From Drone Imagery to Ecological Intelligence
What a Sagebrush Study Tells Us About the Future of Forestry, Restoration and Biodiversity Monitoring
For years, one of the easiest ways to explain the value of drones in forestry and land management was simple: they let us see more.
More area. More detail. More frequently. Often at a lower cost than traditional aerial surveys.
But a recent study published in Rangeland Ecology & Management points toward something considerably more interesting.
The future may not be about seeing more.
It may be about measuring more of what matters.
Researchers studying big sagebrush (Artemisia tridentata) asked a deceptively practical question: could an ordinary RGB drone help identify individual plants with high reproductive potential, allowing restoration teams to concentrate seed collection on the most productive plants?
The answer was a qualified yes.
And buried inside that answer are some important lessons for forestry, biodiversity monitoring, restoration and the growing use of artificial intelligence in natural resource management.
The Problem Starts With Seed, Not Technology
Large-scale restoration isn't simply a matter of having enough people or equipment to put plants in the ground.
You need the right biological material in the first place.
For native species, that can become complicated quickly.
Plants can be highly adapted to local environmental conditions, meaning seed collected from a distant or genetically mismatched population may establish poorly, survive at lower rates or perform differently over the long term. The researchers describe identifying highly reproductive local plants as an important step in building resilient restoration programs.
Big sagebrush makes an especially interesting case.
It is a foundational species across large portions of the western United States and plays an important role in biodiversity and ecosystem function. At the same time, sagebrush ecosystems face pressure from wildfire, climate change, invasive species and development.
Finding productive seed sources across vast, remote and environmentally complex landscapes is therefore not a small logistical problem.
So the researchers turned to drones.
But what they extracted from the imagery is more interesting than the imagery itself.
An RGB Camera Became a Measurement Tool
The study used a DJI Mavic 2 Pro equipped with a conventional 20-megapixel RGB camera.
The researchers flew low-altitude missions with substantial image overlap and processed those photographs using structure-from-motion, or SfM.
Instead of ending with a nice aerial photograph, overlapping images were reconstructed into three-dimensional information about the vegetation.
They generated point clouds and canopy height models, segmented individual sagebrush crowns, and extracted structural and spectral measurements from those plants.
In other words, the photograph became data.
The researchers could begin examining characteristics such as plant height, crown geometry, variation within the canopy and changes in structure, then compare those measurements with flower stalk counts collected in the field.
That distinction matters enormously.
We often talk about drones as image-collection platforms.
Increasingly, that undersells them.
A properly designed drone survey can become a repeatable measurement system.
Structure Told a Bigger Story Than Colour
One of the most interesting findings was that structural characteristics of the sagebrush were more influential than basic colour information when predicting flower production.
The model was not simply asking whether a plant looked greener, darker or different from the vegetation surrounding it.
Characteristics derived from three-dimensional structure, including canopy height and crown geometry, contained meaningful information about reproductive output.
That's an important concept well beyond sagebrush.
In forestry, the operational question might not be:
What colour is the vegetation?
It might be:
How tall is it?
How dense is it?
How is height distributed across the block?
Where is regeneration succeeding?
Where is competition increasing?
Where has vegetation changed since the previous survey?
Where is mortality concentrated?
How has the stand responded to treatment, harvesting, drought, fire or another disturbance?
Those are fundamentally different questions.
And increasingly, they can be answered by extracting structure, position, pattern and change from remotely sensed data.
This Is Where Drones Begin to Intersect With Biodiversity
Biodiversity is frequently discussed at a very high level.
Protect biodiversity.
Improve biodiversity.
Monitor biodiversity.
But managing ecological complexity requires something much more practical:
the ability to observe where that complexity exists and how it changes.
A field ecologist or forester can collect extraordinarily detailed information on the ground, but there is an unavoidable constraint on how much landscape can be physically sampled.
Satellite imagery can cover enormous areas, but individual plants and fine vegetation structure can disappear at that scale.
High-resolution drone surveys occupy an increasingly useful space between the two.
That doesn't mean an RGB drone can somehow "measure biodiversity" with a single flight.
It can't.
Species composition, genetics, wildlife use, soil conditions, ecological interactions and many other dimensions of biodiversity may require completely different data and expertise.
But drones can help provide another piece of the puzzle: spatially explicit evidence about vegetation structure, distribution and change.
And that can matter enormously.
The sagebrush study illustrates the point particularly well because the researchers weren't simply trying to determine where sagebrush existed.
They were trying to identify differences between individual plants within the population.
Which plants were reproducing?
Which were producing more?
Where were those individuals located?
Could managers use those differences to improve restoration decisions?
That is a much richer ecological question than presence or absence.
But Then the AI Hit a Wall
This may be the most important part of the study.
The researchers ultimately retained 561 plants that could be reliably matched between field measurements and drone-derived canopy segments. Flower production ranged from zero to 709 stalks per plant.
Their best model achieved a mean absolute prediction error of approximately 100 flower stalks.
Given that some plants produced more than 700 stalks, that's promising.
It isn't magic.
And the researchers didn't pretend that it was.
They deliberately tested whether the model could generalize beyond the conditions on which it had been trained.
When tested against a completely different site, performance declined.
But when the researchers tried predicting an entirely different year, performance deteriorated dramatically.
Median prediction error increased from approximately 108 flower stalks to 354, accompanied by enormous uncertainty.
That's not a footnote.
It's a lesson.
Nature Moves. Models Have to Move With It.
There is an understandable temptation surrounding artificial intelligence to treat "accuracy" as if it were a permanent characteristic.
A model achieves a particular accuracy today, so that number becomes a feature of the model.
Ecological systems don't work that way.
Rainfall changes.
Temperature changes.
Drought happens.
Fire happens.
Vegetation grows.
Competition changes.
Disease appears.
Phenology shifts.
Management interventions alter the landscape.
The relationship between what a sensor observes and what is happening biologically can change with those conditions.
The authors themselves concluded that applying UAV models to new years may require additional field validation and training data.
That isn't a weakness of drone analytics.
It's a reminder of what responsible ecological analytics should look like.
AI should support field knowledge, not pretend that field knowledge has become unnecessary.
The Same Caution Applies to Seed
There's another important limitation in the study.
Flower production isn't seed production.
A plant producing a large number of flower stalks does not automatically produce the largest quantity of viable seed.
Pollination, seed maturation, genetics, stress tolerance and other biological factors still matter.
The researchers specifically warn that selecting plants based solely on fecundity could favour characteristics that trade off against other useful adaptations.
Again, that's useful.
The strength of remote sensing isn't that it eliminates ecology.
Its strength is that it can help ecologists, foresters and land managers target their attention more intelligently.
Instead of inspecting everything equally, analytics can help identify where the questions are.
One Flight Is a Snapshot. Repeated Flights Become a Record.
This leads to what may ultimately be the bigger opportunity for forestry and land management.
Think about what happens when the same landscape is surveyed repeatedly.
Year one gives you a snapshot.
Year two gives you change.
Year three begins to give you trajectory.
After several years, the organization may possess something much more valuable than a collection of orthomosaics.
It begins building a spatial record of:
What was there.
What was done.
What changed afterward.
For forestry organizations, that historical record could support a range of operational questions.
Was regeneration successful?
Where is stocking below expectations?
How quickly is vegetation developing?
Which areas are responding differently?
Where is mortality occurring?
Did a silvicultural treatment produce the intended result?
How is a burned area recovering?
Where is competitive vegetation becoming a problem?
How is vegetation structure evolving along a utility corridor?
Where do conditions consistently differ from the rest of the block?
That is where the value of the dataset can begin compounding.
The important product is no longer a drone flight.
It's institutional ecological memory.
From Mapping Forests to Understanding Change
This is also where platforms such as Biodrone become particularly interesting to us at Canopy Dynamics.
The objective isn't simply to make another map.
It's to take properly collected aerial data and turn it into information that can support actual forestry and land-management workflows, whether that involves vegetation structure, stocking, regeneration, damage, treatment planning or documenting change.
And there is an important distinction here.
Technology should not replace the forester, ecologist, silviculture professional or restoration specialist making the decision.
Quite the opposite.
The most useful analytics should give those professionals better evidence about where to look, what has changed and where intervention may be warranted.
The domain expert still provides the context.
The technology expands what they can see.
Biodiversity Becomes More Measurable When Change Becomes Visible
There is also a larger environmental opportunity here.
Biodiversity monitoring doesn't necessarily require finding one universal biodiversity metric.
It may instead involve building multiple layers of evidence.
Vegetation structure.
Species distribution.
Habitat characteristics.
Regeneration.
Disturbance.
Recovery.
Mortality.
Landscape connectivity.
Treatment response.
Changes in those variables through time.
Not every layer will come from a drone. Some will come from field observations, LiDAR, satellites, genetics, wildlife surveys, climate information or other sources.
But high-resolution drone-derived measurements can increasingly contribute to that larger picture.
And perhaps more importantly, they can make those observations geographically explicit.
We don't simply know that something changed.
We can begin to know where it changed.
That distinction turns monitoring into management.
The Photograph Is No Longer the Product
A relatively inexpensive RGB drone helping researchers identify potentially high-producing sagebrush plants may initially sound like a narrow scientific application.
It isn't.
It represents a much broader transition already occurring across forestry, agriculture, conservation and land management.
We are moving from:
imagery to measurement,
from measurement to interpretation,
and from interpretation to decisions.
The sagebrush study also provides a useful warning against getting carried away with the technology.
The model wasn't universally transferable.
Not every plant could be reliably segmented.
Flower stalks didn't guarantee viable seed.
Field validation remained important.
Those limitations don't make the research less compelling.
They make it more useful.
Because the future of drone-based ecological intelligence probably won't be built around one perfect sensor or one permanent AI model.
It will be built around repeatable data collection, validated analytics, multiple sources of evidence and datasets that improve our understanding of landscapes over time.
The drone is becoming less important than the record it helps create.
And ultimately:
The photograph is no longer the product. The understanding is.
This article was inspired by “Drone-Based Monitoring of Reproductive Potential in a Foundational Shrub Species,” by Ryan Scott Wickersham and colleagues, published in the July 2026 issue of Rangeland Ecology & Management. The study evaluated the use of RGB UAV imagery and structure-from-motion analysis to estimate flower stalk production in big sagebrush.
