Introduction
Use of dicamba in dicamba-resistant soybean can provide an alternative for management of herbicide-resistant broadleaf weeds (Flessner et al. Reference Flessner, McElroy, McCurdy, Toombs, Wehtje, Burmester, Price and Ducar2015; Spaunhorst and Bradley Reference Spaunhorst and Bradley2013). In 2017, however, off-target movement of dicamba resulted in 2,708 dicamba-related injury investigations by state departments of agriculture in the United States (Bradley Reference Bradley2017). Approximately 1.46 million ha of soybean were reported by state extension weed scientists as injured by off-site movement of dicamba. Unfortunately, issues with dicamba off-site movement continued in 2018, and as of July 15, around 600 cases of dicamba injury to various crops were being investigated by the state departments of agriculture. University weed scientists estimated that approximately 445,000 ha of soybean were injured by dicamba (Bradley Reference Bradley2018).
Symptoms most often associated with soybean exposed to dicamba are leaf cupping and leaf surface crinkling. These symptoms are easily recognized and cause concern regarding the effect of dicamba on crop yield. This uncertainty places growers in a predicament in regard to the economic value of crop inputs for the remainder of the growing season. Foster and Griffin (Reference Foster and Griffin2018), using a 0 (no injury) to 5 (severe injury) scale, reported upper canopy leaf cupping in soybean of 4.7 and leaf surface crinkling of 4.4 15 d after the V3/V4 (Fehr and Caviness Reference Fehr and Caviness1977) exposure to dicamba at 0.6 g ha−1 (1/1,000 of 560 g ha−1 use rate). Following exposure to 0.6 g ha−1 dicamba at R1/R2, terminal leaf cupping of 4.4 and leaf surface crinkling of 3.4 were observed. Meta-analysis of data from over seven decades of simulated drift experiments showed soybean yield losses of basically zero for exposure to dicamba at 0.56 g ha−1 during vegetative stage and approximately 1% during flowering (Egan et al. Reference Egan, Barlow and Mortensen2014).
In most research evaluating crop response to auxin herbicides, injury has been assessed using a visual rating of 0 (no crop injury/no effect) to 100% (plant death), with specific injury criteria represented in ratings often not provided. Crop injury ratings in response to dicamba have shown an increase with increasing rate (Al-Khatib and Peterson Reference Al-Khatib and Peterson1999; Foster and Griffin Reference Foster and Griffin2018; Griffin et al. Reference Griffin, Bauerle, Stephenson, Miller and Boudreaux2013; Johnson et al. Reference Johnson, Fisher, Jordan, Edmisten, Stewart and York2012). When yield data associated with various dicamba rates are available, injury ratings could be used to pinpoint rate of exposure and to predict yield loss. The lack of specificity in assigning injury ratings, along with the variability in ratings expected among individuals, however, could affect the ability to accurately pinpoint exposure rate and to predict yield loss.
Egan and Mortensen (Reference Egan and Mortensen2012) evaluated soybean injury following exposure to dicamba using a 0 to 100 scale in 10-point increments, with description of injury adapted from Andersen et al. (Reference Andersen, Clay, Wrage and Matthees2004) and Behrens and Lueschen (Reference Behrens and Lueschen1979). Injury ratings of 10 represented “slight crinkle of leaflets of terminal leaf”; 20 indicated “cupping of terminal leaflets, slight crinkle of leaflets of second leaf, growth rate normal”; and 50, “no expansion of terminal leaf, second leaf size one-half that of control. Axillary leaf buds unable to open and develop”. In dose–response experiments, soybean injury was quantified 14 d following dicamba applied at V4 at 0.006 to 56.1 g ha−1 in one year and at 0.006 to 561 g ha−1 the second year. For each year, ln-linear regression was used to produce the inverse relationship describing ln (dose) as a function of injury value 14 d after treatment (DAT). Log-linear models produced a correlation between observed injury and treatment dose of r2 = 97.3 in year 1 and r2 = 96.6 in year 2.
Robinson et al. (Reference Robinson, Simpson and Johnson2013) evaluated soybean injury following exposure to dicamba at 0 to 22.7 g ha−1. Visual ratings were made using a scale of 0 to 100% in 10-point increments with description of the injury provided. Injury ratings of 10% represented “slight reduction in height or canopy volume, cupped or bubbled leaves on less than or equal to the upper 10% of the plant, bent petioles, and chlorosis or necrosis”; 20% indicated “moderately crinkled leaflets (extended across less than or equal to the upper 20% of the plant), curled petioles, reduced height and canopy volume, cupped terminal leaflets”; and 50%, “very high reduction of plant height (less than or equal to 50% of the plant) with little likelihood of recovery from the apical meristem, new growth suppressed, formation of pods reduced or malformed, some leaf and stem tissue becomes necrotic, petioles and stem show severe twisting”. Nonlinear regressions of soybean injury as affected by dicamba rate were used in dose–response models to predict seed yield loss as related to dicamba injury. The effective dose (ED) 14 DAT to cause 20% injury (ED20) was 0.68 g ha−1 dicamba for exposure at V2 and V5 and 0.94 g ha−1 for exposure at R2.
The value of any model lies in its ability to accurately predict an outcome. Egan and Mortensen (Reference Egan and Mortensen2012) and Robinson et al. (Reference Robinson, Simpson and Johnson2013), using visible injury ratings that were specific as to injury criteria and level of injury, were able to develop models predictive of dicamba rate of exposure. Follow-up research, however, was not conducted to determine effectiveness of the visible injury rating scales in predicting dicamba exposure rate and soybean yield loss.
Foster and Griffin (Reference Foster and Griffin2018) described response of indeterminate soybean to dicamba at vegetative and reproductive growth stages using overall visible injury and plant height reduction (0 to 100%), but also identified 14 injury criteria and quantified severity of injury for each using a 0 to 5 scale. The objectives of the present research were (1) to utilize data from previous research conducted by Foster and Griffin (Reference Foster and Griffin2018) to develop soybean yield prediction models based on growth stage and days after exposure to dicamba, (2) to conduct experiments with dicamba rates and application timings the same as those used to develop the models and to compare level of injury for the variables included in each model with those reported previously, and (3) to validate the models by comparing predicted percent yield reduction for each rate of dicamba using the appropriate model with yield reduction observed in two experiments.
Materials and methods
In previous research conducted by Foster and Griffin (Reference Foster and Griffin2018), dicamba was applied at rates of 0.6, 1.1, 2.2, 4.4, 8.8, 17.5, 35, 70, 140, and 280 g ae ha−1 (1/1,000 to one-half of 560 g ha−1 use rate) to soybean at V3/V4 (third/fourth node with two to three fully expanded trifoliates) or at R1/R2 (open flower at any node on main stem/open flower at one of the two uppermost nodes on main stem) (Fehr and Caviness Reference Fehr and Caviness1977). Data for 14 injury criteria rated on level of severity (0 to 5, with 0 = no injury, 1 = slight, 2 = slight to moderate, 3 = moderate, 4 = moderate to severe, and 5 = severe) along with overall visible injury and plant height reduction (0 to 100%, with 0 = none and 100 = plant death), and canopy height collected 7 and 15 DAT were analyzed using multiple linear regression (SAS Institute Inc., Cary NC 27513). Through use of a forward-selection/stepwise procedure (Anonymous 2018b, 2018c), all possible simple linear regressions are calculated, and the first variable is selected based on highest probability (Pr > F). At each step another variable is selected, and the resulting P value is reflective of the contribution of all variables selected at that point in the analysis. The selection process continues until the addition of a variable no longer meets the minimum criteria of P = 0.50. Results of the forward-selection procedure for 7 and 15 DAT for V3/V4 and R1/R2 exposure to dicamba are presented in Tables 1 and 2.
Summary of the forward-selection procedure for 7 and 15 d after treatment (DAT) of dicamba at V3/V4 with variables presented in the order of selection using a minimum criterion of P = 0.50. Bolded variables were included in the final model.a
a Variables in the analysis included 12 injury criteria observed following V3/V4 application (0 to 5 scale with 0 = no injury, 1 = slight, 2 = slight to moderate, 3 = moderate, 4 = moderate to severe, and 5 = severe) along with overall soybean visible injury and height reduction (0 to 100% with 0 = no injury/height reduction and 100 = plant death) and canopy height.
b For the 7 and 15 DAT models, the top six variables were selected based on high probability (Pr > F) and maximizing model R2.
Summary of the forward-selection procedure for 7 and 15 d after treatment (DAT) of dicamba at R1/R2 with variables presented in the order of selection using a minimum criterion of P = 0.50. Bolded variables were included in the final model.a
a Variables in the analysis included 10 injury criteria observed following R1/R2 application (0 to 5 scale with 0 = no injury, 1 = slight, 2 = slight to moderate, 3 = moderate, 4 = moderate to severe, and 5 = severe) along with overall soybean visible injury and height reduction (0 to 100% with 0 = no injury/height reduction and 100 =plant death) and canopy height.
b For the 7 and 15 DAT models, the top six variables were selected based on high probability (Pr > F) and maximizing model R2.
Because the prerequisite for inclusion in the initial model was a minimum P value of ≤ 0.50 and the goal was to use the model to develop a field application for yield loss prediction, variables were further selected based on high probability and maximum R2 (Tables 1 and 2). Multiple regression equations including the variables selected for the final models were used in the validation experiments.
Validation experiments
Experiments to validate the dicamba yield prediction models were conducted in 2016 at the LSU AgCenter, Central Research Station, Ben Hur Research Farm (30.363°N, 91.163°W) in Baton Rouge, LA, and at the Northeast Research Station (31.941°N, 91.233°W) in St. Joseph, LA. The soil type and soil classification for the experiment in Baton Rouge was a Mhoon silt loam (fine-silty, mixed, nonacid, thermic Typic Fluvaquent) with a pH of 6.3 and organic matter (OM) of 1.9%; in St. Joseph it was a Sharkey clay (very-fine, smectitic, thermic Chromic Epiaquert) with a pH of 6.1 and OM of 2.3%.
The indeterminate soybean cultivar ‘Asgrow 4835’ (relative maturity 4.8) was used at both locations. Planting dates in 2016 were May 10 in Baton Rouge and May 9 in St. Joseph, and seeding rate was 300,000 seed ha−1. On the same day of planting, S-metolachlor plus sulfentrazone (Authority® Elite; FMC Corp., Philadelphia, PA 19104) at 1,760 g ai ha−1 was applied in Baton Rouge and glufosinate-ammonium (Liberty® 280 SL herbicide; Bayer CropScience LP, Research Triangle Park, NC 27709) at 590 g ai ha−1 plus glyphosate (Roundup PowerMax®; Monsanto Co., St. Louis, MO 63167) at 1,260 g ae ha−1 plus flumioxazin plus pyroxasulfone (Fierce®; Valent U.S.A. Corp., Walnut Creek, CA 94596) at 160 g ai ha−1 were applied in St. Joseph. In Baton Rouge, glyphosate (Roundup PowerMax®) was applied at 870 g ha−1 when weeds were 5 to 8 cm tall and around 2 wk later to eliminate weed competition. In St. Joseph, clethodim (Select®; Valent U.S.A. Corp., Walnut Creek, CA 94596) was applied at 280 g ai ha−1 to eliminate grass weed competition and was followed 1 wk later with an application of glyphosate at 870 g ha−1. Fungicides and insecticides were applied at both locations beginning at R3 (Fehr and Caviness Reference Fehr and Caviness1977) as needed based on LSU AgCenter recommendations (Anonymous 2018a).
Dicamba rates and application timings were the same as those used in the previous research to develop the models (Foster and Griffin Reference Foster and Griffin2018). The diglycolamine salt of dicamba (Clarity® herbicide; BASF Corp., Research Triangle Park, NC 27709) was used, and nonionic surfactant at 0.25% (Induce; Helena Chemical Co., Collierville, TN 38017) was added to all treatments; a nontreated control was included for comparison. A randomized complete block design with a factorial arrangement of treatments (growth stage by dicamba rate) and four replications were used for both locations.
Dicamba treatments were applied using a CO2-pressurized backpack sprayer calibrated to deliver 140 L ha−1 spray volume at 270 kPa. Sprayers were fitted with 11002 Turbo TeeJet Induction flat-fan spray nozzles (TeeJet® Technologies, Spraying Systems Co., Wheaton, IL 60189), and spray pressure was 270 kpa, resulting in ultra-coarse spray droplets. Spray boom height above the canopy was 46 cm, and ground speed was 5 km h−1. Wind speed at application was no more than 4.8 km h−1. Specific dates for dicamba application at each location along with rainfall received 0 to 4 DAT and average minimum/maximum air temperature, soil temperature, and percent relative humidity 0 to 4 DAT are shown in Table 3. In Baton Rouge, plots consisted of two rows spaced 76 cm apart with a nontreated border area between plots of 152 cm. In St. Joseph, treated plots were two rows spaced 101 cm apart with a nontreated border area between plots of 202 cm. Movement of dicamba between adjacent treatments was not observed.
Rainfall, average minimum (min) and maximum (max) air and soil temperatures, and average relative humidity 0–4 d after herbicide treatment (DAT) for experiments conducted in Baton Rouge and St. Joseph, LA, to validate the dicamba yield loss prediction model.
a Soybean growth stages included V3/V4 (third/fourth node with two to three fully expanded trifoliates) and R1/R2 (open flower at any node on main stem/open flower at one of the two uppermost nodes on main stem).
Data were collected for each dicamba rate for the variables included in the final models based on soybean growth stage and days after exposure (Tables 1 and 2), and averaged across locations. Soybean was combine-harvested on October 11, 2016 in Baton Rouge and on October 6, 2016 in St. Joseph, and yields were adjusted to 13% moisture. To normalize the data, soybean yield for each dicamba rate based on growth stage from the validation experiments and from calculations using the models was expressed as percent yield reduction using the appropriate nontreated control.
$$\eqalign{\hskip-15pt{\rm{Soybean \ yield \ reduction }}\left( \% \right)\cr = {[ {( {{\rm{yield \ of \ nontreated \ control }}}}}{{{-{\rm{ yield \ of \ dicamba \ treatment}}} )}} / \cr \quad{{{\rm{yield \ of \ nontreated \ control}}}]\times }100}$$
To validate the models, predicted percent yield reduction and observed percent yield reduction were compared.
Results and discussion
Development of the models
For V3/V4 exposure 7 DAT, 14 of the 15 variables met the 0.50 significance level for inclusion in the initial model using the forward-selection/stepwise procedure (Table 1). The initial model for 15 DAT included 11 of the 15 variables. Variables that did not meet the significance level were terminal leaf epinasty 7 DAT and upper canopy leaf cupping, upper canopy pale leaf margins, overall visible injury, and canopy height 15 DAT. Terminal leaf cupping and upper canopy leaf rollover/inversion were not observed following V3/V4 exposure to dicamba 7 or 15 DAT (Foster and Griffin Reference Foster and Griffin2018) and were not included in the analysis.
For R1/R2 exposure 7 DAT, 10 of the 13 variables met the 0.50 significance level for inclusion in the initial model (Table 2). The model for R1/R2 15 DAT included 11 of the 13 variables. Variables that did not meet the significance level for inclusion in the model were terminal leaf chlorosis, overall visible injury, and canopy height 7 DAT, and upper canopy leaf rollover/inversion and overall visible injury 15 DAT. Upper canopy leaf cupping, upper canopy pale leaf margins, lower leaf soil contact, and lower stem base swelling were not observed following R1/R2 exposure to dicamba 7 or 15 DAT (Foster and Griffin Reference Foster and Griffin2018).
To simplify the models, the number of variables was reduced based on high probability (Pr > F) and maximizing model R2 at each step in the forward-selection procedure. Variables selected for the V3/V4 7 DAT final model were plant height reduction, lower leaf soil contact, lower stem base lesions/cracking, canopy height, overall visible injury, and upper canopy leaf surface crinkling (Table 1). Probability level and model R2 were 0.0005 and 0.901, respectively, when the six variables were included. For the V3/V4 15 DAT final model, variables selected were lower stem base lesions/cracking, plant height reduction, terminal leaf epinasty, leaf petiole droop, leaf petiole base swelling, and stem epinasty (Table 1). Probability level and model R2 when the six variables were included were 0.0007 and 0.920, respectively.
Variables selected for the R1/R2 7 DAT final model were plant height reduction, lower stem base lesions/cracking, leaf petiole droop, upper canopy leaf rollover/inversion, leaf petiole base swelling, and stem epinasty (Table 2). Probability level and model R2 were 0.1401 and 0.915, respectively, when the six variables were included. For the R1/R2 15 DAT final model, variables selected were lower stem base lesions/cracking, terminal leaf chlorosis, leaf petiole base swelling, stem epinasty, terminal leaf necrosis, and terminal leaf cupping (Table 2). Probability level and model R2 when the six variables were included were 0.0114 and 0.936, respectively.
Validation experiments—data collected for variables
Soybean plants were actively growing when dicamba was applied at both growth stages and at both locations. For the V3/V4 and R1/R2 applications, rainfall of 18 and 72 mm, respectively, was received within 4 DAT at Baton Rouge and 32 and 40 mm, respectively, at St. Joseph (Table 3). Average maximum air temperature 0 to 4 DAT for the applications ranged from 31 C to 34 C and average maximum relative humidity from 93% to 98%.
V3/V4 Application 7 and 15 DAT models
Soybean plant height reduction (variable 1, 7 DAT model and variable 2, 15 DAT model) increased as dicamba rate increased from 0.6 to 280 g ha−1 from an average of 5% to 78% 7 DAT and from 19% to 100% 15 DAT (Table 4). Overall injury (variable 5, 7 DAT model) was 16% at 0.06 g ha−1 dicamba and increased to 92% at 280 g ha−1. The steady increase in both plant height and overall injury in response to increasing dicamba rate has been reported by others (Al-Khatib and Peterson Reference Al-Khatib and Peterson1999; Griffin et al. Reference Griffin, Bauerle, Stephenson, Miller and Boudreaux2013; Foster and Griffin Reference Foster and Griffin2018).
Data collected from validation experiments for the six variables selected for the V3/V4 soybean yield prediction models for 7 and 15 d after treatment s(DAT) with dicamba.a
a Data averaged for Baton Rouge and St. Joseph, LA. For V3/V4, model variables (Var) were: Var 1 = plant height reduction, Var 2 = lower leaf soil contact, Var 3 = lower stem base lesions/cracking, Var 4 = canopy height, Var 5 = overall visible injury, Var 6 = upper canopy leaf surface crinkling for 7 DAT, and Var 1 = lower stem base lesions/cracking, Var 2 = plant height reduction, Var 3 = terminal leaf epinasty, Var 4 = leaf petiole droop, Var 5 = leaf petiole base swelling, Var 6 = stem epinasty for 15 DAT.
b Plant height reduction and overall visible injury evaluated on scale of 0 = no height reduction/injury to 100 = plant death; other variables evaluated on scale of 0 to 5 with 0 = no injury, 1 = slight, 2 = slight to moderate, 3 = moderate, 4 = moderate to severe, and 5 = severe.
In contrast to what was observed for plant height reduction and injury, lower leaf soil contact (variable 2, 7 DAT model) and lower stem base lesions/cracking (variable 3, 7 DAT model and variable 1, 15 DAT model) were not observed for dicamba rates of 0.6 to 8.8 g ha−1 (Table 4). For 17.5 g ha−1 dicamba, injury level for lower leaf soil contact and lower stem base lesions/cracking was either 1 or 2 and increased to 4 and 5 for 35 to 280 g ha−1. Response to dicamba rate for both variables was in agreement with previous research used to generate the models (Foster and Griffin Reference Foster and Griffin2018).
Canopy height (variable 4, 7 DAT model) for the nontreated control averaged 38.5 cm and decreased 68% as dicamba rate increased to 280 g ha−1 (Table 4). Upper canopy leaf surface crinkling (variable 6, 7 DAT model) was 2 or 3 for 0.6 to 8.8 g ha−1 dicamba, but was not observed at ≥ 17.5 g ha−1 (Table 4). Injury ratings for terminal leaf epinasty, leaf petiole droop, leaf petiole base swelling, and stem epinasty (variables 3, 4, 5, and 6, 15 DAT model) in most cases were 0 or 1 for 0.6 and 1.1 g ha−1 dicamba, 2 or 3 for 4.4 and 8.8 g ha−1, and 4 and 5 for ≥ 17.5 g ha−1 (Table 4). Ratings for these variables in most cases were within 1 to 2 units of data used to develop the models (Foster and Griffin Reference Foster and Griffin2018).
R1/R2 Application 7 and 15 DAT models
Soybean plant height reduction (variable 1, 7 DAT model) increased from an average of 2% to 60% as dicamba rate increased from 0.6 to 280 g ha−1 (Table 5). Lower stem base lesions/cracking (variable 2, 7 DAT model and variable 1, 15 DAT model) was not observed for 0.6 to 8.8 g ha−1 dicamba 7 DAT and for 0.6 to 4.4 g ha−1 15 DAT (Table 5). Injury of 3 to 5 was noted for 70 to 280 g ha−1 7 DAT and for 17.5 to 280 g ha−1 15 DAT. For data used to develop the models, lower stem base lesions/cracking was not observed 7 or 15 DAT where dicamba was applied at 0.6 to 17.5 g ha−1, and injury was 3 to 5 at 7 and 15 DAT for 70 to 280 g ha−1 dicamba (Foster and Griffin Reference Foster and Griffin2018).
Data collected from validation experiments for the six variables selected for the R1/R2 soybean yield prediction models for 7 and 15 d after treatment (DAT) with dicamba.a
a Data averaged for Baton Rouge and St. Joseph, LA. For R1/R2, model variables (Var) were: Var 1 = plant height reduction, Var 2 = lower stem base lesions/cracking, Var 3 = leaf petiole droop, Var 4 = upper canopy leaf rollover/inversion, Var 5 = leaf petiole base swelling, Var 6 = stem epinasty for 7 DAT, and Var 1 = lower stem base lesions/cracking, Var 2 = terminal leaf chlorosis, Var 3 = leaf petiole base swelling, Var 4 = stem epinasty, Var 5 = terminal leaf necrosis, Var 6 = terminal leaf cupping for 15 DAT.
b Plant height reduction; height reduction evaluated on scale of 0 = no height reduction/injury to 100 = plant death; other variables evaluated on scale of 0 to 5 with 0 = no injury, 1 = slight, 2 = slight to moderate, 3 = moderate, 4 = moderate to severe, and 5 = severe.
For leaf petiole base swelling (variable 5, 7 DAT model and variable 3, 15 DAT model), injury level was 1 or 2 for 0.6 and 1.1 g ha−1 dicamba, 3 for 4.4 g ha−1, and 4 or 5 for ≥ 8.8 g ha−1 (Table 5). For leaf petiole droop (variable 3, 7 DAT model), injury level was 1 or 2 for 0.6 to 2.2 g ha−1 dicamba, 3 for 4.4 to 17.5 g ha−1, and 4 or 5 for ≥ 35 g ha−1 (Table 5). For upper canopy leaf rollover/inversion (variable 4, 7 DAT model), injury level was 1 or 2 for 0.6 to 35 g ha−1 dicamba and 3 or 4 for the higher rates (Table 5).
Stem epinasty (variable 6, 7 DAT model and variable 4, 15 DAT model) was 1 or 2 for 0.6 to 2.2 g ha−1 dicamba 7 DAT and for 0.6 and 1.1 g ha−1 15 DAT (Table 5). An injury level of 4 was observed for stem epinasty at 17.5 g ha−1 dicamba 7 DAT and at 4.4 g ha−1 15 DAT. Terminal leaf chlorosis (variable 2, 15 DAT model) injury level was 3 or 4 for 2.2, 4.4, and 8.8 g ha−1 and was 0 or 1 for the other rates (Table 5). Terminal leaf necrosis (variable 5, 15 DAT model) was 0 or 1 for 0.6 to 2.2 g ha−1 dicamba, 2 for 4.4 and 8.8 g ha−1, and 4 or 5 for the higher rates (Table 5). Terminal leaf cupping (variable 6, 15 DAT model) injury was 5 for dicamba at 0.6 to 2.2 g ha−1, 3 and 1 for 4.4 and 8.8 g ha−1, respectively, but injury was not observed for the higher rates (Table 5). High levels of terminal leaf cupping at the lower dicamba rates and slight to no injury at the higher rates were also observed for data used to generate the 15 DAT model (Foster and Griffin Reference Foster and Griffin2018).
Validation experiments—soybean yield
Data collected for the six variables specified by the models for experiments conducted in Baton Rouge and St. Joseph were used in multiple linear regression equations to predict soybean yield. Using the respective nontreated controls, predicted percent yield reduction for each dicamba rate was compared with observed percent yield reduction to evaluate accuracy of the models and consistency between locations.
V3/V4 7 DAT model
The soybean yield prediction equation for V3/V4 7 DAT is:
$$ { \hat Y\left( {{\rm{kg h}}{{\rm{a}}^{-1}}} \right) = [93.29 - 0.30 \left( {{\rm{plant \ height \ reduction}}} \right) \\ \quad - 3.77 \left( {{\rm{lower \ leaf \ soil \ contact}}} \right)\\ \quad -4 \left( {{\rm{lower \ stem \ base \ lesions}}/{\rm{cracking}}} \right) \\ \quad - 0.76 \left( {{\rm{canopy \ height}}} \right) \\ \quad - 0.27 ({\rm{overall \ visible \ injury}}) \\ \quad + 1.71 \left( {{\rm{upper \ canopy \ leaf \ surface \ crinkling}}} \right)]^\prime 67.2 } $$
Because canopy height is included in the equation, yield for the nontreated control is calculated by entering canopy height data for the nontreated control and entering zeros for the other variables where injury was not observed (Table 4). The conversion factor 67.2 is used to express yield in kg ha−1.
Observed yields for the nontreated controls for the V3/V4 application were 4,510 kg ha−1 at Baton Rouge and 5,480 kg ha−1 at St. Joseph. Using the V3/V4 7 DAT model, predicted percent yield loss for dicamba at rates of 0.6 to 8.8 g ha−1 underestimated observed yield loss 10 to 17 percentage points at Baton Rouge and by 9 to 16 points at St. Joseph (Table 6). The close agreement between locations was also observed for 35 g ha−1, where predicted yield loss overestimated observed yield loss by 32 and 28 points at Baton Rouge and St. Joseph, respectively. Differences between locations, however, were quite large for 17.5 g ha−1 dicamba, with predicted yield loss overestimating observed yield loss by 29 percentage points at Baton Rouge and underestimating observed yield loss by 1 percentage point at St. Joseph. Averaged over locations, the V3/V4 7 DAT model underestimated observed yield reduction by 11 to 16 percentage points for 0.6 to 8.8 g ha−1 dicamba.
V3/V4 15 DAT model
The soybean yield prediction equation for V3/V4 15 DAT is:
$$ \eqalign{\widehat Y( {\rm kg \ h{a^{-1}}} ) =\ {{[ {65.18-4.08 \,( {{\rm{lower \ stem \ base \ lesions}}/{\rm{cracking}}} )}}}\cr {{{-0.46 \,( {{\rm{plant \ height \ reduction}}} )}}} \cr {{{+ 5.38 \,( {{\rm{terminal \ leaf \ epinasty}}} )}}} \cr {{{-5.92 \,( {{\rm{leaf \ petiole \ droop}}} ) }}} \cr {{{+ 4.21 \,( {{\rm{leaf \ petiole \ base \ swelling}}} )}}} \cr {{{-3.77 \,( {{\rm{stem \ epinasty}}} )} ]} \times } \\ \,67.2}$$
For dicamba at 0.6 to 4.4 g ha−1, predicted percent yield reduction using the 15 DAT model underestimated observed yield reduction by as much as 6 points or overestimated by as much as 3 points at Baton Rouge (Table 6). At St. Joseph, predicted yield reduction for 0.6 to 4.4 g ha−1 dicamba was either the same as the observed yield reduction or predicted yield reduction overestimated observed yield reduction by as much as 4 points. For the two locations, predicted yield reduction overestimated observed yield reduction by 2 and 13 points for 8.8 g ha−1, by 16 and 23 points for 17.5 g ha−1, and by 26 and 27 points for 35 g ha−1. Averaged across locations, the V3/V4 15 DAT model for 0.6 to 4.4 g ha−1 dicamba underestimated or overestimated observed yield loss by 1 and 3 points. For 8.8 g ha−1 dicamba, predicted yield reduction overestimated observed yield reduction by an average of 8 points.
Difference in percent soybean yield reduction between predicted yield based on data collected using the V3/V4 7 and 15 d after treatment (DAT) models and observed yield for dicamba applied in the validation experiments.a
a Predicted yield determined using the prediction equations: Ⓨ (kg ha−1) = [93.29 – 0.30 (plant height reduction) – 3.77 (lower leaf soil contact) – 4.25 (lower stem base lesions/cracking) – 0.76 (canopy height) – 0.27 (overall visible injury) + 1.71 (upper canopy leaf surface crinkling)] × 67.2 for 7 DAT and Ⓨ (kg ha−1) = [65.18 – 4.08 (lower stem base lesions/cracking) – 0.46 (plant height reduction) + 5.38 (terminal leaf epinasty) – 5.92 (leaf petiole droop) + 4.21 (leaf petiole base swelling) – 3.77 (stem epinasty)] × 67.2 for 15 DAT.
b A minus (–) value implies that predicted value underestimated observed yield reduction, whereas a plus (+) value implies that predicted value overestimated observed yield reduction.
R1/R2 7 DAT model
The soybean yield prediction equation for R1/R2 7 DAT is:
$$\eqalign{{ \widehat Y( {{\rm{kg \ h}}{{\rm{a}}^{-1}}} ) = {[ {68.11-0.77 \ ( {{\rm{plant \ height \ reduction}}} )}}} \cr {{{\hskip-12.8pc -6.93 \ ( {{\rm{lower \ stem \ base \ lesions}}/{\rm{cracking}}} )}}}\cr \hskip-12.8pc{{{-1.60 \ ( {{\rm{leaf \ petiole \ droop}}} ) }}}\cr \hskip-12.8pc{{{+ 1.93 \ ( {{\rm{upper \ canopy \ leaf \ rollover}}/{\rm{inversion}}} ) }}}\cr {{\hskip-12.8pc{- 2.95 \ ( {{\rm{leaf \ petiole \ base \ swelling}}} ) }}}\cr {{{\hskip-12.8pc+ 1.78 \ ( {{\rm{stem \ epinasty}}} )} ] \,\:\!\times } 67.2 } }$$
Observed yields for the nontreated controls for the R1/R2 application were 4,520 kg ha−1 at Baton Rouge and 5,390 kg ha−1 at St. Joseph. Predicted yield reduction for dicamba at 0.6 to 8.8 g ha−1 underestimated observed yield reduction by 12 to 22 percentage points at Baton Rouge (Table 7). At St. Joseph for 0.6 to 8.8 g ha−1 dicamba, however, predicted and observed yield reduction was the same or predicted yield reduction overestimated observed yield reduction by 1 and 4 points or underestimated yield reduction by 7 points. For 17.5 and 35 g ha−1 dicamba, predicted yield loss was underestimated by 8 to 12 percentage points at the locations. The inconsistency in the performance of the R1/R2 7 DAT model between the locations for the lower dicamba rates is probably indicative of the intensity of symptom expression as related to environmental conditions during the 7 d following exposure. Averaged across locations, the R1/R2 7 DAT model underestimated observed yield loss by 4 to 9 percentage points for 0.6 to 4.4 g ha−1 dicamba. Yield loss was underestimated by an average of 10 to 15 points for 8.8, 17.5, and 35 g ha−1 dicamba.
Difference in percent soybean yield reduction between predicted yield based on data collected using the R1/R2 7 and 15 d after treatment (DAT) models and observed yield for dicamba applied in the validation experiments.a
a Predicted yield determined using the prediction equations: Ⓨ (kg ha−1) = [68.11 – 0.77 (plant height reduction) – 6.93 (lower stem base lesions/cracking) – 1.60 (leaf petiole droop) + 1.93 (upper canopy leaf rollover/inversion) – 2.95 (leaf petiole base swelling) + 1.78 (stem epinasty)] × 67.2 for 7 DAT and Ⓨ (kg ha−1) = [69.82 – 10.37 (lower stem base lesions/cracking) – 3.92 (terminal leaf chlorosis) – 4.68 (leaf petiole base swelling) + 3.90 (stem epinasty) – 2.46 (terminal leaf necrosis) – 1.70 (terminal leaf cupping)] × 67.2 for 15 DAT.
b A minus (–) value implies that predicted value underestimated observed yield reduction, whereas a plus (+) value implies that predicted value overestimated observed yield reduction.
R1/R2 15 DAT model
The soybean yield prediction equation for R1/R2 15 DAT is:
$${ \hat Y\left( {{\rm{kg \ h}}{{\rm{a}}^{-1}}} \right) = {[ {69.82-10.37 \left( {{\rm{lower \ stem \ base \ lesions}}/{\rm{cracking}}} \right) \\ \quad -3.92 \left( {{\rm{terminal \ leaf \ chlorosis}}} \right) \\ \quad - 4.68\left( {{\rm{leaf \ petiole \ base \ swelling}}} \right) \\ \quad + 3.90 \left( {{\rm{stem \ epinasty}}} \right) \\ \quad -2.46 \left( {{\rm{terminal \ leaf \ necrosis}}} \right) \\ \quad -1.70\left( {{\rm{terminal \ leaf \ cupping}}} \right) ] \times 67.2}}} $$
For dicamba at 0.6 to 4.4 g ha−1 using the R1/R2 15 DAT model, predicted yield loss underestimated observed yield loss by 2 to 5 percentage points at Baton Rouge, but overestimated observed yield loss by 8 to 12 points at St. Joseph (Table 7). For 8.8 g ha−1 dicamba, predicted yield loss underestimated observed yield loss by 14 points at Baton Rouge and by 2 points at St. Joseph. Averaged across locations, the R1/R2 15 DAT model overestimated observed yield loss by 3 to 5 percentage points for 0.6 to 4.4 g ha−1 dicamba. For 8.8 to 140 g ha−1 dicamba, the model underestimated average yield loss by as much as 8 points or overestimated yield loss by as much as 6 points.
In conducting the validation study the goal was to evaluate the models for consistency across experiments and for accuracy in predicting response. Consistency between validation experiments in predicting yield loss from dicamba at 0.6 to 4.4 g ha−1 was observed for the V3/V4 7 DAT, V3/V4 15 DAT, and R1/R2 15 DAT models (Tables 6 and 7). The rates of 0.6 to 4.4 g ha−1 dicamba are important, because in many cases soybean exposure from off-site movement of dicamba can be attributed to vapor drift representing 0.56 g ha−1 (0.1% of the applied rate of 560 g ha−1) (Egan and Mortensen Reference Egan and Mortensen2012; Grover et al. Reference Grover, Yoshida and Maybank1972) and to spray particle drift from an adjacent field representing 5.6 g ha−1 (1% of applied rate) (Brown et al. Reference Brown, Carter and Stephenson2004; Carlsen et al. Reference Carlsen, Spliid and Svensmark2006; de Jong et al. Reference de Jong, de Snoo and van de Zande2008; US EPA 2006; Wang and Rautman Reference Wang and Rautman2008).
In respect to accuracy in yield loss prediction for 0.6 to 4.4 g ha−1 dicamba, the V3/V4 7 DAT model underestimated observed yield loss by an average of 11 to 16 percentage points, the V3/V4 15 DAT model underestimated or overestimated observed yield loss by 1 and 3 percentage points, and the R1/R2 15 DAT model overestimated observed yield loss by 3 to 5 percentage points (Tables 6 and 7). Although yield loss prediction for the R1/R2 7 DAT model was inconsistent for the experiments for 0.6 to 4.4 g ha−1 dicamba, yield loss using the model underestimated observed yield loss by an average of 4 to 9 percentage points (Table 7). Hock et al. (Reference Hock, Knezevic, Martin and Lindquist2006) evaluated the performance of WeedSOFT (a free online tank-mix calculator) for predicting soybean yield loss from weed competition and selected a variation of 20% above or below the observed yield loss as an indicator of accuracy. For comparison, in the present study, level of accuracy for the V3/V4 and R1/R2 15 DAT models in predicting yield loss for 0.6 to 4.4 g ha−1 dicamba was within ±5 percentage points.
For the six-variable models used to predict yield in the validation study, lower stem base lesions/cracking was included in each of the models (V3/V4 and R1/R2 for 7 and 15 DAT) and was ranked as the first variable in the 15 DAT models (Tables 1 and 2). Soybean height reduction, leaf petiole base swelling, and stem epinasty were included in three of the models and leaf petiole droop in two of the models. Variables included in only one model were lower leaf soil contact, canopy height, overall visible injury, and upper canopy leaf surface crinkling (V3/V4 7 DAT); terminal leaf epinasty (V3/V4 15 DAT); upper canopy leaf rollover/inversion (R1/R2 7 DAT); and terminal leaf chlorosis, necrosis, and cupping (R1/R2 15 DAT).
Overall visible injury was not included as a variable in the 15 DAT models. For auxin herbicides where multiple symptoms are expressed, assigning a single injury rating on a 0 to 100% scale to represent multiple injury criteria would require that individual criteria be assigned a level of severity and ranked as to contribution to overall injury. Such a rating system would be subjective, and ratings would be expected to vary among individuals. Egan et al. (Reference Egan, Barlow and Mortensen2014), in summarizing meta-analysis from over seven decades of simulated drift experiments, reported that visible injury ratings made 12 to 16 d after dicamba exposure, as was most commonly reported, generally overestimated soybean yield loss, and that plants exposed to dicamba in the vegetative stage can generally recover from low to moderate injury.
Although presence of moderate to severe upper canopy leaf cupping and leaf crinkling from dicamba exposure are often used to predict soybean yield loss, neither of the variables were selected in the V3/V4 or R1/R2 soybean yield loss prediction models reported herein. Al-Khatib and Peterson (Reference Al-Khatib and Peterson1999) stated that symptoms from soybean exposure to dicamba that usually are worrisome to growers, such as cupping of terminal leaf, crinkling, and leaf stunting, occur at rates much lower than required to reduce yield. Weidenhamer et al. (Reference Weidenhamer, Triplett and Sobotka1989) reported foliar aberrations of crinkling and cupping of terminal leaves, leaf margin injury, and size reduction in soybean at dicamba rates as low as 0.06 g ha−1, much lower than needed for yield reduction.
In the present study, the models were developed using indeterminate maturity group 4.8 and 5.1 soybean cultivars exposed to dicamba under hot and humid growing conditions in Louisiana (Foster and Griffin Reference Foster and Griffin2018). In experiments used to develop the models and to validate the models, soybean irrigation and timely rainfall avoided drought stress, providing soybean plants the opportunity to recover from injury. Weeds were controlled and fungicides and insecticides were used as needed to maximize soybean yield potential. The models, however, are limited to V3/V4 or R1/R2 exposure and to injury assessment at no more than 15 d. Based on when soybean is planted in the mid-South and when at-planting and initial in-crop applications of dicamba are made, off-site exposure to sensitive soybean would most likely occur during the vegetative stage or at early flowering. Around 15 d is when injury expression is highly visible, and plant symptoms can become less noticeable over time, as a result of plant recovery or death (Griffin et al. Reference Griffin, Bauerle, Stephenson, Miller and Boudreaux2013). The yield loss prediction models described herein could be a useful tool for early-season evaluation of dicamba injury to susceptible soybean. An application for use in field assessment has been developed and is available through the LSU AgCenter.
Author ORCiDs
James L. Griffin https://orcid.org/0000-0003-1470-3535
Acknowledgments
Partial funding to support this research was provided by the Louisiana Soybean and Grain Research and Promotion Board. Appreciation is expressed to the LSU AgCenter Central Research Station, Ben Hur Research Farm in Baton Rouge, LA, and the Northeast Research Station in St. Joseph, LA, for providing land area to support this research. No conflicts of interest have been declared.
