Yield performance stability of adapted and improved cowpea in the Equatoria region of South Sudan

South Sudan has diverse agro-ecologies with high potential of agricultural production for different crops (Galla and Ngalamu, 2013), including cowpea [Vigna unguiculata (L.) Walp]. The interest among farmers for cowpea production is increasing in the Equatoria region of South Sudan for  food security and household income. However, the lack of  quality seed and low grain yield of local varieties are the major production constraints among farmers. This has prompted the introduction of improved cowpea genotypes from different sources to select stable high yielding ones to boost production. Testing genotypes across locations are important for the identification of stable and high yielding genotypes and the target environment (Yan et al., 2000). However, genotype x environment interaction (GEI) plays a significant role in determining a stable high yielding genotype. A combined analysis of variance (ANOVA) can explain GEI and reveal the main effects, but it cannot explain the effects resulting from the interaction (Asnake et al., 2013). Hence, determination of varietal performance across test environments and ideal stable genotype(s) is crucial. Thus, the information drawn from such studies will facilitate the selection and release of the best performing genotypes to farmers.
        
Following the proposal of Gabriel (1971), the genotype and genotype by environment (GGE) biplot technique was used to display the GGE of multi-environment trials (MET) data which is referred to as GGE biplot (Yan, 2002). The GGE biplot analysis is the most powerful statistical tool to determine the pattern of genotypic responsiveness across environments (Kaya et al., 2006). It displays both genotypic (G) and genotype × environment (GE) effects. However, the yield is the combined effects of G, E and GEI; only G and GEI are pertinent and concurrently significant in genotype evaluation (Yan et al., 2007). When selecting genotypes across contrasting environments for yield stability, plant breeders look for a non-crossover type of GEI or preferably the absence of GEI for general adaptation (Thomason and Phillips, 2006). Nevertheless, GEI is important for exploitation by GGE biplot to target suitable genotypes to a specific environment or group of environments.
        
GGE biplot uses visual analysis of MET data (Sozen et al., 2018) and clearly identifies the best variety for specific environments and enables identification of MET. The MET can occur in more than one agro-ecological zone, defined by similar abiotic and biotic stresses (Cooper and DeLacy, 1994). Based on GGE biplot analysis of yield performance, genotypes can be assessed for both broad and specific adaption (Sozen et al., 2018) which is crucial for assigning a particular genotype to a specific environment. The application of GGE biplot has helped breeders to select superior and high yielding genotypes since the package integrates both yield performance and stability (Kang, 1993, Kang and Magari, 1996; Fan et al., 2007). The objective of this study was to evaluate the yield stability performance of nine cowpea genotypes across six environments in South Sudan.

The test locations were Palotaka (PAL) and Pajok (PAJ), in Magwi County, and Juba in Juba (JUB) County. PAL and PAJ are in the Greenbelt while JUB falls in Hills and Mountains agro-ecologies. The characteristics of the test locations are presented in Table 1. The genetic materials consisted of seven improved cowpea genotypes (IT90K-277-2, IT07K-274-2-9, IT07K-211-1-8, IT08K-193-15, IT07K-297-3, ACC004, and IT08K-1 80-11) introduced from the International Institute of Tropical Agriculture (IITA) and two farmers’ preferred varieties (Mading Bor II and Titinwa). Based on yield performance, the best nine genotypes were selected from 147 cowpea genotypes evaluated in a preliminary yield trial. Mading Bor II (MBR II) is an adapted variety collected from Bor in the Flood Plains agro-ecology, while Titinwa, a landrace commonly grown by farmers, collected from Mundri in the Ironstone Plateau agro-ecology of South Sudan. Titinwa was used as a local check, because of its easy cooking ability.
 

        
The nine genotypes were evaluated in a randomized complete block design with three replications in each test location. The first evaluation was conducted during the first season in JUB and PAL (2014A), and PAJ in the second season (2014B), whereas in 2015, the genotypes were evaluated at the three sites in the second season (2015B). Each genotype was planted in a 4 m long, four-row plot with 0.50 m spacing between rows and 0.40 m between hills within a row. Two seeds were planted in a hill and thinned to one plant after 21 days to attain a population density of 33,333 plants ha^-1 in each location.
        
Data were recorded in each plot using the cowpea descriptors developed by the International Board for Plant Genetic Resources (IBPGR) now Bioversity International. Grain yield adjusted to 12% moisture content was computed from the grain weight of harvested pods per plot. Although we measured many traits in this study, only data on grain yield and yield components are presented.
        
The ANOVA was performed using GenStat 18^th Edition, considering all effects random except genotypes following the procedures of Vargas et al., (2013). Each season and location were considered as an environment. Analysis of variance was conducted for each environment separately and combined across the six environments to determine whether GEI was significant. Repeatability estimates were computed for grain yield and other six traits to determine the precision of the trial (Allard, 1999). Phenotypic correlations were computed for these traits. To compute yield performance stability using GGE biplot (Yan, 2002), test environments were coded with the corresponding environmental characteristics and season as shown in Table 1. Mean grain yields of the nine cowpea genotypes from six environments were used to construct the GGE biplots. The analysis was based on singular value decomposition (SVD) of the first two principal components (PC1 and PC2) to identify genotypes adapted to specific or broad environments. The mean yield data used to generate the GGE biplot were environment centered without transformation which was then decomposed into principal components (PCs) via SVD Yan (2002). 

Analysis of variances
 
Combined ANOVA showed that environment had highly significant effects on grain yield, number of seeds per pod, number of branches per plant and number of days to 50% flowering, whereas year had significant interactions with genotypes for grain yield, number of pods per plant, number of branches per plant, and days to 50% flowering (Table 2). Estimates of repeatability varied with traits, with a number of seeds per pod scoring the highest (0.95) and grain yield (0.21) had the lowest score. Environment accounted for 60 to 75% of total variability in grain yield and yield components. Our results corroborate with the findings of other workers (Putto et al., 2008; Singh et al., 2018) who reported that environment accounts for 50 - 80% of the total variation in MET data.
 

        
The mean grain yield of all genotypes was 735.8 kg ha^-1. The best genotype (IT90K-277-2) produced grain yields of 1,626.4 kg ha^-1 and the worst one (IT07K-274-2-9) produced 343.7 kg ha^-1. The grain yield of the local check, Titinwa, was 628.3 kg ha^-1. Interestingly, the adapted variety MBR II was among the top three genotypes that - surpassed the local check Titinwa by 44 to 64.3%. Two of the best three genotypes (IT90K-277-2, IT07K-211-1-8) are improved lines introduced from IITA. MBR II had a competitive performance with the two best genotypes in PAJ during the second season in 2014. These genotypes had consistent performance during the second season with limited rains, suggesting that they can also adapt to marginal areas (Cooper et al., 1997; Meseka et al., 2016). 
 
Phenotypic correlations
 
Phenotypic correlations between grain yield and six yield components showed that plant height, number of branches per plant, number of pods per plant, number of seeds per pod, days to 50% flowering and 100-seed weight were positively and significantly correlated with grain yield (Table 3). The strongest correlations were detected between grain yield and number of pods per plant and number of branches per pant (r = 0.92; P<0.01) and between grain yield and plant height (r = 0.89; P<0.01). This result indicated inherent contributions of these traits to grain yield showing a positive linear relationship between grain yield and the yield components, suggesting that improving one of these traits could positively influence grain yield. These traits can be used as an index for selecting high yielding cowpea genotypes. Our results corroborate the findings of Souza et al., (2007) and Menna et al., (2015) who reported significant positive correlations of similar traits in cowpea. However, Aybegun et al., (2018) reported a negative and strong association between grain yield and number of days to 50% flowering in field pea.
 

 
GGE Biplot analysis
 
In this study, GGE biplot explained 88.4% of the total yield variation in the test environments. The vertex for “which-won-where-what” view identified IT07K-211-1-8 as the best genotype at PAJ, while IT90K-277-2 was the best genotype at PAL and JUB. MBR II had yield potential above average across the six environments. The check variety, Titinwa and most of the genotypes were not adapted to a specific location (Fig1 and Table 2). The poor performance of Titinwa and most of the genotypes across test environments can be attributed to GEI effects and their genetic backgrounds. However, different genotypes produced the highest grain yield in different test environments. The differential ranking of the genotypes across test environments demonstrates the existence of a mixture of crossover and non-crossover types of GEI as a common case with MET data (Kaya et al., 2006). 
 

        
Yield performance and stability of the nine cowpea genotypes determined by an average environment coordinate (AEC) method (Yan and Hunt, 2001; Yan, 2002) was defined by contributions of PC1 and PC2 in GGE biplot. A line passing through the AEC and biplot origin (average environment axis) serves as abscissa of the AEC, while the ordinate of AEC is the line that passes through the origin and perpendicular to the AEC abscissa (Fig 2). The AEC ordinate separates genotypes with above average grain yields from those with below average yields. Genotypic stability is crucial to sustainable performance across environments, the longer the projection to AEC ordinate, the greater tendency to more variability of genotypic performance across test environments. Consequently, IT90K-277-2 was identified as high yielding and most stable, while IT07K-211-1-8 was high yielding but unstable. MBRII was stable but had a lower yield compared to IT90K-277-2 and IT07K-211-1-8. Titinwa and five other genotypes had poor yield performance but were stable.
 

 
        
In the polygon view of the discriminating environments, JUB and PAL were strongly correlated and most discriminatory (Fig 3), suggesting that removing one of the two locations as a test environment would not lead to any loss of information and will cut down on resources that can be used in other locations (Yan et al., 2000; Yan and Rajcan, 2002; Yan and Tinker 2005; Kaya et al., 2006; Asnake et al., 2013). Yan and Tinker (2005) suggested that test environments that are non-discriminating provide no information on the genotypes and therefore should not be used as test environments. The first rainy season in PAJ characterized by high relative humidity and rainfall provided a perfect environment for insect pests and diseases, making it unsuitable for growing cowpea. Our results suggest that farmers in PAJ should grow improved cowpea varieties during the second season and concentrate on growing other crops like groundnuts (Arachis hypogaea), maize (Zea mays L.) and sorghum [Sorghum bicolor (L.) Moench] that  require adequate rainfall which occurs during the first season. Although PAJ and PAL are grouped under the Greenbelt, PAL falls in mid-altitude (>1000 masl), while PAJ (893 masl) and JUB (496 masl) fall in the lowland that requires different germplasm classifications.
 

The present study revealed that environment interacted significantly with genotypes accounting for more than 60% of the total variability in grain yield and six yield components. Two improved genotypes introduced from IITA had high grain yields compared to farmers’ preferred varieties. Interestingly, MBRII had competitive yield potential and other desirable traits compared to the two best-improved genotypes. IT90K-277-2 was the most stable and high yielding genotype that can also be grown in other lowland areas with similar environmental conditions. The study revealed that farmers in high rainfall lowland areas should avoid growing improved cowpea during the first season due to high insect and disease pressure.  Farmers in lowland areas in Sub-Saharan Africa can tap into the stable performance of the best three genotypes for production and marketing to improve food security and household incomes.

This study is made possible by the project grant (2013 PASS 002) awarded by the Program for African Seed Systems (PASS) of the Alliance for a Green Revolution in Africa (AGRA).