The analyses of variance demonstrated that the effects of the genotypes and of the
GE interaction were significant (p ≤ 0,01) for all 43 trials (supplementary material 3), which suggests the existence of different mega-environments within the study region, such a fact that allows the use of the GGE Biplot methodology.
The use of the first two principal components did not explain at least 70% of the total variation of 21 of the trials evaluated. Thus, for these trials, the environments were partitioned in two groups according to the average environment axis (AEA), as recommended by (Zdziarski et al. 2019), and, later on, a new GGE Biplot analysis was carried out for each one of the groups.
At the end of the GGE biplot analyses, 59 graphs of stratification were obtained (supplementary material 2), being 24 for the 2011/12 crop season; 15 for the 2012/13 crop season; six for the 2013/14 one and 14 for the 2015/16 crop season. By means of these graphs, matrices of coincidence were obtained and represented through networks of similarity.
To carry out the global analysis, nine coinciding municipalities among the four crop seasons were used. Based on the analyses (Figure 4), the standard of grouping of the municipalities throughout the four years revealed the existence of four mega-environments. Municipalities that belong to the same mega-environment promote similar effects on the tested genotypes, reducing the effect of the GE interaction. In this sense, they can be considered environmentally redundant to represent the identified mega-environment.
From the information generated by the network of similarity, it was possible to observe the independence of the towns of Bela Vista do Paraíso and Rolândia in relation to the other towns. Neither are not part of a mega-environment, which demonstrates that they have different environmental features from the others. An important feature to be mentioned is that the trials performed in the town of Bela Vista do Paraíso were irrigated, while the others are carried out in dry conditions. Thus, for the region being studied, these are considered essential environments for the implementation of soybean trials. This makes the representativeness of the study region more efficient, ensuring greater reliability in the selection and recommendation of soybean cultivars. It was also possible to notice that the towns of Maracaju, Palotina and Mamborê are present in more than one mega-environment, which demonstrates that they can represent more than one mega-environment.
The GGE Biplot methodology has been used to direct the planning of breeding programs in several regions and situations (Marcolin & Vezzetti 2017). In the present work, it was possible to identify municipalities capable of optimizing the implementation of a network of multi-environment trials for macro-region two of soybeans in Brazil, as established by (Kaster 2012).
It is important to reiterate that, for the region to be well represented, the testing sites must represent the environmental heterogeneity of the region; thus, environmentally redundant municipalities that do not contribute to the representativeness of the entire region must be replaced or ruled out. It is also important to point out that the towns which are environmentally distinct from the others must be maintained; this way there will be a better representativeness in the region through the complementarity of these sites.
Keeping this in mind, the optimization of the network of trials for the region being studied can be employed. It was found that the four mega-environments identified have intercessions; in these intercessions, towns are allocated and they can represent the respective mega-environments. Thus, it is recommended that these trials be performed in four of the nine coinciding environments evaluated, which are: Palotina, Maracaju, Bela Vista do Paraíso and Rolândia. The towns of Bela Vista do Paraíso and Rolândia were included because they are considered essential environments in the representativeness of the soybean region being studied, as they are different from the others. The towns of Palotina and Maracaju were included in the environment to carry out the trials because both belong to more than one mega-environment, and so they can represent them. Palotina represents mega-environments ME1 and ME2 and Maracaju represents mega-environments ME3 and ME4.
In the recommendation of the municipalities to represent the region being studied, by means of an analysis of network of similarity considering all the crop seasons evaluated and the coinciding towns between them, it was possible to achieve a 55,6 % decrease in the number of municipalities included in the evaluations. The resources that before were directed to all these towns that provide redundant information of genotypes can be reallocated to the inclusion of other trials in municipalities that have not been adopted in the trial network yet, and/or can be directed to the inclusion of more genotypes to be evaluated in the representative towns of this network of trials.
For the analyses to be conclusive, the GGE Biplot method requires at least three years of trials so that the formation of mega-environments is analyzed ( Yan & Frégeau-Reid 2018). Thus, in addition to the joint analysis of the four crop seasons, analyses of crop seasons combined in trios were also carried out. This way, it is possible to view some standards and/or groupings that did not occur when the matrix of global environmental coincidence was used. The analyses were performed with the coinciding towns among the trio of the evaluated crop seasons.
In the first analysis considering three crop seasons simultaneously (2011/12, 2012/13 and 2013/14), the formation of four mega-environments formed by seven towns was observed (Figure 5a). When comparing this analysis to the global analysis of the crop seasons, it is observed that Mamborê, Palotina and Cafelândia remained together in the same mega-environment. However, Palotina remains together with Cafelândia and also with Toledo, forming other two mega-environments, which is consistent with the choice of Palotina to represent these towns in the global analysis.
The towns of Maracaju and Palotina are still part of more than one mega-environment, which reinforces their similarity to the towns belonging to all the mega-environments that these are part of. The towns of Rolândia and Bela Vista do Paraíso, like in the global analysis, were not allocated in a mega-environment, which allows us to confirm that they environmentally differ from the others, and were very important in the formation of the testing sites.
DOR = Dourados; NV = Naviraí; MCJ = Maracaju; BV = Bela Vista do Paraíso; RL = Rolândia; CF = Cafelândia; PLT = Palotina; TL = Toledo; MCH = Marechal Cândido Rondon; FRT = Floresta e MAB = Mamborê.
For the trio of crop seasons 2011/12, 2012/13 and 2015/16, eight mega-environments were identified (Figure 5b). For this analysis, the towns of Maracaju, Palotina, Cafelândia and Rolândia were present in more than one mega-environment. No town was considered discrepant with the others; all of them were allocated in a mega-environment. Thus, the towns of Maracajú, Cafelândia and Palotina could represent the other sites. Such an analysis emphasizes that the towns of Palotina and Maracaju are environmentally similar to more than one town and can represent them in these trials for this region.
By analyzing crop seasons 2011/12, 2013/14 and 2015/16 together, 11 coinciding towns were evaluated, from which nine mega-environments were formed (Figure 5c). For this analysis, three towns that had not been studied yet were added, which are: Floresta and Marechal Cândido Rondon in the state of Paraná, and Naviraí in the state of Mato Grosso do Sul.
No municipalities were considered independent from the others, all of them belong to a mega-environment. It is interesting to notice that, although they are geographically closer and belong to the same state, the towns of Naviraí, Maracaju and Dourados are not allocated in the same mega-environment, which demonstrates that a shorter linear distance between two sites does not necessarily imply environmental homogeneity between them.
The towns of Maracaju, Cafelândia, Rolândia and Mamborê are once again allocated in more than one mega-environment. The towns of Marechal Cândido Rondon and Floresta, being studied for the first time in this analysis, are also in more than one mega-environment; besides that, they are allocated in the same mega-environment, which allows us to conclude that they are environmentally redundant and that only one of these sites will be selected for the implementation of the experiments. For this case, Floresta is the town selected, since Marechal Candido Rondon can also be represented by Maracaju and does not represent the town of Dourados.
In the evaluation, where the crop seasons of 2012/13, 2013/14 and 2015/16 were analyzed, nine coinciding towns were involved, of which seven formed five mega-environments (Figure 5d). As in the global analysis, the towns of Bela Vista do Paraíso and Rolândia proved to be discrepant from the others, not belonging to a mega-environment. Once again, the towns of Maracaju and Palotina are included in more than one mega-environment, and can represent other towns.
These networks of similarity traced with three crop seasons together confirmed some standards that could be observed in the global analysis, and also allowed for the study of two towns that were not present in such an analysis. Thus, the analyses in trio can generate important information, besides being conclusive.
Because not all the towns that participated in the trials for this region could be studied in the global analysis and in the analyses in trio, individual analyses for each crop season were also carried out; hence, all towns could be studied. Although the individual analysis is not conclusive, it can help in decision-making and it can also serve to characterize a specific crop season. Apart from that, these individual analyses can help the breeder to view grouping standards of existing environments and if they are repeated throughout the crop seasons.
In the 2011/12 crop season, the VCU trials were evaluated in 15 municipalities, which were grouped in eight mega-environments (Figure 6a).
It can be noticed that the towns of Palotina, Maracaju, Mamborê, Rolândia, Marechal Candido Rondon and Floresta presented ambiguity, belonging to more than one mega-environment. The towns of Bela Vista, Iporá and Toledo, in the state of Paraná, do not belong to a group, which suggests that they have environmental features that are specific and different from the others. Thus, these towns may have been important for the complementarity of the environments tested in the experimental network.
In this analysis, it is also possible to observe that Dourados, Naviraí and Maracaju, which are in the same state and are physically closer to each other, were not allocated in the same mega-environment. The town of Abelardo, in the state of Santa Catarina and the towns of Palotina, Campo Mourão and Mamborê belong to the same group. The town of Florínea, in the state of São Paulo and the towns of Marechal Candido Rondon and Floresta, in the state of Paraná, are also allocated in the same mega-environment. This reinforces the fact that the closest towns and/or which are in the same state will not necessarily be environmentally similar.
For the 2012/13 crop season, 17 towns were evaluated, of which 13 formed 9 mega-environments (Figure 6b). The town of Ponta Porã (PP), in Mato Grosso do Sul, presented a significant coincidence with the towns of Bela Vista do Paraíso (BV) and Santa Terezinha de Itaipu (STP) and Toledo (TL), demonstrating that the environmental influence of this town on the evaluated genotypes is similar to the other cited towns. The same happens to the towns of Cafelândia, Toledo, Marechal Candido Rondon and Naviraí, which also present a significant coincidence with other towns belonging to other mega-environments.
The other towns, Sidrolândia (SI), Ubiratã (UB), Mamborê (MAB) and Francisco Alves (FA), in the state of Paraná, did not combine in one mega-environment, which indicates a possible divergence of environmental influence on the genotypes between them and the other towns belonging to the mega-environments formed for this crop season.
In the 2013/14 crop season, VCU trials were implanted in 18 towns, of which seven were grouped into three mega-environments (Figure 6c). It is noticed that, like in the previous crop season (2012/13), the towns of Dourados, Maracaju and Naviraí remain allocated in the same mega-environment. When evaluating the grouping obtained for the 2015/16 crop season, it was verified that of the 16 towns evaluated, eight formed six mega-environments (Figure 6d).
By means of the analyses carried out, it was possible to identify some stratification standards throughout the crop seasons. The towns of Naviraí (NV) and Maracaju (MCJ) remained in the same mega-environment in all the evaluated crop seasons. As for the towns of Bela Vista do Paraíso (BV), Toledo (TL) and Rolândia (RL), they appear far from the others, not remaining in any group in the crop seasons of 2011/12, 2013/14 and 2015/16.