Mathematical ruminant nutrition models can be used to integrate our knowledge of feed, intake, and digestion and passage rates upon feed energy values, escape of dietary protein, and microbial growth efficiency. They can be valuable tools for estimating animal requirements and nutrients derived from feeds in each unique farm production scenario, and thus can have an important role in providing information that can be used in the decision-making process to enhance the feeding system (Tedeschi et al., 2005b). By accounting for farm-specific animal, feed, and environmental characteristics, more accurate prediction of dietary nutrient requirements for maintenance, growth and milk production of cattle and nutrient excretion in diverse production situations is possible (Fox et al., 2004).

In the United States, livestock farms are under increasing pressure to reduce nutrient accumulation on the farm and manure nutrient excretions in order to meet environmental regulations (Fox et al., 2006). The Natural Resources Conservation Service (NRCS), an office of the United States Department of Agriculture (USDA), has identified the need to improve feed management in concentrated animal feeding operations (CAFO) to reduce manure nutrients. The USDA-NRCS has developed a national conservation practice standard for feed management (#592; USDA-NRCS, 2003) to be used as part of the nutrient management (#590; USDA-NRCS, 2006) planning process. The purpose of a feed management plan is (1) to supply the quantity of available nutrients required by livestock while reducing the quantity of nutrients excreted, and (2) to improve net farm income by feeding nutrients more efficiently.

The development of feeding and nutrient management plans is complex and requires the integration of a large amount of research and knowledge information. Therefore, mathematical nutrition models can be used to assist in the deployment of technology that meets governmental regulations by facilitating the application and development of site-specific plans. Furthermore, mechanistic models more accurately account for animal and crop requirements, and manure and soil management than fixed, tabular guidelines because they can be customized and calibrated for site-specific characteristics and recommendations (Tedeschi et al., 2005a; Tedeschi et al., 2005b).

The identification of cattle requirements and formulating diets to meet those requirements more accurately is the best current strategy to minimize nutrient output per kg of meat or milk produced. The terms precision feeding and phase feeding have been widely used to describe nutrient management practices that result in reduced excretion of nutrients by CAFO. Both terms refer to a more precise nutrition system, where nutritionists meet cattle nutritional needs without supplying nutrients in excess, reducing outputs and inputs. Phase feeding of protein or protein withdrawal is a systematic method that applies precision feeding concepts to different phases of animal growth to accurately meet their nutrient requirements during the feeding period. Phase feeding involves formulating and providing more specific rations during growth-specific periods as the animal matures (Vasconcelos et al., 2007).

Mathematical models of ruminant nutrition have been employed for over three decades (Chalupa and Boston 2003) and have stimulated improvements in feeding cattle. More complete data sets available in recent years combined with different mathematical approaches have allowed us to improve nutrition models. Several mathematical models of ruminant nutrition have been develop in the past (Tedeschi et al. 2005b) and it is likely that frequency of use will increase to support decision making not only in the nutrition of cattle, but also for other aspects including farm economics, animal management, and assessment of environmental impact (Tylutki et al. 2004).

The development and application of mathematical models are essential in several branches of the scientific research domain. Notably, predictive models are used to estimate the outcomes of experiments that cannot be practically (or ethically) conducted, directly measured, are cost prohibitive, or simply because there is plenty of available data and the collection of new data is neither justifiable nor acceptable. Even though, models are generally accepted by the scientific community, the identification of their adequacy for predictive purposes is extremely important in building confidence and acceptance of the predictions in broader situations.

The need to evaluate the correctness of model predictions has been widely discussed and several techniques have been proposed (Easterling and Berger, 2002; Hamilton, 1991; Tedeschi, 2006). Nonetheless, most evaluations are superficial and provide little or no information regarding the ability of a model in predicting future outcomes. This can be partially explained because most mathematical models are designed to be static, deterministic, and range-dependent, implying that there is a range of optimum predictive ability and often they have a narrower and site-specific application rather than a broader one. A second reason is related to the difficult in assessing the suitability of mathematical models due to the intrinsic unaccounted for variation of the database; thus, affecting the results of the evaluation process. A thorough and unbiased evaluation of a model is a requisite not only to build confidence in the model’s predictions, but also in designing more resilient models. Lastly, a third reason lies in using the evaluation process to prove the rightness and robustness of a mathematical model or even to promote its acceptance and usability by others (Sterman, 2002).