7.4 Rationale for Representative Sampling
7.4 Rationale for
Representative Sampling
The primary goal of data collection for IREAs is
to characterize as many ecological patterns of interest as possible. The constraints of most IREAs,
which de facto limit the number of samples collected, require several possibly contradictory sampling design attributes: efficiency in the recovery
of ecological patterns and processes, full representation of patterns and processes, geographic replication, logistical benefits, and cost effectiveness.
Commonly used random sampling procedures do
not necessarily have these attributes. The goal of
the data collection phase of an IREA is acquiring
data that are representative of the environments to
be characterized under severely limited sample
sizes. Sampling theory emphasizes randomization
in order to provide the probability structure for statistical analysis or to give credibility to the statistical model used (Gillison and Brewer, 1985; see
also Chapter 6). Simple random sampling does not
capture the full range of variability ofthe biotic and
abiotic components of ecosystems at regional
scales unless the sampling intensity is very high
(Pielou, 1974; Orl6ci, 1978; Gauch, 1982).
The data collection process for IREAs should
consider three aspects of pattern recognition: (1)
the delineation of the pattern itself (e.g., a specific
forested ecosystem); (2) the frequency and distribution of patches of the pattern (i.e., spatial distribution, number, and size of stands in a forested
ecosystem) (Godron and Forman, 1983; Gillison
and Brewer, 1985); and (3) the detection of boundaries or rates of change between samples (Fortin,
1994, 1999). In landscapes, patch frequency and
distribution vary as a scale-sensitive function of environmental complexity and of the level of resolution of the ecological classifications used to characterize the pattern (Gillison and Brewer, 1985;
Bissonette, 1997; Mladenoff et aI., 1997; Hargis et
al., 1998; Bourgeron et ai. 1999). Landscape configuration variability should be analyzed in terms
of the driving variables (the abiotic factors) controlling the biotic components of the ecosystem
(Bourgeron et aI., 1994b).
To obtain data representative of all patterns and
processes, IREAs have made use of a series of stratified sampling schemes (SSs) (see Chapter 6).
These schemes have been employed successfully
to provide both accuracy and efficiency in the recovery of patterns and statistical validity over large
heterogeneous areas with mostly unknown patterns. SSs used in IREAs fall into three broad cat97
egories: stratified random sampling, stratified
semirandom sampling, and multistage stratified
random and semirandom sampling. They have been
used in terrestrial and aquatic ecosystems and to
sample plant and animal species and communities.
7.4.1 Stratified Random Sampling
Stratified random sampling (SRS) (see Chapter 6)
is used to stratify a region based on relevant data,
such as abiotic variables (e.g., slope, aspect) or
ecosystem units. Random sampling is then conducted within each stratum. The specific objectives
of a project determine whether the number of samples in a stratum is proportional to its occurrence
or is a fixed size. SRS has been used for a variety
of biological assessments (e.g., Davis et aI., 1990;
Aspinall and Veitch, 1993; Walker et aI., 1993).
Stohlgren et ai. (1997) used SRS as part of a
methodology for rapid assessment of plant diversity patterns at landscape scales. The strata were
six vegetation types of interest; the same number
of samples was allocated to each type. Ecological
classifications (see Chapter 22) have often been
used as the basis for environmental stratification
for sampling and monitoring. In Great Britain, the
Institute of Terrestrial Ecology (ITE) land cover
classification is used as the basis for stratification
for national-scale surveys of land use and cover, as
well as for monitoring and evaluation of land-use
policies (e.g., Bunce et al., 1996b).
7.4.2 Stratified Semirandom Sampling
Stratified semirandom sampling (SSRS) is a variant of SRS. Stratification is conducted as for SRS,
but randomly selected potential sites may be eliminated based on additional criteria. For example,
Neave et ai. (1996) conducted representative sampling of bird assemblages in a 7600-km2 area of
southern Australia containing open Eucalyptus forest. Forested sites were stratified by combinations
of temperature, precipitation, and nutrient-supply
classes, resulting in identification of 24 environmental domains. Potential sites for sampling bird
species and their habitats were evaluated based on
their suitability defined by road accessibility.
Twenty-three of the domains were sampled with
165 I-ha plots, efficiently covering the range of environmental variability by sampling a very small
fraction of the study area.
A methodology known as gradient directed transect (gradsect) sampling is a variant of SSRS. This
approach, first described by Gillison and Brewer
(1985), is based on the distribution of patterns
7.4 Rationale for
Representative Sampling
The primary goal of data collection for IREAs is
to characterize as many ecological patterns of interest as possible. The constraints of most IREAs,
which de facto limit the number of samples collected, require several possibly contradictory sampling design attributes: efficiency in the recovery
of ecological patterns and processes, full representation of patterns and processes, geographic replication, logistical benefits, and cost effectiveness.
Commonly used random sampling procedures do
not necessarily have these attributes. The goal of
the data collection phase of an IREA is acquiring
data that are representative of the environments to
be characterized under severely limited sample
sizes. Sampling theory emphasizes randomization
in order to provide the probability structure for statistical analysis or to give credibility to the statistical model used (Gillison and Brewer, 1985; see
also Chapter 6). Simple random sampling does not
capture the full range of variability ofthe biotic and
abiotic components of ecosystems at regional
scales unless the sampling intensity is very high
(Pielou, 1974; Orl6ci, 1978; Gauch, 1982).
The data collection process for IREAs should
consider three aspects of pattern recognition: (1)
the delineation of the pattern itself (e.g., a specific
forested ecosystem); (2) the frequency and distribution of patches of the pattern (i.e., spatial distribution, number, and size of stands in a forested
ecosystem) (Godron and Forman, 1983; Gillison
and Brewer, 1985); and (3) the detection of boundaries or rates of change between samples (Fortin,
1994, 1999). In landscapes, patch frequency and
distribution vary as a scale-sensitive function of environmental complexity and of the level of resolution of the ecological classifications used to characterize the pattern (Gillison and Brewer, 1985;
Bissonette, 1997; Mladenoff et aI., 1997; Hargis et
al., 1998; Bourgeron et ai. 1999). Landscape configuration variability should be analyzed in terms
of the driving variables (the abiotic factors) controlling the biotic components of the ecosystem
(Bourgeron et aI., 1994b).
To obtain data representative of all patterns and
processes, IREAs have made use of a series of stratified sampling schemes (SSs) (see Chapter 6).
These schemes have been employed successfully
to provide both accuracy and efficiency in the recovery of patterns and statistical validity over large
heterogeneous areas with mostly unknown patterns. SSs used in IREAs fall into three broad cat97
egories: stratified random sampling, stratified
semirandom sampling, and multistage stratified
random and semirandom sampling. They have been
used in terrestrial and aquatic ecosystems and to
sample plant and animal species and communities.
7.4.1 Stratified Random Sampling
Stratified random sampling (SRS) (see Chapter 6)
is used to stratify a region based on relevant data,
such as abiotic variables (e.g., slope, aspect) or
ecosystem units. Random sampling is then conducted within each stratum. The specific objectives
of a project determine whether the number of samples in a stratum is proportional to its occurrence
or is a fixed size. SRS has been used for a variety
of biological assessments (e.g., Davis et aI., 1990;
Aspinall and Veitch, 1993; Walker et aI., 1993).
Stohlgren et ai. (1997) used SRS as part of a
methodology for rapid assessment of plant diversity patterns at landscape scales. The strata were
six vegetation types of interest; the same number
of samples was allocated to each type. Ecological
classifications (see Chapter 22) have often been
used as the basis for environmental stratification
for sampling and monitoring. In Great Britain, the
Institute of Terrestrial Ecology (ITE) land cover
classification is used as the basis for stratification
for national-scale surveys of land use and cover, as
well as for monitoring and evaluation of land-use
policies (e.g., Bunce et al., 1996b).
7.4.2 Stratified Semirandom Sampling
Stratified semirandom sampling (SSRS) is a variant of SRS. Stratification is conducted as for SRS,
but randomly selected potential sites may be eliminated based on additional criteria. For example,
Neave et ai. (1996) conducted representative sampling of bird assemblages in a 7600-km2 area of
southern Australia containing open Eucalyptus forest. Forested sites were stratified by combinations
of temperature, precipitation, and nutrient-supply
classes, resulting in identification of 24 environmental domains. Potential sites for sampling bird
species and their habitats were evaluated based on
their suitability defined by road accessibility.
Twenty-three of the domains were sampled with
165 I-ha plots, efficiently covering the range of environmental variability by sampling a very small
fraction of the study area.
A methodology known as gradient directed transect (gradsect) sampling is a variant of SSRS. This
approach, first described by Gillison and Brewer
(1985), is based on the distribution of patterns