Machine learning for modeling animal movement.

摘要: Animal movement drives important ecological processes such as migration and the spread of infectious disease. Current approaches to modeling animal tracking data focus on parametric models used understand environmental effects behavior fill in missing data. Machine Learning Deep learning algorithms are powerful flexible predictive tools but have rarely been applied In this study we present a general framework for predicting that is combination two steps: first behavioral states second animal's velocity. We specify at individual level well collective movement. use Random Forests, Neural Recurrent Networks compare performance one step ahead long range simulations. results against custom constructed Stochastic Differential Equation (SDE) model. apply approach high resolution ant found methods outperformed SDE model prediction. The did comparatively better simulating behaviour. Of Long Short Term Memory (LSTM) best also Forest LSTM gull migratory demonstrate generalizability framework. deep easier compared traditional which can restrictive assumptions. However, machine less interpretable than models. type should be determined by goal study, if prediction, our provides evidence could useful tools.