Genotype X Environment Interaction and Yield Stability in Early- Maturing Cowpea (Vigna unguiculata (L.) Walp.) Landraces in Ethiopia

'e study was conducted to estimate the effects of genotype, environment, and genotype× environment interaction on grain yield and yield-related traits and to identify stability genotype. At six environments, twenty-four cowpea landraces and one check were evaluated in a 5× 5 triple lattice during the 2019 cropping season. Data were collected on yield and yield-related traits.'e analysis of variance for each environment and across environments showed significant differences among genotypes, environments, and GEI for most traits including yield. Environment, genotype, and GEI showed 27.45%, 20.9%, and 49.55% contribution to the total sum of squares, respectively, for grain yield. 'is indicated that the environments were diverse and most of the variation in grain yield was caused due to interaction and environmental means. G24 (2632 kg ha) and G16 (2290 kg ha) were the highest yielder and stable genotypes with mean grain yields above the grand mean (2049.28 kg ha) and standard check (2273 kg ha). G24 and G16 were the most stable genotypes according to cultivar superiority, Wricke’s ecovalence, regression coefficient, and devotion from regression stability models.


Introduction
Cowpea is an annual herbaceous and self-pollinated grain legume crop that belongs to family Fabaceae and genus Vigna. It is a diploid species with 2n � 2x � 22 chromosomes [1]. e precise primary center of origin and the exact center of diversity of cowpea is not known, but according to OECD [2], Ethiopia is the primary center of diversity of cowpea and more than 66.5% of the arable land is very suitable for cowpea production.
Cowpea is drought resistant and adapted to marginal soil due to the nitrogen-fixing ability that makes it a useful staple crop for farmers in harsh environments under moisture stress and hot temperatures [3,4]. It is widely cultivated and consumed globally, especially in the arid and semiarid tropics and sub-Saharan Africa [5]. As documented by Abate et al. [6], about 38 million households (194 million people) grow cowpea in sub-Saharan Africa that accounts for about 84% of the world production. Particularly, production in Nigeria and Niger accounts for more than 45% and nearly 15% of the world's total production, respectively, with more than 4 million hectors of cultivated land each. Nigeria is supposed to be the highest producer of cowpea in the world with the production of 850,000 tons followed by Niger and Brazil, whereas Niger is the largest cowpea exporter in the world with an estimate of 215, 000 tons exported annually [1,7]. e crop is cultivated primarily for its edible seeds, young leaves (as a vegetable), pods seeds, soil improvement, and forage value [8,9]. Due to its fast growth habit, it covers the ground rapidly and, therefore, limits weed competition and soil erosion [10,11].
In Ethiopia, cowpea is mostly produced in the Amhara region (North and South Wollo, Waghimra, and Oromia special zones), Tigray region, Oromia region (East Shewa and East and West Harerghe zones), Gambella (Anywaa Zone), and SNNPR: South Omo, Gamo Gofa, and Wolayta zones [8,9]. As documented by the aforementioned authors, Ethiopia has a great diversity of cowpea landraces (local farmers' varieties) in many traits and over 83% of cultivated cowpeas in Ethiopia are landraces. Cowpea production and utilization in Ethiopia is very low as compared to other African countries, though the country is claimed to be the center of diversity and/ or origin and has a high potential for the production of the crop.
Despite the fact that cowpea is a multipurpose grain legume, its productivity is very low and unstable across environments and years due to biotic factors (weeds, insects, and diseases), abiotic factors (soil type, altitude, rainfall distribution, and intensity), and scarcity of widely adaptable and early-maturing varieties [12,13].
e variability in environments such as location effect, seasonal fluctuation, and their interaction highly influences the production of cowpea in relation to yield potential [14,15].
Genotype × environment interaction (GEI) is a challenge for plant breeders since it complicates variety recommendation because of the inconsistency of best-yielding genotypes across cropping environments and season. e effect of GEI is most commonly minimized by repeating the experiment at several sites in a single year, or over several crop seasons in a single site, or both [16,17].
Even though there are many opportunities for breeders to develop cowpea varieties possessing different agronomic characteristics and inherent tolerance to a wide range of biotic and abiotic factors, progress of cowpea breeding in Ethiopia is very limited either in exploiting the available genetic variability in the country or from introduction of improved varieties. Moreover, information on the effect of genotype, environment, and their interaction and the performance stability of cowpea landraces is scanty. erefore, evaluating cowpea landraces in contrasting environments to generate such information is imperative. In view of this, this research was initiated with the following objectives: (i) To estimate the effects of genotype, environment, and genotype × environment interaction on grain yield and yield-related traits (ii) To assess the stability of cowpea landraces for yield across environments

Description of the Study Area.
e experiment was conducted at six cowpea-growing environments during the 2019 main cropping season in Ethiopia. e environments were as follows: Sekota, Kobo, Sirinka, Melkassa, Miesso, and Babile. e test environments were selected on the basis of representativeness for cowpea-producing environments. ese six locations representing different agroecologies of cowpea-growing areas in Ethiopia and descriptions of the areas are presented in Table 1.

Experimental Materials and Design.
Twenty-four Ethiopian cowpea landraces (Table 2) collected from the Amhara region, Tigray region, Oromia region, Gambella, and SNNPR were evaluated along with one check (released variety: Kanketi). e experiment was conducted using a 5 × 5 triple lattice design at each location, and each genotype was assigned randomly. e seeds were planted on the first week of July at a plot size of 4 m × 2.4 m (9.6 m 2 ) having four rows, with an interrow spacing of 60 cm and 20 cm between seeds within the row. e net harvest area was 4.8 m 2 per plot, the central two rows. e spacing between plots and blocks was 1 m and 1.5 m, respectively. All agronomic management practices were carried out equally and properly as per the recommendations.

Data Collection and Data
Analysis. Data for plant height, number of pods per plant, and number of seeds per pod were collected on the basis of five sample plants which were randomly taken from the two central rows, and the average of five samples was used for analysis, while days to 50% flowering, days to 90% maturity, grain yield, and 100 seed weight were collected on the basis of the net plot (i.e., two central rows).
Different statistical software packages were used to analyse the data. SAS [18] was used for analysis of variance of the individual environments and combined over locations [19]. SPAR 2.0 was used for regression coefficient (bi) and deviation from regression (S 2 di), and GenStat (18 th edition, 2015) was used for cultivar performance (p i ) and Wricke's ecovalence (Wi). e combined analysis of variance across the environment was carried out using a mixed model (genotype: fixed, location: random) in order to determine the differences between genotypes across environments, among environments and their interaction. e F-max method of Hartley [20] was used to assess the homogeneity of error variances prior to doing combined analysis over the environment. However, location four (Melkassa) was an outlier in terms of error variance; hence, it was removed from the combined analysis. Due to this, the combined analysis of variance for grain yield and yield-related traits of 25 cowpea genotypes tested was conducted for five environments. Various stability analyses were computed for grain yield. Mean comparison using Duncan's Multiple Range Test (DMRT) was performed to explain the significant differences among the means of genotypes and environments.

Analysis of Variance for Individual Environments.
Analysis of variance for grain yield (kg ha −1 ) of the 25 cowpea genotypes tested in six environments is presented in Table 2. Analysis of variance for each environment showed a highly significant (P < 0.01) variation in grain yield among cowpea genotypes in all tested environments.
is indicated the presence of performance variation among the tested cowpea genotypes for grain yield. Likewise, El-Shaieny et al. [21]; Santos et al. [22]; Ishiyaku et al. [23]; and Tariku [24] in cowpea and Abele [25] and Tullu [26] in common bean reported large variation in grain yield performance among different tested genotypes across environments.
Among the testing environments, the general performances of the genotypes for grain yield were the highest at Miesso as compared to the other six environments with a mean grain yield of 2528.8 kg ha −1 followed by Melkassa (2487.1 kg ha −1 ) and Kobo (1986.6 kg ha −1 ). e superior performance of genotypes at Miesso and Melkassa can be attributed to the uniform distribution of rainfall during the season (Table 3). e lowest grain yield was obtained at Sirinka with a mean yield of 1676.1 kg ha −1 . is may be due to heavy rainfall and hail that occurred at the vegetative stage. Babile was the 2 nd least-performing environment with mean grain yield of 1729.7 kg ha −1 , which may be due to a small amount and erratic rainfall during the growing season.

Combined Analysis of Variance over Five Environments.
For combined analysis of variance, the homogeneity of error variances was tested using the F-max method of Hartley [20], which is based on the ratio of the larger mean square of error (MSE) from the separate analysis of variance to the smaller mean square of error as F − ratio � larger MSE/ smaller MSE . If the larger error mean square is not threefold larger than the smaller error mean square, the error variance was considered homogeneous [27] and allowed pooled analysis across environments. However, location four (Melkassa) was an outlier in terms of error variance; hence, it was removed from the combined analysis. Due to this, the combined analysis of variance for grain yield and yield-related traits of 25 cowpea genotypes tested was conducted for five environments (Tables 4 and 5). e analysis showed that cowpea grain yield was significantly (P < 0.01) affected by the environment, genotype, and GEI. e significance of GEI indicated that the relative performances of the genotypes were not consistent across the test environments and the environments had different effects on the yield potential of the genotypes. is, in turn, suggested the need to conduct further analysis on the genotype by environment interaction to understand the nature of the interaction and to accomplish stability analysis for identifying genotypes with stable performances across environments and/or those that fit for specific adaptation. In agreement with this finding, El-Shaieny et al. [28]; Santos et al. [22]; and Ishiyaku et al. [23] in cowpea and Firew [29]; Abele [25]; and Tadele et al. [30] in common bean reported the significant effects of the genotype, environment, and genotype by environment interaction on yield and yieldrelated traits.
In this study, the largest variance was explained by the genotype by environment interaction (49.55%), followed by the environment effect (27.45%), genotype effect (20.9%), and 2.07% of replication effect (Table 4). Only 9.21% of total variation remained unexplained (as error). A large yield variation explained by the GEI and environment indicated that genotypes had various performances across environments and the environments were diverse, with large difference among environmental means causing most of the variation in grain yield. In line with this, many researchers reported the dominance of the environmental effect in the combined analysis of grain yield (Abele [25]; Tariku [24,26]). Hence, the GEI maintained or exerted about more than double times larger effect than the genotype, which complicates the selection of superior and adaptable genotypes. Pham and Kang [31] indicated that, since GEI minimize the usefulness of genotype effects alone, it is, thus, imperative that yield levels, adaptation, and stability are taken into account in the multilocation trial. erefore, simultaneous consideration of both high mean yield (genotype main effect) and GEI stability is very crucial in selecting superior genotypes from evaluated cowpea genotypes.
Combined analysis of variance revealed a highly significant (P < 0.01) variation among the tested genotypes for days to 50% flowering, days to 90% maturity, plant height, number of pods per plant, number of seeds per pod, grain yield, and hundred seed weight (Table 6). In this study, the genotype × environment interaction showed highly significant differences and the tested genotypes showed rank change and strong interaction. erefore, the mean performance of genotypes cross the tested environments was interpreted using simple effects (i.e., mean of each genotype at each location) to provide relevant information.

Analysis of Variance for Grain Yield from Eberhart and
Russell's Model. e significant mean squares for G × E interaction in the general ANOVA (Table 4) suggested the importance of partitioning the genotype by the environment interaction effect into the environment (linear), G × E (linear) interaction effects (the sum of squares due to regression or b i ), and unexplained deviation from linear regression (pooled deviation mean squares or S 2 di). e results of ANOVA for grain yield from Eberhart and Russell's model are presented in Table 7.
In the joint regression analysis of variance for grain yield, there was a highly significant difference for genotypes and a significant difference for genotype x environment (linear) sources of variation. is indicated contrasts between the environments and the occurrence of the differential response of genotypes across environments. e linear and nonlinear parameters provide an adequate account of the dynamic response of genotypes to changing environments and are used with the mean performance of genotypes to assess the potentialities of different genotypes [32]. e significant G × E (linear) interaction indicated that the stability parameter "b i " estimated by the linear response to change in environment was not the same for all genotypes, and the variability of genotypes for performance is partially unpredictable.
ese results are in agreement with those reported by Sarvamangala et al. [33] and Nunes et al. [34] in cowpea.
According to Eberhart and Russell's [21] model, a stable genotype should have high mean yield, the regression coefficient approximating to a unity (b i � 1), and deviation from regression (S 2 di) nearly equal to zero. e regression coefficient (b i ) measures the response of genotypes to environments. When the regression coefficient of the genotype is nonsignificantly different from unity (b i � 1), the genotype is said to have average response to environmental change and suitable to both poor and good environments, while if the b i value of genotypes is significantly greater than unity (b i > 1), the genotype is highly responsive to environmental change and it is suitable only for a good environment. When the genotype b i value is significantly less than unity (b i < 1), it indicates that the genotype is less responsive and it is suitable for poor environments [35]. Accordingly, many genotypes with mean grain yield lower and above the grand mean had b i values which were nonsignificantly different from unity (b i � 1), indicating their average response to both poor and good environments (Table 8). From those genotypes, G16, G20, G24, G2, and G7 had positive and nearly a unity (bi � 1) regression coefficient values that indicated average responsiveness and that the genotypes were suitable for all environments. ese genotypes had also higher mean yield than overall mean of genotypes, suggesting that they are desirable for cultivation in all environments. In contrast to this, genotypes G5, G6, G10, G15, and G23 had the b i values significantly greater than unity, indicating that these genotypes are highly responsive to environmental change and are suitable only for good environments. Similarly, Rodriguez [36] and Adewale et al. [37] reported differences in cowpea genotypes for their responsiveness and stability for grain yield tested at different environments. e mean sums of squares due to pooled deviation from regression were highly significant (P ≤ 0.01) for grain yield of all genotypes except G21, indicating the importance of a nonlinear genotype by environment interaction, and the genotypes were unstable across environments (Table 8).    Likewise, Nigussie [38] have reported that twelve out of fifteen common bean genotypes evaluated at six environments had S 2 di values significantly different from zero. From the tested genotypes, 11 genotypes had lower grain yield than the grand mean of genotypes, which showed that these genotypes were not desirable for cultivation though they were stable. e most stable genotypes with the lowest S 2 di values were G21, G14, G10, G24, G20, and G4, which ranked 1 st , 2 nd , 3 rd , 4 th , 5 th , and 6 th for deviation from regression, respectively, from the lowest to highest. However, genotype G21 showed a nonsignificant deviation from regression and negative S 2 di value with lower grain yield, indicating its suitability for unfavourable environments. e most unstable genotypes with the highest S 2 di values were G19, G23, G18, and G3, which ranked 22 nd , 23 rd , 24 th, , and 25 th correspondingly among the tested genotypes. erefore, these genotypes are best fit for specific adaptation in favourable environments where there are high input levels. If the mean yield, regression coefficient value (b i ), and the deviation from the regression (S 2 di) are considered together, then the most stable genotype would be G24 with a mean grain yield of 2632 kg ha −1 (ranked 1 st in yield), 9 th in regression coefficient (b i � 0.91 close to 1), and 4 th in deviation from the regression. Genotype G2 with a mean grain yield of 2276.23 kg ha −1 and G20 (2213.16 kg ha −1 ) would be the 2 nd and 3 rd most stable genotypes among the tested. Two genotypes (G5 and G23) among thirteen high yielder genotypes had a significant deviation of regression coefficient (b i ) values from unity, with a significant deviation of S 2 di value from zero (S 2 di > 0), indicating that they were unstable and desirable for cultivation in favourable environments. In agreement to this, Tariku [24] in cowpea and Firew [29], Abele [25], and Tullu [26] in common bean had identified the stable genotypes which are desirable for cultivation in all