Skip to content

Aws google satellite embedding v1

rslearn.data_sources.aws_google_satellite_embedding_v1.GoogleSatelliteEmbeddingV1

Google Satellite Embedding v1 (annual satellite embeddings from AlphaEarth) at 10 m/pixel resolution on AWS Open Data. The data is stored as Cloud-Optimized GeoTIFFs organized by year and UTM zone.

Available years: 2018-2024.

See https://registry.opendata.aws/aef-source/ for details. No credentials are needed (public bucket).

Configuration

{
  "class_path": "rslearn.data_sources.aws_google_satellite_embedding_v1.GoogleSatelliteEmbeddingV1",
  "init_args": {
    // Required cache directory for the index file. Unless prefixed by a protocol
    // (like "file://..."), it is joined with the dataset path.
    "metadata_cache_dir": "cache/gse",
    // Whether to apply de-quantization (default true). The raw data is int8; with
    // de-quantization enabled, values are mapped to float32 in [-1, 1] using:
    //   ((value / 127.5) ** 2) * sign(value)
    // Raw nodata is -128; after de-quantization it becomes -1.0.
    // Set false to keep raw int8 values.
    "apply_dequantization": true
  }
}

This data source supports direct materialization ("ingest": false).

Available Bands

The band names for the 64 embedding channels are "A00" through "A63".

  • When apply_dequantization is true (default): float32 values in [-1, 1] (nodata = -1).
  • When apply_dequantization is false: raw int8 values (nodata = -128).