Case Study · Drone-based Weed Detection with John Deere

Goal

From drone imagery, produce a weed-detection application map that a John Deere sprayer can execute in-field.

Data flow

1. Drone flight delivers RGB and multispectral tiles.
2. Imagery lands as a STAC collection in the Spatio-Temporal API — Zarr DataCube with RGB and band layers.
3. The detection service writes weed polygons as features in a feature collection (POST /api/v2/collections/{id}/items).
4. A second service produces an application map raster layer from the polygons, formatted for the John Deere Operations Center.
5. The map is pushed to the machine via the John Deere connector.

APIs used

• Spatio-Temporal API — imagery in (Zarr layers), detections + application map out (features + raster layer)
• Farm API — field boundaries
• Activity API — the planned spraying activity, with the application map as an attachment on the record
• Integrations · John Deere — pushing the application map
• Service Registry — the chain of two services