Predicted Soybean Yield Loss As Affected by Emergence Time of Mixed-Species

Weed Communities
Author(s): Mark R. Jeschke, David E. Stoltenberg, George O. Kegode, Christy L. Sprague, Stevan Z.
Knezevic, Shawn M. Hock, and Gregg A. Johnson
Source: Weed Science, 59(3):416-423.
Published By: Weed Science Society of America
DOI: http://dx.doi.org/10.1614/WS-D-10-00129.1
URL: http://www.bioone.org/doi/full/10.1614/WS-D-10-00129.1

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, and
environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published
by nonprofit societies, associations, museums, institutions, and presses.
Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of
BioOne’s Terms of Use, available at www.bioone.org/page/terms_of_use.
Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries
or rights and permissions requests should be directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research
libraries, and research funders in the common goal of maximizing access to critical research.
Weed Science 2011 59:416–423

Predicted Soybean Yield Loss As Affected by Emergence Time of

Mixed-Species Weed Communities
Mark R. Jeschke, David E. Stoltenberg, George O. Kegode, Christy L. Sprague, Stevan Z. Knezevic, Shawn M. Hock, and
Gregg A. Johnson*
Potential crop yield loss due to early-season weed competition is an important risk associated with postemergence weed
management programs. WeedSOFT is a weed management decision support system that has the potential to greatly reduce
such risk. Previous research has shown that weed emergence time can greatly affect the accuracy of corn yield loss
predictions by WeedSOFT, but our understanding of its predictive accuracy for soybean yield loss as affected by weed
emergence time is limited. We conducted experiments at several sites across the Midwestern United States to assess
accuracy of WeedSOFT predictions of soybean yield loss associated with mixed-species weed communities established at
emergence (VE), cotyledon (VC), first-node (V1), or third-node (V3) soybean. Weed communities across research sites
consisted mostly of annual grass species and moderately competitive annual broadleaf species. Soybean yield loss occurred
in seven of nine site-years for weed communities established at VE soybean, four site-years for weed communities
established at VC soybean, and one site-year for weed communities established at V1 soybean. No soybean yield loss was
associated with weed communities established at the V3 stage. Nonlinear regression analyses of predicted and observed
soybean yield data pooled over site-years showed that predicted yields were less than observed yields at all soybean growth
stages, indicating overestimation of soybean yield loss. Pearson correlation analyses indicated that yield loss functions
overestimated the competitive ability of high densities of giant and yellow foxtail with soybean, indicating that adjustments
to competitive index values or yield loss function parameters for these species may improve soybean yield loss prediction
accuracy and increase the usefulness of WeedSOFT as a weed management decision support system.
Nomenclature: Giant foxtail, Setaria faberi Herrm., SETFA; yellow foxtail, Setaria pumila (Poir.) Roemer & J.A.
Schultes, SETLU; corn, Zea mays L.; soybean, Glycine max (L.) Merr.
Key words: Bioeconomic model, competitive index, decision support system, integrated weed management,
WeedSOFT.

Glyphosate-resistant corn and soybean cropping systems Such findings suggest an important role for weed manage-
have improved weed management compared with the tactics ment decision support systems in selecting optimal tactics for
used before the adoption of glyphosate-resistant technology protecting crop yield and economic return in glyphosate-
(Kruger et al. 2009). However, potential corn and soybean resistant cropping systems. Furthermore, the long-term success
yield loss due to early-season weed competition is an important of managing problematic weeds in glyphosate-resistant cropping
risk associated with POST glyphosate programs. Previous systems is expected to require the development of multifaceted
research across the North Central Region found that the integrated weed management programs that include glyphosate
optimum timing for initial glyphosate application to avoid corn as well as other weed management tactics (Kruger et al. 2009).
grain yield loss was when weeds were less than 10 cm in height WeedSOFT is a weed management decision support system that
and when corn growth was not more advanced than the fourth- has the potential to identify optimal management tactics by
leaf vegetative stage (V4; Ritchie et al. 1997a) (Gower et al. utilizing information on weed density, weed and crop growth
2003). In the northeastern United States, a study conducted stages, treatment efficacy, and environmental conditions to
across 10 site-years found that at high weed densities, the predict crop yield loss (Neeser et al. 2004). Weed species are
duration of weed competition preceding glyphosate application assigned a competitive index (CI) value ranging from 0 to 10,
at eighth-leaf (V8) corn reduced yield on average by 15% with 10 representing the highest level of competitiveness.
(Myers et al. 2005). In Wisconsin, glyphosate-resistant soybean Competitive index modifier (CIM) values adjust the CI based
yield losses associated with early-season weed competition on relative weed and crop growth stages to determine an
averaged 9.3 and 3.1% across 30 fields in 2008 and 40 fields in adjusted competitive index (ACI). The products of the ACI
2009, respectively (Fickett et al. 2009). values and weed densities are summed to yield the total
competitive load (TCL). Therefore, WeedSOFT can account
for the effect of relative emergence time within a crop–weed
DOI: 10.1614/WS-D-10-00129.1 community on crop yield loss.
* First and second authors: Graduate Research Assistant and Professor, Several studies have addressed the accuracy of WeedSOFT
Department of Agronomy, 1575 Linden Drive, University of Wisconsin,
Madison, WI 53706; third author: Assistant Professor, Department of Plant yield loss predictions. In a study conducted across the north
Sciences, North Dakota State University, Fargo, ND 58105; fourth author: central states, Schmidt et al. (2005) reported similar observed
Associate Professor, Department of Crop and Soil Sciences, Michigan State and predicted corn-yield losses over 21 site-years, but
University, East Lansing, MI 48824; fifth and sixth authors: Associate Professor variability of predictions occurred at sites with a high density
and Graduate Research Assistant, Agronomy and Horticulture, University of
Nebraska, Haskell Agricultural Laboratory, Concord, NE 68728-2828; seventh of a single weed species regardless of CI, at sites with a
author: Associate Professor, Department of Agronomy and Plant Genetics, predominant broadleaf species with a CI greater than 5, and
University of Minnesota, Southern Research and Outreach Center, Waseca, MN at sites with moderate to severe drought stress. At sites in
56093. Current address of first author: Pioneer Hi-Bred International, Inc., 7300 Nebraska, Missouri, and Illinois, Hock et al. (2007) found
NW 62nd Ave, P.O. Box 1004, Johnston, IA 50131-1004; current address of
third author: Department of Agriculture, 800 University Drive, Northwest
that WeedSOFT provided better weed control predictions for
Missouri State University, Maryville, MO 64468-6001. Corresponding author’s broadleaf than grass weed species in corn, but also that the
E-mail: [email protected] model provided excellent herbicide-efficacy predictions for

416 N Weed Science 59, July–September 2011

Table 1. Soil characteristics, soybean variety, and planting date for research sites in 2004 and 2005.
Site Year Soil type pH Organic matter Variety Planting date
%
East Lansing, MI 2004 Capac sandy clay loam 7.0 4.1 Asgrow 2107 May 29
2005 Capac sandy clay loam 7.0 4.1 Asgrow 2107 May 4
Waseca, MN 2004 Webster clay loam 6.4 6.3 Asgrow 2105 May 19
Fargo, ND 2004 Fargo-Ryan silty clay 7.2 6.8 Asgrow 0801 May 27
2005 Fargo-Ryan silty clay 7.2 6.8 Asgrow 0801 May 20
Prosper, ND 2005 Perella-Bearden silty clay loam 7.3 3.3 Asgrow 0801 May 20
Concord, NE 2004 Kennebec silty clay loam 6.7 4.0 Agripro 2502 May 22
Arlington, WI 2004 Plano silt loam 6.3 4.2 Asgrow 2106 May 6
2005 Plano silt loam 6.3 4.2 Asgrow 2106 May 5

high crop-yield potential environments, indicating a good treatment. Weed communities were established relative to
potential for practical use for herbicide recommendations. soybean growth stage: weeds that emerged at the same time as
However, Jeschke et al. (2009) found that WeedSOFT pre- soybean (VE), and at the cotyledon (VC), first-node (V1), and
dictions tended to overestimate the competitiveness of later- third-node (V3) stages. Glyphosate was applied as needed
emerging mixed-species weed communities in corn across to maintain weed-free conditions prior to targeted weed
seven site-years. emergence times. Tillage consisted of chisel plowing followed
In a study that included highly competitive broadleaf weed by disking or field cultivation for seed bed preparation at most
species in soybean (high CI values), Hock et al. (2006a) found sites. Glyphosate-resistant soybean varieties were selected
that soybean yield loss varied across locations, and that early- based on site-year conditions and were seeded at a targeted
emerging weeds were associated with greater soybean yield rate of 494,000 seeds ha21 in rows spaced 19-cm apart. Plot
loss than later-emerging weeds. Yield loss was 52 and 21% size was 3.0 m by 9.1 m.
when common sunflower (Helianthus annuus L.) emerged at
emergence (VE) and first-node (V1) soybean stages (Ritchie Data Collection. Two 25-cm by 76-cm quadrats were
et al. 1997b), respectively. Although broadleaf weeds were established in each plot for collection of weed community
associated with greater soybean yield loss than grass weeds, this data, except at the Fargo, ND, site in 2004 where four
outcome was attributed in part to the relatively low density of quadrats were established in each plot. Weed community data
grass weeds. In related research, revised weed CI values were recorded separately for each quadrat. Weed density
improved predicted soybean yield loss, but inconsistencies in within the quadrats was measured for each species 2 wk
predictions for most weed species suggested that additional following community establishment and prior to soybean
modifications to WeedSOFT were needed to further improve harvest when weed biomass was at or near maximum. Soybean
soybean yield loss predictions (Hock et al. 2006b). grain was harvested by machine and adjusted to 13%
Results of recent research (Jeschke et al. 2009) showed that moisture.
the predictive accuracy of WeedSOFT for corn yield loss as
affected by weed emergence time could be greatly improved
by modifying growth stage CIM values. In soybean, our Statistical Analysis. Soybean yield data for each experiment
understanding of weed emergence time effects on the were analyzed using linear mixed-effects models fit by REML
predictive accuracy of WeedSOFT is limited, particularly in S-Plus software Version 7.0,1 with block considered as a
for mixed-species communities. Consequently, we conducted random effect and weed community establishment time
experiments at several sites across the Midwestern United considered as a fixed effect (Pinheiro and Bates 2000). Yield
States to determine the accuracy of WeedSOFT predictions of loss associated with weed community establishment times was
soybean yield loss in mixed-species weed communities. determined by treatment contrasts relative to the season-long
weed-free treatment (Pinheiro and Bates 2000). Assumptions
of normal distribution of data were assessed using normal
Materials and Methods probability plots of the residuals (Crawley 2002).
Soybean yield loss associated with mixed-species weed
Research Sites. Field experiments were conducted in a total communities was compared to potential yield loss predicted
of nine site-years across the Midwestern United States in 2004 by the ADVISOR module of WeedSOFT version 11.0.18.2
and 2005 (Table 1). Experiment sites were selected locally Weed CI values in WeedSOFT are based on regional and
based on the presence of natural infestations of at least two of local research as well as expert opinion; consequently, values
the following weed species: common lambsquarters (Cheno- can differ among state-specific versions of WeedSOFT.
podium album L., CHEAL), giant ragweed (Ambrosia trifida Therefore, yield loss was predicted for the Michigan,
L., AMBTR), velvetleaf (Abutilon theophrasti Medik., ABUTH), Nebraska, and Wisconsin experiments using the respective
redroot pigweed (Amaranthus retroflexus L., AMARE), tall water- state-specific versions of WeedSOFT. Wisconsin WeedSOFT
hemp [A. tuberculatus (Moq.) Sauer, AMATA], woolly cupgrass was used for analysis of Minnesota and North Dakota
[Eriochloa villosa (Thunb.) Kunth, ERBVI], giant foxtail, barnyard- experiments, as these states do not have state-specific versions
grass [Echinochloa crus-galli (L.) Beauv., ECHCG], yellow foxtail, of the software. Yield loss predictions were generated based on
and large crabgrass [Digitaria sanguinalis (L.) Scop., DIGSA]. weed community composition 2 wk following community
establishment. It was assumed that all weeds were less than
Field Procedures. The experimental design at each site was a 5-cm tall within 2 wk of emergence. In order to accurately
randomized complete block with four replications of four validate the yield-loss module in WeedSOFT, it was necessary
times of weed community establishment and a weed-free to base yield-loss predictions on soybean growth stage at

Jeschke et al.: Predicted soybean yield loss N 417

Table 2. WeedSOFT version 11.0 competitive index modifier (CIM) values value representative of soybean growth stage. The exponential
used to adjust competitive index (CI) values based on soybean and weed decay equation was chosen empirically to describe revised
growth stage.
CIM values relative to that of CIM values currently in the
CIM value ADVISOR module.
Soybean
Weed growth stage
growth Stage
stagea numberb Very smallc Small Medium Tall
Results and Discussion
V1 1 1.00 1.25 2.00 2.50
V2 2 0.80 1.00 1.50 2.00 Weed Community Composition Across Site-Years. Twenty-
V3 3 0.60 0.75 1.25 1.50 three weed species were observed across experiment site-years
V4 4 0.40 0.50 0.90 1.20
V5 5 0.30 0.35 0.65 0.75 (Table 3). The most abundant weed species (based on density)
R1 6 0.20 0.25 0.50 0.50 across site-years were either giant foxtail or yellow foxtail,
R2 7 0.17 0.20 0.40 0.40 except in Nebraska where tall waterhemp was the most
R3 8 0.15 0.18 0.30 0.39 abundant species. Common lambsquarters, common ragweed
R4 9 0.14 0.17 0.28 0.38
R5 10 0.13 0.16 0.27 0.37 (Ambrosia artemisiifolia L., AMBEL), redroot pigweed, and
R6 11 0.12 0.15 0.27 0.36 velvetleaf were the most abundant broadleaf weed species across
R7 12 0.11 0.15 0.26 0.35 site-years.
R8 13 0.10 0.15 0.25 0.35 Proportional density of grass and broadleaf weed species
a
Abbreviations: V1 to V5, first- to fifth-node vegetative stages, respectively; varied among site-years (Figure 1). Weed communities at
R1, beginning bloom; R2, full bloom; R3, full pod; R5, beginning seed; R6, full Wisconsin and Fargo, ND, in 2004 and 2005 and Prosper,
seed; R7, beginning maturity; and R8, full maturity (Ritchie et al. 1997b). ND, in 2005 had a greater proportion of grass species than
b
Values used to represent crop growth stages in regression analysis.
c
Weed growth stage categories are expressed in inches and are specific for each
broadleaf species, whereas weed communities at Minnesota,
weed species. Nebraska, and Michigan in 2004 had a greater proportion
of broadleaf species. At Michigan in 2005, the proportional
density of grass and broadleaf species was similar. Total weed
the time of targeted weed growth stage rather than weed densities were typically greatest in weed communities that
community establishment time. Yield observed in the weed- emerged at the same time as soybean (VE), and decreased with
free treatments was used as the weed-free yield on which to successive community establishment time. Total weed density
base yield loss predictions. In practice, the weed-free yield ranged from 40 to 680 plants m22 among weed communities
must be estimated based on cropping history, as it is not that emerged with soybean.
known at the time when weed management decisions are
made. However, for this analysis, observed crop yields were Weed Community Composition and Soybean Yield by
used in order to reduce error for validation of the crop yield Site-Year. Michigan 2004. Giant foxtail and wild mustard
loss model. Pearson correlations were used to determine weed (Sinapis arvensis L., SINAR) were the most abundant of eight
community characteristics associated with error in crop yield weed species at the site (Table 3). Observed yield associated
predictions (Crawley 2002). with weed communities established at VE, VC, and V1 soybean
Comparisons of predicted and observed soybean yield was 42, 89, and 87% of weed-free yield (4,200 kg ha21),
pooled over site-years were used to determine prediction respectively (Table 4). Yield associated with weed communities
accuracy associated with CIM values in the WeedSOFT established at V3 soybean did not differ from weed-free
(version 11.0.18; Table 2) and to evaluate potential changes yield. Predicted yield was less than observed yield for weed
to the CIM values for improved predictive accuracy of the communities established at VE and VC soybean, and was 24
soybean yield loss model. Although weed CI values may differ
and 59% of weed-free yield, respectively. Predicted yield for
among state-specific versions of WeedSOFT, CIM matrices are
weed communities established at V1 and V3 soybean were 81
the same among versions. Nonlinear regression analyses of
and 88% of weed-free yield, respectively, and was similar to
predicted and observed soybean yield pooled over site-years
were conducted based on soybean growth stage 2 wk follow- observed yield.
ing community establishment, at the time of weed density
measurements. The Gompertz equation (Ratkowsky 1990) was Michigan 2005. Weed community composition (Table 3) and
used to describe the effect of weed establishment timings on densities (Figure 1) did not differ greatly between 2004 and
predicted and observed soybean yields (Equation 1): 2005. Giant foxtail and common lambsquarters were the most
abundant of seven weed species in 2005 (Table 3). Observed
Y ~A exp ({B exp ({KT )) ½1 yield associated with the weed community established at VE
where Y is yield as a percentage of the weed-free crop, A is the soybean was 59% of weed-free soybean (3,460 kg ha21),
upper asymptote, B and K are parameters that determine the whereas yield associated with the weed community established
shape of the curve, and T is a numerical value representative of at V1 soybean was slightly greater than weed-free yield
soybean growth stage. (Table 4). Soybean yield associated with weed communities
An exponential decay equation was used to describe established at VC and V3 soybean did not differ from that
proposed CIM values as a function of soybean growth stage of weed-free yield. Predicted yield was less than observed
(Equation 2): yield for weed communities established at VE, VC, and V1
Y ~Y0 exp ({kX ) ½2 soybean, with predicted yields of 22, 32, and 61% of weed-
free yield, respectively. Predicted yield associated with the
where Y is the growth stage CIM value, Y0 represents CIM at weed community established at V3 soybean was 100% of
growth stage zero, k is a rate constant, and X is a numerical weed-free yield, as no weeds were observed.

418 N Weed Science 59, July–September 2011

Table 3. Weed species abundance rank and competitive index (CI) values in WeedSOFT version 11.0 for weed species observed at research sites in 2004 and 2005.
Abundance rank
Prosper,
Michiganb Minnesotac Nebraskad Fargo, NDc NDc Wisconsinc CI value
Weed
speciesa 2004 2005 2004 2004 2004 2005 2005 2004 2005 Michigan Nebraska Wisconsin
ABUTH 3 7 5 3 — — — 7 7 2.0 4.2 1.0
AMAPO —e 5 — — — — — — — 1.5 — —
AMARE 7 — 2 — — 2 4 5 — 1.5 — 0.9
AMATA — — — 1 — — — — — — 2.5 —
AMBEL 4 4 3 — 6 7 5 11 4 4.0 — 0.9
CHEAL 5 2 4 — — 3 6 4 3 3.0 — 1.5
CIRAR — — — — 8 9 — — — — — 0.6
DIGSA — — — — — — — 3 5 — — 0.3
EPHMAf — — — — 2 — — — — — — 1.0
ERICA — — — — — — — 10 — — — 1.5
HIBTRf — — — — 3 — 10 — — — — 1.0
KCHSC — — — — — — 8 — — — — 0.75
MALNEf — — — — — 4 3 — — — — 1.0
POLCO — — — — 5 6 7 — 6 — — 0.6
POLPY 8 8 — — — — — 6 — 1.5 — 0.5
SALREf — — — — — 8 — — — — — 1.0
SETFA 1 1 1 2 — — — 1 1 0.8 2.0 0.75
SETLU — — — — 1 1 1 2 2 — — 0.25
SINAR 2 3 — — 4 5 — — — 2.0 — 0.75
SOLPT 6 6 — — — — — 8 8 0.5 — 0.25
SOLSA — — — — — — 2 — — — — 0.25
TAROF — — — — 7 — — 9 — — — 0.6
XANST — — — — — — 9 — — — — 1.25
a
Abbreviations: ABUTH, velvetleaf; AMAPO, Powell amaranth (Amaranthus powellii S. Wats.); AMARE, redroot pigweed; AMATA, tall waterhemp; AMBEL,
common ragweed; CHEAL, common lambsquarters; CIRAR, Canada thistle [Cirsium arvense (L.) Scop.]; DIGSA, large crabgrass; EPHMA, spotted spurge; ERICA,
horseweed [Conyza canadensis (L.) Cronq.]; HIBTR, Venice mallow; KCHSC, kochia [Kochia scoparia (L.) Schrad.]; MALNE, common mallow; POLCO, wild
buckwheat (Polygonum convolvulus L.); POLPY, Pennsylvania smartweed (Polygonum pensylvanicum L.); SALRE, lanceleaf sage (Salvia reflexa Hornem.); SETFA, giant
foxtail; SETLU, yellow foxtail; SINAR, wild mustard; SOLPT, eastern black nightshade (Solanum ptychanthum Dunal); SOLSA, hairy nightshade; TAROF, dandelion
(Taraxacum officinale G.H. Weber ex Wiggers); and XANST, common cocklebur (Xanthium strumarium L.).
b
Yield predictions used CI values from Michigan WeedSOFT version 11.0.
c
Yield predictions used CI values from Wisconsin WeedSOFT version 11.0.
d
Yield predictions used CI values from Nebraska WeedSOFT version 11.0.
e
Weed species not observed.
f
Weed species not included in WeedSOFT.

Minnesota. Five weed species were observed, with giant foxtail, establishment times, and was 28 and 38% of weed-free yield,
redroot pigweed, and common ragweed the most abundant respectively. Weed density data were not collected for
(Table 3). Observed yield associated with weed communities weed communities established at V1 and V3 soybean, and
established at VE and VC soybean was 40 and 78% of weed- consequently, yield was not predicted for these communities.
free yield (3,650 kg ha21), respectively (Table 4). Observed
yield associated with weed communities established at V1 and Fargo, ND, 2004. Yellow foxtail, spotted spurge [Chamaesyce
V3 soybean did not differ from weed-free yield. Since weed maculata (L.) Small, EPHMA], and Venice mallow (Hibiscus
density data were not collected for the weed community trionum L., HIBTR) were the most abundant of eight weed
established at VE soybean, yield was not predicted for this species for this site-year (Table 3). Observed yield did not
community establishment time. However, predicted yield differ from weed-free yield (1,390 kg ha21) for any weed
did not differ from observed yield for weed communities community establishment time (Table 4). The relatively low
established at VC and V1 soybean, but was less than observed weed-free yield was associated with below normal rainfall in
yield for the weed community established at V3 soybean. June and August (data not shown). Predicted yield was less
Predicted yields were 68, 98, and 87% of weed-free yield for than observed yield for each establishment time, and was 39,
weed communities established at VC, V1, and V3 soybean, 56, 74, and 87% of weed-free yield for weed communities
respectively. established at VE, VC, V1, and V3 soybean, respectively.

Nebraska. Tall waterhemp, giant foxtail, and velvetleaf were Fargo, ND, 2005. Total weed community density at Fargo,
the only weed species observed at this site-year (Table 3). ND, was greater in 2005 than in 2004 (Figure 1). Nine weed
Total weed community density was low relative to the other species were observed; the most abundant weed species were
site-years (Figure 1). Observed yield associated with the weed yellow foxtail, redroot pigweed, and common lambsquarters
community established at VE soybean was only 40% of weed- (Table 3). In contrast to 2004, yields associated with weed
free yield (3,120 kg ha21), whereas observed yields associated communities established at the VE and VC soybean were less
with other weed establishment times did not differ from than weed-free yield (3,260 kg ha21), with yields of 63 and
weed-free yield (Table 4). Predicted yield was less than 81% of weed-free yield, respectively (Table 4). Also in contrast
observed yield for both VE and VC weed community to 2004, rainfall was much above normal in June and August

Jeschke et al.: Predicted soybean yield loss N 419

 much less than observed yield for each weed community
establishment time. Predicted yields were 23, 26, 48, and
67% of weed-free yield for weed communities established at
VE, VC, V1, and V3 soybean, respectively.

Wisconsin 2005. Weed community composition in 2005 was

similar to that in 2004. Grass species composed the majority of
the weed community (Figure 1), with giant foxtail, yellow
foxtail, and common lambsquarters the most abundant of eight
weed species observed (Table 3). In contrast to 2004, observed
yield associated with the weed community established at VE
soybean was less than weed-free yield (4,100 kg ha21), and was
only 17% of weed-free yield (Table 4). Also in contrast to
2004, rainfall was below normal for most of the growing
season (data not shown). Observed yield associated with weed
communities established at later soybean growth stages did not
differ from weed-free yield. Predicted yield associated with the
weed community established at VE soybean was 22%, which
was greater than observed yield. Since the asymptote parameter
of the soybean yield loss function in WeedSOFT is limited to
80% yield loss (Neeser et al. 2004), prediction accuracy is
reduced for yield losses greater than 80%. The Wisconsin 2005
Figure 1. Density of grass and broadleaf weed species associated with weed site-year was the only instance across site-years where this
communities established at soybean emergence (VE), cotyledon (VC), first-node parameter restriction likely limited prediction accuracy.
(V1), and third-node (V3) stages in 2004 and 2005. Weed density measurements Predicted yield associated with the weed community established
were taken 2 wk following community establishment. Data not collected indicated at VC soybean was 56% of weed-free yield, and was much less
by ‘‘nc.’’ Error bars represent the standard error of mean values.
than the observed yield of 99% of weed-free yield. For weed
communities established at V1 and V3 soybean, predicted
(data not shown). However, similar to 2004, predicted yield yields did not differ from observed yields, and were 92 and 97%
was less than observed yield for each weed community of weed-free yield, respectively.
establishment time, except V3 soybean. Predicted yield was
23, 34, 65, and 85% of weed-free yield for weed communities
established at VE, VC, V1, and V3 soybean, respectively. Soybean Yield Pooled Over Site-Years. Nonlinear regression
analyses of predicted and observed soybean yield data pooled
over site-years showed that predicted yields were less than
Prosper, ND. Yellow foxtail, hairy nightshade (Solanum observed yields at all soybean growth stages (Figure 2). This
physalifolium Rusby, SOLSA), and common mallow (Malva result is indicative of a general overestimation of soybean yield
neglecta Wallr., MALNE) were the most abundant of 10 weed loss potential associated with weed communities. The fitted
species observed (Table 3). Total weed density associated with value for predicted yield at V1 soybean was 27% of weed-free
the weed community established at VC soybean was the soybean, whereas observed yield was 60% of weed-free
greatest among weed community establishment times for soybean. Underprediction of yield at V1 soybean indicates
this site-year, whereas for other site-years, weed density was that factors other than growth stage CIM values contributed
typically greatest for weed communities established at VE to the inaccuracy of predicted soybean yield. To isolate the
soybean (Figure 1). Observed yields associated with weed effects of CIM values on prediction accuracy, TCL values
communities established at VE and VC soybean were less were adjusted (multiplied by 0.0775) such that average
than weed-free yield (2,650 kg ha21); soybean yield was not predicted and observed yields were equal at V1 soybean
affected by weed communities established at V1 and V3 (Figure 3). Following this adjustment, parameters of the
soybean (Table 4). Although weed density associated with the nonlinear model fit to predicted yields did not differ from
community established at VE soybean was less than at VC parameters for observed yield (p 5 0.8104). These results
soybean, yield associated with establishment time at VE was indicate that the soybean growth stage CIM values adequately
less than that at VC, and was 41 and 59% of weed-free yield, represent variability in soybean yield loss potential associated
respectively. Predicted yield was less than observed yield for with weed community establishment times. Therefore, inac-
each weed community establishment time, and was 24, 24, curacy of soybean yield predictions is likely due to other
50, and 67% of weed-free yield for weed communities model parameters (discussed below).
established at VE, VC, V1, and V3 soybean, respectively.
Improving Accuracy of Predicted Soybean Yield Loss.
Wisconsin 2004. The weed community was dominated by Pearson correlation analyses (Crawley 2002) were conducted
grass species (Figure 1), with giant foxtail, yellow foxtail, and to determine the relationship between prediction error
large crabgrass the most abundant of 11 weed species observed (observed yield vs. predicted yield) and weed community
(Table 3). Observed yield associated with each weed com- characteristics. Soybean yield prediction error associated with
munity establishment time was similar to weed-free yield specific weed species could be indicative of inaccuracy in weed
(3,860 kg ha21; Table 4), even though weed densities were CI values. Since weed communities consisted largely of grass
relatively high (Figure 1). In contrast, predicted yield was species and moderately competitive broadleaf species, total

420 N Weed Science 59, July–September 2011

Table 4. Weed-free, predicted, and observed soybean yield associated with weed communities established at emergence (VE), cotyledon (VC), first-node (V1), and
third-node (V3) soybean.
Soybean yield (6 SE)
Site Year Time of weed establishment Weed-free Observed Predicted
21
kg ha ----------------------------------------- % of weed-free yield ----------------------------------------
Michigan 2004 Weed-free 4,200 6 130
VE 42 6 2 (, 0.001)a 24 6 1
VC 89 6 4 (0.030) 59 6 8
V1 87 6 6 (0.012) 81 6 7
V3 96 6 6 (0.298) 88 6 7
2005 Weed-free 3,460 6 170
VE 59 6 5 (, 0.001) 22 6 1
VC 96 6 4 (0.438) 32 6 1
V1 112 6 5 (0.042) 61 6 7
V3 101 6 7 (0.954) 100 6 ,1
Minnesota 2004 Weed-free 3,650 6 170
VE 40 6 5 (, 0.001) ncb
VC 78 6 5 (0.004) 68 6 11
V1 93 6 6 (0.257) 98 6 2
V3 99 6 4 (0.824) 87 6 7
Nebraska 2004 Weed-free 3,120 6 40
VE 40 6 6 (, 0.001) 28 6 1
VC 89 6 4 (0.058) 38 6 3
V1 102 6 2 (0.669) nc
V3 104 6 4 (0.444) nc
Fargo, ND 2004 Weed-free 1,390 6 60
VE 96 6 8 (0.522) 39 6 8
VC 96 6 3 (0.545) 56 6 9
V1 97 6 6 (0.616) 74 6 13
V3 102 6 9 (0.870) 87 6 4
2005 Weed-free 3,260 6 170
VE 63 6 8 (, 0.001) 23 6 1
VC 81 6 5 (0.019) 34 6 2
V1 96 6 5 (0.515) 65 6 6
V3 95 6 7 (0.460) 85 6 4
Prosper, ND 2005 Weed-free 2,650 6 110
VE 41 6 6 (, 0.001) 24 6 1
VC 59 6 16 (0.002) 24 6 1
V1 94 6 2 (0.587) 50 6 7
V3 102 6 6 (0.908) 67 6 3
Wisconsin 2004 Weed-free 3,860 6 120
VE 94 6 4 (0.102) 23 6 1
VC 93 6 5 (0.078) 26 6 1
V1 102 6 5 (0.729) 48 6 3
V3 95 6 4 (0.176) 67 6 6
2005 Weed-free 4,100 6 160
VE 17 6 1 (, 0.001) 22 6 , 1
VC 99 6 4 (0.838) 56 6 3
V1 100 6 8 (0.940) 92 6 3
V3 104 6 6 (0.582) 97 6 1
a
Values in parentheses are p-values at a 5 0.05.
b
Weed community data not collected.

and proportional density of grass and broadleaf weeds, total soybean yield loss. They suggested that yield loss function
density, and number of weed species were used to describe parameters in soybean may need to be weed species-specific to
weed communities. Prediction error was positively correlated adequately represent soybean yield loss potential, citing
with grass density, grass proportional density, total weed variability in canopy architecture plasticity among broadleaf
density, and weed species number, and negatively correlated weed species. Since grass species vary greatly from broadleaf
with broadleaf proportional density (Table 5). Grass popula- weeds in canopy architecture (Moechnig et al. 2003), it is
tions across site-years were typically composed of either yellow reasonable to suggest that yield loss function parameters
or giant foxtail, or a combination of these two species. These specific to grass weeds may improve accuracy of soybean yield
results suggest that WeedSOFT overestimated the competitive predictions.
ability of high densities of giant and yellow foxtail in soybean. Although our research showed that WeedSOFT typically
Adjustments to CI values for these species may improve overestimated soybean yield loss, the model serves an
soybean yield prediction accuracy; however, the association of important role as a decision support system. Previous research
prediction error with grasses and high weed densities suggests has shown considerable variation in the onset of the critical
that crop yield loss function parameters may be limiting time of weed removal in glyphosate-resistant soybean across
accuracy of soybean yield predictions as well. Hock et al. years, tillage systems, and row spacing (Mulugeta and
(2006b) found that using revised CI values provide some, Boerboom 2000). WeedSOFT has the potential to provide
but not major, improvements in WeedSOFT for predicting a conservative but useful estimate of soybean yield loss based

Jeschke et al.: Predicted soybean yield loss N 421

 Table 5. Pearson correlation coefficients between prediction error and weed
community characteristics for soybean experiments conducted in 2004 and 2005.
Weed community characteristic Prediction errora
Broadleaf density 0.1193
Grass density 0.2362**b
Broadleaf proportional density 20.2261*
Grass proportional density 0.3275***
Total weed density 0.2309**
Number of weed species 0.3961***
a
Prediction error is the difference between observed yield and predicted yield.
b
An asterisk ‘‘*’’ denotes p , 0.05, ‘‘**’’ denotes p , 0.01, and ‘‘***’’ denotes
p , 0.001.

on observed densities of multiple weed species, crop and

weed growth stages, and environmental conditions. However,
future research should address adjustments to competitive
index values and soybean yield loss function parameters for
Figure 2. Observed soybean yield (m) and soybean yield predicted by giant foxtail and yellow foxtail, and perhaps other grass weed
ADVISOR (%) associated with weed communities established at soybean
emergence (VE), cotyledon (VC), first-node (V1), and third-node (V3) stages
species, to improve the prediction accuracy of WeedSOFT.
pooled over site-years: Michigan in 2004 and 2005, Minnesota in 2004,
Nebraska in 2004, Fargo, ND, in 2004 and 2005, Prosper, ND, in 2005, and
Wisconsin in 2004 and 2005. Soybean stages represent soybean growth stage at Sources of Materials
the time of weed community density measurements 2 wk following weed 1
community establishment. Categorical growth stages were converted to a S-Plus for Windows, Version 7.0. Insightful Corporation. 1700
continuous numerical scale for regression analyses: 1 5 first-node (V1), Westlake Avenue North, Seattle, WA 98109-3044. http://www.
2 5 second-node (V2), 3 5 third-node (V3), 4 5 fourth-node (V4), insightful.com/.
5 5 fifth-node (V5), 6 5 beginning bloom (R1), and 7 5 full bloom (R2) 2
soybean. Observed soybean yield pooled over site-years was described by
WeedSOFT Decision Support System, Version 11.0.18. Uni-
Y 5 99.96 exp(21.346 exp(20.9928X)) (r2 5 0.42). Predicted soybean yield versity of Nebraska-Lincoln, P.O. Box 830915, Lincoln, NE 68583-
pooled over site-years was described by Y 5 84.74 exp(22.001 exp(20.5692X)) 0915. http://weedsoft.unl.edu/.
(r2 5 0.58).

Acknowledgments
This research was funded in part by USDA-CSREES NC-202
Regional Research Project ‘‘Characterizing Weed Population
Variability for Improved Weed Management Decision Support
Systems to Reduce Herbicide Use,’’ and in part by the College of
Agricultural and Life Sciences, University of Wisconsin-Madison
(M. Jeschke and D. Stoltenberg). G. Kegode thanks Mark Ciernia at
NDSU for technical help in conducting this research.

Literature Cited
Crawley, M. J. 2002. Statistical Computing: An Introduction to Data Analysis
Using S-Plus. West Sussex, England: John Wiley & Sons. Pp. 305–322.
Fickett, N. D., D. E. Stoltenberg, C. M. Boerboom, and C. M. Hammond.
2009. Estimated economic losses from early weed competition in Wisconsin
corn and soybean fields. North Cent. Weed Sci. Soc. Proc. 64:93.
Gower, S. A., M. M. Loux, J. Cardina, S. K. Harrison, P. L. Sprankle, N. J. Probst,
T. T. Bauman, W. Bugg, W. S. Curran, R. S. Currie, R. G. Harvey, W. G.
Johnson, J. J. Kells, M.D.K. Owen, D. L. Regehr, C. H. Slack, M. Spaur, C. L.
Figure 3. Observed soybean yield (m) and soybean yield predicted by Sprague, M. VanGessel, and B. G. Young. 2003. Effect of postemergence
ADVISOR (%) associated with weed communities established at soybean glyphosate application timing on weed control and grain yield in glyphosate-
emergence (VE), cotyledon (VC), first-node (V1), and third-node (V3) stages resistant corn: results of a 2-yr multistate study. Weed Technol. 17:821–828.
pooled over site-years: Michigan in 2004 and 2005, Minnesota in 2004, Hock, S. M., S. Z. Knezevic, W. G. Johnson, C. Sprague, and A. R. Martin.
Nebraska in 2004, Fargo, ND, in 2004 and 2005, Prosper, ND, in 2005, and 2007. WeedSOFT: effects of corn-row spacing for predicting herbicide efficacy
Wisconsin in 2004 and 2005. Predicted values reflect the use of a multiplier to on selected weed species. Weed Technol. 21:219–224.
TCL values that resulted in average predicted yield equaling observed soybean Hock, S. M., S. Z. Knezevic, A. R. Martin, and J. L. Lindquist. 2006a. Soybean
yield at the V1 growth stage (CIM 5 1) and that eliminated error from the row spacing and weed emergence time influence weed competitiveness and
model not associated directly with the CIM values. Soybean stages represent competitive indices. Weed Sci. 54:38–46.
soybean growth stage at the time of weed community density measurements 2 wk Hock, S. M., S. Z. Knezevic, A. R. Martin, and J. L. Lindquist. 2006b.
following weed community establishment. Categorical growth stages were Performance of WeedSOFT for predicting soybean yield loss. Weed Technol.
converted to a continuous numerical scale for regression analyses: 1 5 first-node 20:478–484.
(V1), 2 5 second-node (V2), 3 5 third-node (V3), 4 5 fourth-node (V4), Jeschke, M. R., D. E. Stoltenberg, G. O. Kegode, J. A. Dille, and G. A. Johnson.
5 5 fifth-node (V5), 6 5 beginning bloom (R1), and 7 5 full bloom (R2) 2009. Weed community emergence time affects accuracy of predicted corn
soybean. Observed soybean yield pooled over site-years was described by yield loss by WeedSOFT. Weed Technol. 23:477–485.
Y 5 99.96 exp (21.346 exp (20.9928X ) ) (r2 5 0.42). Predicted soybean yield Kruger, G. R., W. G. Johnson, and S. C. Weller, et al. 2009. U.S. grower
pooled over site-years was described by Y 5 99.09 exp (21.321 exp (20.9190X ) ) views on problematic weeds and changes in weed pressure in glyphosate-
(r2 5 0.62). An F-test indicated no statistical difference between the regression resistant corn, cotton, and soybean cropping systems. Weed Technol.
parameters of the two models (p 5 0.8104). 23:162–166.

422 N Weed Science 59, July–September 2011

Moeching, M. J., C. M. Boerboom, D. E. Stoltenberg, and L. K. Binning. Pinheiro, J. C. and D. M. Bates. 2000. Mixed-Effects Models in S and S-Plus.
2003. Growth interactions in communities of common lambsquarters New York: Springer. 530 p.
(Chenopodium album), giant foxtail (Setaria faberi ), and corn. Weed Sci. Ratkowsky, D. A. 1990. Handbook of Nonlinear Regression Models. Statistics,
51:363–370. textbooks and monographs, Vol. 107, 241 p. New York: Marcel Dekker.
Mulugeta, D. and C. M. Boerboom. 2000. Critical time of weed removal in Ritchie, S. W., J. J. Hanway, and G. O. Benson. 1997a. How a corn plant
glyphosate-resistant Glycine max. Weed Sci. 48:35–42. develops. Special Rep. 48. Ames, IA: Iowa State University. 22 p.
Myers, M. W., W. S. Curran, M. J. VanGessel, B. A. Majek, B. A. Scott, D. A. Ritchie, S. W., J. J. Hanway, H. E. Thompson, and G. O. Benson. 1997b. How a soybean
Mortensen, D. D. Calvin, H. D. Karsten, and G. W. Roth. 2005. The effect of plant develops. Special Rep. 53. Ames, IA: Iowa State University. 20 p.
weed density and application timing of weed control and corn grain yield. Schmidt, A. A., W. G. Johnson, and D. A. Mortensen, et al. 2005. Evaluation of
Weed Technol. 19:102–107. corn (Zea mays L.) yield-loss estimations by WeedSOFTH in the north central
Neeser, C., J. A. Dille, G. Krishman, D. A. Mortensen, J. T. Rawlinson, A. R. region. Weed Technol. 19:1056–1064.
Martin, and L. B. Bills. 2004. WeedSOFTH: a weed management decision
support system. Weed Sci. 52:115–122. Received August 27, 2010, and approved February 28, 2011.

Jeschke et al.: Predicted soybean yield loss N 423