关键词: 燕麦干草, 近红外光谱, 品质成分, 预测模型, 偏最小二乘法, 种质资源

Abstract: To develop a rapid quality detection model for oat （Avena sativa） hay applicable across environments， 849 samples were collected at three sites （Baoding and Zhangjiakou in Hebei， Guyuan in Ningxia） for two years. Crude protein （CP）， neutral detergent fiber （NDF）， acid detergent fiber （ADF）， soluble sugars （SS）， and total potassium （K） were measured. Prediction models were established using near-infrared spectroscopy （NIRS） combined with partial least squares regression （PLSR）. All five traits showed broad， near-normal distributions， indicating good sample representativeness. The CP， SS， and K models performed well， with cross-validation coefficients （R²） ≥ 0.89， ratio of performance to deviation （RPD） ≥ 3.0， and validation set correlations （r） of 0.84， 0.86， and 0.83， respectively， demonstrating strong external predictive ability. The NDF and ADF models achieved R² ≥ 0.85 and RPD ≥ 2.5， with r values of 0.78 and 0.79， meeting requirements for routine screening. Based on model predictions and forage quality standards， 56 high-quality forage oat accessions were identified. This study provides an efficient technical platform for high-throughput phenotyping and quality-based breeding in oat germplasm.

Key words: Oat hay, Near-infrared spectroscopy, Quality components, Prediction model, Partial least squares, Germplasm resources