Behavioral patterns as indicators of resilience after parturition in dairy cows.

During the transition phase, dairy cows are susceptible to develop postpartum diseases. Cows that stay healthy or recover rapidly can be considered to be more resilient in comparison to those that develop postpartum diseases. An indication of loss of resilience will allow for early intervention with preventive and supportive measures before the onset of disease. We investigated which quantitative behavioral characteristics during the dry period could be used as indicators of reduced resilience after calving, using noninvasive Smart Tag neck and Smart Tag leg sensors in dairy cows (Nedap N.V.). We followed 180 cows during 2 wk before until 6 wk after parturition at 4 farms in the Netherlands. Serving as proxy for loss of resilience, as defined by the duration and severity of disease, a clinical assessment was performed twice weekly and blood samples were taken in the first and fifth week after parturition. For each cow, clinical and serum value deviations were aggregated into a total deficit score (TDS total). We also calculated TDS values relating to inflammation, locomotion, or metabolic problems, which were further divided into macro-mineral and liver-related deviations. Smart Tag neck and leg sensors provided continuous behavioral activity signals of which we calculated the average, variance, and autocorrelation during the dry period. Diurnal patterns in the behavioral activity signals were derived by fast Fourier transformation and the calculation of the nonperiodicity. To select significant predictors of resilience, we first performed a univariate analysis with TDS as dependent variable and the behavioral characteristics that were measured during the dry period, as potential predictors with cow as experimental unit. We included parity group as fixed effect and farm as random effect. Next, we performed multivariable analysis with only significant predictors, followed by a variable selection procedure to obtain a final linear mixed model with an optimal subset of predictors with parity group as fixed effect and farm as random effect. The TDS total was best predicted by average inactive time, nonperiodicity ruminating, nonperiodicity of bouts standing up and fast Fourier transformation stand still. Average inactive time was negatively correlated with average eating time, and these 2 predictors could be exchanged with only little difference in model performance. Our best performing model predicted TDS total at a cutoff level of 60 points, with a sensitivity of 79.5% and a specificity of 73.2% with a positive predicted value of 0.69 and a negative predicted value of 0.83. The models to predict the other TDS categories showed a lower predictive performance as compared with the TDS total model, which could be related to the limited sample size and therefore, low occurrence of problems within a specific TDS category. Furthermore, more resilient dairy cows are characterized by high averages of eating time with high regularity in rumination and low averages of inactive time. They reveal high regularity in standing time and transitions from lying to standing, in the dry period. These behaviors can be used as indicators of resilience and allow for preventive intervention during the dry period in vulnerable dairy cattle. However, further examination is still required to find clues for adequate intervention strategies.

Invited review: Big Data in precision dairy farming.

Insight into current scientific applications of Big Data in the precision dairy farming area may help us to understand the inflated expectations around Big Data. The objective of this invited review paper is to give that scientific background and determine whether Big Data has overcome the peak of inflated expectations. A conceptual model was created, and a literature search in Scopus resulted in 1442 scientific peer reviewed papers. After thorough screening on relevance and classification by the authors, 142 papers remained for further analysis. The area of precision dairy farming (with classes in the primary chain (dairy farm, feed, breed, health, food, retail, consumer) and levels for object of interest (animal, farm, network)), the Big Data-V area (with categories on Volume, Velocity, Variety and other V's) and the data analytics area (with categories in analysis methods (supervised learning, unsupervised learning, semi-supervised classification, reinforcement learning) and data characteristics (time-series, streaming, sequence, graph, spatial, multimedia)) were analysed. The animal sublevel, with 83% of the papers, exceeds the farm sublevel and network sublevel. Within the animal sublevel, topics within the dairy farm level prevailed with 58% over the health level (33%). Within the Big Data category, the Volume category was most favoured with 59% of the papers, followed by 37% of papers that included the Variety category. None of the papers included the Velocity category. Supervised learning, representing 87% of the papers, exceeds unsupervised learning (12%). Within supervised learning, 64% of the papers dealt with classification issues and exceeds the regression methods (36%). Time-series were used in 61% of the papers and were mostly dealing with animal-based farm data. Multimedia data appeared in a greater number of recent papers. Based on these results, it can be concluded that Big Data is a relevant topic of research within the precision dairy farming area, but that the full potential of Big Data in this precision dairy farming area is not utilised yet. However, the present authors expect the full potential of Big Data, within the precision dairy farming area, will be reached when multiple Big Data characteristics (Volume, Variety and other V's) and sources (animal, groups, farms and chain parts) are used simultaneously, adding value to operational and strategic decision.

Indicators of resilience during the transition period in dairy cows: A case study.

The transition period is a demanding phase in the life of dairy cows. Metabolic and infectious disorders frequently occur in the first weeks after calving. To identify cows that are less able to cope with the transition period, physiologic or behavioral signals acquired with sensors might be useful. However, it is not yet clear which signals or combination of signals and which signal properties are most informative with respect to disease severity after calving. Sensor data on activity and behavior measurements as well as rumen and ear temperature data from 22 dairy cows were collected during a period starting 2 wk before expected parturition until 6 wk after parturition. During this period, the health status of each cow was clinically scored daily. A total deficit score (TDS) was calculated based on the clinical assessment, summarizing disease length and intensity for each cow. Different sensor data properties recorded during the period before calving as well as the period after calving were tested as a predictor for TDS using univariate analysis of covariance. To select the model with the best combination of signals and signal properties, we quantified the prediction accuracy for TDS in a multivariate model. Prediction accuracy for TDS increased when sensors were combined, using static and dynamic signal properties. Statistically, the most optimal linear combination of predictors consisted of average eating time, variance of daily ear temperature, and regularity of daily behavior patterns in the dry period. Our research indicates that a combination of static and dynamic sensor data properties could be used as indicators of cow resilience.

Applicability of day-to-day variation in behavior for the automated detection of lameness in dairy cows.

Lameness is a major problem in modern dairy husbandry and has welfare implications and other negative consequences. The behavior of dairy cows is influenced by lameness. Automated lameness detection can, among other methods, be based on day-to-day variation in animal behavior. Activity sensors that measure lying time, number of lying bouts, and other parameters were used to record behavior per cow per day. The objective of this research was to develop and validate a lameness detection model based on daily activity data. Besides the activity data, milking data and data from the computerized concentrate feeders were available as input data. Locomotion scores were available as reference data. Data from up to 100 cows collected at an experimental farm during 23 mo in 2010 and 2011 were available for model development. Behavior is cow-dependent, and therefore quadratic trend models were fitted with a dynamic linear model on-line per cow for 7 activity variables and 2 other variables (milk yield per day and concentrate leftovers per day). It is assumed that lameness develops gradually; therefore, a lameness alert was given when the linear trend in 2 or more of the 9 models differed significantly from zero in a direction that corresponded with lameness symptoms. The developed model was validated during the first 4 mo of 2012 with almost 100 cows on the same farm by generating lameness alerts each week. Performance on the model validation data set was comparable with performance on the model development data set. The overall sensitivity (percentage of detected lameness cases) was 85.5% combined with specificity (percentage of nonlame cow-days that were not alerted) of 88.8%. All variables contributed to this performance. These results indicate that automated lameness detection based on day-to-day variation in behavior is a useful tool for dairy management.

Detection model for mastitis in cows milked in an automatic milking system.

Automated detection of diseases (such as mastitis) in dairy cows might be an alternative for detection by observation during milking - especially when using an automatic milking system (AMS). An outline of a detection model is given. This detection model includes time-series models for two variables (milk yield and electrical conductivity of milk), with interpolation on previous values. The model is flexible in the number of variables actually used. Parameter values and the residual variances are updated by linear regression after each milking. Alerts for mastitis are given when the residuals fall outside given confidence intervals. A data set with 111 cows for 16 months (on average, 58 lactating cows per day) was used to test the model. Depending on the chosen confidence interval, 42-44 out of 48 cases of clinical mastitis were detected; the remaining cases were not detected because not all data needed were available. These results were better than the results obtained with the model usually used on the farm. The number of false-positive alerts depended on the chosen confidence interval and was higher than the number found with the model usually used.

Application of fuzzy logic in automated cow status monitoring.

Sensors that measure yield, temperature, electrical conductivity of milk, and animal activity can be used for automated cow status monitoring. The occurrence of false-positive alerts, generated by a detection model, creates problems in practice. We used fuzzy logic to classify mastitis and estrus alerts; our objective was to reduce the number of false-positive alerts and not to change the level of detected cases of mastitis and estrus. Inputs for the fuzzy logic model were alerts from the detection model and additional information, such as the reproductive status. The output was a classification, true or false, of each alert. Only alerts that were classified true should be presented to the herd manager. Additional information was used to check whether deviating sensor measurements were caused by mastitis or estrus, or by other influences. A fuzzy logic model for the classification of mastitis alerts was tested on a data set from cows milked in an automatic milking system. All clinical cases without measurement errors were classified correctly. The number of false-positive alerts over time from a subset of 25 cows was reduced from 1266 to 64 by applying the fuzzy logic model. A fuzzy logic model for the classification of estrus alerts was tested on two data sets. The number of detected cases decreased slightly after classification, and the number of false-positive alerts decreased considerably. Classification by a fuzzy logic model proved to be very useful in increasing the applicability of automated cow status monitoring.

[Automated detection of estrus and mastitis in dairy cows]. / Automatische detectie van bronst en mastitis bij melkkoeien.

The development and test of detection models for oestrus and mastitis in dairy cows is described in a PhD thesis that was defended in Wageningen on June 5, 2000. These models were based on sensors for milk yield, milk temperature, electrical conductivity of milk, and cow activity and concentrate intake, and on combined processing of the sensor data. The models alert farmers to cows that need attention, because of possible oestrus or mastitis. A first detection model for cows, milked twice a day, was based on time series models for the sensor variables. A time series model describes the dependence between successive observations. The parameters of the time series models were fitted on-line for each cow after each milking by means of a Kalman filter, a mathematical method to estimate the state of a system on-line. The Kalman filter gives the best estimate of the current state of a system based on all preceding observations. This model was tested for 2 years on two experimental farms, and under field conditions on four farms over several years. A second detection model, for cow milked in an automatic milking system (AMS), was based on a generalization of the first model. Two data sets (one small, one large) were used for testing. The results for oestrus detection were good for both models. The results for mastitis detection were varying (in some cases good, in other cases moderate). Fuzzy logic was used to classify mastitis and oestrus alerts with both detection models, to reduce the number of false positive alerts. Fuzzy logic makes approximate reasoning possible, where statements can be partly true or false. Input for the fuzzy logic model were alerts from the detection models and additional information. The number of false positive alerts decreased considerably, while the number of detected cases remained at the same level. These models make automated detection possible in practice.