SP)
What is sampled?
________________________________________________________________________________
►
SP-A
Study area units are randomly sampled and
used units are randomly sampled.
►
SP-B
Study area units are randomly sampled and
unused units are randomly sampled.
►
SP-C
Used units are randomly sampled and
unused units are randomly sampled.
►
SP-D
Available units are randomly sampled and
classified as used or unused.
________________________________________________________________
Design I
►
►
►
►
Units used by the population of animals are recorded (but, use
by individual animals is not possible to record).
For example, in this design aerial or ground surveys are used
to locate animals.
Variables that potentially influence selection of units by
animals are measured at the locations. For example,
vegetation or forage types, food availability, slope, aspect, and
density of roads in a plot centered at the locations might be
measured at each location and used in a model to predict the
relative probability of selection of locations by animals in the
population.
Maps, aerial photographs, or GIS might be used to provide
sample or census data on study area plots or pixels.
Design II
In some cases, the study area is defined for a population of animals, but individual
animals are identifiable and habitat units selected can be recorded for unique animals.
Four examples of this design are provided:
1. A random sample of uniquely identified animals is obtained from the population so
that a sample of habitat units selected by a given animal can be recorded. Also, a
sample of study area units is selected. Predictor variables are measured on units
selected by the ith animal, and on the sample of study area units. Predictor variables
might be measured in the field or from aerial photographs, GIS, or maps.
2. Animals are trapped, radio-marked, and their home ranges measured with the
assumption that animals captured provided a random sample. A Design II study
would involve comparison of proportions of resource types and other variables in
home ranges to the same variables measured on similar sized regions randomly
sampled from the entire study area.
3. Habitat within each animal’s home range is compared with the habitat in the entire
study area.
4. Compared food selected by individuals of one species with individuals of a similar
species with random samples of food from the entire study area.
Design III
In this design, individuals are uniquely identified (usually by radio transmitters)
or collected for stomach samples. Data from each animal are analyzed to
provide a Resource Selection Function for each animal. Two examples of this
design are provided:
1. The animals in a sample are radio-marked, and the relocations of an animal
provide a sample of resource units selected by that animal. Resource units
within an animal’s home range also are sampled. Predictor variables are
measured on each sampled unit to contrast used units with units within each
home range.
2. Individual animals might be collected and stomach analysis performed on
each. Predictor variables are measured on prey or food items (e.g., species,
color, size). These data are then contrasted to measurements from a sample of
prey or food items collected in a certain size buffer surrounding the collection
site.
Comparison of Designs
►
Design I has been the most commonly used in the past;
however, it has the least specific information. Inferences can be
made to resource selection by the population of animals with
the assumption that study design and sampling protocol
adequately samples habitat units selected by the population and
in the study area.
►
Designs II and III tend to be preferred because data are obtained
on individual animals and their habitat or food selection. Thus,
variation in habitat selection among gender or age classes can
be analyzed. However, cost of a resource selection study
usually increases when individual animals are captured, marked,
and tracked.
Assumptions for Designs I-III
► Design
I: that study design and sampling protocol
adequately samples habitat units selected by the
population and in the study area and there is an
independence of locations of used units.
► Designs
II and III: that sample of animals was
collected by a random procedure, not on the fact that
data on individual animals might have lacked
independence.
Models for Resource Selection
Functions (RSFs)
►
Resource units usually are defined as
individual items of food (with food
selection) or blocks of land or points
on the landscape (with habitat
selection). Each resource unit is
characterized by the values that it
possesses for certain predictor
variables (also called independent
variables or covariates) X1, X2, ..., Xp,
representing characteristics such as,
size and color of food items, or the
distance from water and the habitat
type of habitat units. Three
mathematical functions (i.e., models
or curves) are involved in studies of
resource selection.
A RSF for the relative probability of use of
resource units with a single variable X (adapted
from McDonald and Manly 2001). The “available”
curve is approximately a normal distribution with a
mean of 20 and variance of 2.5. The “used” curve
is approximately a normal distribution with a mean
of 22 and variance of 1.9. These 2 distributions
define the RSF.
Assumptions for Estimation of RSFs
There are 6 major assumptions for estimating RSFs.
1. The researcher is interested in ranking habitat units (food units) in a study area based
on the relative probabilities of selection by animals.
2. Predictor variables to be measured on sampled units are correlated with the
probability of selection and do not change appreciably during the study period.
3. Measurement errors for predictor variables, X1, X2, ..., Xp, are relatively small in
comparison to variation from unit to unit.
4. In Design I and II studies, animals in the population have equal access to all units in
the study area. If this is not the case, e.g., if animals are territorial, Design III should be
used.
5. Study area units are randomly sampled.
6. Selected units are randomly sampled or the probability of detection of selected units
is approximately constant. If the probability of detection of use of a sampled unit is
highly dependent on the vegetation type or other predictor variables then other more
complex study designs are required .
Defining Study Areas and
Measuring Habitat Selection
Selection and use of a particular area or unit of habitat by an animal are the result of
proximate and/or ultimate predictor (independent) variables.
Proximate Variables: those features used as cues when an animal evaluates a site
(habitat unit). They may include structural features such as understory cover, canopy
height, slope, density of roads in the unit or distance to water. The presence/absence of
other animals that may act as competitors or predators also may influence habitat
selection. Animals may use such features as cues, but they may not be the same as the
variables that have resulted in evolutionary associations between animals and habitat.
Ultimate Variables: those parameters that affect an individual's abilities to reproduce,
obtain food, and avoid predators are examples of ultimate variables that influence
habitat selection.
Studies of habitat selection usually involve measure of proximate variables and food
availability. However, with adequate data on ultimate variables, for example,
measurements of predator abundance and competition, a more complete understanding
of habitat selection can be obtained.
Techniques for
Detection of Habitat
Selection
Direct and indirect methods have been
used to detect wildlife habitat
selection.
Direct methods include observation,
capture, and radio-telemetry whereas
indirect methods are dependent on
some evidence of animal activity
within an area or specific site (e.g.,
bed sites, browsed twigs, feces, nests,
or tracks).
These measures may be used to
detect use of units along systematic
transects, in a small-mammal
trapping grid, or with other sampling
designs appropriate to the animal of
interest.
Representative methods
to examine habitat use
patterns. “A” illustrates
Design II and sampling
protocol A: available
habitat is inventoried
and compared to the
composition of an
animal’s home range.
“B” illustrates Design I
and sampling protocol
A: random samples are
compared to
characteristics of a
sample of sites where
use has been detected,
such as nests or roost
sites. “C” illustrates
Design I and sampling
protocol C: systematic
plots (or points) are
established and features
are compared between
sites where use was
detected (via captures,
tracks, feces, radio
relocations, etc.) and
sites where use did not
occur.
Direct Methods for Detection of
Habitat Selection
Direct observations of animals may allow economical sampling of a large segment
of the study area and to distinguish activities within vegetation types. Collection of
data can be combined with aerial or other survey procedures. Problems to consider
are differential visibility among vegetation types and the difficulty of recording
observations during nocturnal periods.
Advantages of animal capture include being able to examine individuals for
age/gender and other characteristics. Capture procedures can be combined with
mark-recapture statistics to estimate abundance. However, differential vulnerability
to capture due to age and gender or other factors may bias results and attractants
may cause animals to select vegetation types that are normally not selected.
Radio-telemetry also can be used to measure habitat selection. Advantages include
being able to examine individuals of known age/gender and other characteristics for
habitat selection. Animals can be located multiple times throughout the day/night
and seasons. Habitat selection for important components (e.g., den site selection or
roost sites) can be studied.
Indirect Methods for Indication of
Habitat Selection
Detection of Tracks:
allows one to sample units economically and detect use
in a large sample of units in a short time by all segments of the populations.
However, the procedure often suffers from lack of good tracking conditions (e.g.,
uniform snow) and different visibility of tracks in different vegetation types.
Detection of Pellet Groups or Scat: measures selection by all segments
of a population. With cleared plots, information on seasonal selection is
obtained, and potentially can be combined with deposition rates to estimate density
of animals. However, defecation rates often are unknown or vary with habitat type
and activity, and decomposition rates may vary among habitats. Visibility and
detection of pellet groups also may vary with vegetation type.
Browsing or Feeding:
may provide evidence of use of habitat units by all
segments of a population. This technique may provide additional information on
food habitats and be combined with studies of carrying capacity. Potential biases
include competition for the same food by other species and food species must be
present before use of a unit can be documented.
Techniques to Define Study Areas
Issues with the definition of the study area are simplified if the scale of selection
is clearly delineated.
First-order Selection: the selection of physical or geographical range of a
species. Few if any habitat selection studies are of first-order selection.
Second-order Selection: results in the home range of an individual or social
group within the physical or geographical range of a species. Second-order
selection is of interest in many habitat-selection studies and will typically require
radio-marking of individuals or social groups (Design II and III studies).
Third-order Selection: selection of sites within the home range. Typically,
home range of an individual or social group is calculated by the minimum convex
polygon method and all units within the home range define the study area.
Fourth-order Selection: the actual procurement of food items from those
available at a feeding site as identified by third-order selection.
Guidelines
The guidelines presented below may be helpful when study area
boundaries are delineated for second-order selection or Design I studies.
Size of the study area should be substantially larger than the home range
of the study species.
Numbers of study animals, groups, or social units present on the study
area should be, as far as possible, adequate for study.
An opportunity should exist for independent locations of animals or
independent location of home ranges within the study area (i.e., as close
as possible to an unbiased random sample of sites selected by animals or
a random sample of home ranges).
Study area boundaries should be chosen with consideration of the
biology of the animal. Physical barriers such as rivers or mountain
ranges might make better boundaries than an arbitrary (geopolitical)
straight line on a map.
Techniques to Measure Food Availability
in Resource Selection Studies
Abundance and distribution of food resources are among the major
environmental features that influence habitat selection. Because food
intake relates to energy needs, reproduction and ultimately to survival,
understanding food selection is a fundamental component of behavioral
ecology.
Wildlife food abundance can be estimated for an area by measuring the
annual production of herbaceous plants, woody stems, fruits, and seeds,
or by assessing the abundance of potential prey.
Availability suggests a food resource is both accessible and usable.
Access to food resources can vary with weather or by the presence of
predators or competitors. In resource selection studies, effects that
modify food abundance might be modeled using covariates such as
snow depth or presence of a predator or competitor.
Grasses and Forbs
►
Clipping and Weighing: dried samples of above ground vegetation is
the most accurate, but most time-consuming, technique for measuring
predictor variables for availability of herbaceous plants. Many techniques
have been developed to more rapidly estimate vegetative biomass and to
avoid destructive sampling. These include:
►
Obstruction of Vision: as measured by Robel range pole methods
Estimating Biomass: in small quadrats.
►
Estimating Percent Cover: by species in small sample plots.
►
Pin Intercept Methods: have been developed using sampling frames
containing rows of pins that are pushed through the vegetation.
►
Notched Boot: while walking through an area, at regular intervals, the
species of plant nearest a notch in the tip of the boot is recorded.
Browse
►
Predictive Equations: have been developed that relate measures of shrub
size to forage production and hence, estimation of the amount of browse in a
study unit as predictor variables. Specific equations must be estimated for
each species and for individual study sites.
►
The Twig-count Method: estimates biomass of browse by calculating
the average weight of edible material in a single twig and multiplying that
value times the number of twigs. A sample of previously browsed twigs is
used to estimate the average browsing diameter for each forage species.
Mass of browsed twigs is then estimated from a collection of twigs clipped to
the size of the average browsed twig.
►
Densities of Twigs: can be estimated from counts on circular plots or
belt transects, and browse biomass is calculated per unit area. Modifications
of this technique include development of equations that use unbrowsed twig
length or basal diameter to estimate twig mass.
Fruits and Seeds
► Fruits
and Seeds: from low-growing herbs and
shrubs can be counted and averaged per plant and
summed over an area to estimate biomass in study
units.
► Hard
Mast Crops: such as acorns, can be collected
in funnel traps that sample an area under the canopy.
Although traps usually prevent animals from taking
mast once it has fallen from the tree, information on
production can be biased if seeds are consumed before
falling to the ground.
Design Considerations and Analysis
General Considerations
Frequently, investigators have focused on
modeling selection of habitat units during
specific periods of time of day or behavior:
feeding, resting, or rearing young. The most
common study design (SP-A) would involve
collection of a sample of selected units, such as
locations of radio-marked animals, and contrast
those units with a sample of units in the study
area. In this case, the models yield estimates of
the relative probability of selection (i.e.,
information to the effect that one unit might be
selected with twice or 3 times the probability
that another unit is selected).
A valid probability sampling procedure is used
to sub-sample a variety of features at each study
site, cell, or home. The features selected to
describe a sampled unit (e.g., litter depth,
understory stem density, canopy closure,
distance from roads, aspect, slope) are assumed
to represent or be highly correlated with
the variables used by animals to evaluate a
site and often include some measurement
of food abundance, cover, and structural
characteristics
An example of nested plots used to sample ground
litter, understory stem density, and overstory
composition (modified from Dueser and Shugart
1978).
Measurement of Landscape
Variables Using a GIS
► A Landscape: is a mosaic of habitat patches in which a patch of
interest is embedded.
► Landscape Variables: include patch size, patch context, and
other habitat characteristics (e.g., density of roads, proportion of
habitat types, or density of edge between habitat types in a buffer
(circle) centered at the site.
►
Because the actual habitat features influencing selection are not
known, measuring several features is appropriate. However, as the
number of features measured becomes large, the chance of detecting
spurious relationships also increases. Therefore, the list of features
to be sampled should be limited to those based on biological
considerations for the relationships between animals and their
habitats.
Standard Statistical Analyses
►
Analytical methods for resource selection studies usually involve
comparison of characteristics of samples or censuses of used units and
samples of units from the study area. The first step should involve
graphical and descriptive comparisons of the distribution of the predictor
variables (also called covariates) that describe each unit for the samples
being compared (e.g., used vs. study area). Patterns described in these
analyses probably will be apparent in any inferential analyses (e.g.,
hypothesis testing). Inferential analyses include:
►
Chi-square Analyses for Categorical Data
►
Modeling Resource Selection
►
Selection Ratios for Categorical Data
►
Resource Selection Function for Categorical Data
►
Estimation of a RSF for Categorical Data Using Logistic Regression
Chi-square Analyses for Categorical Data
►
For data based on resource categories (e.g., vegetation types, food types, or
categorized continuous predictors), a Chi-square test can provide an omnibus answer
to the question: is there evidence of selection or not? This test is appropriate when
individual observations of selected units are considered independent. Chi-square
analyses appear most appropriate in the Design I case and are not appropriate when
several animals have multiple (dependent) relocations.
►
The form of test statistic that is most commonly used for this purpose is the Pearson
statistic, which takes the form
►
where Oi is an observed sample frequency, Ei is the expected value of Oi according
to the hypothesis being considered, and the summation is over all resource
categories.
►
►
Eighty percent of all cells should have expected values of 5 or more, otherwise, the
standard Chi-square distribution may not be an accurate approximation to the
sampling distribution of the statistic.
Modeling Resource Selection
Selection Ratios for Categorical Data
►
The selection ratio (SR) for
a given resource category is
the ratio of the proportion
used to the proportion of the
category in the study area.
If the ratio is close to 1,
there is evidently no
selection. Values smaller
than 1 indicate selection
against that category; large
values indicate selection for
the category.
►
SR = % of locations
% of Study area
Selection ratios, and
standardized selection ratios
are calculated with 95%
Bonferroni adjusted confidence
limits (CL). If CL do not
overlap, then values are
considered different.
Resource Selection Function for Categorical
Data Using Logistic Regression
►
Logistic regression is a specialized regression tool for working
with multiple continuous and discrete variables. However,
logistic regression also can be used to analyze the effect of a
single categorical response variable on resource selection (e.g.,
the effects of the 4 habitat categories on selection of locations
by an animal.
►
The technique is presented as a means to analyze selection
among categories of a categorical variable and as an
introduction to estimation of a RSF when there are multiple
variables.
Multiple Continuous and Discrete
Variables
►
The estimation of RSFs is presented as a unified theory for study of
the relationships between selection of units and multiple predictor
variables measured on those units. These functions allow one to
estimate the relative probabilities that habitat units were selected and
to rank units according to their value for use by wildlife for the
study area, population, and time period.
►
Modeling of resource selection is introduced for computation of
selection ratios among categories of vegetation types or food types.
Logistic regression is the basic tool by which all example sets of
data in this chapter are analyzed; it is used to analyze data to
illustrate the concepts of relative probabilities of selection among
vegetation types. Information theory and maximum likelihood
procedures are used for selection of multiple predictor variables for
the RSFs.
SUMMARY
►
►
►
Habitat use and food selection by wildlife in comparison to available habitat and
food is used to identify important habitat patches and food resources. Selection
is defined if resources are used disproportional to their availability at a given
scale: geographical range of a species, home range of an individual or social
group, selection of sites within a home range, or procurement of food items at a
feeding site. Generally, resource selection studies should consider selection at
more than 1 scale and should be replicated in time and space.
Study area and selected units should be randomly sampled. Resource units are
defined as individual items of food (with food selection) or blocks of land or
points on the landscape (with habitat selection). Models for Resource Selection
Functions were developed as a unified theory for study of the relationships
between selection of units and predictor variables measured on those units.
These functions allow one to estimate the relative probabilities that habitat units
were selected and to rank units according to their value for use by wildlife for
the study area, population, and time period.
Logistic regression is the basic tool by which data are analyzed; it is used to
analyze data to illustrate the concepts of relative probabilities of selection among
vegetation types.