Efficiency of visual selection of agronomic traits for soybean production in a protected environment

The objectives of this work were to evaluate the efficiency of visual selection in soybean under a protected environment, to determine if the ability of an evaluator to select visually depends on their experience with the crop, and to study the efficiency of visual selection of traits associated with productivity to obtain income gains. For that, three soybean populations (F3, F4, and F5) were used, composed of 20 families each, conducted in a greenhouse in a randomized block design, with three replicates and experimental units, two plants per pot. In these, five evaluators with different levels of experience with the crop made the visual selection for each plant. Evaluators differ in their ability to identify more productive plants, but this ability is not associated with the evaluator's experience with the crop. In the search for more productive soybean plants, the number of pods per plant is the characteristic that most assists visual selection, provided there is uniformity of experimental conditions and genetic variability.


| INTRODUCTION
The main goal of most soybean breeders is to raise the productive potential of the species by selecting superior individuals. Among the practices used to increase productivity, visual selection is the oldest, simplest, and probably the most used method in plant breeding over time. The ability of breeders to visually identify genotypically superior individuals (Weatherwax, 1954) is used.
One of the first discussions on visual selection occurred in the work published by Engledow and Wadham (1925). They found that visual selection depends on the breeder's experience, which was corroborated by Bowman et al. (2004) and Salmon and Larter (1978).
However, Townley-Smith and Hurd (1973) found no difference in the abilities of experienced and inexperienced selectors.
Several reports in the literature indicate that visual selection is of low efficiency. Shebeski (1967) has shown that in F 2 populations of wheat, visual selection was no better than casual selection. Dahiya et al. (1984) verified in F 3 chickpea populations that the yield of visually selected families was not higher than those obtained by chance. Marques et al. (2000), when evaluating populations of half-siblings, verified that the efficiency of visual selection is small on average for maize crops.
However, other results demonstrate the efficacy of the method. Byth et al. (1969) reported that visual classification was useful in identifying soybean F 3 lines with high grain yield under favorable growth conditions. In cotton F 4 lines, Bowman et al. (2004) identified correlations between visual classification and real cotton lint yield of genotypes evaluated by up to 70%.
Among the factors that contribute to the lack of coherence between the results of the studies that seek to quantify the efficiency of visual selection is the heterogeneity of the experimental area, in which the more heterogeneous the area is, the greater the probability of visual selection is not efficient (Gardner, 1961;Méndez, 1971;Molina, 1983). An alternative to improve this strategy would be to conduct the populations to be evaluated in a greenhouse with a more uniform environment.
Another factor to consider is that classic soybean breeding programs can take 8 to 12 years between crossing and the release of a new variety . However, several modern techniques, known as "speed breeding," make it possible to reduce this time to 4 to 7 years, by reducing the breeding cycle, which is the time between the first crosses and the use of cultivars (Ceccarelli, 2015).
Among these techniques is the use of nurseries in low-latitude regions and/or a greenhouse with a controlled environment that allows two to four cycles of self-fertilization per year. It is still common, under these conditions, to speed up the program, the selection of individuals within segregating populations in early generations (Bernardo, 2010;Silva et al., 2017), and in many cases, visual selection is performed, based on phenotypic aspects, between and among within, or only within, segregating populations.
Thus, despite the suspicions of efficacy reported by Gardner (1961), Méndez (1971), and Molina (1983), and of being applied in several soybean breeding programs, the efficiency of visual selection of soybean plants in a controlled environment still needs to be better elucidated.
Considering that the mathematical model that explains the grain yield in soybean is multiplicative: grain yield = total number of pods per plant Â average number of grains per pod Â average weight of grain, increases in yield can be obtained by selecting genotypes with a higher number of grains per pod or with more pods per plant, which in turn can be obtained with an increasing number of lateral branches.
Significant effects of the influence of the primary yield components on soybean productivity were reported by Bárbaro et al. (2007), Iqbal et al. (2003), Khan et al. (2000), and Nogueira et al. (2012). However, there have been no reports of the efficiency of visual selection on these traits aiming to increase soybean yield.
The objectives of this study were to evaluate the efficiency of visual selection in soybean under uniform environmental conditions, to determine if the ability of an evaluator to select visually depends on their experience with the crop, and to study the efficiency of visual selection of traits associated with productivity to obtain yield gains.

| MATERIAL AND METHODS
To carry out this study, three soybean populations, F 3 , F 4 , and F 5 , Pest and disease control was carried out according to the recommendations for soybean cultivation (Sediyama et al., 2015). The experimental precision was evaluated by the estimation of the accuracy presented by Resende (2007), given by the estimator:  height and number of lateral branches, feasibility for mechanized harvesting, and resistance to early threshing (Hamawaki et al., 2006).
The plants were harvested individually. The number of pods, the average number of seeds per pod, the number of lateral branches, and the grain weight (g/plant) were obtained from each plant. The plants with the highest grain weight were identified and used to estimate the coincidence with the data obtained by the evaluators.
A common method to evaluate the efficiency of visual selection has been to compare the level of coincidence of the selected plants with the individuals with the highest real yield. This strategy was used by Bowman et al. (2004), Cutrim et al. (1997), and Dahiya et al. (1984). However, the approach may not be the best one, since two individuals may present very similar weights and, since the selection is made in absolute value, the coincidence will not be complete.
Thus, to estimate the capacity of the different visual selection alternatives to identify plants with higher grain weight, the linear regression method was adopted with the model where Yi is the number of genotypes with greater grain weight in the proportions of 10%, 20%, 30%, and 40% in population i; Xi is the number of coincident plants obtained by visual selection with Yi in the referred proportions; β0 is the intercept; β1 is the slope of the regression line; and Ԑi is the random error of simple linear regression.
Regression analyses were performed using Genes software (Cruz, 2013).

| RESULTS AND DISCUSSION
All the traits studied showed significant genotypic effects in the populations in which they were evaluated, indicating that the selective process can be successfully performed (Table 1).
The coefficient of experimental variation (CVe) is the most commonly used way to measure an experiment's quality. Except for the number of seeds per pod in the F 5 population (Table 1), the CVe values of the variables in all populations were within limits reported in the literature (Almeida et al., 2011;Cargnelutti Filho et al., 2009;Nogueira et al., 2012;Resende & Duarte, 2007). Thus, the experimental accuracy can be considered high.
However, the use of CVe to evaluate the precision of an experiment has been questioned, since this estimate does not consider the coefficient of genetic variation or the number of repetitions (Resende, 2007). Therefore, the estimate of accuracy b r b gg is more adequate for assessing accuracy since it considers all of these factors simultaneously. This author conceptualizes accuracy as a parameter that refers to the correlation between the true genotypic value of the genetic material and that estimated or predicted based on the information from the field experiments. In this context, the greater the estimate of accuracy approaches the unit, the more accurate the experiment is. The efficiency of the different visual selection alternatives in identifying the most productive plants was obtained by linear regression Yi = β0 + β1Xi + ei. Visual selection will be able to identify the most productive individuals if β0 = 0 and β1 = 1 statistically and if the coefficient of determination is higher than 70%. This approach was used by Silva et al. (2015) to compare the efficiency of the individual BLUP method (BLUPIS) in the selection within sugarcane families to the individual BLUP method.
Even for random selection, the determination coefficient values were higher than 70% (Table 2), so the comparison between the different selection procedures was restricted to the significance of the values of β0 and β1.
In some situations, visual selection was efficient in identifying more productive plants, such as the selection of plants by the highest number of pods by the most experienced evaluator in the F 3 and F 4 populations, being superior to random selection, which did not meet the efficacy criteria in any population (Table 3). This indicates that visual selection may be effective in some circumstances.
However, for rice crops, Cutrim et al. (1997) reported that progenies derived from the plants selected visually in the F 2 generation and those that were randomly selected did not differ in grain yield in the F 4 generation evaluations. These authors also reported that the lack in agreement of this study may be related to the significant interaction between the genotypes and the environment since the environment in which the F 2 lineages were selected was not the same in which the F 4 generation was evaluated.
Contrary to that proposed by Gardner (1961), Méndez (1971), and Molina (1983), the uniformity of the experimental area alone does not guarantee success in visual selection (Table 3), since even though all populations present high levels of accuracy (Table 1), the proportion of different visual selection alternatives, which showed significant efficiency, was not the same for all populations, being 50%, 30%, and 0% for the populations F 3 , F 4 , and F 5 , respectively, indicating that other factors influence the process. Even with this variation, no influence of the evaluator's level of experience on the selective success was observed, and there is even an example in the F 3 population in which the inexperienced selector 1 was able to elect more productive plants more often than the most experienced selector. In the example in question, the inexperienced evaluator -1 was successful with the selection for pod number, number of lateral branches, and agronomic value. In contrast, most of the evaluators were successful only when selected for pod number and agronomic value.
Evaluating the efficiency of visual selection on the yield of irrigated rice grains, Cutrim et al. (1997) also found that the selectors differed in the ability to identify productive plants visually and that this ability is not associated with crop experience. However, Bowman et al. (2004) found that breeders with more than 30 years of experience were more efficient than selectors with 4 years of experience in the visual selection of cotton progenies.
Among the traits related to the productive capacity of soybean, which allowed a greater number of selectors to visually identify the most productive individuals was the number of pods per plant in the F 3 population, allowing the effectiveness of 80% of the evaluators.
This percentage drops to 20% in the F 4 population and 0% in the F 5 population.
Studying the visual efficiency in F 3 populations of chickpeas, Dahiya et al. (1984) also observed that the number of pods per plant is the most important yield component for this crop and that it was the criterion that provided the selection of the most productive lineages. The conclusion, however, cannot be extrapolated in all cases.
Visual selection for traits associated with production, for example, did not generate correlated changes in the yield of winter wheat grains (Ud-Din et al., 1993). Blanche et al. (2009) observed that the efficiency of visual selection will be greater, the higher the phenotypic standard deviation, indicating that the visual perception capacity of a trait, and therefore, the selective success will depend on the level of phenotypic variation of the same.
Phenotypic values observed experimentally are associated with genotypic values, as demonstrated by high levels of accuracy (Table 1). The expressive drop in selective efficiency may be related to the magnitude of the genetic variance of the number of pods in each population, being 132.56, 23.88, and 20.32 in the F 3 , F 4 , and F 5 populations, respectively (Table 1). Thus, the greater the genotype variance of a population for a given trait, the greater the chance of success with visual selection, a fact that corroborates with Kohel and White (1963) and Thomson (1973).
The present study demonstrates that depending on the magnitude of the genetic variance in a population, as long as there is uniformity in the experimental conditions, the indirect visual selection for soybean yield through the number of pods can be efficient, even at the current level of soybean breeding.
The agronomic value of soybean plants was the second trait used as a selection criterion that provided the greatest number of selectors to identify the most productive individuals visually. This is an additional attribute resulting from the combination of all other productivity-related traits, such as the number of pods, number of seeds per pod, plant height, susceptibility to lodging, disease resistance, and dehiscence resistance, among others.
Following this criterion for visual selection, effectiveness was achieved in the F 3 population of 60% of the evaluators. In the F 4 generation, an efficiency of 40% of the selectors along with the number of seeds per pod was obtained, demonstrating the possibility of using this criterion in visual selection in soybean populations to increase production ( Table 3).
The present result corroborates Lopes et al. (2001), who affirmed that through visual selection, the agronomic value may represent an increase in efficiency in the selection of better genotypes, especially in the initial evaluations of yield, when a large number of genotypes are generally available, which will contribute to reducing the time and resources spent in the initial evaluation phase.

| CONCLUSIONS
The selectors differ in their ability to identify more productive plants Abbreviations: AV, agronomic value; NLB, number of lateral branches; NPP, number of pods per plant; NSP, number of seeds per pod. * β0 equal to 0 and β1 equal to 1 by Student's t test at 5% probability; number of pods per plant (NPP); number of seeds per pod (NSP); number of lateral branches (NLB); agronomic value (AG). most assists visual selection in the search for more productive individuals.

CONFLICT OF INTEREST
The authors declare no conflict of interest. All authors have read and agreed to the published version of the manuscript.

AUTHOR CONTRIBUTION
Conceptualization (

DATA AVAILABILITY STATEMENT
Data sharing or accessibility does not apply because the manuscript reviews previously published research.