From orthomosaic to inventory: what you can and cannot measure from the air

An orthomosaic is not a big photo

An orthomosaic is what you get after correcting hundreds of photos for camera perspective and terrain relief and stitching them into a single image where every pixel has a coordinate. That is the difference from an aerial photo: a mosaic can be measured. Two trees 60 pixels apart at 3 cm per pixel are 1.8 m apart, and that distance still holds at the far end of the plot.

It also pays to read its defects, because all three get mistaken for crop problems. Seams between flight lines show up as straight lines where the tone changes: they come from blending photos, not from the soil. Stretched or smeared canopies at the edges come from parallax: a tall tree seen from the side leans over, and with short overlap the software could not straighten it. Ghosts —half a tractor, a person repeated— are objects that moved between photos.

All of this is decided in the flight, not in processing: see the guide on how to fly an orchard for tree counting.

What you can measure

With a mosaic of sufficient resolution and a plot boundary, this is what we deliver as inventory and what you can verify yourself against the image:

• Position and count per plant. Every detected canopy carries its coordinate. The count is per block and per plot, not a loose total.
• Canopy or rosette diameter. Measured on the image and calibrated against field measurements (see the next section).
• Gaps and missing plants. Once plants are grouped into rows, empty positions become explicit: how many are missing and where they are.
• Planting geometry. Row orientation, real row and in-row spacing, and linear meters of row to budget irrigation, pruning and harvest.
• Actually planted area. The flown polygon excludes roads, maneuvering areas and edges. It almost never matches the area on the deed, and for any per-hectare figure we use the flown one.
• Canopy height, when the flight produced an elevation model and ground is visible between rows: height is the difference between the canopy surface and the terrain beside it. Under closed canopy, where no ground shows, the number stops being reliable.
• Per-plant vigor, only if the flight was multispectral: NDVI, NDRE and the rest are computed per canopy, not per image patch.

What you cannot measure, however good the image looks

This list is the one that prevents disappointments, and it matters as much as the previous one:

• Fruit. What sits under the foliage is invisible from above. Yield is not counted: it is estimated with a model that combines count, canopy size, age and the plot's own past harvests.
• Whatever lives under the canopy. Covered weeds, irrigation, young plants under a large tree, trunk damage. The drone sees the roof of the crop.
• Internal maturity. Brix, moisture or sugars have no signature from the air. In agave, maturity is approximated with diameter and age, not with color.
• Pest species. You see the symptom —a patch losing vigor, a dead tree—, not the agent causing it. Someone on the ground confirms that, which is what the field reporting app is for.
• Two canopies that touch. Under closed canopy two merged trees can be counted as one. It is mitigated with more resolution, flying when shadows are short and reviewing on screen, but it is the real limit of automatic counting in a mature orchard.

Image diameter is not field diameter

This is the nuance almost nobody explains. What you measure on the mosaic is the extent of visible foliage: fuzzy edges, hanging leaves, self-shadow. What an agronomist measures with a tape is a different thing, and the two differ systematically, not randomly.

That is why the diameter we deliver is calibrated: real field measurements fit a curve that converts image diameter into field diameter, and a separate curve is fitted per planting cycle, because a three-year-old plant and an eight-year-old one do not look the same from above. Without that calibration, the image number is good for ranking plants against each other, not for stating centimeters.

The practical consequence: when you order an inventory, ask what the diameter was calibrated against and with how many measurements. If the answer is that it was not calibrated, what you have is a ranking, not a measurement.

From what you see to what you decide

The inventory is not the end; it is the input to four concrete decisions. The count per block against theoretical density tells you where to replant and how much material to buy. Diameter ranks plots by maturity and builds the harvest window. Gaps with coordinates become a work order for the crew. And the series of flights —the same plot two or three times a year— shows real growth per block and catches losses before they are visible on foot.

Two honesty notes. First: in every inventory there are plants the image leaves in doubt, and a person reviews those on screen. Second: when a missing position is filled in by interpolating the row, it is flagged as interpolated and not mixed with the detected ones; whoever receives the file has to be able to separate what was seen from what was inferred. That is how we deliver plant and tree counting.

What to ask your flight provider for

If someone else flies for you, this is what they have to hand over for the imagery to be usable for inventory:

• The raw photos of the flight, not just the finished mosaic: with them it can be reprocessed if something went wrong
• The orthomosaic as a GeoTIFF with its coordinate system declared, not resampled to a coarser resolution and not aggressively lossy-compressed
• The resolution in cm per pixel and the overlap used, in writing
• The elevation model, if one was produced
• The plot polygon that was flown, so you know what area is really covered