Key Results
- 303 autonomous flights across three DJI drone models over the same 50 ha area
- Consistent 5 cm target ground resolution and 80-90% overlap on every mission
- 84 RGB dates and 57 multispectral dates spanning 3.5 years, October 2020 to May 2024
- A seasonal greenness signal resolved at tree, species, and stand level
Goal: tracking individual tree crowns across seasons in dense tropical forest
Tropical forests drive a large share of the global carbon balance, and they are the least well documented. To connect what happens in a single tree crown to broad climate models, academic researchers need high-resolution imagery of the same forest, repeated over years.
Satellites cannot do this in the equatorial tropics, where cloud cover wipes out most optical passes, and the resolution is too coarse to isolate one tree. Drones can, but only if every flight lands the camera over the same ground at the same resolution. Miss that and you cannot compare one date to the next.
The site is a demanding one: a dense rainforest at Paracou, French Guiana, with about 3000 mm of rain a year and 1617 mapped tree crowns across the 50 ha study area.
Why repeatable drone acquisition was the constraint
Comparing imagery across dates only works if the acquisitions match. The target here was a 5 cm ground sampling distance (GSD) with 80 to 90% image overlap, held across every flight and every drone in the study.
That is not a one-time setup. Full coverage of the 50 ha needed three flights per session, each around 20 minutes and several hundred images. The team ran this repeatedly across 3.5 years, with three different aircraft: a Phantom 4 Advanced and a Mavic 2 Pro for RGB, and a Phantom 4 Multispectral for calibrated bands. Each drone flew at a different altitude, between 100 and 150 m, to hit the same 5 cm resolution given its sensor.
They also worked without RTK (real-time kinematic) GNSS or ground control points (GCPs). Deploying reference targets above a closed canopy is not practical, and dense foliage blocks RTK signal. So the acquisition side had to be clean and repeatable on standard positioning alone.
Planning 303 autonomous missions in UgCS
Every one of the 303 flights was planned and flown autonomously in UgCS drone flight planning software.
The software held the mission parameters constant: the 5 cm target resolution, the 80 to 90% overlap, and the flight lines over the 50 ha area, repeated across dates and across all three drones. Because each aircraft needed a different altitude to reach the same GSD, UgCS let the team configure resolution and overlap per platform and fly the survey mission hands-off, which kept the whole dataset comparable.
The high overlap served two purposed It is required for stereophotogrammetric 3D reconstruction, and it built the image redundancy the team relied on later to detect and fix small spatial shifts between survey dates. UgCS delivered consistent, dense, repeatable coverage, and it is the foundation the rest of the analysis was built on.
The spatial and spectral corrections themselves were done in post-processing, using Agisoft Metashape with the Time-SIFT and Arosics methods, which pulled residual misalignments between mosaics down to a few centimeters. Those tools need a stable, redundant set of acquisitions to work on, which is what UgCS missions produced.
Result: a four-year phenology signal at tree level
The repeated coverage turned into a usable biological record.
Across the long time series, 84 RGB dates and 57 multispectral imagery dates, the team resolved a clear seasonal greenness cycle at the stand level, peaking in the dry season. They tracked defoliation and leaf flush on individual crowns of species like Parkia nitida and Recordoxylon speciosum, and even caught a single branch cycling out of sync with the rest of its tree.
For equipment choice, the useful finding is that cheaper RGB sensors produced usable greenness data, not just the dedicated multispectral unit. Tree-level and stand-level phenology monitoring is achievable in dense tropical forest with consumer drones, provided the acquisition stays consistent across years.
Technical Specifications
What consistent acquisition made possible
The science here depended on comparing the same forest across years, at a resolution fine enough to see one tree change. That is only possible if every flight matches the last.
UgCS handled that side. It flew 303 autonomous missions over the same 50 ha with three different consumer DJI drones, holding a 5 cm target resolution and 80 to 90% overlap on each one, for 3.5 years. The consistency of that coverage, and the image redundancy it built in, let the team detect a seasonal phenology signal at tree, species, and stand level using off-the-shelf equipment.
Repeatable autonomous acquisition, run at scale over long periods, is the job UgCS did in this study, and it made the four-year record possible.
Reference
Barbier, N., Ploton, P., Tulet, H., Viennois, G., Leblanc, H., Burban, B., Réjou-Méchain, M., Verley, P., Ball, J., Feurer, D., & Vincent, G. (2026). Monitoring tropical forests with light drones: ensuring spatial and temporal consistency in stereophotogrammetric products. ISPRS Open Journal of Photogrammetry and Remote Sensing, 19, 100114. https://doi.org/10.1016/j.ophoto.2025.100114
Licensed under CC BY 4.0.