Stratification can be used alongside regular sampling too. For example, we could
take every seventh square from a stratum where a bird is thought to be common,
but every fourteenth square from a stratum where it is thought to be rare
(see Nemeth and Bennun 2000 for a similar approach).
There is, however, a possible bias in systematic sampling, in that this method
might over- or under-sample certain features that are regularly distributed in the
landscape. For example, it might be that parallel roads are the same distance apart
as our lines of samples, leading to over- or under-sampling of areas near roads. In
reality, however, such biases are very rare, although we need to be aware of them.
In summary, regular sampling has much to recommend it and it has probably
been under-used in the past.
An attractive alternative is to integrate the strengths of random and regular
sampling by using a randomized Latin squaredesign (Figure 2.9(b)), in which
each column and each row holds one, and only one, sampling unit. Sampling
units are drawn randomly from the rows and columns on the condition that
every row and column can only contain a single square, which ensures balanced
coverage of the area. This pattern of sampling can be repeated across the study
area and within larger sampling units.
2.3 Field methods
In the section above, we considered the key question of how we choose where
to make our counts. Now we must consider how to choose between counting
methods. Although we have presented survey design as a linear process, in reality,
there should be a strong feedback loop in which the sampling strategies and
Field methods| 35(a) (b)Fig. 2.9(a) An example of a regular sampling method, and (b) a randomized Latin
square design. Survey squares are shaded.