Mapping recovery: How drones could help rangelands bounce back from wildfire

High above the rugged terrain of Idaho and eastern Oregon, drones are helping scientists see the land in a whole new way, and it could transform how we heal landscapes scarred by wildfire.

Led by Boise State graduate student Maisha Maliha and Associate Professor Trevor Caughlin, in collaboration with the U.S. Department of Agriculture's Agricultural Research Service, researchers used drone imagery to classify plant species across vast terrain. Their study reveals drones technology's power to capture landscape-scale ecology, leading to more efficient and sophisticated land management practices.

Associate Professor Trevor Caughlin and graduate student Maisha Maliha collecting data at Castle Rocks State Park, Idaho.

Challenges to land management

Land management groups, such as the Bureau of Land Management, rely on accurate data about biodiversity and invasive species when making decisions. These decisions affect everything from ecosystem health and sustainability to rangeland productivity and livestock profitability.

"The fundamental challenge of rangeland management is that individual plants are important, but the spatial extent that we have to manage is huge," Caughlin said. "Native species help maintain the soil, support biodiversity and provide nutritious and resilient forage for livestock. Invasive species degrade these ecosystem services and promote wildfire. So knowing exactly what plants are out there is really important."

"Individual plants are important, but the spatial extent that we have to manage is huge."

Trevor Caughlin

Historically, ecologists and land stewardship organizations have relied on satellite imagery to get a bird's-eye view of vegetation across large landscapes. However, satellite images lack the resolution necessary to identify individual plant species - information that is vital for early detection of invasive species and understanding forage availability for grazing. Thankfully, machine learning and drone technology are now poised to transform rangeland management by bringing precision agriculture to the open range.

Drones: a cost-effective alternative

Unoccupied Aerial Vehicles (UAVs), commonly known as drones, come in a wide range of models, with prices ranging from $30 for recreational "toy" drones to hundreds of millions for military defense drones. Commercial drones entered the market in 2006 and have evolved considerably since then. Those used for mapping plants in the wild and for agricultural applications cost anywhere from $4,000 to $60,000.

Boise State graduate students Maisha Maliha and Anna Roser conducting fieldwork.

The price disparity among commercial drones is largely due to the spectral resolution of their sensors. Think of drone imaging as drawing a landscape with a box of crayons. A standard, base-level drone works with three colors: red, green and blue (hence, RGB sensors). RGB sensors can draw the picture, but definition is limited. Multispectral sensors operate more like a curated, problem-specific eight-pack of crayons; drones with these offer more definition at a mid-range price. At the high end of the spectrum are hyperspectral sensors - a full pack of crayons capable of capturing hundreds of shades and textures. Unfortunately, hyperspectral sensors are extremely expensive and require immense data processing, making them less accessible.

Thankfully, when it comes to plant identification, it turns out that you don't need the full box of crayons to accurately portray rangeland ecology.

Findings: 90% accuracy for a fraction of the cost

Maliha and Caughlin's study shows that drones with RGB sensors are capable of successfully identifying 18 sagebrush steppe plant varieties with more than 90% accuracy, making it one of the most comprehensive UAV rangeland studies. Their models even distinguished similar species, such as low sagebrush and big sagebrush.

RGB imagery models also performed nearly as well as their multispectral counterparts, indicating that even lower-cost models offer an efficient solution to land management challenges.

"This study shows how machine learning and UAV imagery can be used to support biodiversity mapping and improve our ability to understand and monitor plant communities," said Maliha, now a data science faculty member at Montgomery College in Maryland. "These findings are especially meaningful because they demonstrate that relatively accessible UAV data, combined with data science methods, can provide a practical approach to ecological research and monitoring."

From imagery to action: guiding post-wildfire restoration

These findings have significant implications for modern land management challenges, including fire restoration.

For example, the Big Grass Wildfire that started in July 2026 along the Oregon-Idaho border has damaged more than 570,000 acres, making it the largest wildfire in Oregon's history. Though nearing total containment, more than half a million acres of scorched rangeland remains across two states. Drone technology can help researchers monitor restoration efforts and track plant recovery across this vast area.

"By mapping plant species and clusters with drones, and then flying drones over areas that are recovering from wildfire, we can help make strategic decisions about where we need to target effort in dropping seed or planting new plants to be able to help the ecosystem recover after a wildfire," Caughlin said.

What's next: scaling the model for Idaho and beyond

Though extensive field training data across a variety of ecosystems will be necessary to improve this technology's transferability, Caughlin recently won a National Science Foundation fellowship that will allow him to continue this work. This fall, he will collaborate with the University of New Mexico's Sevilleta Long-Term Ecological Research Station in a distinct rangeland landscape to refine models for species classification.

Excited by this opportunity, Caughlin said, "Having a diverse array of sites and environmental conditions across the Western United States will help us build a generalizable model that can identify plant species in Idaho and beyond."

Share This

• Share on Facebook
• Share on X
• Share on LinkedIn
• Share through Email