Anyone created a interactive map on google alpha earth data?

[raybell-md]

I was excited to try https://www.linkedin.com/posts/kylebarrongeo_native-cloud-optimized-geotiff-support-in-activity-7445513650326614017-xAR0 on google alpha earth. This data is non-trivial as it has 64 bands.

Sadly you'll need a GCP project to test it

store = GCSStore(
    bucket="alphaearth_foundations",
    client_options={
        "default_headers": {"x-goog-user-project": os.environ.get("PROJECT_ID")}
    }
)
geotiff = await GeoTIFF.open(
    "satellite_embedding/v1/annual/2024/18N/xger0wzpms5jl5nhv-0000000000-0000000000.tiff",
    store=store,
)

I couldn't get any data to show up in the Map.

I tried one band as gray scale and another as psudeo RGB (https://developers.google.com/earth-engine/tutorials/community/satellite-embedding-01-introduction#visualize_the_satellite_embedding_dataset).

Note: I am running on google colab which may affect displaying of the map.

Curious if anyone has got this to work

   # 3. Render Callback (Dynamic Grayscale)
   def render_tile(tile) -> EncodedImage:
       # Get just the first band (index 0)
       band_data = tile.array.data[0, :, :].astype(np.float32)

       # Calculate min/max dynamically so we ALWAYS see contrast
       d_min = np.nanmin(band_data)
       d_max = np.nanmax(band_data)

       if d_min == d_max:
           # If the tile is completely empty/uniform, return a transparent image
           img = Image.new("RGBA", (band_data.shape[1], band_data.shape[0]), (0, 0, 0, 0))
       else:
           # Scale the data to 0-255 based on its own min/max
           gray = np.clip((band_data - d_min) / (d_max - d_min) * 255, 0, 255).astype(np.uint8)
           # Create a Grayscale image
           img = Image.fromarray(gray, mode="L")

       # Encode and return
       buf = io.BytesIO()
       img.save(buf, format="PNG")
       return EncodedImage(data=buf.getvalue(), media_type="image/png")

   # 4. Display the Map widget in Jupyter
   layer = RasterLayer.from_geotiff(geotiff, render_tile=render_tile)

   # This must be the last line in the cell to render interactively!
   Map(layer)

[kylebarron]

The problem isn't that the data has 64 bands; the problem is that this data was exported to COG incorrectly and the data is "flipped". I.e. the origin is in the bottom left instead of the (usual) top left.

developmentseed/deck.gl-raster#261
developmentseed/deck.gl-raster#321

I tested this on the JS side but it still wasn't rendering correctly.

I think it might be easier to render this from the AEF GeoZarr store, which hopefully will connect to Lonboard in a month or so