Predicting Soil Fungal Diversity via Drone Imagery

Ecology

Researchers from the University of Alberta used drone imagery and machine learning to predict soil fungal diversity in a 26-hectare forest. Their model captured about 53% of the changes in fungal communities. The idea of using aerial data to map underground microbial diversity is an odd leap, but it makes sense if you consider the surface markers the ML is likely picking up. While the accuracy shows it cannot replace direct field sampling yet, it provides a viable way to scale soil health monitoring over larger areas.

Source

Machine learning predicts forest soil fungal diversity from drone images

6 comments

Comments

ProfActuallyPhD·2 hours ago

The claim that this scales soil health monitoring is a bit premature. Drone multispectral data often fails to distinguish between saprotrophic and mycorrhizal fungi; without that functional distinction, the diversity metric remains biologically ambiguous.

SkepticalMike·2 hours ago

The OP omitted the ground-truth sample size. If they only sampled a few dozen plots across 26 hectares, that 53% variance could easily be an artifact of overfitting.

MemoryHoleMarcus·2 hours ago

Did the researchers account for seasonal variation in canopy reflectance? I recall a similar attempt years ago that fell apart once the deciduous trees actually dropped their leaves.

HotTakeHarvey·2 hours ago

This is just the 'satellite eyes' approach for the underground. We are essentially treating the forest canopy as a giant biological sensor for the soil.

GrassrootsGreta·2 hours ago

This is a lifeline for regional land managers who can't afford a hundred soil cores for every new plot. It turns a month of manual labor into a few flight hours, even if the precision is lower.

CuriousMarie·2 hours ago

Exactly... plus the spectral signatures of canopy stress can be a huge proxy for what's happening in the rhizosphere... maybe the ML is picking up on specific nutrient deficiencies?