A 50-m eelgrass model turns national monitoring into restoration-scale maps
WHAT THE EVIDENCE SAYS
Using 37,497 national monitoring observations from 2011–2022, researchers trained a 50-m random-forest model for eelgrass across Denmark's roughly 7,000-km coastline. Independent test accuracy was 80.3% with AUC 0.878; modeled depth limits agreed with field observations across 85 water bodies (r=0.779), and light availability was the strongest predictor.
WHY IT MATTERS / OUR READ
Spatial models become more decision-useful when they predict an ecological threshold and show where confidence is high enough to guide field verification or restoration investment. The important product is not just a suitability map but an uncertainty-aware decision surface.
CAVEATS / UNCERTAINTY
The Danish-trained model is not directly transferable to other coasts without local observations and environmental predictors. It estimates distribution and restoration feasibility rather than guaranteeing establishment or persistence after planting.
SOURCE
Estuarine, Coastal and Shelf Science · Original evidence
View original evidence ↗
Denmark's machine-learning framework predicts eelgrass extent and depth limits with independent validation, while making uncertainty visible enough to support site-level planning.
Read the Wild Lands assessment →