Monitoring road development in Congo Basin forests with multi-sensor satellite imagery and deep learning
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Road development has affected many remote tropical forests around the world and has accelerated human-induced deforestation, forest degradation and biodiversity loss. The development of roads in tropical forests is largely driven by industrial selective logging, which can provide a sustainable source of revenue for developing countries while avoiding more detrimental forms of forest degradation or deforestation. Understanding the dynamics and impacts of road development is challenging, because road inventories in remote tropical forests have been largely incomplete or outdated. In this study, we present novel remote sensing-based methods for automated monitoring of road development and apply them across the Congo Basin forest region, an area characterized by increasing road development rates driven by logging activities. We trained a deep learning model with Sentinel-1 and -2 satellite imagery to map road development on a monthly basis at 10 m spatial scale, leveraging the complementary value of radar and optical imagery. Applying the model across the Congo Basin forest, we present a vectorized map of road development from January 2019 until December 2022, demonstrating an F1-score of 0.909, a false detection rate of 4.2% and a missed detection rate of 14.9%. In total, we mapped 35,944 km of road development in the Congo Basin forest during the four years, with at least 78% apparently related to logging activities, mainly located in the western part of the region. We estimate that 30% of the detected road openings were previously abandoned logging roads that were reopened. In addition, 23% of detected road development was located in areas considered to be intact forest landscapes. The road monitoring methods demonstrated in this study can facilitate several crucial forest management and conservation objectives in the tropics, such as assessing ecological and climate impacts related to selective logging, monitoring illegal or unsustainable activities, and providing a basis for improved understanding and evaluation of human impacts on forests at large scale. More information, including a full overview of the Congo Basin forest road map, can be found at: https://wur.eu/forest-roads . • Sentinel-1 and -2 data and deep learning were used to map monthly road development. • A false detection rate of 4.1% and a missed detection rate of 14.9% were achieved. • 35,944 km of road development was mapped in the Congo Basin forest for 2019-2022. • At least 78% of road development was related to industrial selective logging.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it