MODEL EVALUATION OF METHANE EMISSION FROM GOATS

Abstract Text: There have been several attempts to develop mathematical models to predict methane (CH₄) emissions because they can be easily applied in practical situations to estimate diet ME. Because studies evaluating the suitability of different mathematical models available for methane emission prediction in goats are lacking, this study was carried out to evaluate 3 empirical models (Blaxter and Clapperton, Moe and Tyrrel, Pelchen and Peters) based on their ease of application, common use and their feasible input variables. The prediction ability of those models was evaluated using a database of two metabolism trials, in which 45 individual measurement of methane emission from goats (averaging 30±2.93 kg BW) was taken using SF₆ technique. Models were evaluated by regressing residual (observed minus predicted) values on the predicted values centered on their mean values. The intercepts of the regression equations were used to estimate mean biases, whereas linear biases were assessed using the slopes of the regression equations. Also, using the same database, an empirical model was developed taking into account the coefficient of determination (R²), forward selection, and Cp of Mallows.  Observed DMI ranged from 280 to 1300 g/d. Observed methane emission ranged from 8 to 22 g/d, which represented methane losses ranging from 3.5 to 11% of dietary GE intake. Results showed a significant mean and linear biases (P<0.001) for all models, showing that these models over predict methane emissions from goats. Moe and Tyrrell model presented the highest linear bias which was -65.5 g/d at the maximum predicted value (86.9 g CH₄/d). Pelchen and Peters model exhibited the lowest magnitude of the linear bias, which was less than 2 g/d at the minimum (11.5 CH₄/d) and -10 g/d at the maximum (27 g CH₄/d) predicted methane values. The linear bias of Blaxter and Clapperton model ranged from approximately 4 g/d (at the minimum predicted values) to -15 g/d (at the maximum predicted values). A simple regression equation was developed using same database. Accordingly, the best-fit linear model that represents goat methane emission was defined as follows (R² = 0.51, RMSE = 2.23, P<0.001, Cp = 11.4): CH₄(g/d) = 4.36±2.07 + 0.17±0.06 × BW (kg) + 0.006±0.001 × DMI (g/d).

Keywords: bias, empirical model, CH₄