Diversifying crop rotations with temporary grasslands - Université de ...

oth by the current crop (wheat vs. lucerne) and the
preceding crops (annuals vs. perennials). All surveys
were performed during the same season (March to May)
of the years 2006, 2007 and 2008. Numbers of fields
surveyed per group and per year are provided in
Table 1. While most of these fields were chosen independently
(space-for-time substitution), some individual
fields were followed for the transitions between young
and old lucerne (b and c, 13 fields) and between old
lucerne and wheat following lucerne (c and d, 12 fields).
Weed species composition was described based on
presence–absence data of all herbaceous plant species in
several quadrats in each field. Crop volunteers were
excluded from all analysis, as they may artificially
increase impact of the preceding crop. Lucerne volunteers
were frequently found in wheat following lucerne.
In wheat, surveys were done in 32 quadrats of 4 m 2
(2 m · 2 m) per field arranged on transects forming an
eight-pointed star in the centre of the field. In lucerne,
surveys were peformed on 30 quadrats of 0.25 m 2
(0.5 m · 0.5 m) which were arranged on two to three
parallel transects. Field edges were avoided in both
cases. Different quadrat sizes were necessary to adequately
describe the weed species composition in both
annual and perennial crops that greatly varied by their
vegetation density (high in perennial lucerne, low in
winter wheat). A statistical method was used a posteriori
to check whether the two methods adequately described
the weed species composition. For each field, we
calculated the ratio of the observed species richness to
the expected total species richness estimated by ChaoÕs
formula (Colwell & Coddington, 1994) using the ÔspecpoolÕ
function in the ÔveganÕ package of R 2.8.1
(Oksanen et al., 2009). This ratio was then compared
between the four groups of fields. There was no
significant variation among mean (F3,416 = 0.67,
P = 0.57) and median values (v 2 = 2.2, df =3,
P = 0.53) of the four groups. On average, the sampled
weed species richness of each field was about 75% of the
estimated total, suggesting that both methods were
equivalent for describing the weed composition and the
relative frequencies of the most important taxa. A total
Table 1 Four groups of fields (treatments) defined to represent
four key stages of crop rotation including annual and perennial
crops
Group Crop and precedent
Nb. of fields surveyed
2006 2007 2008 Total
a Wheat after annual crops 87 41 56 184
b Lucerne 1 year 14 8 13 35
c Lucerne 2–6 years 55 53 51 159
d Wheat after lucerne 4 17 21 42
Total 160 119 141 420
of 161 weed taxa were distinguished, comprising 129
species and 32 genera or species groupings that posed
identification difficulties. Presence–absence data of each
quadrat were used to calculate the relative frequency of
each taxon in the field, which was used as a proxy of the
species abundance at the field scale.
Statistical analysis
Ó 2010 INRA
Journal Compilation Ó 2010 European Weed Research Society Weed Research 50, 331–340
Arable weeds and perennial lucerne 333
Community composition
Rare species (80 taxa present on less than 10 fields out of
420) were excluded from multivariate analysis, as they
may unduly influence the results (Kenkel et al., 2002).
Canonical Discriminant Analysis (CDA; Kenkel et al.,
2002) was used as a constrained ordination method to
visualise the differences in weed species composition
between the four groups of fields. Analysis of Similarities
(ANOSIM; Clarke, 1993) was used for testing differences
in species composition. This randomisation-based
method is recommended for analysing large multivariate
data sets containing many zeroes (Sosnoskie et al., 2006)
and does not require assumptions about multivariate
normality (Kenkel et al., 2002). We used the Bray–
Curtis dissimilarity measure and 10 000 permutations.
After the global analysis, we tested the pairwise differences
between all groups and reported the Bonferronicorrected
P-values. Lastly, Indicator Species Analysis
(ISA; Dufrene & Legendre, 1997) was used to identify
the most representative weed species of the four groups
of fields. Indicator values (IV) are calculated for each
species in each pre-defined group varying between 0
(species absent from all fields of that group) and 100
(species present with highest abundances in all fields of
the group, thus Ôperfect indicationÕ). Indicator values are
tested for statistical significance using a randomisation
technique (4999 permutations of the fieldÕs group
memberships).
Functional groups
All 161 weed taxa were sorted into eight a priori defined
functional groups (FG). Grasses were divided into
annual and perennial species, broad-leaved species into
annual, perennial and ÔintermediateÕ species (comprising
biennials and species varying between annual and
perennial life cycles). Annual broad-leaved species
constituted the largest group. This was therefore further
split according to morphology, opposing ÔuprightÕ (erect
morphology since seedling stage), ÔclimbingÕ (species
winding on neighbouring plants), ÔrosetteÕ (circular
arrangement of the first leaves near to the soil surface)
and ÔotherÕ (comprising all other morphologies). For
each field, relative frequencies of the FGs were calculated
by dividing the sum of the frequencies of all species
in each FG by the sum of species frequencies across all