1 UMR1213 Herbivores, INRA, VetAgro Sup, Clermont Université, Université de Lyon, Saint-Genès-Champanelle, France.
2 INRA, UE0326 Domaine Expérimental du Pin, EXMES, France.
Appl Spectrosc. 2017 Sep;71(9):2164-2176. doi: 10.1177/0003702817712740. Epub 2017 Jun 6.
The forage feed value determined by organic matter digestibility (OMD) and voluntary intake (VI) is hard and expensive. Thus, several indirect methods such as near infrared reflectance (NIR) spectroscopy have been developed for predicting the feed value of forages. In this study, NIR spectra of 1040 samples of feces from sheep fed fresh temperate forages were used to develop calibration models for the prediction of fecal crude ash (CA), fecal crude protein (CP), fresh forage OMD, and VI. Another 136 samples of feces were used to assess these models. Four calibration strategies were compared: (1) species-specific calibration; (2) family-specific calibration; (3) a global procedure; and (4) a LOCAL approach. The first three strategies were based on classical regression models developed on different sample populations, whereas the LOCAL approach is based on the development models from selected samples spectrally similar to the sample to be predicted. The first two strategies use feces-samples grouping based on the species or the family of the forage ingested. Forage calibration data sets gave value ranges of 79-327 g/kg dry matter (DM) for CA, 65-243 g/kg DM for CP, 0.52-0.85 g/g for OMD, and 34.7-100.5 g DM/kg metabolic body weight (BW) for VI. The prediction of CA and CP content in feces by species-specific fecal NIR (FNIR) spectroscopy models showed lower standard error of prediction (SEP) (CA 15.03 and CP 7.48 g/kg DM) than family-specific (CA 21.93 and CP 7.69 g/kg DM), global (CA 19.83 and CP 7.98 g/kg DM), or LOCAL (CA 30.85 and CP 8.10 g/kg DM) models. For OMD, the LOCAL procedure led to a lower SEP (0.018 g/g) than the other approaches (0.023, 0.024, and 0.023 g/g for species-specific, family-specific, and global models, respectively). For VI, the LOCAL approach again led to a lower SEP (6.15 g/kg BW) than the other approaches (7.35, 8.00, and 8.13 g/kg BW for the species-specific, family-specific, and global models, respectively). LOCAL approach performed on FNIR spectroscopy samples gives more precise models for predicting OMD and VI than species-specific, family-specific, or global approaches.
本研究利用绵羊采食新鲜温带牧草后粪便的近红外反射光谱（NIR）数据，建立了预测粪便粗灰分（CA）、粗蛋白（CP）、新鲜牧草有机物消化率（OMD）和采食量（VI）的校正模型。此外，还使用了另外 136 个粪便样本对这些模型进行了评估。比较了 4 种校正策略：（1）种特异性校正；（2）科特异性校正；（3）全局校正；（4）局部校正。前三种策略基于不同样本群体开发的经典回归模型，而局部校正策略则基于与待预测样本光谱相似的样本开发校正模型。前两种策略基于所采食牧草的种类或科对粪便样本进行分组。牧草校正数据集的 CA 值范围为 79-327 g/kg 干物质（DM），CP 值范围为 65-243 g/kg DM，OMD 值范围为 0.52-0.85 g/g，VI 值范围为 34.7-100.5 g DM/kg 代谢体重（BW）。种特异性粪便近红外（FNIR）光谱模型预测 CA 和 CP 含量的标准预测误差（SEP）较低（CA 为 15.03 和 CP 为 7.48 g/kg DM），优于科特异性（CA 为 21.93 和 CP 为 7.69 g/kg DM）、全局（CA 为 19.83 和 CP 为 7.98 g/kg DM）或局部（CA 为 30.85 和 CP 为 8.10 g/kg DM）模型。对于 OMD，局部校正法的 SEP（0.018 g/g）低于其他方法（种特异性、科特异性和全局模型的 SEP 分别为 0.023、0.024 和 0.023 g/g）。对于 VI，局部校正法的 SEP（6.15 g/kg BW）再次低于其他方法（种特异性、科特异性和全局模型的 SEP 分别为 7.35、8.00 和 8.13 g/kg BW）。局部校正法应用于 FNIR 光谱样本，可建立更精确的 OMD 和 VI 预测模型，优于种特异性、科特异性或全局方法。
