Conference Agenda

The number of automation technologies for dairy cattle farming has increased in recent years. The influence of individual technologies on animal welfare has already been investigated, but information about the impact of increasing farm automation through various combinations of these technologies is missing. The aim of this study was to examine whether the automation level on a dairy farm affect welfare indicators. Therefore, 32 trial farms were categorized into varying automation levels using a newly developed classification system. The Welfare Quality® Assessment protocol for dairy cows was used to conduct welfare assessments. Overall welfare scores and individual measures from the protocol were compared across farms with differing automation levels. No significant differences were observed in overall welfare scores, suggesting that the influence of automation does not exceed other farm-related factors influencing animal wellbeing (e.g. housing environment, management). Significant effects of milking, feeding, and bedding systems on the behavior of cattle were observed. Thereby, higher automation levels positively affect the human–animal relationship and led to positive emotional states. Farms with higher automation levels had significantly lower scores of severe lameness and dirtiness of lower legs. Thus, a higher automation level could help to improve animal welfare on dairy farms.

Heat and cold stress in pig housing arise from coupled animal thermoregulation and barn microclimate. Yet mechanistic animal models are seldom available in a reusable, co-simulation-ready form. This work develops a modular Modelica implementation of a mechanistic thermoregulation model for a growing pig (≈100 kg), parameterised from published literature. The model represents core and skin heat storage with internal heat transfer and determines boundary heat transfer via convection, long-wave radiation using a mean radiant temperature proxy, and latent/sensible losses through respiration and skin evaporation. Thermoregulatory feedback is implemented using temperature-error signals (ε) that modulate pulmonary ventilation (panting) and metabolic heat production; a hot-side controller follows established heat-stress formulations and a cold-side extension enables thermogenesis under cool conditions. Results to date show stable simulation under annual indoor-climate forcing at a 10 min step. The model reports heat-flow partitions (metabolic, convective, radiative, evaporative, respiratory) that support diagnostics and sensitivity analysis. Responses to warm versus cool episodes follow expected trends, with increased respiratory losses during heat load and increased metabolic heat during cold load. Next steps will benchmark outputs against established heat- and cold-stress indices, and couple the model with barn physics to assess mitigation strategies and energy–welfare trade-offs.

Sensor integration enables the classification of behavioral data while taking microclimatic influences into account. This study presents an approach incorporating these effects into the analysis of animal behavior, providing insights for precision livestock farming. The experiment was conducted on nine heifers, equipped with a neck-mounted sensor that continuously records tri-axial acceleration, temperature (°C), relative humidity (%), light intensity (lux), and air pressure (hPa) on pasture conditions. Signals were aggregated by 15-minute intervals. Random Forest Regression (RFR) was used to calculate activity deviation from microclimate variables, which were assessed using R², RMSE, and MAE. RFR classification determined activity levels. Classification performance was evaluated using F1 score, precision, recall, and confusion matrix analysis. RFR's limited predictive performance (RMSE = 88.10, MAE = 58.59) shows microclimate alone doesn't describe activity deviance. But analysis recognized humidity, air pressure, and temperature as the most significant predictors. Temporal cyclic variables contribute less, showing environmental dominance over diurnal effects. Results suggest that animal-proximal microclimate sensing provides meaningful contextual information for activity monitoring in pasture-based systems. Even though environmental variables do not fully determine activity magnitude, they measurably contribute to activity-state differentiation. This finding supports the integration of microclimate sensing into data-driven pasture management frameworks.

The pig sector faces increasing challenges concerning pig welfare and production in a changing climate. Automatic sorting systems enable housing of large groups (>150 pigs), the system separates feeding areas from functional areas using a sorting scale. Appropriate system design is essential to accommodate pigs at different weights under varying thermal conditions, as suboptimal design compromise welfare and production efficiency. This study investigated space usage in two parallel units (200 pigs per unit; 209 m²) on a commercial farm during May to October in 2024. Seven top-view cameras per unit recorded images during 48-hour periods once weekly throughout the production periods. Functional areas on the images were analyzed using computer vision. Pig occupancy (pixel-based fraction) and number of pigs were extracted with an image-processing pipeline using pre-trained foundation models (GroundingDino and Segment Anything Model; no task-specific training or fine-tuning). Validation demonstrated a mean average precision (mAP) of 91.1 compared to manual scoring. Linear mixed-effect models revealed significant differences in occupancy between different functional areas over time. Temperature significantly affected occupancy patterns. Computer vision proved suitable for quantifying space use in large-group housing under commercial conditions, highlighting the importance of improving system design to support animal welfare.

Determination of the ventilation rate in naturally ventilated livestock buildings is essential for the design and control of the ventilation system as well as in the estimation of the gaseous emissions from the buildings. The aim of this study is to quantify the ventilation rate of a naturally ventilated pig barn with an outdoor exercise yard. One-year on-farm measurements of air velocities at two locations at the sidewall opening and three locations at the yard opening of the barn were conducted. Meanwhile, the outdoor meteorological parameters including air temperature, relative humidity, wind speed and wind direction as well as the number and mass weight of the pigs in the barn were recorded. The results showed a strong seasonal variation of the ventilation rate with a daily averaged value of 0.94 m³· s^-1, 6.83 m³· s^-1, and 2.34 m³· s^-1 in April, August, October 2023, respectively. Correlation analysis revealed that the ventilation rate was largely affected by the opening size, air temperature, and outdoor wind speed, whereas the effects of pigs number and body weight were comparatively minor. The results provide information for ventilation performance evaluation and ventilation control strategies development for this type of pig housing systems.

Housing environment plays a key role in animal welfare. Monitoring space use of farm animals improves understanding of how animals interact with and adapt to their environment. In this work, we demonstrate how recent advances in computer vision, particularly pre-trained foundation models, can be used to detect individual animals without task-specific training, and that the resulting detections can serve as a proxy for estimating area occupancy.

We combined the vision–language object detector GroundingDINO and Segment Anything Model (SAM) for detecting, counting and segmenting animals within predefined functional areas to estimate occupancy over time. We evaluated the detector on two datasets of different species: (1) a large group housing system for pigs (8 weeks, 12 cameras, 600 images, 18 functional areas), and (2) a dairy cow facility (6 days, 4 cameras, 448 images, 8 functional areas).

Performance of the detector was assessed using mean Average Precision (mAP@[.5:.95]), yielding scores of 91.1 for Dataset 1 and 89.3 for Dataset 2. The results indicate that pre-trained foundation models enable effective zero-shot detection for livestock monitoring, enabling faster development of monitoring systems and supports data-driven approaches to improving animal welfare and housing design.

A key principle in achieving source-oriented NH₃ emission reduction in dairy houses is limiting urease activity in urine puddles by an integrated floor cleaning approach. The approach combines optimized floor design, targeted disinfectant application and compatible manure removal methods. Neutral electrolysed oxidising water (N-EOW) has been proven to effectively limit urease activity on grooved solid floors. To explore the combined manure removal and N-EOW application strategies, two complementary studies were conducted. Firstly, five manure removal methods (including different combinations of scraping, brushing and high-pressure cleaning) were tested, followed by standardized N-EOW spraying, to assess changes in floor urease activity. The most promising treatment combinations were further examined to determine their long-term effects on urease activity and NH₃ emissions. Results showed that N-EOW effectively reduced urease activity in floors that have been wet-brushed or high-pressure cleaned, achieving reductions of 39% and 64%, respectively. After high pressure cleaning once, N-EOW spraying once per day reduced NH₃ emissions by 71% from the floor, while maintaining long-term low urease activity. Overall, this study provides practical guidance for floor management strategies that enable efficient, source-based NH₃ emission reduction in dairy houses.