segclust2d

segclust2d segments and clusters bivariate and multivariate time series to detect stationary phases in movement data for interpreting animal behavioral modes and home-range shifts.

Key Features:

• Segmentation and Clustering: Performs segmentation of time series into stationary phases and clusters those segments based on their statistical characteristics.
• Stationarity Assumption: Identifies phases assumed to be stationary, characterized by specific means and variances over limited time periods.
• Minimum Segment Length Parameter: Uses a user-specified minimum segment length to constrain segmentation and prevent over-fragmentation of biologically relevant phases.
• Multi-scale and Multivariate Input: Applies to bivariate coordinates and more generally multivariate time series, including derived metrics such as speed and turning angle.
• Comparative Performance: Demonstrates competitive performance versus methods based on hidden Markov models and Ornstein–Uhlenbeck processes and does not require initial parameter guesses.

Scientific Applications:

• Identify Behavioral Modes: Distinguishes small-scale behaviors (e.g., feeding, resting, transit) using derived movement metrics such as speed and turning angle.
• Detect Home-Range Shifts: Identifies temporary home ranges and shifts at larger spatial scales from bivariate relocation coordinates.
• Animal Movement Ecology (Biologging): Analyzes biologging time series to interpret movement processes and behavioral states in ecological studies.

Methodology:

Segments input time series into stationary phases, clusters segments by their characteristics, uses a user-defined minimum segment length to guide segmentation, assumes segments are defined by specific means and variances, and has been compared to hidden Markov model and Ornstein–Uhlenbeck process–based methods without requiring initial guesses.