This study aimed to select superior barren-tolerant germplasm, establish an evaluation model, and identify diagnostic indicators in foxtail millet germplasm resources from Shanxi, so as to provide the theoretical basis and methodological support for the efficient breeding of barren-tolerant foxtail millet varieties.

617 local foxtail millet varieties were used as experimental materials. Trials were carried out under low-fertility stress in Taiyuan (23TY, 24TY) and under normal fertility conditions in Dongyang (23DY, 24DY) from 2023 to 2024. Eleven indicators were measured, including stem node number (SNN), stem diameter (SD), panicle diameter (PD), plant height (PH), peduncle length (PeL), panicle length (PaL), leaf length (LL), leaf width (LW), spike weight per plant (SWP), grain weight per plant (GWP), and thousand-grain weight (TGW). A comprehensive evaluation and identification were carried out using multiple analytical methods, such as normal distribution test, barren tolerance coefficient difference analysis, correlation analysis, cluster analysis, and regression analysis.

The results showed significant differences in agronomic traits, with coefficients of variation ranging from 9.23% to 48.05%. Specifically, the coefficient of variation for LW in 23TY was the smallest (9.23%), while the coefficient for GWP in 24TY was the largest (48.05%). K-S test results indicated that, except for SNN, PaL, LW, and SWP, other traits showed varying degrees of normal distribution in different environments (P>0.05). Correlation analysis showed that SWP and GWP exhibited a significant positive correlation with the highest correlation coefficient, followed by SNN with PH, PaL with LL. In addition, PeL and SNN showed a significant negative correlation in 23TY and 23DY; LL and PH, LL and SNN, PeL and PD had a significant negative correlation in 23TY; and PeL was significantly negatively correlated with PH in 23DY. Principal component analysis results revealed that 11 phenotypic traits were converted into 2 comprehensive indices, with a cumulative explanation rate of 59.68%. Using the membership function method calculated the barren tolerance (D value), and cluster analysis was performed. The varieties were classified into three categories: Class Ⅰ included 13 barren-sensitive varieties with D values ranging from 0.09 to 0.28; Class Ⅱ included 595 moderately barren-tolerant varieties with D values ranging from 0.40 to 0.81; Class Ⅲ included 9 barren-tolerant varieties with D values ranging from 0.84 to 0.93. Maorangu was the most barren-tolerant variety from Yangquan city. Through multiple regression analysis, a predictive evaluation model for barren tolerance was established: Y=-0.031+0.129X₉+0.167X₁₀+0.141X₆+0.233X₈+0.205X₄ (R²=0.951, P<0.001).

Using multivariate statistical analysis methods, it is reliable to evaluate and predict the barren tolerance of foxtail millet germplasm. Five phenotypic traits are selected as evaluation indicators for barren tolerance: plant height, panicle length, leaf width, spike weight per plant, grain weight per plant. The Maorangu variety from Yangquan city was the strongest barren tolerance.