TO FIX OR NOT TO FIX?
The modern monitor's dilemma
The following article by Vicki Nelson was written as a discussion
paper for a recent workshop on the Great Barrier Reef Marine Park
Authority/Department of Environment video monitoring program. There had
been some debate as to whether it was more appropriate to use random
transects instead of fixed transects, and Vicki's paper was intended to
highlight some the of the issues involved, and to list the pros and
cons of each approach. Since the issue of whether to use fixed or
random transects in monitoring programs is a common one, I think it is
appropriate to give this paper wider exposure. Vicki has recently left
the Research and Monitoring Section to take up a position in the
private sector. We wish her well in her career.
Jamie Oliver, Monitoring Coordinator
|
MONITORING OF REEF BENTHOS:
fixed v. random transect
Vicki Nelson
ny decision regarding the choice between fixed or random
transects should be based on the objectives of the monitoring. For
example, if the main objective of monitoring is to extend a separate
project which uses fixed transects, then sites and transects must be
fixed. If the objectives don't define a priori which method to use,
deciding which approach to use should be based on knowledge of spatial
variation, the cost of reallocating sites, whether permanent sites have
advantages for other reasons, and the power of tests from each
approach.
The issues are as follows.
Statistical Issues
The re-survey of fixed transects has one major advantage:
- The error variance in the Analysis of Variance (ANOVA) test is not
inflated by natural spatial variability. Green (1993) says
`Re-randomisation does nothing more than cloud the comparison of
differences without truly adding error degrees of freedom.' Fixed
transects are more appropriate than re-randomised transects (within
fixed sites) when we are interested in differences in temporal changes
among sites because the error term in a repeated measures analysis will
be within-site temporal variation (as opposed to spatial variation
among sites).
Some possible disadvantages of a re-survey of fixed transects:
- Often fewer fixed transects are needed to achieve
the same level of power as random transects. Green
(1993) claims that this poses questions about the degree of coverage of
the site by transects (i.e. representativeness). However, if power is
similar between the two methods, then the transects must by definition
be equally representative (since power and representativeness are both
functions of within-site spatial variation).
- Green also claims that the robustness of the test is sensitive to
small degrees of freedom in the error term. For the same power, fewer
transects are sampled using fixed transects than random ones. This
means that assumptions are more difficult to check and that violations
are more likely to be serious.
- If transects are fixed and one is lost or can't be sampled for some
reason, there is a statistical problem. Repeated measures ANOVAS do not
deal with missing values very well. Either the missing cell must be
estimated (with associated problems) or all data for that replicate
must be omitted (after 20 years, this could be a bit of a
disappointment).
- Fixed transects are usually not completely fixed. There is an error
associated with resampling fixed transects.
- An error degrees of freedom at each site is lost each time you
sample if you use multivariate repeated measures analysis of variance.
When the number of times exceeds the total number of replicates, the
test is impossible (Green 1993). This means that for a long-term
monitoring program, you have to have a lot of replicates or you may not
be able to carry out an analysis of the program because there are not
enough error degrees of freedom to do the test. This problem is
enhanced with more complicated ANOVA models (e.g. unbalanced, nested or
multifactorial designs). The multivariate approach to repeated measures
analysis is useful because there are no assumptions about correlations
through time (compound symmetry, i.e. the correlation between time 1
and time 2 must be the same as the correlation between time 1 and time
6). However, there are ways to adjust for violations
of this assumption using the univariate approach but the power of the
univariate test is diminished.
A site needs to be sampled more intensively if transects are
reallocated each time because each time you revisit the site, you're
resampling it. That means that the estimate of spatial variation will
itself vary among times so that there is a confounding of temporal with
spatial variability. In order to minimise the probability of confusing
change over time with spatial variability, you need to make sure that
the variance among transects at a site is small relative to temporal
variation. The only way to do that without fixing the transects is to
sample a lot of them.
Logical Issues
The Department of Environment's monitoring program was initially set
up to complement the Australian Institute of Marine Science Long-term Monitoring Program as well as to answer questions about local
management issues. If the data is to be comparable with that of the
Australian Institute of Marine Science, the designs should be as
similar as possible. However, there are some differences in the
programs. The Australian Institute of Marine Science only samples the
north-east corners of reefs, while sites are distributed around reefs
in some of the Department of Environment's sampling. Because of the
heterogeneity in designs even among the different Department of
Environment programs, there is currently no way of using the data in a
broad-scale analysis of all monitoring done by the Department of
Environment. There are two ways of dealing with this. The first is to
accept it and design programs according to local objectives only. The
second is to impose a rigid sampling strategy (the Long-term Monitoring
Program's strategy) on all monitoring done by the Department of
Environment. Both of these approaches are reasonable, but for different
reasons. If the Long-term Monitoring Program's strategy is adopted by
all Regions, the monitoring will be comparable across Regions and
between the Department of Environment and the Australian Institute of
Marine Science. If local Department of Environment Regions define their
objectives and design programs to suit their Region, monitoring will
not be comparable among programs, but the monitoring is more likely to
have the ability to detect defined changes if it is appropriately
designed. Under this second scenario, there is no problem with using
random transects as opposed to fixed, as long as the sampling is well
designed.
A major problem could arise if it is decided in the middle of a
program that random transects are more appropriate. The problem with
changing some of the sites from fixed sites with fixed transects to
fixed sites with random transects within sites is that data will not be
comparable across time at those sites, nor will sites with fixed
transects be able to be compared with those with random transects. That
means we are constrained to a site-by-site or reef-by-reef approach,
rather than taking the broader perspective initially intended
(comparing among regions).
Logistic Issues
There are logistic difficulties with fixed transects in areas with
low visibility. It can be difficult to relocate fixed transects,
especially if markers are missing. These difficulties can be overcome
by marking transects with more than one stake (the Australian Institute
of Marine Science marks 50-metre transects with stakes every 10 metres)
and fixing sites with a global positioning system. Careful mapping of
sites should also help to relocate fixed transects. Nevertheless, the
problem of finding transects with limited field time and constrained
budgets is not trivial. Under circumstances with low visibility and
limited time, random unmarked transects make sense. The challenge is to
make sure that the transects are truly representative of the site and
that there are enough of them to provide a good (i.e. small) estimate
of within-site variability.
Aesthetic/Ethical Issues
In areas such as the Whitsunday Islands (Queensland)
or other highly used sites, the question of whether it is reasonable to
scatter stakes all over the reef is a serious one. Visible stakes
reduce the beauty of an area for visitors, but probably don't affect
the ecology of the place. Whether transects are fixed or not will
depend on the social impact of stakes rather than any ecological issue.
Summary
There are both advantages and disadvantages of using either fixed or
random transects. The primary reason for choosing one over the
other should be which one suits the purpose of the monitoring program
better. Issues such as logistical difficulty and statistical problems
enter into the decision making process AFTER the decision about
objectives has been made. The issue is by no means a simple one and
decisions over which type of approach to use must be considered well
from many points of view.
Reference
Green, R.H. 1993, Application of repeated measures designs in
environmental impact and monitoring studies,
Aust. J. Ecol. 18(1): 81-98.
|
back to RR homepage | back to contents this issue | to next page
Great Barrier Reef Marine Park Authority
PO Box 1379 TOWNSVILLE QLD 4810. Phone: (077) 500 700, Fax: (077) 726
093
E-mail:
registry@gbrmpa.gov.au