The plant materials used in this study included 65 aromatic rice lines (Table 1). Twenty-seven came from Africa Rice Saint Louis Station as part of the Korea-Africa Food and Agriculture Cooperation Initiative (KAFACI). Thirty-three recombinant lines came from our national rice germplasm at the Institut de l’Environement et de Recherches Agricoles (INERA) and the remaining five lines were obtained from the Africa Rice Senegal core collection. The Orylux6 rice variety, widely used in Burkina Faso by farmers, was included among the 65 lines and used as control.

Table 1.List of plant materials used in this study.

^aKAFACI: Korea Africa for Food and Agriculture Cooperation Initiative; INERA: Institut de l’Environement et de Recherches Agricoles. ^bRIL: Recombinant Inbreed Line; DH: Double Haploid.

2.2.1. Field Trials

The study was conducted at the INERA agricultural research station of Kamboinsé (12.4467˚ N, −1.5625˚ W) over two consecutive years. Each year included two cropping seasons: a rainy season (June-October) and a dry hot season (November-May). Consequently, the experiments were evaluated in four environments defined as season–year combinations: 2022 rainy season (E1), 2022 dry season (E2), 2023 rainy season (E3), and 2023 dry season (E4). These four environments were used in the genotype × environment interaction analysis.

Field trials were arranged following a Randomized Complete Block Design (RCBD) with three replications. Elementary plots (1.2 m²) were spaced 40 cm apart lengthwise and 50 cm apart widthwise, each bearing 20 plants in 20 cm × 20 cm spacings. Soil preparation consisted of flat ploughing followed by crushing, levelling and application of bottom fertilizer composed of 2.5 t ha⁻¹ of organic fertilizer and 200 kg ha⁻¹ of Nitrogen Phosphate Potassium (NPK 14-23-14). Rice seedlings were transplanted 14 days after germination. Urea (46%) was used as a booster at a rate of 150 kg ha⁻¹ in three equal fractions at 10, 30 and 60 days after transplanting. Irrigation was punctual in the event of pockets of drought in the wet season and permanent in the dry season. Yield (RDT) was determined from paddy rice harvested on a net plot area of 2.16 m², excluding border rows to minimize edge effects. Grain weight was measured and adjusted to a standard moisture content of 14% before conversion to yield expressed in t ha⁻¹.

2.2.2. Data Collection and Analysis

Field data were collected on nine qualitative traits (Table 2) and 10 quantitative traits. The qualitative traits were mainly DUS (Distinction, uniformity, Stability) descriptors [10] including basal leaf sheath pigmentation, panicle exertion, awn development, flag leaf attitude, stigma color, panicle attitude, glume length class, lemma color, and ligule shape. Quantitative variables were the following:

i) 50% panicle heading cycle (CSE) as the number of days from sowing to 50% panicles heading.

ii) 85% maturity cycle (CSM), representing the number of days from sowing to maturity of 85% of plants.

iii) plant height (HP), corresponding to the height of five random plants (cm) from each plot at maturity stage.

iv) Panicles number (NP) referring to the number of fertile panicles assessed on five random plants from each plot at crop maturity.

v) Panicles length (LP), corresponding to the length (cm) of the panicles from five random plants at crop maturity.

vi) Paddy grain length (Lg) indicating the distance (mm) between the two ends of the paddy grain on the longitudinal plane from 10 random grains for each line after harvest.

vii) Paddy grain width (lg) corresponding to the distance (mm) between the two ends of the paddy grain on the transverse plane from 10 random grains for each line after harvest.

viii) Paddy grain length/width ratio indicating ratio between the length and width of 10 random individual grains after harvest.

ix) Weight of 1000 paddy grains (PMG) measured using an electronic scale with a precision of 0.001g.

x) Paddy grain yield per hectare (RDT)indicating the grain production of the entire plot (t ha⁻¹).

Table 2.List of qualitative traits studied.

2.2.3. Participatory Varietal Selection

A total of 21 female and 19 male farmers from the surrounding localities were included in the participatory varietal selection to show their preference among 65 rice lines. Participants used their own criteria for making their decisions, but they were interviewed individually.

2.2.4. Statistical Analysis

A combined analysis across the four environments was performed using the following linear model:

where Y i j k is the observed value of genotype k in replication j within environment i; μ is the overall mean; E_i is the fixed effect of the i^th environment; R j ( E i ) is the effect of replication j nested within environment i;G_k is the effect of genotype k; ( G E ) i k is the genotype x environment interaction; and ε i j k is the residual error.

Quantitative agronomic data were subjected to combined analysis of variance using the Additive Main Effects and Multiplicative Interaction (AMMI) model to partition the total variation into Genotype (G), Environment (E), and genotype × environment (G × E) interaction effects. The interaction component was further decomposed into Interaction Principal Component Axes (IPCAs) to assess the pattern and magnitude of genotype stability across environments. To complement AMMI, the Weighted Average of Absolute Scores from the BLUP of the GEI effects (WAASB) index was calculated to quantify genotype stability by integrating all significant interaction principal components into a single stability metric [11]. Principal Component Analysis (PCA) was performed on standardized quantitative traits using genotype mean values across environments to explore multivariate relationships among agronomic variables and to identify the main sources of phenotypic variation. The use of genotype means allowed the characterization of overall genotype performance while reducing the influence of within-environment experimental variation despite the presence of significant genotype × environment interaction. For qualitative morphological descriptors, Multiple Correspondence Analysis (MCA) was conducted to examine the structure of categorical data and to detect morphologically distinct genotype groups. Hierarchical Clustering on Principal Components (HCPC) was applied to the MCA results to classify genotypes into homogeneous clusters. All statistical analyses were conducted using R software [12], primarily with the following main packages: metan [13], FactoMineR [14], factoextra [15], and corrplot [16].

Table 3 summarizes the responses of rice lines for the quantitative traits. The combined analysis of variance revealed highly significant effects of genotype (F_64;3632 = 44.14; p < 0.001), environment (F_3;3632 = 276.04; p < 0.001), and genotype × environment interaction (F_192;3632 = 8.77; p < 0.001). Coefficients of variation ranged from 3.4% to 27.2%. The mean grain yield was higher in 2022 (7.65 t ha⁻¹) compared to 2023 (7.02 t ha⁻¹), representing a reduction of approximately 8.3% in the second year. Yield variability also differed between years. The Coefficient of Variation (CV) reached 28.69% in 2022 and increased to 31.61% in 2023, indicating greater dispersion of genotype performance under the environmental conditions in 2023.

The 50% panicle heading cycle (CSE) parameter ranged from 59.33 days in rice line Kayan to 98 days (rice line KD) with an average of 87.18 days. The control line Orylux6 had a CSE of 79 days which was significantly different (p < 0.05) from CSE of 47 lines.

CSM ranged from 79.25 days (rice line Kayan) to 127.83 days (rice line Moussoni), with an average of 114.56 days. The Orylux6 control matured in 105 days. A total of 45 lines had a maturity cycle significantly different from that of the control. Plant height (HP) varied between 91.67 cm (in line K20-9) and 175.33 cm (in line K20-60) with an average of 116.4 cm.

Table 3.Summary of quantitative trait values.

^aCSE-50% = time (days) from sowing to 50% panicle heading; CSM-85% = time (days) from sowing to 85% crop maturity; HP = plant height (cm); lg = grain width (mm); Lg = grain length (mm); ratio = grain length/grain width ratio; LP = panicle length (cm); NP = panicles number; PMG = weight of 1000 grains (g); RDT = grain yield (t ha⁻¹), CV = Coefficient of Variation, P = Probability.

The mean plant height for the control Orylux6 was 115.8 cm and differed significantly from plant height in 40 lines. Panicle length (LP) ranged from var 21.51cm (rice line Kossi2) to 29.91cm (rice lines K20-60 and Basmati370), the average LP being 25.59 cm. With a mean LP of 24.85 cm, the control line differed significantly from seven other lines. Per plant panicle number (NP) was lowest in Remar11 (NP = 6.9) and highest in Fangatigui (NP = 19.35) with an average of 13.01 across the lines. The NP in control line Orylux6 was 16.0 which made it significantly different from 30 lines. The 1000-grain weight (PMG) was also highly variable, ranging between 16.56 g in line Kayan and 34.15 g in line Remar13. The overall average PMG was 25.41 g. The control had a PMG of 20.84 g and differed from 60 lines. Variable grain dimensions were found in grain length (Lg: 6.27 - 11.51 mm), grain width (lg: 2 - 2.71 mm) and grain ratio (2.7 - 5.1). The overall average values for the three parameters were 9.23 mm (Lg), 2.33 mm (lg) and 3.9 (ratio), respectively. Grains of the control line Orylux6 were 9.04 mm long, 2.03 mm wide and 4.5 ratio, respectively. In total, the control line differed from 48 lines in Lg, 59 lines in lg and 59 lines in Lg/lg ratio. Finally, grain yield (RDT) strongly discriminated among the evaluated lines, ranging from 4.94 t ha⁻¹ (in line Kayan) to 9.7 t ha⁻¹ (in line Fangatigui) and averaging at 7.1 t ha⁻¹. The control produced 6.7 t ha⁻¹, which made it significantly different from 19 rice lines.

The combined ANOVA revealed a highly significant genotype × environment (G × E) interaction (F_192;3632 = 8.77; p < 0.001). Therefore, the additive main effects and multiplicative interaction (AMMI) model was used to partition interaction patterns. Genotype effects were highly significant (p < 0.001), indicating substantial genetic variability for grain yield among the evaluated lines. Although the main environmental effect was not significant (p = 0.143), the magnitude of the G × E Interaction (GEI) confirmed differential genotype responses across environments.

The first two interaction principal component axes (IPCA1 and IPCA2) explained 60.3% and 38.9% of the GEI sum of squares, respectively, accounting for 99.2% of the total interaction variation. The statistical significance of these axes indicated that the interaction was largely structured rather than random, supporting the adequacy of the AMMI2 model for describing genotype performance.

To complement AMMI2, the Weighted Average of Absolute Scores (WAASB) was computed to quantify stability. The joint evaluation of mean yield and WAASB in a performance × stability quadrant enabled the identification of genotypes combining high productivity and broad adaptation (high yield, low WAASB) (Figure 1). This integrated approach strengthened selection decisions by simultaneously considering productivity and stability across environments.

The WAASB stability analysis identified several genotypes combining high productivity and stability across environments (Table 4). Kossi4 and IR841 showed the highest mean grain yield (9.7 t ha⁻¹) with WAASB values of 0.1622 and 0.3108, respectively, indicating strong performance with acceptable stability across environments. Other genotypes such as Burkindi, Sahel210, and K20-Germ4 exhibited very low WAASB values, reflecting high stability, although their mean yields were comparatively lower.

Figure 1. Performance × Stability quadrant base mean grain yield and WAASB index. Quadrants (Q1 - Q4) are indicated along with their description.

Table 4.High yielding and stable rice genotypes identified in the study.

Correlation analysis (Table 5) indicated a strong positive association between CSE and CSM (r = 0.96). Plant height (HP) was positively correlated with LP (r = 0.64) and negatively correlated with NP (r = −0.58). NP showed negative correlations with lg (r = −0.40) and PMG (r = −0.32) but was positively associated with the ratio variable (r = 0.39). A positive correlation was also observed between Lg and PMG (r = 0.70), while the ratio trait was positively correlated with Lg (r = 0.55) and negatively correlated with lg (r = −0.64). Grain yield (RDT) was moderately correlated with CSM (r = 0.52), CSE (r = 0.46), and NP (0.30).

Table 5.Pearson correlation matrix among agronomic traits^a.

^aCSE = time (days) from sowing to 50% panicle heading; CSM = time (days) from sowing to 85% crop maturity; HP = plant height (cm); lg = grain width (mm); Lg = grain length (mm); ratio = grain length/grain width ratio; LP = panicle length (cm); NP = panicles number; PMG = weight of 1000 grains (g); RDT = grain yield (t ha⁻¹); correlation coefficients with a star sign are significant at p < 0.05.

Principal Component Analysis (PCA) revealed that the first two components explained 39.5% of the total phenotypic variation among the evaluated rice lines, with PC1 and PC2 accounting for 20.7% and 18.8% of the variance, respectively (Figure 2). PC1 was primarily associated with phenological traits, namely days to 50% panicle heading (CSE; 0.80) and days to 85% maturity (CSM; 0.81), together with grain-related traits such as grain length (Lg; 0.57) and thousand-grain weight (PMG; 0.45). This pattern indicates that differences in crop cycle duration and grain morphology were major sources of variation among genotypes.

The strong alignment between CSE and CSM reflects the close developmental relationship between flowering and maturity timing. PC2 was characterized by positive loadings for plant height (HP; 0.57) and small grain length (Lg; 0.70), and negative loadings for number of panicles (NP; −0.59) and the ratio trait (−0.47), suggesting a contrast between plant architectural traits and reproductive allocation. Grain yield (RDT) showed relatively low loadings on both PC1 (0.22) and PC2 (0.004), indicating that yield variation was not strongly structured along the first two principal components but rather distributed across multiple trait dimensions.

Figure 2. Principal component analysis biplot of evaluated traits. CSE = time (days) from sowing to 50% panicle heading; CSM = time (days) from sowing to 85% crop maturity; HP (cm) = plant height (cm); lg = grain width (mm); Lg = grain length (mm); ratio = grain length/grain width ratio; LP = panicle length (cm); NP = panicles number; PMG = weight of 1000 grains (g); RDT = grain yield (t ha⁻¹).

Multiple Correspondence Analysis (MCA) performed to explore the structure of qualitative agromorphological traits indicated that the first two dimensions explained 23% and 22% of the total inertia, respectively (Figure 3). This accounted for 45% of the overall variation. The projection of individuals on the factorial plane revealed a clear structuring of genotypes. Hierarchical Clustering on Principal Components (HCPC) identified three distinct morphological groups.

The association between cluster membership and categorical traits was highly significant, as confirmed by chi-square tests for flag leaf attitude (p = 3.89 × 10⁻²⁷), ligule shape (p = 1.09 × 10⁻¹⁶), lemma color (p = 1.09 × 10⁻¹⁶), panicle attitude (p = 7.68 × 10⁻¹⁵), and awn presence (p = 9.39 × 10⁻¹¹). Cluster characterization based on v-test statistics revealed that architectural traits were the primary drivers of separation. Cluster 1 was strongly defined by erect flag leaf attitude (v-test = 6.64, p < 0.001) and the presence of awns (v-test = 5.79, p < 0.001), while semi-erect leaves and awn absence were significantly underrepresented (negative v-tests). Cluster 2, representing most genotypes, was significantly associated with semi-erect flag leaf attitude (v-test = 7.12, p < 0.001), awn absence (v-test = 5.10, p < 0.001), and semi-drooping panicle attitude (v-test = 2.22, p < 0.05). In contrast, Cluster 3, composed of two highly divergent genotypes, was characterized by drooping panicle attitude (v-test = 3.49, p < 0.001) and strongly reflexed flag leaf attitude (v-test = 3.49, p < 0.001), with complete absence of semi-erect leaf types.

Figure 3. Factorial map of genotypes based on Multiple Correspondence Analysis (MCA) of qualitative agromorphological descriptors.

As summarized in Table 6, participatory varietal selection revealed similar preferences between men and women farmers across evaluated criteria (p > 0.05 for all traits). Productivity was highly prioritized by both groups (80% of men vs. 71.42% of women), although the difference was not statistically significant (χ² = 0.033, p = 0.86). Early maturity was slightly more emphasized by women (71.42%) than men (60%), but this difference was also non-significant (χ² = 0.318, p = 0.57). Similarly, plant height was considered important by 76.19% of women compared to 60% of men (χ² = 0.802, p = 0.37), while grain length was valued by 90.48% of women and 80% of men (χ² = 0.332, p = 0.56). Panicle number received the highest overall priority, being selected by all men (100%) and 90.48% of women, without significant gender differences (χ² = 0.427, p = 0.51).