Classification of Winter Wheat Genotypes Based on Grain Yield Components Using K-means Clustering in Multi-environment Trials

Abstract

Understanding the interaction between grain yield and its main components is essential for improving wheat productivity and guiding breeding strategies. In this study, K-means clustering was applied to classify 36 winter wheat genotypes evaluated in multi-environment trials across four locations in Bulgaria over four growing seasons. The analysis was based on grain yield (GY), number of grains m^–2 (NGM), and thousand grain weight (TGW), representing key processes of yield formation.

The optimal number of clusters was determined using the Elbow method and Silhouette coefficient, which identified k = 2 as the most robust solution, while k = 4 provided additional biological resolution. The two-cluster classification revealed a clear separation of genotypes into contrasting yield formation strategies defined by the trade-off between grain number and grain weight.

The four-cluster solution further refined genotype classification, identifying intermediate and high-performing groups, including genotypes combining high grain number and high yield. This suggests that the classical trade-off between yield components can be partially overcome.

Overall, the results confirm the dominant role of grain number in yield determination, while highlighting the compensatory contribution of grain weight. The application of K-means clustering provides a robust framework for identifying biologically meaningful genotype groups and supports its use as a practical tool for selection in wheat breeding programmes under variable environmental conditions.