The field trial was conducted at the Forage Reproduction Base in Xihai Town, Haiyan County, Haibei Tibetan Autonomous Prefecture, Qinghai Province (36°59’36” N, 100°52’48” E). This region is characterized by a continental plateau climate and is situated at an elevation of 3,156 m. The mean annual temperature is approximately 0.5 °C, while the mean annual precipitation totals about 369.5 mm, predominantly occurring between July and September. The diurnal temperature variation is significant. The average annual wind speed is 3.3 m/s, and the area enjoys abundant sunshine, receiving approximately 2,970 hours of sunlight annually. The frost-free period extends for about 92 days. Throughout 2023, the mean monthly temperature and precipitation were 3.1 °C and 55.13 mm, respectively (Figure 1). In 2024, the corresponding values throughout the year were 3.5 °C and 62.8 mm. The meteorological data were sourced from the China National Meteorological Science Data Center. The soil is classified as Chestnut soil, exhibiting a pH of 8.31. The soil organic carbon content measures 21.15 g/kg, total nitrogen is 3.06 g/kg, and total phosphorus is 0.27 g/kg.

The experimental materials were obtained from the Qinghai-Tibet Plateau (Figure 2; Table 1). This experiment was conducted using a common garden design, in which soil conditions, water and fertilizer inputs, and management practices were strictly controlled to minimize interference from local environmental factors. As a result, the observed differences in agronomic traits and yield among the accessions can be primarily attributed to their genetic background and the adaptive effects associated with their original elevations. The experimental site was prepared by deep plowing and leveling, followed by the removal of surface stones and weed residues. A randomized complete block design (RCBD) was employed, consisting of five replications. Sowing occurred in mid-May 2022 using the hill-seeding method, with each hill planted with 4–5 seeds at a depth of 3–4 cm. Both plant and row spacing were set at 50 cm. Each plot measured 3 m × 5 m and was surrounded by a 1 m wide buffer row, with a 1 m wide working path maintained between adjacent plots. Diammonium phosphate (DAP) was applied as a basal fertilizer at a rate of 75 kg·ha^-1. Following seedling emergence, thinning was conducted to ensure that only a single healthy seedling, free of visible pests or diseases, remained per hill. The plants were maintained under natural growth conditions throughout the trial period, without artificial irrigation, topdressing, or grazing. Weed control was performed three times in 2022 and four times annually in both 2023 and 2024.

Systematic sampling was conducted during the full blooming stage of E. sibiricus in 2023 (the second growth year) and 2024 (the third growth year). Five uniformly growing plants, exhibiting no evident disease or pest symptoms, were selected from each experimental plot using a systematic sampling method. A calibrated steel tape was employed to measure the vertical distance from the base to the highest growth point, recorded as absolute plant height, as well as the length and width of the flag leaf. Stem diameter was measured with a vernier caliper, and the total number of tillers was recorded simultaneously. Upon completion of the morphological measurements, the plants were cut at ground level. Fresh biomass was determined immediately after removing impurities such as attached soil, litter, and other plant species. The samples were then sealed and transported to the laboratory, where they were fixed at 105 °C for 30 minutes and subsequently dried at a constant temperature of 75 °C until a constant weight was achieved to ascertain dry matter content. At the full ripening stage (late August for both years), seeds were manually collected from individual plants and weighed.

Data were initially organized using Microsoft Excel 2016. Statistical analyses were conducted with SPSS 27.0 (Chicago, IL, USA). The normality of the measured traits—including plant height, leaf length, leaf width, stem diameter, effective tillers, total tiller number, forage yield, and seed yield across different elevations of E. sibiricus—was assessed using the Kolmogorov-Smirnov test, along with an evaluation of the homogeneity of variance. A one-way analysis of variance (ANOVA), followed by the least significant difference (LSD) multiple comparisons test (α = 0.05), was performed. To examine trait variations among different germplasm resources, a two-way ANOVA was employed. Data analysis and visualization were performed using R version 4.4.1. K-means clustering analysis was executed using the factoextra and ggplot2 packages in R to identify key phenotypic characteristics. A Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) multi-criteria decision-making model was implemented via the plyr package to comprehensively evaluate agronomic traits, as well as individual plant forage and seed production performance. Mantel tests were utilized to analyze the correlation strength between agronomic traits and production performance indicators. A structural equation model (SEM) was constructed using the piecewise SEM package to quantify the direct and indirect effects, along with path coefficients, of various factors on forage and seed yield, thereby elucidating the regulatory pathways underlying yield formation. Standardized effect values were visualized and finalized using Origin 2022. The Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is an intuitive and practical multi-criteria decision-making method. It ranks alternatives by calculating the distance between each alternative and the fictitious “ideal best solution” and “ideal worst solution,” and then derives the relative closeness to the ideal solution for ranking (Wang et al., 2022b). Structural Equation Modeling (SEM) is a multivariate statistical analysis method that integrates factor analysis and path analysis. It is primarily used to test whether the proposed theoretical model—which hypothesizes relationships among observed variables and latent variables, as well as among latent variables themselves—fits the data well (Cheung, 2018).

This study examined the agronomic traits and yield of E. sibiricus during its second and third years. In the second year, accession 2019–005 exhibited the tallest plant height at 75.67 cm, significantly surpassing all other accessions (P < 0.05; Figure 3a). Accession 2019–002 displayed the largest stem diameter at 3.57 mm and the highest number of effective tillers, with 59 tillers per plant, demonstrating significant superiority over 2019-001 (P < 0.05; Figure 3b). The longest leaf length, measuring 45.2 cm, and the highest seed yield of 40.41 g were recorded for 2019-006, both significantly exceeding those of 2019-002 (P < 0.05; Figures 3c, h). Accession 2019–001 exhibited the greatest leaf width at 1.2 cm and the highest total tiller number, with 137 tillers per plant, significantly outperforming 2019-003 (P < 0.05; Figures 3d, f). Accession 2019–003 produced the highest forage yield at 266.85 g, significantly surpassing all other accessions (P < 0.05; Figure 3g). In the third year, accession 2019–001 exhibited superior performance in plant height (174.02 cm), leaf width (1.63 cm), number of effective tillers (117 tillers per plant), and total tiller number (270 tillers per plant), all of which were significantly greater than those of the other accessions (P < 0.05; Figures 3a, d–f). Accession 2019–003 achieved the largest stem diameter at 4.03 mm and the highest forage yield at 488.57 g, both significantly exceeding those of 2019-001 (P < 0.05; Figures 3b, g). The longest leaf length (21.26 cm) and the highest seed yield (77.14 g) were recorded in 2019-006, significantly greater than those of 2019-004 (P < 0.05; Figures 3c, h).

The results of the two-way ANOVA demonstrated that both elevation and plant year exerted highly significant main effects (P < 0.01) on plant height, leaf length, leaf width, number of effective tillers, total tiller number, forage yield, and seed yield. Stem diameter was significantly affected solely by the elevation factor. The interaction between elevation and age was highly significant (P < 0.01) for plant height, stem diameter, number of effective tillers, and total tiller number. In contrast, no significant interaction effects were detected for leaf length, leaf width, forage yield, or seed yield (Table 2).

The Mantel test analysis revealed significant positive correlations between forage yield per plant and several morphological traits, including plant height, stem diameter, leaf length, leaf width, and total tiller number (Figure 4). Additionally, seed yield per plant exhibited significant positive correlations with plant height, stem diameter, leaf length, leaf width, and effective tiller number. A notable positive correlation was also identified between effective tiller number and total tiller number. Moreover, both effective tiller number and total tiller number demonstrated significant positive correlations with plant height and leaf length.

A comprehensive multi-trait evaluation was performed on six wild E. sibiricus accessions sourced from varying elevations. K-means clustering analysis, based on key agronomic traits (Figure 5), indicated that these accessions could be categorized into two distinct clusters. The first cluster, which included accessions 2019-001, 2019-002, 2019-003, and 2019-006, represented 59% of the total and exhibited superior phenotypic performance. In terms of elevational distribution, Cluster I comprised accessions from elevations of 3000 m to 4000 m, whereas Cluster II contained accessions from elevations exceeding 4000 m.

To identify appropriate elevational gradients for wild E. sibiricus, a comprehensive evaluation model was developed utilizing a multi-criteria decision-making approach known as TOPSIS. This model incorporated several key parameters, including plant height, stem diameter, leaf length, leaf width, number of effective tillers, total tiller number, forage yield, and seed yield (Figure 6). The evaluation results revealed a descending order of similarity scores: 2019-001 > 2019-006 > 2019-002 > 2019-003 > 2019-005 > 2019-004. Accession 2019–001 achieved the highest similarity score of 0.74, indicating superior overall performance, whereas accession 2019–004 recorded the lowest score of 0.15. Therefore, the elevational region from which 2019–001 was collected (3147 m) may represent the optimal elevational gradient for wild E. sibiricus.

To elucidate the pathways and quantify the path coefficients through which elevation, year, and their interaction influence forage and seed yield, a structural equation model (SEM) was developed. The results demonstrated a good model fit (P = 0.732, Fisher’s C = 5.237). Elevation exerted an indirect effect on forage yield by influencing plant height and stem diameter. Year also had an indirect impact on forage yield through its effects on plant height, leaf length, leaf width, and total tiller number. The interaction between elevation and year indirectly influenced forage yield via plant height, stem diameter, and total tiller number. Leaf width and stem diameter had positive direct effects on forage yield, with path coefficients of 0.408 and 0.230, respectively. In contrast, plant height exhibited a significant negative direct effect, with a path coefficient of -0.749 (Figure 7a).

Elevation indirectly influenced seed yield by affecting plant height. Additionally, year had an indirect impact on seed yield through its effects on plant height, leaf length, leaf width, and the number of effective tillers. The interaction between elevation and year also exerted an indirect influence, mediated by plant height and the number of effective tillers. Leaf length, leaf width, and the number of effective tillers exhibited positive direct effects on seed yield, while plant height showed a negative direct effect. This model demonstrated a good fit (P = 0.563, Fisher’s C = 2.968) (Figure 7c).

Using the standardized path coefficients derived from the piecewise structural equation modeling (SEM), the contributions of each factor were quantified through effect decomposition. In the context of forage yield formation, plant height (total effect: -0.549) and leaf width (total effect: 0.408) emerged as the primary drivers. Path analysis indicated that elevation, year, and their interaction predominantly influenced forage yield by mediating these two traits (Figure 7b). Regarding seed yield formation, year, leaf length, and effective tiller number demonstrated the highest total effects, with values of 0.98, 0.67, and 0.41, respectively. Consequently, elevation and year primarily affected seed yield by modulating leaf length and effective tiller number (Figure 7d).