1.1. Selection of species

Study objectives will always determine which species are selected for trait measurement. For species-level
analyses of trait variation, and for identifying general strategies or syndromes of resource use, or trade-offs at
the local, regional or global scale (e.g. Reich et al. 1997; Westoby et al. 1998; Díaz et al. 2004; Wright et al.
2004; Gubsch et al. 2011), species or populations from a broad range of environments and phylogenetic
groups should be selected. For questions about evolution, the choice of species may be based on the
inclusion of representatives of different enough phylogenetic groups, or on other phylogenetically relevant
criteria (such as being members of particular clades), with little consideration about their abundance in situ.
In contrast, when trying to understand how environmental variables shape vegetation characteristics, or how
vegetation characteristics affect local flows of matter and energy (e.g. primary and secondary production,
carbon, water and mineral nutrient cycling), the main criterion for species selection should be local
abundance. In those cases, species should be selected that collectively make up for ~80% of cumulative
relative abundance, following Garnier et al. (2004) and Pakeman and Quested (2007) (see specifics for
abundance measurements below). Exceptions may be made if this criterion would imply measurements for
an impracticably large numbers of species, e.g. communities with unusually high species richness per unit
area, especially combined with a very high evenness. Examples are tropical rainforests and fynbos
vegetation, in which well over 100 species per plot may be needed to reach the 80% biomass threshold.

In forests and other predominantly woody vegetation, the most abundant species of the understorey may
also be included (e.g. when the research question relates to the whole-community or ecosystem level), even
if their biomass is much lower than that of the overstorey woody species. In predominantly herbaceous
communities, species contribution to a particular community may vary with time during a growing season.
As a first step, we suggest that the relative abundance and the traits should be measured at the time of peak
standing biomass of the community. This does not always apply to reproductive structures, which obviously
have to be measured when they are present and fully developed, which sometimes does not coincide with the
time of maximum vegetative growth.

For comparing sites or for monitoring trends in ecosystem-level properties across environmental
conditions (e.g. pollution, or different regional climate or fertility levels), indicator species can be selected on
the basis of the sensitivity of their trait values to the environmental factor of interest, and their importance
locally and regionally, as well as for the ease with which they can be found and identified in the field
(independent of their relative abundance) (Ansquer et al. 2009; De Bello et al. 2011). In this sense, it may be
useful to distinguish ‘variable’ traits from more ‘stable’ traits (Garnier et al. 2007). Although most traits
show some variation within species along environmental gradients, or in response to specific environmental
changes, the intraspecific variation of so-called ‘stable traits’ is low compared with their interspecific
variation. The reverse is the case for so-called ‘variable traits’, which implies that they should preferably be
measured in more than one site or condition across the habitat range (Garnier et al. 2007). By contrast,
‘stable traits’ can be measured for any representative population from the entire gradient. Traits known to
often be ‘variable’ include vegetative and reproductive plant height, mineral nutrient concentration in leaves,
onset of flowering, branching architecture and spinescence. Traits that are relatively ‘stable’ include
categorical traits, such as life form, clonality, dispersal and pollination modes, and to a lesser degree
photosynthetic type (C3 or C4). Some quantitative traits such as leaf and stem dry matter content, or leaf
toughness can be ‘stable’ along certain gradients, e.g. of nutrients or disturbance, but not along others, e.g. a
light gradient (cf. Poorter et al. 2009). Species may therefore vary in which quantitative traits are stable
across given gradients, so tests should be made before a trait is taken to be stable for a given species (Albert
et al. 2010, 2012; Hulshof and Swenson 2010; Messier et al. 2010; Moreira et al. 2012).

Appendix 1 gives a rough indication of the within-species variability (coefficients of variation; i.e.
standard deviation divided by the mean, hereafter CV) for some of the quantitative traits described in the
present handbook, along with the more frequently used units and the range of values that can be expected.
Appendix 1 summarises field data collected in several studies for a wide range of species coming from
different environments. Because of the low number of replicates generally used, each of the individual
estimates bears an uncertainty (and CV will likely increase as scale increases); however, by looking at the
range of CVs calculated across a wide range of species, a reasonable estimate of the typical within-speciesat-a-site variability can be obtained. We, therefore, present in Appendix 1 the 20th and 80th percentiles of the CV distribution.

How species abundance should be measured to determine the species making up 80% of cumulative
abundance (e.g. whether to lay out transects, select points or quadrats at random or systematically, or to
follow a different method) is beyond the scope of the present handbook and is extensively covered in plantecology and vegetation-science textbooks. However, it should be noted that different methods are relevant to
different ecological questions and associated traits (Lavorel et al. 2008; see also Baraloto et al. 2010b,
specifically for tropical forest). Taxon-free approaches that do not require species identification offer an
alternative to estimates of relative abundance, and effectively capture the contribution of more abundant
species. These include measuring traits regardless of species identity, along a transect (‘trait-transect’
method, Gaucherand and Lavorel 2007), or for individuals rooted nearest to random sampling points, as long
as the canopy structure is quite simple (‘trait-random’ method – Lavorel et al. 2008). Methods of taxon-free
sampling have also been applied to tropical forests, being, in this case, strongly based on the frequency or
basal area of individual trees (Baraloto et al. 2010b). Trait values obtained through these methods can differ
from those obtained using the standard approach of selecting robust, ‘healthy-looking’ plants for trait
measurement (see Section 1.2).