Bridging the Raunkiæran Shortfall through specimen-anchored, AI-derived trait infrastructure

Abstract

Biodiversity informatics has substantially improved access to taxonomic names and species occurrences, yet functional trait knowledge remains severely constrained by the Raunkiæran shortfall: traits are sparse, weakly standardized, rarely specimen-anchored, and poorly suited for inference under rapid global change. We argue that this limitation is not primarily a data-availability problem, but an infrastructure problem - specifically, a persistent gap between digitized biodiversity evidence and reproducible trait synthesis. Here, we outline a practical pathway for bridging this gap grounded in a coherent project lineage. FishNet 2 provides a specimen-backed foundation for geographic discovery and distributional inference. FishAIR establishes quality-aware, standardized, and AI-ready representations of specimen imagery. Biology- and hierarchy-guided machine learning, developed through BGNN and Imageomics, enables interpretable, scalable phenomic extraction. Specimen-anchored synthesis efforts, including Morphological Barcoding and the Global Fish Trait Atlas, translate extracted phenotypes into population-aware trait reference models that preserve provenance, intraspecific variation, and uncertainty. We further situate this architecture within a broader observatory framework, the Living Environmental Phenomics Observatory, and generalize its principles through AIDEN-SD as a model for interoperable, AI-ready data products. Together, this lineage reframes traits as infrastructure-derived ecological observables, enabling trait-based ecology, conservation, and global change ecology, including climate-impact assessment, at scales commensurate with rapid environmental change.