Eckhard Stamer

Automation of oestrus detection in dairy cows: a review

Abstract The economic importance of traits like longevity, health and reproduction has increased compared to milk yield in dairy cows. Effective oestrus detection is important for improved reproduction. Commonly, oestrus detection is performed by visual observation, but this is particularly difficult on large dairy farms because of short observation periods during feeding and milking. As a result of technical progress in monitoring cows using computers, automatic oestrus detection has become possible. In many studies different traits have been analysed for utilisation in automatic oestrus detection. The best results were found for detection using pedometers. Results of oestrus detection varied depending on the used threshold value, the number of cows, housing and treatment of cows and the utilised method of time series analysis. The detection rate of most investigations is sufficiently high at 80–90%. Error rates between 17 and 55% and specificities between 96 and 98% indicate a large number of false positive oestrus warnings. The main problem of automatic oestrus detection is to reduce the false positive alerts. In recent years several authors have combined different traits with the objective of improving detection rates. Best multivariate analyses results were found for combinations with activity. Further research should be performed using data from a commercial dairy farm. A comparison of different time series methods and multivariate analysis of traits would be useful.

Evaluation of five lactation curve models fitted for fat:protein ratio of milk and daily energy balance

Selection for milk yield increases the metabolic load of dairy cows. The fat:protein ratio of milk (FPR) could serve as a measure of the energy balance status and might be used as a selection criterion to improve metabolic stability. The fit of different fixed and random regression models describing FPR and daily energy balance was tested to establish appropriate models for further genetic analyses. In addition, the relationship between both traits was evaluated for the best fitting model. Data were collected on a dairy research farm running a bull dam performance test. Energy balance was calculated using information on milk yield, feed intake per day, and live weight. Weekly FPR measurements were available. Three data sets were created containing records of 577 primiparous cows with observations from lactation d 11 to 180 as well as records of 613 primiparous cows and 96 multiparous cows with observations from lactation d 11 to 305. Five well-established parametric functions of days in milk (Ali and Schaeffer, Guo and Swalve, Wilmink, Legendre polynomials of third and fourth degree) were chosen for modeling the lactation curves. Evaluation of goodness of fit was based on the corrected Akaike information criterion, the Bayesian information criterion, correlation between the real observation and the estimated value, and on inspection of the residuals plotted against days in milk. The best model was chosen for estimation of correlations between both traits at different lactation stages. Random regression models were superior compared with the fixed regression models. In general, the Ali and Schaeffer function appeared most suitable for modeling both the fixed and the random regression part of the mixed model. The FPR is greatest in the initial lactation period when energy deficit is most pronounced. Energy balance stabilizes at the same point as the decrease in FPR stops. The inverted patterns indicate a causal relationship between the 2 traits. A common pattern was also observed for repeatabilities of both traits, with repeatabilities being largest at the beginning of lactation. Additionally, correlations between cow effects were closest at the beginning of lactation (r(c)=-0.43). The results support the hypothesis that FPR can serve as a suitable indicator for energy status, at least during the most metabolically stressful stage of lactation.

Short communication: Genetic relationships among daily energy balance, feed intake, body condition score, and fat to protein ratio of milk in dairy cows

Postpartum energy status is critically important to health and fertility, and it remains a major task to find suitable indicator traits for energy balance. Therefore, genetic parameters for daily energy balance (EB) and dry matter intake (DMI), weekly milk fat to protein ratio (FPR), and monthly body condition score (BCS) were estimated using random regression on data collected from 682 Holstein-Friesian primiparous cows recorded between lactation d 11 to 180. Average energy-corrected milk (ECM), EB, DMI, BCS, and FPR were 32.0 kg, 9.6 MJ of NE(L), 20.3 kg, 2.95, and 1.12, respectively. Heritability estimates for EB, DMI, BCS, and FPR ranged from 0.03 to 0.13, 0.04 to 0.19, 0.34 to 0.59, and 0.20 to 0.54. Fat to protein ratio was a more valid measure for EB in early lactation than DMI, BCS, or single milk components. Correlations between FPR and EB were highest at the beginning of lactation [genetic correlation (r(g)) = -0.62 at days in milk (DIM) 15] and decreased toward zero. Dry matter intake was the trait most closely correlated with EB in mid lactation (r(g) = 0.73 at DIM 120 and 150). Energy balance in early lactation was negatively correlated to EB in mid lactation. The same applied to DMI. Genetic correlations between FPR across lactation stages were all positive; the lowest genetic correlation (0.55) was estimated between the beginning of lactation and early mid lactation. Hence, to improve EB at the beginning of lactation, EB and indicator traits need to be recorded in early lactation. We concluded that FPR is an adequate indicator for EB during the energy deficit phase. Genetic correlations of FPR with ECM, fat percentage, and protein percentage showed that a reduction of FPR in early lactation would have a slightly negative effect on ECM, whereas milk composition would change in a desirable manner.

Improving oestrus detection by combination of activity measurements with information about previous oestrus cases

Abstract In this study, the potential benefit of combining the traits activity and period since last oestrus for oestrus detection was investigated. The simultaneous analyses of these traits in a fuzzy logic model should reflect the consideration of the dairy farmer regarding his judging of oestrus warnings. The analyses involved 862 cows, each with one verified oestrus case. Information about previous oestrus or inseminations were available for 373 cows. For comparison only, the trait activity was analysed in a preliminary investigation by an univariate fuzzy logic model. The sensitivity was 91.7% and the error rate was 34.6%. By considering the previous information in a multivariate fuzzy logic model, the sensitivity decreased to 87.9% and the error rate improved to 12.5%. The simultaneous analysis of cows with and without previous information in the oestrus detection model resulted in an increase in error rate to 23.8%, due to the high number of cows without previous information. The obtained results indicate that the information about previous oestrus cases is suitable for multivariate oestrus detection.

Genetic analysis of mastitis data with different models

The aim of this study was to analyze different mastitis data sets with different statistical models and compare results. Data recording took place on 3 commercial milk farms with an average herd size of 3,200 German Holstein cows. Recording started in February 1998 and was completed in December 2005. During this period, 63,540 treatments for clinical mastitis were recorded. Five different data sets were analyzed and the number of cows varied between 12,972 and 13,618, depending on the data set. Data collection periods contained either the first 50 or the first 300 d of lactation. When the data-recording period ended after 50 d of lactation, data sets were analyzed with a lactation threshold model (LTM), a multiple threshold lactation model (MTLM), and a test-day threshold model (TDTM). In the LTM analysis, mastitis was treated as a binary trait coded as 0 (no mastitis) or 1 (mastitis), whereas in MTLM mastitis, codes were between 0 and 4, depending on the number of estimated days with mastitis. The TDTM treated each day as a single observation coded similarly to that of the LTM. When the data collection period included the first 300 d of lactation, data sets were analyzed with the LTM or MTLM only, because the TDTM was computationally infeasible. Mastitis frequencies in LTM data sets were 25.8 and 39.2%, and 26.9 and 39.2% in MTLM data sets, when data recording ended after 50 and 300 d of lactation, respectively. The mastitis frequency in the TDTM data set was 5.2%. Respective heritability estimates of liability to clinical mastitis were 0.08 and 0.09 using the LTM, and 0.08 and 0.11 using the MTLM. When the TDTM was used, the estimated heritability was 0.15. Rank correlation between breeding values of the different data sets ranged between 0.40 and 0.97. Rank correlation between the LTM and MTLM were higher (0.78 to 0.97) than those between these 2 models and the TDTM (0.40 to 0.59).The MTLM combined the positive effects of both the LTM, with respect to the size of the data sets, and the TDTM, with respect to the lack of information.

Implementation of multivariate cumulative sum control charts in mastitis and lameness monitoring.

This study analyzed the methodology and applicability of multivariate cumulative sum (MCUSUM) charts for early mastitis and lameness detection. Data used were recorded on the Karkendamm dairy research farm, Germany, between August 2008 and December 2010. Data of 328 and 315 cows in their first 200 d in milk were analyzed for mastitis and lameness detection, respectively. Mastitis as well as lameness was specified according to veterinary treatments. Both diseases were defined as disease blocks. Different disease definitions for mastitis and lameness (2 for mastitis and 3 for lameness) varied solely in the sequence length of the blocks. Only the days before the treatment were included in the disease blocks. Milk electrical conductivity, milk yield, and feeding patterns (feed intake, number of trough visits, and feeding time) were used for the recognition of mastitis. Pedometer activity and feeding patterns were used for lameness detection. To exclude biological trends and obtain independent observations, the values of each input variable were either preprocessed by wavelet filters or a multivariate vector autoregressive model. The residuals generated between the observed and filtered or observed and forecast values, respectively, were then transferred to a classic or self-starting MCUSUM chart. The combination of the 2 preprocessing methods with each of the 2 MCUSUM sum charts resulted in 4 combined monitoring systems. For mastitis as well as lameness detection requiring a block sensitivity of at least 70%, all 4 of the combined monitoring systems used revealed similar results within each of the disease definitions. Specificities of 73 to 80% and error rates of 99.6% were achieved for mastitis. The results for lameness showed that the definitions used obtained specificities of up to 81% and error rates of 99.1%. The results indicate that the monitoring systems with these study characteristics have appealing features for mastitis and lameness detection. However, they are not yet directly applicable for practical implementations.

Genetic parameters for lameness and claw and leg diseases in dairy cows

Lameness in dairy cows is a serious welfare and economic problem in dairy production. The majority of all lameness cases seem to stem from claw and leg diseases. Indirect selection on claw health potentially might be feasible with lameness as indicator trait. Therefore, the genetic parameters for the 2 traits were estimated by applying both linear and threshold models. In addition, the impact of environmental effects, parity, and stage of lactation was analyzed. In total, 8,299 locomotion scores (1-5) of 326 dairy cows and 708 claw and leg disease diagnoses or treatments of 335 dairy cows from the dairy research farm Karkendamm (Institute of Animal Breeding and Husbandry, Christian-Albrechts-University, Kiel, Germany) were analyzed. Lameness was defined by a locomotion score of ≥ 3. Days in milk were limited to the range of 10 to 350 d. To quantify the effect of the claw disease digital dermatitis, a second data set without this disease was built; 52.8 and 36.4% (without digital dermatitis) of the cows were treated at least once; 47.2% of the cows were clinically lame at least at one time. Genetic parameters were estimated bivariately using the average information restricted maximum likelihood procedure as implemented in the DMU software package. The heritability estimates derived from the threshold model were about twice as large as the values based on the linear model. For lameness, the threshold heritability increased from 0.15 to 0.22 and decreased for the diseases from 0.24 to 0.22 after exclusion of digital dermatitis. The genetic correlations were high and even increased from 0.60 to 0.72 after the exclusion of digital dermatitis, which suggests that lameness (locomotion score) seems to be a good indicator for claw and leg diseases. Digital dermatitis seems to affect the mobility of the dairy cow less strongly than other claw and leg diseases.

Analysis of water intake and dry matter intake using different lactation curve models

The objective was to evaluate 6 different lactation curve models for daily water and dry matter intake. Data originated from the Futterkamp dairy research farm of the Chamber of Agriculture of Schleswig-Holstein in Germany. A data set of about 23,000 observations from 193 Holstein cows was used. Average daily water and dry matter intake were 82.3 and 19.8 kg, respectively. The basic linear mixed model included the fixed effects of parity and test-day within feeding group. Additionally, 6 different functions were tested for the fixed effect of lactation curve and the individual (random) effect of cow-lactation curve. Furthermore, the autocorrelation between repeated measures was modeled with the spatial (power) covariance structure. Model fit was evaluated by the likelihood ratio test, Akaikes and Bayesian information criteria, and the analysis of mean residual at different days in milk. The Ali and Schaeffer function was best suited for modeling the fixed lactation curve for both traits. A Legendre polynomial of order 4 delivered the best model fit for the random effect of cow-lactation. Applying the error covariance structure led to a significantly better model fit and indicated that repeated measures were autocorrelated. Generally, the best information criteria values were yielded by the most complex model using the Ali and Schaeffer function and Legendre polynomial of order 4 to model the average lactation and cow-specific lactation curves, respectively, with inclusion of the spatial (power) error covariance structure. This model is recommended for the analysis of water and dry matter intake including missing observations to obtain estimation of correct statistical inference and valid variance components.

Calibration of an automated California mastitis test with focus on the device-dependent variation

The aim of the paper was to estimate the accuracy of the metrology of an installed indirect on-line sensor system based on the automated California Mastitis Test (CMT) with focus on the prior established device-dependent variation. A sensor calibration was implemented. Therefore, seven sensors were tested with similar trials on the dairy research farm Karkendamm (Germany) on two days in July 2011 and January 2012. Thereby, 18 mixed milk samples from serial dilutions were fourfold recorded at every sensor. For the validation, independent sensor records with corresponding lab somatic cell score records (LSCS) in the period between both trials were used (n = 1,357). From these records for each sensor a polynomial regression function was calculated. The predicted SCS (PSCS) was obtained for each sensor with the previously determined regression coefficients. Pearson correlation coefficients between PSCS and LSCS were established for each sensor and ranged between r = 0.57 and r = 0.67. Comparing the results with the correlation coefficients between the on-line SCS (OSCS) and the LSCS (r = 0.20 to 0.57) for every sensor, the calibration showed the tendency to improve the installed sensor system.