C‑DRONE
Aerial view of Paris and the Eiffel Tower at sunset

C-DRONE GUIDE · 5 AUGUST 2026

Estimating orchard yield by drone: fruit counting, calibration, prices

An orchardist's whole season hinges on a number known only at the end: the tonnage. Downstream contracts are negotiated weeks before picking, labour is booked ahead, crates, bins and cold storage are sized on an estimate — traditionally, a few hand-counted trees and a lot of experience. Drone imagery combined with neural networks changes the basis of the calculation: photograph every tree, detect and count the visible fruit, calibrate on a few ground plots, and extrapolate to the block with a measured error. Here is what the method can do today, its honest limits, and what it costs.

Published on 5 August 2026, reviewed on 29 August 2026 — regulations in force as of August 2026.

What drone counting really measures

The principle: a very high resolution RGB flight — often at an oblique angle to see the trees' sides — produces images of every row, in which a trained detection model (convolutional neural network) identifies and counts the visible fruit. Two corrections separate a gimmick from a tool: the occlusion correction (part of the fruit hides behind foliage; the visible-to-real ratio is calibrated on hand-counted plots) and the size calibration (mean diameter measured on a sample converts count to tonnage). The result: an estimate per tree, per row and per block, with its confidence interval.

The achievable accuracy is documented: Orly Enrique Apolo-Apolo's team showed in 2020 in the European Journal of Agronomy, on a citrus orchard monitored by drone and deep learning, an estimation error of about 5% per tree — against nearly 14% for an experienced technician's visual estimate (see the study on Google Scholar). The order of magnitude transfers to trellised apples and pears; small or heavily hidden fruit (plums, cherries) demands more calibration plots.

What a reliable early estimate is for

Four decisions rest directly on the number:

The same flight feeds neighbouring diagnostics: per-tree vigour, missing or declining trees (the orchard counterpart of missing vine counting), and — in spring — frost risk mapping, which protects precisely the yield being measured here.

How a campaign runs

The flight window depends on species and objective: after physiological drop for a stabilised load estimate, three to six weeks before harvest for the logistics estimate — when fruit is big enough to detect yet there is still time to act. The sequence:

Flying over an orchard is regulatorily simple (rural area, low height); hail nets, however, force flying above the structures with an adapted oblique angle — a point to flag at the quoting stage.

2026 prices

Orders of magnitude observed in France in 2026 (excl. VAT):

Set against the stakes: a few percentage points of error shaved off an apple orchard's estimate means a better-negotiated contract, better-targeted thinning and a harvest without logistical breakdowns. This service draws on our drone agriculture offer: request a quote stating species, area, training system (goblet, axis, trellis, nets) and expected harvest date.

Request a free quote

Also worth reading