Continuous welfare assessment of dairy cows at individual level
Animal welfare on farms is currently assessed using human-evaluation protocols, which provide a single record of herd condition at a specific moment. Integrating sensor-based data and farm records, continuous information on each animal's welfare can be obtained. This study aims to create an alg...
| Autores: | , , , , , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2026 |
| País: | España |
| Institución: | Universitat Autònoma de Barcelona |
| Repositorio: | Dipòsit Digital de Documents de la UAB |
| Idioma: | inglés |
| OAI Identifier: | oai:ddd.uab.cat:324771 |
| Acceso en línea: | https://ddd.uab.cat/record/324771 https://dx.doi.org/urn:doi:10.1016/j.biosystemseng.2025.104365 |
| Access Level: | acceso abierto |
| Palabra clave: | Precision livestock farming Unsupervised machine learning Welfare index Animal-based indicators Accelerometer Rumen bolus |
| Sumario: | Animal welfare on farms is currently assessed using human-evaluation protocols, which provide a single record of herd condition at a specific moment. Integrating sensor-based data and farm records, continuous information on each animal's welfare can be obtained. This study aims to create an algorithm to assess individual dairy cow welfare, contributing to the goal of building a platform to inform producers and consumers about dairy cattle welfare. It was built based on the Five Domains model of animal welfare. 221 cows from four commercial free-stall barn farms in Spain and Italy were fitted with accelerometry collars and rumen boluses and monitored for 92 days. Individual data were collected daily. Accelerometers recorded time spent ruminating, eating, lying, walking, and standing within a 24-h interval. Boluses recorded rumen pH and temperature every 10 min, averaged over 24 h. Farm records included parity, veterinary treatments, and milk conductivity. The model provides a daily global welfare index per cow, categorised into health, nutrition, and environment scores. Behaviour and mental state were not included due to a lack of relevant sensor data. Scores range from 0 to 10, indicating the likelihood of the cow experiencing welfare-compromising conditions. Normal thresholds, based on scientific literature, were set for each trait. The algorithm detected daily deviations in traits, assuming that cows with welfare issues deviate from normal behavioural and physiological patterns. When a cow's welfare index decreased, affected domains could be identified, enabling farmers to address potential welfare issues and implement corrective measures. |
|---|