The AusBeef model for beef production: II. sensitivity analysis

Abstract

The present study evaluated the behaviour of the AusBeef model for beef production as part of a 2 × 2 study simulating performance on forage-based and concentrate-based diets from Oceania and North America for four methane (CH₄)-relevant outputs of interest. Three sensitivity analysis methods, one local and two global, were conducted. Different patterns of sensitivity were observed between forage-based and concentrate-based diets, but patterns were consistent within diet types. For the local analysis, 36, 196, 47 and 8 out of 305 model parameters had normalized sensitivities of 0, >0, >0·01 and >0·1 across all diets and outputs, respectively. No parameters had a normalized local sensitivity >1 across all diets and outputs. However, daily CH₄ production had the greatest number of parameters with normalized local sensitivities >1 for each individual diet. Parameters that were highly sensitive for global and local analyses across the range of diets and outputs examined included terms involved in microbial growth, volatile fatty acid (VFA) yields, maximum absorption rates and their inhibition due to pH effects and particle exit rates. Global sensitivity analysis I showed the high sensitivity of forage-based diets to lipid entering the rumen, which may be a result of the use of a feedlot-optimized model to represent high-forage diets and warrants further investigation. Global sensitivity analysis II showed that when all parameter values were simultaneously varied within ±10% of initial value, >96% of output values were within ±20% of the baseline, which decreased to >50% when parameter value boundaries were expanded to ±25% of their original values, giving a range for robustness of model outputs with regards to potential different ‘true’ parameter values. There were output-specific differences in sensitivity, where outputs that had greater maximum local sensitivities displayed greater degrees of non-linear interaction in global sensitivity analysis I and less variance in output values for global sensitivity analysis II. For outputs with less interaction, such as the acetate : propionate ratio and microbial protein production, the single most sensitive term in global sensitivity analysis I contributed more to the overall total-order sensitivity than for outputs with more interaction, with an average of 49, 33, 15 and 14% of total-order sensitivity for microbial protein production, acetate : propionate ratio, CH₄ production and energy from absorbed VFAs, respectively. Future studies should include data collection for highly sensitive parameters reported in the present study to improve overall model accuracy.