Birdflow is a modeling framework for estimating individual movement trajectories from population level data. The model takes in eBird Status & Trends abundance estimates and proxy for energetic cost and outputs a time heterogeneous Markov chain which estimates the track distribution of the given species.
This codebase allows for the gpu-accelerated training and use of Birdflow models in python and gives an example of how to process an abundnce estimate so that it can be used with Birdflow. For a more in depth discussion of the model and experimental results for 11 species in North America, see our paper here.
The code is written in python 3.X and depends on the following libraries:
- jax
- haiku
- optax
- scipy
It should be up to date with the latest versions of those libraries
For an example of how to run the code, see the jupyter notebook birdflow_demo.ipynb
For any questions or help, contact Miguel Fuentes at mmfuentes@umass.edu
This project is licensed under the MIT License - see the LICENSE file for details