Reproduce “SatBird: Bird Species Distribution Modeling with Remote Sensing and Citizen Science Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mélisande Teng, Amna Elmustafa, Benjamin Akera, Yoshua Bengio, Hager Radi Abdelwahed, Hugo Larochelle, David Rolnick
Mila researchers introduced SatBird, a novel dataset and task designed for predicting multi-species bird encounter rates using a combination of remote sensing and citizen science data. They demonstrated that integrating satellite imagery with environmental rasters through deep learning models significantly improves the accuracy of bird distribution predictions across diverse geographical regions.
Key insights
- Combining satellite imagery and environmental rasters is crucial, yielding a significant performance increase compared to using either data source alone.
- Deep learning models, particularly ResNet-18 with multi-modal inputs, effectively learn complex relationships for multi-species distribution modeling.
- Jointly predicting encounter rates for multiple species with a single model offers slightly better performance than training individual models, supporting the capture of ecological interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.