A Method for Avian Flight Behavior Pattern Mining Based on Avian Radar Data

This paper proposes a method for mining bird behavior patterns based on avian radar to address the lack of in-depth understanding of avian behavioral intent in traditional airport bird situation analysis. The method is based on real-world bird track data collected by a digital array avian radar. Multi-dimensional kinematic “track profiles” including average speed， average altitude， average trajectory straightness， root mean square of horizontal turn rate， and total trajectory duration are constructed. An unsupervised partitioning clustering method， which aims to minimize the sum of squared distances from samples to their assigned centroids， was then employed for pattern mining. The optimal number of clusters is determined to be five by combining the Elbow method and objective evaluation metrics such as the Calinski-Harabasz Index. At K=5， the Calinski-Harabasz Index reached a peak of 10263.09， strongly supporting this choice of K. Furthermore， to validate clustering effectiveness， a comparison with a Gaussian mixture model demonstrated the superiority of the selected method， both quantitatively and visually. The study identified five statistically distinct flight behavior patterns： mid-low altitude high-speed transit， low-altitude slow milling， mid-low altitude meandering movement， mid-low altitude long-duration loitering， and high-altitude migration/anomaly. By combining uniform manifold approximation and projection dimensionality reduction visualization with typical 3D trajectory reconstruction analysis， the existence and separability of these behavior patterns in the feature space were visually verified. This research elevates bird situation analysis from traditional target identification to an understanding of flight intent， providing a new perspective and technical support for fine-grained assessment and intelligent early warning of airport bird strike risks.