Principal component analysis (PCA) of 21 traits in 130 rice (
Oryza sativa L.) landraces identified eight principal components with eigenvalues >1, explaining 71.10% of total phenotypic variation, indicating substantial genetic diversity (Table 1). PC1 explained 16.16% of the variation and was mainly associated with plant height at maturity (X
3), flag leaf length (X
4), panicle length (X
6), internode length (X
7), number of productive tillers (X
8), fresh root weight (X
12), fresh shoot weight (X
13), pollen fertility (X
19) and single plant yield (X
21). These results suggest that plant architecture, biomass accumulation and reproductive efficiency are major determinants of yield variability, consistent with
Anderson (1972);
Rahangdale et al., (2021); Chandraker et al., (2024) and
Khatun et al., (2023); Aisya et al., (2026).
PC2 accounted for 13.21% of the variation and was largely influenced by biomass-related traits, plant height (X
3) and pollen fertility (X
19), indicating the importance of biomass production and physiological efficiency. PC3 explained 10.34% of the variation and was associated with days to 50% flowering (X
1), days to maturity (X
2), grains per panicle (X
9), 1000-seed weight (X
10), leaf area coverage (X
18) and single plant yield (X
21), highlighting the role of source-sink relationships in productivity. Similar observations were reported by
Gunasekaran et al., (2017); Choudhary et al., (2022); Kumar et al., (2026) and
Kim et al., (2024) (Fig 1). The remaining PCs contributed smaller proportions, confirming that yield, biomass and reproductive traits were the major contributors to genetic divergence.
UPGMA cluster analysis based on Euclidean distance grouped the 130 landraces into four clusters (Table 2; Fig 2), confirming substantial diversity, as also reported by
Singh et al., (2020); Sahu et al., (2021); Swarup et al., (2021) and
Sinha et al., (2023) and
Islam et al., (2020); Kumar et al., (2026). Cluster I exhibited superior performance for yield-related traits, including plant height (X
3), panicle length (X
6), productive tillers (X
8), grains per panicle (X
9), single plant yield (X
21) and pollen fertility (X
19), indicating high yield potential. Cluster II showed moderate but stable performance, whereas Cluster III recorded lower yield traits and pollen fertility, reflecting poor reproductive efficiency. Cluster IV exhibited greater biomass accumulation through fresh and dry shoot weight (X
13 and X
15), although this was not accompanied by higher grain yield. Similar relationships were reported by
Sheela et al., (2020); Lakshmi et al., (2022); Kumar et al., (2021); Thakur and Sarma (2023) and
Krishna et al., (2022).
Inter-cluster distance analysis revealed maximum divergence between Cluster III and Cluster IV, followed by Cluster I and Cluster IV, suggesting greater scope for genetic recombination through hybridization. Minimum divergence occurred between Cluster I and Cluster II, indicating close genetic similarity, in agreement with
Singh et al., (2020) and
Sinha et al., (2023).
The PCA biplot clearly separated high- and low-performing genotypes based on trait associations (Fig 3). Traits such as pollen fertility (X
19), fresh root weight (X
12), fresh shoot weight (X
13), panicle length (X
6), internode length (X
7), productive tillers (X
8) and single plant yield (X
21) showed longer vectors, indicating greater contributions to variability. Similar interpretations were reported by
Christina et al., (2021), Khaire et al., (2022) and
Choudhary et al., (2022). Trait contribution analysis identified single plant yield (X
21), productive tillers (X
8), flag leaf length (X
4) and leaf width (X
5) as major contributors to genetic divergence and useful selection criteria for yield improvement, whereas days to flowering (X
1) and maturity (X
2) contributed less to divergence (Table 3), as also reported by
Manoj et al., (2022), Satyanarayana et al., (2023) and
Madhukumar et al., (2023); Balachandran et al., (2026).
Pollen fertility (X
19) showed a strong positive association with yield-related traits and significantly influenced PCA clustering patterns, highlighting its importance in grain setting and reproductive success. It serves as an important indirect selection criterion for improving rice grain yield, as reported by
Reddy et al., (2021) and
Kim et al., (2024). The scree plot showed a gradual decline in variance contribution across principal components, with PC1, PC2 and PC3 accounting for 16.2%, 13.2% and 10.3% of total variation, respectively, while PC4-PC6 contributed smaller proportions (Fig 4). Similar patterns were reported by
Pokhrel et al., (2020); Chandraker et al., (2024) and
Khatun et al., (2023).
Cluster I exhibited superior yield traits, productive tillers (X
8), panicle length (X
6) and high pollen fertility (X
19), indicating high yield potential. Cluster IV showed greater biomass accumulation through higher fresh and dry shoot weight (X
13 and X
15), suggesting stronger vegetative vigor and adaptability. Cluster III had lower pollen fertility and poor yield performance, indicating reduced reproductive efficiency. Similar relationships between physiological traits, reproductive efficiency and adaptability were reported by
Lakshmi et al., (2022); Kim et al., (2024) and
Krishna et al., (2022).
The study identified genetically diverse landraces and key yield-related traits useful for rice breeding. Hybridization between Cluster III × Cluster IV and Cluster I × Cluster IV may effectively exploit heterosis, as suggested by
Singh et al., (2020); Sahu et al., (2021) and
Sinha et al., (2023). Productive tillers (X
8), panicle length (X
6), grains per panicle (X
9), single plant yield (X
21) and pollen fertility (X
19) were identified as important selection traits, while flag leaf length (X
4), leaf width (X
5) and biomass-related traits (X
12 -X
15) may also contribute to improvement.
Cluster mean analysis revealed substantial variation among clusters. Cluster I recorded higher values for plant height, panicle length, productive tillers, grains per panicle, pollen fertility and single plant yield, making it a valuable high-yielding parental source. Cluster IV exhibited greater biomass accumulation, whereas Cluster III showed lower pollen fertility and grain yield but greater genetic divergence.
Maximum inter-cluster divergence was observed between Cluster III and Cluster IV and between Cluster I and Cluster IV (Table 4), suggesting good potential for heterosis breeding. Cluster I × Cluster IV crosses may combine superior yield with high biomass, while Cluster III × Cluster IV crosses may generate diverse segregants. Similar findings were reported by
Singh et al., (2020), Sahu et al., (2021) and
Swarup et al., (2021). Traits such as productive tillers (X
8), panicle length (X
6), grains per panicle (X
9), pollen fertility (X
19) and single plant yield (X
21) can serve as effective selection criteria, while biomass-related traits may contribute to adaptation and stress resilience (Fig 5).
Multiple linear regression analysis revealed a strong association between the studied traits and grain yield (R = 0.955), with the model explaining 91.1% of total variation (R
2 = 0.911). The adjusted R
2 value (0.896) and low standard error (8.25) confirmed model reliability (Table 5), consistent with findings of
Gunasekaran et al., (2017) and
Sudeepthi et al., (2017).
Regression analysis identified number of productive tillers per plant (NPT), number of grains per panicle (NGP) and thousand-grain weight (TGW) as the most influential traits affecting grain yield. These traits showed highly significant positive effects (P<0.001); agreeing with
Pandya and Sarial (2015);
Singh et al., (2020) and
Singh et al., (2023). NGP had the highest standardized coefficient (β = 0.630), followed by NPT (β = 0.511) and TGW (β = 0.472) (Table 6) indicating that grain number per panicle was the strongest determinant of yield. Similar observations were reported by
Jan et al., (2017); Adhikari et al., (2018). In contrast, the remaining agronomic, physiological and growth traits did not show significant direct regression effects on grain yield, although some may contribute indirectly, supporting earlier reports by
Gunasekaran et al., (2017) and
Pandya and Sarial (2015).
