Feeding behaviour can be used as an important indicator to support animal management. However, using feeding behaviour as a tool for dairy cow management an automatic sensor system is needed. Hence, the objective of this study was to setup, test and validate a ultra-high frequency (UHF) radio-frequency identification (RFID) system for measuring time dairy cows spent at the feed fence using two types of passive UHF ear tags. In a first experiment, the reading area of the system was evaluated in two antenna positions. Subsequently, the UHF RFID system was validated with video observations and compared to the measurements of chewing time of a noseband pressure sensor and of the time spent at the feed fence registered by a sensor system with real-time localisation. Differences in the reading area were detected between the two antenna positions and types of ear tag. The antenna position leading to less false positive registrations was chosen for the experiment with cows. The validation with video data showed a high average sensitivity (93.7 ± 5.6%, mean ± standard deviation), specificity (97.8 ± 1.1%), precision (93.8 ± 2.3%) and accuracy (96.9 ± 0.9%) of the UHF RFID system for measuring the time spent at the feed fence. The comparison with the noseband pressure sensor and the real-time localisation resulted in high correlations with a correlation coefficient of r = 0.95 and r = 0.93, respectively. However, substantial absolute differences between the three systems pointed out differences between direct and indirect measures of feeding behaviour in general and between the different sensors in particular. Thus, detailed considerations are necessary before interpreting automatically measured feeding data generally.

The analysis of multispectral imagery (MSI) acquired by unmanned aerial vehicles (UAVs) and mobile near-infrared reflectance spectroscopy (NIRS) used on-site has become increasingly promising for timely assessments of grassland to support farm management. However, a major challenge of these methods is their calibration, given the large spatiotemporal variability of grassland. This study evaluated the performance of two smart farming tools in determining fresh herbage mass and grass quality (dry matter, crude protein, and structural carbohydrates): an analysis model for MSI (GrassQ) and a portable on-site NIRS (HarvestLab^TM 3000). We compared them to conventional look-up tables used by farmers. 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, corresponding to an age of two weeks and four to six weeks, respectively. We found that neither the performance of the smart farming tools nor the performance of the conventional approach were satisfactory for use on multi-species grasslands. The MSI-model performed poorly, with relative errors of 99.7% and 33.2% of the laboratory analyses for herbage mass and crude protein, respectively. The errors of the MSI-model were indicated to be mainly caused by grassland and environmental characteristics that differ from the relatively narrow Irish calibration dataset. The On-site NIRS showed comparable performance to the conventional Look-up Tables in determining crude protein and structural carbohydrates (error ≤ 22.2%). However, we identified that the On-site NIRS determined undried herbage quality with a systematic and correctable error. After corrections, its performance was better than the conventional approach, indicating a great potential of the On-site NIRS for decision support on grazing and harvest scheduling.

Sensor technologies that measure grazing and ruminating behaviour as well as physical activities of individual cows are intended to be included in precision pasture management. One of the advantages of sensor data is they can be analysed to support farmers in many decision-making processes. This article thus considers the performance of a set of RumiWatchSystem recorded variables in the prediction of insufficient herbage allowance for spring calving dairy cows. Several commonly used models in machine learning (ML) were applied to the binary classification problem, i.e., sufficient or insufficient herbage allowance, and the predictive performance was compared based on the classification evaluation metrics. Most of the ML models and generalised linear model (GLM) performed similarly in leave-out-one-animal (LOOA) approach to validation studies. However, cross validation (CV) studies, where a portion of features in the test and training data resulted from the same cows, revealed that support vector machine (SVM), random forest (RF) and extreme gradient boosting (XGBoost) performed relatively better than other candidate models. In general, these ML models attained 88% AUC (area under receiver operating characteristic curve) and around 80% sensitivity, specificity, accuracy, precision and F-score. This study further identified that number of rumination chews per day and grazing bites per minute were the most important predictors and examined the marginal effects of the variables on model prediction towards a decision support system.