Estimating yield before harvest

Counting fruit on the tree is harder than it looks, because a third of the crop is hidden. Here is what research says about doing it well.

A yield estimate multiplies three things: trees, fruit per tree and fruit weight. Each has its own error, and the largest usually sits in the fruit count.

Much of the crop is out of sight

Only about 60 to 70 percent of the crop is visible from outside a citrus tree.¹ Some fruit sit inside the canopy where no camera position can see them, so automatic counting never finds all of them.² Counts must therefore be converted to yield with a correction, and a correction trained on one season was not robust in the next, so it needs recalibrating every season.³

Systematic beats intuition

In nine mango orchards, farmer estimates based on hand counts of around 5 percent of trees were off by 26 percent on average against packhouse counts. Image-based counts reached 11 to 13 percent.⁴ In a Navelina orange orchard, a drone and deep learning model estimated total orchard yield with a 7.2 percent error, against 13.7 percent for an expert technician.¹ In Brazil, a phone video pipeline reached R² of 0.85 for yield once at least 30 percent of fruit were detected and counted.²

Drop and weight matter

Fruit continue to drop until harvest, and the rate varies between blocks and years; an average drop figure cannot simply be assumed.⁵ Fruit weight changes with size, which is why measuring the size distribution, not just counting, makes the tonnage reliable.