Burned area mapping across the Arctic-boreal zone (1985-2020) with Landsat and Sentinel-2 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
Wildfires in the Arctic-boreal zone have increased in frequency over recent decades, carrying substantial ecological, social, and economic consequences. Remote sensing is crucial for mapping burned areas, monitoring wildfire dynamics, and evaluating their impacts. However, existing high-latitude burned area products suffer from significant discrepancies, particularly in Siberia, and their coarse spatial resolutions limit accuracy and utility. To address these gaps, we developed a convolutional neural network model to map burned areas at a 30-meter resolution across the Arctic-boreal zone using Landsat and Sentinel-2 imagery. Our model achieved promising results, with an Intersection Over Union (IOU) of 0.77 and an F1 score of 0.85 on unseen test data, performing better in North America (IOU=0.84) than Eurasia (IOU=0.72) due to differences in fire regimes and data quality. Predictions for six representative years showed our model’s burned area closely matched the median values of Landsat, MODIS, and VIIRS-based products, although alignment varied annually and spatially. Visual assessments indicated our approach was generally more accurate, notably in detecting unburned vegetation islands within fire perimeters missed by other products. This research has numerous potential applications, such as analyzing feedback between vegetation and burn patterns, characterizing spatial dynamics of unburned islands, and improving carbon emission estimates through detailed burn severity assessments. Here we have provided the primary series of scripts used to achieve the above results. In these scripts we use historical vector fire polygons to download imagery from Landsat 5, 7, 8, 9 and Sentinel-2 to train a deep learning model called a UNet++ in the Arctic-boreal zone. Imagery is downloaded from Google Earth Engine, while all other processing is done locally. The series of 6 scripts describes main steps from downloading training data, pre-processing it, training the model, and applying the model across the Arctic Boreal Zone. All scripting is done in python through .py scripts and Jupyter notebooks (.ipynb). Our study area includes Alaska, Canada and Eurasia, and we trained our model on all historical fire polygons from 1985-2020.
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) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| 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 it