Machine learning for modeling animal movement.

Area of Science:

• Ecology
• Computational Biology
• Movement Ecology

Background:

• Animal movement is crucial for ecological processes like migration and disease spread.
• Current models often use parametric approaches, which can have restrictive assumptions.
• Machine learning and deep learning have been underutilized in animal movement analysis.

Purpose of the Study:

• To present a general framework for predicting animal movement using machine learning and deep learning.
• To compare the performance of various machine learning models against a Stochastic Differential Equation (SDE) model.
• To assess the generalizability of the framework across different species and movement scales.

Main Methods:

• A two-step framework: predicting behavioral states, then predicting velocity.
• Utilized Random Forests, Neural Networks, and Recurrent Neural Networks (including LSTM).
• Compared predictions against a custom Stochastic Differential Equation (SDE) model using ant and gull movement data.

Main Results:

• Individual-level machine learning/deep learning models outperformed the SDE model for one-step-ahead predictions.
• The SDE model showed better performance in long-range movement simulations.
• Long Short Term Memory (LSTM) individual-level models achieved the best results in long-range simulations.

Conclusions:

• Machine learning and deep learning offer flexible and powerful tools for animal movement prediction.
• Model choice depends on study goals; ML/DL are advantageous for predictive tasks.
• The proposed framework demonstrates broad applicability in movement ecology.