Geoscience Reference
In-Depth Information
groupings were, however, not significant (
Figure 4
E and
Table A2
). Both
types of groupings suffer from averaging effects, but none as severe as with
species averaging, because only individuals actually engaged in a predation
event are aggregated into interacting nodes. The extent of averaging effects
caused by size-class grouping will depend on the number of size classes used
and how skewed the size distribution of individuals are within a size class.
A large enough number of size classes would ultimately render a network
equal to the individual-level data with one predator individual or prey per
size class and hence no skew. As the middle of a size-class' range (on a log-
scale) is used as the mass representing a size class, the worst case scenario
would be few size classes with size ranges chosen, so individuals assigned to a
size class fall close to one of the edges of it. Here, following
Woodward et al.
(2010)
, we used as many size classes as there were species in the system, to be
able to make a fair comparison with the taxonomic grouping. The averaging
effects of the link average grouping would presumably be less pronounced
relative to the size-class grouping, as here the mean of the predator or prey
individuals assigned to a node is used. Still, this distribution could be skewed
by the presence of a few very large or small predator or prey individuals.
To understand the difference observed between the two groupings, we
must bear in mind that each type of grouping suffers from different effects
of aggregation and sampling. If we examine the difference in the TH-mass
relationship (
Figure 7
), where we find that the size-class-based approach
suggests a steeper slope, we can see how these biases might arise. First of
all, the taxonomic approach ignores ontogenetic niche shifts, thereby obscur-
ing patterns of individuals within a given species altering the TH at which
they are feeding depending upon their size (
Jennings et al., 2008; Nakazawa
et al., 2010
). By ignoring this process, we will tend to reduce the TH of a
taxonomic node by using the mean and this could then be responsible for
reducing the overall TH scaling in the taxonomic food web compared to that
derived using size classes.
Alternatively, a well-sampled size-class-based network will tend to mix
individuals that in reality differ in their TH in a way which will tend to inflate
the TH of the node. For example, if we consider a large individual of a
primary consumer (a basal species in the webs presented here), which could
be placed in the same class as a predator as such that the size class will now
have a TH of two or above but contain a basal species. Size-class-based webs
are also especially vulnerable to sampling effects, as a size class that has no
predator individuals in it but does have a resource individual will appear to
be a basal node even if it contains no individuals of a taxonomically basal
species (see
Figure 7
). On balance, if the sampling effect is less than the
inflation of a node's TH, then the slopes will appear steeper.
Clearly, both approaches for assessing the relationship between TH and
body mass across the whole community have their own limitations and
