Conference Agenda

The main goal of this study was to assess the cleanliness of three pig farms using Adenosine triphosphate (ATP) bioluminescence and BioFinder tests after standard washing, and to establish hygiene guidelines before hot-water washing. Three sampling zones were examined: the feeder area, the resting area, and the defecation area. Eleven pens were assessed, with two sampling points per zone for both ATP and BioFinder analyses. Due to ongoing changes to pen flooring, up to three different floor materials were observed in feeder zones. Additional measurements were conducted in finishing rooms with slated flooring and small concrete plates in the feeder areas. ATP tests were carried out 24 h after washing and before disinfection. For the BioFinder reactions, evaluation was performed 1 h after washing. Results showed that feeder and defecation zones had higher Relative Light Unit (RLU) than resting areas, and that floor material influenced cleanliness. In maternity units, concrete in the feeder areas had RLU values like those of typically dirtier zones. BioFinder reactions followed similar results with more intense reactions near feeders. The findings indicate greater accumulation of microbial and organic residues in feeder areas, likely associated with surface wear and animal behavior.

Over 90% of Swiss dairy cows have access to pasture, yet accurately quantifying herbage dry matter intake (hDMI) from pasture remains challenging. This study estimated hDMI under three grazing durations (2, 8 and 16 h/day) using three different approaches: 1) RumiWatch System (RWS) uses eating and rumination behaviour. 2) Net energy balance (NEB) calculates theoretical hDMI based on animal performance, feed intake and pasture energy content. 3) Herbometer based on pre- and post-grazing sward surface height using an electronic raising plate meter. Two groups of 20 Holstein- and Brown-Swiss lactating cows received a partial mixed ration plus concentrates. One group had in addition access to pasture across three periods 2, 8 or 16 h/day. A crossover design was applied, so both groups experienced all treatments. The trial was conducted in summer 2023 and each period lasted six days. At 2 h grazing, the hDMI per cow was 1.3 and 1.6 kg/day with NEB and herbometer, while RWS equations yielded negative values. Similar results were found at 8 h grazing. At 16 h grazing, the three approaches produced consistent results: 3.5 to 4.6 kg/day. The RWS approach seems unsuitable for short grazing durations, likely due to calibration under long grazing conditions.

Ammonia (NH₃) and methane (CH₄) emissions from naturally ventilated dairy barns represent a major source of impact on the environment, contributing significantly to air quality degradation and climate change.

This study aimed at assessing daily and seasonal gas concentrations to estimate NH₃ and CH₄ emissions from a naturally-ventilated dairy barns in Mediterranean climate. Gas concentrations (i.e., NH₃, CH₄ and CO₂) and climatic variables (i.e., temperature, relative humidity, wind speed, and wind direction) have been collected during one year at different horizontal and vertical sampling locations. Gas concentrations were spatially analysed and the CO₂ mass balance method was applied to estimate emissions.

The results showed the daily and seasonal variability of gas concentrations and emissions in a naturally-ventilated semi-open Mediterranean barn. The highest NH₃ concentrations and emission rates were observed during the hot season, whereas CH₄ concentrations and emissions exhibited their highest values during the cold season. The yearly emission factors for this barn typology were 16.43 ± 9.80 kg y⁻¹ LU⁻¹ and 150.55 ± 93.15 kg y⁻¹ LU⁻¹ for NH₃ and CH₄, respectively. These findings highlight the need for climate-specific emission assessments and dynamic modelling approaches to support both mitigation strategies and the refinement of emission inventories in warm-temperate regions.

Feeding dairy cows at herd level, despite their genetic variations, results in challenges such as greenhouse gas emissions, nutrient inefficiencies, and resource use inefficiency at individual level. Individualized dynamical feeding strategy offers a potential solution to these challenges. Such approaches depend on mechanistic models. While these models are validated using controlled experimental datasets, their performance under commercial farm conditions remains uninvestigated. This study investigates model parameter identification for a validated mechanistic dairy cow model using real-world farm data. The dataset comprised herd-level roughage composition, and individual cow measurements including daily concentrate intake, milk yield, milk fat and protein content from a commercial farm. An offline nonlinear system identification problem was implemented in CasADi and solved with IPOPT. The genetic and physiological parameters together with initial state values were estimated by minimizing the prediction error. The dataset was divided into training (65%) and validation (35%) subsets and achieved a fit of 72% for milk production on the training data and 71% on the validation data. These preliminary results show stable optimization convergence as well as a relatively low prediction error under noisy farm conditions. Ongoing work focuses on extending the study to an online parameter estimation for real-time individualized feed optimization.

Caged-layer farming is the dominant mode of automated layer production.However, current methods rely on detached, single-task models that cause computational redundancy and lack data synergy. These isolated systems struggle with dynamic occlusion on conveyors and fail to provide the granular traceability required for individual cage management. This study proposes an integrated monitoring framework that synergizes instance segmentation, machine learning, and spatiotemporal tracking. First, Egg-Seg was developed by incorporating Multi-Scale Convolutional Attention (MSCA) modules and Dice Loss, significantly enhancing the segmentation of irregular defects and egg boundaries. Second, the Egg-Weight model was established using Standard Ellipse Fitting and the LightGBM algorithm, effectively mitigating geometric distortions caused by dynamic occlusion. Finally, the Egg-Trace method, combining conveyor motion analysis with ByteTrack, implemented a marker-based "Time-Cage" mapping logic to achieve precise counting and traceability. Experimental results demonstrate that Egg-Seg achieved an mAP50-95 of 93.5% for multi-class targets, outperforming mainstream segmentation models. The Egg-Weight model attained an R² of 0.966 and an MAE of 0.83 g. Moreover, the system achieved an AEC of 100%, an AET of 99.16%, and an AEPC of 97.32%. This framework digitizes multi-dimensional production metrics (quality, weight, count, location) at the individual cage level, providing technical support for precision livestock farming.