Skip to content

Reference

Using the engine without QGIS

embed_cd/ imports no QGIS. It runs and tests standalone on numpy, scipy and GDAL, which is what lets the whole thing be developed and verified outside the application.

from embed_cd import job

grid, tiles, hist, partial = job.run(
    bbox=(-125.4, 49.6, -125.2, 49.8),   # lon/lat
    year_a=2019, year_b=2024,
    out_dir="run", dst_crs="EPSG:32610", res_m=10.0,
    cell_m=160.0,                        # pool embeddings for the classifier
)

dst_crs must be a CRS in which a metre is a metre — see Detail and resolution.

Other entry points: embed_cd.objects.polygonize and attach_vectors cut objects and give them embeddings; embed_cd.head.fit_from_classes is the classifier; embed_cd.store handles the GeoPackage and label files.

Citations

  • AlphaEarth Foundations — Brown, C.F. et al. (2025). AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data. arXiv:2507.22291
  • The dataset — Satellite Embedding V1, Earth Engine Data Catalog
  • Comparing embeddings for change — Burns, M. (2026). Rethinking Change Detection and Attribution: How You Compare Satellite Embeddings Matters. Google Earth blog. Link
  • One-vs-all — Rifkin, R. & Klautau, A. (2004). In Defense of One-Vs-All Classification. JMLR 5:101–141.
  • Open-set recognition — Scheirer, W.J., Rocha, A., Sapkota, A. & Boult, T.E. (2013). Toward Open Set Recognition. IEEE Trans. Pattern Analysis and Machine Intelligence 35(7):1757-1772. doi:10.1109/TPAMI.2012.256
  • Reject option — Chow, C.K. (1970). On optimum recognition error and reject tradeoff. IEEE Trans. Inf. Theory 16(1):41–46. · Fumera, G., Roli, F. & Giacinto, G. (2000). Reject Option with Multiple Thresholds. Pattern Recognition 33(12):2099–2101.
  • Change magnitude vs change type — Cohen, W.B. & Fiorella, M. (1998). Comparison of methods for detecting conifer forest change with Thematic Mapper imagery. In Lunetta, R.S. & Elvidge, C.D. (eds), Remote Sensing Change Detection, pp. 89-102. Ann Arbor Press. · Kennedy, R.E., Yang, Z. & Cohen, W.B. (2010). Detecting trends in forest disturbance and recovery using yearly Landsat time series: LandTrendr. Remote Sensing of Environment 114(12). doi:10.1016/j.rse.2010.07.008
  • Otsu — Otsu, N. (1979). A threshold selection method from gray-level histograms. IEEE Trans. Systems, Man, and Cybernetics 9(1):62–66.

Licences and attribution

AlphaEarth Foundations Satellite Embedding V1 — Google and Google DeepMind, CC BY 4.0. Global, every year 2017–2025, read from public cloud-optimized GeoTIFFs on source.coop.

Sentinel-2 cloudless reference imagery — EOX IT Services, containing modified Copernicus Sentinel data. CC BY-NC-SA 4.0, non-commercial only for 2018 onward (2016 is CC BY 4.0). If your deliverable is commercial, do not ship these tiles in it. Commercial licences: cloudless.eox.at.

EMBED-CD itself — GPL-2.0-or-later.