Home · Sensing & AI
Pillar 2 of 2Imagery becomes useful when it stops being a map and becomes a visit list.
Sentinel-2 vegetation indices are computed per parcel, compared against that block's own history, and any unusual decline is raised as an anomaly candidate — whose first action is always the same: an inspection at the spot.
Four sources, one land geometry.
Every signal is normalised to the same parcel geometry and timeline, so last week's satellite observation can be placed directly alongside this morning's inspection result.
Sentinel-2 Satellite Imagery
Four operational indices are computed per parcel: crop greenness, canopy vitality, chlorophyll condition, and vegetation moisture. Every observation carries its valid-pixel percentage and cloud level — without both, a number cannot be judged fit to use.
Weather & Agroclimate
Forecasts are pulled once an hour server-side, not once per browser, then stored as a single versioned artefact complete with its creation time, data provenance, and offline copy.
Soil Sensors & Drones
The registry for sensors and drone missions — orthophoto, multispectral, thermal, tree canopy, change detection — is in place. An estate with no sensors at all still runs fully; the screen stays empty until the first device is registered rather than being filled with sample numbers.
Field Scouting & Inspection
Imagery findings are proven or dismissed on the ground. It is the inspection result that decides whether a candidate ends as an action, or as a false positive that is still recorded together with its reason.
Eight metrics, each with its own status.
The first four are used for daily decisions. The rest are shown exactly as they are, with their status label — an uncalibrated metric is far more dangerous hidden than it is labelled.
| Metric | What it measures — and what it does not | Source & resolution |
|---|---|---|
| Crop GreennessOperations | General greenness and vegetation activity. Not a direct measure of crop health. | Sentinel-2 L2A · 10 m |
| Canopy VitalityOperations | Vegetation condition and density, particularly under a relatively dense canopy. Not a definitive measure of productivity. | Sentinel-2 L2A · 10 m |
| Chlorophyll ConditionOperations | Changes in chlorophyll condition and earlier indication of crop stress. Not a nutrient deficiency diagnosis without an inspection. | Sentinel-2 L2A · 20 m |
| Vegetation MoistureOperations | Water content in the crop canopy. Not soil moisture. | Sentinel-2 L2A · 20 m |
| Soil Moisture EstimateExperimental | An estimate of how wet or dry the soil is, from satellite and supporting models. Not an actual soil sensor reading. | Sentinel-1, weather & models · 20 m |
| Biomass EstimateExperimental | A relative estimate of biomass density. It does not show tonnes per hectare while the model remains uncalibrated. | Derived models & calibration data · 20 m |
| Soil Organic Carbon EstimateExperimental | An estimate of soil organic carbon content from models and supporting samples. Not a laboratory result. | Models & laboratory samples · 20 m |
| Land Surface TemperatureNot yet connected | Land and canopy surface temperature at the moment of satellite observation. Not air temperature. | Landsat thermal · awaiting connection |
* Status reflects the product as it stands today, not a roadmap. An experimental metric changes status only after field calibration at the estate in question.
An anomaly is a candidate, not a diagnosis.
Each parcel is compared against its own history, not against neighbouring blocks. The rules are open and configurable: a decline of at least 8% against the median of previous valid observations, a minimum of three comparison observations, and an observation with valid pixels below 60% never becomes an alert outright.
1 · Detected
The decline crosses the threshold. Severity follows its magnitude, not a hunch.
2 · Quality review
Heavy cloud and low valid-pixel counts are filtered out here, before anyone's time is spent.
3 · Needs verification
The candidate becomes an inspection task at the spot. This is always the first action.
4 · Confirmed
Proven, it becomes an action; unproven, it is recorded as a false positive together with its reason.
One rule we do not negotiate: an index decline never turns automatically into a fertiliser instruction. It becomes a visit first.
Three rules that make this layer worth trusting.
Not because the models are clever, but because every number can be asked where it came from and every doubt has somewhere to surface.
Recommendations that can be explained
A deterministic rules engine, not a black box. Every recommendation carries the data that triggered it, and must be verified by an agronomist before becoming an action.
Credentials never in the browser
Imagery requests go through a server-side proxy that holds the provider credentials. Provider status is shown as it is — including when credentials have not been entered or the service is unreachable.
The unknown does not become zero
A failed observation is marked unavailable rather than filled with a substitute number. Data availability has a screen of its own so anyone can check it.
See the index series over your own blocks.
Send us your estate boundaries; we will show the series for all four operational indices together with their valid-pixel coverage — before you decide anything.