Species–Community Contextual Fit Index (SCCFI): a word embedding-derived ecological indicator from co-occurrence data

Ecological indicators derived from standard biodiversity surveys are needed for monitoring community change where functional trait and environmental data are incomplete. Here, we propose the Species–Community Contextual Fit Index (SCCFI), an embedding-derived indicator that quantifies how well a species fits the ecological context represented by its local community. Inspired by the distributional hypothesis in natural language processing, we treated species as words and local communities as sentences, and used Word2Vec to learn species embeddings from neighbourhood co-occurrence patterns. SCCFI was calculated as the cosine similarity between a focal species embedding and the leave-one-out embedding of its surrounding community. We evaluated the framework in two forest dynamics plots: Dinghushan, China, and Barro Colorado Island, Panama. Species embeddings predicted functional traits and species-level environmental associations, indicating that contextual similarity in the embedding space captures ecological resemblance among species. Community embeddings reconstructed soil nutrient and topographic gradients more accurately than weighted averaging and count-based approaches, supporting their use as representations of local ecological context. SCCFI was positively associated with local abundance and negatively associated with mortality for most species in both forests. These results show that SCCFI translates co-occurrence-derived embeddings into an interpretable indicator of species–community contextual fit. Because SCCFI requires only co-occurrence data, it provides a scalable tool for biodiversity monitoring, early warning of population decline and conservation prioritisation in data-limited ecosystems.