Interação genótipo × ambiente e dimensionamento amostral para estatísticas de precisão em ensaios de soja

Abstract

Soybean grain yield is a relevant characteristic that needs further understanding in highland and lowland scenarios, where there is a lack of comparative theoretical references, mainly regarding genotype × environment interaction. Another little approached aspect about these scenarios is the reliability of estimates of this variable, which initiates by the sampling process, normally performed empirically, causing an elevated bias when samplings are not representative. Therefore, the aims of this study were to verify the effects of genotype × environment interaction on soybean grain yield in highlands and lowlands of subtropical climate and to compare the adaptability and stability methodologies; to analyze the behavior of experimental precision statistics in front of the variations in the number of collected plants per experimental unit in highlands and lowlands; to define the optimal sample size per experimental unit for experimental precision statistics; and to propose predictive models for estimating the precision of experiments with soybean. Field trials were carried out during the 2017/2018 agricultural harvest in two locations of Rio Grande do Sul, on three sowing dates, totaling six experiments. In the first study, 2.70 m2 per plot were harvest and the grain yield of 20 genotypes was measured in both testing locations. With the collected data the significance of the interaction factor was verified and this factor was partitioned into simple and complex components. Next, linear bi-linear models were implemented, Additive Main Effect and Multiplicative Interaction (AMMI), Best Linear Unbiased Prediction (BLUP) and Genotype plus Genotype-Environment interaction (GGE), for verifying the stability of cultivars, with posterior comparison of methodologies through uncertainty statistics and Pearson’s correlation coefficient. In the second study, grain yield was assessed per plant, in 20 plants per plot, using 30 genotypes in the highland location and 20 genotypes in the lowland location, totaling 9,000 measured plants. Thirteen precision statistics were estimated and sample size per experimental unit was determined per statistic, simulating scenarios of 1, 2, ..., 1000 plants; consequently, predictive models for each statistic were parameterized, based on the number of collected plants. The results demonstrated greater grain yields in the highlands, where the second sowing date expressed the highest values. The complex component of interaction represented 82.11 %, which allowed inferring cases of genotype ranking alteration. Agreement between the GGE and BLUP methodologies was observed. The statistics were overestimated in smaller sample scenarios per experimental unit. With the increase of collected plants, exponentially proportional reductions of the confidence interval width of the calculated statistics were verified. This allowed proposing experimental precision prediction models, via confidence interval width and sample size per experimental unit. The sampling of 18 plants per experimental unit was enough for estimating experimental precision statistics. With the performed studies, a greater understanding of the highland and lowland edaphic scenarios on factors that aid cultivars recommendation and experimental planning became possible.