To advance the application of near-infrared （NIR） spectroscopy in assessing the quality of mulberry （Morus alba L.） silage， a total of 105 silage samples were collected. Interval Partial Least Squares （iPLS） was employed to extract informative spectral intervals， and rapid NIR prediction models were subsequently developed using Partial Least Squares （PLS）. The results indicated that the models for crude protein， acid detergent fiber， neutral detergent fiber， and pH exhibited the best performance and were suitable for practical prediction （R²cv， R²p>0.8； RPDcv， RPDp>2.5）. Models for dry matter， lactic acid， and condensed tannins demonstrated moderate performance and were adequate for approximate estimation （2<RPDcv<2.5； 2<RPDp<2.5）. In contrast， the model for hydrolyzable tannins required further refinement （RPDp<2）. Collectively， these findings suggest that models constructed based on characteristic spectral intervals offer enhanced stability and predictive accuracy. This approach provides a theoretical basis and technical support for the efficient production and utilization of mulberry silage， while also offering valuable insights for the development of NIR analytical instrumentation.

Mulberry silage  /  Silage  /  Near-Infrared Spectroscopy （NIRS）  /  Characteristic spectrum extraction  /  Chemometrics