Experimental Description

The effects of soybean row spacing, target densities, and cultivar on the CWFP were evaluated at 6 site-years in southern Manitoba. Field experiments were conducted during the 2016 and 2017 growing seasons at research farms near Carman (49.490348°N, 98.038191°W) and St-Adolphe (49.691525°N, 97.127181°W) and a producer’s field in the rural municipality of Whitemouth (2016: 50.024792°N, 96.036137°W; 2017: 50.001285°N, 96.050290°W). Fields were managed using fall and/or spring conventional tillage practices before establishment of the experiments. Soils at Carman were a fine loamy clay (Rignold series) and at St-Adolphe, experiments were conducted on a heavy clay soil (Scanterbury series). Experiments at Whitemouth were conducted in 2016 on a moderately fine loamy sand (Kiplin series), and in 2017 on the B horizon of a Baynham:Katimik:Stead (5:3:2) soil series, from which the organic A horizon had previously been removed (ca. 1970s). Other soil characteristics and the previous crop at each site year are listed in Table 1. At Carman and St-Adolphe, 40 kg ha⁻¹ of actual phosphate was applied before seeding to meet the fertility requirements for soybean in each year. At the Whitemouth locations, an unknown rate of liquid dairy manure was applied in 2013 where the 2016 field experiments were located and in both years preceding the 2017 field experiments. No additional fertilizer was applied to the soybean at the Whitemouth site-years. To facilitate symbiotic nitrogen fixation in soybean, a commercial peat-based granular inoculant containing 2 × 10⁹ viable cells of Bradyrhizobium japonicum (Kirch.) Jordan g⁻¹ (Cell-Tech, Monsanto BioAg, Winnipeg, MB, R3T 6E3, Canada) was applied in the seed row at a rate of 3.6 kg ha⁻¹ at the time of seeding.

Table 1.

Important field dates and soil characteristics for three locations in Manitoba over 2 yr.

^a If emergence occurred in June, an asterisk (*) was placed in front of the date. Previous crop refers to the preceding stubble at the site-year.

^b Residual nutrient status and soil characteristics reported for top 15 cm of the soil profile and were completed by AgVise Laboratories, Northwood, ND. Abbreviations used in this table reflect residual nitrates (NO₃⁻), residual phosphorous as determined by the Olsen test (Olsen-P), residual potassium (K₂0), residual sulfur (SO₄²⁻), organic matter (OM), and soluble salts (sol. salts).

^c Row-spacing experiment was reseeded on June 12 and its emergence occurred on June 26.

^d Alfalfa (Medicago sativa L.); Oat (Avena sativa L.); Wheat (Triticum aestivum L.).

Three different experiments were established at each location in each year of the study. The effects of row spacing were investigated in the first experiment, a second experiment investigated the effect of plant target density, and a third experiment evaluated the effect of soybean cultivar on the duration of the CWFP. The large size of these experiments and unknown spatial variability of resident weed populations precluded combining all treatments into a single experiment. Each of these three experiments was established as a randomized complete block design with a split-plot treatment arrangement. Experiments were composed of four replicated blocks with either two (row-spacing experiment) or three (soybean density and variety experiments) main plots with nine subplots (time of weed management) within each main plot. Subplot sizes were 2.5-m wide by 6-m long. In the row-spacing experiment, main plot treatments included narrow (19-cm) and wide (76-cm) soybean row spacing. In the soybean density experiment, target densities of 333,500, 444,600, and 666,900 plants ha⁻¹ (0.75X, 1.0X, and 1.5X, respectively, of locally recommended standard target densities) were compared, while the variety experiment compared three soybean cultivars from the same commercial breeding program with different maturity ratings and plant architecture/stature. GR soybean cultivars used for the variety experiment were ‘DeKalb® 22-60’ (DKB2260), ‘DeKalb® 23-60’ (DKB2360), and ‘DeKalb® 24-10’ (DKB2410) (Monsanto Canada, Winnipeg, MB, R3T 6E3, Canada). These soybean cultivars were chosen from one breeding program that differed in time to maturation and plant height. The cultivar DKB2260 is a short-statured plant that matures at about 2,275 CHU (corn heat unit); DKB2360 is a tall-statured plant that matures at 2,350 CHU; and DKB2410 is a medium- to tall-statured plant that matures at about 2,425 CHU. In the row-spacing experiment, DKB2360 was seeded at a density 444,600 plants ha⁻¹, and the same variety was used in the target density experiment. In the target density and variety experiments, all soybeans were seeded at an intermediate row spacing of 37.5 cm.

Five subplots were kept weed-free until a specific soybean developmental stage, with one season-long weedy and three season-long weed-free controls included. Developmental stages at which treatments were applied in these studies included soybean with expanded unifoliate leaves (VC), the specific number of fully expanded trifoliate leaves (V1, V2, or V4), and the beginning of the flowering period (R1) (Fehr et al. Reference Fehr, Caviness, Burmood and Pennington1971). Soybean development was recorded weekly until the R1 stage to approximate soybean development based on accumulated GDD₅ (growing degree days with a base of 5 C). In each subplot treatment, weed removal ceased at the specific soybean developmental stage, and the resident “natural” weed community was allowed to recruit and interfere with soybean for the remainder of the growing season. Weed removal until the designated soybean developmental stage was achieved by applying glyphosate (358 g ae ha⁻¹; Monsanto Canada, Winnipeg, MB R3T 6E3, Canada) and bentazon (889 g ai ha⁻¹; BASF Canada, Mississauga, ON L5R 4H1, Canada) in mixture. These active ingredients were chosen for their efficacious weed control and safety on soybean. Herbicides were applied using a bicycle-wheel push-type sprayer equipped with a 2-m boom set at 50 cm above the crop canopy. Four AirMix 110-01 (Greenleaf Technologies, Covington, LA 70433) nozzles were spaced at 50 cm along the boom and calibrated to apply 100 L water volume ha⁻¹ at 276 kPa.

Explanatory measurements of actual soybean density and soybean plant heights were taken at the V3 and R4 developmental stages, respectively. In each subplot, actual soybean density was determined by counting soybean along two 1-m lengths of row, while soybean heights were determined by measuring five individual plants at the highest point along the main stem. Aboveground weed shoot biomass was collected in lieu of weed density counts from two 0.09-m² quadrats within each subplot at the R5/R6 soybean developmental stage and oven-dried at 65 C until equilibrium. Weeds were sorted by species at time of collection and processed individually.

Air temperature and precipitation data were obtained from the Manitoba Ag-Weather program database (Manitoba Agriculture 2017) for all but the Whitemouth 2017 site-years (Table 2; Figure 1). Manitoba Agriculture recorded temperatures and precipitation at Carman and St-Adolphe in 2016 and 2017. Temperatures used for analysis at Whitemouth 2016 were obtained from the Environment Canada Pinawa weather station. At Whitemouth 2017, a weather station manufactured by ONSET Computer Corporation (Bourne, MA 02532) was installed directly in the field, adjacent to the experiments. Temperature was captured using a S-THB-M002 temperature and relative humidity sensor connected to a H21-001 HOBO field data logger, while a RG2-M rain gauge recorded precipitation. Long-term 30-yr weather averages were retrieved from Environment Canada. For determination of the CWFP, air temperatures were converted to GDD using a base temperature of 5 C beginning the day soybeans were seeded (Knezevic et al. Reference Knezevic, Evans, Blankenship, Acker and Lindquist2002).

Table 2.

Mean monthly air temperature during the growing season at three locations in Manitoba over 2 yr.

^a Calculated between 1981 and 2010 from data compiled by Environment Canada.

Figure 1.

Total monthly precipitation during the growing season at three locations in Manitoba over 2-yr. The 30-yr precipitation normal (yellow bar) was determined from data collected by Environment Canada between 1981 and 2010.

Statistical Analysis

Statistical analysis was performed using SAS Studio v. 14.1 (SAS Institute, Cary, NC 27513). To determine differences in weed-free soybean yields, actual soybean plant densities, soybean plant heights, and total weed biomass from weedy subplots, data were subjected to ANOVA using a linear mixed model (PROC MIXED) approach (Littell et al. Reference Littell, Stroup, Milliken, Wolfinger and Schabenberger2006). For each experiment and variable, fixed effects used in the linear mixed models were the main experimental treatment (row spacing, target density, or cultivar) and site-year. Location and year effects were also analyzed as fixed effects to determine the significance of their interaction and were combined when the interaction was significant. The duration of the weed-free period was included as a fixed effect when soybean heights and weed-free yields were analyzed to determine whether differences existed between the weed-free and R1 subtreatments. The experimental block was included as a fixed effect when analyzing the total weed biomass to determine whether weed biomass was uniform across each experiment within site-year. Random effects were the experimental block nested within site-year and the main treatment plot nested within the interaction between the experimental block and site-year. Conformation of the residuals to the Gaussian “normal” distribution was determined using the Shapiro-Wilk statistic (Littell et al. Reference Littell, Stroup, Milliken, Wolfinger and Schabenberger2006). Lund’s test (Lund Reference Lund1975) was used to determine whether extreme outliers were present and required further examination. Homoscedasticity was tested by visual inspection of residual versus predicted values (Kozak and Piepho Reference Kozak and Piepho2018) and was corrected using the group option in the repeated statement to minimize the Akaike information criterion (Littell et al. Reference Littell, Stroup, Milliken, Wolfinger and Schabenberger2006). Using the PDMIX800 macro (Saxton Reference Saxton1998), Fisher’s protected LSD at a significance level of 5% (α = 0.05) was used to separate the means.

Relative soybean yield was determined for each subplot treatment as its proportion of the mean weed-free yield in each main plot treatment within each block. To determine differences among main plot treatments within experiments within site-years, relative soybean yield was modeled to GDD₅ for each subplot treatment (developmental stage at which weed management was terminated) with the Gompertz function (Equation 1) using PROC NLMIXED in SAS described by and adapted from Knezevic et al. (Reference Knezevic, Evans, Blankenship, Acker and Lindquist2002). The procedure was repeated to determine differences among site-years within main plot treatments, and among main plot treatments with site-years combined. To determine differences between the intermediate and narrow or wide row spacings, an ex-post analysis was performed using the same procedure on the data from the row-spacing experiment and the intermediate row-spacing treatments of the same variety and densities from the density and variety experiments.

([1]) $${\rm Soybean \,\,yiel{d_{GDD}}(\% ) =} \,A \times {e^{ - B \times {e^{ - k \times \rm{GDD}}}}}$$

In Equation 1, parameter A describes the upper asymptote of the curve or, biologically, the maximum relative potential yield of the crop. The B parameter infers the curve’s inflection point. To determine the location of the inflection point on the abscissa (i.e., GDD₅), the value of the B parameter must be transformed to the natural-log scale and divided by k. This yields the inflection point of the Gompertz function in thermal time, which is of biological significance (Tjørve and Tjørve Reference Tjørve and Tjørve2017). Finally, the k parameter is the maximum slope of the curve, which is located at the inflection point, and refers to how quickly soybean seed yield reaches its potential maximum. All three parameters for main plot models were compared using single-degree-of-freedom estimates (Knezevic et al. Reference Knezevic, Evans, Blankenship, Acker and Lindquist2002). Within experiment, random effects used in the nonlinear mixed model were experimental block and the main plot treatment nested within the experimental block, and for the combined analysis, site-year, the experimental block nested within site-year, and the main plot nested within the interaction between experimental block and site-year. Initial parameters were optimized by choosing a set of values minimizing the negative log likelihood, following the sample code used by Coffey (Reference Coffey2016). In brief, a range of probable values was specified for each parameter and set for stepwise increases within that range. To achieve convergence of the procedure, three strategies were employed as necessary: (1) a bounds statement was invoked to keep variance estimates greater than or equal to zero, (2) the relative gradient convergence criterion was set to zero (Kiernan et al. Reference Kiernan, Tao and Gibbs2012), and (3) the optimization algorithm was set to either quasi-Newton, double-dogleg, Newton-Raphson, or conjugate-gradient (SAS Institute 2017). GDD values corresponding to 95% and 97.5% of potential maximum soybean yield were determined for each model by rearranging Equation 1 once models were built. These values reflect a 5% and 2.5% acceptable yield loss (AYL) and were then associated with the nearest soybean developmental stage at individual site-years.

Critical Weed-Free Period

Row spacing, target density, and cultivar were all effective at reducing the duration of the CWFP in soybean. Within experiments and site-years, differences among treatments were due to differences in the B parameters of the Gompertz function only (Table 3), while no differences were observed among the asymptote A or the slope k parameters. Parameter B, particularly in the absence of differences in the k parameter, reflects the inflection point of the Gompertz function (Tjørve and Tjørve Reference Tjørve and Tjørve2017) and thereby infers differences in the duration of the CWFP. Using these parameters, the CWFP was determined at the 5% AYL level, which is typical (Knezevic et al. Reference Knezevic, Evans, Blankenship, Acker and Lindquist2002), and at the 2.5% AYL level, which may be more pragmatic, as it more closely approaches the visual threshold used by practitioners and also accounts for the generally low cost of glyphosate (Stewart et al. Reference Stewart, Nurse, Van Eerd, Vyn and Sikkema2011).

Table 3.

Gompertz parameter estimates and contrast P-values for three cultural weed management experiments in soybean.

^a Model parameters (see Equation 1) were estimated using maximum-likehood approximation.

^b No statistical differences were observed between A or k parameters within experiment within site-year.

^c Soybean density of 1.0X is equivalent to 444,600 plants ha⁻¹.

Row-spacing Experiment

Seeding soybean in narrow rows reduced the duration of the CWFP compared with soybean seeded in wide rows (Table 3). This was observed in the combined analysis and at 3 of 6 site-years (Carman 2016 and 2017 and Whitemouth 2016). A similar trend was also observed at both St-Adolphe site-years, although these differences were not statistically significant. When all site-years were combined, the CWFP in soybean grown in narrow rows was shortened by 104 GDD at the 5% AYL level (Table 4; Figure 2A), which equated to about 1.5 soybean developmental stages. At the individual site-years, the narrow-row treatment shortened the CWFP by 85 to 94 GDD at the 2.5% AYL level and 77 to 156 GDD at the 5% AYL level compared with wide rows (Table 4). This range of GDDs corresponded to between one (Carman 2016) and three (Carman 2017) soybean developmental stages at both AYL levels. Studies in other regions have found similar results (Hock et al. Reference Hock, Knezevic, Martin and Lindquist2006; Nice et al. Reference Nice, Buehring and Shaw2001; Rasool et al. Reference Rasool, Mahajan, Yadav, Hanif and Chauhan2017), as wide-row spacing extends the period of light penetration into the canopy (Puricelli et al. Reference Puricelli, Faccini, Orioli and Sabbatini2003; Steckel and Sprague Reference Steckel and Sprague2004). In this experiment, the canopy did not close completely in the wide-row treatments at any of the site-years (data not shown). An open canopy facilitates weed seedling recruitment by lengthening the seedling recruitment period (Batlla and Benech-Arnold Reference Batlla and Benech-Arnold2014), and increases weed interference with the crop (Légère and Schreiber Reference Légère and Schreiber1989). In the narrow-row soybean treatments, canopy closure occurred between the third (V3) and fourth (V4) trifoliate leaf stages (data not shown).

Table 4.

End of the critical weed-free period for soybean in growing degree days and their corresponding developmental stages for three cultural weed management experiments.^a

^a Abbreviations: AYL, acceptable yield loss; GDD₅, growing degree days with a base of 5 C; SDS, soybean developmental stage; VC, unifoliate developmental stage; V1 to V5, first to fifth trifoliate developmental stage; R1, beginning of flowering stage. Bold values indicate significant differences among treatments within experiment based on B parameter contrast analysis found in Table 3.

^b Soybean density of 1.0X is equivalent to 444,600 plants ha⁻¹.

Figure 2.

Graphical representation of the critical weed-free period of soybean for three cultural weed management experiments in Manitoba. Site years were combined and Equation 1 ( $Soybean\,\,yiel{d_{GDD}}(\% ) = A \times {e^{ - B \times {e^{ - k \times GDD}}}}$ ) was fit to treatments within experiments. Parameter estimates and contrasts can be found in Table 3. Standard target density (1.0 X) was 444,600 soybean plants ha⁻¹. Abbreviation: AYL, acceptable yield loss.

A combination of site-specific factors appeared to contribute to the detection of differences in the CWFP in response to soybean row spacing. At the 3 site-years at which the CWFP was affected by row spacing, peak-season weed biomass was greatest (Figure 3), indicating the necessity of a competitive weed community for observing significant effects on the CWFP. The 3 site-years at which the wide-row spacing increased the duration of the CWFP (Table 3) also had a number of similarities among their soil parameters, including the lowest soil organic matter (OM) content, the lowest spring Olsen-P levels, and among the lowest spring mineral nitrogen content, and soybean were seeded earlier and emerged earlier at these sites compared with others (Table 1). Low soil OM and low nutrient status are often associated with poor crop productivity; however, low soil nitrogen (Geddes and Gulden Reference Geddes and Gulden2018) and early planting (Lenssen Reference Lenssen2008) would be expected to shift the competitive balance toward the nitrogen-fixing soybean crop. This did not happen, and therefore other factors clearly also influenced our observations. In Manitoba, soybean do not respond to soil Olsen-P levels (Bardella Reference Bardella2016).

Figure 3.

Aboveground weed shoot biomass for three soybean cultural weed management experiments in Manitoba. Within experiment, least-squares mean separation is indicated by different letters above columns according to Fisher’s protected LSD (P < 0.05)

In addition, these 3 site-years and St-Adolphe 2016 had significantly lower soybean plant stand densities at V4 in the wide-row treatments (Table 5). However, while it appears that this may have influenced the results, using actual stand density as a covariate when modeling the CWFP did not affect the interpretation of the results. At the 3 significant site-years, soybean densities in the wide-row-spacing treatments were 62% to 75% of those observed in the narrow-row treatments. At St-Adolphe 2016, where row spacing did not affect the CWFP, soybean stand density in the wide-row treatment was only 57% that in the narrow-row treatment. Seedling or plant attrition is commonly observed in wide-row production systems (De Bruin and Pedersen Reference De Bruin and Pedersen2008; Weiner and Freckleton Reference Weiner and Freckleton2010). At St-Adolphe and Whitemouth 2017, soybean stand densities did not differ between row-spacing treatments (Table 3). Taken together, a combination of edaphic factors, site-specific self-thinning, and site-specific weed community characteristics appeared to contribute to the efficacy of the row-spacing treatments on the CWFP. The importance of edaphic factors, including soil nutrient status (Geddes and Gulden Reference Geddes and Gulden2018; Mohammadi and Amiri Reference Mohammadi and Amiri2011) and weed biomass (Martin et al. Reference Martin, Van Acker and Friesen2001; Van Acker et al. Reference Van Acker, Weise and Swanton1993b), to the outcome of weed–crop interference is well established.

Table 5.

Actual soybean plant densities for three cultural weed management experiments in Manitoba.^a

^a Fisher’s protected LSD (P < 0.05) was used for least-squares mean letter separation. Letters are presented beside the means if the treatment by site-year interaction was significant.

^b Soybean density of 1.0X is equivalent to 444,600 plants ha⁻¹.

The intermediate row–spacing (37.5-cm) treatment from both the density and the variety comparison experiments with the same target density and variety were compared with the narrow (19-cm) and wide (76-cm) row–spacing treatments to better understand the effect of row spacing on the duration of the CWFP. In this ex-post analysis, intermediate soybean row spacing shortened the CWFP by up to 90 GDD₅ (5% AYL) compared with wide-row spacing (B parameter P = 0.0493). No differences were observed between the 19-cm and 37.5-cm soybean row spacing. These results must be interpreted with caution, however, as these experiments were not designed for direct comparison among these treatments. The large size of the individual experiments led to spatial variability in the resident weed communities among experiments at some of the site-years (Figure 3), which may have influenced these results.

Target Density Experiment

The combined analysis revealed that the low (0.75X) soybean plant densities lengthened the CWFP compared with standard (1.0X) or increased (1.5X) soybean plant densities (Table 3; Figure 2B). No differences in the CWFP were observed between the 1.0X and 1.5X plant densities in the combined analysis or in the individual experiments (Table 3). Overall, the CWFP was extended by 85 and 118 GDD in low-density soybean stands compared with the standard and increased stand densities, respectively, at 5% AYL (Table 4; Figure 2B), which corresponded to roughly two soybean developmental stages. At the individual site-years, low soybean plant densities extended the CWFP from 57 to 240 GDD at the 2.5% AYL level and 64 to 179 GDD at the 5% AYL level in the lowest- compared with the highest-density treatments (Table 4). This corresponded to between one and two soybean developmental stages. An extended CWFP between the low and standard soybean density treatments was found only in the combined analysis, but not in any of the individual experiments. It is unclear why this occurred. In the individual density experiments, the CWFP was affected by soybean density at 3 of 6 site-years (Table 3). Two of these 3 site-years also showed a significant row-spacing effect on the CWFP. The importance of plant densities to crop productivity, particularly under weed interference, is well known in soybean (Arce et al. Reference Arce, Pedersen and Hartzler2009; Nice et al. Reference Nice, Buehring and Shaw2001) and other crops (Ball et al. Reference Ball, Ogg and Chevalier1997; Fradgley et al. Reference Fradgley, Creissen, Pearce, Howlett, Pearce, Döring and Girling2017; O’Donovan et al. Reference O’Donovan, Newman, Harker, Blackshaw and Mcandrew1999). The dose–response work by Redlick et al. (Reference Redlick, Duddu, Syrovy, Willenborg, Johnson and Shirtliffe2017a) eloquently showed the trade-off between plant densities and herbicides for effective weed management and the importance of plant stand densities for improving herbicide efficacy and reducing the necessity for herbicides for weed management in lentil (Lens culinaris Medik.) (Redlick et al. Reference Redlick, Syrovy, Duddu, Benaragama, Johnson, Willenborg and Shirtliffe2017b). Increasing soybean plant densities by 50% above the standard density did not translate into a shortening of the CWFP. This was surprising, as increasing soybean densities several-fold above standard densities resulted in continued improvements to the competitive effect and competitive response in soybean when subjected to interference from volunteer B. napus, the dominant weed in soybean in western Canada (Mierau et al. Reference Mierau, Johnson, Gulden, Weber, May and Willenborg2019). The findings here suggest an upper limit to the effect of soybean density on the CWFP in soybean; it must be noted that at most site-years, the CWFP at the standard soybean plant densities was short (V1–V2) with little room for further improvement (Table 4). While increasing soybean target densities above the standard treatment did not result in a more competitive crop, when weeds were not controlled, seed yields in the highest soybean plant densities were 11% and 24% greater than the standard and lowest target densities (ANOVA P = 0.0002), respectively. Therefore, high soybean densities contributed to the competitive response in soybean without contributing to the competitive effect.

Unlike in the row-spacing experiments, the 3 site-years where the CWFP was affected by soybean density do not share similar edaphic characteristics. The soil characteristics at St-Adolphe 2016 were quite different from those of the 2 other site-years where an effect was observed (Table 1). In contrast to the row-spacing experiments, significant differences in the CWFP were observed at site-years with the lowest inherent midseason weed biomass (Figure 3), where soybean in the weed-free treatments at the R4 stage were tallest (Table 6) and where differences in stand densities between the greatest and lowest soybean densities at the R4 developmental stage were largest (287,000 to 340,000 plants ha⁻¹ vs. 210,000 to 260,000 plants ha⁻¹) (Table 5). Soybean plant height under weed-free conditions was not different among plant density treatments, which in conjunction with low weed biomass suggests limited shade avoidance response (Green-Tracewicz et al. Reference Green-Tracewicz, Page and Swanton2012). Therefore, it seems that differences in the CWFP in response to soybean density were associated with site-years at which resource limitations and weed interference were lowest (Table 7).

Table 6.

Mean soybean plant heights for three cultural weed management experiments in Manitoba.^a

^a Fisher’s protected LSD (P < 0.05) was used for least-squares mean letter separation. Letters are presented beside the means if the treatment by site-year interaction was significant.

^b Soybean density of 1.0X is equivalent to 444,600 plants ha⁻¹.

Table 7.

Mean weed-free soybean yields for three cultural weed management experiments in Manitoba.^a

^a Fisher’s protected LSD (P < 0.05) was used for least-squares mean letter separation. Letters are presented beside the means if the treatment by site-year interaction was significant.

^b Soybean density of 1.0X is equivalent to 444,600 plants ha⁻¹.

Variety Experiment

The effect of soybean cultivar on the CWFP was unique to each location and consistent over the 2 study years at Carman and at Whitemouth (Table 3). At Carman, the duration of the CWFP was lengthened in the most rapidly maturing soybean cultivar, DKB2260, compared with the other two cultivars, DKB2360 and DKB2410. No differences in the duration of the CWFP were observed between DKB2360 and DKB2410 at Carman in both years. Similar results were observed in the combined analysis, in which the CWFP was 14 and 77 GDD longer for DKB2260, the quickest-maturing variety in this selection, compared with DKB2360 and DKB2410, respectively (Table 4; Figure 2C). This equated to about one soybean developmental stage. Shorter-season cultivars develop more rapidly than longer-season cultivars (Cober and Morrison Reference Cober and Morrison2011; Morrison et al. Reference Morrison, Voldeng and Cober1999; Place et al. Reference Place, Reberg-Horton, Dickey and Carter2011c). In these experiments, DKB2260 began anthesis up to 7 d earlier and matured up to 10 d before the other two cultivars. In addition, row closure occurred at a later developmental stage in DKB2260 (R1/R2 vs. V4 for the other two cultivars), and when different, DKB2260 plants were shorter in maximum height (Table 6) than DKB2360 or DKB2410 plants. These characteristics, in combination with the high weed biomass at Carman (Figure 3), likely contributed to these results at this location and overall. At Whitemouth, soybean cultivar had no effect on the CWFP in either year (Table 3). The differences in soybean plant height among the three soybean cultivars at Whitemouth were greater than at the other locations (Table 6), and therefore, plant height did not appear to be a key factor contributing to the duration of the CWFP. Reasons for the lack of a cultivar effect on the CWFP on soybean at Whitemouth are not clear.

Inconsistent results were observed with the CWFP of DKB2360 at the St-Adolphe location. In 2016, the CWFP in DKB2360 was up to 67 GDD longer than in the other two cultivars. The opposite was observed in 2017, when the CWFP in DKB2360 was 176 and 272 GDD shorter than in DKB2260 and DKB2410, respectively (Table 4). DKB2360 was taller at R4 than both other cultivars at all site-years except at St-Adolphe 2016, where it grew to the same height as the other cultivars (Table 6). None of the other measured morphological characteristics showed an obvious link to the observed difference in the CWFP in this cultivar at this site-year. Favorable environmental conditions at St-Adolphe in 2016 triggered a large weed seedling recruitment event 6 d before soybean emergence. Earlier weed emergence may have intensified weed interference (Korres et al. Reference Korres, Norsworthy and Mauromoustakos2019; Van Acker et al. Reference Van Acker, Weise and Swanton1993b) with soybean at this site-year. It is possible that DKB2360 is more sensitive to more intense early weed interference than the other two cultivars, which consequently would have led to a longer CWFP under these circumstances in 2016. In 2017, uncharacteristically dry spring soil conditions delayed soybean and weed emergence at this site-year until June 17 and June 22, respectively.

Thousand-seed weights among the soybean cultivars were not the same, and seed lots could not be standardized for this difference. DKB2260 consistently had a smaller average seed size (157.4 g 1,000-kernel weight [TKW]) than DKB2360 (185.9 g TKW) and DKB2410 (187.4 g TKW). Large seed size has been directly associated with increased seedling vigor (Fatichin et al. Reference Fatichin, Zheng and Arima2013; Place et al. Reference Place, Reberg-Horton, Carter and Smith2011b, Reference Place, Reberg-Horton, Dickey and Carter2011c) and the competitive response to weed interference in soybean (Jannink et al. Reference Jannink, Orf, Jordan and Shaw2000; Jordan Reference Jordan1993; Place et al. Reference Place, Reberg-Horton, Dickey and Carter2011c) and wheat (Triticum aestivum L.) (Stougaard and Xue Reference Stougaard and Xue2004). Seed size may have contributed to the results of this study, although the results at some of the individual locations indicate that other factors were more influential in determining the CWFP among soybean cultivars.

Weed Effects

Among site-years, the greatest accumulation of midseason weed biomass was generally observed at Carman (Figure 3), and this contributed to the greatest soybean yield loss in weedy treatments. At site-years with very low midseason weed biomass, minimal soybean yield loss was observed. Among experiments within site-years, weed biomass differed between treatments in the target density experiment only. High soybean density decreased weed shoot biomass by 44% compared with the low soybean density (P = 0.0398). Reductions in weed biomass are known to occur as a result of increased densities of soybean (Liebert and Ryan Reference Liebert and Ryan2017; Place et al. Reference Place, Reberg-Horton, Dunphy and Smith2009). The lack of differences among treatments in the variety and row-spacing experiments was likely due to differences in the range in midseason weed biomass observed among experiments within site-years. In the variety experiment, about a 4-fold difference in the midseason weed biomass was observed among treatments within site-years, while in the row-spacing and plant density experiments, this range expanded to about 7- and 10-fold, respectively. Total weed shoot biomass differed among experiments at Whitemouth 2016 and Carman 2017 (data not shown). At each of these site-years, weed biomass was greatest in the row-spacing experiment followed by the variety experiment and then the density experiment. At Whitemouth 2016, a heavy rainstorm during the spring weed seedling recruitment period influenced weed species recruitment and contributed to spatial variation among these experiments, with decreased weed seedling recruitment in the density experiment due to extended waterlogged soil conditions. Soybean performance appeared not to be affected by this event. At Carman 2017, poor soybean emergence in the row-spacing experiment resulted in the experiment being reseeded a month later. A shallow-tillage pass was used to terminate the initial experiment, which, combined with timely precipitation and hot temperatures, increased weed recruitment compared with the other established experiments.

In addition to midseason weed biomass, differences in the midseason weed communities among these experiments may have also contributed to the observed differences. Weed species richness among site-years ranged from 7 species at Carman 2016 to 14 species at Whitemouth 2017, and dominant weed species were not the same among the site-years. Warm-season grasses such as barnyardgrass [Echinochloa crus-galli (L.) P. Beauv.], yellow foxtail [Setaria pumila (Poir.) Roem. & Schult.], and green foxtail [Setaria viridis (L.) P. Beauv.] were dominant and ubiquitous at Carman and St-Adolphe in both years, but not at Whitemouth. Avena fatua, a highly competitive cool-season grass (Beckie et al. Reference Beckie, Francis and Hall2012; Rathmann and Miller Reference Rathmann and Miller1981), was a dominant grassy weed species at St-Adolphe and Whitemouth 2016. This was not unexpected, as S. viridis and A. fatua have been the dominant grassy midseason weeds in the Canadian NGP for decades (Leeson et al. Reference Leeson, Thomas, Hall, Brenzil, Andrews, Brown and Van Acker2005). Due to a greater array of broadleaf weed species, dominant broadleaf weeds varied more among site-years than grasses. In 2016, the dominant broadleaf weeds at Carman and Whitemouth were redroot pigweed (Amaranthus retroflexus L.) and wild buckwheat [Fallopia convolvulus (L.) Á. Löve], while at St-Adolphe, the primary broadleaf weeds were pale smartweed [Persicaria lapathifolia (L.) Delarbre] and ladysthumb (Persicaria maculosa Gray). In 2017, common lambsquarters (Chenopodium album L.) was observed at all locations, albeit sparsely. At Whitemouth, A. retroflexus and shepherd’s-purse [Capsella bursa-pastoris (L.) Medik.] were the prominent broadleaf weeds. Amaranthus retroflexus and other members of the genus have been found to be highly competitive with soybean (Butts et al. Reference Butts, Vieira, Latorre, Werle and Kruger2018; Légère and Schreiber Reference Légère and Schreiber1989; Van Acker et al. Reference Van Acker, Weise and Swanton1993b), which likely influenced the results in 2016. At St-Adolphe in 2017, few broadleaf weed species were observed; nevertheless low densities of volunteer B. napus were present in the row-spacing experiment. Brassica napus can be highly competitive with soybean (Geddes and Gulden Reference Geddes and Gulden2018; P Gregoire, personal communication).

Soybean Weed-Free Yield

Weed-free yields were different among site-years and reflect differences in the carrying capacities, which may also have contributed to the sensitivity of the CWFP in soybean. Overall, greater soybean yields were observed in 2016 than 2017 in all experiments (Table 6), as 2017 was an uncharacteristically dry growing season (Table 2). In 2016, narrow-row soybean produced between 19% and 29% more seed yield than wide-row soybean. Lodging occurred in the wide-row soybean treatment during seed fill at St-Adolphe in 2016 and may have influenced weed-free yields in this experiment (Cooper Reference Cooper1971). No differences in soybean yield were observed between the row-spacing treatments in 2017. In the target density experiment, soybean yield in the low-density treatment was 5% and 8% lower than the standard and high-density treatments. These results agree with other studies that observed lower soybean seed yield at decreased population densities (Cox and Cherney Reference Cox and Cherney2011; De Bruin and Pedersen Reference De Bruin and Pedersen2008). No yield differences were observed between the standard and high-density treatments, suggesting the law of constant final yield was operative in the weed-free treatments in these experiments (Weiner and Freckleton Reference Weiner and Freckleton2010). In the variety experiment, differences in soybean seed yield among cultivars were only observed at Whitemouth, the northernmost location of these studies. In 2016 at this location, DKB2360 produced 9% and 33% greater yield than DKB2410 and DKB2260, respectively, and DKB2410 produced 26% and 16% greater yield than DKB2260 in 2016 and 2017, respectively.

Determination of the CWFP using nonlinear mixed models proved highly effective at elucidating the effects of cultural practices on the competitive ability of soybean and the potential selection pressure for HR weeds. Overall, the range of soybean row-spacing treatments in these experiments appeared to be more effective at reducing the CWFP (up to three soybean developmental stages) than the range of soybean densities (up to two soybean developmental stages). The duration of the CWFP, however, was more extended (V3 to V5) in the row-spacing experiments than in any of the density treatments (maximum V2). Compared with row spacing and plant density, the effect of variety on the CWFP was relatively small (fewer than two soybean developmental stages). While potentially interesting, comparisons of results among experiments should be interpreted with caution. All three experiments were adjacent to each other at each site-year, yet these were large experiments and inherent spatial variation in midseason weed biomass (Figure 3) and community composition were observed among site-years and experiments within site-year. Implementation of these cultural practices was shown to reduce the CWFP in soybean by up to several developmental stages in the NGP. While the cultural weed management practices were evaluated separately in these experiments, they are generally more effective when used in combination (Swanton et al. Reference Swanton, Mahoney, Chandler and Gulden2008). Economic (i.e., equipment and seed costs), social (i.e., “my neighbor does it that way”), and technical (i.e., lack of information on weed competitiveness of specific cultivars) factors play a role in why primary producers have not more readily adopted these practices in this region or elsewhere. Combining these techniques (narrow-row spacing and standard or increased plant densities with a regionally appropriate cultivar choice) may further shorten the CWFP in soybean grown in the NGP, thereby potentially further reducing the need for herbicides and the selection pressure for HR weed biotypes.