The experimental site was located at the National Forage Germplasm Nursery in Haibei County, Qinghai Province (on the Tibetan Plateau) (100eau)mtal E, 3600eau)m N) at an altitude of 3156 m. The region has a typical plateau continental climate, characterized by thin air, abundant sunshine, strong solar radiation, and highly variable weather conditions. There are only two seasons: cold and warm, with a large diurnal temperature range. The average annual temperature is 8.9 °C, with no absolute frost-free period; the frost-free period is approximately 30 d per year, and the growing season for plants is about 120 d. Precipitation and heat occur in the same period, with an average annual precipitation of approximately 375 mm, mainly concentrated from July to September. From June 2022 to September 2024, the average monthly temperature and precipitation were 2.6 °C and 95.9 mm, respectively (Figure 1). The average annual sunshine duration is approximately 2980 hours, and the average annual evaporation is approximately 1400 mm. The chestnut soil is the main soil type, with soil organic matter, total nitrogen, total phosphorus, and available phosphorus contents of 32.5 g/kg, 1.6 g/kg, 1.4 g/kg, and 2.2 mg/kg, respectively, and a soil pH of 8.4 (Shi et al., 2024).

The experimental materials were sourced from different ecological regions of the Tibetan Plateau (Table 1). After field classification and screening, 10 Elymus nutans varieties with better productivity and stronger adaptability were selected from 600 wild resources for the experiment (15-010, 16-123, 16-209, 17-067, 17-213, 18-023, 18-044, 18-175, 19-210, and 20-080). The experiment was initiated by sowing in early June 2021. Prior to sowing, the experimental site was deeply plowed and leveled, and stones, roots, and other debris were removed.

The experiment was conducted continuously for three years, using a completely randomized block design with six replications. To fully exploit the productivity of the tested materials, a single-plant cultivation method was employed. Seeds were first placed in seedling trays for germination (model: external dimensions 540 mm × 280 mm, 105-hole tray, hole diameter 35 mm at the top, 10 mm at the bottom, depth 40 mm, and individual hole volume 25 ml). The seedling trays were filled with field soil and nutrient soil mixture (1:1) and germinated in a greenhouse. To improve seedling survival rates, the seedling trays were transferred to the experimental site after germination for acclimatization. When the seedlings reached the two-leaf-one-heart stage, they were transplanted into the field. The plant and row spacing in the experimental field were both set at 80 cm, with each experimental plot measuring 3 m × 5 m. A 1.5 m wide protective border was established around the experimental field, using Elymus sibiricus cv. Qingmu No. 1 as the planting material for the protective border. The pathways between plots were set at 1 m wide. Compound phosphate fertilizer ((NH₄)₂HPO₄) was used as the base fertilizer, applied at a rate of 75 kg/hm². After the experimental seedlings survived, irrigation, fertilization, and grazing were prohibited. Conventional field management was carried out in the experimental field. In the 1-year (2021), weeding was performed three times. In the second (2022), third (2023), and fourth (2024) years, intertillage and weeding were conducted twice each (after regreening and before jointing), along with plant disease, pest, and rodent control.

Sampling and index measurements were conducted during the flowering period of Elymus nutans in August 2022 (second year), 2023 (third year), and 2024 (fourth year). In each experimental plot, 10 plants with similar growth vigor and no obvious pests and diseases were randomly selected. For each resource, plant height (natural height), crown breadth (average width of forage in the north-south and east-west directions), and grass height (height of basal leaves) were measured using a tape measure. The number of tillers (number of branches growing from the base of the main stem) per plant was counted, and the size of the leaf (flag leaf) was measured in terms of length and width. The plants were then weighed to calculate the dry grass yield per plant.

Forage yield: During the early flowering stage of Elymus nutans, 10 individual plants with consistent growth vigor were randomly selected in each plot (excluding marginal effects) and cut at the ground level. After removing weeds, the samples were immediately weighed to record the fresh weight. The Elymus nutans samples were then placed in large envelopes and brought back to the laboratory. They were first dehydrated at 105 °C for 30 min, followed by drying at 75 °C until a constant weight was achieved, which was recorded as the dry weight (Wang et al., 2022; Wang et al., 2025).

The data were initially analyzed using Excel 2019. Normality (Kolmogorov-Smirnov test) and homogeneity of variance (Levene’s test) were assessed for plant height, leaf length, leaf width, crown breadth, grass height, number of tillers, dry-fresh ratio, and dry grass yield per plant of different Elymus nutans resources using SPSS 19.0 (SPSS 19.0, Chicago, IL, USA) (Edgington, 1965; Odzimierz Bryc, 1995). Subsequently, one-way analysis of variance (ANOVA) with least significant difference (LSD) multiple comparisons was used to analyze the significance of these indices at the 0.05 level of significance. Two-way ANOVA was employed to evaluate the effects of years, varieties, and their interactions on the agronomic traits and forage yield of Elymus nutans.

K-means clustering analysis was conducted on the key traits of Elymus nutans resources using the factoextra and ggplot2 packages in R. Based on the K-means clustering results, the TOPSIS model, was applied using the plyr package to comprehensively evaluate the agronomic traits and forage yield per plant of Elymus nutans resources.

The Random Forest model was constructed using the randomForest package in R 4.0.2 (R Development Core Team) to rank the importance of the factors influencing forage yield. The significance of the model and individual factors was assessed using the rfUtilities and rfPermute packages. To further elucidate the factors affecting the forage production performance of individual Elymus nutans plants, a piecewise structural equation model (SEM) was built using the piecewiseSEM package in R 4.0.2, based on the Random Forest analysis, to explore the processes and pathways influencing forage yield in different Elymus nutans resources. All figures were generated using the Origin 2021 software.

Piecewise Structural Equation Model (Piecewise SEM): Piecewise SEM decomposes complex causal relationships into multiple simple linear models. Each model corresponds to a path in the path diagram and can be fitted and evaluated independently of the others.

We applied piecewise structural equation modeling to analyze the interactions and pathways between agronomic traits (plant height, number of tillers, crown breadth, grass height, and dry-fresh ratio) and dry grass yield per plant in Elymus nutans.

TOPSIS model: The TOPSIS model is a multi-criteria decision-making tool. It determines the best option by comparing the distance of each alternative to the ideal and anti-ideal solutions. It is widely used for evaluation and selection tasks.

The weighting procedure of the TOPSIS model with the entropy-weight method was executed in four steps (n = number of accessions, m = number of indicators).

Normalization – construct matrix P

p_ij = x_ij / Σ_i x_ij (i = 1…n, j = 1…m)

Entropy value e_j

q_ij = p_ij / Σ_i p_ij

e_j = −(ln n)⁻¹ Σ_i(q_ij ln q_ij)

Dispersion coefficient

d_j = 1 − e_j

Weight determination

w_j = d_j / Σ_j d_j (Σ_j w_j = 1)

Weights were then applied to generate the weighted normalized matrix V = p_ij w_j for the subsequent TOPSIS ranking.

The TOPSIS model was applied to evaluate Elymus nutans accessions. This includes data standardization of agronomic traits (plant height, grass height, leaf length, leaf width, crown breadth, and number of tillers) and dry grass yield per plant. Subsequently, we constructed a weighted standardized matrix. The ideal and anti-ideal solutions were determined. Distances of each accession to both solutions were calculated. Finally, closeness coefficients (0-1, closer to 1 is better) were computed to identify the optimal germplasm resources.

The phenotypic characteristics and dry grass yield per plant of Elymus nutans resources varied significantly among the different years (Figure 2). Plant height, grass height, leaf length, leaf width, crown breadth, number of tillers, and dry grass yield per plant were highest in the third year, followed by the second year, and lowest in the fourth year, with significant differences among years (P < 0.05). In the second year, 18–023 had the highest plant height and leaf length (P < 0.05), widest leaf width (P < 0.05), most tillers per plant (P < 0.05), and highest dry grass yield per plant (P < 0.05), which were 109.00 cm, 15.67 cm, 1.02 cm, 62.50 tillers per plant, and 81.32 g, respectively (Figures 2A, C, D, F, G). It also had the highest grass height, at 22.00 cm (Figure 2B). 15-010, 16-123, and 18–023 had the largest crown breadth, 95.83–102.50 cm (Figure 2E). 17–213 had the highest dry-fresh ratio, at 0.41 (Figure 2H). In the third year, 15–010 and 18–023 had the highest plant and grass heights (P < 0.05), measuring 120.83–128.83 cm and 32.50–34.66 cm, respectively (Figures 2A, B). 18–023 had the longest leaf length and widest leaf width, at 17.60 cm and 1.03 cm, respectively (Figures 2C, D). 15–010 had the largest canopy width, at 127.53 cm (Figure 2E). 18–023 had the most tillers per plant and highest dry grass yield per plant (P < 0.05), with 433.33 tillers per plant and 82.83 g, respectively (Figure 2F, G). 17–213 had the highest dry-fresh ratio, at 0.49 (Figure 2H). In the fourth year, 18–023 had the highest plant height, grass height, leaf length, crown breadth, number of tillers, and dry grass yield per plant, which were 99.03 cm, 19.01 cm, 13.16 cm, 86.67 cm, 52.51 tillers per plant, and 79.33 g, respectively (Figures 2A, B, C, E, F, G).

The results of the two-way ANOVA (Table 2) indicated that year and variety, as well as their interaction, had highly significant effects on plant height, grass height, leaf length, leaf width, crown breadth, number of tillers, and dry grass yield per plant of the experimental materials (P < 0.01).

To effectively evaluate the comprehensive performance of the 10 excellent Elymus nutans resources, K-means clustering analysis was conducted based on key agronomic traits and yield-related characteristics of the resources. The 10 Elymus nutans resources were clustered into two groups. The P-values between the two groups were 0.006 for plant height, 0.037 for grass height, 0.028 for leaf length, 0.015 for leaf width, 0.009 for crown breadth, 0.003 for number of tillers, and 0.021 for dry grass yield per plant. Among these, plant height, crown breadth, and number of tillers showed extremely significant differences (Table 3). Group 1 accounted for 48% of the resources, whereas group 2 accounted for 52%. The resources in the second group mainly included 18-023, 15-010. The resources in the second group mainly included 16-123, 17-067, 18-175, 18-044, 17-209, 20-080, 19-210, and 17-213 (Figure 3A).

To further identify the Elymus nutans resources with the best overall performance, the TOPSIS model (Technique for Order Preference by Similarity to an Ideal Solution) was applied to comprehensively evaluate the dry grass yield per plant, plant height, grass height, leaf length, leaf width, crown breadth, number of tillers, and dry-fresh ratio of different Elymus nutans resources. The respective weight coefficients were 0.068 for plant height, 0.146 for grass height, 0.231 for leaf length, 0.105 for leaf width, 0.079 for crown breadth, 0.168 for number of tillers, 0.106 for dry grass yield per plant, and 0.096 for dry-fresh ratio (Table 4). The results showed that the degree of fit, from highest to lowest, was 18-023, 15-010, 18-044, 18-175, 17-213, 16-123, 20-080, 16-209, 19-210, and 17-067. Among them, 18–023 had the highest degree of fit at 0.93, while 17–067 had the lowest degree of fit at 0.11 (Figure 3B). Comprehensive evaluation results indicated that 18–023 had the best overall performance and was the optimal Elymus nutans resource for cultivation.

To determine the effects of plant height, grass height, leaf size (leaf length, leaf width), crown breadth, number of tillers, and dry-fresh ratio on the dry grass yield per plant, a Random Forest model was used to rank the importance of these factors (Figure 4). The results indicated that these factors collectively explained 79.17% of the variation in dry grass yield. Plant height, number of tillers, dry-fresh ratio, grass height, crown breadth, leaf length, and leaf width all had highly significant effects on forage yield (P < 0.01).

Based on the results of the Random Forest model analysis, factors that significantly influenced the forage production performance of Elymus nutans were selected to construct a piecewise structural equation model (Figure 5). This model was used to explore the processes and pathways through which year, variety, and their interactions affect the forage production performance of Elymus nutans. The results showed that the piecewise structural equation model had a good fit (P = 0.974, Fisher’s C = 0.051). Year indirectly affected the dry grass yield per plant through its effects on plant height, crown breadth, grass height, and dry-fresh ratio. Variety indirectly affected the dry grass yield per plant through its effects on grass height and dry-fresh ratio. Plant height, crown breadth, grass height, and dry-fresh ratio all had significant positive effects on the dry grass yield per plant, with path coefficients of 0.257, 0.782, 1.285, and 0.433, respectively (Figure 5A).

The direct, indirect, and total effects of each factor were calculated based on the standardized path coefficients in the piecewise structural equation model (Figure 5B). The results showed that for the dry grass yield per plant of Elymus nutans resources, plant height, number of tillers, and dry-fresh ratio had the highest total effect values, at 0.757, 0.703, and 0.686, respectively. Therefore, year and variety mainly affected the forage yield of Elymus nutans through their effects on plant height, number of tillers, and dry-fresh ratio.