Environ Sci Pollut Res (2017) 24:13121–13135
DOI 10.1007/s11356-017-8796-9
RESEARCH ARTICLE
Simulating changes in cropping practices in conventional
and glyphosate-resistant maize. II. Weed impacts on crop
production and biodiversity
Nathalie Colbach 1,2 & Henri Darmency 1 & Alice Fernier 1 & Sylvie Granger 1 &
Valérie Le Corre 1 & Antoine Messéan 3
Received: 16 June 2016 / Accepted: 9 March 2017 / Published online: 6 April 2017
# Springer-Verlag Berlin Heidelberg 2017
Abstract Overreliance on the same herbicide mode of action
leads to the spread of resistant weeds, which cancels the ad-
vantages of herbicide-tolerant (HT) crops. Here, the objective
was to quantify, with simulations, the impact of glyphosate-
resistant (GR) weeds on crop production and weed-related
wild biodiversity in HT maize-based cropping systems differ-
ing in terms of management practices. We (1) simulated cur-
rent conventional and probable HT cropping systems in two
European regions, Aquitaine and Catalonia, with the weed
dynamics model FLORSYS; (2) quantified how much the pres-
ence of GR weeds contributed to weed impacts on crop pro-
duction and biodiversity; (3) determined the effect of cultural
practices on the impact of GR weeds and (4) identified which
species traits most influence weed-impact indicators. The sim-
ulation study showed that during the analysed 28 years, the
advent of glyphosate resistance had little effect on plant bio-
diversity. Glyphosate-susceptible populations and species
were replaced by GR ones. Including GR weeds only affected
Responsible editor: Philippe Garrigues
Electronic supplementary material The online version of this article
(doi:10.1007/s11356-017-8796-9) contains supplementary material,
which is available to authorized users.
* Nathalie Colbach
Nathalie.Colbach@inra.fr
Antoine Messéan
Antoine.Messean@inra.fr
1
Agroécologie, AgroSup Dijon, INRA, University Bourgogne
Franche-Comté, F-21000 Dijon, France
2
INRA, UMR1347 Agroécologie, BP 86510, 17 rue Sully,
F-21065 Dijon, France
3
Eco-Innov, INRA, 78850 Thiverval-Grignon, France
functional biodiversity (food offer for birds, bees and cara-
bids) and weed harmfulness when weed effect was initially
low; when weed effect was initially high, including GR weeds
had little effect. The GR effect also depended on cultural prac-
tices, e.g. GR weeds were most detrimental for species equi-
tability when maize was sown late. Species traits most harmful
for crop production and most beneficial for biodiversity were
identified, using RLQ analyses. None of the species present-
ing these traits belonged to a family for which glyphosate
resistance was reported. An advice table was built; the effects
of cultural practices on crop production and biodiversity were
synthesized, explained, quantified and ranked, and the optimal
choices for each management technique were identified.
Keywords GM crop . Model . Weed . Glyphosate resistance .
Cropping system . Biodiversity . Yield gap . Harmfulness .
Agroecology
Introduction
Herbicide-tolerant (HT) crops, particularly those tolerant to
glyphosate, are grown on large acreages in some regions of
the world (James 2013). Though these varieties simplify weed
management (Bonny 2016; Brookes and Barfoot 2009;
Shaner 2000), the overreliance on glyphosate has led to the
development of resistant weeds (Bonny 2016; Heap 2016;
Powles 2008). In contrast to the widespread annual use of
glyphosate in HT maize and HT soybean rotations in the
USA, the acreage of HT maize in Europe is almost nil today,
and maize is often rotated with other crops that are not treated
with glyphosate. However, glyphosate is frequently used dur-
ing summer fallow, and cases of weed resistance to glyphosate
have already been detected in arable fields with annual crops
(Collavo and Sattin 2014). Moreover, the introduction of HT
13122
crops into the cropping systems may lead to other changes in
cultural practices which can also favour the evolution of her-
bicide resistance, e.g. simplified rotations and/or simplified or
no tillage (Beckie 2009; Boerboom 1999; Chauvel et al. 2009;
Colbach et al. 2016b; Colbach et al. 2017; Friesen et al. 2000;
Lievin et al. 2013; Moss and Clarke 1994; Moss et al. 2007).
This shift to herbicide-resistant biotypes potentially not on-
ly increases weed harmfulness for crop production (e.g. yield
loss, harvest contamination, field infestion, Mezière et al.
2015b) but can also impair biodiversity. Indeed, weed flora
is a major part of wild plant biodiversity in arable lands and
provides habitat and food resources to a range of animals in
agricultural landscapes, among which pollinators (Bretagnolle
and Gaba 2015) or crop auxiliaries (Taylor et al. 2006). The
management of glyphosate-tolerant crops can also lead to
shifts to different weed species, which has already been
analysed in recent studies (Bigler and Albajes 2011; Bürger
et al. 2015; Heard et al. 2003). However, the particular effect
of herbicide resistance on weed-related functional biodiversity
(e.g. weeds as food resources) has yet to be investigated.
Because of long-term effects and the multiplicity of the
implicated factors, herbicide resistance is increasingly inves-
tigated via modelling and simulations (Cavan et al. 2000;
Colbach et al. 2016b; Gressel and Segel 1990; Maxwell
et al. 1990; Neve et al. 2003; Renton et al. 2014). These
models are, however, limited to a single species and neglect
the impact of weeds on crop production and biodiversity.
Therefore, in the companion paper (Colbach et al. 2017), we
included a simple herbicide resistance submodel in the
existing multispecific FLORSYS model. FLORSYS quantifies
the effects of cropping systems and weather on multiannual
weed dynamics and a series of indicators reflecting weed
harmfulness for crop production and weed contribution to
wild plant biodiversity and functional biodiversity (Colbach
et al. 2014a). Weed harmfulness is defined here as any nega-
tive effect for the farmer, in terms in production amount and
quality, work load and reputation.
Here, the objective was to evaluate, with simulations, the
impact of glyphosate-resistant (GR) weeds, on crop produc-
tion and biodiversity in cropping systems including HT maize.
The tested cropping systems consisted of current conventional
situations and future systems including HT maize with on-
crop glyphosate applications as well as other probable chang-
es in cultural practices (e.g. simplified rotation and tillage).
These cropping systems were identified in a previous simula-
tion study from surveys and expert opinion in two large
maize-growing European regions, Aquitaine in South-
Western France and Catalonia in North-Eastern Spain
(Bürger et al. 2015). In this previous study, we already
analysed the impact of weed flora (consisting mainly of winter
and spring species) on production and biodiversity in these
cropping systems. Here, we continued the analysis, working
with an extended weed flora consisting of additional and more
Environ Sci Pollut Res (2017) 24:13121–13135
diverse species and discriminating glyphosate-susceptible and
glyphosate-resistant populations. We analysed (1) the relation-
ship between the presence of GR weeds and weed impact on
crop production and biodiversity, (2) how this impact was
affected by cultural practices and (3) whether species traits
other than glyphosate resistance influenced weed impacts.
Ultimately, the goal is to assess whether cropping systems
including HT crops are sustainable in the long term, in the
event of herbicide resistance.
Material and methods
The virtual field FLORSYS
FLORSYS is a virtual field on which cropping systems can be
experimented and a large range of crop, weed and environ-
mental measures estimated, with a series of indicators trans-
lating weed contribution to biodiversity and weed harmfulness
for crop production (hence weed-impact indicators,
section BAssessing weed impacts on crop production and
biodiversity^ below). The structure of FLORSYS is presented
in detail in previous papers (Colbach et al. 2014b; Colbach
et al. 2014c; Gardarin et al. 2012; Mezière et al. 2015b;
Munier-Jolain et al. 2014; Munier-Jolain et al. 2013) and the
main aspects are described in the companion paper (Colbach
et al. 2017).
The input variables of FLORSYS consist of (1) a description
of the simulated field (daily weather, latitude and soil charac-
teristics), (2) all the simulated cultural operations in the field
and (3) the initial weed seed bank. These input variables in-
fluence the annual life cycle which applies to annual weeds
and crops, with a daily time-step. FLORSYS parameters (sec-
tion A in supplementary material online) are currently avail-
able for 25 frequent weed species covering the main ecologi-
cal niches of temperate European cropping systems and in-
cluding the species frequently found in European maize crops:
Abutilon theophrasti (EPPO code ABUTH), Alopecurus
myosuroides (ALOMY), Amaranthus retroflexus (AMARE),
Ambrosia artemisiifolia (AMBEL), Avena fatua (AVEFA),
Capsella bursa-pastoris (CAPBP), Chenopodium album
(CHEAL), Datura stramonium (DATST), Digitaria
sanguinalis (DIGSA), Echinochloa crus-galli (ECHCG),
Galium aparine (GALAP), Geranium dissectum (GERDI),
Matricaria perforata (MATIN), Mercurialis annua
(MERAN), Panicum miliaceum (PANMI), Poa annua
(POAAN), Polygonum aviculare (POLAV), Fallopia
convolvulvus (POLCO), Polygonum maculosa (previously
P. persicaria, POLPE), Senecio vulgaris (SENVU), Sonchus
asper (SONAS), Solanum nigrum (SOLNI), Stellaria media
(STEME), Veronica hederifolia (VERHE) and Veronica
persica (VERPE).
Environ Sci Pollut Res (2017) 24:13121–13135
FLORSYS was evaluated with independent field data, show-
ing that daily species densities and, particularly, densities av-
eraged over the years were generally well predicted and
ranked in the model’s original region, i.e. Burgundy
(Colbach et al. 2016a). At southern latitudes, weeds flower
too early, and a corrective patch was used here to improve
the prediction of weed flowering dates.
Mutation and glyphosate resistance
The details of how glyphosate resistance was modelled are
given in the companion paper (Colbach et al. 2017) and are
summarized here. Although several studies show that glyph-
osate resistance in weeds can depend on several genes
(Sammons and Gaines 2014), glyphosate resistance is as-
sumed here to depend on a single gene. The combination of
a wild allele W and a mutant, resistant allele R at one unique
locus leads to three genotypes: susceptible (WW), semi-
dominant resistant (WR) and resistant (RR). These differ in
terms of (1) plant mortality after glyphosate (0.30 and 0.10 for
WR and RR populations, respectively, at 3 l/ha) and (2) repro-
ductive fitness (seed production of WR and RR plants is 0.90
and 0.75 of glyphosate-susceptible seed production in the ab-
sence of herbicides, respectively).
Twelve species considered in FLORSYS can potentially
develop resistance to glyphosate in the model (ALOMY,
AMARE, AMBEL, AVEFA, CHEAL, DIGSA, ECHCG,
PANMI, POAAN, MATIN, SENVU and SONAS).
These species belong to families for which glyphosate
resistance was reported (see references in Colbach et al.
2017). Each time ovules or pollen are produced, a tiny
proportion (10−6) mutate from glyphosate-susceptible to
resistant. Outcrossing species are pollinated with the
pollen of all flowering plants of the same species;
selfing species self-pollinate.
The verisimilitude of the simulated effects of cultural
practices on weed densities and the advent of glyphosate
resistance has already been discussed in the companion
paper, stressing that pleiotropic effects, cross-resistance
with other herbicides and gene flow from neighbouring
fields could modify the conclusions based on the simula-
tions (Colbach et al. 2017). The companion paper also
demonstrated, via a sensitivity analysis, that the values
chosen for the model parameters driving glyphosate resis-
tance (for which few data are available) would not notably
affect simulation outcome.
Assessing weed impacts on crop production
and biodiversity
The weed densities simulated by FLORSYS are translated into a
set of annual indicators depicting the weed flora impact on
crop production and biodiversity (Mezière et al. 2015b; see
13123
section A.4 online). Four indicators reflect the weed harmful-
ness for crop production, discriminating direct (crop yield
loss, harvest pollution, i.e. weed seeds and debris harvested
with the crop seeds), technical (harvesting problems due to
green weed biomass blocking the combine) and sociological
harmfulness (field infestation by weed biomass during crop
growth). The latter reflects the farmer’s worry of being con-
sidered incompetent by his peers, irrespective of any weed
effect on crop production.
Weed-mediated biodiversity indicators comprise two indi-
cators of wild plant biodiversity, i.e. (1) species richness (the
number of weed species) and (2) species equitability (Pielou’s
index). Three other indicators assess functional biodiversity,
i.e. weed contribution to feed other organisms in the agro-
ecosystems, considering the seasons of activity and food
shortage: (3) weed seeds on soil surface in autumn and winter
to feed field birds, (4) lipid-rich seeds on soil surface in sum-
mer to feed carabids and (5) weed flowers in spring and sum-
mer to feed domestic bees.
Cropping systems are always assessed over several years as
farmers are usually not only worried about the effect of the
current year’s weed plants on crop production but also about
the future effect of the many seeds produced by this year’s
weeds (Mezière et al. 2015b).
Simulation plan
The simulation plan was described in detail by Bürger et al.
(2015) and again more superficially in the companion paper
(Colbach et al. 2017). Only a short summary is given here,
more details can be found in section B of supplementary ma-
terial online. Simulations were carried out for Aquitaine in
South-Western France and Catalonia in North-Eastern Spain.
Typical maize-based cropping systems, as well as probable
systems with HT maize, were simulated. To decorrelate the
effects of the changes in cultural practices (e.g. no till) accom-
panying HT maize from those of glyphosate treatments, the
HT scenarios were also simulated with conventional herbicide
programmes, resulting in a total of 45 scenarios. The tested
systems also differed in terms of crop diversity in the rotation,
tillage frequency and intensity, sowing dates, etc. Each sce-
nario was simulated over 28 years and was repeated with 10
different weather series consisting of 29 randomly chosen
weather years, using the same 10 series for each scenario.
The initial seed bank was identical for all simulations and
consisted of the 25 species currently included in FLORSYS.
Relative species abundance was based on regional flora as-
sessments (Colbach et al. 2016a), and initial GR proportions
resulted from preliminary simulations (Colbach et al. 2017)
and ranged from 10−6 to 10−5 for the potentially mutating
species.
The same simulation plan (45 scenarios × 10 weather rep-
etitions) was then repeated without GR weeds, starting with
13124
only glyphosate-susceptible populations and applying a zero
mutation rate for all species. The comparison of the output
produced by the two simulation series allowed us to assess
how much weed impact was due to GR weeds.
Analysed outputs and statistics
Statistical analyses were run with version 9.4 of SAS and
version 3.3.0 of R (R Development Core Team, 2016). In
the first step, potential trade-offs between the annual weed-
impact indicators were analysed to evaluate, for instance,
whether weed harmfulness increased with increasing contri-
bution to biodiversity (which makes reconciling production
and biodiversity difficult), whether all biodiversity contribu-
tions increased or all harmfulness criteria decreased in the
same conditions (which simplifies management). To do this,
Pearson correlations (PROC CORR of SAS) and principal
component analysis (R library FactoMineR) were calculated
for the 45 scenarios, 10 weather repetitions and simulated
28 years. These analyses were carried out separately for results
from simulations with and without GR weeds. Then, data
from simulations with and without GR weeds were pooled,
and the differences in indicator values in simulations with vs.
without GR weeds were analysed as a function of weather
repetition, region, indicator values in GR-free simulations
and the interaction between the latter two, with linear regres-
sions using PROC GLM of SAS.
In the next step, only the last eight simulated years were
used, when differences between simulations with and without
GR weeds were highest. In the second step, analyses of vari-
ance were run for each region (Aquitaine or Catalonia) with
PROC GLM of SAS to explain indicator values as a function
of time since simulation onset (as qualitative variable), pres-
ence of GR weeds, cropping system as well as interaction
between the latter two as fixed effects and weather repetition
as block effect. Least-square means of weed-impact indicators
were compared for cropping systems, using least-significant
difference tests. For each cropping system, the differences in
indicators from simulations with vs. without GR weeds were
compared to zero, using t tests.
In the third step, indicator values averaged over the last
eight simulated years were analysed with linear models using
PROC GLMSELECT of SAS as a function of the same syn-
thetic cropping system descriptors (e.g. average number of
superficial operations per year and proportion of winter crops
in the rotation) as those used in the companion paper (Colbach
et al. 2017). Further variables were added here to account for
the particularities of the biodiversity indicators (e.g.
mouldboard ploughing during the carabid-activity period),
resulting in a total of 17 descriptors. These descriptors as well
as region were used as independent variables and were added
sequentially (forward selection) to the linear model in order to
optimize the Schwarz Bayesian information criterion. The
Environ Sci Pollut Res (2017) 24:13121–13135
final model was chosen among the successive models as the
one with the lowest predicted residual sum of square with
cross validation.
Finally, RLQ analyses were used to identify pertinent rela-
tionships between indicator variables (averaged over the last
8 years and the 10 repetitions) and species traits, using the R
library ade4 (Chessel et al. 2004). The RLQ analysis was
initially developed to analyse correlations between cultural
practices (R matrix) and species traits (L matrix) via weed
species densities (L matrix). Here, the practices were replaced
by indicator variables. Only trait-indicator relationships sig-
nificant at p = 0.05 were considered.
As in the companion paper (Colbach et al. 2017), we used
population densities instead of species densities, considering
WW, WR and RR populations of a given species as three
distinct species that differed in two traits, i.e. glyphosate resis-
tance and population fitness cost for resistance. Instead of 25
species, we thus worked with 49 populations. The same spe-
cies traits as in the companion paper were used, i.e. traits
related to seed bank processes, plant growth, morphological
plasticity and glyphosate resistance. In total, 21 traits were
used in the analysis (section A.2.1 of supplementary material
online).
Results
Weed and GR weed impact over time
When there were no GR weeds, weed harmfulness and con-
tribution to biodiversity usually took a few years of increase or
decrease before stabilizing at values depending on the weed-
impact indicator and the cropping system. For instance, bird-
food offer dropped from a score of 7.9 at simulation onset to
5.2 during the first year in the unploughed, early-sown maize
monoculture and then oscillated between 5.1 and 5.3 during
the rest of the simulation (Fig. 1a). If the field was tilled with a
mouldboard plough, the indicator stabilized at 5.5 to 5.7
(Fig. 1b). In some cases, the initial decrease or increase took
10 years (e.g. decrease in species richness in untilled maize
monoculture, Fig. 1c) or more (e.g. the increase in field infes-
tation in ploughed monoculture had not yet reached an equi-
librium after 25 years, Fig. 1d).
When GR weeds were included in the simulations, there
usually was no difference in weed impact during the ten first
simulated years, compared to the simulations without GR
weeds (Fig. 1). Later, the impact of GR weeds greatly differed
among cropping systems. In some systems, there was no real
difference in weed impact (Fig. 1a). In a few situations, in-
cluding GR weeds in the simulations increased weed harm-
fulness tremendously after a few years, e.g. field infestation
increased to a maximum of 0.7 vs. 0.3t /ha in GR-free simu-
lations (Fig. 1d). The same could happen for biodiversity, e.g.
Environ Sci Pollut Res (2017) 24:13121–13135
Fig. 1 Diversity in the response of weed impact on crop production and
biodiversity when including glyphosate resistant (GR) weeds in the sim-
ulations (red circles) compared to simulations without glyphosate-
resistant weeds (black triangles). a Fast initial decrease and no effect of
GR weeds (example of bird food offer in unploughed field). b Fast initial
decrease and later huge increase with GR weeds (example of bird food
with GR weeds, bird offer rose to more than 7 vs. less than 5.8
in GR-free simulations (Fig. 1b). GR weeds could also dete-
riorate biodiversity, e.g. species richness decreased by two
species when including GR weeds (Fig. 1c).
Correlations between weed-impact indicators
The principal component analysis between annual weed-
impact indicator values in simulations with GR weeds identi-
fied three groups (Fig. 2): (1) weed harmfulness indicators, (2)
species richness and functional biodiversity indicators (i.e.
weed-based food offer indicators) and (3) species equitability,
Fig. 2. There was no change when simulations were run with-
out GR weeds (section C.1.2 online). Weed harmfulness indi-
cators were all highly correlated (Pearson correlation coeffi-
cient >0.91, p < 0.0001, section C.1.1 online). Biodiversity
indicators were less correlated, with carabid food and bee food
13125
offer in ploughed field). c Slow initial decrease and later small decrease
with GR weeds (example of species richness in untilled field). d Very
slow initial increase and later huge increase with GR weeds (example of
field infestation in ploughed field). All are examples of early sown HT
maize monoculture in Aquitaine and means of ten weather repetitions
the most correlated (0.86, p < 0.0001). Because of the high
correlation among harmfulness indicators, only crop yield loss
was considered in the following sections. The other results can
be found in section C of supplementary material online. Bee
food results are also only presented online.
Indicator values in simulations with vs. without GR weeds
were also highly correlated, with Pearson correlation coeffi-
cients ranging from 0.57 each for bird food and species equi-
tability to 0.90 for carabid food and 0.91 for bee food offer
(p < 0.0001, section C.1.3 online). Whatever the weed-impact
indicator, values in simulations including GR weeds were
higher than in GR-free simulations when indicator values
were low in GR-free simulations (positive intercept values in
Table 1). For instance, when species richness was close to zero
in GR-free simulations, there were 3.8 and 5.4 additional spe-
cies in GR-including simulations in Aquitaine and Catalonia,
respectively. Generally, the difference in biodiversity between
13126 Environ Sci Pollut Res (2017) 24:13121–13135

were large in GR-free simulations, including GR weeds de-

creased weed impact. For instance, when species richness was
high (e.g. 22 species) in simulations without GR weeds, there
were four (3.84–0.34 × 22) and two species (5.45–0.36 × 22)
less in respectively Aquitaine and Catalonia after including
GR weeds.
In conclusion, GR weeds increased weed impact when this
impact was initially low; they decreased the impact when it
was initially high. This tendency was valid for all indicators
(i.e. significant intercept and regression values in Table 1) but
only explained a small part of the difference between simula-
tions with vs. without GR weeds. Indeed, the total explained
variability (R2) of the linear models of Table 1 was quite low
even though interactions with region and the effect of weather
repetitions were included.

Weed impact in cropping systems

Fig. 2 Principal component analysis of annual weed-impact indicators
simulated with FLORSYS with a 10−6 mutation rate in Aquitaine and
Catalonia, with 21–24 cropping system scenarios per region and 10
weather repetitions per scenario. Biodiversity indicators are in green
and harmfulness for crop production in red
simulations with vs. without GR weeds was highest in
Catalonia (largest intercept values) and the difference in harm-
fulness was largest in Aquitaine.
The more indicator values increased in GR-free simula-
tions, the more the additional effect due to GR weeds de-
creased (i.e. negative regression parameter in Table 1). The
decrease in additional biodiversity was strongest in Catalonia
(most negative regression parameter) and the decrease for
harmfulness was largest in Aquitaine. When indicator values
To make the effect of GR weeds more visible, only the eight
last simulated years were analysed here. Whatever the region
and the analysed weed-impact indicator, the effects of weather
repetition and time since simulation onset were either not sig-
nificant or negligible (partial R2 close to zero in Table 2). The
largest effect was due to the cropping system (largest partial
R2). In each region, variations were smallest for species rich-
ness which varied two- to threefold (e.g. from 10 species to
nearly 19 in Catalonia, lines 14 and 8 in Table 2—B) and
biggest for crop yield loss, which increased from −0.11% in
the late-sown maize monoculture (line 8) to 83% in the
unploughed diverse rotation in Catalonia (line 3).
Surprisingly, weeds sometimes increased crop yield, e.g. in
the early-sown unploughed maize monoculture in Aquitaine
(negative yield loss in line 10 in Table 2—A). There, weeds

Table 1 Linear regressions of

difference of annual weed-impact Weed-impact indicator Intercept Regression parameter R2
indicator values in simulations
with vs. without glyphosate- Aquitaine Catalonia Aquitaine Catalonia
resistant (GR) weeds as a function
of indicator values from GR-free Wild plant biodiversity
simulations, region, interaction Species richness 3.84 A 5.45 B −0.34 a −0.36 a 0.19
between both as well as
Species equitability 0.13 A 0.17 B −0.40 a −0.52 b 0.23
interaction between region and
weather repetition with PROC Functional biodiversity
GLM of SAS Bird food 0.47 A 0.93 B −0.08 a −0.16 b 0.09
Carabid food 0.33 A 0.32 A −0.09 a −0.11 a 0.06
Bee food 0.30 A 0.51 B −0.28 a −0.29 a 0.16
Harmfulness for crop production
Crop yield loss 5.48 A 4.13 A −0.32 a −0.18 b 0.15
Harvest pollution 0.34 A 0.26 A −0.26 a −0.15 b 0.13
Harvesting problems 0.37 A 0.38 A −0.24 a −0.16 b 0.12
Field infestation 0.26 A 0.17 B −0.42 a −0.20 b 0.21

Region × weather repetition was also significant at p = 0.05. Intercept values and regression parameters followed
by the same letters for a given indicator were not significantly different at p = 0.05
Environ Sci Pollut Res (2017) 24:13121–13135 13127

Table 2 Effect of weed flora and glyphosate-resistant (GR) weeds on
biodiversity and crop production in HT maize cropping systems during
the last eight simulated years evaluated with analyses of variance of weed
impact indicators as a function of simulated cropping system, presence vs.
absence of GR weeds, time and weather repetition. Mean indicator values
(Mean) per cropping system are coloured from red (lowest value of a
given column) to green (highest value) for biodiversity indicators and
vice versa for harmfulness indicators; values of a given column followed
by the same letter are not significantly different at p = 0.05. Differences
(Diff) between simulations with and without GR weeds are coloured in
red for deterioration (decreased biodiversity or improved harmfulness)
and green for improvement (vice versa); empty cells show non-
significant differences at p = 0.05

A. Aquitaine
Species
Species richness Bird food Carabid food Yield loss
Cropping system equitability
Mean Diff Mean Diff Mean Diff Mean Diff Mean Diff
1 Soya/Maize/Wheat/Maize 15.62 B 0.30 GH -0.061 6.01 C 2.96 C 41.53 CB -20.05
2 Soya/HTmaize/Wheat/HTmaize 16.49 A 0.29 H 6.09 C 3.05 C 43.62 B -10.05
3 + no mouldboard plough 9.34 H 0.37 F 3.28 F 2.30 D 19.76 D
4 Wheat/HTmaize 10.42 G 0.34 GF 0.92 G 2.02 E 20.30 D
5 + no mouldboard plough 14.16 C 0.24 I 0.059 6.58 B 4.66 B 60.16 A
6 HTmaize monoculture 11.17 F 0.12 J 5.52 D 1.88 E 1.08 E
7 + early sowing 10.56 GF 2.65 0.52 E -0.174 6.14 C 1.04 1.31 F 1.01 1.86 E 13.97
8 + late sowing 12.89 E 0.35 F 5.44 ED 2.05 E 0.26 E
9 + no mouldboard plough 8.09 I 0.70 B 5.25 E 0.59 G 0 E
10 + no plough + early sowing 6.32 J 0.76 A 5.17 E 0.01 I -7.20 F
11 + no till 13.78 DC -0.93 0.24 I 9.65 A 8.69 A 39.48 CB
12 + no till + 2nd glyphosate 14.05 DC 0.22 I 9.63 A 8.70 A 37.59 C
13 + no till + 2nd gly. + early sow. 13.48 DE -2.06 0.23 I 9.64 A 8.70 A 38.15 C
14 + catch crop killed with glyph. 4.37 K 0.59 D 0.04 H 0.26 H -0.12 E
15 + catch crop killed with tillage 4.84 K 0.65 C 0.06 H 0.76 G -0.28 E
Partial R²
Cropping system 0.58 0.57 0.82 0.89 0.43
Weather repetition 0.01 0.00 0.00 ns 0.00 ns 0.01
Time 0.01 0.00 ns 0.00 0.00 0.04
Presence of GR weeds 0.00 ns 0.00 0.00 ns 0.00 0.00 ns
Interaction system x GR weeds 0.01 0.01 0.00 ns 0.00 0.01

B. Catalonia
Species richness Species equitability Bird food Carabid food Yield loss
Cropping system Mean Diff Mean Diff Mean Diff Mean Diff Mean Diff
1 Wheat/Alfalfa/Maize 14.96 F 0.37 BC 3.62 G 4.47 C 72.65 B -17.63
2 Wheat/Alfalfa/HTmaize 13.44 G 0.37 BCD 3.13 H 4.06 D 56.13 DC
3 + no mouldboard plough 12.16 I 0.41 A 5.69 F 5.39 B 83.72 A
4 Wheat/HTmaize 18.1 BC 0.36 ECD 3.91 G 4.01 D 74.11 B -8.61
5 + no mouldboard plough 18.67 BA 0.4 BA 6.85 CD 5.6 B 59.97 C
6 HTmaize monoculture 15.6 E 0.36 ECD 0.10 7.18 CB -0.74 2.42 FE -0.52 7.77 F
7 + early sowing 14.53 F 0.32 F 7.31 B -0.51 2.01 G -0.79 2.48 G
8 + late sowing 18.74 A 0.16 H 5.68 F 2.65 E -0.11 G
9 + no mouldboard plough 12.71 HI 0.34 EF 0.054 6.53 ED 0.97 I -0.47 0.27 G
10 + no plough + early sow. 13.21 HG 0.26 G -0.084 6.2 E 0.94 I 0.02 G
11 + no till 16.48 D 0.37 BECD 9.93 A 8.89 A 51.31 D
12 + no till + 2nd glyphosate 18.02 C 0.37 ECD 9.65 A 8.64 A -0.49 41.34 E
13 + no till + 2nd glyph + early sow. 16.86 D -1.66 0.34 EFD -0.064 9.86 A 8.85 A 42.91 E
14 + catch crop killed with glyph. 10.74 J 0.83 0.36 ECD 1.22 I 1.69 H 0.68 G
15 + catch crop killed with tillage 16.68 D 0.12 I -0.048 5.43 F 2.33 F 0.72 G
Partial R²
Cropping system 0.42 0.23 0.71 0.81 0.60
Weather repetition 0.01 0.03 0.01 0.01 0.01
Time 0.03 0.00 ns 0.00 0.00 ns 0.03
Presence of GR weeds 0.00 ns 0.00 ns 0.00 ns 0.00 0.00
Interaction system x GR weeds 0.01 0.02 0.00 0.00 0.00
13128
delayed maize emergence by 2–5 days compared to simula-
tions excluding weeds, which protected the crop from late
frost in several weather repetitions (section C.4 online).
When looking in detail at the tested cropping systems, it
appeared that switching from conventional maize varieties
and herbicides to HT maize and glyphosate had no big effect
on weed-impact indicators (lines 2 vs. 1 in Table 2). Other
scenarios showed much bigger changes. For instance, both
biodiversity (except species equitability) and harmfulness in-
creased tremendously in no-till systems, irrespective of the
region, the glyphosate frequency or the sowing dates (lines
11–13 vs. 9). Conversely, introducing an additional crop (bar-
ley catch crop in Aquitaine, triticale cash crop in Catalonia)
before maize decreased biodiversity (except species equitabil-
ity) and harmfulness (lines 14–15 vs. 6). In Catalonia, the
deleterious effect of the additional crop on biodiversity was
attenuated or even cancelled out if tillage instead of glypho-
sate was used to clean the field after the additional crop (line
15 vs. 6).
The effect was less clear for other scenarios. For instance,
unploughed scenarios performed differently, depending on the
region, the rotation and the indicator. In Aquitaine, with the
exception of species equitability, biodiversity deteriorated in
the unploughed vs. ploughed diverse rotation (line 3 vs. 2 in
Table 2—A) and maize monoculture (9 vs. 6) but improved in
the unploughed vs. ploughed maize/wheat rotation (5 vs. 4).
The effect on harmfulness was similar: decrease in the diverse
rotation, no effect in maize monoculture, large increase in
maize/wheat.
Including GR weeds in the simulations had little consistent
effect. The impact of GR weeds depended very much on the
cropping system (only the interaction between the two factors
was significant in Table 2). It was negligible for most indica-
tors (partial R2 close to zero) and only significant in a few
cropping systems. In Aquitaine, early-sown maize monocul-
ture was the scenario where including GR weeds had the big-
gest effect (line 7 in Table 2—A), with an additional 14% of
yield loss, adding more than two species to species richness,
increasing trophic resources by +1 point, and decreasing spe-
cies equitability by 0.17. In other scenarios, the presence of
GR weeds decreased yield loss (e.g. diverse rotation, lines 1
and 2) or biodiversity (e.g. untilled, early-sown maize mono-
culture with two glyphosate sprayings, line 13). In Catalonia,
including GR weeds changed weed-impact indicators more
often, frequently decreasing biodiversity (particularly in maize
monocultures, e.g. lines 6,7, 9, 12 in Table 2—B) and, more
rarely, yield loss (e.g. conventional diverse rotation, line 1).
Effect of cultural practices
The variability in effects of changes in cultural practices (e.g.
no-plough) was not only due to interactions with rotation or
region but also to smaller modifications in cultural practices
Environ Sci Pollut Res (2017) 24:13121–13135
that accompanied the analysed major changes (e.g. increase in
tillage frequency and/or herbicide applications). These hidden
modifications were taken into account in the regression anal-
yses of weed impact indicators as a function of cultural prac-
tices in Table 3. Generally, biodiversity indicators were lower
in Aquitaine than in Catalonia (negative regression parameter
of line 3 in Table 3), except species equitability which was
larger. Weed harmfulness was also lower in Aquitaine than in
Catalonia.
Ten cultural practices influenced crop yield loss. Late
maize sowing increased weed harmfulness (line 5 in
Table 3), though a delay of nearly four weeks would be nec-
essary to result in and additional yield loss of 10% (i.e. divid-
ing 10 by the regression parameter 0.391 = 25.5 days). Adding
cover or cash crops (lines 7 and 8) that lasted less than
5.5 months (=84.5/15.21) increased harmfulness whereas
each additional month of cover subtracted 15% yield loss.
Each additional superficial tillage operation removed 10%
yield loss (line 9). Tillage was more effective if the first oper-
ation was delayed relative to previous crop harvest (line 17),
with an additional 1% yield-loss reduction for every 10 days
delay. Mouldboard ploughing reduced harmfulness even
more, particularly when carried out before winter crops (lines
12 and 13). The later the field was ploughed relative to previ-
ous harvest, the more yield loss was reduced (line 20) because
more time was left for superficial tillage. Glyphosate treat-
ments reduced harmfulness (lines 22 and 23), particularly in-
crop applications which subtracted 14% yield loss.
Harmfulness though increased when herbicides were delayed
(line 21).
The effect of cultural practices on biodiversity indicators
mostly depended on the same factors than their effect on
harmfulness indicators, but with varying effects. For instance,
late maize sowing increased species richness and carabid food
but decreased species equitability and bird food (line 5).
Generally, tillage decreased biodiversity (lines 9, 16) though
the effect greatly depended on timing (e.g. lines 16 vs. 14 and
lines 17, 19 and 20). For instance, early post-harvest tillage
was better for carabid food but worse for bird food (line 17).
Indeed, late tillage leaves seeds longer on soil surface during
the critical bird-feeding period (Oct. to March); early tillage
leaves less time for seeds to be exposed to rain before burial,
which makes seeds more dormant and thus persistent,
delaying weed emergence and seed production until the next
carabid-feeding period (April–Sept.). In contrast to harmful-
ness, fallow glyphosate applications decreased food-offer in-
dicators more than in-crop glyphosate (lines 22 and 23). The
effect on plant biodiversity was different: species richness and
species equitability increased when the broad-spectrum glyph-
osate was used respectively during fallow and in crops instead
of selective products. Generally, the later herbicides were ap-
plied to crops, the better for plant biodiversity and carabid-
food offer (line 21).
Environ Sci Pollut Res (2017) 24:13121–13135 13129

Table 3 Effect of cultural practices, in interaction with the presence of Catalonia. Regression parameters estimated with linear models using for-
glyphosate-resistant (GR) weeds, on weed-impact indicators averaged ward selection and cross-validation. All variables kept in the final model
over the last 8 years simulated with F LOR S YS in Aquitaine and were significant at p = 0.05. Empty cells show non-significant effects
Crop
Cropping system descriptor Species Species Bird Carabid yield
(average over rotation) Mean [min, max] richness equitability food food loss
[0,22]§ [0.04,0.79] [0,9.8] [0, 8.8] [-4,99]
1 R-Square 0.504 0.192 0.644 0.751 0.444
2 Intercept 16.900 2.187 36.122 -1.937 -112.8

3 Region is Aquitaine (vs. Catalonia) {yes, no} -4.613 0.100 -0.629 -1.371 -16.73
4 Proportion of winter crops 0.16 [0, 0.50]
5 Maize sowing date (Julian days#) 123 [101, 152] 0.002 -0.00405 -0.053 0.024 0.344
6 With GR weeds -0.00410
7 Additional crop before maize {yes, no} -13.76 0.246 -8.277 95.881
8 Duration additional crop (if any) $ 5.7 months [5, 7] 2.014 -0.021 1.402 -16.89
9 Superficial tillage operations (other than roll) 1.7 [0, 3] -1.998 0.031 -2.401 -2.309 -11.50
10 Superficial tillage operations from April to Sept. 0.30 [0,2] nt.
11 Superficial tillage operations from Oct. to March 1.05 [0,2] nt.
12 Mouldboard plough before spring crops 0.36 [0, 1] nt. nt. nt. nt. .
13 Mouldboard plough before winter crops$ 0.10 [0, 1] nt. nt. nt. nt. -13.32
14 Summer ploughing (April to Sept.) 0.2 [0, 1] 3.658 -0.164 0.336 0.795
15 With GR weeds -0.124
16 Winter ploughing (Oct. to March) 0.2 [0, 1] -0.648 -2.453 -2.014
17 Time from harvest to 1st Tillage (if any)$ 102 days [1, 212] 0.002 0.005 -0.015 -0.081
18 With GR weeds 0.004
19 Time from last tillage (if any) to cash crop sowing$ 7 days [1, 19]
20 Time from harvest to ploughing (if any) $ 140 days [6, 216] -0.008 0.000 -0.003 -0.098
21 Time from cash crop sowing to 1st herbicide 37 [1, 104] 0.046 0.000 -0.004 0.010 0.595
22 Glyphosate during fallow 0.19 [0,1] 0.618 -0.095 -1.363 -0.770 -4.320
23 Glyphosate after HT maize sowing 0.54 [0,1] -1.496 0.075 -0.126 -16.94
$
Variable tested in interaction with another (presence or absence of additional crop, tillage or ploughing)
§
5 and 95% percentiles of indicator values

Few of these effects depended on the presence of GR
weeds in the simulations, i.e. only three interactions between
cultural practices and GR-weed presence were significant in
Table 3 and all three concerned plant biodiversity. Late maize
sowing decreased species equitability more when GR weeds
were included in the simulations (line 6) whereas the summer
ploughing was less deleterious when including GR weeds
(line 15). Delaying the first post-harvest tillage was more ben-
eficial for species richness in simulations with vs. without GR
weeds (line 18).
Which species and traits determine weed impact?
Similarly to weed-impact indicators, total weed densities
greatly varied among cropping systems, from less than
0.01 plants/m2 averaged over all simulated years and weather
repetitions in the unploughed early-sown maize monoculture,
to 3467 plants/m2 in the untilled monoculture in Aquitaine
(section C.3.1). Differences between simulations with vs.
without GR weeds varied comparatively little, with densities
divided by 4 in the Catalonia early-sown maize monoculture
and multiplied by more than 11 in the Aquitaine equivalent.
Differences between simulations with vs. without GR weeds
also depended on weed species. In Aquitaine, the densities of
most weed species decreased when including GR weeds, with
decrease ranging from a factor 2 (G. aparine) to 7
(D. sanguinalis) in simulations with vs. without GR weeds.
Two species were more abundant when including GR weeds,
i.e. C. album and S. asper which increased sixfold (section
C.3.2 online). In Catalonia, only S. asper was more abundant
in GR-including simulations (+10%); all other species de-
creased in simulations when GR weeds were included, with
13130 Environ Sci Pollut Res (2017) 24:13121–13135

C. album (−34%) and G. aparine (−49%) decreasing the least
and D. sanguinalis the most (−97%). The species whose den-
sities increased most when including GR weeds where those
with a high seed area/mass ratio (regression parameter = 0.42,
p = 0.001) and a large specific plant width (0.00165,
p = 0.0007, R2 = 0.36, section C.4 online), i.e. species with
fast-germinating seeds and mostly investing biomass into
plant width.
Species traits most pertinent for understanding weed-
impact indicators are described by Pearson correlation
coefficients determined via an RLQ analysis (Table 4). The
two traits that we are most interested in are population fitness
cost for resistance and glyphosate resistance. The latter slight-
ly reduced species equitability (Pearson correlation coefficient
of −0.16 in line 22 in Table 4) whereas the effect of fitness cost
was not significant (results not shown). Weed impact on crop
production and biodiversity almost entirely depended on other
species traits, none of which were correlated to glyphosate
resistance in the 25 species used in the present simulations,
except monocot status (Fig. 3)

Table 4 Relationships between weed species traits or characteristics
and weed-impact indicators averaged over the last eight simulated years
in simulations with and without GR weeds, identified by fourth-corner
analyses preceded by the RLQ analysis. Pearson correlation coefficients r
between indicators and traits, and tests of the null hypothesis that species
are distributed independently of their preferences for scenarios and of
their traits (highest p values of permutation models permuting scenarios
or species). Only species traits significantly correlated to at least one
indicator at p = 0.05 are listed here

Biodiversity indicators Harmfulness indicators

Crop
Species trait Species Species Bird Harvest
yield
richness equitability food pollution
loss
1 Monocot species (instead of dicot) -0.15
2 Seed mass 0.36 -0.80
3 Seed lipid content 0.22
4 Seed area/mass 0.24 -0.26
Emergence season#
5 Sept 0.15 -0.19
6 Oct-Nov 0.29 0.27
7 Dec-Feb -0.82 0.28
8 April -0.40
9 May-June -0.29 -0.27
10 Emergence season (length) 0.25 0.29
11 Base temperature 0.11
12 Specific leaf area 0.46 0.67 0.60
13 Leaf biomass ratio -0.21
14 Climbing species (yes, no) 0.38 -0.73 0.71 0.61
15 Maximum plant height 0.70 0.63
16 Maximum plant width 0.24
Shading sensitivity§
17 Leaf biomass ratio 0.28
18 Specific leaf area -0.34 0.79 -0.39 -0.28
19 Specific plant width$ 0.74 -0.34
20 Specific plant height$ 0.14
21 Harvest index& 0.26 -0.80 0.28
22 Glyphosate resistance -0.16
§
A positive shading sensitivity parameter indicates that the associated species parameter increases with shading intensity
$
Specific plant height (or width) is plant height (or width) divided by plant biomass
&
Ratio of mature seed biomass vs total above-ground biomass
Environ Sci Pollut Res (2017) 24:13121–13135
Fig. 3. Principal component analysis of species traits of the 25 weed
species included in FLORSYS. Red boxes show traits highly correlated to
crop yield loss (based on the results of Table 4) and the potential of
glyphosate resistance of species. SLA is specific leaf area, HM and WM
specific plant height and widths, LBR leaf biomass ratio, and shadeTRAIT
is the sensitivity of a species trait to shading
The species that were most harmful for crop production in
the simulations were efficient in producing large leaf areas
(positive correlation with specific leaf area in line 12 in
Table 4) without investing too much biomass (negative corre-
lation with leaf biomass ratio in line 13); they were often
climbing species (line 14), potentially tall (line 15) and
invested more biomass into seed production (line 21). When
shaded by neighbouring plants (shading sensitivity parame-
ters, lines 17–20), the most harmful species were those that
did not reduce their leaf area (negative correlation with shad-
ing sensitivity of specific leaf area in 18) or plant width (line
19). The effect of emergence-related traits depended on the
analysed harmfulness indicators. Crop yield loss was in-
creased by species with a low pre-emergent seedling loss
(i.e. dicotyledonous vs. monocotyledonous species, line 1)
and a large potential emergence season (line 10), i.e. species
that can emerge in winter and spring crops. Harvest pollution,
which is an issue in wheat but less so in maize (at least with the
here simulated cropping systems and weed species), increased
with species emerging in late autumn and early winter (lines
6–7), i.e. after winter wheat was sown, and decreased with
those emerging in late spring (line 9), i.e. after maize sowing.
The effect of species traits on biodiversity indicators
depended on the analysed indicator. Species equitability
responded similarly to harmfulness. Species richness was in-
creased by fast-emerging species (positive correlation with
seed lipid content and seed area/mass ratio, lines 3–4) that
were able to increase their leaf biomass when shaded by
neighbouring plants (shading sensitivity of leaf biomass ratio,
13131
line 17). The response of food-offer indicators such as bird-
food offer usually were the opposite of the harmfulness re-
sponse, with a particular detrimental effects of traits leading to
seed disappearance at the onset of the feeding the period. For
instance, bird food offer (which is crucial during October to
March) was greatly decreased if the species emergence season
included December to February (lines 5, 7) but increased if the
species temperature was high (line 11), thus hampering winter
germination.
Discussion
The present paper (1) quantified how much the presence of
GR weeds contributed to weed impacts on crop production
and biodiversity, (2) determined the effect of cultural practices
on the impact of GR weeds and (3) identified which species
traits most influence weed-impact indicators and whether
these are included in GR weeds.
Implications for managing glyphosate resistance
The consequences of glyphosate-tolerant maize and of the
accompanying changes in cultural practices for weed impacts
on biodiversity and crop production have already been
discussed in a previous paper (Bürger et al. 2015). Here, we
will focus on the impact of glyphosate resistance in weeds.
The first major result of the present study was the demon-
stration that glyphosate resistance only rarely affected the var-
iables of crop production and biodiversity that we studied here,
during the 28 years after switching to HT maize, in the partic-
ular case of the maize-based cropping systems tested here and
with herbicide-resistance model used here (see the following
section for a discussion of the limits of the approach). The
biggest effects were due to cultural practices and region. This
is consistent with field surveys reporting that in-crop glyphosate
treatments had no effect (Schwartz et al. 2015) or improved
(Young et al. 2013) plant biodiversity which mostly depended
on geographical location (Schwartz et al. 2015; Young et al.
2013). Here, including GR weeds in the simulations increased
weed impact only when this impact was originally low; when
weed impact was already high, the GR weeds had no more
effect or weed impact was lower than when disregarding GR
weeds. This applied even when GR densities were quite high as
in no-till systems (see results in Colbach et al. 2017). In the
systems tested here, GR populations tended to replace
glyphosate-susceptible populations and/or species without in-
creasing weed densities, particularly when the habitat-carrying
capacity was reached as in the no-till systems.
The second key result was the identification of the weed
species traits that were responsible for the biggest crop yield
loss in the tested cropping systems, i.e. (1) traits that allow
species to avoid preventive tillage-based techniques via higher
13132 Environ Sci Pollut Res (2017) 24:13121–13135

seed persistence (e.g. smaller seed area/mass) when these
techniques are applied, (2) traits that determine plant ability
to compete with crop plants for light by maximizing space
occupation and shade avoidance and (3) traits that optimize
conversion of light into seed production via a high specific
leaf area and harvest index (part of above-ground plant bio-
mass attributed to seed production). The analysis of the cor-
relation among species traits showed that none of the poten-
tially GR species simulated here present these traits. If a spe-
cies would, however, combine glyphosate resistance with the
traits that make weeds more competitive toward crops, crop
production would probably be much more affected by GR
weeds than in the present simulations.
The same conclusions apply for weed-related functional
biodiversity, i.e. including GR weeds only has an occasional
and small effect. This remains true even in those cases where
plant biodiversity is affected. In other words, the weed com-
munity composition changes but the trait composition remains
similar. Again, this conclusion only applies with the weed
community used here where none of the species combined
glyphosate resistance with traits that are beneficial or detri-
mental for functional biodiversity.
Consequently, weed management should not focus only on
resistant weeds as such, but on species presenting the trait
combinations that are most harmful for crop production and/
or beneficial for biodiversity. These are synthesized in Table 5.
This advice is consistent with the conclusions of the compan-
ion paper (Colbach et al. 2017) which, in common with many
papers on herbicide resistance (Colbach et al. 2016b; Moss
and Clarke 1994; Renton and Flower 2015; Vencill et al.

Table 5 Synthetic advice table for managing weed impact in maize-based cropping systems in Aquitaine and Catalonia, based on the present results (in
bold) and the companion paper (in italic; Colbach et al. 2017)

Cropping system component Advice in order to

Control harmfulness Promote biodiversity (species richness,

trophic resources)

% of winter crops Reduce

Cash crop sowing date Delay
Maize sowing date Do not delay Delay except for carabid food
Additional crop before maize ≥6 months but leave time for tillage ≥6 months
No till Avoid Yes
Superficial tillage
Frequency High Low
Timing Before weeds become big, during non
dormancy of most harmful species
1st post-harvest tillage Delay Delay except for carabid food
Last pre-sowing tillage With sowing
Mouldboard plough Summer only
Before maize Yes
Before winter crops Yes
Efficiency Highest if short-living and superficially
emerging seeds
Risk Leaves the most competitive species
Timing Delay to leave time for surface seeds Early
to germinate
Glyphosate Efficient (as long as no resistance), Fallow spraying for species richness,
better in crops than during fallow none otherwise
Risk Selects glyphosate-resistant weeds if no till
Timing of herbicides Do not delay Delay except for bird food
Target weed traits Voluminous plants$ and high efficiency#, Small seeds, no emergence in late autumn/early
even when shaded&, generalist§ in winter, slowly growing plants, large plant
diverse rotation and autumn-emerging area and leaf area when shaded
if many winter crops in rotation
$
Climbing, potentially tall
&
Do not reduce plant width and specific leaf area when shaded
#
High specific leaf area with low biomass ratio and high harvest index
§
Long potential emergence season
Environ Sci Pollut Res (2017) 24:13121–13135
2012; WRAG 2015), concludes that the practices for avoiding
herbicide-resistant weeds are the same as those best for con-
trolling weeds as such, and that many of these practices are
also efficient for controlling herbicide-resistant weeds. The
novelty of our study was to detect the traits of the weed species
that should be targeted, depending on the cropping system
objective in terms of ecosystem services, and to identify the
practices targeting these different traits.
Limits of the present results
The limits of the glyphosate-resistance submodel and the sim-
ulation plan have already been discussed in the companion
paper (Colbach et al. 2017). Here, we will focus on weed
harmfulness and contribution to biodiversity.
We already started to discuss the limits of the FLORSYS
model for predicting weed impact on biodiversity and crop
production in previous simulation studies (Bürger et al.
2015; Mezière et al. 2015a). The comparison of the present
results based on 25 weed species with our previous study
testing the same cropping systems with only 16 species
(Bürger et al. 2015) shows the limits of the present approach.
The same tendencies were observed and similar conclusions
were reached in both studies but the amplitude of the effects
changed. In the present study, rotation was less efficient to
reduce weed harmfulness, and biodiversity was less affected
by changes in cultural practices. This is consistent with the use
of additional species here, leading to a larger range of species
traits and covering more diverse niches (e.g. summer-
emerging weed species) in the simulated fields. Further simu-
lation studies with more diverse cropping systems and weed
floras that differ in terms of species numbers and diversity will
be necessary to draw more generic conclusions on the effects
of cultural practices, glyphosate resistance and weed impact
on crop production and biodiversity.
None of the cropping systems tested here reconciled
low weed harmfulness with high biodiversity contribution,
and several biodiversity indicators were correlated to
harmfulness indicators. Moreover, all harmfulness indica-
tors were highly correlated, indicating that the cropping
systems tailored to target a given harmfulness issue would
also control all other harmfulness issues (or vice versa).
This was not true in another simulation study investigating
more diverse cropping systems in two other French re-
gions (Mezière et al. 2015a). These authors found
cropping systems that reconciled crop production and bio-
diversity, and they observed a diversity of biodiversity and
harmfulness profiles (e.g. systems with high yield loss and
low harvest contamination). This was not surprising inso-
far as these authors worked with diverse and longer rota-
tions (including multiannual grasslands) and more varied
management techniques, including mechanical weeding.
13133
Implications for other types of herbicide resistance
The adoption of stacked GM crops tolerant to both glyphosate
and a synthetic auxin (2,4-D or dicamba) will dramatically
increase the use of dicamba and 2,4-D. The risk that new
resistances evolve or already evolved resistances to these her-
bicides expand is high. Indeed, resistance has currently been
reported in several tens of species (Heap 2016). As was re-
cently discovered, resistance to synthetic auxins can be caused
by mutations within auxin transporter genes (Goggin et al.
2016). The genetic mechanism of resistance is thus similar
to the one considered in our model for glyphosate.
Moreover, as glyphosate, dicamba is applied post-emergence.
Consequently, the simulations presented here could be extend-
ed to model the evolution of resistance to synthetic auxins,
alone or after a first evolution of resistance to glyphosate, or
for modelling the simultaneous evolution of both resistances.
Conclusion
The present simulation study with maize-based cropping sys-
tems showed that glyphosate resistance did not notably
change the impact of weeds on biodiversity and crop produc-
tion during the first 28 years after introducing HT crops and
that weed impact almost entirely depended on cultural prac-
tices. This though only applies as long as no weed species
combines glyphosate resistance with the traits that were
shown to cause the greatest yield or biodiversity losses. The
novel conclusions drawn here were made possible by combin-
ing three approaches, i.e. a weed dynamics model driven by
cultural practices and weather, a genetic mutation and heredity
model, and a series of indicators of weed impact on crop
production and biodiversity. As a result, a generic advice table
was proposed, on how to optimize cultural practices depend-
ing on the planned production and biodiversity objectives, and
in which conditions their application is needed. This advice
applies to maize-based systems; further simulations are nec-
essary to establish generic rules and/or advice specific to dif-
ferent regions and cropping-system types.
Acknowledgements
This project is supported by INRA, the European project
AMIGA (Assessing and Monitoring Impacts of Genetically
modified plants on Agro-ecosystems, FP7-KBBE-
2011-5-CP-CSA), the French project CoSAC (ANR-14-
CE18-0007) and the research programme BAssessing and re-
ducing environmental risks from plant protection products^
funded by the French Ministries in charge of Ecology and
Agriculture.
13134
References
Beckie HJ (2009) Herbicide resistance in weeds: influence of farm prac-
tices. Prairie Soils & Crops Journal 2:17–23
Bigler F, Albajes R (2011) Indirect effects of genetically modified herbi-
cide tolerant crops on biodiversity and ecosystem services: the bio-
logical control example. J Verbr Lebensm 6:79–84. doi:10.1007/
s00003-011-0688-1
Boerboom CM (1999) Non chemical options for delaying weed resis-
tance to herbicides in Midwest cropping systems. Weed Technol
13:636–642
Bonny S (2016) Genetically modified herbicide-tolerant crops, weeds,
and herbicides: overview and impact. Environmental Management
57:31–48. doi:10.1007/s00267-015-0589-7
Bretagnolle V, Gaba S (2015) Weeds for bees? A review. Agron Sustain
Dev 35:891–909
Brookes G, Barfoot P (2009) GM crops: global socio-economic and
environmental impacts 1996–2007
Bürger J, Granger S, Guyot SHM, Messean A, N (2015) Simulation study
of the impact of changed cropping practices in conventional and GM
maize on weeds and associated biodiversity. Agric Syst 137:51–63.
doi:10.1016/j.agsy.2015.03.009
Cavan G, Cussans J, Moss SR (2000) Modelling different cultivation and
herbicide strategies for their effect on herbicide resistance in
Alopecurus myosuroides. Weed Res 40:561–568
Chauvel B, Guillemin JP, Colbach N (2009) Evolution of a herbicide-
resistant population of Alopecurus myosuroides Huds. In a long-
term cropping system experiment. Crop Prot 28:343–349
Chessel D, Dufour AB, Thioulouse J (2004) The ade4 package. I. One-
table method. R News 4(4):5–10
Colbach N et al (2014a) The role of models for multicriteria evaluation
and multiobjective design of cropping systems for managing weeds.
Weed Res 54:541–555. doi:10.1111/wre.12112
Colbach N, Busset H, Roger-Estrade J, Caneill J (2014b) Predictive
modelling of weed seed movement in response to superficial tillage
tools. Soil Tillage Res 138:1–8
Colbach N, Collard A, Guyot SHM, Mezière D, Munier-Jolain NM
(2014c) Assessing innovative sowing patterns for integrated weed
management with a 3D crop:weed competition model. Eur J Agron
53:74–89. doi:10.1016/j.eja.2013.09.019
Colbach N et al (2016a) Uncertainty analysis and evaluation of a com-
plex, multi-specific weed dynamics model with diverse and incom-
plete data sets. Environ Model Softw 86:184–203. doi:10.1016/j.
envsoft.2016.09.020
Colbach N, Chauvel B, Darmency H, Delye C, Le Corre V (2016b)
Choosing the best cropping systems to target pleiotropic effects
when managing single-gene herbicide resistance in grass weeds. A
blackgrass simulation study. Pest Manag Sci 72:1910–1925. doi:10.
1002/ps.4230
Colbach N, Fernier A, Le Corre V, Messean A, Darmency H (2017)
Simulating changes in cropping practices in conventional and
glyphosate-resistant maize. I. Effect on weeds. Environmental
Science and Pollution Research. doi:10.1007/s11356-017-8591-7
Collavo A, Sattin M (2014) First glyphosate-resistant Lolium spp. bio-
types found in a European annual arable cropping system also af-
fected by ACCase and ALS resistance Weed Res 54:325–334 doi:
10.1111/wre.12082
Friesen LJS, Ferguson GM, Hall JC (2000) Management strategies for
attenuating herbicide resistance: untoward consequences of their
promotion. Crop Prot 19:891–895. doi:10.1016/S0261-2194(00)
00116-2
Gardarin A, Dürr C, Colbach N (2012) Modeling the dynamics and
emergence of a multispecies weed seed bank with species traits.
Ecol Model 240:123–138. doi:10.1016/j.ecolmodel.2012.05.004
Environ Sci Pollut Res (2017) 24:13121–13135
Goggin DE, Cawthray GR, Powles SB (2016) 2,4-D resistance in wild
radish: reduced herbicide translocation via inhibition of cellular
transport. J Exp Bot 67:3223–3235
Gressel J, Segel LA (1990) Modelling the effectiveness of herbicide ro-
tations and mixtures as strategies to delay or preclude resistance.
Weed Technol 4:186–198
Heap I (2016) The international survey of herbicide resistant weeds.
Available www.weedscience.com. Internet.
Heard MS et al (2003) Weeds in fields with contrasting conventional and
genetically modified herbicide-tolerant crops. I. Effects on abun-
dance and diversity. Philosophical Transactions of the Royal
Society of London Series B-Biological Sciences 358:1819–1832
James C (2013) Global Status of Commercialized Biotech/GM Crops:
2013. http://www.isaaa.org/resources/publications/briefs/46/default.
asp. Accessed 26/09/2014
Lievin J, Waller F, Duroueix F, BONIN L, Quillot E, Rodriquez A (2013)
R-sim: un outil web qui evalue le risque de developpement de resis-
tances aux herbicides. Paper presented at the AFPP—22e
Conference du COLUMA, Journees internationales sur la lutte
contre les mauvaises herbes, Dijon France, 10–12 Decembre 2013
Maxwell BD, Roush ML, Radosevich SR (1990) Predicting the evolution
and dynamics of herbicide resistance in weed populations. Weed
Technol 4:2–13
Mezière D, Colbach N, Dessaint F, Granger S (2015a) Which cropping
systems to reconcile weed-related biodiversity and crop production
in arable crops? An approach with simulation-based indicators. Eur
J Agron 68:22–37. doi:10.1016/j.eja.2015.04.004
Mezière D, Petit S, Granger S, Biju-Duval L, Colbach N (2015b)
Developing a set of simulation-based indicators to assess harmful-
ness and contribution to biodiversity of weed communities in
cropping systems. Ecol Indic 48:157–170. doi:10.1016/j.ecolind.
2014.07.028
Moss SR, Clarke JH (1994) Guidelines for the prevention and control of
herbicide-resistant black-grass (Alopecurus myosuroides Huds).
Crop Prot 13:230–234
Moss SR, Perryman SAM, Tatnell LV (2007) Managing herbicide-
resistant blackgrass (Alopecurus myosuroides): theory and practice.
Weed Technol 21:300–309. doi:10.1614/wt-06-087.1
Munier-Jolain NM, Guyot SHM, Colbach N (2013) A 3D model for light
interception in heterogeneous crop:weed canopies. Model structure
and evaluation. Ecol Model 250:101–110. doi:10.1016/j.ecolmodel.
2012.10.023
Munier-Jolain NM, Collard A, Busset H, SHM G, Colbach N (2014)
Modelling the morphological plasticity of weeds in multi-specific
canopies. Field Crop Res 155:90–98. doi:10.1016/j.fcr.2013.09.018
Neve P, Diggle AJ, Smith FP, Powles SB (2003) Simulating evolution of
glyphosate resistance in Lolium rigidum I: population biology of a
rare resistance trait. Weed Res 43:404–417
Powles SB (2008) Evolution in action: glyphosate-resistant weeds threat-
en world crops. Outlooks on Pest Management 19:256–259. doi:10.
1564/19dec07
Renton M, Flower KC (2015) Occasional mouldboard ploughing slows
evolution of resistance and reduces long-term weed populations in
no-till systems. Agric Syst 139:66–75. doi:10.1016/j.agsy.2015.06.
005
Renton M, Busi R, Neve P, Thornby D, Vila-Aiub M (2014) Herbicide
resistance modelling: past, present and future. Pest Manag Sci 70:
1394–1404. doi:10.1002/ps.3773
Sammons RD, Gaines TA (2014) Glyphosate resistance: state of knowl-
edge. Pest Manag Sci 70:1367–1377. doi:10.1002/ps.3743
Schwartz LM et al (2015) Seedbank and field emergence of weeds in
glyphosate-resistant cropping Systems in the United States. Weed
Sci 63:425–439. doi:10.1614/ws-d-14-00089.1
Shaner DL (2000) The impact of glyphosate-tolerant crops on the use of
other herbicides and on resistance management. Pest Manag Sci 56:
320–326
Environ Sci Pollut Res (2017) 24:13121–13135
Taylor R, Maxwell B, Boik R (2006) Indirect effects of herbicides on bird
food resources and beneficial arthropods. Agric Ecosyst Environ
116:157–164
Vencill WK et al (2012) Herbicide resistance: toward an understanding of
resistance development and the impact of herbicide-resistant crops.
Weed Sci 60:2–30. doi:10.1614/ws-d-11-00206.1
WRAG (2015) Guidleines for minimizing the risk of glyphosate resis-
tance in the UK. http://webarchive.nationalarchives.gov.uk/
13135
20151023155227/http://www.pesticides.gov.uk/Resources/CRD/
Migrated-Resources/Documents/W/WRAG_Glyphosate_
resistance_guidelines_June_2015.pdf. doi:http://webarchive.
nationalarchives.gov.uk/20151023155227/http://www.pesticides.
gov.uk/Resources/CRD/Migrated-Resources/Documents/W/
WRAG_Glyphosate_resistance_guidelines_June_2015.pdf
Young BG et al (2013) Agricultural weeds in glyphosate-resistant
cropping systems in the United States. Weed Sci 61:85–97