AI-assisted synthesis reveals hidden cross-scale degradation in global mangrove ecosystems under human pressure
Year: 2026
Type: preprint
Abstract: Recent satellite evidence shows that global mangrove extent, after a net loss of ~ 2,900 km² between the 1980s and 2010, has since rebounded by ~ 2,000 km², largely through natural seaward expansion and regeneration. Yet this spatial recovery does not necessarily imply the recovery of biodiversity or ecosystem functioning. Satellites record where mangroves grow and species inventories record which organisms persist, but neither captures how plant physiology or metabolism respond to human and climatic pressures beneath the canopy. These responses are measured in thousands of local field and experimental studies, but they remain scattered and unsynthesized at global scale, and pooling them manually has reached practical limits in scalability and traceability. Here, using a human-in-the-loop Retrieval-Augmented Generation (RAG) pipeline, we screened 9,097 articles and built the Quantitative Global Mangrove Biodiversity Risk Database (QGMBRD), compiling 343 independent effect sizes (log response ratios of impacted versus reference conditions) from 6,744 driver-specific studies, spanning landscape to microbial scales under climate change, land-use change, pollution and biological invasion. Despite a nearly fivefold increase in mangrove biodiversity research between 2001 and 2025 (from 1,320 to 6,288 publications). The spatial multilevel meta-analysis further reveals consistent negative responses from macroscopic landscapes to microscopic processes. Crucially, relying solely on traditional taxonomic metrics can yield false-negative conclusions of stability where evidence is scarce and noisy. By contrast, functional metabolic and physiological indicators robustly capture significant degradation and reveal a heavier tail of extreme decline—28.8% of functional observations exceed critical degradation thresholds (|LnRR| > 1), compared with 20.5% for taxonomic metrics. Thus, recent spatial recovery should not be equated with recovery of ecological integrity, global monitoring frameworks must integrate cross-scale functional indicators to diagnose hidden biological impairment. Broadly, our human-centric, AI-augmented framework improves the scalability and traceability of cross-scale ecological evidence synthesis.
Source: Research Square
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225788525