Distributional ecology: A word-embedding AI framework for multi-level ecological prediction and species persistence

This repository provides the full computational workflow for the Distributional Ecology Framework, demonstrated using the Barro Colorado Island (BCI) forest dynamics plot in Panama. The framework adapts the distributional hypothesis from natural language processing to ecology by treating species as “words” and communities as “sentences,” enabling multi-level ecological prediction through word-embedding models. Using large forest plot data, the framework learns species representations from co-occurrence patterns and applies them to: Functional trait prediction Environmental gradient inference Community-level environmental reconstruction Similarity-based occurrence and persistence analysis Species embeddings capture ecological information across biological levels, linking species persistence to their community context. Community embeddings reliably reconstruct soil and topographic variables and reveal relationships between embedding similarity, species abundance, and mortality. This repository reproduces the main analyses corresponding to Figures 2–4 of the manuscript: Distributional ecology: A word-embedding AI framework for multi-level ecological prediction and species persistence The workflow includes: Data preprocessing pipelines Species- and community-level embedding analyses Jupyter notebooks demonstrating the full pipeline Supporting Python modules R scripts for weighted averaging comparison Although demonstrated on the BCI forest plot, the workflow is generalizable to other forest datasets that follow a similar structure. ⚠️ The inital data are not redistributed due to data use agreements and must be requested directly from the respective data providers which mention in readme.md. This repository provides a scalable, data-driven foundation for applying distributional ecology to community prediction, environmental reconstruction, and species persistence analysis across ecological systems.