michaelfburke/iap-embed-za
Domaine:
environment and energygeospatialType de record:
project
Créateur:
mic
Hôte:
A research project that applies geospatial foundation model embeddings to map and predict invasive alien plant (IAP) distributions across South Africa.
# IAP-Embed-ZA
**IAP-Embed-ZA** is a research project that applies geospatial foundation model embeddings to map and predict invasive alien plant (IAP) distributions across South Africa. It appears to be the first South African IAP mapping study built around Google AlphaEarth Satellite Embeddings.
*Per-pixel `P(presence | Acacia mearnsii)` over South Africa: AlphaEarth + Sentinel + WorldClim features extracted on a 0.1° grid via Earth Engine, then scored with a spatially-CV-evaluated RandomForest. Grey cells lack 2024 Sentinel-2 coverage after cloud-masking.*
## Research Goal
Use **Google AlphaEarth Satellite Embeddings V1** (64-dimensional, 10 m, annual 2017–2024) as primary input features for machine learning models that predict IAP presence and distribution at biome and/or species level.
## Dataset Summary
Only datasets currently flowing through the pipeline are listed. `data/processed/occurrences_raw.csv` is sourced via GBIF, and at present all records resolve to iNaturalist research-grade observations.
| Dataset | Source URL | Access Method | Resolution | Coverage |
|---|---|---|---|---|
| GBIF (iNaturalist research-grade records, filtered to SA) | Invasive and Introduced Pla… | `pygbif` / GBIF API | Point records | Global, filtered to South Africa |
| AlphaEarth Satellite Embeddings V1 | developers.google.com | Google Earth Engine (`ee`) | 10 m, 64 bands | Global annual composites (2017–2024) |
| Sentinel-2 SR Harmonized | developers.google.com | GEE (`ee`) | 10 m | Global |
| Sentinel-1 GRD | developers.google.com | GEE (`ee`) | ~10 m | Global |
| WorldClim v1 bioclim variables | developers.google.com | GEE (`ee`) | ~1 km | Global |
| SRTM elevation | developers.google.com …