The form of the sample unit must be decided during the planning
stages of the survey or monitoring program as the choice of sample
unit will depend on the aims. Remember that precision and accuracy
of the estimate will be affected by sampling criteria choice.
Sampling criteria include
size
shape
number
arrangement
errors
Size
How large a quadrat should be used?
Quadrat size depends on the morphology of the species, the homogeneity
of vegetation and the local conditions of the study site.
- small quadrats (e.g. 1 metre squared; with a side scale of 1:2,
e.g. 0.71 m x 1.41 m) for small plants
- large quadrats (e.g. 10 m2;
2.24 m x 4.47 m) for scrub and woodland and more widely spaced
individuals.
The quadrat should be slightly larger than the 'minimal area'
of that vegetation. It may be necessary to undertake a pilot study
to determine the appropriate quadrat size.
Shape
What shape of quadrat should be used?
The more edge to the quadrat per unit area, the more possibility
of miscounting individuals. The ratio of length of edge to inside
area in a circle is less than that for a square. That of a square
is less than that of a rectangle (i.e. edge length:area = circle<square<rectangle).
The effects of your choice on the accuracy of your sample is species
and habitat specific. The choice of quadrat shape may be guided
by the patterns within the vegetation being sampled, or by the logistics
of sampling. For example, when sampling diverse low heath, it may
be easier to keep track of your data taking in a long and narrow
quadrat, than in a square one.
Number
How many quadrats will be used?
The decision of how many quadrats to sample requires balancing
the (a) need to include enough quadrats to capture the true variabiltiy
within the vegetation being sampled with (b) the available labour,
time and cost. The actual number you should include in your sample
depends on the variability between qudrats (i.e. within the area
being sampled). Your sample should include enough quadrats to capture
the true variabiltiy within the vegetation being sampled.
A plot of the running mean and standard error (a measure of variability
about the mean) of your results against the number of quadrats sampled
will provide an indication of whether you have captured that variability.
For example, you will have collected data from sufficient quadrats
when the standard error begins to stabilise.
A decisions about how many qudrats to sample may also be based
upon the total area sampled. Often 30% of the total area is assumed
to capture a representative sample. The variance within a sample
would be less than the variance between samples but what is acceptable
will depend on the degree of heterogeneity of the encountered species
and the degree of accuracy required.
The most effective and efficient way of optimising your sample
size is to aim to maximise the degrees of freedom of your analysis.
Refer to notes for SBI209
Topics 4-6 for examples of the use of degrees of freedom and
how these effect confidence in your conclusions, based on sample
sizes.
Arrangement
How will the quadrats be oriented?
Where will the quadrats be placed in the study area?
Quadrat arrangement can be
- Selective - quadrats are arranged subjectively to include
representative areas or areas with some special feature, such
as the species under study.
- Random - each sample by definition must have an equal
chance of being chosen. Samples may be positioned by using pairs
of random numbers as distances along two axes positioned at right
angles to each other. Many statistical tests assume that data
have been randomly collected.
- Regular (systematic) - samples positioned using a grid,
or transects arranged linearly and contiguously are examples of
systematic sampling. This method is often used in the study of
pattern. It may be useful as a preliminary method of identifying
variability in an area, or for mapping. This method, however does
not account for sub-areas within the study area. Hence some sub-areas
may be undersampled and some over sampled. Limited statistical
tests are available. It can give a very distorted result if the
area is not homogenous.
- Restricted random (partial random) - area under study
is subdivided and then each subdivision is sampled at random.
- Stratified - stratified sampling involves dividing the
area of study into relatively homogenous sub-areas and then sampling
each of these at a frequency according to its area.
|
For more information
Please view
weblinks
http://www.unb.ca/forestry/
centers/cwru/soe/plantbio.htm
"Plant biodiversity in Natural, mixed-species Forests and
Silvicultural plantations in the vicinity of Fundy National
Park".
Case study within the Greater Fundy Ecosystem Research Project
by Cam Andrea Veinotte, Bill Freedman and Wolfgang Maass (1996).
This study aims to examine the effect of clear cutting and
plantation establishment on the region's floral biodiversity.
Details are provided of the sampling regime
which included, within each stand, 30 quadrats of 1m2
established along a transect, in order to
sample the ground vegetation present within the site.
|
|
|
|
5.2 (iii)
Please read this case study and identify what sampling methodology
they have used.
We will discuss this during out next online tutorial.
|
Errors
Errors are deviations of observations from true population values.
For example
- Bias - refers to deviations in a constant direction.
Bias may result from
- inaccurate measuring instruments
- incorrect technique
- consistent misapplication of technique or sampling design
- consistent misidentification.
- Indeterminate errors - deviations are equally likely
to be in both directions. This may result from
- personal factors such as fatigue, inattention, response
to weather
- sampling errors - because only a fraction of the whole target
area is measured, the mean value of sampled data will always
differ from the true mean of the population.
Example
A sampling strategy with simple stratification
on vegetation types is likely to bias sampling against
the floristically-rich ecotonal areas where major vegetation types
overlap. A systematic method such as Gradsect avoids
this bias.
Gillison & Brewer (1985) showed that gradsects capture more information
than randomly placed quadrats of similar length. This paper details
a gradsect survey in NSW which aimed to determine the type and range
of eucalypt and rainforest communities present, and to use the results
to provide a stratified sample for a fauna survey
to establish the correlation between fauna and floristic composition.
Bias due to lack of accessibility to certain grid
points was made explicit by restricting sampling to within 0.5 km
of tracks accessible to 4WD vehicles.
|
For more information
Please read
Please continue to read your textbook,
pages 111-138.
Reading 5
Goldsmith et al. (1976) for the criteria for selecting quadrat
size, placement, orientation etc. This reading will also assist
you to consider the concept of 'minimal area'.
Further Readings
Sampling Methods/Criteria
(1) Shifley, S.R. and Schlesinger, R.C. (1994) 'Sampling
guidelines for old-growth forests in the Midwest, USA'
Natural Areas Journal 14 (4): 258-268.
In this study the relation between sampling intensity
in old forests and precision of estimates for
(i) measures of stand density
(ii) woody debris, and
(iii) species richness/diversity, were examined. Effects of
plot size are discussed.
(2) Menges, E.S. and Gordon, D.R. (1996) 'Three levels of
monitoring intensity for rare plant species' Natural Areas
Journal 16 (3) 227-237.
Three hierarchical levels of sampling are used.
Level one focuses on species occurrence (presence/absence).
Level two involves a quantitative assessment of abundance
or condition, often in terms of percent cover, density or
frequency.
Level three involves demographic monitoring of marked individuals.
The three levels can be efficiently nested.
(3) Overton, W.S. and Stehman, S.V. (1996) 'Desirable design
characteristics for long-term monitoring of ecological variables'
Environmental and Ecological Statistics 3 (4)
: 349-361.
This paper discusses the difference in demands that a sampling
strategy for long-term monitoring has compared
to a survey designed for a single time period.
(4) Johnson, R.R. and Higgins, K.F. (1998) 'Bias in quadrat-derived
estimates of number of prairie wetlands' Wetlands 18:3.
This study aimed to evaluate the magnitude of bias
in estimates of the number of temporary, seasonal, semipermanent,
and total wetlands from samples of square quadrats of different
sizes.
|
back to sampling concepts
|