Field Anomaly Detection

Flags where vegetation diverges, unexpectedly, from its local field trend — at the 4×4 m patch level, weeks before a problem is visible to the eye.

Fields are monitored episodically; by the time a problem is visible it has cost weeks of loss. ETD's Field Anomaly Detection flags locations where vegetation behaviour diverges significantly from the surrounding local trend between two cloud-free observations. Detection runs at 2 m using NDVI from enhanced multispectral, identifying anomalies at the 4×4 m patch level — one twenty-fifth of a native Sentinel-2 pixel. Thresholds are derived automatically per field, season and image pair. A disease focus of three or four plants, a first-week weed patch, or a single failed irrigation emitter becomes detectable at the moment it begins. The same mechanism surfaces nutrient deficiency, waterlogging, drought stress, and even pipeline leaks along corridors.