Reproduction pack for: What can satellite embedding fields tell us about shallow coral-reef habitats? A decodability survey of AlphaEarth over the Great Barrier Reef (v1)

Public reproduction pack for: Cropp, A. (2026). What can satellite embedding fields tell us about shallow coral-reef habitats? A decodability survey of AlphaEarth over the Great Barrier Reef. EarthArXiv preprint (its DOI will be added to this record as a related identifier once assigned). Contents (one zip, alphaearth-gbr-decodability-scripts-v1.zip): scripts/ — peer-review scripts p00–p05 plus the shared library rppub.py. p00_download_data.py acquires every required dataset from its public source (cache-first, SHA-256 verifiable against the release deposit); p01–p05 reproduce the cross-reef leave-one-reef-out transfer, the pooled 67-reef analysis, the cross-era transfer, the operational-threshold deployment, and the independent Squidle+/RLS spot-check. decoders/ — released Heron-trained ridge decoder coefficients, calibrated operational thresholds and the coral-loss axis (derived model parameters; no ground-truth records beyond the aggregate intercept disclosure stated in the folder README). data-deposit/ — the manifest of the exact AlphaEarth embedding windows used (8,670 files, SHA-256 per file, per-window source URL); the window files themselves are deposited separately as the embedding-window archive (DOI listed as a related identifier). DATA-AVAILABILITY.md — every dataset, its licence, and the three reproducibility tiers. datasets/ — starts empty; filled by p00_download_data.py. The pack reproduces the fully public tier of the study end-to-end (train and test on public data) and executes the public-test-data results from the released decoder weights. The Heron leave-one-site-out and change-regime tables require the restricted ReefCloud ground truth, available on request from its custodian. Requirements: Python 3, packages listed in scripts/requirements.txt. The scripts, documentation and released coefficient files are MIT-licensed. No datasets are redistributed in this pack; each dataset downloaded by the scripts remains under its own licence, listed in DATA-AVAILABILITY.md. This research was supported by the Commonwealth through an Australian Government Research Training Program Scholarship, undertaken through the PhD@Work arrangement at James Cook University hosted by OSIBOT PTY LTD.