Abstract Presently, there are no models for predicting nitrogen excretion (NE) in buffaloes, which are required for preparation of inventory of NE. Thus, this study aimed to develop statistical models to predict NE from animal and dietary characteristic variables. A dataset of 481 treatment means from 143 publications was constructed, which contained at least NE and nitrogen intake (NI) data. Simple and multiple regression models were developed using the datasets containing all animals, growing or lactating buffaloes. The models that predicted faecal NE (g/d) with high precision and accuracy were faecal NE (g/d) = 13.4 (±3.30) + body weight (BW; kg) × 0.023 (±0.0083) – crude protein (CP) content of diet (g/kg) × 0.080 (±0.0211) + NI (g/d) × 0.288 (±0.0148) [RMSPE = 26.5%, with 98% of mean square prediction error (MSPE) being random error; R2 = 0.77] for all animals, and faecal NE (g/d) = 17.7 (±4.04) + BW (kg) × 0.033 (±0.0167) – ADG (kg/d) × 10.2 (±2.99) – CP (g/kg) × 0.052 (±0.026) + NI (g/d) × 0.231 (±0.0324) [RMSPE = 27.9%, with 96.4% of MSPE being random error; R2 = 0.49] for growing animals only. The best predicted regression equations for urinary NE (g/d) = −10.8 (±5.15) + BW (kg) × 0.019 (±0.0129) + CP (g/kg) × 0.056 (±0.0298) + NI (g/d) × 0.334 (±0.0206) [RMSPE = 45.5%, with 90% of MSPE from random error; R2 = 0.65] for the dataset containing all animals; and urinary NE (g/d) = 4.23 (±3.89) – BW (kg) × 0.039 (±0.0188) – ADG (kg/d) × 13.2 (±4.51) + NI (g/d) × 0.421 (±0.0313) [RMSPE = 33.2%, with 92.4% of MSPE accounting random error; R2 = 0.61] for only growing animals. In lactating buffaloes only, no models containing milk yield as a predictor were reliable for predicting NE perhaps due to paucity of studies included in the dataset. Prediction models for urinary NE had usually greater RMSPE compared with the models for faecal NE. The equations developed in the present study were found suitable for estimation of NE factors from different categories of buffaloes with different BW, ADG and feeding conditions. The models developed in the present study would be useful for preparation of global inventory of NE and separate estimation of urinary and faecal NE in buffaloes.

中文翻译：

---

使用饮食和动物变量预测水牛生产系统中的氮排泄

摘要 目前，还没有用于预测水牛的氮排泄 (NE) 的模型，而这些模型是编制 NE 清单所必需的。因此，本研究旨在开发统计模型，从动物和饮食特征变量预测 NE。构建了来自 143 篇出版物的 481 种处理方法的数据集，其中至少包含 NE 和氮摄入 (NI) 数据。使用包含所有动物、生长中或哺乳期水牛的数据集开发了简单和多元回归模型。以高精度和准确度预测粪便 NE (g/d) 的模型为粪便 NE (g/d) = 13.4 (±3.30) + 体重 (BW; kg) × 0.023 (±0.0083) – 粗蛋白 (CP)日粮含量 (g/kg) × 0.080 (±0.0211) + NI (g/d) × 0.288 (±0.0148) [RMSPE = 26.5%，其中 98% 的均方预测误差 (MSPE) 为随机误差；R2 = 0。77] 对于所有动物，粪便 NE (g/d) = 17.7 (±4.04) + BW (kg) × 0.033 (±0.0167) – ADG (kg/d) × 10.2 (±2.99) – CP (g/kg ) × 0.052 (±0.026) + NI (g/d) × 0.231 (±0.0324) [RMSPE = 27.9%，其中 96.4% 的 MSPE 是随机误差；R2 = 0.49] 仅适用于生长中的动物。尿 NE (g/d) 的最佳预测回归方程 = -10.8 (±5.15) + BW (kg) × 0.019 (±0.0129) + CP (g/kg) × 0.056 (±0.0298) + NI (g/d) ) × 0.334 (±0.0206) [RMSPE = 45.5%，其中 90% 的 MSPE 来自随机误差；R2 = 0.65] 对于包含所有动物的数据集；和尿 NE (g/d) = 4.23 (±3.89) – BW (kg) × 0.039 (±0.0188) – ADG (kg/d) × 13.2 (±4.51) + NI (g/d) × 0.421 (±0.0313) ) [RMSPE = 33.2%，其中 92.4% 的 MSPE 考虑了随机误差；R2 = 0.61] 仅适用于生长中的动物。仅在哺乳期水牛中，没有包含作为预测因子的产奶量的模型对于预测 NE 是可靠的，这可能是因为数据集中包含的研究很少。与粪便 NE 的模型相比，尿 NE 的预测模型通常具有更高的 RMSPE。发现本研究中开发的方程适用于估计具有不同 BW、ADG 和饲养条件的不同类别水牛的 NE 因子。本研究中开发的模型将有助于准备全球 NE 清单以及分别估算水牛的尿液和粪便 NE。发现本研究中开发的方程适用于估计具有不同 BW、ADG 和饲养条件的不同类别水牛的 NE 因子。本研究中开发的模型将有助于准备全球 NE 清单以及分别估算水牛的尿液和粪便 NE。发现本研究中开发的方程适用于估计具有不同 BW、ADG 和饲养条件的不同类别水牛的 NE 因子。本研究中开发的模型将有助于准备全球 NE 清单以及分别估算水牛的尿液和粪便 NE。