Geoscience Reference
In-Depth Information
using species body mass averages (e.g.
Cohen et al., 1993
) or individual-level
feeding interaction data (
Barnes et al., 2010
). It was then possible to test if
these relationships were altered by the choice of resolution or grouping in a
consistent fashion across systems.
II. METHODS
In general, the relationship between body mass and nine response variables
divided into three sets was examined using different linear regression models.
Firstly, in Size Structure Dimension set #1: Trophic orderings (Section
II.C.1
),
prey body mass, predator-prey body mass ratio (PPMR) and trophic height
(TH) were regressed against different aggregations of predator and species
body masses. Secondly, in Size Structure Dimension set #2: Diet variation
(Section
II.C.2
), the variance and range of predator's prey body masses, and
the in-degree (generalism) of predators were regressed against various aggre-
gations of predator and species body masses. Lastly, in Size Structure
Dimension set #3: Predator variation (Section
II.C.2
), the variance and
range of prey's predator body masses, and the out-degree (vulnerability) of
prey were regressed against different aggregations of prey and species masses.
In order to search for species-based and size-class patterns, we expressly
needed to analyse food webs constructed from individual-level data (
Ings
et al., 2009; Woodward and Warren, 2007; Woodward et al., 2010
). Specifi-
cally, we used datasets of predator-prey interactions in a given locality for
which the species identity and size estimates were directly observed (e.g. via
gut contents analysis) for both predator and prey individuals involved in
each interaction. Data were from the four systems described by
Woodward
et al. (2010)
plus a further three previously unpublished food webs, from
Chilean rivers. The former are described in detail elsewhere (
Woodward
et al., 2010
), whereas the latter systems are described in greater detail
below, and the key characteristics of all seven are summarised in
Table 1
.
The raw data of individual predator-prey interactions were aggregated
using a taxonomic-based approach to generate different levels of resolution,
or binned into size classes using a size-class-based approach to provide
groups, which could be contrasted with the taxonomic aggregations. Com-
parisons of the response variables could then be done, either between the
levels of resolution or between the two approaches to grouping. For details
on aggregations conducted, see Section
II.B
; for details on response variables
analysed, see Section
II.C
; and for details on regression models used, see
Section
II.D
.
