Evaluating satellite data and deep learning for identifying direct deforestation drivers in Cameroon
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
Deforestation rates have been increasing in the Congo Basin in recent years, especially in Cameroon. To support actions to slow deforestation, Earth Observation (EO) has been used extensively to detect forest loss, but approaches to automatically identify specific drivers of deforestation in a level of detail that allows for intervention prioritisation (e.g. focusing on specific areas and actions, designing measures to address specific drivers) have been rare. In this paper, using a new country-specific dataset created for this task, we test whether deep learning with optical satellite data can reliably identify direct drivers of deforestation in Cameroon. We compare the effectiveness of two types of freely available optical satellite imagery of differing spatial resolutions: Landsat-8 (30 m) and NICFI PlanetScope (4.77 m). Since it can be challenging to know which collections are best suited for specific applications, we tested different ones to find the optimal approach. Our detailed classification strategy includes fifteen direct deforestation drivers for forest loss events taking place between 2015 and 2020. We obtain a macro-average F1 score of 0.77 with Landsat-8 data, and a macro-average F1 score of 0.65 with NICFI PlanetScope. Despite a coarser spatial resolution, Landsat-8 performs better than NICFI PlanetScope overall, including for small-scale drivers, although results vary by class. Using only a single-image approach, we achieve F1 scores above 0.65 for all classes except ‘Oil palm plantation’, ‘Hunting’ and ‘Fruit plantation’ with Landsat-8. Our results demonstrate the potential of this approach to monitor and analyse land-use changes leading to deforestation with more refined classes than before. Further, our study demonstrates the potential of leveraging existing available datasets and straightforwardly adapting a generalised framework for other regions experiencing rapid deforestation with only a relatively small amount of location-specific data.
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 | 0.001 | 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