This article discusses the notion that the invisibility of the animalness of the animal constitutes a fundamental obstacle to change within current production systems. It is discussed whether housing animals in environments that resemble natural habitats could lead to a re-animalization of the animals, a higher appreciation of their moral significance, and thereby higher standards of animal welfare. The basic claim is that experiencing the animals in their evolutionary and environmental context would make it harder to objectify animals as mere bioreactors and production systems. It is argued that the historic objectification of animals within intensive animal production can only be reversed if animals are given the chance to express themselves as they are and not as we see them through the tunnel visions of economy and quantifiable welfare assessment parameters.

The objectives of this study were to investigate the individual variation, repeatability and correlation of methane (CH4) production from dairy cows measured during 2 different years. A total of 21 dairy cows with an average BW of 619±14.2 kg and average milk production of 29.1±6.5 kg/day (mean±s.d.) were used in the 1st year. During the 2nd year, the same cows were used with an average BW of 640±8.0 kg and average milk production of 33.4±6.0 kg/day (mean±s.d.). The cows were housed in a loose housing system fitted with an automatic milking system (AMS). A total mixed ration was fed to the cows ad libitum in both years. In addition, they were offered concentrate in the AMS based on their daily milk yield. The CH4 and CO2 production levels of the cows were analysed using a Gasmet DX-4030. The estimated dry matter intake (EDMI) was 19.8±0.96 and 23.1±0.78 (mean±s.d.), and the energy-corrected milk (ECM) production was 30.8±8.03 and 33.7±5.25 kg/day (mean±s.d.) during the 1st and 2nd year, respectively. The EDMI and ECM had a significant influence (P<0.001) on the CH4 (l/day) yield during both years. The daily CH4 (l/day) production was significantly higher (P<0.05) during the 2nd year compared with the 1st year. The EDMI (described by the ECM) appeared to be the key factor in the variation of CH4 release. A correlation (r=0.54) of CH4 production was observed between the years. The CH4 (l/day) production was strongly correlated (r=0.70) between the 2 years with an adjusted ECM production (30 kg/day). The diurnal variation of CH4 (l/h) production showed significantly lower (P<0.05) emission during the night (0000 to 0800 h). The between-cows variation of CH4 (l/day, l/kg EDMI and l/kg ECM) was lower compared with the within-cow variation for the 1st and 2nd years. The repeatability of CH4 production (l/day) was 0.51 between 2 years. In conclusion, a higher EDMI (kg/day) followed by a higher ECM (kg/day) showed a higher CH4 production (l/day) in the 2nd year. The variations of CH4 (l/day) among the cows were lower than the within-cow variations. The CH4 (l/day) production was highly repeatable and, with an adjusted ECM production, was correlated between the years.

Management and monitoring systems may enable the farmers to enhance production results and reduce labor time. The aim of this paper is to develop a dynamic monitoring system for mortality rates of sows and piglets. For this purpose a model for mortality rates is implemented using a Dynamic Generalized Linear Model. Variance components are pre-estimated using an Expectation-Maximization algorithm applied on a dataset containing data from 15 herds, each of them including observations over a period ranging from three to nine years. Data are registrations of events for insemination, farrowing (including stillborn and live born), number of weaned piglets and death of sows. The model provides reliable forecasting on weekly basis. Detection of impaired mortality rate is performed by statistical control tools that give warnings when the mortality (rate) shows sudden or gradual changes. For each herd, mortality rate profile, analysis of model components over time and detection of alarms are computed for two categories, namely sows and piglets.

Pigs are known to be particularly sensitive to heat and cold. If the temperature becomes too low, the pigs will grow less efficiently and be more susceptible to diseases such as pneumonia. If the temperature is too high, the pigs will tend to foul the pen, leading to additional risks of infection. Furthermore, unpublished data show that the temperature within a single section of grower/finisher pigs can vary considerably from pen to pen, and previous studies have shown that pigs can be significantly affected by wind, even when not directly exposed to it. To address this latter concern, some pig producers and research stations have implemented a shielding to prevent winds from blowing between separate sections of the pig housing buildings. However, according to our search of the literature, no published studies have ever investigated the effectiveness of such shielding.To determine the significance of the effects of wind shielding, linear mixed models were fitted to describe the average daily weight gain and feed conversion rate of 1271 groups (14 individuals per group) of purebred Duroc, Yorkshire and Danish Landrace boars, as a function of shielding (yes/no), insert season (winter, spring, summer, autumn), start weight and interaction effects between shielding and start weight and shielding and insert season. Such a model was fitted separately to the data collected for each breed. Shielding was found to have significant interaction effects with season (p=0.007) and start weight (p=0.0002) for Duroc pigs, but no effect could be shown for Yorkshire or Danish landrace.To determine the effect of a group׳s placement relative to the central corridor of a grower/finisher station, a similar model was fitted to the data for Duroc pigs, replacing shielding with distance from the corridor (1st, 2nd, 3rd or 4th pen). The effect could not be tested for Yorkshire and Danish Landrace due to lack of data on these breeds. For groups of pigs above the average start weight, a clear tendency of higher growth rates at greater distances from the central corridor was observed, with the most significant differences being between groups placed in the 1st and 4th pen (p=0.0001). A similar effect was not seen on smaller pigs. Pen placement appears to have no effect on feed conversion rate.No interaction effects between shielding and distance to the corridor could be demonstrated. Furthermore, in models including both factors, the effect of distance to corridor completely dominated over the effect of shielding, suggesting that shielding should at most be considered of secondary importance.

Good management in animal production systems is becoming of paramount importance. The aim of this paper was to develop a dynamic monitoring system for farrowing rate. A farrowing rate model was implemented using a dynamic generalized linear model (DGLM). Variance components were pre-estimated using an expectation-maximization (EM) algorithm applied on a dataset containing data from 15 herds, each of them including insemination and farrowing observations over a period ranging from 150 to 800weeks. The model included a set of parameters describing the parity-specific farrowing rate and the re-insemination effect. It also provided reliable forecasting on weekly basis. Statistical control tools were used to give warnings in case of impaired farrowing rate. For each herd, farrowing rate profile, analysis of model components over time and detection of alarms were computed. The model provided a good overview of the development of the parity specific farrowing rate over time and the control charts were able to detect impaired results. Suggestions for future improvements include addition of parity-specific control charts, calibration of the charts for use in practice and inclusion of a sow effect in the farrowing model.

Two concentrates (MELK and VEM) with two different carbohydrate compositions were supplemented during milking in an Automatic Milking System (AMS). The objectives of this study were to estimate the effect of the concentrates on CH4 emission from dairy cows and to investigate the precision of the CO2-method when measuring in an AMS for different length of time. Holstein cows (n=36) were used with mean body weight of 660kg (SD=75.13) and average milk production of 31.7kg (SD=8.98), mixed parity and mixed lactation. Cows were allocated in two groups (n=18). After an adaptation period (period 1), each group received either 100% MELK (More Energy Lactating Cows; a newly introduced feeding system) or 100% VEM (Feed Value System for milk production) during periods 2 and 3. Besides, both groups were fed the same Total Mixed Ration (TMR) ad libitum in the stable. Air samples in the AMS from a point near the cows head were analysed every 20s using the Gasmet equipment based on Fourier Transform Infrared (FTIR) Spectroscopy Technique. The equipment ran continuously for 15 days over the three measurement periods (5 days×3 periods) with a 14 days waiting time in between the periods. Individual records of the CH4 and CO2 concentrations in the cows breath was calculated after subtracting the CH4 and CO2 concentrations in the stable air from the measured concentrations. The CH4:CO2 ratio was then multiplied with the calculated total CO2 production by the individual cows to get the quantitative CH4 production. Milk production and total dry matter intake (DMI, kg/day) were very similar in the two groups. The supplemented concentrate was allocated according to the individual milk yield and the intake ranged from 1.60 to 7.30kg/day in MELK cows and from 2.06 to 7.20kg/day in VEM cows. No significant difference was found for CH4 production in MELK and VEM groups over the three periods. A linear positive relation between the CH4 (g/day) and energy corrected milk (ECM, kg/day) production and the feed intake (DMI, kg/day) was observed for the entire period. The calculated CO2 and CH4 production were very similar in the two groups throughout the entire measurement period. The analysis of the precision of the CO2-method, using a 95% significance level, indicated that showing a difference of 9 or 5% in methane production requires a measuring period of 5 or 15 days, respectively, when using 18 cows per group. The study shows no effect of a limited change in supplementation of starch and sugar on CH4 production through feeding concentrates MELK or VEM in the AMS. To obtain an effect of changing the carbohydrate composition of the diet on the CH4 production, it is likely that a larger change in the diet is necessary. This can only efficiently be done by changing the TMR part of the diet.

This paper suggests methods to monitor sows׳ activity before, during and after farrowing using sensor data. The progress of parturition is analysed from video recordings for a total of 19 sows, of which half was provided with straw (S), and half received no straw (NS). The pre-partum high active behaviour is defined as the hours when the sows performed more than 50% active behaviour per hour, allowing for 1h of resting. It is characterised by its duration, intensity and its last hour as compared to the onset of parturition. No difference is observed for the duration and last hour of pre-partum high activity for Group S and Group NS. Nevertheless, the intensity indicates that Group S tends (p=0.07) to be more active (80%) than Group NS (70%) during the pre-partum high activity. The last hour of the pre-partum high active behaviour and the increase of the Lying Active behaviour characterise the onset of farrowing; and a reduction of the number of activity shifts characterises the end of farrowing. No difference of activity is observed for sows farrowing at night vs. in the daytime. Finally, results indicate that sows with a long pre-partum high activity also have more of long birth intervals (more than three times the median) during farrowing. In conclusion, the methods appear promising to monitor the activity around farrowing, and is less time consuming than video analyses. A better monitoring of these phases can potentially result in a better welfare and a reduction of piglet mortality.

This application note presents a newly developed surveillance module for monitoring reproduction performances in dairy herds. It is called Critical Control Point and is part of a recently developed management tool, Dairy Management System. This management tool is commercialized as software intended both for farmers, extension officers, breeding advisors and veterinarians. Insemination and conception rates, for cows and heifers, are modeled at the herd level using Dynamic Generalized Linear Models for binomial data. The results are updated and monitored on a weekly basis, using control charts, and alarms are provided when the performances are below target values. Both the number of observed inseminations and pregnancies, and the insemination and pregnancy rates are monitored. The components of the user interface are presented and some comprehensive graphs, accessible to the user, illustrate the herd’s performances over the last 52 weeks.

This paper reviews the use of information from animal-based monitoring systems, management information systems (MIS) and decision support systems (DSS) in pig production. As technology evolves, more and new types of data can be automatically collected. This increases the pace at which new systems are being developed. The collected data are processed into information, which should be used at the herd level. As these systems are being developed, there is a need to evaluate the value of the information provided to the user. In this review, the concepts of data and information are first described as well as the factors influencing the value of information, e.g. measuring devices and data/information processing. Then, after presenting the technologies and recordings for obtaining data used in the studies included in this paper, the value of information of the systems is reviewed according to their expected benefits in terms of production and management, animal health and welfare, and economical impact. The aim and value of DSS and other MIS are thereafter reviewed. Few studies have evaluated the economical benefits of using technologies to obtain new information, in terms of the number of piglets or the economical value. To quantify the value of information, it is suggested to develop a model frame with a detailed level of information, which allows one to estimate the value of including or excluding one piece of information. A more precise quantification of the value, and benefits, of information can in the future assist the development of more focused systems, in which implementation may be facilitated by reducing the uncertainty of the payback period.

The aim of the present study was to investigate the use of accelerometer sensors to estimate grazing time. The estimated grazing time was furthermore combined with bite frequency data in order to model grass intake. Differing levels of stocking densities and grass height were used. Two field experiments were conducted: one in 2009 (EX1) using 20 Holstein cows with 7h daily grazing and ad libitum feeding inside, and another in 2010 (EX2) using 10 Holstein cows with 7.5h daily grazing and restricted feeding inside. For both experiments, data collected were (i) activity data measured by accelerometers, (ii) manually registered bite counts and (iii) estimation of grass intake from energy requirements. In EX2 the necessity of tight sensor fixation was tested. Head mounted accelerometers were used for estimation of grazing time, which was computed using threshold values of raw downloaded data from one axis.Loosely mounted sensors attached to and hanging from the neck collar, compared to tightly mounted sensors on the head of cows did not result in significantly different estimations of grazing time. Bite count recordings showed cow individual differences in bite frequency (ranging from 48 to 62bitesmin−1) for the same day on the same paddock. The best estimation of grass intake was for cows which were fed restricted indoors (≈30% of diet). This was modelled by using grazing time and bite frequency and resulted in prediction intervals ranging from ±1.2 to ±1.4kgDMcow−1day−1 for continuous grazing with an initial grass height of 11cm. Adding individual bite frequency per cow to the model together with the grazing time, reduced the intake prediction interval from an average of ±2.3kgDMcow−1day−1 to ±1.3kgDMcow−1day−1 in a continuous grazing system.

The in vitro gas production (GP) technique has been widely used for feed evaluation. However, variability in results limits useful comparisons. Results from a ring test undertaken in four laboratories (Italy – IT, Spain – SP, Wales – WA and Denmark – DK) using the same wireless equipment (ANKOM Technology), same substrates and same laboratory protocol are presented, including calculation of repeatability and reproducibility according to ISO 5725-2. Hay, maize starch and straw samples and units without sample (blanks) were incubated in five repetitions using rumen inoculum from cows (DK, IT and WA) or sheep (SP). Curves, corrected for blanks, were fitted using an exponential regression model with a lag time. The following variables were considered: (i) GP24 and GP48: raw values at 24 and 48 h (mL/g DM), corrected for blanks; (ii) A: asymptotic GP (mL/g DM); (iii) T1/2: time when half A is produced (h); (iv) GPMR: maximum predicted GP rate (mL/h); (v) L: lag time (h). A mixed model including laboratories as random effect was used. A significant interaction between substrate and laboratories was found for all variables except A. The most repeatable and reproducible results were observed for A and GP48. The results from this ring test suggest the need for more standardisation, particularly in the procedures that occur outside the laboratory.

This paper describes a supervised learning approach to sow-activity classification from accelerometer measurements. In the proposed methodology, pairs of accelerometer measurements and activity types are considered as labeled instances of a usual supervised classification task. Under this scenario sow-activity classification can be approached with standard machine learning methods for pattern classification. Individual predictions for elements of times series of arbitrary length are combined to classify it as a whole. An extensive comparison of representative learning algorithms, including neural networks, support vector machines, and ensemble methods, is presented. Experimental results are reported using a data set for sow-activity classification collected in a real production herd. The data set, which has been widely used in related works, includes measurements from active (Feeding, Rooting, Walking) and passive (Lying Laterally, Lying Sternally) activities. When classifying 1-s length observations, the best method achieved an average recognition rate of 74.64%, for the five activities. When classifying 2-min length time series, the performance of the best model increased to 80%. This is an important improvement from the 64% average recognition rate for the same five activities obtained in previous work. The pattern classification approach was also evaluated in alternative scenarios, including distinguishing between active and passive categories, and a multiclass setting. In general, better results were obtained when using a tree-based logitboost classifier. This method proved to be very robust to noise in observations. Besides its higher performance, the suggested method is more flexible than previous approaches, since time series of any length can be analyzed.

A possibility for a new farm application to optimize grazing management focuses on estimating grazing time. In two experiments, activity data were collected using accelerometers attached to grazing animals. Grass intake was estimated by the method of energy balance. Grazing time was computed from acceleration data (GRacc). GRacc data showed a sensitivity of 74% and specificity of 82%, when compared to manual observations. GRacc was tested for correlation with grass intake: the best correlation (0.82) was observed for cows with restricted feeding in the barn, and strip grazing on grass with a height of 16 cm.

Monitoring animal production results in real time is a challenge. Existing management information systems (MIS) in pig production are typically based on static statements of selected key figures. The objective of this paper is to develop a dynamic monitoring system for litter size at herd and sow level, with weekly updates. For this purpose, a modified litter size model, based on an existing model found in the literature, is implemented using dynamic linear models (DLMs). The variance components are pre-estimated from the individual herd database using a maximum-likelihood technique in combination with an Expectation–Maximization (EM) algorithm applied on a larger dataset with observations from 15 herds. The model includes a set of parameters describing the parity-specific mean litter sizes (herd level), a time trend describing the genetic progress (herd level), and the individual sow effects (sow level). It provides reliable forecasting with known precision, on a weekly basis, for future production. Individual sow values, useful for the culling strategy, are also computed. In a second step, statistical control tools are applied. Shewhart Control Charts and V-masks are used to give warnings in case of impaired litter size results. The model is applied on data from 15 herds, each of them including a period ranging from 150 to 800 weeks. For each herd, the litter size profile, the litter size over time, the sow individual effect and sow economic value, are computed. Perspectives for further development of the model can take into account indices including conception rate, service rate, mortality rate etc. Such a model can be used as a basis for developing a new, dynamic, management tool.

This article suggests and assesses two different monitoring methods for detecting sows parturition using series of three-dimensions acceleration measurements previously classified into activity types. Two groups of sows are monitored: a first group (n = 9) provided with straw (S), and a second group (n = 10) where no straw is provided (NS); two types of activity are taken into account: high active behaviour (corresponding to feeding, rooting and nest building behaviours) and total active behaviour (including any active activity type). The first method suggests modeling sows’ diurnal pattern of activity using a saw-tooth function for the probability of being active and monitoring the series using a Dynamic Generalized Linear Model (DGLM). The second method is based on a cumulative sum of hourly differences of activity, from day-to-day. Both methods use a threshold value, optimized for each group, to detect the onset of farrowing. Best results in terms of sensitivity and specificity are observed for the cumulative sum method, using individual variance and monitoring high active (sensitivity = 100%; specificity = 100%) and total active behaviours (sensitivity = 100%; specificity = 95%). Results of the DGLM method indicate a sensitivity of 100% and a specificity of 89% in average for both group S and NS. Observing the occurrence of alarm times, the DGLM method allows (i) earlier detection of farrowing: 15 h before the onset of farrowing, for both groups, as compared to 9–12 for the other methods; and (ii) a better distribution of alarms, i.e. minimize the number of alarms occurring within the last 6 h before farrowing.

Automatic monitoring of activity levels in broiler chicken flocks may allow early detection of irregular activity patterns, indicating potential problems in the flock. Leg disorders are the main welfare concern for modem broiler chickens. Dynamic control of broiler activity during the growing period may improve the muscular-skeletal development thereby reducing leg disorders and improving welfare of the animals. The undisturbed activity of groups of broiler chickens was investigated in three steps. The first step applied a model, which was able to filter out outliers in the data stream of automatically recorded activity from overhead video cameras. The second step described the undisturbed levels of activity in groups of broiler chickens over the course of a day in week 1, 2 and 3. The third step applied a method to detect deviations in activity level, thereby giving an indication of a level change in activity within the flock of broilers. The results indicate the potential for automatic detection of deviations in activity level in flocks of broiler chickens. From these methods it is possible to develop automatic monitoring systems, which can notify the producer when the activity in the broiler flock deviates from an expected level at a given age. Such monitoring system may improve the welfare of commercial broiler chickens. (C) 2011 IAgrE. Published by Elsevier Ltd. All rights reserved.

This paper discusses the main architectural alternatives and design decisions in order to implement a sows’ activity classification model on electronic devices. The different possibilities are analyzed in practical and technical aspects, focusing on the implementation metrics, like cost, performance, complexity and reliability. The target architectures are divided into: server based, where the main processing element is a central computer; and embedded based, where the processing is distributed on devices attached to the animals. The initial classification model identifies the activities performed by the sows using a multi-process Kalman filter having, as input, 3-axes data from accelerometers. However, the power demanding hardware resources to run the filters require frequent battery recharges, making its use unsuitable in the current state-of-the-art. It motivated the development of a heuristic classification approach, focusing on the resource constrained characteristics of embedded systems. The new approach classifies the activities performed by the sows with accuracy close to 90%. It was implemented as a hardware module that can easily be instantiated to provide preprocessed information to models in order to detect important situations in the sows’ life, e.g. the onset of farrowing.

This article suggests a method for classifying sows’ activity types performed in farrowing house. Five types of activity are modeled using multivariate dynamic linear models: high active (HA), medium active (MA), lying laterally on one side (L1), lying laterally on the other side (L2) and lying sternally (LS). The classification method is based on a Multi-Process Kalman Filter (MPKF) of class I. The performance of the method is validated using a Test data set. Results of activity classification appear satisfying: 75–100% of series are correctly classified within their activity type. When collapsing activity types into active (HA and MA) vs. passive (L1, L2, LS) categories, results range from 96 to 100%. In a second step, the suggested method is applied on series collected for 19 sows around the onset of farrowing, including 9 sows that received bedding materials (57 sow days in total) and 10 sows that received no bedding material (61 sow days in total). Results indicate that there is a marked (i) increase of active behaviours (HA and MA, p<0.001) and (ii) decrease of lying laterally (L1 and L2) behaviours starting 20–16h before the onset of farrowing; during the last 24h before parturition, the averaged time spent lying laterally in a row decreases and the number of changes of activity types for HA and MA increases. These behavioural changes occur for sows both with and without bedding material, but are more marked when bedding material is provided. Straightforward perspectives for applications of this classification method for monitoring activity types are, e.g. automatic detection of farrowing and detection of health disorders.