Drone software for agronomists

Two layers, often confused

“Agronomist drone software” covers two quite different jobs, and a lot of frustration comes from buying one when you needed the other.

The capture layer gets the imagery: flight planning, route and altitude, overlap settings, the sensor payload, and the pipeline that stitches the frames into a georeferenced orthomosaic. It answers the question did we photograph the whole field properly?

The analysis layer starts once the imagery exists. It reads the bands, computes vegetation indices, separates canopy from soil, compares this pass against the last one, and turns all of that into something with a location and a recommended action attached. It answers the question what changed, where, and what should we do about it?

Terravect does both, so nobody has to work out how to survey their own land: we arrange the flying, and the imagery comes straight to us. Capture is the well-understood half, though — a sensor, a flight plan and a stitching pipeline. What separates one survey from another is the analysis, and that part is ours: our own processing pipeline and classification models, and a voxel-based 3D path algorithm that reads through the depth of a canopy rather than across the top of it.

What the sensor sees that the eye doesn't

A standard camera records three broad bands — red, green and blue. A multispectral sensor records several narrow, deliberately chosen bands, typically including green, red, red edge (around 700–730 nm) and near-infrared (NIR, beyond roughly 760 nm).

Those last two matter because of how leaves behave. Healthy chlorophyll absorbs red light strongly, while the internal structure of a healthy leaf reflects near-infrared strongly. The result is a steep jump in reflectance between red and NIR — the red edge. When a plant comes under stress, that jump flattens and shifts, and it does so before the canopy turns a colour you would notice from the tractor seat.

This is the whole basis of the method: you are measuring a physiological signal, not a colour. It is why a survey can flag a problem in a block that still looks perfectly green from the headland.

From an index map to a decision

Most tools will happily hand you an NDVI map. A colour-ramped field is a starting point, not an answer, and three things stand between it and a decision:

• Is the variation real or an artefact? Differences in sun angle, cloud, sensor calibration or flight altitude all move index values. A change between passes has to be separated from a change in conditions.
• Is it the crop or the ground? Where the canopy is incomplete, a pixel mixes plant with bare soil, cover crop or shadow. The index reads the mixture, not the plant.
• What kind of stress is it? A low-vigour patch is a question, not a diagnosis. Water stress, disease and nutrient deficiency can all depress the same index, and they call for different responses.

That interpretive step is the work. It is also the step that decides whether a survey changes what anyone does on Monday morning.

Why location matters more than you'd think

A finding without a precise, navigable location is a finding nobody acts on. “There is stress in the north-west quarter” sends somebody walking. A finding pinned to a specific row, plot or zone on a model of the ground sends them straight to it.

This is where a 3D digital twin earns its place over a flat map. Terrain is part of the explanation: slope, aspect and drainage account for a great many of the patterns that show up in a vigour map, and they are invisible in a flat orthomosaic.

Repeat passes are the point

A single survey tells you what the field looks like today. A sequence of surveys tells you what is happening — which is nearly always the more useful question. Change detection is where drone data stops being a picture and starts being a monitoring system.

For that to work, passes have to be comparable: consistent sensor, consistent calibration, and a consistent frame of reference to lay each pass over. Otherwise you are comparing flights, not fields.