Spatiotemporal Mapping of Grazing Livestock Behaviours Using Machine Learning Algorithms.

Area of Science:

• Ecological science
• Animal behavior studies
• Sustainable land management

Background:

• Grassland health is often assessed using vegetation indices (NDVI, LAI).
• Livestock behavior's role in grassland degradation is understudied, especially spatial and temporal patterns.
• Understanding these patterns is crucial for effective grassland management.

Purpose of the Study:

• To investigate spatiotemporal livestock behavior patterns under different grazing management systems and grazing-intensity gradients (GIGs).
• To compare the effectiveness of continuous versus rotational grazing in mitigating grazing pressure based on livestock behavior.
• To provide data-driven insights for sustainable grazing strategies.

Main Methods:

• Utilized high-resolution GPS tracking data of livestock in Wenchang, China.
• Employed machine learning classification, specifically K-Nearest Neighbours (KNN) with SMOTE-ENN resampling, for accurate behavior analysis.
• Analyzed spatial clustering and temporal concentration of livestock activities across varying GIGs.

Main Results:

• The KNN model achieved high accuracy (F1-scores of 0.960 and 0.956) for classifying grazing patterns.
• Continuous grazing showed persistent spatial clustering and increased temporal peaks in activity even at reduced intensity.
• Rotational grazing demonstrated more even temporal activity and reduced spatial clustering with decreased GIGs.

Conclusions:

• Livestock spatial behavior patterns are critical indicators of grassland degradation.
• Rotational grazing is more effective than continuous grazing in managing grazing pressure through behavioral distribution.
• Integrating livestock behavior analysis into monitoring provides valuable insights for sustainable grassland management.