Combining drone data and machine learning can help cover a lot more ground in monitoring forest soil health, University of Alberta research shows.
Using both tools to map and monitor soil fungal diversity - a key indicator of a healthy forest ecosystem - proved highly effective and could help ease the need for taking boots-on-the-ground soil samples over huge areas of forest, says Dr. Cameron Carlyle, a professor in the Faculty of Agricultural, Life & Environmental Sciences and a co-author on the study.
"To understand patterns across the landscape relies heavily on manual soil collection and DNA sequencing across many locations, which means it's labour-intensive and expensive," he notes. "But by integrating remote sensing data - information collected from drones - with soil measurements and machine learning, the research provides a more cost-effective and scalable way to map out fungal soil diversity."
Keeping track of that diversity is vital, adds Wanwan Yu, who led the research as a visiting PhD researcher in Carlyle's lab.
"Soil fungi are central to forest function, influencing nutrient cycling, decomposition and tree growth," she notes. "Without that monitoring, we risk losing critical information on biodiversity patterns, ecosystem stability and the ability to manage and conserve forests effectively, particularly in the face of environmental change."
The study, conducted in a 40-year-old planted forest located in a nature reserve in China, looked at alpha diversity - the number of different fungal species living in one specific spot - and beta diversity - how much the types of fungal species change from one area to another across the forest.
The team collected 538 soil samples from a 26-hectare area and used DNA sequencing to identify the different types of fungi present. Drones were also flown over the forest to collect high-resolution images, measure tree heights and track how much light the leaves reflected, gauging factors such as chlorophyll content and water levels.
The information was then processed by a smart computer program called a random forest model to test whether the drone data could reliably predict what was happening in the soil fungi community.
The results showed that host tree species, the landscape and soil properties jointly shaped fungal patterns.
Soil fungal diversity changes significantly depending on tree species, as well as among trees of the same species, by creating different micro-environments that shape the fungal community.
Nor was there one single driving factor in soil fungal diversity. Rather, the variety and the shifting patterns of fungal species across the forest were influenced by different combinations of environmental factors, the study showed.
The research also revealed that machine learning was successful in predicting about 53 per cent of the beta diversity of fungi, and did a moderate job of predicting alpha diversity, ranging from 28 to 45 per cent, depending on the specific measurement used.
While the combined high-tech approach used in the study can't completely replace more detailed, localized information provided only through hands-on field sampling of soil, it could improve the coverage and efficiency of monitoring general diversity patterns across huge, varied areas of forest, Yu suggests.
"It helps extend information from limited sampling points - which can miss important spatial patterns and vulnerable biodiversity hotspots - to broader landscapes that would be difficult or costly to monitor using traditional methods alone."
The work provides a practical way to include belowground biodiversity in forest restoration, conservation and long-term monitoring efforts, Carlyle adds.
"This tool allows forest managers, reclamation companies and conservation researchers to efficiently monitor underground soil health to support better long-term land management and restoration."
