Assessing yield stability in African yam bean

Maintaining yield stability in the African yam bean (AYB) (Sphenostylis stenocarpa) under year-to-year variability is crucial to its sustained productivity. Exploring yield stability in crops is vital in identifying how stable and consistent the yield of such crops is. Cultivation of AYB, an underutilized traditional legume in a specific environment, will further popularize the crop and enhance the acceptance as a cheap protein source thereby reducing hunger and malnutrition especially in regions where climate change has negatively affected legume crop production. Field trials were carried out to study the performance of 23 AYB genotypes in four-year environments. Two seeds of each genotype were sown in a single 5m row plot spaced at 1m between and within rows; the trial was conducted during the cropping season of 2011, 2012, 2013, and 2014 and was laid out in a randomized complete block design (RCBD) with three replications. At harvesting, five plants from each row were separately harvested; seeds of all the sampled plants in each plot were bulked and weighed, and the seed yield per plant was then determined. A combined analysis of variance (ANOVA) was performed to test for the significance of genotypes, year, and genotype by year interaction. Before combined ANOVA, a test for homogeneity of residual variances was performed using Bartlet’s test; stability of the genotypes over the years was ascertained numerically and graphically using additive main effects and multiplicative interaction and Genotype X Genotype X Environment interaction (GGE) biplot analyses. Rainfall distribution between 680 and 1,700mm with an average temperature of 28.50°C under sandy-clay soil type supported high and stable seed yield production in AYB. This environment was found adequate during the 2014 (E1) growing season. Genotypes TSs118, TSs12, TSs109, TSs148, TSs5, TSs61, and TSs69 produced an above-average mean yield across the years and were found to be productive and stable in all the year environments. TSs82 and TSs6 with aboveaverage mean seed and tuber yield can be considered for cultivation where seed and tuber dual-purpose production is to be maximized, while TSs111, TSs49, and TSs96 with high tuber yield ranking above average total tuber yield can be further enhanced for tuber production.


Introduction
Achieving yield stability under increased year-to-year yield variation is an important breeding goal for indigenous legumes as their relevance for food security is becoming more prominent in Africa (Thiyagu et al. 2012). Again, the use of stability parameters is important in defining and identifying crop genotypes with consistent yield production (Clarkea et al. 2019). African yam bean (AYB) (Sphenostylis stenocarpa) is an important pulse crop grown predominantly in the South Western parts of Nigeria and some African nations. Aside from producing seeds, some accessions also bear tubers (Apata and Ologhobo 1990; Ikhajiagbe et al. 2007). Recently, the crop has gained a renewed interest by researchers, not because of its high-nutritional value alone, but also the ability of the crop to grow over a wide range of climatic and soil conditions (Aremu et al. 2007; Arogundade et al. 2014; Malumba et al. 2016). Furthermore, AYB contributes to soil fertility improvement through atmospheric nitrogen fixation (Assefa and Kleiner 1997). Despite these advantages in recent years, the total acreage for AYB production is on a constant declining trend, leading to low productivity (Linneman 1995; Aremu and Ibirinde 2012; Arogundade et al. 2016) when compared with other grain legumes. This is ascribed to yield instability, particularly in AYB producing areas, characterized by high interannual yield variability (Sood et al. 2011). This emphasizes the need for developing high yield stable genotypes to survive the challenges of meteorological factors fluctuation. Similarly, this will help developing economies that are characterized by low income and high unemployment rates such as Nigeria to improve their food nutritional qualities ; Asaleye et al. 2018; Popoola et al. 2018). The fact that the performance of AYB is strongly influenced by variability in the environments, at the very least, and those representing yearly weather variation, makes it a barrier for improving yield potential (Yadava et al. 2011). Multi-environment trials are typically used in plant breeding programs to evaluate genetic materials over a range of years in a target environment (Holtsford and Ellstrand 1992; Odeseye et al. 2018; Ibrahima et al. 2019). However, when comparing these genetic materials, their interaction with the environment is often expected to a certain degree.
Genotype by environment interaction (G × E) usually refers to the differential response of genotypes for a given trait across environments and is considered an important source of year-to-year variation in crop yield (Owusua et al. 2018; Seyouma et al. 2019). Therefore, the most trustworthy approach to manage G × E interaction is the adoption of stable cultivars in a cropping system (Adeniyan et al. 2007). For developing high-yield stable cultivars, it becomes imperative to identify stable sources through the analysis of G × E interaction in cultivar evaluation trials conducted over the years. The development of high yielding stable AYB cultivars is itself a strenuous task due to the influence of several factors that affect plant growth, which is documented to be compounded by year-to-year variability in weather (Kang and Pham 1991; Adewale and Dumet 2011). Therefore, understanding changes in yield are important for harnessing genotypic stability. To understand the influences of year-toyear variability in weather on genotype yield stability, a repetitive field testing and trait evaluation are conducted while employing statistical tools that directly test for the presence of G × E and measure individual genotype stability. The existence of G × E and yield stability can be evaluated using various multivariate techniques (Ene-Obong and Okoye 1992; Farias et al. 2017). A number of outstanding reviews of most of the current statistical tools to test and measure genotypic stability have been published (Baye 2002;Adewale et al. 2010;Aremu et al. 2018). Among them, the most widely used is the additive main effects and multiplicative interaction (AMMI) analysis, which is a hybrid model that incorporates both analyses of variance and principal component analysis, has been considered as a powerful statistical tool for interpreting G × E interaction and extensively used in plant breeding research (Cornelius et al. 1992;Das et al. 2010).
The attainment of stable genotypes and identification of testing/breeding environments with respect to targeted traits is the ultimate goal of any breeding program (Adewale and Dumet 2011). On this basis, estimating the magnitude of G × Y interaction on yield stability in AYB has a crucial role to play here because yield stability acts as a buffer against crop failure for resource-poor farmers in Africa. Thus, the exploration of considerable variation among cultivars to identifying stable genotypes for seed yield is highly essential for the sustainability of the smallholder farming system in sub-Saharan Africa. It has previously been reported that the selection of AYB for yield stability is a complex process due to the influence of G × E that affect yield (Holhs 1995). When unpredictable environmental factors, such as year-to-year variation results in significant G × E interaction, crop improvement scientists need to develop stable cultivars that perform reasonably well under a range of environmental conditions.
Consequently, genotype main effect plus genotype × environment interaction (GGE also called site regression analysis, a graphical analysis method of multienvironment data) has been widely exploited in the identification of most adapted stable genotypes in target environments (Gauch and Zobel 1996;Cartera et al. 2018;Ngailo et al. 2019). In the AYB cultivar breeding program, yield (pod or tuber) stability is an important breeding objective, next to yield potential (Cornelius et al. 1992). A multi-environment trial in plant breeding remains a challenge as a result of genotype × environment interaction (GEI). The choice of statistical analysis plays a key role in determining accuracy in the recommendation of both productive and stable test environments. To date, very few studies have been reported on performance trials of AYB, and the studies concluded that the selection of superior genotypes for yield is impacted by G × E (Anley et al. 2013). However, no stability study has been performed using a multiyear yield trial for a diverse range of AYB cultivars, representing a large share of national genetic variation of AYB germplasm collections. The collections for this study are maintained in the gene bank of the International Institute of Tropical Agriculture (IITA). Hence, this study obtains information on seed yield performance of 23 AYB genotypes using G × Y interaction analysis and hence assesses the joint value of AMMI and GGE techniques for the identification of stable and high yielding AYB genotypes for the use of farmers and specific crop breeding research.

Genetic materials and field trials
Four field experiments were conducted between 2011 and 2014 at the Teaching and Research Farm of Ladoke Akintola University of Technology, Ogbomoso (longitudes 4′100E, Latitude 8′100 N, 149 masl), Nigeria. The study site is representative of the Guinea Savannah Zone of Southern Nigeria. The soil type of this site is sandyclayey; there are two main cropping seasons, early cropping season (April-July) and late cropping season (August-November). AYB is mostly planted during the early cropping season. The weather parameters prevailing during the experimental years were taken and are presented in Table 1.
Twenty-three (23) accessions of AYB (presented in Table  2) obtained from diverse eco-geographical origins of Nigeria were sourced from the gene bank of the IITA, Ibadan, in 2010. These cultivars were selected for this study based on their diversity for various agronomic traits.
The experimental sites were cleared, plowed, and harrowed. Two seeds of each genotype were sown in single 5 m row plots spaced at 1 m between and within rows. This gives a population density of 10,000 plants per hectare. The trial was conducted at each year (April 30th, May 1st, May 31st, and June 20th during the cropping season of 2011, 2012, 2013, and 2014, respectively) and was laid out in a randomized complete block design (RCBD) with three replications. Due to the effect of climate change observed during the cropping season, the experiment was not carried out on the same date of each year. However, the early season was put into consideration to ensure little variation in the weather condition. This was done by planting the genotypes during the early cropping season between April and June for the four years considered. Likewise, this will also enhance the understanding of how climate and weather change affects legume crop production. The blocks were separated by a 2-m alley to reduce interblock plant competition. Interplots were separated by 1-m spacing between and within rows. The four trials were conducted during the rainy season. Hand weeding was carried out as at when needed to maintain a clean field. Two seeds were sum per stand to give two-plant stands and a total of ten plants. Only one of each inner plant stand was harvested to give five plants from each row which were separately harvested. Seeds of all the sampled plants in each plot were bulked and weighed, and the seed yield per plant (g) was then determined by dividing by five.

Statistical analyses
A combined analysis of variance (ANOVA) for RCBD across the years (environments) was performed to test for the significance of genotypes, year, and genotype by year interaction using the general linear model (GLM) procedure of the Statistical Analysis System (SAS 9.1) statistical package (SAS, 2004). Before combined ANOVA, a test for homogeneity of residual variances was performed using Bartlet's test (Steel et al. 1996). The stability of the genotypes over the years was ascertained numerically and graphically using AMMI and GGE biplot analyses. The AMMI was employed in this study. The grain yield data were subjected to AMMI analysis which combines ANOVA with additive and multiplicative parameters into a single model (Gauch 2013). In this study, the replicate effect was removed in combining the data. According to Gauch (1988), the replicate effect tends to make the data noisy, in which the estimated parameter may not reflect the actual mean. The main effect was partitioned into genotypic and environmental effects, while the nonadditive effect was due to genotype by environment interaction and is detailed in the AMMI equation used for analysis. This study follows Gauch (1988) and stated the AMMI model as follows: In equation (1), i = (AYB genotypes 1, 2, 3,…,23); j = (year environments) 1, 2,….4; Y ij is the observed mean yield of ith genotype in the jth environment; µ is the grand mean; G i is the ith genotypic effect, E j is the jth environment effect, n is the number of principal component analysis (PCA) axes considered, which is judged based on an empirical consideration of F-test of significance, λk is the eigenvalue of the PCA axis k; αik and γjk are the ith genotype jth environment PCA scores for the PCA axis k, θ ij is the residual; and n is the number of PCA axes retained in the model. The noise sum of the square in the interaction sum of square (SS) is estimated by multiplying df G × E by error mean square (MS).
AMMI stability value (ASV) was calculated using SAS 2.0 package of 2000 following the formula developed by Purchase (1997): In equation (2), SSIPCA1 is the SSs of interaction principal component analysis 1; SSIPCA2 is the SSs of interaction principal component analysis 2; IPCA1 is interaction principal component analysis one, and IPCA2 is interaction principal component analysis two.

Mean performance of genotypes
AYB flowering and pod formation start one-hundred days after germination. In this experiment, the dates to flowering across the four environments were toward the end of early planting seasons (between July and end of August), when humidity is low and atmospheric temperature is also getting warm (27-35°C). Flowering marks the end of the accumulation of photosynthates from source to sink.

AMMI analysis
Consequent upon growing the 23 AYB genotypes in different cropping years, the G × E interaction effects from the combined AMMI analysis were statistically significant (P < 0.001) ( are not suitable to be grown within 27-35°C temperature with low rainfall distribution ( Table 1) because the resultant seed yield is below the average. In addition to winning environment ranking, ASV and ranking analysis identified G1 (TSs118), G19 (TSs61), and G3 (TSs12) as having above-average mean yield and are, therefore, the most stable genotypes across the growing year seasons ( Table 5). Better still, the 1PCA scores of these three AYB genotypes are between 0 (−0.76) and <2 (1.93) indicating small interaction effect with the environment, whereas G17 (TSs58) (11.87) followed by G20 (TSs69) (−9.51) and G5 (TSs111) (11.98) with first principal component analysis (IPCA) scores >2 are adjudged to be less stable and more sensitive to environmental changes.

GGE biplot analysis
Considering the GGE biplot analysis, 77.8 percent of the total variation was jointly explained by PC 1 and PC 2, accounting for 45% and 32.8% of the total variation, respectively. The interaction pattern based on genotype and environment is visualized in Figure 1. The genotype groups that were positioned farthest from the biplot origin were joined to form the polygon, with genotypes at the vertex contributing most to the interaction, and presenting the highest yielding genotypes (22,20,19) as also captured in Table 4 in the mean yield across the environments. Furthermore, the GGE biplot distributed the genotypes based on stability and seed yield ( Figure  2). The average environment coordinates (AECs) (representing an average performance of the genotypes) and its ordinates (representing the stability of performance) indicated that genotypes 5, 18, 12, 7, and 15 on the left side of the AEC had a low yield in all the environments. Contrarily, genotypes (22, 20, and 19) on the right side had high yield across the environments. Genotype1 (TSs118), G3 (TSs12), G6 (TSs148), and 19 (TSs61) are most stable as they are located almost on the AEC ordinate.
Similarly, the GGE biplot revealed the associations among environments and identified potential megaenvironments ( Figure 3). The GGE biplot for seed yield in the 23 AYB genotypes showed that the four years/ environments were distributed into two mega-environments,

Discussion of findings
The combination of years using AMMI and GGE biplots identified the yield performance of the 23 genotypes as affected by the prevailing environmental factors (rainfall, temperature, humidity, etc.) in each of the four years of this study. The analysis revealed significant variations among the genotypes and years, which in turn explained a considerable degree of variation which can be exploited when selecting genotypes for specific environmental conditions. The differential response of AYB genotypes for seed yield to the differences in weather parameters across the four-year environments indicated the presence of G × E, which justified further analysis of the stability of performance with yield and environment using AMMI and GGE.
From the AMMI analysis model, the fact that the yield pattern of each genotype accounted for a higher percentage of the G × E SS indicates that there is considerable selectivity pattern in the model using IPCA axis scores arising from individual genotype yield performance. These findings agree with the earlier report of Gauch (2006), which revealed that significant IPCA1 and successive axes in AMMI model captured interaction exclusively in a diatonic   GM = grand mean; YSI = yield selection index; ASV = AMMI stability value.  The E1 had near-zero IPCA scores of 0.631 indicated a small interaction effect but a below-average mean yield performance. Small interaction is an indication that the environment has little influence on the genotypes. Following the weather data in 2011 (Table 1), the nearzero 1PCA interpretation some of the genotypes recorded an above-average mean seed yield. According to Gauch et al. (2011) and Kuang et al. (2014), genotypes with above-average mean yield across growing environments can be adjudged as stable genotypes. This explains that the prevailing rainfall pattern, temperature, and humidity of the growing season did not significantly affect the performance of some of the AYB but affected 10 of the 23 genotypes, namely as follows: G1 (TSs118), G2 (TSs10), G3 (TSs12), G6 (TSs148), G9 (TSs6), G13 (TSs116), G14 (TSs48), G19 (TSs61), G20 (TSs69), and G22 (TSs82). These genotypes recorded an average mean yield above the grand mean (418.8 kg/ha) across the four years cropping environments and are therefore adjudged productive. This result criticizes the use of only mean performance across environments ( Table 4) to adjudge genotype performance. No wonder the earlier research efforts of Yan (2002) and the later work on maize genotypes across nine environments (Mahalingam et al. 2006;Yan et al. 2007) stated that if a genotype or an environment has opposite directional signs with the PCA score, their interaction is negative no matter the quantum of the yield value.
On the genotype and environment AMMI stability ranking, the use of the GGE biplot and PCA scores further identified the winning genotype in the year environment.
Mean yield ranking and scores using IPCA1 and IPCA2 ( Table 5) ranked G20 (TSs69), G22 (TSs82), and G19 (TSs61) as best seed yielding genotypes. That G20 and G22 recorded high yield of 533.0 kg/ha and 524.0 kg/ha, respectively, notwithstanding, the opposite signs on their IPCA scores (−9.51; −6.79) indicated negative interaction with the growing environments and therefore not winning in any of the four environments (Figure 1). Genotypes 13 (TSs116) and G19 (TSs61) demonstrated positive interaction when grown under specific rainfall of 608-1,500 and the environmental temperature not exceeding 340°C. These two specific genotypes are seen to have won in the 2013 environment and hence adjudged adaptable and stable for the environment. Kuang et al. (2014) found 3 genotypes of maize out of 9 in 20 environments to be productive and stable using AMMI analysis. This result further confirms the findings of Gauch and Zobel (1997) and also supported by the efforts of Purchase 1997 who emphasized that ASV gives a balance measurement between IPCA scores, such that the lower the IPCA scores, the more specifically adapted the genotype to the study year/environment. Again, the more recent work of Clarkea et al. (2019), identified stability performance in the yield of commercial sorghum hybrids to be affected by stress (water and temperature) in the environment. The ranking patterns of ASV and yield stability (YSI) are almost the same ( Table 5). This implies that genotype selection can follow either of both rankings patterns.
The scaling in Figure 2 centers on genotype mean yield and stability across the four-year environments. This agrees with Yan (2002) scaling, where singular values are partitioned into environment scores for comparing genotype relationships among year environments. Following Zobel et al. (1988), the main effect principal additive effects of genotypes and environment adequately explain the genotype means and environment means in relation to individual genotype mean performance across the fouryear environments ( Table 4). The differential seed yield performance may be attributed to the inconsistent lower rainfall distribution in the 2011 and 2013 growing years, which consequently determined the temperature and humidity flow ( Table 1). No wonder, Silveira et al. (2013) reported that phenotypic stability is a function of both climatic and soil conditions. The GGE biplot analysis identified some genotypes as performing similarly in a given test year environment, the discriminating power of the genotypes relative to the ideal environment was revealed. Furthermore, the analysis suggested that the years/environments used for the evaluation of the AYB could be separated into two mega-environments, and where mega-environments (large) are used, test environment representatives are identified. The similarity of the genotypes in their performance observed in the growing years of 2011 and 2013 suggested a positive correlation between them, which was also obvious from the acute angle formed between the vectors of the two years.
Even though the 2012 season is somewhat closer to the 2013 cosine angle, genotype performances within these two environments are not the same. Whereas 2014 and 2012 with the longest vectors in comparison with the remaining two years/environments suggested that these years/environments were the most effective for genotypic differentiation based on seed yield in AYB. From these observations, it is inferred that field trials with a few test years might give the same level of information, which was otherwise obtained by evaluating under the highly correlated years, thereby reducing the cost associated with field evaluation, without precision in identifying the best performance for seed yield , ASV and YSI comparatively support IPCA scores values in selecting stable genotypes and have been successfully used in the selection of wheat and corn for specific location adaptation. Following large IPCA scores, G5 (TSs-111; 11.98), G17 (TSs-58; 11.87), and G11 (TSs-95; 6.78) were the more responsive and contributed largely to the interaction component and may be considered for specific adaptation. Those with smaller IPCA scores G1 (TSs118), G3 (TSs12), G6 (TSs148), G19 (TSs61), and G20 (TSs69) are identified as the most stable genotypes across environments.

Conclusion
The study assessed the performance of 23 AYB genotypes for seed yields for four years (2011, 2012, 2013, and 2014) cropping seasons using Additive Main Effect and Multiplicative Interaction (AMMI) and GGE biplot to determine the main effects due to genotype (G), environment (E) and GEI in varying year environments in one location. The use of AMMI in genotype yield analysis using single value-based data remains unique. This method separates genotype, environment, and the interaction components to showcase specific genotype productivity in a given environment. Genotype stability studies are important in classifying genotype adaptation and consistency in yield performance. Such studies will reduce efforts and save costs in engaging unpredictable field locations for farming and crop breeding research. The use of GGE and AMMI tools in this research identified G1 (TSs118), G3 (TSs12), G6 (TSs148), G19 (TSs61), and G20 (TSs69) as the most productive and adaptable to any range of the three weather characteristics within the cropping seasons of AYB. The year effect becomes relevant in multi-environment breeding, where weather reports are cumulatively harnessed. It is therefore important to collate accurate weather forecast decisions before engaging these seven genotypes for yield production, location notwithstanding. Better still, G4 (TSs-109), G9 (TSs-6), and G20 (TSs-69) have the dual-purpose production capacity to give an aboveaverage mean seed and tuber yield. The weather factors in the growing seasons of 2011 and 2012 are almost the same and supported below-average mean yield performance (251.8 kg/ha; 299.7 kg/ha) as evident in the GGE-environment AMMI biplot ( Figure 4) and therefore can be a representative of the year environment. The year 2014, with the highest rainfall, relative humidity, and lowest average temperature, is most productive and stable (635.8 kg/ha). A prevailing environment with high rainfall, with consequently high relative humidity and a prevailing low temperature, can be determined and utilized for growing the earlier identified stable genotypes (G1 (TSs-118), G3 (TSs-12), G4 (TSs-109), G6 (TSs-148), G9 (TSs-6), G19 (TSs-61), G20 (TSs-69)). In achieving a meaningful breeding program, precision in genotype yield stability performance selection using AMMI ranking and GGE biplots either ASV or YSI can be interchangeably engaged during genotype stability selection.