| Authors | Abreu-Dias |
| Paper (PDF) | View paper |
| Category | Planning & Inventory › Remote sensing & mapping |
0
Automated individual tree species identification using drones and artificial intelligence represents one of the most rapidly advancing frontiers in precision forestry, combining remote sensing, computer vision, and machine learning to deliver spatially explicit species maps at stand and landscape scales. Traditional field-based species inventories are labour-intensive, time-consuming, and practically limited in spatial coverage, making scalable automated alternatives increasingly attractive for both commercial forestry and conservation. Drone-based platforms — particularly fixed-wing and multirotor UAVs equipped with RGB, multispectral, hyperspectral, or LiDAR sensors — can collect fine-resolution canopy data at low cost compared to manned aircraft or satellite systems. These data are then processed through AI pipelines that typically combine individual tree crown (ITC) delineation algorithms with species classification models. Convolutional neural networks (CNNs), including architectures such as ResNet, EfficientNet, and vision transformers, have shown strong performance when trained on labelled crown-level image patches, while ensemble approaches fusing spectral and structural features from LiDAR point clouds with optical imagery further improve classification accuracy, particularly for spectrally similar species. Systematic reviews of this field highlight several recurring challenges: training data scarcity for rare species, transferability of models across sites with different canopy structures, seasonal variation in spectral signatures, and the difficulty of separating co-dominant species in dense, mixed-species forests. In Australian forestry contexts — including native hardwood management in Victoria and Queensland, and softwood plantation monitoring in New South Wales and South Australia — automated species identification supports timber volume estimation, health monitoring, and biodiversity assessments. Species such as mountain ash (Eucalyptus regnans), alpine ash, radiata pine (Pinus radiata), and a range of native eucalypt species have been the focus of UAV-based classification research within Australia. Progress in this field is closely tied to improvements in sensor miniaturisation, open benchmark datasets, and the adoption of transfer learning techniques that allow models pre-trained on large datasets to be fine-tuned for local species assemblages with limited labelled samples.
This tool appears promising because it provides faster, more accurate, and lower-cost species identification that can support better forest management, but it is not yet a proven tool. Benefits remain mostly indirect and real-world risk reduction has not been clearly demonstrated, meaning adoption-risk and uncertainty still limit its suitability
| Field | Value |
|---|---|
| Company | Abreu-Dias |
| Type | Paper |
| Year | 2025 |
| FWPA RD&E | 7.3 |
| Remarks | 0.0 |
| Case Studies | 0 |