Author: Komárek J., Rous J., Klouček T. (2025). The bigger, the better? Sample size effects in drone-estimated forest height. Journal of Forestry Research. In press.
Forests are central to the Earth system. They harbour biodiversity, regulate water and nutrient cycles, protect soil, and store vast amounts of carbon. Their structure is a key parameter shaping biomass, canopy dynamics, ecosystem services, and resilience to disturbance. Taller trees often represent larger carbon pools, but they also complicate how we monitor carbon turnover and forest productivity. With global forests facing mounting pressures from climate change, land-use conversion, and increasing disturbance regimes, accurate measurements of their structure are more critical than ever.
Why Tree Height Matters
Tree height is more than a simple dimensional attribute. It directly links to above-ground biomass, productivity, and forest resilience. Errors in height estimation propagate into carbon stock assessments, biodiversity models, and evaluations of forest services such as flood regulation or protection against natural hazards. In other words, the quality of height data influences how we manage and conserve forests in practice.
While satellites and conventional aircraft cover large extents, drones offer unparalleled spatial and temporal resolution at local to regional scales. They are relatively flexible and increasingly integrated with multispectral cameras, lidar, and even thermal sensors. This technological progress enables single-tree level detection, canopy reconstructions, and detailed monitoring of structural and spectral change. Yet, drone operations face familiar constraints like limited payload and endurance, heavy data volumes, and sensitivity to weather and lighting conditions.
Recent studies highlight how different drone-based methods can achieve impressive accuracies, often comparable to traditional field measurements. Image matching and drone-mounted lidar are the two most common approaches. Image matching reconstructs canopy surfaces from overlapping photographs, while lidar penetrates through gaps to capture canopy and sub-canopy structure. Both deliver height estimates, but their performance depends heavily on site conditions and sampling design.
Insights from a Comparative Review
To better understand the influence of study design and site type on drone-based tree height estimation, we reviewed 36 peer-reviewed case studies worldwide. Each reported sample size, site type, method, and quantitative error metrics.
Several patterns emerged:
- Sample size matters. Small studies (<150 trees) showed a mean error of 1.12 m but with large variability. Intermediate studies (150–300 trees) performed worse (1.37 m) with highest variability (SD 1.00 m). In contrast, large datasets (>300 trees) achieved stable errors with the lowest variability (SD 0.45 m). Larger samples appear to buffer outliers and stabilise performance.
- Site type strongly influences accuracy. Natural forests, with their irregular canopies and species diversity, consistently exhibited higher errors (mean 1.41 m). Cultivated plantations performed better (mean 1.21 m), while urban greenery (often more homogeneous) showed the lowest errors (mean 0.92 m).
- Sensors differ in structured vs. complex environments. In plantations, lidar outperformed image-based methods (mean error 0.44 m vs. 0.91 m). However, in structurally complex natural forests, the difference between lidar (1.58 m) and photogrammetry (1.69 m) was marginal.
These findings emphasise that both methodological choices and environmental complexity drive performance.
Practical Implications for Drone Remote Sensing
The review highlights several operational lessons:
- Robust study design is essential. Larger sample sizes and replication across conditions increase reliability. Small, single-site studies may overestimate accuracy or underestimate variability.
- Site context must guide technology choice. While photogrammetry suffices for open, homogeneous canopies, lidar becomes indispensable in plantations or when full vertical structure is required. In dense natural forests, both methods face challenges from occlusion and structural heterogeneity.
- Weather and field conditions remain limiting. Drones are highly sensitive to light, wind, and precipitation. Even with advances in endurance and automation, long-term monitoring programmes must account for real-world operational limits.
- Standardisation is urgently needed. Without common protocols for data acquisition, processing, and error reporting, cross-study comparisons remain difficult. Transparency in workflows is critical for building robust, comparable datasets.
Moving Forward
Drone technology has transformed the scale and detail of forest monitoring, bridging the gap between field surveys and satellite observations. Yet, our synthesis shows that accuracy is not universal. It depends on sensor choice, site complexity, and study design. By integrating drones into carefully planned, standardised monitoring frameworks, while acknowledging their limitations, remote sensing researchers and practitioners can generate consistent, reliable data for forest management and climate reporting.
Ultimately, drones are not replacing satellites or fieldwork, but they are reshaping how we connect the dots between canopy detail and global change. In forestry, where every metre of tree height carries ecological and economic weight, this technological shift is nothing short of transformative.
