Identification of high performing rice cultivars under coastal delta conditions through Principal Component and Cluster Analysis

Introduction

Stability of rice cultivars to the changing climatic conditions has become a key issue for sustainable crop production. CSA promotes the use of cultivars that are both resilient to environmental pressures and efficient in resource utilization, and are capable of sustaining high productivity (FAO, 2013). The ability to identify such cultivars involves assessing them across multiple agronomic traits and is beyond grain yield. There is substantial variability in cultivar traits related to growth duration, biomass production, canopy development and yield between rice cultivars, leading to a range in their ability to adapt to specific agro-ecological conditions. Many agronomic characters are interrelated and the evaluation of these traits separately may not be sufficient to indicate the overall performance or the relationships between cultivars. With this in mind, multivariate statistical methods can be used to integrate more than one of these traits to assess cultivar performance (Hair et al., 2019) [5] . Principal Component Analysis (PCA) is a multivariate powerful technique that allows the conversion of correlated variables into a reduced dimension of principal components that captures most of the variance (Yali and Nepir, 2026) [17] . This allows identification of specific traits that vary between cultivars and allows for the correct visualisation of genotype-trait relationships (Jolliffe, 2002) [7] . Hierarchical Cluster Analysis (HCA) is used to categorize cultivars into groups of cultivars that are similar in overall distance from one another (Johnson & Wichern, 2007) [6] , thus providing support for identification of superior cultivars for breeding and cultivation. Multivariate statistical techniques have been widely applied in rice improvement programmes to characterize genetic diversity, classify cultivars and identify promising genotypes for sustainable production under diverse environmental conditions (Nachimuthu et al., 2014; Hair et al., 2019) [5,11] . Such analyses provide valuable information for cultivar selection by integrating multiple growth and yield traits into a single analytical framework. Therefore, the present study employed Principal Component Analysis and Hierarchical Cluster Analysis to evaluate seven rice cultivars grown under the coastal agroecosystem of Karaikal.

Materials and methods

Seven rice cultivars, namely CO 51, ADT 46, KKL (R) 2, IR 20, BPT 5204, Karuppukavuni, and CR 1009, representing short, medium and long-duration groups, were evaluated under field conditions at PAJANCOA & RI, Karaikal, India, during Kharif 2025. The experiment was conducted in a randomized block design with three replications following the recommended package of practices (TNAU & DOA, 2020). Multivariate statistical analysis was performed using R software (version 4.x.x) to evaluate the performance of seven rice cultivars based on eleven growth and yield traits. The dataset comprised plant height at harvest (PH @H), number of tillers at harvest (NT @H), leaf area index at flowering (LAI @F), number of productive tillers at harvest (NPT @H), dry matter production at harvest (DMP @H), panicle length (LP), panicle weight (WP), test weight (TW), grain yield (GY), straw yield (SY) and days to harvest (DAYH). Prior to analysis, all variables were standardized to zero mean and unit variance to eliminate differences in measurement scales. Principal Component Analysis (PCA) was performed using the FactoMineR package (Le et al., 2008) [9] , and the principal components were retained based on the Kaiser criterion (eigenvalue > 1). The amount of contribution of each variable to the principal components were examined using loading values and contribution percentages. The relationships between cultivars, agronomic traits were graphically displayed using scree plots, variable correlation plots, treatment score plots and PCA biplots generated using the factoextra package (Kassambara & Mundt, 2020) [8] . Standardized trait data was used for Hierarchical Cluster Analysis (HCA). The dissimilarity matrix was generated according to the Euclidean distance and clusters were generated by Ward's minimum variance (Ward.D2) method (Ward, 1963; Teague and Awosile, 2026) [15,16] . The dendrogram was used to classify the rice cultivars based on their growth and yield attributes. The result of HCA was validated using K-means clustering. The within cluster sum of squares (WSS) elbow method was used to find the optimum number of clusters, and the cultivars were clustered in three clusters by using 25 random initializations (nstart = 25). Agronomic traits were clustered and the cluster-wise means were calculated for interpretation. The FactoMineR, factoextra, cluster and tidyverse packages were used to create all the graphical visualization and multivariate analyses in R.

Results and discussion Principal Component Analysis

The eleven agronomic traits used here to measure the variance among the seven cultivars of rice were well captured by PCA. The first three principal components had eigenvalues exceeding one (PC1 = 4.79, PC2 = 4.11 and PC3 = 1.29) and accounted for 92.68% of the total variation between cultivars. PC1 explained 43.59% of the total variance, PC2 explained 37.33% and PC3 explained 11.76%, showing that these components captured a significant portion of the variance in the data. The contribution of all other principal components was less than 6% and so contributed very little to the cultivar discrimination. The sum of variances above the 90% threshold shows that the selected principal components were able to explain an adequate amount of the variation between rice cultivars, which is necessary for good interpretation of multivariate analysis (Abdi & Williams, 2010; Rencher & Christensen, 2012; Azimov et al., 2026) [1,2,13] (Table 1; Figure 1). The scree plot also showed that the majority of information was contained within the first three principal components, with the eigenvalues dropping quickly after PC3, in line with the selection of these components for further interpretation (Figure 1).

Contribution of Agronomic Traits to Principal Components

The loading matrix showed that some agronomic traits were highly involved in the separation of cultivars. Number of tillers at harvest had a strong negative correlation (-0.866) with PC1, while plant height at harvest (0.509), leaf area index at flowering (0.852), panicle weight (0.841), days to harvest (0.909), dry matter production (0.588), grain yield (0.589) and test weight (0.600) had positive correlations. Biomass accumulation, as reflected in dry matter production, grain yield and straw yield, was one of the most important factors that accounted for the observed variability and was an important component in differentiating the performance of cultivars ( [10,18] .

Cluster Analysis

The rice cultivars were clustered into three clusters based on their similarity in growth and yield performance using the hierarchical cluster analysis. The dendrogram showed good discernment of the cultivars into groups that have different agronomic characteristics, reflecting high phenotypic variability of the materials evaluated (Figure 7). The elbow method was used to identify the best number of clusters for the K-means analysis and the within cluster sum of squares (WCSS) values were plotted against k values. The elbow plot was very clearly convex at k = 3, suggesting three clusters were the most suitable number for partitioning the rice cultivars (Figure 8). The seven cultivars of rice were divided into three groups by K-means clustering that was based on optimal cluster number obtained by the hierarchical clustering. Cluster I comprised ADT 46 (V2), KKL (R) 2 (V3), IR 20 (V4), BPT 5204 (V5) and CR 1009 (V7), indicating relatively similar agronomic performance. The trait combination of Karuppukavuni (V6) differed from the other varieties and it was grouped separately due to its higher plant height, panicle weight and lower grain yield than the other varieties (Table 3; Figure 9). The similarity of the hierarchical and K-means cluster shows that the cultivar classification is reliable and there is a high level of phenotypic diversity among the rice cultivars evaluated. Other clustering methods are extensively adopted for the classification of genotypes according to several quantitative traits (Rao, 1952; Johnson & Wichern, 2007) [6,12] .

Identification of High-Performing Rice Cultivars

Combination of this interpretation with that of cluster analysis showed the cultivar CR 1009 (V7) to be the most superior, due to its high grain yield (6,298 kg ha -1 ), high dry matter yield (15,944 kg ha -1 ), highest LAI at flowering (7.27) and the longest crop duration (149 days), which makes this cultivar a good prospect for coastal delta conditions. KKL (R) 2 (V3) and ADT 46 (V2) also performed well and had higher LAI (6.31 and 6.54 respectively) and a balance in growth and yield attributes. Karuppukavuni (V6), on the other hand, had an unusually high plant height (180.96 cm) and panicle weight (4.15 g), but grain yield (2,892 kg ha -1 ) was relatively low, which suggests that a higher vegetative growth and biomass accumulation alone does not necessarily translate into higher grain productivity and differences in source-sink balance and assimilate partitioning efficiency between the two varieties (Smith et al., 2018) [14] . CO 51 (V1) had low productivity as compared with the rest, lower LAI at flowering (2.74) and shorter crop duration (106 days) making it distinct (Table 3; Figures 4, 7 and 8). This information demonstrates the value of multivariate analysis to the identification of superior rice cultivars, based on multiple growth and yield traits instead of grain yield. Other applications of multivariate techniques for identification of superior crop genotypes by combining several yield components are reported by Mohammadi & Prasanna (2003) [10] .

Conclusion

The diversity among cultivars of rice was found using multivariate analysis, which was effective. PCA identified the major traits contributing to cultivar variation, while hierarchical and K-means clustering grouped cultivars into distinct clusters with three optimal groups confirmed by the elbow method. These findings demonstrate the usefulness of multivariate approaches for cultivar characterization, selection of promising genotypes and supporting future rice improvement programmes.

References

1. Abdi H, Williams LJ. Principal component analysis. Wiley Interdiscip Rev Comput Stat. 2010;2(4):433-459.
2. Azimov A, Shavkiev J, Koraev A. Genetic diversity, population structure, and salinity tolerance of upland cotton (Gossypium hirsutum L.) germplasm revealed by SSR markers. 2026.
3. Cruz CD, Regazzi AJ, Carneiro PCS. Modelos biométricos aplicados ao melhoramento genético. 4th ed. Viçosa: UFV; 2012.
4. Food and Agriculture Organization of the United Nations. Climate-smart agriculture sourcebook. Rome: Food and Agriculture Organization of the United Nations; 2013.
5. Hair JF, Black WC, Babin BJ, Anderson RE. Multivariate data analysis. 8th ed. Cengage Learning; 2019.
6. Johnson RA, Wichern DW. Applied multivariate statistical analysis. 6th ed. Pearson Prentice Hall; 2007.
7. Jolliffe IT. Principal component analysis. 2nd ed. Springer; 2002.
8. Kassambara A, Mundt F. factoextra: Extract and visualize the results of multivariate data analyses. R package version 1.0.7; 2020.
9. Le S, Josse J, Husson F. FactoMineR: An R package for multivariate analysis. J Stat Softw. 2008;25(1):1-18.
10. Mohammadi SA, Prasanna BM. Analysis of genetic diversity in crop plants salient statistical tools and considerations. Crop Sci. 2003;43(4):1235-1248.
11. Nachimuthu VV, Robin S, Sudhakar D, Raveendran M, Rajeswari S, Manonmani S. Evaluation of rice genetic diversity and variability in a population panel by principal component analysis. Indian J Sci Technol. 2014;7(10):15551562.
12. Rao CR. Advanced statistical methods in biometric research. New York: John Wiley & Sons; 1952.
13. Rencher AC, Christensen WF. Methods of multivariate analysis. 3rd ed. John Wiley & Sons; 2012.
14. Smith MR, Rao IM, Merchant A. Source-sink relationships in crop plants and their influence on yield development and nutritional quality. Front Plant Sci. 2018;9:1889.
15. Teague E, Awosile B. Distribution and diversity of Salmonella serovars from beef cattle before, at, and after processing in relation to human-derived serovars. Front Vet Sci. 2026;13:1897484.
16. Ward JH Jr. Hierarchical grouping to optimize an objective function. J Am Stat Assoc. 1963;58(301):236-244.
17. Yali W, Nepir G. Assessment of genetic diversity among elite sorghum [Sorghum bicolor (L.) Moench] genotypes based on agro-morphological traits using principal component and cluster analyses under drought-stressed environments. Ecol Genet Genomics. 2026;100503.
18. Yan W, Kang MS. GGE biplot analysis: A graphical tool for breeders, geneticists, and agronomists. Boca Raton: CRC Press; 2003.