Prediction of total mixed ration composition for dairy buffaloes using portable near-infrared spectrophotometers

Laboratory methods (LAB) for feed analysis are reliable but time-consuming and expensive. Near-infrared reflectance spectroscopy (NIRS) offers a non-destructive alternative, and portable devices have increased interest for on-farm real-time monitoring of feed composition. However, their accuracy relative to LAB remains insufficiently validated. This study evaluated two portable NIRS instruments compared with LAB methods for chemical characterisation of total mixed rations (TMR) for dairy buffaloes. 230 TMR samples were collected, scanned with both devices, and analysed with LAB methods. Parameters assessed included dry matter (DM), ether extract (EE), ASH, crude protein (CP), neutral detergent fibre, acid detergent fibre, acid detergent lignin, and starch. Agreement among methods was examined using Bland-Altman and Passing-Bablok analyses, while differences were evaluated with Friedman ANOVA. Principal Component Analysis (PCA) was applied to explore the multidimensional structure of the dataset. Finally, Passing-Bablok regression was used to correct NIRS bias and reassess agreement with the reference method. Good agreement was observed for DM (mean differences 3.12% for NIRS_A, 0.99% for NIRS_B) and moderate for EE (<0.5%), while larger discrepancies occurred for fibre fractions, CP, ASH, and starch. Passing-Bablok regression revealed significant constant and proportional biases, particularly for fibre and starch. Bias correction improved agreement, with DM and EE predictions closely matching LAB values (e.g. DM: slope = 1.00, intercept = 0.95). The PCA indicated that NIRS devices capture overall compositional patterns, with NIRS_B showing lower variability. In conclusion, portable NIRS devices cannot yet replace laboratory analyses for buffalo TMR; however, after bias correction, they may have potential as tools for on-farm monitoring.