The experiment was conducted in two extreme ecological sites. The first site was located in Limpopo Province (Loskop research station [25.1773°S, 29.3936°E with average temperatures between 16°C and 31°C]) during 2019–2020 cropping season. The second site was established in North West province (Molelwane research farm [25^o48′00″S, 25^o 38′21″E with the average temperatures ranging from 22°C to 34°C]) in 2020–2021 cropping season. The soil type at both locations was classified as the Hutton series according to the South African soil classification system (Kasirivu, Mterechera & Dire 2011) and is reddish with a sandy loam texture. The total rainfall during the growing seasons at Molelwane trial site was 590 mm in 2019–2020 and 561 mm in 2020–2021, while at Loskop it was 429 mm in 2019–2020 and 442 mm during the 2020–2021 growing season.

Twenty Bambara groundnut landraces were obtained from the Agricultural Research Council (ARC) gene bank, Pretoria, South Africa. A randomised complete block design (RCBD) with three replications was used. Each plot was 2.7 cm × 60 cm with inter- and intra-row spacing of 60 cm × 30 cm. Each plot had 2 rows, with each row consisting of 10 plants. Sowing was done on a flat bed, with one seed sown at each station. Weeds were controlled by hoeing. No fertiliser was applied at both sites (Gerrano et al. 2015).

Table 1 shows the quantitative data that were collected (on three randomly selected plants averaged per plot) during the pre-harvest and post-harvest growth stages of the crop. Qualitative morphological data were collected using Bambara groundnut descriptors (IPGRI/IITA/BAMNET 2000) on the traits presented in Table 2.

Statistical Analysis System (SAS 2021) version 9.4 was used to analyse the data. Data subjected to analysis of variance (ANOVA) to explore differences among test landraces. Mean separation was carried out using Fisher’s protected least significant difference (LSD) at the 5% significance level of probability. Correlations between yield and yield-related contributing traits and principal component analysis (PCA) were also performed to reveal variability among characters. Cluster and biplot analyses were performed to study similarities and dissimilarities between the tested landraces based on the traits recorded. For qualitative morphological data, the phenotypic frequency distributions of the phenotypic characters were computed. Shannon Weaver diversity index was calculated from the phenotypic frequencies to analyse and interpret the phenotypic diversity for each trait (Zavinon et al. 2019): H'=1−∑i=1n(loge)Pi[Eqn 1], where Pi is the proportion of accessions in the ith class of an n-class descriptor and n is a phenotypic class for a descriptor.

The phenotypic, genotypic, and environmental variances, and coefficient of variation were calculated according to the formula described by Khan et al. (2020):

Environmental variance: δ2e=MSE[Eqn 2]

Genotypic variance: δ2p=MSG−MSErs[Eqn 3]

Phenotypic variance: δ2p=σ2g+σ2e[Eqn 4], where mean square of genotypes (MSG) is the mean square due to genotype, MSE is the mean square of error (environmental variance), and r is the number of replications and s the number of sites.

Phenotypic coefficient of variance: PCV=(δ2g)GM*100[Eqn 5]

Genotypic coefficient variance: GCV=δ2g/GM×100[Eqn 6], where:

δ²p = phenotypic variance

δ²g = genotypic variance

GM = grand mean of the character studied

Estimation of heritability in a broad sense: broad sense heritability (h²), expressed as the percentage of the ratio of genotypic variance (δ²g) to the phenotypic variance (δ²p) was calculated using the following formula (Owusu et al. 2021): h2=δ2gσ2p×100[Eqn 7]

Genetic advance (GA) was estimated as per the formula given by Khan et al. (2020): GA=δ2g×δ2pδ2p[Eqn 8], where:

GA-expected genetic advance

δ²p = phenotypic variance

δ²g = genotypic variance

K = the constant selection differential at 5% selection intensity (k = 2.063)

This article followed all ethical standards for research without direct contact with human or animal subjects.

The ANOVA showed that there was significant variation among the tested Bambara groundnut landraces for traits recorded (Table 3 and Table 4). Results on vegetative traits revealed that Bamb3 had the highest number of days to 50% flowering at 84.2, whereas Bamb6 had the shortest number of days to 50% flowering at 75 (Table 5). Bamb4 had the highest number of nodes per stems at 25, whereas Bamb16 had the least at 14.8. Bamb13 had the highest number of branches at 16.3, whereas Bamb16 had the lowest at 9.7. Bamb20 had the longest days to reach maturity at 175.8, whereas Bamb11 had the shortest at 165.8. Bamb2 had the highest chlorophyll content of 53.6 g/mL, whereas Bamb20 had the least chlorophyll content at 28.8 g/mL. Bamb20 also recorded the longest leaflets at 15.6 mm. Bamb2 had the shortest leaflet at 12.3 mm, value of Bamb6 had the tallest plant with a value of 24.6 mm, whereas Bamb11 had the shortest plant at 14.3 mm. Variation was also observed in yield and yield-related traits (Table 4), where Bamb4 recorded the tallest pod length at 21.7 mm whereas Bamb12 recorded 18.3 mm. Bamb14 had the widest pod width at 12.6 mm, while Bamb17 had a pod width of 10.2 mm. Bamb6 had the longest seed length at 14.3 mm, whereas Bamb4 had the shortest at 10.3 mm. Bamb7 had the biggest seed width at 11.8 mm, while Bamb11 had the smallest at 10.5 mm. Bamb6 had the heaviest 100 seed weight per plot at 59.3 g, whereas Bamb19 had the lightest at 39.9 g. Bamb8 had the highest seed yield per plant at 71.6 g and the highest number of pods per plant at 141.3, whereas Bamb3 had the least yield per plant and Bamb20 had the lowest number of pods per plant at 44.

Pearson’s correlation coefficients showed interrelationships among the traits recorded (Table 5). It was observed that a number of branches per stem had a significant positive correlation with plant height (r = 0.50) and seed length, and seed width (r = 0.50). Moreover, a highly significant positive correlation was observed between chlorophyll content and the number of pods per plant (r = 0.74), terminal leaflet length and leaf area (r = 0.71) and terminal leaflet width and leaf area (r = 0.89). There was a significant negative correlation between the date of maturity and number of pods per plant (r = -0.50), pod width, and 100 seed weight (r = -0.66). Furthermore, there was a positive correlation between days to 50% flowering and pod length (r = 0.40). The number of nodes per stem had a positive correlation with leaf area (r = 0.44), terminal leaflet width (r = 0.39) and stand count (r = 0.40). The number of branches per stem also showed a significantly positive correlation with the number of pods per plant (r = 0.41). Positive correlations were observed for seed length with chlorophyll content (r = 0.45), seed width with chlorophyll (r = 0.41), and number of pods per plant (r = 0.43), 100 seeds weight with yield per plant (r = 0.35), terminal leaflet width with plant height (r = 0.40), and finally on leaf area and plant height (r = 0.40). A set of negative correlation was also observed on days to 50% flowering with 100 seed weight (r = -0.3365) and yield per plant (r = -0.38), date of maturity with seed width (r = -0.39), 100 seed weight (r = -0.40), and yield per plant (r = -0.40), pod length with seed length (r = -0.36), pod width with plant height (r = -0.39), seed length with terminal leaflet width (r = -0.35), and finally on chlorophyll with terminal leaflet length (r = -0.35).

The analysis performed on the 17 traits across the two locations revealed that the first six principal components (PCs) with Eigen values > 1.00 accounted for 78.34% (Table 6) of the variability. Component loadings higher than +0.30 were regarded as significant. The first PC accounted for 20.74% variability and most of the traits that grouped under this component were vegetative traits (Table 6), including the number of nodes per stem, number of branches per stem, plant height, terminal leaflet width, and leaf area. The second PC accounted for about 19.58% of the total variability and variables found under this component were yield and vegetative traits, namely, seed length, seed width, number of pods per plant, and chlorophyll content. The third PC accounted for 14.07%, which comprised the following yield components: days to 50% flowering, pod width, and pod length. The fourth PC accounted for 9.2% of the total variation, comprising vegetative components such as the number of nodes per stem, date of maturity, plant height, and stand count. The fifth PC accounted for 8.48% of the total variability comprising both vegetative and yield components, namely, leaf area, seed mass per plant, and terminal leaflet length. The sixth PC accounted for about 6.27% of the total variation, comprising a yield component, namely, the mass of seeds per plant. Finally, the analysis also showed negative loadings above 0.3 which were regarded as variables with negative influence on the PCs, namely, 100 seed weight (third principal); days to 50% flowering, pod length, (fourth principal); 100 seed weight (fifth principal); and days to 50% flowering, seed length, plant height (sixth principal).

The variation observed among the Bambara groundnut landraces used in this study is depicted in the PCA biplot (Figure 1). The landraces were divided into four quadrants, and accounted for 84.65% of the variation for PC1 and PC2, representing 73.88% and 10.77%, respectively. It was observed that landraces Bamb5, Bamb16, Bamb19, and Bamb20 were in the first quadrant, where a strong positive association was observed between terminal leaflet width and date of maturity. Furthermore, Bamb3, Bamb4, Bamb7, Bamb17, and Bamb18 were found in the second quadrant, where a strong positive correlation was observed between plant height and pod width, days to 50% flowering, number of nodes per stem and stand count, and pod length. Plant height and pod width had a strong influence on PC2, whereas a number of pods per plant and chlorophyll had a strong effect on PC1. Bamb1, Bamb2, Bamb10, and Bamb17 were positioned at the centre of the quadrant and were stable accessions for the tested traits across seasons and environments, whereas Bamb3, Bamb4, Bamb7 and Bamb8 were distinct accessions that were located far from the origin. These accessions have peculiar genes and/or alleles that separated them from the test accessions. Hence, Bamb3 and Bamb4 were positively associated with the NNs, Bamb7 and Bamb8 were positively associated with NPP and YPP, respectively. Terminal leaflet length and 100 seed weight showed a strong influence on PC2. Finally, the genotypes Bamb6, Bamb9, Bamb11, Bamb12, Bamb13, Bamb14 and Bamb15 were found on the last quadrant where the correlation was observed between terminal leaflet length and leaf area. Bamb3 and Bamb7 had a strong but negative influence on PC2 and the characters that contributed to that influence were mass of seeds per plant, days to 50% flowering and number of nodes per stem. Bamb8 and Bamb20 showed a strong but negative influence on both PC1, and the characters that contributed to that influence were number of pods per plant and date of maturity.

Hierarchical cluster analysis of landraces, based on the similarities of traits, was conducted using the complete linkage method and a dendrogram was constructed (Figure 2). The cluster analysis results corroborate those of the PC biplot analysis (Figure 1). The dendrogram was divided into five major clusters with approximately 0.8 genetic similarities. The dendrogram grouped landraces with similarities as they appear in Table 3, where it was noted that Bamb1, Bamb4, Bamb5, and Bamb18 occurred in the same quadrant in the biplot and are grouped. Their association was due to the following traits: plant height, pod width, pod length, number of nodes per stem, and days to 50% flowering. Another similarity was observed with Bamb9, Bamb11, Bamb12, Bamb13, and Bamb14, which were associated with leaf area, terminal, leaflet length, and leaflet width. Furthermore, few dissimilarities were observed in certain landraces which were noticed to be further away from the other landraces. Bamb3 dissimilarities were due to the number of nodes per stem and days to 50% flowering, and Bamb7 dissimilarities were due to a number of pods per plant and chlorophyll content, while Bamb8’s dissimilarities were due to mass of seeds per plant, and finally Bamb20’s dissimilarities were due to date of maturity.

The investigation of variances (genotypic and phenotypic), the genotypic coefficient of variance and phenotypic coefficient of variance (PCV), broad sense heritability, and genetic advance are presented in Table 7. The results show that the genotypic variance reflects on phenotypic variance because relatively high values of both genotypic and phenotypic variance were obtained on 100 seed weight, yield per plant, chlorophyll content, and number of pods per plant. The results indicate that genotypic variance varied from 0.93 for terminal leaflet width to 3397.3 for a number of pods per plant. For phenotypic variance, the least value was also found on the terminal leaf width (3.66) and the top most being on the number of pods per plot (6498.03). The results further showed a greater variation in the PCV, ranging from 3.93% (date of maturity) to 67.33% for number of pods per plant. The highest values of 67.33% and 93.12% were recorded for GCV and PCV in the number of pods per plot. The following traits having values > 20% for GCV and PCV are considered high and those include number of nodes per stem, number of branches per stem, 100 seed weight, yield per plant, chlorophyll, terminal leaflet length, and number of pods per plant. Seed length and terminal leaflet width had medium PCV and GCV values (10% – 20%). Typically, the estimated values of heritability in the broad sense were high (h²b > 30) for almost all traits evaluated (Table 8). The heritability values ranged from 10.17% (pod length) to 57.22% (chlorophyll content). Moderate values (30% < h²b < 60%) were obtained for the following traits: number of nodes per stem (32.06%), number of branches per stem (36.32%), date of maturity (44.51%), 100 seed weight (42.28%), yield per plant (44.69%), chlorophyll content (57.22%), and number of pods per plant (52.29%). The genetic advance, as a percentage of the mean, ranged from 3.71% for pod length to 100.45% for number of pods per plant. The following traits, listed in their ascending order, had the high values of genetic advance (> 20%): number of nodes per stem (28.08), number of branches per stem (33.51%), hundred seed weight (34.83%), chlorophyll (62.09%), terminal leaflet length (69.95%), yield per plant (78.83%), and number of pods per plant (100.45%).

The leaflet shape of landraces differed significantly during plant growth. It was observed that most of the landraces (Table 8) produced elliptic leaflets (35%) followed by those with oval leaflets (30%), 20% had lanceolate leaflets, and 15% had round leaflets (Figure 3). The results also indicated that the colour of the flower and seed shape was 100%. That is, all the landraces studied produced yellow flowers and oval seed shapes (Figure 4 and Figure 5). A high percentage was also observed on green fully expanded leaves (80%), with the least being 20% (dark green fully expanded leaflets). With regards to the dark pigmentation on leaves, it was observed that most of the landraces produced leaves with zero pigments (85%) and only a few landraces had dark pigmentation (15%). There was variation in vegetative growth with three different types of growth habit; namely, bunch type, semi-bunch, and spreading type (Figure 5). Among these types, it was observed that semi-bunch type was the most common (65%) followed by the bunch type (25%), and the spreading type was less frequent (10%) (Table 8). It was observed that seed colour (Figure 4) varied significantly, producing eight different colours with different frequencies; namely in their descending order: black, red and yellowish (15%), cream with purple and brown (10%) and cream, light purple, and brown with purple dots (5%). In addition, pod colour and texture were recorded, where purple little grooves recorded the highest frequency of (60%) followed by black much grooved (15%) and the least was observed on yellowish-white pods (5%) (Figure 6). These results are not different from Kambou et al. (2020) who also recorded a high frequency of smooth texture on pods followed by smooth textured pods, which were dominant (Figure 6).

The Shannon Weaver index was relatively low (Table 8) for the 10 qualitative traits that were evaluated. The index ranged between 0.01 (leaf colour, seed shape, and flower shape) and 1.76 (seed colour). Compared to the rest of the qualitative traits studied, seed colour was the most polymorphic, followed by terminal leaflet shape (1.34).

The mean days to 50% flowering recorded in this study revealed significant variation among the tested Bambara groundnut landraces (Table 3). In the current study, it was observed that days to 50% flowering ranged from 75 to 84 days, which was higher than the range (51–65) that was reported by Abu and Buah (2011); Unigwe et al. (2016) on the characterisation of Bambara groundnut landraces and their evaluation by farmers in the Upper West Region of Ghana; morphological evaluation of selected accessions in South Africa, respectively, which also reported the lowest values for days to 50% flowering of 41–67 days, respectively. Such variation could be caused by several environmental factors such as temperature, altitude and soil conditions as well as spatial and temporal genotypic factors affecting flowering in Bambara groundnut. According to Unigwe et al. (2016), similar factors may be responsible for the variation in days to flowering. Number of branches per stem ranged from 9.67 to 16.3 (Table 4), which was higher than the values of 10.9–12.3 reported by Effa et al. (2016). This might be due to the presence of variation in the growing environments (Gerrano et al. 2013). One hundred seed weight ranged from 38.7 g to 59.3 g, with an average of 51.8 g. This variation is possibly due to different seed sizes for different landraces and this corroborates the findings by Valombola et al. (2019). Plant height ranged from 19.3 mm to 24.7 mm with an average of 23.2 mm, this was shorter than the range of height 26 mm – 27 mm, and 22.5 mm – 23.8 mm recorded by Effa et al. (2016) on fertilised landraces in south eastern Nigeria which were not significant and were shorter than the accessions selected in a study by Onwubiko, Uguru and Chimdi (2019). The possible explanation for this variation is that no fertiliser was used in the current study (Gerrano et al. 2015). Variations in seed length and seed width may be due to inherent differences in seed size and shape (Gerrano et al. 2017). Landraces showed that pod length and pod width ranged from 18.2 mm to 21.6 mm and 10.1 mm – 12.7 mm, which is similar to the Bambara accessions from Burkina Faso as reported by Kambou et al. (2020) whose findings were slightly higher than those found in the current study. The yield per plant ranged from 22.8 g to 79 g, which was much less than the findings recorded by Kambou et al. (2020) (100–378 g) and Massawe et al. (2005), which ranged from 84 g to 112.7 g. However, the findings of the current study showed that yield per plant was higher than the results that were recorded by Abedije et al. (2018), which ranged from 13 g to 51 g. These variations could be caused by several factors including environmental features, different seed sizes and the number of pods produced per plant and even the location in which they are planted. Similarly, Gerrano et al. (2013) and Siwale et al. (2022) reported the existence of phenotypic variation among diverse Bambara groundnut accessions in South Africa. Moreover, a study that was conducted in Nigeria also reported the presence of phenotypic variation among tested Bambara groundnut landraces (Esan, Oke & Ogunbode 2023). Zongo et al. (2023) also reported significant differences among the Bambara groundnut genotypes in Niger.

Correlation analysis is important in plant breeding because it measures the degree of genetic and non-genetic association between traits. In this study, the correlation coefficient suggested that the agromorphological traits were positively correlated (Table 5). It was observed that the number of branches per and plant height had a significant correlation (r = 0.50), which suggests that selecting plants with a high number of branches may lead to genotypes with longer internodes, and eventually the number of leaves will also be higher, resulting in higher yields Moreover, a highly significant and positive correlation was observed between chlorophyll content and the number of pods per plant (r = 0.74). This simply means that landraces with high chlorophyll content should be considered for selection. This is because the higher the chlorophyll content, the more fruits will be produced. Also, highly significant correlations were observed between terminal leaflet length with leaf area and terminal leaflet width (r = 0.71) and leaf area (r = 0.89). These results agree with those of Gerrano, Jansen van Rensburg and Adebola (2013) who also recorded a positive correlation between these traits, and attributed their results to the relationship between leaf length and leaf width. The study also showed a negatively significant correlation between the date of maturity and the number of pods per plant (r = -0.05). Esan et al. (2023) found significant positive correlations for most of the traits evaluated in Nigerian landraces.

The results of the PCA of the 17 quantitative morphological traits measured are presented in Table 6. The first six PCs contributed 78.35% of the variability among the 20 Bambara groundnut genotypes evaluated. Components with Eigenvalues >1 accounted for 78.34% cumulative variance and component loadings with ± 0.3 were regarded significant (Kutcher, Ferguson & Cohen 2013) and their traits were measured for the effect each has on the total cumulative variability. The number of nodes per stem, number of branches per stem, terminal leaflet width, leaf area and plant height accounted for 20.74% of the cumulative variance, suggesting that they should be considered in the selection for good agronomic performance (Valombola et al. 2019). Days to 50% flowering, correlated positively with PC3 in the current study, but negatively with PC3 on accessions from North Central Namibia (Valombola et al. 2019). Terminal leaflet width correlated positively with PC1 in the current study, but negatively with genotypes that were evaluated for selection by Mohammed et al. (2019). Generally, the analysis of 17 traits showed that PC1 consisted of several traits that contributed to the greatest variation followed by PC2. Mohammed (2014) recorded similar results where they observed that seed traits like seed length and seed width correlated positively with PC2. Principal biplot analysis showed how strongly each trait is associated with the other. The biplot also showed the association of the traits to the genotypes (Table 3).

The PC further explained agro-morphological similarities and variation among Bambara groundnut landraces relative to 17 measured traits (Figure 1). The quantitative traits led to the grouping of landraces into four quadrants representing 84.65% of the total variation. Bamb5, Bamb16, Bamb19, and Bamb20 were close to each other, implying that they share common traits among them. The shared traits are date of maturity, days to 50% flowering, plant height and terminal leaflet width. Again, landraces that are scattered far apart within the axes would mean that they are distantly related to other landraces within the same quadrant. Similar findings were recorded by Mohammed (2014). PC 1 and PC2 accumulated 73.88% and 10.77% of the total variation, respectively. This suggests that the first two principal components explained the large variations that existed among the accessions based on the extent of the traits contributing towards these variations.

The results indicated that landraces were grouped based on the relationship they have among them (Figure 2). For instance, Bamb1, Bamb4, Bamb5, and Bamb18 were grouped in one cluster and it was noticed that even in the principal biplot they are found in the same quadrant. The same applied to landraces that were scattered far apart from others in the biplot, for example, like Bamb7, Bamb8, and Bamb20. This could be due to some unique traits that are not shared between those landraces or that those landraces have a common origin in geographical representation and genetic background. These results agree with those recorded by Valombola et al. (2019) where it was found that genotypes grouped in one cluster had a common origin.

According to Owusu et al. (2021), selection for trait improvement does not only depend on available genetic variation but also on the degree of heritability for such variation. The ANOVA between genotypic and phenotypic variance showed that phenotypic values were relatively higher than genotypic values for all traits (Table 7). This is an indication that traits were influenced by the environment. Khan et al. (2020) reported similar findings, where the GCV and PCV were categorised as low as (0% – 10%), intermediate (10% – 20%) and high (> 20%). Both PCV and GCV values in the current study were from medium to high for most of the traits. This indicates that genetic progress could be made on either of the traits with high values through selection (Owusu et al. 2021). The results further indicated that most of the traits had medium to strong heritability and genetic advance, except for traits such as days to 50% flowering, pod length, pod width, seed length, seed width, terminal leaflet width, leaf area, plant height, and stand count. According to Owusu et al. (2021), selection could be easier if heritability is higher than 70%.

The results showed that, among the three types of growth habit, the semi-bunched morphotype was the most frequent followed by bunch type and the spreading type (Table 8). Similar observations were recorded by Ntundu et al. (2006) on landraces from Tanzania. Eighty percent of the landraces had a green fully expanded leaflet and 35% had elliptic leaflet shape. Bonny et al. (2019) and Khan et al. (2021) observed similar findings. The majority of pod textures had grooves (60%) and few smooth (5%). The Shannon weaver index indicated that seed colour (H’ = 1.76) and terminal leaflet shape (H’ = 1.34) were the most diversified traits out of 10 traits measured. These findings are not different from the findings by Zavinon et al. (2019).