Buildings and roads detected from 0.5 meter resolution Maxar satellite imagery, Alaskan communities, 2018-2023
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
We created HABITAT (High-resolution Arctic Built Infrastructure and Terrain Analysis Tool), a deep learning-based, high-performance computing-enabled mapping pipeline to automatically detect buildings and roads from high-resolution Maxar satellite imagery in Arctic communities. The code is made available at https://github.com/PermafrostDiscoveryGateway/HABITAT. The pipeline is based on a ResNet50-UNet++ semantic segmentation architecture trained on a training dataset comprised of building and road footprint polygons manually digitized from Maxar satellite imagery across the circumpolar Arctic (including Alaska, Russia, and Canada). From imagery of 285 Alaskan communities acquired between 2018-2023, we detected approximately 250,000 buildings and storage tanks (comprising a 41.76 million square meter footprint) and 15 million meters of road. Building (including storage tanks) footprint polygons and road centerlines were strictly mapped within the boundaries of Alaskan communities (both incorporated places and census designated places). After the deep learning model detected building and road footprints, post-processing was performed to smooth out building footprints, extract centerlines from road footprints, and remove falsely-detected infrastructure. In particular, a buffer is created around developed land cover identified by the 2016 Alaska National Land Cover Database map, and model predictions that fall outside of the buffer are assumed to be confused with non-infrastructure land cover.
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 | 0.001 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.005 |
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