A Study on the Infrageneric Classification of Hordeum Using Multiple Methods: Based on Morphological Data

Abstract

The genus Hordeum (barley) represents an essential group within the Poaceae family, comprising diverse species with significant ecological and economic importance. This study aims to improve the infrageneric classification of Hordeum by integrating multiple analytical approaches based on morphological data. A comprehensive dataset of key morphological traits was compiled from a wide range of Hordeum accessions, including representatives from all major taxonomic groups within the genus. Understanding and classifying the evolutionary traits of barley species, particularly in terms of environmental adaptation, pest resistance, and productivity improvement, is essential. DNA-based classification methods allow precise molecular-level analysis but are resource-intensive, especially when large-scale processing is required. This study addresses these limitations by employing an integrative approach combining hierarchical clustering, Principal Component Analysis–Linear Discriminant Analysis (PCA-LDA), and Random Forest (RF) to analyze the compiled morphological datasets. Morphological clustering via hierarchical analysis revealed clear taxonomic distinctions, achieving 86.0% accuracy at the subgenus level and 83.1% at the section level. PCA-LDA further refined classification by identifying key traits such as seed width, area, and 100-seed weight as primary contributors, achieving perfect accuracy for the Hordeum section and high accuracy for species like Hordeum vulgare and Hordeum spontaneum. RF analysis enhanced classification performance, achieving 100% accuracy at the section level and high accuracy for species with sufficient data. This approach offers a new framework for classifying diverse barley species and contributes significantly to data-driven decision-making in breeding and conservation efforts, supporting a deeper understanding of barley’s adaptive evolution in response to environmental changes.

1. Introduction

The genus Hordeum, belonging to the Poaceae family, comprises approximately 33 species and 45 subspecies worldwide. These species are primarily distributed across temperate regions, each adapting to specific native habitats and environmental conditions. Within the Hordeum genus, the widely cultivated barley (Hordeum vulgare) is prominent. Additionally, it includes wild barley (Hordeum spontaneum) and several wild species utilized for livestock feed and pasture development. From the data on wild barley species in the Royal Botanic Gardens, Kew database (“Hordeum species profile”), Hordeum stenostachys has narrow and long spikes and thrives well in high-altitude mountainous regions. Hordeum brachyantherum has short, compact spikes with relatively short awns and grows in wetlands and coastal areas of North America. Hordeum jubatum possesses spikes with long awns and adapts well to cold environments such as those in North America and Siberia. Hordeum murinum is characterized by hairy stems and leaves, making it well-suited to arid environments. Hordeum marinum grows in coastal areas and regions with high salinity. Hordeum procerum is tall with relatively long spikes and adapts well to low temperatures and high altitudes, such as in mountainous regions. Hordeum intercedens has short awns and a restricted habitat, being found in North and Central America. Barley exhibits remarkable environmental adaptability, thriving in diverse ecosystems ranging from cold highlands to arid desert areas. Given its significance as a global food and ecological resource, barley is pivotal in research focused on preserving and utilizing genetic diversity [1,2,3,4,5]. Classifying various barley species and understanding their evolutionary traits is critical for developing and disseminating resilient cultivars, especially as climate change poses significant challenges [6]. Traditional DNA-based classification methods enable precise molecular-level analysis, yet they require extensive time and resources, making them less practical for rapid, high-throughput classification. These methods also demand specialized laboratory equipment and skilled expertise, posing limitations for real-time or large-scale applications [7].

To date, research on phenotypic diversity within barley species has primarily focused on evaluating the phenotypic and genetic changes associated with environmental adaptation in Hordeum vulgare subsp. spontaneum, known as a wild relative of cultivated barley [8,9,10], and among a number of other barley species; studies on interspecies phenotypic characteristics have been limited to a few species [11,12,13,14]. Although phenotypic classification of plants presents numerous challenges, recent studies have explored machine learning techniques for efficient classification and dimensionality reduction. The combination of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) has gained attention. PCA-LDA is recognized for its ability to extract key variables from large datasets and improve classification accuracy. PCA effectively reduces dimensions by identifying essential features, while LDA maximizes class separability based on these features [15,16].

Furthermore, studies have shown that the Random Forest (RF) algorithm, which combines multiple decision trees, is capable of accurately classifying specimens based on multiple variables and complex interactions [17]. RF assesses the importance of variables during classification, improving accuracy by emphasizing distinguishing features. This approach is particularly advantageous for rapidly classifying samples across diverse variables and enabling in-depth analysis of phenotype–phylogeny interactions [18].

The classification of barley species remains challenging, due to their morphological diversity and environmental adaptations. This study proposes a novel approach to classifying diverse barley species by integrating hierarchical clustering based on individual similarities, PCA-LDA for effective dimensionality reduction and accuracy improvement, and RF for detailed classification based on multiple characteristic traits. This combined approach helps to effectively classify the Hordeum genus at different taxonomic levels (section and species).

2. Materials and Methods

2.1. Plant Materials and Cultivation Methods

In this study, 215 accessions representing 17 species of the Hordeum genus were cultivated in greenhouse at the National Agrobiodiversity Center (latitude 35°, longitude 127°), National Institute of Agricultural Sciences, Rural Development Administration, Republic of Korea. For phenotypic observations, 25 plants per accession were sown in the experimental plots on 22 November 2023 and all 25 individuals were evaluated. The experiment was conducted using a Completely Randomized Design (CRD), with three replicates. Following germination, the barley plants underwent a winter period for vernalization. Agronomic practices were uniformly applied according to the Standards for Research and Analysis of Agricultural Science and Technology [19].

2.2. Morphological Data Collection

Morphological traits were collected from each accession. A total of 17 morphological traits were collected, including days to flowering (when the spikes emerging from the pedicel make up more than 60% of the total), days to maturity (when the yellowing of the inflorescence exceeds 40% of the total), culm length, spike length, number of triple-spikelets, various spikelet characteristics (e.g., awn length and glume length for lateral and central spikelet), and flag leaf characteristics (area, length, and width). Seed characteristics (100-seed weight, area, length, width, and roundness) data were collected after harvest (3–10 September 2024). The characteristics of seeds (length, width, area, roundness) and flag leaves (length, width, area) were measured using Image J software (version 1.54k).

2.3. Trait-Based Species Classification

The classification of Hordeum species based on the 17 morphological traits was conducted using three different methods: hierarchical clustering, PCA-LDA, and Random Forest. The goal was to evaluate the potential for accurate species classification based on these traits.

2.3.1. Hierarchical Clustering

Hierarchical clustering utilized Euclidean distance to calculate similarities between accessions, with McQuitty’s WPGMA (Weighted Pair Group Method with Arithmetic Mean) employed as the clustering algorithm. The resulting dendrogram visually grouped morphologically similar accessions. The clustering results derived from the dendrogram were compared to the existing species labels, to evaluate the clustering accuracy.

2.3.2. Principal Component Analysis—Linear Discriminant Analysis (PCA-LDA)

PCA was applied to reduce the 17 morphological traits to the top 10 principal components, simplifying the dataset and enhancing classification performance while maximizing variance among individuals. The reduced dataset was then subjected to Linear Discriminant Analysis (LDA) for species classification. LDA optimized group separation by maximizing between-group variance and minimizing within-group variance.

2.3.3. Random Forest

The RF model was trained using morphological-trait data, with each decision tree leveraging various combinations of traits to reliably classify species based on their unique trait patterns.

2.4. Performance Evaluation

The performance of the classification models, PCA-LDA and RF, was assessed by splitting the data into training (70%) and test (30%) sets. Evaluation was conducted based on a confusion matrix, deriving metrics such as accuracy, precision, recall, and F1-score to compare model performance. A 5-fold cross-validation approach was employed to prevent overfitting. Additionally, the mean absolute weight of each trait was calculated to evaluate its relative importance in the classification process, enabling the identification of key traits that contributed most significantly to species differentiation.

2.5. Statistical Analysis

Hierarchical clustering was performed using the R (version 4.2.1) statistical software (https://www.r-project.org/ (accessed on 25 September 2024)), and prior to clustering, all variables were standardized using z-score transformation to account for differences in scale. The distances between observations were calculated using the “hclust” and “dist” functions in the R program, and the dendrogram was visualized using the “cutree” function.

PCA-LDA and RF statistical analyses and modeling were performed using Python (version 3.9, https://www.python.org/ (accessed on 7 October 2024)) and the Scikit-learn library. Data preprocessing and visualization were conducted using Python libraries such as Pandas, NumPy, Matplotlib, and Seaborn. For RF analysis and parameter optimization, Scikitlearn’s GridSearchCV was utilized.

3. Results

3.1. Morphological Trait Variation in the Hordeum Genus

The analysis of morphological traits across different taxonomic levels revealed significant variations within the Hordeum genus. These variations were observed at the subgenus, section, and species levels (Table 1 and Table 2). At the subgenus level, the Hordeum subgenus showed significantly larger values in traits such as flag leaf area, flag leaf length, flag leaf width, and 100-seed weight, compared to the Hordeastrum subgenus. This suggests that Hordeum generally has larger leaves and heavier seeds, which may contribute to increased productivity or resilience. In contrast, the Hordeastrum subgenus shows smaller values in culm length but has a higher number of triple-spikelets, highlighting distinct growth characteristics. Among sections, larger culm length and flag leaf area were found in the Hordeum section compared to Trichostachys and Marina, indicating vigorous growth characteristics. Intermediate trait values were found in the Trichostachys, while smaller culm length, flag leaves, and seed size were measured compared to the Hordeum section. Smaller culm length and flag leaves were observed in the Marina section, indicating compact growth habit.

Figure 1 provides a visual representation of these agronomic traits across the five sections: Hordeum, Stenostachys, Trichostachys, Marina, and Nodosa. Significantly larger flag leaf area, length, and width were found in the Hordeum compared to the other sections (**** p < 0.0001). In contrast, the smallest leaf dimensions were measured in the Marina section, while minimal differences in leaf size were observed between Nodosa and other sections. The number of samples per sections varies, which may affect results, particularly in the Nodosa section, represented by only one accession.

Noticeable differences were observed in seed traits. Significantly larger seed area, length, and width, along with a higher 100-seed weight, were measured in the Hordeum compared to the other sections (**** p < 0.0001). Higher seed roundness was found in the Hordeum section, while no significant differences in seed roundness were detected among the other sections. The smallest seed dimensions and the lightest seeds were found in Stenostachys section. Other structural traits further differentiated these sections. For instance, Trichostachys exhibited the longest central and lateral-spikelet glumes, which were significantly longer than those of Hordeum and Marina (**** p < 0.0001), while Trichostachys had relatively long glumes, though not as pronounced as Stenostachys. Culm length was another distinguishing trait: Hordeum had significantly longer culms compared to Stenostachys, Trichostachys, and Marina (**** p < 0.0001), while Marina showed the shortest culms with statistical significance (**** p < 0.0001).

Additional structural traits, such as awn length and spike morphology, were used to further highlight differences among the sections. The Hordeum section had significantly longer awns on both central and lateral-spikelet compared to other sections (**** p < 0.0001), while Stenostachys and Marina exhibit significantly shorter awns (**** p < 0.0001). Significant variation in spike length was also observed. The shortest spikes were consistently exhibited by Marina (**** p < 0.0001), whereas significantly higher number of triple-spikelets was found in the Stenostachys compared to the other sections (**** p < 0.0001). The Hordeum and Trichostachys sections flowered significantly earlier than other sections, particularly Stenostachys (**** p < 0.0001). While minimal differences were generally observed in days to maturity, species within Stenostachys, such as Hordeum. bogdanii and Hordeum lechleri, have very short maturing periods, whereas Hordeum cordobense and Hordeum stenostachys exhibited the longest durations to maturity. In summary, Table 1 and Table 2 provide a comprehensive overview of these statistical differences, and Figure 1 illustrates the distribution of these traits across sections, underscoring the distinct agronomic profiles of each group. These findings offer valuable insights into the diversity, growth patterns, and adaptive strategies within these taxonomic levels.

3.2. Classification Using Hierarchical Clustering Analysis

Hierarchical clustering analysis using Euclidean distance and the McQuitty (WPGMA) algorithm demonstrated clear clustering of species within the Hordeum and Hordeastrum subgenus, which were divided into two main classes and nine subclasses (Figure 2). Class2 consisted entirely of accessions from the Hordeum subgenus, achieving an accuracy of 100%. achieving an accuracy of 100%. In contrast, Class1 was primarily composed of Hordeastrum subgenus accessions (60.5%), but also included some Hordeum subgenus accessions (39.5%), indicating a misclassification rate, due to these Hordeum subgenus accessions being grouped within Hordeastrum. The overall classification accuracy at the subgenus level was 86.0% (Table 3).

At the section level (Table 4), Subclasses5 through 9 were entirely composed of Hordeum section accessions, achieving a classification accuracy of 100% across all subclasses and demonstrating strong clustering consistency. Subclass5 predominantly consisted of Hordeum spontaneum, known as the wild relative of H. vulgare, indicating clear phenotypic distinctions between wild and cultivated forms. In Subclass2, the Marina section accounted for 73.3% of the accessions, although some accessions from the Stenostachys and Nodosa sections are included, resulting in a misclassification rate of 3.7%. Subclass3 consisted solely of Trichostachys section accessions, which should ideally grouped within Class2 under the Hordeum subgenus. However, due to similarities to Hordeastrum in characteristics such as culm length, seed traits, and flag leaves, it appears to have been misclassified. The overall classification accuracy at the section level was 83.1%, indicating that morphological traits align well with sectional groupings.

3.3. Classification Using PCA-LDA Analysis

To enhance the effectiveness of the LDA analysis, a preliminary PCA was conducted, to optimize variable selection. The PCA results indicated that PC1 and PC2 together accounted for 65.9% of the total variance, with PC1 alone contributing 52.1%. In order to maximize classification accuracy, the top 10 traits with the highest contributions to PC1—including seed width, seed area, and 100-seed weight—were selected as the most influential variables (Table 5). These selected traits were then used in the subsequent LDA analysis to refine the classification of Hordeum accessions based on their morphological and agronomic characteristics. This process allowed for a more streamlined and focused LDA, improving the interpretability and accuracy of the clustering results, as confirmed by Figure 3.

The confusion matrix for section classification (Figure 4A) demonstrates a high classification accuracy among the different sections. The Hordeum section had all 44 samples correctly classified, achieving perfect accuracy. The Stenostachys section showed eight correct classifications and one misclassification, while the Trichostachys section had five correct classifications and one misclassification. The Marina section had three out of six samples correctly classified, resulting in three misclassifications. The Nodosa section was not included in the dataset, due to limited data availability. Overall, the classification accuracy at the section level indicates that the model effectively distinguishes sections based on their morphological characteristics.

The confusion matrix for species classification (Figure 4B) shows varying levels of accuracy across different species. Specifically, H. vulgare achieved perfect accuracy, with all 34 samples correctly classified, and H. spontaneum also achieved high accuracy with 10 correct classifications. In contrast, Hordeum murinum had five correct classifications with one misclassification, while Hordeum procerum showed two misclassifications. Additionally, Hordeum stenostachys and H. marinum each had one misclassification. These results highlight the need to improve differentiation among species with similar traits. Resources such as Hordeum bogdanii and Hordeum brachyantherum were not classified, due to limited sample sizes.

The analysis of LDA feature weights for section classification (Figure 5A) revealed the most influential morphological traits distinguishing the sections within the Hordeum genus. Seed width was the most significant trait, followed closely by flag-leaf width and seed length. Seed area was also found to play a crucial role in differentiating the sections, underscoring the importance of seed-related characteristics in morphological classification. Other important traits, including central-spikelet awn length and flag-leaf length, were identified as contributing positively but to a lesser extent. The number of triple-spikelet, lateral-spikelet awn length, and 100-seed weight were found to have lower weights, indicating their roles in distinguishing sections were comparatively minor. Flag-leaf area had the least impact among the selected features, reflecting its limited contribution to section differentiation.

In the context of species classification (Figure 5B), the LDA feature weights revealed similar trends, with seed width again being identified the most significant trait, followed by seed length and flag-leaf width. The seed area was also identified as a key feature, strengthening its relevance in distinguishing between species. Flag-leaf length and central-spikelet awn length were influential as well, contributing to the classification, but not to the same extent as the primary traits. Lateral-spikelet awn length, number of triple-spikelets, 100-seed weight, and flag-leaf area showed lower weights, indicating their lesser importance in species-level differentiation. These findings highlight the pivotal role of seed-related traits in both section and species classifications within the Hordeum genus. The consistent importance of these characteristics across different taxonomic levels suggests that they are fundamental to the morphological differentiation of Hordeum accessions.

3.4. Classification Using RF Analysis

The confusion matrix for section classification (Figure 6A) using the RF algorithm demonstrated a high level of accuracy across all section labels. Hordeum section was classified with perfect accuracy, showing 44 correctly predicted instances with no misclassifications. The Marina, Stenostachys, and Trichostachys sections also achieved perfect accuracy, each with all true instances correctly classified (six for Marina, nine for Stenostachys, and six for Trichostachys). Nodosa, however, did not appear in the dataset, as shown by the absence of any predicted or actual labels for this section. This matrix suggests that the model achieved 100% classification accuracy at the section level, highlighting the distinctive morphological and agronomic traits that separate each section in the dataset.

The confusion matrix for species classification is presented in Figure 6B, with variation in classification accuracy among species. H. vulgare was classified accurately with 34 correct predictions, and H. spontaneum followed closely with 10 correct predictions, reflecting strong clustering of these species, based on their distinct characteristics. Other species like Hordeum. marinum, H. murinum, H. lechleri, and H. cordobense also achieved high classification accuracy, with six, six, three, and two correct predictions, respectively. However, some species used in this study, such as H. bogdanii, H. brachyantherum, and Hordeum capense, have a limited number of samples, which prevented them from being used as test data, despite being included in the training dataset, resulting in no correctly classified samples. Nevertheless, this confusion matrix effectively illustrated species-level differentiation for those species with distinct characteristics.

The decision tree for section classification (Figure 7A) provides a clear structure for distinguishing between sections. The first split was based on 100-seed weight (threshold: 2.12), seed area (threshold: 6.7), and seed length (threshold: 6.4), effectively separating Hordeum and Trichostachys from other sections. Subsequent splits were focused on spike length and lateral-spikelet glume length, to further differentiate the sections. This decision tree highlighted the fact hat the Hordeum section was primarily characterized by seed traits that distinguished it from other sections. Among other sections, lateral-spikelet glume traits, awn characteristics, and spike length played key roles in their classification. The decision tree for species classification using the RF model (Figure 7B) emphasized the key decision points leading to the classification of Hordeum species. The first split was based on 100-seed weight, with a threshold of 2.12 effectively separating H. vulgare from other species such as H. murinum, H. marinum, and H. cordobense. The species H. marinum, H. murinum, H. lechleri, and H. procerum were classified based on lateral-spikelet glume length (threshold: 2.38). To further distinguish H. marinum, traits such as spike length, days to flowering, and central-spikelet awn length were utilized. This clarity in branching underscores the critical role of seed-related traits in ensuring accurate species classification.

On the other hand, H. marinum was identified as a separate class, based on traits like flag-leaf length and central-spikelet awn length. Although the tree depth was adjusted for visualization due to overlapping classifications among multiple species, an overall analysis of feature weights confirmed that lateral-spikelet awn, central-spikelet awn, and seed traits were the most critical features for distinguishing species. The RF feature importance analysis for section classification (Figure 8A) highlights the key morphological traits contributing to the differentiation of sections within the Hordeum genus. Seed area emerged as the most significant feature, displaying the highest importance value, followed by seed width and 100-seed weight, which also played crucial roles in section differentiation.

Other notable features included central-spikelet awn length and culm length, though their relative importance was lower. Seed length, lateral-spikelet awn length, and spike length showed moderate importance, indicating their consistent contributions to section classification. In contrast, traits such as days to flowering, flag-leaf length, seed roundness, and days to maturity exhibited low importance values, reflecting their relatively minor roles in distinguishing sections. In the analysis for species classification (Figure 8B), lateral-spikelet awn length was identified as the most significant trait, surpassing all other features in importance. This was followed by seed width and seed area, which were also crucial for distinguishing among the Hordeum species. Central-spikelet awn length and spike length were important, reinforcing their relevance in species classification. Other traits, such as seed length, flag-leaf area, and 100-seed weight, showed moderate importance, contributing to species differentiation to a lesser extent. In contrast, traits like flag-leaf width, flag-leaf length, days to flowering, and days to maturity exhibited very low importance, reflecting their secondary roles in species classification.

The classification performance metrics for the PCA-LDA and RF analyses are presented in Table 6. The PCA-LDA method had an accuracy of 92.3% for section classification and 90.8% for species classification. The precision and recall metrics were commendable, with 92.0% and 92.3% for section classification and 94.3% and 90.8% for species classification, respectively. The F1-score for section classification was 92.1%, while it was slightly higher, at 91.6%, for species classification. In contrast, the RF model had superior performance across all metrics, achieving a perfect classification accuracy of 100% for both section and species classifications. This included a perfect precision and recall score of 100% and an F1-score of 100%, indicating the model’s effectiveness in distinguishing between the various sections and species within the Hordeum genus.

Table 7 illustrates the results of the 5-fold cross-validation conducted for both PCA-LDA and RF. The mean accuracy for PCA-LDA was 91.6% for section classification and 89.3% for species classification, with standard deviations of 3.15% and 3.8%, respectively. In contrast, RF showed consistent accuracy, with values of 95.3% for section classification and 93.0% for species classification. Overall, the results indicate that RF significantly outperforms PCA-LDA in both section and species classifications, suggesting that this model may be more suitable for accurately classifying Hordeum accessions based on their morphological traits. The cross-validation results further confirmed the robustness of the RF model in providing reliable classification outcomes.

At each decision point (node), the trait threshold represents the boundary value used to split the data based on a specific trait. For example, at the first decision point in (A), if the 100-seed weight is ≤2.21, the data flow to the left branch; otherwise, they flow to the right branch. The Gini coefficient indicates the impurity at each node. A value close to 0 signifies that the data are more evenly split, while a higher value implies that the data contain a mix of different classes. Samples represent the number of data points (samples) at each node. Value indicates the number of samples belonging to each class (section) at a given node. For instance, in (A), at the first decision point, value = [1, 18, 23, 31, 142] corresponds to the following sections: Hordeum, Marina, Nodosa, Stenostachys, and Trichostachys, with 142 samples in Hordeum, 18 in Marina, 1 in Nodosa, 23 in Stenostachys, and 31 in Trichostachys. In (B), the value reflects the number of samples for each species in the following order: H. bogdanii, H. brachyantherum, H. californicum, H. capense, H. cordobense, H. fuegianum, H. hybr., H. intercedens, H. jubatum, H. lechleri, H. marinum, H. murinum, H. procerum, H. roshevitzii, H. spontaneum, H. stenostachys, and H. vulgare. Class represents the class with the highest probability of being predicted at that node.

4. Discussion

The materials used in this study were provided by the National Agrobiodiversity Center (NAS) and utilized as research resources. However, some resources could not be used, due to limited seed quantities or unsuccessful germination in the field, resulting in certain species being analyzed with only one or a very small number of samples. Therefore, 25 individuals were planted per accession to observe multiple traits and reduce phenotypic errors, while cultivating them in the same field to minimize environmental variability. Additionally, phenotypic traits across 17 species were investigated to enhance classification accuracy.

From an evolutionary perspective, plant phenotypes such as seed traits, leaf characteristics, glume, awn, plant height, and flowering time adapt to environmental factors like climate, soil, and available resources [20,21,22,23,24,25,26,27]. Based on this, it can be inferred that plants belonging to the five sections of the Hordeum genus (Hordeum, Marina, Trichostachys, Stenostachys, and Nodosa) have developed unique phenotypic and physiological traits in response to the environmental conditions of their habitats. These traits are interpreted as the result of adaptation to environmental stressors and natural selection within each section. The Hordeum section exhibited prominent characteristics such as greater culm length and seed traits (length, width, area, and weight). In contrast, the Marina section displayed the smallest culm length and flag-leaf traits (length, width, and area). This is attributed to its distribution in coastal regions of Europe and Asia, where the plants are exposed to saline soils and strong winds. Such adaptations are interpreted as survival strategies to reduce transpiration and minimize water loss under saline and arid conditions. The Trichostachys section exhibited the longest glumes in both the central and lateral-spikelet. In wheat, glumes are known to perform various functions, including floret and seed formation and protection, water regulation, and even assisting photosynthesis, in some species [23]. These traits are believed to enhance water retention capabilities, enabling plants to adapt to dry savannas and temperate-to-subarctic regions. The Stenostachys section displayed distinct characteristics, such as longer spike lengths and a higher number of triple-spikelets. In the Nodosa section, represented by H. capense, data analysis was not possible, due to a single sample; however, its morphological characteristics showed trends similar to those of the Marina section.

The phylogenetic analysis of Hordeum species based on phenotypic traits revealed a tree that aligns remarkably well with previously published classifications derived from DNA-based studies. This congruence between phenotypic clustering and molecular phylogenies strengthens the reliability of both approaches in uncovering evolutionary relationships and origins within the genus. The strong clustering of H. procerum, H. cordobense, and H. stenostachys within Subclass 1 reflects findings from prior DNA-based studies that utilized markers such as 5S rDNA and nuclear HTL genes [28,29]. Both phenotypic and molecular data suggest a shared evolutionary trajectory for these species, shaped by interspecific hybridization and introgression events, as previously reported in studies on polyploid Hordeum species [29,30].

The dominance of H. marinum subspecies in Subclass 2, along with the inclusion of H. roshevitzii, Hordeum intercedens, H. brachyantherum subsp. californicum, and H. capense, aligns with DNA-based phylogenies that identified similar groupings [30,31]. Prior studies demonstrated that H. brachyantherum subsp. californicum and H. roshevitzii played key roles as progenitors of tetraploid and hexaploid species in this group. The alignment of phenotypic and DNA-based data underscores the robustness of these findings and highlights the influence of hybridization events in New World species [32,33].

The exclusive clustering of H. murinum subspecies in Subclass 3 based on phenotypic traits is consistent with their distinct position in DNA-based phylogenetic studies. These studies have consistently demonstrated that the Xu genome in H. murinum forms a unique clade, separate from other Hordeum genomes [28,31]. The coherence of this group in both molecular and phenotypic analyses emphasizes its genetic and ecological divergence.

The phenotypic grouping of H. jubatum and H. lechleri in Subclass 4 is in agreement with molecular studies that identified H. brachyantherum subsp. californicum and H. roshevitzii as their progenitors [30,31]. The shared evolutionary history suggested by both approaches highlights the significant role of intercontinental dispersal and adaptation in shaping the diversity of these species [34]. The clustering of H. vulgare subsp. spontaneum in Subclass 5 aligns with its established role as the wild progenitor of cultivated barley, as revealed in DNA-based analyses. The monophyly observed in both phenotypic and molecular phylogenies reflects the genetic-bottleneck effects associated with barley domestication [31,33].

The consistency between phenotypic and molecular phylogenies reinforces the use of integrative approaches in understanding Hordeum evolution. While phenotypic analyses are valuable, combining them with genome-wide molecular studies will enhance resolution and address remaining uncertainties. Additionally, incorporating ecological and biogeographical data will provide a more comprehensive understanding of the evolutionary dynamics within this economically and scientifically important genus.

Overall, this study demonstrates the effectiveness of using morphological traits for classification within the Hordeum genus. However, classification proved challenging for some species or taxonomic categories, due to limited resources. Despite these limitations, high classification accuracy was observed in several sections and species, with the RF model achieving the highest accuracy. This suggests the potential for breeding strategies leveraging morphological differences and further classification research. As data continue to accumulate over time, they may not only be utilized more effectively for classification, but also applied to other machine learning models to identify optimal classification methods.

5. Conclusions

The 17 phylogenetic traits of barley examined in this study exhibit considerable variation. In particular, the Hordeum subgenus shows more prominent characteristics in seed and flag-leaf size, compared to the Hordeastrum subgenus. At the section level, Hordeum has larger seeds and flag-leaves, Trichostachys is distinguished by its prominent lateral- spikelet glumes, Marina has smaller leaf sizes, and Stenostachys is characterized by its spike length and the number of triple-spikelets. These differences suggest the potential for selecting traits suited to specific agricultural or environmental conditions, providing a favorable approach for analyzing the phenotypic and phylogenetic relationships we aimed to investigate. The results of the classification performance of our three methods (hierarchical clustering, PCA-LDA, and RF) emphasize clear morphological and agronomic differences at the subgenus, section, and species levels. These findings provide valuable insights for targeted breeding strategies aimed at particular environmental or agricultural requirements.

Author Contributions

Conceptualization, N.R. and P.S.; methodology, M.H., H.-C.K., H.Y. and D.-S.Y.; data curation, N.R. and G.-T.C.; data analysis, P.S., H.-J.W. and M.H.; writing—original draft preparation, P.S.; writing—review and editing, N.-J.C. and N.R. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Relevant data are incorporated within the manuscript. Additional data can be obtained upon direct request from the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. Comparison of 17 agronomic traits among different sections: Hordeum, Stenostachys, Trichostachys, Marina, and Nodosa. Box plots represent the distribution of 17 agronomic traits for each section. Statistical significance was determined using Wilcoxon rank sum tests for pairwise comparisons between sections. Significance levels are indicated by asterisks: **** p [less than] 0.0001, *** p [less than] 0.001, ** p [less than] 0.01, * p [less than] 0.05, and “ns” denotes no significant difference. The results reveal significant differences in days to flowering across several sections, with Hordeum and Stenostachys showing particularly notable differences from other sections. (A) Days to flowering, (B) Days to maturity, (C) Culm length (cm), (D) Number of triple-spikelets, (E) Central-spikelet awn length (cm), (F) Central-spikelet glume length (cm), (G) Lateral-spikelet awn length (cm), (H) Lateral-spikelet glume length (cm), (I) Spike length (cm), (J) Flag-leaves area (cm2), (K) Flag-leaves length (cm), (L) Flag-leaves width (cm), (M) 100-seed weight (g), (N) Seed area (mm2), (O) Seed length (mm), (P) Seed width (mm), (Q) Seed roundness.

Figure 2. Dendrogram of the hierarchical clustering of species based on Euclidean distance and McQuitty’s (WPGMA) clustering method. The analysis grouped the species into two main classes (Class1, Class2), further divided into nine subclasses (Subclass 1–9), as indicated within the dendrogram. Each branch represents a unique taxonomic group based on the morphological traits analyzed, with clustering patterns reflecting similarity levels among the species. Class1 primarily consists of 60.5% accessions from the Hordeastrum subgenus, while Class2 is entirely composed of accessions from the Hordeum subgenus. The five sections (Hordeum, Stenostachys, Trichostachys, Marina, and Nodosa) are color-coded for clear visualization of sectional affiliations.

Figure 3. Linear Discriminant Analysis (LDA) classification results based on the top 10 traits selected from PCA (A,B). (A) The left scatter plot shows the classification of Hordeum sections using two LDA components, illustrating clear separation among the five sections: Hordeum, Stenostachys, Trichostachys, Marina, and Nodosa. (B) The right scatter plot presents species-level classification, where individual Hordeum species are distinctly separated based on the same two LDA components. These visualizations demonstrate the effectiveness of the selected traits in distinguishing both sections and species within the Hordeum genus, with high clustering consistency among groups.

Figure 4. Confusion matrices for LDA classification of Hordeum sections and species based on PCA-selected traits (A,B). The analysis was conducted using 70% of the data as training data and 30% as testing data. (A) The left matrix shows the section-level classification results, where most sections (Hordeum, Marina, Nodosa, Stenostachys, and Trichostachys) exhibit high accuracy, with only minor misclassifications observed among similar sections. (B) The right matrix illustrates the species-level classification, with distinct separation among Hordeum species, though some species show slight misclassification, due to close morphological similarities. These confusion matrices provide a visual representation of the classification performance, highlighting the effectiveness of PCA-LDA in distinguishing sections and species within the Hordeum genus.

Figure 5. Trait importance for section and species classification using PCA-LDA (A,B). (A) The feature importance for section classification, (B) the feature importance for species classification. Traits include days to flowering, days to maturity, culm length, spike length, number of triple-spikelets, lateral-spikelet awn length, central-spikelet awn length, flag-leaf area, flag-leaf length, flag-leaf width, 100-seed weight, seed area, seed length, and seed width.

Figure 6. Confusion matrices for section and species classification using RF analysis based on 17 traits (A,B). The analysis was conducted with 70% of the data as training data and 30% as testing data. (A) The left confusion matrix shows the section-level classification results, demonstrating high accuracy in distinguishing sections such as Hordeum, Marina, Nodosa, Stenostachys, and Trichostachys. (B) The right confusion matrix presents species-level classification results, where most Hordeum species were accurately classified, although slight misclassifications occurred for some species. These confusion matrices visually illustrate the effectiveness of RF in classifying both sections and species within the Hordeum genus, based on the selected 17 traits.

Figure 7. Visualization of decision tree from random forest model (A,B). This figure shows a simplified visualization of a single decision tree derived from the RF model, with a limited depth. Each node represents a feature-based data split, and includes the Gini impurity value, along with the number of samples classified into each class. From top to bottom, the figure illustrates the progressive division of data into increasingly specific classes. (A) represents the Section classification results, distinguishing the data at a higher hierarchical level to identify different sections. (B) shows the Species classification results, focusing on finer distinctions to classify species within each section.

Figure 7. Visualization of decision tree from random forest model (A,B). This figure shows a simplified visualization of a single decision tree derived from the RF model, with a limited depth. Each node represents a feature-based data split, and includes the Gini impurity value, along with the number of samples classified into each class. From top to bottom, the figure illustrates the progressive division of data into increasingly specific classes. (A) represents the Section classification results, distinguishing the data at a higher hierarchical level to identify different sections. (B) shows the Species classification results, focusing on finer distinctions to classify species within each section.

Figure 8. Trait importance for section and species classification using RF analysis based on 17 traits (A,B). (A) The feature importance for section classification, where seed length, culm length, seed area, and spike length are among the traits with the highest importance in distinguishing Hordeum sections. (B) The feature importance for species classification, with lateral-spikelet awn length, seed width, and seed area showing the highest contributions in differentiating Hordeum species. These graphs highlight the key traits that play a significant role in classifying both sections and species within the Hordeum genus, demonstrating the effectiveness of RF in identifying essential morphological traits for accurate classification.

Table 1

Agricultural trait characteristics of the Hordeum genus, presented by subgenus, section, and species.

Categories	Subcategories	No. of Accession	DF (Day)	DM (Day)	CL (cm)	SpL (cm)	NTS	LSAL (cm)	LSGL (cm)	CSAL (cm)	CSGL (cm)
Subgenus	Hordeum	169	144 ± 8	38 ± 5	94.8 ± 19.5	9.9 ± 2.2	30.1 ± 6.2	9.3 ± 5.6	1.7 ± 0.8	12.2 ± 4.5	1.6 ± 0.8
	Hordeastrum	46	169 ± 23	36 ± 11	57.6 ± 25.2	8.6 ± 3.4	53.9 ± 14.7	0.6 ± 0.4	2.0 ± 1.0	1.5 ± 0.9	2.0 ± 1.2
Section	Hordeum	139	145 ± 8	38 ± 5	101.8 ± 13.1	9.9 ± 2.3	28.0 ± 4.3	10.3 ± 5.7	1.4 ± 0.4	13.9 ± 3.0	1.3 ± 0.4
	Trichostachys	30	141 ± 7	39 ± 3	62.5 ± 8.7	9.9 ± 1.3	39.5 ± 5.3	4.5 ± 0.9	3.1 ± 0.6	4.6 ± 0.8	2.9 ± 0.5
	Marina	22	157 ± 13	34 ± 6	40.0 ± 7.3	6.0 ± 0.7	46.8 ± 5.2	0.7 ± 0.3	2.0 ± 0.3	1.5 ± 0.4	2.1 ± 0.3
	Nodosa	1	160	37	93	7.8	42.3	0.4	1.9	1.3	1.4
	Stenostachys	23	181 ± 26	37 ± 14	72.9 ± 25.3	11.0 ± 3.2	61.3 ± 17.4	0.5 ± 0.6	2.0 ± 1.3	1.6 ± 1.3	2.0 ± 1.6
Species	H. vulgare subsp. spontaneum	27	137 ± 5	38 ± 4	109.0 ± 9.8	11.8 ± 1.3	30.1 ± 2.6	1.3 ± 4.3	1.8 ± 0.4	14.8 ± 2.0	1.7 ± 0.5
	H. vulgare subsp. vulgare	110	147 ± 8	37 ± 6	100.1 ± 13.3	9.4 ± 2.3	27.6 ± 4.5	12.5 ± 3.4	1.3 ± 0.4	13.6 ± 3.1	1.2 ± 0.3
	H. vulgare var. hexastichon	1	139	35	93.2	9.6	22.3	11.7	0.8	13.8	0.8
	H. hybr.	1	139	41	95.6	11.5	29.7	10.2	1.4	13	1.3
	H. murinum spp.	15	142 ± 9	38 ± 3	65.4 ± 8.6	10.0 ± 1.1	40.6 ± 4.2	4.2 ± 1.0	2.9 ± 0.6	4.3 ± 0.9	2.9 ± 0.5
	H. murinum subsp. glaucum	4	138 ± 8	35 ± 2	66.2 ± 5.8	8.6 ± 0.7	44.0 ± 7.4	4.4 ± 0.6	3.1 ± 0.3	4.6 ± 0.3	2.6 ± 0.1
	H. murinum subsp. leporinum	11	142 ± 5	40 ± 2	57.2 ± 7.5	10.3 ± 1.5	36.4 ± 4.3	4.9 ± 0.6	3.5 ± 0.5	5.1 ± 0.6	3.1 ± 0.3
	H. marinum spp.	11	153 ± 14	35 ± 8	41.0 ± 6.3	6.1 ± 0.7	46.0 ± 4.7	0.7 ± 0.2	2.0 ± 0.2	1.6 ± 0.3	2.0 ± 0.2
	H. marinum subsp. gussoneanum	10	162 ± 11	33 ± 4	37.0 ± 5.7	5.9 ± 0.6	47.8 ± 6.1	0.7 ± 0.3	2.0 ± 0.3	1.5 ± 0.5	2.0 ± 0.3
	H. marinum subsp. marinum	1	151	37	58.4	6.7	45	1.1	2.5	0.8	2.5
	H. capense	1	160	37	93	7.8	42.3	0.4	1.9	1.3	1.4
	H. bogdanii	1	211	15	61.7	9.3	52.3	0.7	0.9	1	0.7
	H. roshevitzii	1	174	32	66.8	8.4	41	0.3	1.5	1	1.5
	H. californicum	1	148	41	47.9	6.4	47.3	0.1	1.4	0.6	1.3
	H. cordobense	4	158 ± 13	57 ± 8	76.7 ± 24.2	10.2 ± 0.8	60.1 ± 14.7	0.1 ± 0.1	0.5 ± 0.1	0.5 ± 0.1	0.5 ± 0.0
	H. intercedens	1	144	40	51.9	7.7	45.7	0.3	2	1.2	1.8
	H. stenostachys	2	164 ± 15	49 ± 6	94.0 ± 14.9	11.1 ± 0.1	59.9 ± 2.6	0.1 ± 0.0	1.1 ± 0.0	0.8 ± 0.0	0.5 ± 0.0
	H. fuegianum	1	225	31	48.2	9.5	55.7	0.1	0.9	0.6	0.7
	H. jubatum	1	183	36	88.5	14.3	79.7	0.8	4.6	5.3	5.6
	H. lechleri	4	215 ± 2	17 ± 1	57.2 ± 4.0	14.0 ± 2.9	77.8 ± 15.6	1.5 ± 0.4	4.0 ± 0.7	3.1 ± 0.3	4.5 ± 0.6
	H. procerum	5	174 ± 9	37 ± 3	99.8 ± 25.6	13.3 ± 2.2	70.8 ± 11.6	0.3 ± 0.2	2.2 ± 0.4	1.8 ± 0.3	2.2 ± 0.4
	H. brachyantherum subsp. brachyantherum	2	195 ± 13	37 ± 9	44.7 ± 14.4	6.2 ± 2.3	31.5 ± 11.6	0.1 ± 0.0	1.3 ± 0.1	0.6 ± 0.0	1.1 ± 0.1

Note: This table provides detailed trait information for different hierarchical classifications within the Hordeum genus, including Hordeum and Hordeastrum subgenera, as well as various sections and species within these groups. Traits include days to flowering(DF), days to maturity (DM), culm length (CL), spike length (SpL), number of triple-spikelets (NTSs), lateral-spikelet awn length (LSAL), lateral-spikelet glume length (LSGL), central-spikelet awn length (CSAL), and central-spikelet glume length (CSGL). Each trait is presented with its average value and the variability within the population, indicated by the standard deviation. This table serves as a comprehensive summary of the morphological and agronomic diversity observed in the Hordeum genus across different taxonomic levels.

Table 2

Agricultural trait characteristics of the Hordeum genus, presented by subgenus, section, and species (continued from Table 1).

Categories	Subcategories	No. of Accession	FLA (cm2)	FLL (cm)	FLW (cm)	100SW (g)	SA (mm2)	SL (mm)	SW (mm)	SR
Subgenus	Hordeum	169	144.8 ± 75.8	15.1 ± 4.1	1.13 ± 0.4	4.2 ± 1.8	22.7 ± 7.0	9.0 ± 1.7	3.1 ± 0.6	0.3 ± 0.1
	Hordeastrum	46	34.0 ± 23.0	8.5 ± 3.9	0.48 ± 0.2	0.4 ± 0.2	4.6 ± 1.2	4.3 ± 0.7	1.3 ± 0.2	0.3 ± 0.0
Section	Hordeum	139	164.6 ± 68.3	16.2 ± 3.4	1.23 ± 0.3	4.8 ± 1.0	25.5 ± 3.5	9.6 ± 1.1	3.4 ± 0.3	0.4 ± 0.1
	Trichostachys	30	53.1 ± 23.1	10.0 ± 2.9	0.62 ± 0.2	1.5 ± 2.0	9.4 ± 2.0	6.1 ± 0.7	1.9 ± 0.2	0.3 ± 0.0
	Marina	22	26.8 ± 8.4	6.6 ± 1.4	0.5 ± 0.1	0.5 ± 0.1	5.1 ± 1.0	4.5 ± 0.6	1.4 ± 0.1	0.3 ± 0.0
	Nodosa	1	18.9	5.6	0.43	0.3	4.7	4.4	1.4	0.3
	Stenostachys	23	41.6 ± 29.9	10.5 ± 4.5	0.46 ± 0.2	0.3 ± 0.1	4.1 ± 1.3	4.1 ± 0.7	1.2 ± 0.2	0.3 ± 0.0
Species	H. vulgare subsp. spontaneum	27	113.7 ± 49.2	13.6 ± 3.0	1.02 ± 0.3	3.9 ± 0.5	23.2 ± 2.0	10.1 ± 0.6	2.9 ± 0.2	0.3 ± 0.0
	H. vulgare subsp. vulgare	110	178.9 ± 65.3	17.0 ± 3.0	1.3 ± 0.3	5.0 ± 1.0	26.1 ± 3.6	9.5 ± 1.1	3.5 ± 0.3	0.4 ± 0.1
	H. vulgare var. hexastichon	1	116.8	14.3	1.03	5.1	27.5	10.4	3.4	0.3
	H. hybr.	1	17.9	4.8	0.43	4.3	24.2	10.1	3.1	0.3
	H. murinum spp.	15	57.9 ± 27.6	10.2 ± 3.4	0.66 ± 0.1	0.8 ± 0.3	9.2 ± 2.3	6.1 ± 0.7	1.9 ± 0.2	0.3 ± 0.0
	H. murinum subsp. glaucum	4	39.2 ± 9.5	8.1 ± 2.5	0.57 ± 0.0	0.6 ± 0.1	7.3 ± 0.8	5.4 ± 0.3	1.7 ± 0.1	0.3 ± 0.0
	H. murinum subsp. leporinum	11	51.5 ± 18.4	10.3 ± 2.4	0.6 ± 0.1	2.7 ± 3.0	10.4 ± 1.3	6.5 ± 0.5	2.0 ± 0.1	0.3 ± 0.0
	H. marinum spp.	11	27.1 ± 7.2	6.9 ± 1.2	0.49 ± 0.1	0.5 ± 0.1	5.2 ± 0.9	4.5 ± 0.6	1.4 ± 0.1	0.3 ± 0.0
	H. marinum subsp. gussoneanum	10	26.1 ± 10.2	6.2 ± 1.6	0.5 ± 0.0	0.5 ± 0.2	5.0 ± 1.2	4.5 ± 0.6	1.4 ± 0.1	0.3 ± 0.0
	H. marinum subsp. marinum	1	30.1	6.6	0.56	0.4	4.8	4.1	1.5	0.4
	H. capense	1	18.9	5.6	0.43	0.3	4.7	4.4	1.4	0.3
	H. bogdanii	1	11.2	4.4	0.31	0.1	2.8	3.8	1	0.3
	H. roshevitzii	1	55.4	9.2	0.76	0.4	4.7	3.8	1.6	0.4
	H. californicum	1	7.8	3.9	0.24	0.3	4.1	4	1.3	0.3
	H. cordobense	4	33.1 ± 10.1	10.6 ± 1.4	0.39 ± 0.1	0.2 ± 0.1	3.1 ± 0.4	3.4 ± 0.2	1.1 ± 0.1	0.3 ± 0.0
	H. intercedens	1	9.4	4.5	0.27	0.3	4.2	3.9	1.4	0.4
	H. stenostachys	2	21.1 ± 2.2	9.9 ± 1.6	0.27 ± 0.0	0.4 ± 0.0	5.2 ± 0.1	4.2 ± 0.2	1.5 ± 0.1	0.4 ± 0.0
	H. fuegianum	1	16.3	7.6	0.27	0.2	2.6	3.9	0.8	0.2
	H. jubatum	1	54.3	12.2	0.55	0.2	4.8	4.5	1.3	0.3
	H. lechleri	4	49.6 ± 23.9	12.1 ± 4.3	0.5 ± 0.1	0.2 ± 0.1	2.6 ± 0.1	3.4 ± 0.1	1.0 ± 0.0	0.3 ± 0.0
	H. procerum	5	82.3 ± 20.8	16.1 ± 1.9	0.7 ± 0.2	0.3 ± 0.1	5.8 ± 0.3	5.1 ± 0.2	1.5 ± 0.0	0.3 ± 0.0
	H. brachyantherum subsp. brachyantherum	2	8.6 ± 1.7	4.3 ± 0.5	0.24 ± 0.0	0.3 ± 0.0	4.0 ± 0.1	4.0 ± 0.2	1.3 ± 0.1	0.3 ± 0.0

Note: This table provides detailed trait information for different hierarchical classifications within the Hordeum genus, including Hordeum and Hordeastrum subgenera, as well as various sections and species within these groups. Traits include flag-leaf area (FLA), flag-leaf length (FLL), flag-leaf width (FLW), 100-seed weight (100 SW), seed area (SA), seed length (SL), seed width (SW), and seed roundness (SR). Each trait is presented with its average value and the variability within the population, indicated by the standard deviation. This table serves as a comprehensive summary of the morphological and agronomic diversity observed in the Hordeum genus across different taxonomic levels.

Table 3

Classification accuracy and misclassification details at the subgenus level based on hierarchical clustering analysis. The accessions were divided into two main classes.

Class	Composition(Subgenus)	No. of Accession (%)	Misclassification (%)	Accuracy (%)
Class1	Hordeum	30 (39.5)	30 (14.0)	185 (86.0)
	Hordeastrum	46 (60.5)
Class2	Hordeum	139 (100)

Table 4

Classification accuracy and misclassification details at the section level based on hierarchical clustering analysis. The clustering divides the accessions into nine subclasses.

Class	Composition(Section)	No. of Accession (%)	Misclassification (%)	Accuracy (%)
Subclass1	Stenostachys	11 (100)	0 (0)	177 (82.3)
Subclass2	Marina	22 (73.3)	8 (3.7)
	Stenostachys	7 (23.3)
	Nodosa	1 (3.3)
Subclass3	Trichostachys	30 (100)	30 (14.0)
Subclass4	Stenostachys	5 (100)	0 (0)
Subclass5	Hordeum	26 (100)	0 (0)
Subclass6	Hordeum	24 (100)	0 (0)
Subclass7	Hordeum	58 (100)	0 (0)
Subclass8	Hordeum	27 (100)	0 (0)
Subclass9	Hordeum	4 (100)	0 (0)

Table 5

Summary of PCA results showing the contribution of different agricultural traits to the first five principal components (PC1-PC5) for Hordeum genus classification.

Characters	PC1	PC2	PC3	PC4	PC5
Seed width	10.71	0.00	0.54	0.41	0.17
Seed area	10.31	0.56	0.67	3.07	0.02
100-seed weight	9.20	0.04	1.93	2.33	0.16
Central-spikelet awn length	8.97	0.92	1.49	4.00	0.08
Seed length	8.44	4.72	0.01	8.02	0.00
Flag-leaf width	7.67	0.12	0.00	12.71	5.12
Flag-leaf area	7.33	0.27	0.00	19.73	3.52
No. of triple-spikelet	7.09	2.62	2.68	8.24	3.29
Lateral-spikelet awn length	6.81	1.36	5.77	0.02	0.32
Flag-leaf length	5.58	8.49	0.06	16.03	0.20
Lateral-spikelet glume length	4.40	9.95	0.26	0.05	27.36
Central-spikelet glume length	4.33	5.41	1.94	0.03	35.50
Culm length	3.41	18.13	0.33	0.16	8.18
Seed roundness	2.90	13.03	0.91	10.52	0.94
Days to flowering	2.65	0.01	29.29	8.75	10.47
Days to maturity	0.14	0.85	53.24	5.94	0.72
Spike length	0.05	33.52	0.87	0.00	3.95
Eigenvalue	8.86	2.34	1.43	1.24	1.16
Variance%	52.1	13.8	8.4	7.3	6.8
Cumulative variance%	52.1	65.9	74.3	81.6	88.4

Note: The table includes eigenvalues, variance explained (%) by each principal component, and cumulative variance (%). PC1 and PC2 together account for 65.9% of the total variance, with PC1 contributing 52.1% and PC2 contributing 13.8%. Traits with the highest contributions to PC1, such as seed width, seed area, and 100-seed weight, were selected as the top 10 influential traits and subsequently used for Linear Discriminant Analysis (LDA), to enhance classification accuracy.

Table 6

Comparison of classification performance metrics (accuracy, precision, recall, and F1-score) between PCA-LDA and RF analyses for section and species classification of the Hordeum genus.

Metric	PCA-LDA Classification		RF Classification
	Section	Species	Section	Species
Accuracy%	92.3	90.8	100.0	100.0
Precision%	92.0	94.3	100.0	100.0
Recall%	92.3	90.8	100.0	100.0
F1-Score%	92.1	91.6	100.0	100.0

Table 7

5-fold cross-validation results for PCA-LDA and RF analyses, showing accuracy (%) for each fold in section and species classification.

Fold Number	PCA-LDA Classification Accuracy%		RF Classification Accuracy%
	Section	Species	Section	Species
1-Fold	88.4	90.7	95.3	93.0
2-Fold	88.4	88.4	100.0	95.3
3-Fold	90.7	83.7	100.0	95.3
4-Fold	95.4	95.3	100.0	97.7
5-Fold	95.4	88.4	99.7	90.7
Mean	91.6	89.3	98.6	94.4
SD	3.15	3.8	1.9	2.4

Note: SD—standard deviation.

References

1. Shakhatreh, Y.; Haddad, N.; Alrababah, M.; Grando, S.; Ceccarelli, S. Phenotypic diversity in wild barley (Hordeum vulgare L. ssp. spontaneum (C. Koch) Thell.) accessions collected in Jordan. Genet. Resour. Crop Evol.; 2009; 57, pp. 131-146. [DOI: https://dx.doi.org/10.1007/s10722-009-9457-8]

2. Von Bothmer, R.; Jacobsen, N. Origin, taxonomy, and related species. Barley; 1985; 26, pp. 19-56.

3. Von Bothmer, R.; van Hintum, T.; Knüpffer, H.; Sato, K. Diversity in Barley (Hordeum vulgare); Elsevier: Amsterdam, The Netherlands, 2003.

4. Blattner, F.R. Phylogenetic analysis of Hordeum (Poaceae) as inferred by nuclear rDNA ITS sequences. Mol. Phylogenetics Evol.; 2004; 33, pp. 289-299. [DOI: https://dx.doi.org/10.1016/j.ympev.2004.05.012] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15336664]

5. Sato, K. History and future perspectives of barley genomics. DNA Res.; 2020; 27, dsaa023. [DOI: https://dx.doi.org/10.1093/dnares/dsaa023]

6. Westerbergh, A.; Lerceteau-Köhler, E.; Sameri, M.; Bedada, G.; Lundquist, P.-O. Towards the Development of Perennial Barley for Cold Temperate Climates—Evaluation of Wild Barley Relatives as Genetic Resources. Sustainability; 2018; 10, 1969. [DOI: https://dx.doi.org/10.3390/su10061969]

7. Shah, S.A.A.; Luo, H.; Pickupana, P.D.; Ekeze, A.; Sohel, F.; Laga, H.; Li, C.; Paynter, B.; Wang, P. Automatic and fast classification of barley grains from images: A deep learning approach. Smart Agric. Technol.; 2022; 2, 100036. [DOI: https://dx.doi.org/10.1016/j.atech.2022.100036]

8. Bian, J.; Cui, L.; Wang, X.; Yang, G.; Huo, F.; Ling, H.; Chen, L.; She, K.; Du, X.; Levi, B. Genomic and phenotypic divergence in wild barley driven by microgeographic adaptation. Adv. Sci.; 2020; 7, 2000709. [DOI: https://dx.doi.org/10.1002/advs.202000709]

9. Hubner, S.; Bdolach, E.; Ein-Gedy, S.; Schmid, K.J.; Korol, A.; Fridman, E. Phenotypic landscapes: Phenological patterns in wild and cultivated barley. J. Evol. Biol.; 2013; 26, pp. 163-174. [DOI: https://dx.doi.org/10.1111/jeb.12043]

10. Aberg, E.; Wiebe, G.A. Classification of barley varieties grown in the United States and Canada in 1945; U.S. Department of Agriculture: Washington, DC, USA, 1946; [DOI: https://dx.doi.org/10.22004/ag.econ.169914]

11. Yang, J.; Yen, C.; Lu, B.; Von Bothmer, R. The genus Hordeum (Gramineae) in China. Variation and distribution. Willdenowia; 1987; 16, pp. 479-490.

12. Mizianty, M. Variability and structure of natural populations of Hordeum murinum L. based on morphology. Plant Syst. Evol.; 2006; 261, pp. 139-150. [DOI: https://dx.doi.org/10.1007/s00606-006-0437-6]

13. Cabİ, E.; DoĞAn, M.; Mavİ, D.Ö. Comparative leaf anatomy of the genus Hordeum L. (Poaceae). Turk. J. Bot.; 2011; 35, pp. 357-368. [DOI: https://dx.doi.org/10.3906/bot-1003-14]

14. Amer, W.M.; Hegazy, A.K.; Azer, S.A. Taxonomic revision of genus Hordeum L. (Gramineae) in Egypt. Int. J. Biodivers. Conserv.; 2013; 5, pp. 198-208.

15. Jayani, P.G.; Lakshika, T.S.T. Computer-aided interpretable features for leaf image classification. arXiv; 2021; arXiv: 2106.08077

16. Ali, S.; Hassan, M.; Kim, J.Y.; Farid, M.I.; Sanaullah, M.; Mufti, H. FF-PCA-LDA: Intelligent Feature Fusion Based PCA-LDA Classification System for Plant Leaf Diseases. Appl. Sci.; 2022; 12, 3514. [DOI: https://dx.doi.org/10.3390/app12073514]

17. Rezaei, M.; Gupta, S.; Diepeveen, D.; Laga, H.; Jones, M.G.K.; Sohel, F. Barley disease recognition using deep neural networks. Eur. J. Agron.; 2024; 161, 127359. [DOI: https://dx.doi.org/10.1016/j.eja.2024.127359]

18. Lee, Y.-H.; Sohn, S.-I.; Hong, S.-H.; Kim, C.-S.; Na, C.-S.; Kim, I.-S.; Jang, M.-S.; Oh, Y.-J. Classification of Convolvulaceae plants using Vis-NIR spectroscopy and machine learning. Environ. Biol. Res.; 2021; 39, pp. 581-589. [DOI: https://dx.doi.org/10.11626/KJEB.2021.39.4.581]

19. Rural Development Administration. Guideline of Agricutural Practices; 2020. Republic of Korea. Available online: https://lib.rda.go.kr/search/mediaView.do?mets_no=000000312330 (accessed on 5 September 2023).

20. Parkhurst, D.F.; Loucks, O. Optimal leaf size in relation to environment. J. Ecol.; 1972; 60, pp. 505-537. [DOI: https://dx.doi.org/10.2307/2258359]

21. Nicotra, A.B.; Atkin, O.K.; Bonser, S.P.; Davidson, A.M.; Finnegan, E.J.; Mathesius, U.; Poot, P.; Purugganan, M.D.; Richards, C.L.; Valladares, F. Plant phenotypic plasticity in a changing climate. Trends Plant Sci.; 2010; 15, pp. 684-692. [DOI: https://dx.doi.org/10.1016/j.tplants.2010.09.008]

22. Ntakirutimana, F.; Xie, W. Unveiling the actual functions of awns in grasses: From yield potential to quality traits. Int. J. Mol. Sci.; 2020; 21, 7593. [DOI: https://dx.doi.org/10.3390/ijms21207593]

23. Li, X.; Tang, Y.; Zhou, C.; Lv, J. Contributions of glume and awn to photosynthesis, ¹⁴C assimilates and grain weight in wheat ears under drought stress. Heliyon; 2023; 9, e21136. [DOI: https://dx.doi.org/10.1016/j.heliyon.2023.e21136] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37886759]

24. Sheehy, J.; Dionora, M.; Mitchell, P. Spikelet numbers, sink size and potential yield in rice. Field Crops Res.; 2001; 71, pp. 77-85. [DOI: https://dx.doi.org/10.1016/S0378-4290(01)00145-9]

25. Lee, Z.; Kim, S.; Choi, S.J.; Joung, E.; Kwon, M.; Park, H.J.; Shim, J.S. Regulation of flowering time by environmental factors in plants. Plants; 2023; 12, 3680. [DOI: https://dx.doi.org/10.3390/plants12213680] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37960036]

26. Wiegmann, M.; Maurer, A.; Pham, A.; March, T.J.; Al-Abdallat, A.; Thomas, W.T.; Bull, H.J.; Shahid, M.; Eglinton, J.; Baum, M. Barley yield formation under abiotic stress depends on the interplay between flowering time genes and environmental cues. Sci. Rep.; 2019; 9, 6397. [DOI: https://dx.doi.org/10.1038/s41598-019-42673-1]

27. Volis, S.; Verhoeven, K.; Mendlinger, S.; Ward, D. Phenotypic selection and regulation of reproduction in different environments in wild barley. J. Evol. Biol.; 2004; 17, pp. 1121-1131. [DOI: https://dx.doi.org/10.1111/j.1420-9101.2004.00738.x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15312084]

28. Hu, Q.; Sun, G. Phylogenetic analysis of two single-copy nuclear genes revealed origin and complex relationships of polyploid species of Hordeum in Triticeae (Poaceae). Genome; 2017; 60, pp. 518-529. [DOI: https://dx.doi.org/10.1139/gen-2016-0179]

29. Baum, B.R.; Edwards, T.; Johnson, D.A. Codependence of repetitive sequence classes in genomes: Phylogenetic analysis of 5S rDNA families in Hordeum (Triticeae: Poaceae). Genome; 2010; 53, pp. 180-202. [DOI: https://dx.doi.org/10.1139/G09-096] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/20237596]

30. Alkhilafi, S. Molecular Phylogeny and Origins of Hordeum Polyploid Species. Master’s Thesis; Saint Mary’s University: Halifax, NC, Canada, 2014.

31. Wang, H.; Sun, D.; Sun, G. Molecular phylogeny of diploid Hordeum species and incongruence between chloroplast and nuclear datasets. Genome; 2011; 54, pp. 986-992. [DOI: https://dx.doi.org/10.1139/g11-063] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22085287]

32. Jin, Y.; Du, X.; Jiang, C.; Ji, W.; Yang, P. Disentangling sources of gene tree discordance for Hordeum species via target-enriched sequencing assays. Mol. Phylogenet Evol.; 2024; 199, 108160. [DOI: https://dx.doi.org/10.1016/j.ympev.2024.108160]

33. Brassac, J.; Blattner, F.R. Species-Level Phylogeny and Polyploid Relationships in Hordeum (Poaceae) Inferred by Next-Generation Sequencing and In Silico Cloning of Multiple Nuclear Loci. Syst. Biol.; 2015; 64, pp. 792-808. [DOI: https://dx.doi.org/10.1093/sysbio/syv035]

34. Blattner, F.R. Multiple intercontinental dispersals shaped the distribution area of Hordeum (Poaceae). New Phytol.; 2006; 169, pp. 603-614. [DOI: https://dx.doi.org/10.1111/j.1469-8137.2005.01610.x]