Browsing by Person "Hart, Leonie Sophia"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Precision grazing: investigating the potential and limits of smart tools and algorithms to support pasture allocation decisions in dairy cattle farming
    (2022) Hart, Leonie Sophia; Dickhöfer, Uta
    Grassland as a feed resource for grazing cattle has untapped potential for environmentally sustainable and economically viable dairy production. Implementing automated tools can increase resource use efficiency of feeding systems that include grazing, especially because the spatiotemporal variability of pastures is high. The relevant tools estimate available herbage mass, its nutritional quality and the current state of an animal’s herbage intake on pasture in nearly real time and allow farmers make informed pasture allocation decisions by interpreting and visualizing sensor data. However, the applied tools and associated algorithms must be accurate and as time efficient as possible. In the present thesis, three studies investigated these requirements under on-farm conditions on temperate permanent grasslands. The aim of the first study was to evaluate existing prediction models that determine herbage scarcity on pasture based on changes in dairy cow behavior. A second study evaluated the performance of three tools in determining fresh herbage mass and nutritional quality of multi-species grasslands: an analysis model for multispectral imagery (MSI), a portable on-site near-infrared reflectance spectroscopy instrument (NIRS) and visual observations done by farmers who used look-up tables. A camera mounted to an unmanned aerial vehicle (UAV) performed MSI. In the third study, the labor input of all tools and a forth tool—a semi-automated rising plate meter (RPM)—was modeled for an exemplary dairy farm to compare the time-efficiency between these work procedures and to identify optimization potential in their workflows. Although some of the investigated approaches were commercially available and ready-to-use, their accuracy and applicability have not been scientifically evaluated under farming conditions. In the first study, a grazing study, twenty lactating Brown Swiss cows grazed in two groups and half-days in six-day grazing cycles where the allocated grazing areas were restricted. Behaviors of individual cows were recorded by using noseband sensors and pedometers that monitored the head, jaw, and leg activity. A generalized linear model and a random forest model predicted daily herbage availability on pasture based on eight feeding behaviors. The predictions (binary classifications) were compared to two individual-animal and day-specific reference indicators for feed scarcity: reduced milk yields and rumen fill scores that undercut normal variation. It was found that the predictive performance of the models in identifying herbage scarcity was low. However, two out of eight behavioral variables were confirmed as suitable predictors: “daily rumination chews” and “bite frequency,” the latter being particularly sensitive when new feed allocation is present in the grazing setup within 24 h. Several steps for further development are needed before the predictions can be applied on commercial farms. In the second study, a grassland study, herbage mass and nutritional quality were estimated using the three tools mentioned above, but NIRS was combined with destructive sampling to determine herbage mass. Surveys were undertaken on 18 multi species grasslands located on six farms in Switzerland throughout the vegetation period in 2018. The sampled plots represented two phenological growth stages. The estimates of the tools were compared with laboratory measurements as a reference. The tools showed unsatisfactory performance for use on multi-species grasslands. Environmental characteristics of the 18 studied plots presented a great challenge for the MSI-based model. An increase in standing herbage mass on the plots was related to the higher error in estimating herbage mass and crude protein, respectively. The on-site NIRS determined fresh herbage quality with a systematic and correctable error. After corrections, its performance (relative prediction error ≤7.7 %) was better than that of the visual observation using look-up tables. In the third study, the labor study, a modeling approach that included time measurements based on video footage was used to study the workflow and the labor input of each tool. The labor time requirement was modeled when tools were applied in different farm sizes (i.e., 6–100 ha) and paddock setups (i.e., 4–45 paddocks on a given pasture area). In addition, we studied the optimization potential in workflows by identifying work elements where time can be saved by outsourcing or replacing them. The labor time requirement differed between the RPM, the on-site NIRS and the UAV (i.e., MSI-based tool) depending on the farm size and paddock setup (0.7–5.9 h per operation). It increased for all tools with an increase in farm size, but labor time requirement was lowest for the RPM. For the UAV, it did not increase noticeably when the division of the grazing area changed. Nevertheless, the potential to save time was identified for the UAV and the NIRS (34.4 min and 10.6 min, respectively). Further development of the behavior- and MSI-based approaches is needed, including the variable selection and expansion (i.e., which cow behaviors and which spectral indices are most relevant; which additional data can improve the prediction). In conclusion, the on-site NIRS has great potential for decision support on pasture allocation and supplementary feeding when looking at its measurement accuracy. However, instead of destructive sampling to determine herbage mass, using an alternative approach such as RPM would contribute to a more sociotechnologically sustainable grazing management using the NIRS.