5.6 Conclusion
termining adequate sample sizes is an iterative
process, recalculating variance and temporal and
spatial replication needs as the study progresses.
Befriend statisticians and sampling design experts.
The biggest design flaw in many ecological assessment programs is biased site selection (Stohlgren
et al., 1998). Why are so many long-term study
plots in flat terrain and close to roads? Krebs (1989,
p. 202) defmes this as "accessibility sampling,"
where "the sampling unit is restricted to units that
are readily accessible," such as where samples of
forest stands are taken only along roads. He also
wrote that "judgmental sampling" occurs when "the
investigator selects on the basis of experience a series of 'typical' sampling units-a botanist may select 'climax' stands of grassland to measure." The
issue here is that these sampling approaches are rejected by statisticians because they cannot be evaluated by the theorems of probability (Krebs, 1989).
Hence they are of limited use for extrapolating assessment results beyond the plots themselves
(Stohlgren, 1994).
The long-term value of data sets also depends on
the intensity and spatial pattern of sampling (Fortin
et aI., 1989; Stohlgren, 1994). Data from sparsely
sampled sites may suggest no effect or trend where
one exists. There is a growing appreciation for the
need for adequate replication in long-term monitoring (Hinds, 1984; Likens, 1991; Stohlgren et al.,
1995b). However, Kareiva and Anderson (1988)
showed that (1) as plot size increased in ecological studies, the number of replicates decreased, and
(2) once plot size reached about 3 m 2 , the number
of replicates was consistently less than 5.
Sample size determination is tricky business.
Typically, appropriate sample sizes are determined
after evaluating between-plot variance from initial
field tests (Krebs, 1989, pp. 173-199). The appropriate sample size is then determined by corresponding the variance to a predetermined level of
accuracy. Since disturbances of some kind or another frequently occur on most landscapes, studies
designed to quantify disturbance effects must anticipate the need for large sample sizes. Interacting natural processes on an already heterogeneous landscape
are further complicated by a range of species-specific
and site-specific responses (Stohlgren et al., 1995b).
Sampling at multiple spatial scales is also becoming more commonplace because it is virtually impossible to inventory an entire landscape or region
(Whittaker et al., 1979; Shmida, 1984; Stohlgren et
al., 1998). Data acquisition for trend analysis must
consider temporal and spatial variability simultaneously.
75
5.5.2 Accuracy Assessments of
Mapped Data
It often proves difficult to assess the accuracy of
maps. Field data, which are needed to assess map
accuracy, are expensive (Kalkhan et aI., 1995).
Checking only large polygons on a map overestimates the accuracy of a map that contains many
small polygons-the typical case. Accuracy assessments of satellite imagery are greatly improved
by double-sampling techniques (Kalkhan et al.,
1995) by which separate error matrices are developed between classifications on the satellite imagery and aerial photography and between aerial
photography and ground observations. These error
matrices provide information on the accuracy of
map categories at various levels of resolution (Congalton, 1991; Stohlgren et aI., 1997a). Other potential sources of map error include (I) sharp delineations of boundaries where gradients or broad
ecotones exist (Stohlgren and Bachand, 1997), (2)
exclusion of rare or heterogeneous habitats in the
classification scheme (Stohlgren et al., 1997c), and
(3) poor map resolution or a large minimum mapping unit (Stohlgren et al., 1997a). Regardless of
the type of data, mapped or not, an understanding
of the quality and limitations of the data will improve their usefulness.
5.6 Conclusion
I conclude with nine suggestions to improve the
data acquisition process for ecological assessments.
1. Clearly articulate goals and objectives. Routinely review the objectives to see if they can
achieve the goals. Modify objectives where necessary.
2. Contact experienced groups with similar objectives. Seek help before designing a complex plan
for data acquisition. Others may prevent you
from "re-creating the wheel" and help you to
avoid mistakes.
3. Work in teams. Data acquisition and management are not a one-person job. Everyone involved in a project must help. Roles and responsibilities must be clearly defined.
4. Commit the necessary funds early, and maintain
them. A good rule of thumb is that 20% to 25%
(or more) of project funds should be allocated
to data acquisition and management.
5. Have a written plan. The plan should include
the goals and objectives, sampling protocols, ex-
termining adequate sample sizes is an iterative
process, recalculating variance and temporal and
spatial replication needs as the study progresses.
Befriend statisticians and sampling design experts.
The biggest design flaw in many ecological assessment programs is biased site selection (Stohlgren
et al., 1998). Why are so many long-term study
plots in flat terrain and close to roads? Krebs (1989,
p. 202) defmes this as "accessibility sampling,"
where "the sampling unit is restricted to units that
are readily accessible," such as where samples of
forest stands are taken only along roads. He also
wrote that "judgmental sampling" occurs when "the
investigator selects on the basis of experience a series of 'typical' sampling units-a botanist may select 'climax' stands of grassland to measure." The
issue here is that these sampling approaches are rejected by statisticians because they cannot be evaluated by the theorems of probability (Krebs, 1989).
Hence they are of limited use for extrapolating assessment results beyond the plots themselves
(Stohlgren, 1994).
The long-term value of data sets also depends on
the intensity and spatial pattern of sampling (Fortin
et aI., 1989; Stohlgren, 1994). Data from sparsely
sampled sites may suggest no effect or trend where
one exists. There is a growing appreciation for the
need for adequate replication in long-term monitoring (Hinds, 1984; Likens, 1991; Stohlgren et al.,
1995b). However, Kareiva and Anderson (1988)
showed that (1) as plot size increased in ecological studies, the number of replicates decreased, and
(2) once plot size reached about 3 m 2 , the number
of replicates was consistently less than 5.
Sample size determination is tricky business.
Typically, appropriate sample sizes are determined
after evaluating between-plot variance from initial
field tests (Krebs, 1989, pp. 173-199). The appropriate sample size is then determined by corresponding the variance to a predetermined level of
accuracy. Since disturbances of some kind or another frequently occur on most landscapes, studies
designed to quantify disturbance effects must anticipate the need for large sample sizes. Interacting natural processes on an already heterogeneous landscape
are further complicated by a range of species-specific
and site-specific responses (Stohlgren et al., 1995b).
Sampling at multiple spatial scales is also becoming more commonplace because it is virtually impossible to inventory an entire landscape or region
(Whittaker et al., 1979; Shmida, 1984; Stohlgren et
al., 1998). Data acquisition for trend analysis must
consider temporal and spatial variability simultaneously.
75
5.5.2 Accuracy Assessments of
Mapped Data
It often proves difficult to assess the accuracy of
maps. Field data, which are needed to assess map
accuracy, are expensive (Kalkhan et aI., 1995).
Checking only large polygons on a map overestimates the accuracy of a map that contains many
small polygons-the typical case. Accuracy assessments of satellite imagery are greatly improved
by double-sampling techniques (Kalkhan et al.,
1995) by which separate error matrices are developed between classifications on the satellite imagery and aerial photography and between aerial
photography and ground observations. These error
matrices provide information on the accuracy of
map categories at various levels of resolution (Congalton, 1991; Stohlgren et aI., 1997a). Other potential sources of map error include (I) sharp delineations of boundaries where gradients or broad
ecotones exist (Stohlgren and Bachand, 1997), (2)
exclusion of rare or heterogeneous habitats in the
classification scheme (Stohlgren et al., 1997c), and
(3) poor map resolution or a large minimum mapping unit (Stohlgren et al., 1997a). Regardless of
the type of data, mapped or not, an understanding
of the quality and limitations of the data will improve their usefulness.
5.6 Conclusion
I conclude with nine suggestions to improve the
data acquisition process for ecological assessments.
1. Clearly articulate goals and objectives. Routinely review the objectives to see if they can
achieve the goals. Modify objectives where necessary.
2. Contact experienced groups with similar objectives. Seek help before designing a complex plan
for data acquisition. Others may prevent you
from "re-creating the wheel" and help you to
avoid mistakes.
3. Work in teams. Data acquisition and management are not a one-person job. Everyone involved in a project must help. Roles and responsibilities must be clearly defined.
4. Commit the necessary funds early, and maintain
them. A good rule of thumb is that 20% to 25%
(or more) of project funds should be allocated
to data acquisition and management.
5. Have a written plan. The plan should include
the goals and objectives, sampling protocols, ex-