Your own Earth Engine layers

An administrator can add map layers that are computed in Earth Engine. They draw next to the built-in layers.

1. Build the layer in the Earth Engine Code Editor.
2. Export it with Export.image.toAsset() into your own Earth Engine project.
3. Give the service account of the app read access to the asset.
4. Register it on Administration → Map layers. Enter the asset ID, the band to draw, a palette, and who can see it.

The layer then appears in the layer window of the map, in the group that you name. It has its own key, and it renders through the same path as each built-in layer.

How the asset and the token fit together

The asset is permanent. The tile URL is not. The app is built around that difference.

Export.image.toAsset() writes your processed grid into a managed record in your own Earth Engine project. It stays there and it does not expire.

But Earth Engine does not serve raw tiles publicly. A call to getMapId() on the asset creates a URL. That URL carries a token, and the token expires within hours.

A function on the server is therefore the bridge:

1. The browser asks the function for a tile URL.
2. The function signs in with the service account.
3. It loads the asset that is already computed. It does not repeat the remap() and visualize() steps that the export did.
4. It calls getMapId() and returns the new URL.
5. Leaflet draws the tiles.

Nothing is cut into thousands of stored z/x/y.png files. There is one asset in Earth Engine, and one small function that creates tokens on request.

The app asks for a new URL approximately every hour. An expired template does not give an error. It serves blank tiles, and blank ground reads as ground with nothing on it, and not as a layer that failed.

Note This is also why the app files the tiles of an Earth Engine layer under the layer and not under the URL. Under the URL, each saved tile would be unreachable when the token changed.

A tree cover layer, from start to end

Classify or threshold the data in the Code Editor. Export it once. Register the asset. The export is what makes the layer cheap to draw, because the work happens one time and not for each tile.

// Ten cover classes over an area of interest, exported once as an asset.
var aoi = ee.Geometry.Rectangle([-105.4, 39.5, -104.6, 40.2]);

var s2 = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')
  .filterBounds(aoi)
  .filterDate('2025-06-01', '2025-09-15')
  .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))
  .median();

var ndvi = s2.normalizedDifference(['B8', 'B4']).rename('ndvi');

// Ten classes, 1..10. Zero stays free to mean nodata, which is what the
// layer's "Hide values below" setting masks out.
var classes = ndvi.multiply(10).ceil().clamp(1, 10).toByte().rename('classification');

Export.image.toAsset({
  image: classes.clip(aoi),
  description: 'cover_classes_2025',
  assetId: 'projects/your-project/assets/cover-classes-2025',
  region: aoi,
  scale: 20,
  maxPixels: 1e10
});

When the export finishes, give the service account of the app read access to the asset. Then register the asset on Administration → Map layers.

Select the Classified cover preset. It fills in the band, the range of 1 to 10, ten different colors and the zero mask. These are the settings that are easy to get wrong. When they are wrong, the layer renders badly instead of giving an error.

The presets

The presets describe the shapes that these rasters usually take: percent cover, a classified grid, canopy height in metres, a signed index, a probability.

No preset carries an asset ID. A preset describes how to paint a raster. Which asset you have is your own business.

Who can see a layer

Each layer carries its own access level: everyone, members or administrators. A finished layer can therefore be public while a draft stays internal.

The layer window lists a layer that the viewer cannot render, and marks it. It does not hide it. When the viewer switches it on, the app explains why it cannot draw it.

Note The app stores an asset ID. It never stores a script. An asset ID names something that is already computed under your own project. A script would be arbitrary compute on the Earth Engine budget of FRMS.

Caution Name the band if the asset has more than one. Earth Engine refuses a palette on an image with several bands, and the layer will not render. Also set Hide values below when the no-data value of the asset is zero. Otherwise the whole world paints as the bottom of the gradient.

Where to learn Earth Engine

Read Learning for a short path into the Code Editor, the data catalogue and the common mistakes.