Measuring dairy cow welfare with real-time sensor-based data and farm records: a concept study.

Welfare assessment of dairy cows by in-person farm visits provides only a snapshot of welfare and is time-consuming and costly. Possible solutions to reduce the need for in-person assessments would be to exploit sensor data and other routinely collected on-farm records. The aim of this study was to develop an algorithm to classify dairy cow welfare based on sensors (accelerometer and/or milk meter) and farm records (e.g. days in milk, lactation number). In total, 318 cows from six commercial farms located in Finland, Italy and Spain (two farms each) were enrolled for a pilot study lasting 135 days. During this time, cows were routinely scored using 14 animal-based measures of good feeding, health and housing based on the Welfare Quality® (WQ®) protocol. WQ® measures were evaluated daily or approximately every 45 days, using disease treatments from farm records and on-farm visits, respectively. WQ® measures were supplemented with daily temperature-humidity index to account for heat stress. The severity and duration of each welfare measure were evaluated, and the final welfare index was obtained by summing up the values for each cow on each pilot study day, and stratifying the result into three classes: good, moderate and poor welfare. For model building, a machine-learning (ML) algorithm based on gradient-boosted trees (XGBoost) was applied. Two model versions were tested: (1) a global model tested on unseen herd, and (2) a herd-specific model tested on unseen part of the data from the same herd. The version (1) served as an example on the model performance on a herd not previsited by the evaluator, while version (2) resembled a custom-made solution requiring in-person welfare evaluation for model training. Our results indicated that the global model had a low performance with average sensitivity and specificity of 0.44 and 0.68, respectively. For the herd-specific version, the model performance was higher reaching an average of 0.64 sensitivity and 0.80 specificity. The highest classification performance was obtained for cows in poor welfare, followed by cows in good and moderate welfare (balanced accuracy of 0.77, 0.71 and 0.68, respectively). Since the global model had low classification accuracy, the use of the developed model as a stand-alone system based solely on sensor data is infeasible, and a combination of in-person and sensor-based welfare evaluation would be preferable for a reliable welfare assessment. ML-based solutions, even with fair discriminative abilities, have the potential to enhance dairy welfare monitoring.

High biosecurity and welfare standards in fattening pig farms are associated with reduced antimicrobial use.

In order to reduce antimicrobial use in pig production, the consequences of insufficient biosecurity and welfare problems need to be known. This study aimed to investigate associations between the number of antimicrobial treatments per fattening pig, and biosecurity, indicators for animal welfare as well as the prevalence of lesions at slaughter. The data used in this study were extracted from the pig health and welfare classification system (Sikava), which gathers data on medicine usage, meat inspection, animal welfare and the condition of farm buildings from over 95% of pig production in Finland. The data were registered during years from 2011 to 2013. Upon antimicrobial prescription, information on the number of fattening pigs treated and the main reason for treatment was recorded. In addition, at least 4 times per year, pig farms registered in Sikava were visited by the farm veterinarian who assessed, among other things, biosecurity and indicators for animal welfare (air quality, condition of facilities, cleanliness, enrichment and stocking density). Finally, data from slaughterhouse inspections were collected (number of carcasses with joint infection, abscesses, lung lesions, pleurisy and liver lesions). For analysis, these datasets were aggregated at the farm level to a quarter of a year. During the studied period, the mean number of antimicrobial treatments per fattening pig per 3 months was equal to 0.09. The main reasons for antimicrobial treatments were musculoskeletal diseases, tail biting and respiratory disorders (42, 33 and 12% of diagnoses, respectively). The meat inspection scoring indicated that as much as 14.7% of all pigs had pleurisy, 5.3% liver lesions and 4.1% abscesses. A standard zero-inflated negative binomial model was used to identify factors associated with the number of antimicrobial treatments per pig. The count of antimicrobial treatments per pig increased with the size of a farm. Regardless of prevalence of lesions, farms with poor drinking equipment, insufficient enrichment and a combination of poor condition of pens and high stocking density were associated with an increased number of antimicrobial treatments for musculoskeletal diseases per pig. Problems with stocking density and enrichment were associated with the number of antimicrobial treatments for tail biting, although these results depended on prevalence of joint infections. Problems with air quality and the combination of poor cleanliness and poor condition of facilities were associated with increased number of antimicrobial treatments due to respiratory diseases. This study suggests that by improving biosecurity and welfare at pig farms, antimicrobial use can be reduced.

Analyses of body weight patterns in growing pigs: a new view on body weight in pigs for frequent monitoring.

Frequent BW monitoring of growing pigs can be useful for identifying production (e.g. feeding), health and welfare problems. However, in order to construct a tool which will properly recognize abnormalities in pigs' growth a precise description of the growth process should be used. In this study we proposed a new model of pig growth accounting for daily fluctuations in BW. Body weight measurements of 1710 pigs (865 gilts and 843 barrows) originating from five consecutive batches from a Danish commercial farm were collected. Pigs were inserted into a large pen (maximum capacity=400) between November 2014 and September 2015. On average, each pig was observed for 42 days and weighed 3.6 times a day when passing from the resting to feeding area. Altogether, 243,160 BW measurements were recorded. A multilevel model of pig growth was constructed and fitted to available data. The BW of pigs was modeled as a quadratic function of time. A diurnal pattern was incorporated into the model by a cosine wave with known length (24 h). The model included pig effect which was defined as a random autoregressive process with exponential correlation. Variance of within-pigs error was assumed to increase with time. Because only five batches were observed, it was not possible to obtain the random effect for batch. However, in order to account for the batch effect the model included interactions between batch and fixed parameters: intercept, time, square value of time and cosine wave. The gender effect was not significant and was removed from the final model. For all batches, morning and afternoon peaks in the frequency of visits to the feeding area could be distinguished. According to results, pigs were lighter in the morning and heavier in the evening (minimum BW was reached around 1000 h and maximum around 2200 h). However, the exact time of obtaining maximum and minimum BW during the day differed between batches. Pigs had access to natural light and, therefore, existing differences could be explained by varying daylight level during observations periods. Because the diurnal amplitude for pig growth varied between batches from 0.9 to 1.4 kg, BW monitoring tools based on frequent measurements should account for diurnal variation in BW of pigs. This proposed description of growth will be built into a monitoring tool (a dynamic linear model) and applied to farm data in future studies.

Monitoring growth in finishers by weighing selected groups of pigs - A dynamic approach.

Application of BW monitoring methods for the whole batch of pigs is not common in commercial herds. Instead, farm managers may regularly weigh a chosen subset of pigs (observed group) and use the obtained information for monitoring, forecasting, and decision support. The objective of this study was to construct a model for growth monitoring and forecasting in pig fattening herds and use the developed model framework to quantify the value of information on BW. The dynamic process of pig growing was described by means of a dynamic linear model (DLM) with Kalman filtering. For this study, data from 9 fattening cycles with the total registration for 9,800 pigs were used. The variance components were estimated by fitting a mixed-effects linear model on selected BW measurements. The obtained model was evaluated on its performance in forecasting the number of pigs ready to deliver from the whole batch and from a particular pen given the level of information on a reference data set consisting of 2 batches (Batch 3 [B1] and Batch 4 [B2]). Scenarios with a different frequency of observations (only 1 selected week, every second week, or weekly) on individual and aggregated levels for an observed group comprising 1 pen (36 pigs, which constitute 7.5% of pigs in a batch) or 2 pens (15.5% of pigs) were analyzed. Moreover, results with only initial herd information and insertion BW at the batch, pen, and pig level were presented. The model can be used for growth monitoring of the batch and for prediction of the number of pigs ready for slaughter in a given week (i.e., with a BW exceeding a threshold, which, in this study, is set to 105 kg). With an increased level of information, both accuracy (measured by the mean absolute deviation [MAD] of actual number of pigs above 105 kg from predicted number) and precision (measured by CV) of the model continue to improve. When monitoring all pigs at insertion and the observed groups every week (15.5% of pigs) compared with predictions based on only initial herd information, the MAD between the observed and predicted number of pigs above 105 kg in a single pen decreased by 1.4 and 2 pigs whereas CV was reduced by 147 and 78% for B1 and B2, respectively. The DLM was able to detect variation between pens already at insertion; therefore, data on initial BW had high value for the prediction procedure. Moreover, the aggregation had a marginal effect on model performance.

The economic and environmental value of genetic improvements in fattening pigs: An integrated dynamic model approach.

The selection of animals for improved performance affects the profitability of pig fattening and has environmental consequences. The goal of this paper was to examine how changes in genetic and market parameters impact the biophysical (feeding patterns, timing of slaughter, nitrogen excretion) and economic (return per pig space unit) results describing pig fattening in a Finnish farm. The analysis can be viewed as focusing on terminal line breeding goals. An integrated model using recursive stochastic dynamic programming and a biological pig growth model was used to estimate biophysical results and economic values. Combining these models allowed us to provide more accurate estimates for the value of genetic improvement and, thus, provide better feedback to animal breeding programs than the traditional approach, which is based on fixed management patterns. Besides the benchmark scenario, the results were simulated for 5 other scenarios. In each scenario, genotype was improved regarding daily growth potential, carcass lean meat content, or the parameters of the Gompertz growth curve (maturing rate [], adult weight of protein [α], and adult weight of lipid mass []). The change in each parameter was equal to approximately 1 SD genetic improvement (ceteris paribus). Increasing , , daily growth potential, or carcass lean meat content increased the return on pig space unit by €12.60, €7.60, €4.10, or €2.90 per year, respectively, whereas an increase in decreased the return by €3.10. The genetic improvement in and resulted in the highest decrease in nitrogen excretion calculated in total or per kilogram of carcass gain but only under the optimal feeding pattern. Simulated changes in the Gompertz growth function parameters imply greater changes in ADG and lean meat content than changes in scenarios focusing on improving ADG and lean meat content directly. The economic value of genetic improvements as well as the quantity of nitrogen excreted during the fattening period largely depends on feeding. Improved genotypes can require changes in pig management pattern. Estimating the influence of the genotype on the nitrogen excretion without considering changes in the management pattern can result in flawed conclusions. To improve overall economic performance and to decrease the environmental footprint of fattening pig production, the pig producer can adjust the herd management pattern according to the pigs' genetics.

Optimal management of replacement heifers in a beef herd: a model for simultaneous optimization of rearing and breeding decisions.

The aim of this study was to provide farmers an efficient tool for supporting optimal decisions in the beef heifer rearing process. The complexity of beef heifer management prompted the development of a model including decisions on the feeding level during prepuberty (age <10 mo), the time of weaning (age, BW, calendar month), the feeding level during the reproductive period (age ≥10 mo), and time of breeding (age, BW, and calendar month). The model was formulated as 3-level hierarchic Markov process. A founder level of the model has 12 states resembling all possible birth months of a heifer. Based on the birth month information from the founder level, for the indoor season (November to April) and outdoor season (May to October), feeding and breeding costs (natural service cost in the outdoor and AI cost in the indoor season) were applied. The optimal rearing strategy was found by maximizing the total discounted net revenues from the predicted future productivity of the Polish Limousine heifers defined as the cumulative BW of calves born from a cow calved until the age of 5 yr, standardized on the 210th day of age. According to the modeled optimal policy, heifers fed during the whole rearing period at the ADG of 810 g/d and generally weaned after the maximum suckling period of 9 mo should already be bred at the age of 13.2 mo and BW constituting 55.6% of the average mature BW. Based on the optimal strategy, 52% of all heifers conceived from May to July and calved from February to April. This optimal rearing pattern resulted in an average net return of EUR 311.6 per pregnant heifer. It was found that the economic efficiency of beef operations can be improved by applying different herd management practices to those currently used in Poland. Breeding at 55.6% of the average mature BW, after a shorter and less expensive rearing period, resulted in an increase in the average net return per heifer by almost 18% compared to the conventional system, in which heifers were bred after attaining 65% of the mature BW. Extension of the rearing period by 2.5 mo (breeding at the age 15.7 mo), due to a prepubertal growth rate lowered by 200 g, reduced the average net return per heifer by 6.2% compared to the results obtained under the basic model assumptions. In the future, the model may also be extended to investigate additional aspects of the beef heifer development, such as the environmental impacts of various heifer management decisions.