Global, Multi-Scale Standing Deadwood Segmentation In Centimeter-Scale Aerial Images
Bibliographic record
Abstract
With tree mortality rates rising across many regions of the world, efficient methods to map dead trees are becoming increasingly important to monitor forest dieback, assess ecological impacts, and guide management strategies. Computer vision combined with the high spatial detail of aerial images from drones or airplanes provides an avenue for mapping dead tree crowns or partial canopy dieback, collectively referred to as standing deadwood. However, current methods for mapping standing deadwood are limited to specific biomes or image resolutions. Here, we present a transformer-based semantic segmentation model that generalizes across forest biomes and a wide range of image resolutions (1-28 cm) for mapping both dead tree crowns and partial canopy dieback. Our approach combines a SegFormer-based transformer architecture for image feature extraction and Focal Tversky Loss to mitigate class imbalance. We used a crowd-sourced dataset of 434 high-resolution aerial images and manual delineations of standing deadwood of vastly varying quality. The orthophotos span all major forest biomes and cover 10,778 hectares. To further mitigate imbalances across biomes, resolutions, deadwood occurrence, and image sources, we developed a four-dimensional sampling scheme that ensures balanced representation during training. Despite the heterogeneous nature of images and crowd-sourced labels, cross-validation results across this heterogeneous dataset demonstrated F1-scores between 0.34 and 0.57 across biomes, with the highest performance observed in temperate forests at approximately 4 cm resolution, and lowest in tropical biomes (0.34). The model evaluation further revealed that in most cases the model performance outperforms the human-made labels. Our analysis revealed resolution-dependent performance variations across biomes, suggesting a relationship between optimal mapping resolution and biome-specific crown characteristics. We make both our model and a machine-learning-ready dataset publicly available on deadtrees.earth to support future research in tree mortality mapping.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".