Field trials were conducted near Ottosdal (26°47′ S, 25°56′ E; altitude 1,490 m), North-West Province, South Africa, during the 2017/2018 (Season 1) and 2018/2019 (Season 2) production seasons. The region has a semi-arid climate (BSk) with a mean annual rainfall of 447 mm (Kottek et al., 2006). Approximately 90% of the annual rainfall occurs in the summer growing season (October to April). Rainfall patterns are highly inconsistent between seasons and dry spells during the growing season are common phenomena.

Soil type is a hard-xanthic Plinthic Haplustox (Soil Survey Staff, 2003). Soil bulk density in the 0–60 cm soil depth was 1.6 g cm⁻³ at the onset of the trial in Season 1. Soil texture was sandy loam with organic matter content of 0.9%. The experimental site has been under no-tillage since 2011. Maize monoculture practices were followed in the field trial and soil cover was approximately 95% in the 2 months following harvest. Strong winds during winter removed a large portion of the crop residues resulting in a soil cover of 35%–40% on the day of planting in each season. Maize monoculture is a common practice across the summer grain production region of South Africa due to favorable markets and livestock feed needs during the winter months (Haarhoff et al., 2020).

Cumulative growing degree days (GDD) were calculated according to Gilmore and Rogers (1958) using daily air temperature data provided by the South African Weather Service. The GDD base temperature was set as 10°C. Air temperature was measured at a weather station approximately 10 km from the trial site. Rainfall was recorded at the trial site using a manual rain gage.

The experimental design was a randomized split-plot design with four blocked replicates. Whole-plots were row spacing (0.52 and 0.76 m), while plant population formed sub-plots, randomly nested within whole-plots (Table 1). These plant population configurations were chosen to achieve similar intra-row spacings in each row spacing as practiced by local producers. Plant populations of between 15,000 and 28,000 plants ha⁻¹ are currently established by rainfed producers under conventional tillage conditions across the local region.

Plot lengths were 20 m and plot width depended on row spacing. Plots with 0.52 m row spacing had 12 rows leading to plot widths of 6.2 m, while the 0.76 m row spacing plots had 10 rows leading to 7.6 m widths. Plots were overplanted at 65, 000 plants ha⁻¹ to ensure a high stand, and hand-thinned to the target plant populations at the fifth-leaf collar (V5) development stage (Ritchie et al., 1993), leaving a stand with uniform intra-row spacing in each treatment. The plots used in Season 1 were also used in Season 2 to include compounding effects of root biomass accumulation in the soil from the use of different plant densities. Maize plant density prior to Season 1 was 25, 000 plants ha⁻¹.

Representative soil samples were taken prior to planting to establish baseline chemical properties. In both seasons, nitrogen was broadcasted prior to planting as urea at 75 kg N ha⁻¹, while 14 kg N ha⁻¹ was band-placed as monoammonium phosphate at planting. Maize was planted by means of direct-drilling, using a 10-row John Deere 2117 no-tillage planter (John Deere Pty (Ltd.), Iowa, United States) and a six-row Jumil 2670-EX POP no-tillage planter [Jumil, Pty (Ltd.), Castelo, Espírito Santo, Brazil] in the 0.76 and 0.52 m row spacing plots, respectively.

The trials were established on 14 December 2017 and 4 January 2019 in Season 1 and 2, respectively. The optimal planting window for achieving maximum maize grain yield potential in the North-West province ranges between mid-November to mid-December. Early-autumn frost may occur at the end of April during kernel filling resulting in complete crop loss. Due to very hot conditions and low rainfall at the onset of Season 2, as recommended to farmers, planting was delayed beyond these dates. The 120-day Pioneer maize hybrid P2864WBR was used in both seasons (DuPont Pioneer Hi-Bred International). This hybrid was selected because it is one of the highest yielding cultivars in the region and commonly planted by local rainfed maize producers. Weeds were chemically controlled with pre-emergence herbicides after planting. Although weed pressure was low, hand-weeding was done throughout the growing seasons if necessary to keep plots weed free.

Total biomass was evaluated after emergence by randomly selecting five plants in each plot at 30, 60, 90, and 120 days after emergence (DAE). At least, 75% of plants reached the sixth-leaf collar (V6) stage at 30 DAE, tasseling (VT) at 60 DAE, the linear development phase of kernel filling (R3–R4) at 90 DAE, and physiological maturity (R5–R6) development stage at 120 DAE. Biomass samples were oven-dried at 60°C for 72 h to remove all moisture.

Intercepted photosynthetically active radiation (IPAR) and LAI were measured at VT, when maximum LAI was achieved, using an LP-80 AccuPAR ceptometer (Decagon Devices Inc., 2017). The 84 cm long probe was placed diagonally across two crop rows, with the two ends of the probe located in adjacent crop rows. This measuring regime is advised for row crops, as it provides a representative sample of the entire PAR environment below and between crop rows. The AccuPAR ceptometer calculates LAI based on the above and below-canopy measurements along with additional variables that relate to the canopy architecture and position of the sun. The IPAR and LAI measurements were done at five random spots within each plot above the leaf canopy (reference measurement, Qa) and at ground level (below-canopy measurement, Qb) between 12:00 and 14:00 on clear and windless days. The IPAR is reported as a percentage and was calculated using Equation 1:

Soil water content was monitored at 2- to 3-week intervals in Seasons 1 and 2 from planting until R5-R6. One galvanized access tube (length 120 cm, diameter 4 cm) was in-stalled per plot using a hand auger (diameter 4 cm) immediately after planting in the middle of two crop rows. A neutron probe (503DR Elite Hydroprobe Model, CPN Inc., Concord, CA, United States) was used to record soil water content at 30, 60, 90, and 120 cm soil depths. To calibrate soil water data, gravimetric soil samples were taken approximately 100 cm from the access tubes at planting (at the same time as the neutron probe readings) using a hand auger (diameter 7 cm) at soil layers 0–30, 30–60, 60–90, and 90–120 cm to determine gravimetric soil water content using the standard gravimetric method (Schmugge et al., 1980). The soil samples were oven-dried for 72 h at 105°C to remove all water. The gravimetric soil water content of each soil sample was converted to volumetric water content by multiplying by the soil bulk density. A linear regression of calibration readings against volumetric water values was calculated and used to calculate volumetric water content from the growing season soil water readings. Volumetric soil water content (cm⁻³ cm⁻³) was then converted to soil water content (mm) per layer by multiplying the volumetric soil water content by the soil layer depth (mm). Crop evapotranspiration (crop ET) was calculated as rainfall minus the change in soil water content (accumulated 0–120 cm soil depth) between subsequent measurements, minus drainage. Runoff was considered negligible as the experimental site is flat (<0.5% slope) and well drained. Water productivity for grain (WPg) and biomass production (WPb) were estimated by dividing maize grain yield and total biomass at R5–R6 by the seasonal crop ET (Hatfield and Dold, 2019).

Maize grain yield was determined by hand harvesting the full length of the center eight and six rows of the 0.52 and 0.76 m plots, respectively. Yield components were determined by randomly selecting 10 plants per plot at harvest. Grain samples were oven-dried at 60°C until constant weight and kernel weight was calculated by weighing a sample of 1,000 kernels. Harvest index was calculated by dividing maize grain yield by biomass as determined at R5–R6. All grain yield data were standardized to a moisture level of 12.5%.

Statistical analyses were performed by using Statistica version 13.5.0.17 (TIBCO Software, 2018). The Restricted Maximum Likelihood (REML) procedure was used to analyze according to the split-plot design. Three treatment factors were specified as fixed effects, i.e., plant population, row spacing and season, as well as the interaction between all three factors. Block, the interaction between block and plant population and block and row spacing were specified as random terms. The REML procedure was followed because the random factors of the dependent variables are also estimated, which allowed the evaluation of the effects of both row spacing and plant population as well as the interactions, despite dissimilar plant population treatments between the 0.52 and 0.76 m row spacings. Fisher’s least significant differences (LSD) test were conducted at a 5% significance level to determine whether interactions among the three factors of interest were significant. The Bonferroni correction test was used as validation of the Fisher’s LSD test to reduce the chances of obtaining false-positive results (type I errors), since multiple pairwise tests were performed on a single set of data. Normality of residuals and homogeneity of variances were tested and fulfilled the assumptions of the statistical model.

The total amount of rainfall during the growing period of Seasons 1 and 2 were 263 and 310 mm, respectively. The amount of rainfall for the 8 weeks prior to planting of trials in Seasons 1 and 2 were 83 and 62 mm, respectively. The distribution of rainfall during the cropping seasons was variable and dry spells occurred in both seasons (Figures 1A,B). Despite the late planting date in Season 2, average air temperature was comparable between seasons with cumulative growing degree days (GDD) totaling 1,404 and 1,386 from seedling emergence (VE) to R5–R6 in Season 1 and 2, respectively.

In Season 1, the total amount of rainfall from VE to 14-leaf collar (V14) was 149 mm, corresponding to a deficit of 70 mm compared to the 30-year average. In spite of the low rainfall during this period, soil water status was adequate and early vegetative growth was not affected by the prevailing growing conditions. A dry spell occurred from 57 to 88 DAE when plants were in the early reproductive development stages (VT to R3–R4). Maize plants across all treatments were under severe water stress, thereby negatively affecting kernel development. From 88 DAE onward, wet conditions prevailed with 102 mm received between R3–R4 and R5–R6 allowing satisfactory kernel filling.

Season 2 was characterized by challenging growing conditions from the onset of the season. Between VE and V14, a total of 138 mm of rainfall was received, with only two rainfall events recording more than 15 mm. Between V10 and R3–R4, a prolonged dry spell combined with high air temperatures occurred. Only 15 mm of rainfall was received between the V10 and R3–R4. At this point in the growing season, rainfall received was 130 mm below the 30-year average. Water-stress conditions negatively affected final vegetative growth, pollination, and ear growth across all treatments. Wet conditions and cool air temperatures characterized the period between R3–R4 and R5–R6, allowing maize plants to conclude the latter stages of kernel filling under stress-free growing conditions.

Both IPAR and LAI were affected by the interaction between row spacing and season (p < 0.05). In Season 1, IPAR was 6.8% greater at the 0.76 m row spacing compared to the 0.52 m row spacing (Table 2; p < 0.05). In Season 2, despite challenging growing conditions, a well-developed leaf canopy cover was present at 0.52 m row spacing indicating a 10.4% greater IPAR compared to the 0.76 m row spacing. No differences in LAI were observed between row spacings in Season 1 (p > 0.05), however, LAI was 21.8% greater at the 0.52 m row spacing compared to 0.76 m row spacing in Season 2 (p < 0.05).

The water content of the soil profile varied over the seasons as a result of variable crop uptake, evaporation rate, and rainfall occurrence (Supplementary Table S1). The soil water content varied with time due to variable plant uptake, evaporation rate, and seasonal rainfall occurrence. Soil water contents were similar at the start and end of the growing season between plant population treatments, irrespective of the row spacing. Also, the timing of water loss from the soil was similar between treatments throughout the growing season. This indicates evaporation at low plant population (< 38,000 plants ha⁻¹) due to poor leaf canopy was similar to the evapotranspiration rate at higher plant populations. Crop ET was only affected by the main effect of season and was 255 and 333 mm in Season 1 and 2, respectively (results not shown; p < 0.05).

Crop ET as a function of IPAR at the 0.52 and 0.76 m row spacings is illustrated in Figure 2. At the 0.52 m row spacing, there was a strong positive response of crop ET to IPAR in both seasons (r² > 0.8; p < 0.05). At the 0.76 m row spacing, there was a positive response of crop ET to IPAR in both seasons, however, this response was weak (r² > 0.3) in both seasons. Crop ET was greater for a given IPAR in the 0.76 m row spacing than in the 0.52 m row spacing, but both showed similar responses to increasing IPAR.

Total biomass at V6 was affected by the interaction of plant population and row spacing, without a season effect (p < 0.05). Total biomass at ≥50,000 plants ha⁻¹ was greater than at lower plant populations at a similar row spacing due to lower interplant competition combined with adequate soil water levels during the first 4 weeks following planting (Table 3; p < 0.05). At 0.76 m row spacing, total biomass at 50,000 plants ha⁻¹ was less than at 40,000 and 50,000 plants ha⁻¹ (p < 0.05).

Total biomass at the VT, R3–R4, and R5–R6 development stages was affected by the interaction between row spacing and season (p < 0.05). There was, however, no response of total biomass at R5–R6 to plant population indicating the trade-off associated with increased plant population in semi-arid environments when available soil resources are insufficient to address the greater demand at higher densities. At VT in Season 1, total biomass was 14% greater at 0.52 m than at 0.76 m row spacing (p < 0.05), while no differences were observed in total biomass between row spacings at R3–R4 and R5–R6 development stages (Table 4; p > 0.05). In Season 2, total biomass was lower at VT, R3–R4, and R5–R6 development stages with both row spacings compared to Season 1 (p < 0.05). Total biomass at 0.76 m row spacing was 32, 30, and 33% more than at the 0.52 m row spacing at the VT, R3–R4, and R5–R6 development stages, respectively.

Total biomass as a function of IPAR at 0.52 and 0.76 m row spacings is illustrated in Figure 3. In Season 1, there was a strong response of total biomass to IPAR at 0.76 m row spacing with increases of 468, 716, and 1,403 kg ha⁻¹ for each additional 10% of IPAR at the VT, R3–R4, and R5–R6 stages, respectively. However, at the 0.52 m row spacing, the response of total biomass was positive at the VT stage (452 kg ha⁻¹ per 10% increase in IPAR) but negative at the later growth stages.

Contrasting responses of total biomass to IPAR were observed in the drier Season 2. Total biomass at 0.52 m row spacing had a negative response to IPAR at VT, R3–R4, and R5–R6 (Figure 3). Although total biomass at 0.76 m row spacing had a positive response to IPAR at VT, the response was weak. At R3–R4 and R5–R6, a weak negative response of total biomass to IPAR was observed at 0.76 m.

Mean grain yield was considerably greater in Season 1 (8,119 kg ha⁻¹) than in Season 2 (7,162 kg ha⁻¹; Table 5). In Season 1, there were no yield differences between plant populations at the 0.52 m spacing or at populations less than 40,000 plants ha⁻¹ at the 0.76 m spacing. However, the crop at 0.76 m spacing and 50,000 plants ha⁻¹ yielded significantly more than all the other treatments. In Season 2 yield declined as population increased at the 0.52 m spacing, although the difference between yields at 50,000 and 60,000 plants ha⁻¹ was not significant (p > 0.05). Yield was greater at the 0.76 m spacing and there were no yield differences between plant populations (p > 0.05).

Kernel weight was similar across all treatments is Season 1 (Table 5; p > 0.05). In Season 2, kernel weight decreased with increasing plant population at 0.52 m row spacing (p < 0.05). Kernels per plant and grain yield per plant decreased with increasing plant population at both spacings and in both years although differences were not always significant (p < 0.05). Grain yield per plant decreased with increasing plant population at 0.52 and 0.76 m row spacing (Table 5; p < 0.05). In Season 1, grain yield per plant was higher at 0.76 m row spacing compared 0.52 m, with the opposite effect observed in Season 2 (p < 0.05). Harvest index remained constant across all treatments in Season 1, however, at 0.52 m row spacing in Season 2, harvest index decreased with increasing plant population (p < 0.05).

Neither crop ET nor WPb was affected by the plant population and row spacing treatments and were 24.58 and 51.85 kg mm⁻¹ in Season 1 and 2 across treatments, respectively (data not shown). The response of WPg to the treatments was similar to the response of grain yield. In Season 1, WPg ranged from 24.8 to 31.3 kg mm⁻¹, with differences between 25,000 and 60,000 plants ha⁻¹ at 0.52 m row spacing and between 50,000 and 20,000 and 40,000 plants ha⁻¹ at 0.76 m row spacing (Table 5; p < 0.05). In Season 2, WPg decreased (p < 0.05) with increasing plant population at 0.52 m row spacing, while WPg remained constant (p > 0.05) across plant population at 0.76 m row spacing. Treatment and seasonal effects on crop ET, WPb, and WPg during 2-week periods throughout the growing season were explored, however, no differences (p > 0.05) were found between treatments.

Grain yield as a function of crop ET at the 0.52 and 0.76 m row spacings is illustrated in Figure 4. In Season 1 at 0.52 m row spacing, a weak negative response of grain yield to crop ET was found, while a positive response in grain yield to crop ET was found at 0.76 m row spacing (Figure 4A). At 0.76 m row spacing, for each additional 10 mm of crop ET, maize grain yield increased by 651 kg ha⁻¹. In Season 2, a strong negative response of grain yield to crop ET was found at 0.52 m row spacing, while a weak positive response of grain yield to crop ET was found at 0.76 m row spacing (Figure 4B). At 0.52 m row spacing, for each additional 10 mm of crop ET, grain yield decreased by 945 kg ha⁻¹.

Grain yield as a function of IPAR at 0.52 and 0.76 m row spacings is illustrated in Figure 5. In Season 1, a negative response of maize grain yield to IPAR was observed at 0.52 m row spacing, while the opposite was true at 0.76 m row spacing (Figure 5A). In Season 2, a strong negative response in grain yield to IPAR was observed at 0.52 m row spacing, while grain yield could not be explained by IPAR at 0.76 m row spacing (Figure 5B).