STELLAR: Spatio-Temporal Environmental Learning with Latent Alignment and Refinement for Long-Tailed Species Distribution Modeling
Researchers have introduced STELLAR, a new framework designed to improve Joint Species Distribution Modeling (JSDM) by addressing spatio-temporal dynamics and the imbalance of rare species. The STELLAR model integrates a Graph-Temporal Encoder, a Context-Anchored Latent Alignment mechanism, and an Imbalance-Aware Decoupled Decoding module. Experiments using the eBird dataset show STELLAR significantly outperforms existing methods, particularly in predicting rare species and understanding species interactions. AI
IMPACT Improves ecological modeling accuracy for rare species, aiding conservation efforts.