Automated Animal Welfare Assessment Under Climate Change Using Deep Learning

Abstract

Climate change is increasingly affecting livestock health and welfare through rising temperatures, humidity variations, heat waves, and extreme environmental conditions. Continuous assessment of animal welfare is therefore essential for sustainable and climate-resilient livestock management. This study presents a deep learning-based framework for automated animal welfare assessment by integrating visual, behavioural, and environmental information. Animal images and video sequences are analysed to identify welfare-related behaviours such as feeding, drinking, walking, standing, lying, resting, and abnormal inactivity. Environmental parameters, including temperature, relative humidity, seasonal conditions, and Temperature-Humidity Index, are integrated with the extracted behavioural features to provide climate-aware welfare assessment. Deep learning models are employed for automated feature extraction, behaviour recognition, temporal pattern analysis, climate-stress detection, and welfare-risk classification. The framework categorizes welfare conditions into normal, moderate concern, and high-risk levels and can generate early warnings when climate-induced behavioural abnormalities are detected. The resulting information can support management interventions related to ventilation, cooling, water availability, shade provision, and feeding schedules. The proposed approach reduces dependence on subjective manual observation while enabling continuous and scalable livestock monitoring. Overall, integrating deep learning with environmental intelligence provides a promising foundation for automated welfare assessment, early climate-stress detection, and sustainable livestock management.