EarthDailySentinel2Biophysical

EarthDaily Sentinel-2 + Biophysical Products

This example shows how to materialize EarthDaily Sentinel-2 imagery alongside derived biophysical products such as LAI, FAPAR, and FCOVER.

The Sentinel-2 imagery and biophysical variables are separate EarthDaily STAC collections. Rather than adding a special combined data source, prepare each layer independently, then align the biophysical layer item groups to the Sentinel-2 item groups by item name before materialization.

Use earthdaily.Sentinel2L2A for the Sentinel-2 layer in this workflow. Its sentinel-2-l2a item IDs match the naming scheme used by the EarthDaily biophysical products.

Layers

Here is a representative dataset configuration snippet. The target biophysical layers use a larger max_matches during prepare so the alignment script has candidate items to choose from.

{
  "layers": {
    "sentinel2": {
      "type": "raster",
      "band_sets": [
        {
          "bands": ["B02", "B03", "B04", "B08"],
          "dtype": "uint16"
        }
      ],
      "data_source": {
        "class_path": "rslearn.data_sources.earthdaily.Sentinel2L2A",
        "init_args": {
          "assets": ["B02", "B03", "B04", "B08"],
          "harmonize": true
        },
        "query_config": {
          "space_mode": "INTERSECTS",
          "max_matches": 4
        },
        "ingest": false
      }
    },
    "lai": {
      "type": "raster",
      "band_sets": [
        {
          "bands": ["lai"],
          "dtype": "float32",
          "nodata_value": 0
        }
      ],
      "data_source": {
        "class_path": "rslearn.data_sources.earthdaily.Biophysical",
        "init_args": {
          "variable": "lai"
        },
        "query_config": {
          "space_mode": "INTERSECTS",
          "max_matches": 20
        },
        "ingest": false
      }
    }
  }
}

Add similar layers for FAPAR and FCOVER by changing the layer name, band name, and variable to fapar or fcover.

Workflow

First prepare the dataset normally:

rslearn dataset prepare --root ./dataset

EarthDaily biophysical item IDs are derived from the Sentinel-2 scene IDs by appending the uppercased variable name. For example, LAI products may be named:

S2C_31TEJ_20250424_0_L2A -> S2C_31TEJ_20250424_0_L2A_LAI

Align the biophysical layer groups to the Sentinel-2 layer groups with that name relationship:

python docs/examples/align_item_groups_by_name.py \
  --root ./dataset \
  --reference-layer sentinel2 \
  --target-layers lai fapar fcover \
  --target-name-template '{reference_item_name}_{target_layer_upper}'

The script rewrites the prepared item groups for lai, fapar, and fcover so they have the same group order as sentinel2. For each reference group, the target item name is formatted from the first Sentinel-2 item name and selected from the target layer's already-prepared candidate groups.

Finally materialize:

rslearn dataset materialize --root ./dataset

Notes

• Run the alignment script after prepare and before materialize.
• Use --dry-run first to check how many windows would be updated.
• Increase the target layers' query_config.max_matches if the expected biophysical item is not available in the already-prepared candidate groups.
• Rewritten target layer data is marked unmaterialized so materialize will read the aligned groups.
• If you need to change the alignment settings, rerun prepare for the target layers to restore their full candidate item groups before rerunning the script.
• The default --group-time-range-source target keeps the matched target layer's prepared request time ranges. This is usually the least surprising option for materialization.
• If target layer names differ from the product suffixes, run the script separately per target layer with a literal template such as --target-name-template '{reference_item_name}_LAI'.