Can we do better than the weatherman?
By Billy Brown, Malia Caputo, and Heather White Feed is the primary expense on dairy farms, so identifying cows that are more feed efficient can pay dividends for dairy farmers and improve sustainability of dairy production. Determination of feed efficiency for individual cows requires feed intake measurement capabilities; however, this is not possible in group-housed dairy cows today. In the absence of having individual intakes, how could we identify cows that are feed efficient (or inefficient) using markers that could easily be accessed from existing data streams on dairy farms? If we could determine individual cow intakes, could we manage cows differently based on their feed efficiency? Are there other benefits to knowing a cow’s feed intake, such as predicting disease or estrus? Feed intake prediction models currently are not precise enough to predict intake for individual cows, but function well to formulate rations for groups of cattle. These models utilize basic cow descriptive factors, such as milk production and components, body weight, stage of lactation, breed, and parity. There has been a recent effort to strengthen feed intake prediction models with new data sources that are becoming available on dairy farms, and predicting feed intake accurately on the cow-level will require incorporation of novel data streams. A multitude of possibilities exist (Figure 1).
Metabolism and smart technologies Technological advancements in the dairy industry have given dairy farmers access to a variety of precision management technologies, including activity monitoring systems and real-time blood analysis devices. These technologies generate information that could be considered as predictor variables. A recent publication from the University of Wisconsin-Madison looked at the value of adding sensor-derived behavior variables and blood metabolites as novel data streams to traditional feed intake prediction models (Martin et al., 2021; J. Dairy Sci. 104:8765-8782). In addition to collecting blood samples, a commercial ear tag sensor system (SMARTBOW; Zoetis) was used to measure activity, rumination, lying time, and location. The researchers used a sequential approach to adding different predictor variables to their models based upon the ease of obtaining each predictor type. Adding the sensor-derived behavioral variables to the traditional predictor variables explained additional 2% of the variation in intake. While nominal, this improvement highlights a unique contribution that alternate data streams have on explaining variation in intake. However, further addition of blood metabolites that are relevant to body energy status did not improve precision nor accuracy of the predictions. Overall, these researchers were predicting feed intake with 82% precision.
Milk fatty acids and predicted transmitting abilities On-farm DHIA milk testing programs may be another source of novel predictor variables. Milk fatty acid analysis is growing in popularity and availability through milk testing organizations. Milk fatty acids are indicative of nutritional and metabolic status and may present a snapshot of the cow’s feed intake patterns. Brown et al. (2022, as recently accepted in JDS) tackled their usefulness as predictor variables in a study using data from 350 cows at the University of Wisconsin-Madison dairy herd. Excitingly, new variables like preformed and de novo fatty acids improved prediction of feed intake over traditional predictor variables by 4 to 8%. Similar results were obtained by other researchers using milk mid-infrared spectroscopy wavelengths (Dórea et al., 2018; J. Dairy Sci. 104:8765–8782). Perhaps one of the most easily accessible candidate predictor variables for feed intake prediction models has been overlooked entirely until this year. Predicted transmitting abilities (PTA) are a measure of the animal’s genetic capability. While traditional PTA for milk production and components have been available for decades, recent work has enabled the development of new PTA related to feed efficiency – notably, PTA for residual feed intake and body weight composite. When offering these PTA for inclusion in prediction models, the PTA for milk and residual feed intake were routinely retained in the final models (Brown et al., 2022) and increased precision of prediction by 3 to 12% over traditional predictor variables. This is an exciting development, and validates the previous work conducted on research farms (where individual feed intake can be determined) to evaluate the genetic aspect of feed efficiency.
When combined, traditional predictor variables in addition to milk fatty acids and PTA in feed intake prediction models explained 67% of the variation in daily feed intake (Brown et al., 2022). The unique aspect about the models developed by these researchers is that they were derived from data on a single day rather than the normal approach of averaging data over a period of weeks or months. This is promising, because it suggests we can accurately predict individual cow feed intake at a single moment in time without having to collect animal data for days or weeks on end. We could theoretically determine feed intake for cows on a routine basis on the farm. Future approaches and considerations As a part of their efforts, the team at the University of Wisconsin-Madison created a suite of models based on different combinations of data sources, which allows the farmer to use the model that best reflects data streams available on their farm (Table 1). For example, if the farmer in a robotic herd has access to body weights and routinely assigns body condition scores, but the DHIA does not include milk fatty acid testing, they could use model 5 based on those needs (Table 1).
The fact that a farmer could predict feed intake on the dairy farm could open a world of possibilities for management aspects on the farm. Could it be used to more accurately inform dairy nutritionists to estimate feed intake when balancing a ration for a group of cows? If a pen of cows is consuming more feed than can fit in one mixer wagon, can we adjust grouping by moving a few cows that are eating the most to another pen so we can ensure feed is mixed appropriately? Could this help predict disease in cows that are not yet showing clinical signs but have unknowingly started to drop in feed intake? Clearly, one of the greatest opportunities is to identify the most feed efficient cows and make management and breeding decisions based on those metrics. The potential of these models to be used is only limited by the creativity of industry.
More work is needed to improve the accuracy of predicting feed intake. The use of smart technologies, milk fatty acids, and PTA as predictor variables indicates there is probably a wealth of data available to us today or in the near future that could prove useful for predicting feed intake. Overall, it appears that continued integration of data streams on dairy farms may help to better monitor and manage our cows in the future. REFERENCES Brown, W.E., M.J. Martin, C. Siberski, J.E. Koltes, F. Peñagaricano, K.A. Weigel, and H.M. White. 2022. Predicting feed intake using point-in-time data streams readily available on dairy farms. J. Dairy Sci. Accepted. Dórea, J. R. R., G. J. M. Rosa, K. A. Weld, and L. E. Armentano. 2018. Mining data from milk infrared spectroscopy to improve feed intake predictions in lactating dairy cows. J. Dairy Sci. 101:5878–5889. doi:10.3168/jds.2017-13997. Available from: http://dx.doi.org/10.3168/jds.2017-13997 Martin, M. J., J. R. R. Dórea, M. R. Borchers, R. L. Wallace, S. J. Bertics, S. K. DeNise, K. A. Weigel, and H. M. White. 2021. Comparison of methods to predict feed intake and residual feed intake using behavioral and metabolite data in addition to classical performance variables. J. Dairy Sci. 104:8765–8782. doi:10.3168/jds.2020-20051.
The authors are, respectively, a postdoctoral research associate, PhD candidate, and associate professor at the University of Wisconsin-Madison. Since submitting this manuscript, Dr. Brown has transitioned to assistant professor at Kansas State University.
