Rules an agronomist can read. Evidence behind every number.
Canopiq combines field computer vision, unsupervised probe anomaly detection and a transparent yield-loss model. Here is what each layer does, and where its limits are.
Five passes behind every capture.
Recover
Each pixel's colour is mapped back to its NDVI or relative canopy temperature through the export's colour ramp.
Zone
The field boundary is found and the crop is split into 16-pixel zones (~1.3 m at 8 cm GSD).
Test
Every zone is scored against the whole field with a robust z-score; agronomically trivial differences are ignored.
Name the cause
Shape decides: rows of outliers, separate spots, broad soft zones, hot zones or cold zones.
Corroborate & price
Probes confirm or question the finding, and the yield model prices it for this season.
Outliers against their own field.
Instead of a fixed NDVI threshold, which breaks across crops, growth stages and sensors, Canopiq compares every zone with the rest of the same field in the same capture. The crop, the light and the camera affect every zone equally, so a problem stands out against its own field.
- Line failure: a run of zone rows where more than 60% of the row is stressed
- Pest / disease: three or more separate deep patches (NDVI down by more than 0.22) close together
- Nutrient zone: a broad, contiguous zone with a moderate NDVI drop
- Water stress: a contiguous hot canopy zone in thermal
- Waterlogging: a contiguous cold patch in thermal
Irrigation cycles cancel out.
Every probe in a field is wetted by the same schedule and dried by the same weather. Dividing each probe's reading by the field median per day gives a ratio near 1.0 through every cycle. The 5-day mean of that ratio is scored with a median/MAD robust z-score, which a single odd reading can't distort.
A ratio well below 1 means water isn't reaching that zone. One well above 1 means it's pooling, from a leak, poor drainage or an over-long run time.
Tonnes and dollars, shown with the working.
Value at risk = hectares affected × expected yield (t/ha) × cause loss share × crop price. The risk score adds severity × confidence, a water-urgency bump and a capped log-value term. Every input is visible, so an agronomist can check the figure.
Grounded, cited, and honest about gaps.
Copilot receives the farm's findings, probe anomalies and scouting history as structured data and answers from it, citing finding IDs. Offline mode answers the common questions directly from the record. When a language model is enabled, it gets the same data and the same rule: if the record can't answer, say so.
What it doesn't do (yet).
Raw multispectral bands
The on-device engine reads rendered NDVI exports. Raw band GeoTIFFs and other indices (NDRE, CWSI) need the cloud pipeline.
Whole-field framing
Hectares assume the capture covers the whole field. Partial captures need the field boundary set in the app.
Diagnosis
Canopiq names the likely cause from the pattern. Pests, diseases and deficiencies still need a scout or lab test to confirm.