🧬 biology

An Information Theory Treatment of Animal Movement Tracks

This paper presents a novel fine-scale approach for analyzing animal movement tracks by applying Shannon's Information Theory to segment data into statistical movement elements and canonical activity modes, enabling entropy-based evaluation, coding efficiency assessment, and parameter optimization to better understand animal behavior in changing landscapes.

Original authors: Wayne M Getz

Published 2026-08-04

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Wayne M Getz

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine trying to understand a story just by looking at a map of footprints left in the snow. You can see where someone walked, but can you tell if they were running from a wolf, strolling to a picnic, or just wandering aimlessly? This is the challenge facing scientists who study how animals move. For decades, they've been able to track animals with GPS collars, recording their location every hour, every minute, or even every second. But knowing where an animal is doesn't automatically tell you what it's doing. To solve this, scientists use a field called "Information Theory." Think of this like a giant library of codes. Just as a computer translates a complex movie into a string of 0s and 1s to store it, scientists want to translate an animal's messy path into a clean code of "movement words." If we can crack the code, we can understand the animal's life story: when it's hunting, resting, or migrating, and how it reacts to changes in its world, like a new predator or a drying river.

This paper, written by Wayne Getz, proposes a new, super-detailed way to crack that code. Instead of looking at the whole path at once, Getz suggests breaking the animal's movement down into tiny, bite-sized chunks, much like breaking a long sentence down into individual letters, then words, and finally sentences. He calls the smallest chunks "Statistical Movement Elements" (or StaMEs). Imagine these as single steps. If you take a few steps at a time and group them together, you get a "word." If you group those words, you get a "sentence" that describes a specific behavior, like "grazing" or "sprinting."

The paper argues that the old ways of doing this—looking for sudden changes in speed or direction—are sometimes too slow or miss the tiny details. Getz's new method uses math from Shannon's Information Theory to measure exactly how much "information" is in the animal's path. It's like a quality control check for the code. The method involves six steps: chopping the path into tiny pieces, sorting those pieces into shapes (StaMEs), stringing them into words, and then grouping those words into "Canonical Activity Modes" (CAMs), which are the animal's main activities. The paper doesn't just say "this works"; it provides a way to measure how accurate the code is. It calculates an "error rate" to see how often the computer guesses the wrong activity, and it uses a special math tool called "Jensen-Shannon divergence" to compare different ways of grouping the data to find the best fit.

The findings suggest that by using this fine-tuned, second-by-second approach, scientists can get a much clearer picture of animal behavior than before. The paper shows that this method can handle high-resolution data (like GPS points taken every second) and can distinguish between different types of movement with high precision. It explicitly rules out the idea that we can always identify the exact biological muscle movements (like "flexing a leg") just from GPS data alone; instead, it focuses on statistical patterns that represent those behaviors. The paper is confident that this approach offers a rigorous new tool for the toolbox, allowing researchers to compare how different animals move and how their "information content" changes as they learn their environment or face new challenges. It doesn't claim to have solved every mystery of animal life, but it provides a solid, mathematical framework to start asking better questions about the stories hidden in the tracks.

Technical Summary: An Information Theory Treatment of Animal Movement Tracks

Problem Statement
The analysis of animal movement tracks has advanced from hourly to second-by-second position recordings. However, methods for segmenting these tracks to extract behavioral information have lagged, typically operating at sub-hourly to minute scales. Existing segmentation techniques, such as Behavioral Change Point Analysis (BCPA) and Hidden Markov Models (HMM), produce variable-length segments. This variability introduces ambiguity regarding the number of elements used to code specific track segments and may struggle to identify rare events or converge on solutions for high-resolution data with multiple behavioral modes. Furthermore, fundamental movement elements (FuMEs) are often unobservable in relocation data alone, requiring auxiliary data (e.g., video, accelerometers) for identification. There is a need for a rigorous, fine-scale approach that operates at the second-by-second level, standardizes movement elements, and provides a theoretical framework for measuring the information content and coding efficiency of movement tracks.

Methodology
The paper proposes a hierarchical, two-level segmentation framework grounded in Shannon's Information Theory. The approach transforms raw relocation data ( $T_{loc}$ ) into standardized statistical elements and higher-order activity modes through a six-task process defined by a functor $M$ :

Segmentation into Statistical Movement Elements (StaMEs): The relocation time series is divided into fixed-length segments of $\mu$ steps. These segments are clustered into $n$ distinct categories based on shape or vector statistics (e.g., speed, turning angle). The centroids of these clusters form a set of symbols $S = \{\sigma_i\}$ , creating a symbolic time series $T_\sigma$ .
Word Generation: Consecutive StaMEs are grouped into "words" of length $m\mu$ steps. This creates a set of words $W$ .
Clustering into Canonical Activity Modes (CAMs): The words are clustered into $k$ categories. The centroids of these clusters define a set of "raw" Canonical Activity Modes (CAMs), denoted $K = \{\kappa_c\}$ .
Rectification of CAMs: To improve coding reliability, words are assigned to "rectified" CAMs ( $\kappa^*_c$ ) based on the dominant cluster membership of their underlying StaME sequences. This step minimizes ambiguity by ensuring that words with identical StaME sequences are consistently assigned to the same CAM type.
Error and Entropy Calculation: The framework calculates the coding error rate ( $E_\kappa$ ) by measuring the proportion of words reassigned during the rectification process. It also computes information entropy ( $H$ ) for the StaMEs, words, and CAMs. The paper distinguishes between entropy assuming no autocorrelation ( $H_0$ ) and entropy accounting for first-order autocorrelation ( $H_1$ ).
Parameter Optimization: The method utilizes the Jensen-Shannon (JS) divergence to compare different clustering outcomes and parameter sets ( $\mu, n, m, k$ ). JS divergence measures the average divergence between the distribution of words within clusters and the overall word distribution, serving as a metric for the efficacy of the segmentation.

Key Contributions

Standardization of Movement Elements: The paper introduces Statistical Movement Elements (StaMEs) and Canonical Activity Modes (CAMs) as fixed-length, standardized coding units, replacing the variable-length segments of traditional BCPA and HMM methods.
Information Theoretic Framework: It applies Shannon's coding theory to animal movement, allowing for the calculation of entropy (information content) and coding efficiency for movement paths.
Quantifiable Error Rates: Unlike previous methods, this approach explicitly calculates the error rate ( $E_\kappa$ ) associated with assigning sequences of movement elements to behavioral categories.
Optimization via Divergence Measures: The use of Jensen-Shannon divergence provides a rigorous, symmetric metric for comparing different clustering algorithms and parameter choices to determine the optimal segmentation strategy.
Hierarchical Abstraction: The framework establishes a clear hierarchy from raw coordinates to StaMEs, then to words, and finally to CAMs, which can subsequently be parsed into Behavioral Activity Modes (BAMs) and Diel Activity Routines (DARs).

Results and Capabilities
The paper does not present a specific case study with empirical results but rather establishes the theoretical and computational machinery. It demonstrates that:

The entropy of a movement track can be quantified in bits, allowing for the comparison of information content across different individuals or conditions.
The coding efficiency of derived StaMEs and CAMs can be measured, enabling the selection of the most informative parameter sets ( $\mu, n, m, k$ ).
The rectification process (Task 5) reduces coding ambiguity, though it introduces a calculable error rate that reflects the reliability of the CAM assignment.
The framework is computationally more manageable than BCPA for large, high-resolution datasets and offers an alternative approach to segmentation that may better address challenges associated with identifying rare events, which variable-length methods like BCPA may struggle to detect.

Significance
The author claims this approach provides a new tool for the "toolbox" of movement ecology, specifically addressing the challenges of global change and altered landscapes. By enabling the analysis of movement at a second-by-second resolution, the method allows researchers to:

Rigorously compare the information content of movement tracks under varying internal (e.g., age, health, physiology) and external (e.g., landscape, weather, predation risk) conditions.
Formulate testable hypotheses regarding how an individual's information processing changes, such as whether information content increases with landscape learning or decreases due to parasitic infection.
Integrate auxiliary data (e.g., accelerometers, heart rate) to refine the identification of movement elements and reduce GPS observation errors.
Move beyond descriptive segmentation to a quantitative, information-theoretic understanding of animal movement behavior, facilitating the prediction of movement sequences and the interpretation of higher-order behavioral structures.

Technical Summary: An Information Theory Treatment of Animal Movement Tracks

More like this