Veterinarians make herd-level decisions with imperfect information. A reproductive program may affect the herd for years. A nutrition change can alter production as well as manure and nutrient flows. A heat-stress intervention may look different depending on the farm, the animals and the weather.
What if some of those decisions could be tested before they were implemented?
That is the long-term vision behind RuFaS, the Ruminant Farm Systems model, an open-source whole-farm simulation platform developed by University of Wisconsin-Madison researchers and recent topic of discussion on a Hoard’s Dairyman webinar. The model is designed to connect the major components of a dairy operation, including animals, manure, soil and crops, and feed storage, allowing researchers to simulate how changes in one area can affect the rest of the farm.
Ultimately, however, the goal is bigger than simulation.
“The long-term vision is that we will be able to connect data back and forth from real farms with the model. So we have a version of the real model in the field and we have a virtual version on the side that can share data back and forth so we have good estimates and we can do very important what-if scenarios and research before it’s done in the field,” says Victor Cabrera, professor and extension specialist at the University of Wisconsin-Madison.
That concept is known as a digital twin: a virtual representation of a real farm that is continually informed by data from the operation itself.
Moving Beyond the Average Herd
One of the most interesting aspects of RuFaS is its ability to represent variation among individual animals.
Rather than simply modeling an “average cow,” RuFaS simulates individual animals with differences in performance and requirements. That creates an opportunity to examine how decisions could affect not just a theoretical herd, but the distribution of animals within it.
That distinction could become increasingly important as more farms collect individual-animal data through activity monitors, milk systems, weighing equipment and other technologies.
The current model can work from farm-level information and simulated animal distributions, but Cabrera envisions eventually incorporating actual individual-animal records. In that scenario, a virtual farm could begin with the same animals as the real operation and continually compare predicted outcomes with what is actually happening.
That creates the possibility of moving from a static model to a feedback loop:
Farm Data → Digital Twin → Prediction → Real-World Putcome → Recalibration
Testing Management Decisions Virtually
RuFaS has already been used to examine reproductive management, including the interaction between heifer and cow reproductive programs. The model can accommodate approaches such as estrus detection, synchronization and activity-monitoring technologies, allowing researchers to examine how decisions made earlier in an animal’s life could affect later herd outcomes.
Heat stress is another potential application.
RuFaS can incorporate environmental data including temperature, relative humidity, precipitation and solar radiation. Cabrera said those data can be used to simulate heat stress and its potential effects on animal performance:
“We will need daily data. And once we have those data we can use it to simulate the heat stress of the animals, and how that would potentially impact the performance of the animal.”
For a veterinarian, the value of that approach isn’t necessarily predicting exactly what will happen. It is creating another way to evaluate possible outcomes before making a change.
A veterinarian could eventually ask questions such as: What happens if this reproductive protocol changes? What happens if heat abatement is increased? What happens if a different feeding strategy is implemented? Which outcomes improve, and which may deteriorate elsewhere in the system?
That whole-farm perspective is central to RuFaS. The model is designed so information moves between its different components, allowing an intervention in one area to produce simulated downstream effects in others.
The Challenge of Making a Digital Twin Useful
The concept comes with a significant challenge: data.
A highly detailed RuFaS simulation can involve roughly 1,200 inputs per farm and more than 12,000 outputs. That creates questions about data availability, accuracy, standardization and validation.
Not every output can realistically be compared with a real-world measurement. Instead, Cabrera says the team expects to identify key performance indicators that can be used to evaluate how closely the model tracks the actual farm.
That validation process will be critical if digital twins eventually move from research environments into routine farm decision-making.
There is also a deliberate effort to keep RuFaS interpretable. Although the team is exploring ways AI could help optimize scenarios, interact with the model and assist with development, AI is not currently driving the underlying simulation. The researchers want the equations and functions behind the outputs to remain transparent.
That may ultimately be one of the most important characteristics of a veterinary decision-support system.
A digital twin won’t replace the veterinarian standing in the barn, examining cows and interpreting what is happening on a particular operation. Instead, it could give that veterinarian something they currently have limited ability to access: a way to explore potential futures for the herd before committing to one.
