Grazing intensity regulates cattle movement and forage-use-efficient behavioral patterns through vegetation and abiotic pathways in a temperate meadow steppe

Abstract

Livestock movement and time allocation are key behavioral processes linking grazing pressure to grassland use efficiency, yet how these behaviors respond to quantified grazing intensity and associated vegetation and abiotic conditions remains insufficiently understood. In this study, we used a long-term controlled grazing experiment in a temperate meadow grassland to examine cattle behavioral dynamics under multiple grazing intensities. GPS collars continuously recorded cattle movements across seasons and grazing treatments. Using movement speed and displacement metrics, we developed and compared six machine-learning models, including Random Forest, XGBoost, and CatBoost, to identify a suitable classifier for distinguishing three behavioral states: resting, grazing, and walking. Model selection considered both classification performance and training time. We further integrated vegetation structural attributes and meteorological variables and applied correlation analysis, Random Forest variable-importance assessment, and piecewise structural equation modeling to identify the key drivers of cattle behavior. The Random Forest model achieved high classification accuracy and computational efficiency and was therefore selected for subsequent behavioral classification. Cattle behavior varied markedly among grazing intensities. Under moderate grazing, cattle exhibited longer grazing and resting durations, reduced walking time, and a more favorable time-allocation pattern with lower non-foraging movement costs. In contrast, heavy grazing substantially increased walking time and daily movement distance, suggesting greater searching effort under reduced forage availability. Movement distance was jointly regulated by vegetation structure and rainfall, whereas the durations of grazing, walking, and resting were more strongly influenced by abiotic factors, particularly rainfall and soil moisture. By integrating GPS-based behavioral monitoring with machine-learning approaches, this study identifies livestock behavioral metrics as practical indicators for assessing grazing pressure and grass–livestock balance. These findings provide quantitative evidence for optimizing grazing intensity and improving adaptive management in temperate meadow grasslands.