Advances in Crop Science and Technology
Open Access

AMMI; GGE biplot; Regression; Stability

Introduction

Soybean (Glycine max (L.) Merrill) is an important legume as good sources of inexpensive protein (40 %) and vegetable oil (26 %) worldwide (Pratap et al., 2012). It can be used directly for food in the household, or processed for soy-milk, cooking oil and a range of other products, including infant weaning food. The poultry industry also uses soybean for feed production. Soybean grain often has a good market demand. The crop residues are also rich in protein and are good feed for livestock or form a good basis for compost manure. The largest global oilseed crop production goes to soybean (53%), followed by rapeseed (15%), cottonseed (10%) and peanut (9%) (Pratap et al., 2012) [1].

s It is used as food, nutritious animal feed and improves soil fertility through nitrogen fixation when used in crop rotation with cereal crops (Pratap et al., 2012). In the last five years soybean production in Ethiopia showed an increment, from 90,000 tons in 2015 to 126,000 tons in 2019 (FAO, 2019). The productivity of soybean in Ethiopia is 2.3 ton ha-1 and higher as compared to African average productivity (1.3ton ha-1), but below the world average (2.8 ton ha-1) in 2019 (FAO, 2019) [2].

The performance of a genotype is dependent on the genetic potential of the variety, the environment where the variety is grown, and the interaction between the genotype and the environment (Yan, 2001; Yan and Hunt, 2001). Breeders evaluate different genotypes across locations in order to develop high yielding, adaptable and stable cultivars over the testing environments or specific locations. A number of analytical tools and models have been used to assess the stability and adaptability of genotypes across environments. The regression model proposed by Eberhart and Russell (1966) allows for the computation of a complete analysis of variance with individual stability estimates and departure from linearity of a regression line [3 ]. The model considers a stable variety as the one with a high mean yield, bi=1 and s2 di=0. Similarly, genotypes with a high s2 di deviate significantly from linearity and have a less predictable response for the given environments. Additive Main effects and Multiplicative Interaction (AMMI) model involves correlation or regression analysis that also relates the genotypic and environmental score derived from a principal component analysis of the genotype by environment interaction matrix to genotypic and environmental covariates. Genotype by Environment interaction studies were conducted for soybean by different researchers in different countries. Stability of a given genotype can also be determined by its response for diverse environments where soybean variety is grown. Research focusing on stability or genotype by environment interactions is necessary for plant breeders to develop genotypes that respond optimally and consistently across environments. Therefore, this experiment was initiated to determine the nature and magnitude of genotype by environment interaction and identify superior and stable soybean genotypes for the diverse environments [4].

Materials and Methods

Germplasm and study sites

Thirteen medium set soybean genotypes including the standard check (Billo) were tested at Bako, Uke and Billo for two consecutive main cropping seasons (2020-2021) (Table 1).

Table 1: Environments used in the study and their main characteristics in Ethiopia.

Experimental design and management

Thirteen medium set soybean genotypes were evaluated in a Randomized Complete Block Design with three replications. A plot consisted of four rows with the spacing of 0.6 m between rows and 0.1 m between plants. Fertilizer rate of 100 kg ha-1 NPS was applied at planting. Management practices were done for all experimental units across location and years according to the recommendations made for the crop and/or location. Two middle rows in each replication were harvested [5]. The grain was adjusted to 10% seed moisture content before weighing to record yield and converted to hectare basis before data analysis (Table 2).

Table 2: Lists of experimental materials and their source used the experiments.

Data analysis

The grain yield data collected at each site were subjected to analysis of variance (ANOVA) followed by combined analysis of variance for all the six sites using SAS statistical software.

Additive main effects and multiplicative interaction (AMMI)

The responses of the genotypes were evaluated with regression (Eberhart and Russel, 1966) and Additive Main-effect and Multiplicative Interaction (AMMI) models GenStat 16 edition software. The linear model proposed by Eberhart and Russell (1966) is: Yij = μi +biIj +S2dij

Where Yij is the mean performance of ith variety (I=1, 2,…, n) environment; μi is the mean of ith variety over all the environments; bi is the regression coefficient which measures the response of ith variety to varying environments; S2dij is the deviation from regression of ith variety in the jth environment, Ij is the environmental index of jth environment [6].

AMMI model (Zobel and Gauch, 1996):

Where Yger is the observed yield of genotype g in environment e for replication r; Additive parameters: μ the grand mean; g α the deviation of genotype g from the grand mean and e β the deviation of environment e; the multiplicative parameters: n

λ the singular value

for interaction principal component axis (IPCA) n, gn γ the genotype eigenvector for axis n, and en δ the environment eigenvector; ge ρ PCA residuals (noise portion) and ger ε error term.

Results and Discussion

Combined analysis of variance

The combined analysis of variance for yield is presented in Table 3. The result revealed that the main effects, genotype (G), location (L) and Year (Y), and the interaction effect G × L, G × Y and G × L ×Y showed a highly significant (P ≤ 0.001) difference for grain yield [7 ].

Significant differences were observed for grain yield among genotypes in all environments (Table 3). This indicated the presence of genetic variability among the genotypes. Environment for grain yield (averaged across genotypes) ranged from 1.09 ton ha-1 at Billo in 2020 to 2.66 ton ha-1 at Uke in 2020. Mean grain yield across environments ranged from 1.41 ton ha-1 (JM-PR142/CLR-15-5C-2) to 2.49 ton ha-1 (TGX-1987-62F) with grand mean of 1.89 ton ha-1. Five genotypes (TGX-1987-62F), (PB-12-2), (PM-12-53), (TGX 1989- 45F) and JMALM/ H3-15-5C-1 gave yield above grand mean (1.89 ton ha-1 ) and the remaining eight genotypes including old and newly released check Korme and Billo gave below the average yield. The mean grain yield combined over location and years showed that genotype TGX-1987- 62F was the top ranking in performance [8].

Table 3: Combined analysis of variance for13 Medium Set soybean varieties evaluated in Western Oromia.

AMMI model analysis

An output of the AMMI model analysis of variance for grain yield is presented in Table 4. This analysis also revealed the presence of highly significant (P< 0.01) differences among medium set soybean genotypes for grain yield. From the total treatment sum of squares, the largest (72.2%) portion was due to environments main effect followed by genotypes main effect (18.6%) and genotype by environment interaction (11.34%). A large yield variation explained by environments indicated the existence of both spatial and temporal diversity in testenvironments, with large differences among environmental means causing most of the variation in grain yield [9]. In line with this result, Tolessa and Gela (2014) reported large yield variation of common bean genotypes due to environments. This also indicates the existence of a considerable amount of deferential response among the evaluated soybean genotypes to changes in growing environments and the differential discriminating ability of the test environments. Substantial percentage (74.36%) of G × E interaction was explained by IPCA1 followed by IPCA-2 (25.66%) and, therefore, used to plot a two-dimensional GGE biplot. Amare and Tamado (2014) and Temesgen et al. (2014) suggested that the most accurate model for AMMI could be predicted by using the first two IPCA [10].

Table 4: Mean Seed Yield (ton ha-1) of Soybean Genotypes evaluated in western Oromia across Locations and Years.

AMMI biplot analysis

AMMI biplot graph (Figure 1) with X-axis plotting IPCA1 and Y-axis plotting IPCA2 scores illustrate stability, adaptability and high yielding of soybean genotypes to the testing environments. It has been reported that the IPCA1 scores of a genotypes in AMMI analysis are an indication of the stability or adaptation over environments (Alberts, 2004) (Table 5).

Figure 1: AMMI biplot showing “which won where” and stable soybean genotypes evaluated at six environments in western Oromia.

Table 5: Partitioning of the explained sum of square (SS) and mean square (MS) from AMMI analysis for grain yield of seven soybean genotypes.

The greater the IPCA scores, negative or positive, the more specific adapted is a genotype to certain environments. According to AMMI biplot, Environments Bako and Uke relatively showed high IPCA scores and contributed largely to GEI. Bako and Uke environments were conducive for best performing soybean genotypes. Genotypes JM DAVS/ALM-15-5A, PI 567061and PM-12-56 was intended to low yielding environment (Figure 1). Based on the IPCA score, PI-12-55 and PB-12-3 were not stable genotypes and as well performed under low yielding environments. TGX-1987-62F and PM-12-53genotypes revealed more static performance across environments in comparison to other medium set soybean genotypes in the trial. PB-12-3 performed to low yielding environments and also was relatively stable (Figure 1). PM-12-53, PB-12-2 and TGX-1987-62F genotypes have relatively lower IPCA by virtue of which they proved to give best grain yield and stability than other genotypes (Figure 1). TGX-1987-62F genotype had the highest grain yield followed by PB-12-2 and PM-12-53 genotypes. Similar results were also reported by Temesgen et al. (2014) on linseed and Niger seed and Adane et al., (2020) on soybean (Table 6).

Table 6: Average Yield AMMI-estimates per environment.

GGE biplot analysis

In GGE biplot (Figure 2), IPCA1 and IPCA2 explained 76.31 and 14.74 %, respectively of soybean genotypes by environment interaction and made a total of 91.05 %. In a study conducted on groundnut by Amare and Tamado (2014) and white lupines by Atnaf et al. (2017), IPCA1 and IPCA2 explained an interaction of 81.8 and 63.4%, respectively, extracted from IPCA1 and IPCA2. An ideal genotype is defined as a genotype which has the greatest IPCA1 score (mean performance) and with zero GEI, as represented by an arrow pointing to it (Figure 2). A genotype is more desirable if it is located closer to the ideal genotypes. Thus, using the ideal genotype as in the center, concentric circles were drawn to help visualize the distance between each genotype and the ideal genotype. Therefore, the ranking based on the genotype-focused scaling assumes that stability and mean yield are equally important [11 ].

Figure 2: GGE-biplot based on genotype-focused scaling for comparison the genotypes with the ideal genotype.

In this study, TGX-1987-62F and PB-12-2 genotypes which fell closest to the ideal genotype were identified as the most desirable genotypes as compared to the rest of the tested medium set soybean genotypes in the trials (Figure 2). Similarly, Dabessa et al. (2016) identified ideal genotypes based on the genotype-focused scaling that assumes stability and high mean yield of studied genotypes. Ideal test environment is an environment which has more power to discriminate genotypes in terms of the genotypic main effect as well as being able to represent the overall environment. But such a type of environment may not exist in real conditions. Therefore, by assuming a small circle which is located in the center of concentric circles and an arrow pointing on it as the ideal environment (Figure 2), it is possible to identify desirable environments which are found closer to the ideal environment (Yan and Rajcan, 2002). Hence, among the testing environments, Bako,located near to this ideal environment was identified as the best desirable testing environment in terms of being the most representative of the overall environments and powerful to discriminate medium set soybean genotypes in the trial [12].

Conclusion and Recommendations

Combined analysis of variance indicated that grain yield performance of the tested medium set soybean genotypes is highly influenced by environment, genotypes, and GEI. This indicates that a particular genotype does not exhibit uniform performance under different environmental conditions or different genotypes may respond differently to a specific environment. The varieties and environment main effects and genotype-by-environment interaction effects are highly significant for medium set soybean genotypes in the trial. The environment contributed most to the variability in grain yield. Genotype TGX-1987-62F was close to the ideal genotype and could thus be used as bench mark for the evaluation of medium set soybean genotypes in western Oromia. Considering mean grain yield and stability simultaneously, PB-12-2 was the best medium set soybean genotype in the trial and is recommended for further evaluation under variety verification trial.

Acknowledgement

This study was funded by Oromia Agricultural Research Institute, Bako Agricultural Research Center. Pulse and Oil Crops Research Team members of Bako Agricultural Research Center are acknowledged for trial management and data collection.

1. Alberts MJA, (2004) Comparison of statistical methods to describe genotype environment interaction and yield stability in multi-location maize trials. M.Sc. Thesis, University of the Free State.

Indexed at, Google Scholar, Crossref

2. Amare K, Tamado T (2014) Genotype by environment interaction and stability of pod yield of elite breeding lines of groundnut (Arachis hypogaea L.) in Eastern Ethiopia. Star Journal 3: 43-46.

Indexed at, Google Scholar, Crossref

3. Atnaf M, Wegary D, Dagne K, Tesfaye K (2017) Genotype by environment interaction and grain yield stability of Ethiopian white lupin (Lupinus Albus L.)Landraces. Experimental Agriculture 1-14.

Indexed at, Google Scholar, Crossref

4. Dabessa A, Alemu B, Abebe Z, Lul D (2016) Genotype by environment interaction and kernel yield stability of groundnut (Arachis hypogaea L.) varieties in Western Oromia, Ethiopia. Journal of Agriculture and Crops 2: 113-120.

Indexed at, Google Scholar, Crossref

5. FAO (2019) FAOSTAT Food and Agriculture Organization of the United Nations Eberhart SA and Russell WA, 1966. Stability parameters for comparing varieties. Crop Sci 6: 36-40.

Indexed at, Google Scholar, Crossref

6. HG Gauch Jr, Zobel RW (1996) Optimal Replication in Selection Experiments. Crop Science 36.

Google Scholar

7. Pratap A, Gupta SK, Kumar J, Solanki RK (2012) Soybean. In: Gupta SK (ed) Technological innovations in major world oil crops. Vol. I, breeding. Springer Science New York 293-321.

Google Scholar

8. Temesgen A, Mammo K, Lule D (2014) Genotype by environment interaction (G x E) and grain yield stability analysis of Ethiopian linseed and Niger seed varieties. Journal of Applied Biosciences 80:7093-7101.

Indexed at, Google Scholar, Crossref

9. Tolessa TT, Gela TS (2014) Sites regression GGE biplot analysis of haricot bean (Phaseolus vulgaris L.) genotypes in three contrasting environments. World Journal of Agricultural Research 2:228-236.

Indexed at, Google Scholar, Crossref

10. Yan W (2001) GGE Biplot - a windows application for graphical analysis of multi-environment trial data and other types of two-way data. Agron J 93: 1111-8.

Indexed at, Google Scholar, Crossref

11. Yan W, Hunt LA (2001) Interpretation of Genotype × Environment Interaction for Winter wheat Yield in Ontario. Crop Science 41: 19-25

Google Scholar

12. Yan W, Rajcan I (2002) Biplot evaluation of test sites and trait relations of soybean in Ontario. Crop Sci 42: 11-20.

Indexed at, Google Scholar, Crossref