Applying a deep learning pipeline to classify land cover from low-quality historical RGB imagery
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
Land use and land cover (LULC) classification is becoming faster and more accurate thanks to new deep learning algorithms. Moreover, new high spectral- and spatial-resolution datasets offer opportunities to classify land cover with greater accuracy and class specificity. However, deploying deep learning algorithms to characterize present-day, modern land cover based on state-of-the-art data is insufficient for understanding trends in land cover change and identifying changes in and drivers of ecological and social variables of interest. These identifications require characterizing past land cover, for which imagery is often lower-quality. We applied a deep learning pipeline to classify land cover from historical, low-quality RGB aerial imagery, using a case study of Vancouver, Canada. We deployed an atrous convolutional neural network from DeepLabv3+ (which has previously shown to outperform other networks) and trained it on modern Maxar satellite imagery using a modern land cover classification. We fine-tuned the resultant model using a small dataset of manually annotated and augmented historical imagery. This final model accurately predicted historical land cover classification at rates similar to other studies that used high-quality imagery. These predictions indicate that Vancouver has lost vegetative cover from 1995-2021, including a decrease in conifer cover, an increase in pavement cover, and an overall decrease in tree and grass cover. Our workflow may be harnessed to understand historical land cover and identify land cover change in other regions and at other times.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it