Environmental Engineering Reference
In-Depth Information
motorised harvesters may operate over tens of kilometres or more, and sampling
units should be defined accordingly. Where the ecology of the species operates on
a different scale to that of the harvest, the finer scale should be used to define
sampling units. The scale of measurement should be similar for abundance and
predictor variables.
Survey design
In order to ensure that gradients in important environmental variables are fully
covered, the ideal survey design is to
stratify
by key variables (Hirzel and Guisan
2002; Vaughan and Ormerod 2003). To do this, identify which habitats or influ-
ences are likely to be most important (harvesting effort will be a key one), categorise
these variables across their ranges, and divide the study area into regions according
to combinations of these categories. Equal sampling effort can then be allocated
to each region. Where the focal species is rare or localised, it may be necessary
to increase sampling effort in sites known to have good abundance (effectively
stratifying by abundance). However, the analysis and interpretation of data collected
in this way is more complicated (Keating and Cherry 2004) and the approach is
not analytically ideal.
Sample size
The
number of survey sites
required to deliver robust measures of association
depends primarily on the number of candidate explanatory variables. A basic
correlative study of harvest impact may be possible with just a handful of sites, but
ideally, if the analysis is to be used predictively, at least ten sites should be covered
for each environmental variable considered.
Analysis and interpretation
The most accessible approach to the analysis of spatial patterns of abundance is to
use a
generalised linear model
to identify significant predictors of species' abun-
dance (for example see Boxes 2.19 and 4.4). This approach is flexible (methods are
available to deal with either binary presence-absence data, raw counts or absolute
abundances), and readily available in standard statistical software (Section 2.7).
However, there are other approaches which might be considered, for example in
order to account for spatial auto-correlation in your data, or in order to cope with
large numbers of candidate predictor variables. Guisan and Thuiller (2005) pro-
vide a useful review of recent developments and current tools in this area.
If spatial analyses are to be used predictively, their reliability should ideally be
validated
by comparing predicted abundance or occupancy with the actual situ-
ation in sites other than those used for the original analysis (Vaughan and Ormerod
2005). This requires further sampling, which should be taken into account when
forecasting the sampling effort required. In the end, though, it is often difficult to
identify good spatial predictors of abundance, even with a great deal of effort.
Constraints on sampling effort, failure to identify good candidate variables, and
