Current Progressive Dairy digital edition

Challenging nutrition models

Tamara Scully for Progressive Dairyman Published on 05 May 2017

Nutrition models often don’t result in the performance expected when they are put into use on the dairy, said Jim Tully of Pine Creek Nutrition Services during the Penn State 2016 Dairy Nutrition Workshop.

Those perfect models on paper don’t always work in practice. When the model isn’t working, there must be a reason. Models have limitations and, often, the limiting factors are a function of management.



Dr. Robert Fry of Atlantic Dairy Consulting joined Tully for a discussion on the use of nutrition models and the real-life reasons why performance doesn’t always match expectations.

Limiting factors

As the vast majority of today’s dairy cow rations are computer-generated using nutrition models, more complex nutritional calculations can occur.

Nutrition models only work when all of the inputs are correct, Fry said. Numerous factors impact the outcome, including the cow, the feeder, forage moisture adjustments, knowledge of the model platform, accurate animal inputs, forage analysis details and ration formulation. With all of the variables involved, the model’s predicted outcome is not necessarily seen on the dairy.

Models such as the Cornell Net Carbohydrate and Protein System (CNCPS) nutritional computer model, account for environmental inputs, animal inputs and feed inputs with over 100 possibilities built into the model. While this allows for precision, it also increases the risk of inaccurate inputs.

If the nutrition model accounts for a change in body condition scoring over a given period, such as in the fresh cow pen, but animals are kept in that pen longer than usual and gain back body condition, the inputs to the model are no longer accurate. As a result, milk production changes might occur that aren’t predicted based on the model’s calculations.


“Output is always relative to inputs,” Fry said.

On the management side of things, overcrowding, ventilation and overall cow comfort are other issues that need to be addressed in order to maximize milk and achieve model predictions. Environmental conditions will affect the model.

If the weather is hazy, hot and humid, for example, “It will slam cows,” Fry said.

Changes in a cow’s maintenance requirements to combat heat stress results in less milk output. Cows won’t eat as much, dry matter intake (DMI) drops, and body condition will decrease as they live off body fat. The rumen rate of passage changes, with an increase in metabolizable energy and a decrease in metabolizable protein. Milk production numbers will change from model predictions.

Metabolizable energy and metabolizable protein “are extremely important” in the CNCPS and come from three places: diet formulation, dry matter intake and body condition scoring changes that may be occurring, Fry said. You must take into account the body condition scoring change you expect to happen for the model to be accurate.

“Energy is usually the eventual limiting factor to milk production,” he said.


DMI has to be accurate, and the actual intake – not the amount fed – must be inputted to the model. If only 96 percent of what should be fed is actually being fed, go back and recalculate the model.

“At the right DMI, milkfat and protein, the model is working,” Fry said. “You get all the inputs right, it works. The model does work if we work it right. It’s the management side of things.”

Managing forage

“The analysis side is a huge factor in diet formulation,” Tully said. “But what is the cutoff, on dollars spent or rate of return?”

Sampling silage at least twice per week, checking the moisture levels, is important. If the dry matter content changes slightly over several days, production will change. Moisture testing can be done easily in the microwave.

“Dry matter is just huge,” Tully said. “The next pound of dry matter fixes a lot.”

Forage samples from every new silo or new cutting should be tested, three times quickly, and averaged. Then, Fry recommended sampling once per month to ensure accuracy in DMI.

“An average of a whole bunch of samples is closer to being correct than any one sample we take,” Fry cautioned. “High-inclusion-rate items get greatest priority (for analysis), especially if it’s a forage.”

Tully recommended producers focus on increasing feed efficiency for the most profitable approach on the dairy.

Working with your feeders via training, monitoring and even monetary rewards can help reduce feeding errors, inefficiency and inconsistency.

“Track what you bought and what you are supposed to feed in order to find the shrink in commodity bags,” he said. Three to 4 percent shrink may be OK, but always work to improve.

Byproduct variability, which can be significant, has to be managed. Feeding a larger number of byproducts might be warranted to reduce the impact of any single ingredient.

“Control what you can control. Getting a handle on the variation is one of those high-priority options because the cow hates it,” Tully said. “We’ve got to stop variation. Consistency matters.”

Working with the model

“The model ... is just not the first thing I go to,” Tully said.

When working with one California Jersey herd, protein in the diet was increased and “the model worked. We were really surprised at how good it worked. But it didn’t pay,” he said.

Based on ingredient pricing, feeding the additional protein, which made the model work as predicted, didn’t increase profit.

In another example, product “failure” caused a situation where the model didn’t work. As we try to “dial in” precise nutrients, problems often arise. An on-farm trial for a new feed product did not match what the model predicted. In vitro and in situ analysis is not the same as real life, and the product did not perform on the farm as it did in the lab, Tully said.

“It’s not the model’s fault,” as many products don’t perform in practice as they do in theory, he said. “Product formulation is a big factor. It’s the input side. Book values are exactly that.”

Nutrition models can precisely predict performance when inputs are accurate. But the model can’t account for inaccurate inputs or management issues that aren’t factored into the equation. Models can predict how changes will impact production, but it’s not possible to take into account every variable.

“The cow is a big factor,” Tully said.  end mark

Tamara Scully, a freelance writer based in northwestern New Jersey, specializes in agricultural and food system topics.