Estimating Avocado Yield Per Tree Before Harvest

Why avocado estimates go wrong

Avocado is a crop where getting the estimate wrong is expensive at scale, not just inconvenient. When the California Avocado Commission studied the problem in the early 1990s, it documented what a bad estimate does: the 1985 crop volume was grossly overestimated by 40 percent. Growers harvested early expecting a large crop, the market was flooded, and by the last quarter fruit was in short supply. The uneven shipment spread produced what the Commission described as tremendous financial loss to many growers.

That history explains why California's industry still runs a formal crop-estimating process: estimate forms mailed to every grower in September for the pre-season estimate, update rounds in January and April, and field experts who walk groves to verify estimates. Growers are not being surveyed for fun. Their per-grove numbers feed packing house schedules, harvest strategy and marketing plans, and the Commission has asked for help with exactly this task on its own crop estimate page.

The hard part is specific to avocado: the fruit hang singly, deep inside a dense, dark canopy, often the same color as the leaves. Green-skin cultivars are hard to see at all. An eyeball estimate from the row side systematically misses the interior of the tree, and the error is not random. It is a consistent undercount that grows with canopy size.

The three numbers every estimate needs

South African avocado work going back decades (the Bavendorf model, adapted to avocado by Köhne at Westfalia Estates in 1985) reduces yield estimation to three parameters:

• Production capacity of the tree (how much crop a tree of that size and health can carry)
• Fruit set density (how many fruit are actually set this season)
• Fruit weight at harvest (which depends on cultivar and the size class the fruit reaches)

That decomposition is useful because it tells you what to measure and when. Tree capacity changes slowly; you can hold it roughly constant from your own history. Fruit set density is the fast-moving variable, and it is what a photo survey measures directly. Fruit weight is a harvest-time sample away from being solid data. Most grower estimates fail by collapsing all three into one eyeballed number per block.

A per-tree photo method for avocado

1. 1Split the grove into blocks by cultivar and age. A 'Hass' tree and a 3-year-old rework block do not share parameters. Estimates are per block; sums come after.
2. 2Photograph every tree, from at least two sides. Geotagged photos let software cluster shots to individual trees automatically. For avocado, take one whole-canopy shot per side plus close-ups where fruit are visible in clusters; move your camera position slightly so you get different angles into the canopy interior.
3. 3Count fruit per photo and roll up per tree with a maximum rule. Where photos of the same tree overlap, take the maximum count across them rather than the sum, or you double-count fruit seen in two frames.
4. 4Estimate the visible fraction honestly. This is the make-or-break number for avocado. Because so much fruit hides in the canopy interior, the visible fraction on a photo count is lower than for most tree crops. Calibrate it once per season: photograph your calibration trees, add up the photo count, then count everything at harvest on those same trees. The ratio is your correction factor, and it is cultivar- and canopy-specific.
5. 5Sort the count into size classes. Fruit weight depends on size, and size varies within a tree. The classic industry calculator counts fruit per tree by size category (the familiar lug sizes, 28 down to 84 and undersize) and multiplies each category's count by its average fruit weight. From photos you can approximate this by judging whether the fruit run large, medium or small, then adjusting at harvest with your actual packout data.
6. 6Apply your drop allowance. Avocados shed fruit from set onward, and the shed continues until harvest. Your January count is not your April crop. Record a drop allowance per block per window, learned from your own history, and state it explicitly in the estimate.
7. 7Store the estimate on the tree record, not a legal pad. Next year's estimate improves by comparing against the previous year's per-tree numbers and actuals. That comparison is where estimation skill actually comes from.

A worked example: a half-hectare 'Hass' block

Suppose you farm 40 'Hass' trees on half a hectare and you are filling in a pre-season estimate form in September. Your photo survey (two passes per tree, counted with the maximum rule) gives these results on a sample of the block:

• Trees A1-A40, mean maximum photo count per tree: 210 fruit
• Visible-fraction calibration from last season (photo count 156, harvest count 234 on two calibration trees): visible fraction about 0.67, so correction factor 1.5
• Corrected fruit per tree: about 315
• Your September sample of 20 picked fruit averaged 230 g, and your packout history says about 8 percent will be dropped, culled or undersize

Estimated marketable yield per tree: 315 × 0.230 kg = 72 kg, minus 8 percent, so about 67 kg per tree. Across 40 trees, the block estimate is roughly 2.7 tonnes. You would give a buyer or a packing house a range (say 2.4-3.0 tonnes) rather than a point number, because the visible fraction and the drop allowance each carry uncertainty. The range is the honest output of the method; the point estimate is just its middle.

In January you repeat the survey. Fruit are bigger, the count per photo drops as fruit shed, and your sample fruit now average 280 g. The block estimate tightens to, say, 2.6 tonnes. By the April update you are planning bins and picking days with the number, and your error at harvest tells you next season's correction factor.

Mistakes specific to avocado

• Reusing a mango-style visible fraction. Avocado hides more fruit than most crops. A correction factor calibrated on another crop undercounts badly.
• Ignoring tree age and size classes. Capacity per tree is not uniform; a block of 20-year trees and a young block do not share a per-tree average.
• Counting at the wrong window. The pre-set flush exaggerates; count after the main fruit drop, then re-check before harvest.
• One eyeball pass per year. The California process uses three estimate windows (September, January, April) precisely because one number in September will be wrong by April. Your own passes should be at least two.
• Not recording actuals per tree. Without harvest-time per-tree counts or weights, your correction factor never improves. The estimate stays a guess forever.

Season timeline for a California-style estimate: September, pre-season photo survey and estimate form. January, update pass (post-drop fruit count, revised size class). April, final pass for harvest planning (bins, pickers, packout booking). At harvest, record actuals and recompute your visible fraction. Repeat annually; the factor converges.

How CanopyTwin fits in

CanopyTwin handles the bookkeeping that makes this method repeatable. You photograph each tree with your phone; GPS clusters the geotagged photos into individually tracked trees on a satellite map, so per-tree counts roll up without double-counting overlapping shots. The AI counts fruit and flowers in survey photos (Orchard tier), every tree keeps a photo history, and CSV exports move the numbers into your harvest plan or your commission's estimate form. Every plan starts with a 30-day free trial, no credit card up front.