Tracking Flowering and Fruit Set, Tree by Tree

Why track flowering at all?

Because the three most expensive mistakes of a season are made in the eight weeks after bloom, while you are still guessing. Under-thinned crop load costs fruit size. Missing a pollination or chilling failure costs the whole block. Over-reacting to a heavy bloom with inputs you did not need costs money. None of these are visible from the gate; all of them show up early in per-tree counts.

This is not a tropical niche habit. In commercial apple production, flower cluster counts per tree are a standard input to crop load management, and the Cornell-led, multi-university PACMAN program has evaluated digital scanning tools specifically for estimating flower cluster number and predicting fruit set after thinning sprays. A review of mango harvest forecasting by Central Queensland University lists weekly estimates of flowering as a core input to a harvest forecast engine, alongside temperature records and fruit load estimates. The practice transfers to any tree crop: count flowers per tree, count set fruit per tree a few weeks later, and the ratio between them tells you things no single glance can.

The single number you should care about: set rate

Fruit drop between flowering and harvest is normal and large. Extension educators in mango country make exactly this point: growers worried about fruit drop are told that some drop is expected, and research presentations to Florida growers are built around separating normal drop from a real problem. California avocado physiologists have documented how much set is lost between flowering and harvest as a matter of course. The consequence for record keeping is simple: an absolute count at any moment is nearly meaningless on its own. What means something is the sequence.

• Flower cluster count per tree at peak bloom: the season's potential.
• Fruit set count per tree 3-4 weeks after petal fall (or panicle tip senescence in mango): what survived the first drop.
• Fruitlet count per tree at the second drop, usually 6-8 weeks after set: what survived the June-style drop.
• Harvest count per tree: the truth, and the denominator that calibrates everything.

Set rate = surviving fruit divided by flower clusters, per tree. Trees with an abnormally low set rate while neighbors are fine are telling you something tree-specific: pollination, nutrition, disease, or a structural problem. That diagnosis is only possible if you counted both endpoints on the same individual tree, which is exactly what a notebook organized by tree number is for.

What to photograph, when

Photo counting beats eyeballing for flowers because flowers come in clusters of hundreds of units, and human counting falls apart above roughly a dozen visible clusters per tree. The practice, per tree:

1. 1Peak bloom: photograph the canopy from two sides, plus close-ups of flowering wood. AI or manual counting of flower clusters per photo rolls up to a per-tree bloom intensity. (Commercial platforms in apples already do this from imaging platforms, which tells you the measurement is robust enough to hang decisions on.)
2. 2Set stage: same two sides, same framing if you can manage it. Comparing a tree's bloom photo against its set photo is the fastest way to see what happened.
3. 3Second drop window: repeat. This pass is where thinning or salvage decisions get made, because per the University of Wisconsin extension's fruit growth rate work, fruitlets destined to drop can be identified by their growth rate well before visible signs appear, and visible signs of drop may not appear for weeks.
4. 4Pre-harvest: fruit count per tree, feeding your yield estimate.
5. 5After harvest: record the actual. Next season's counts mean more with last season's error attached.

A worked example: 12 'Nam Doc Mai' trees

Say you track 12 trees. Peak bloom photos give you flower clusters per tree; the set-stage pass 4 weeks later gives you fruit per tree. Here is the table for four of them (illustrative numbers):

• Tree 01: 46 flower clusters → 38 fruit at set → set rate 83% → on track
• Tree 02: 51 clusters → 12 fruit at set → set rate 24% → investigate (edge row, shaded side faces a windbreak)
• Tree 03: 44 clusters → 0 fruit at set → total set failure → check for cold during bloom, mark for canopy management
• Tree 04: 39 clusters → 41 fruit at set → suspicious (more fruit than clusters suggests counting error or clustered sets) → recount in the next pass

Block average set rate: about 55% across all 12 trees. That single number, plus the per-tree spread, tells you the block is having a normal mango season except for two trees worth a field visit. Without the counts, you would have walked past tree 02's problem until it became a bare spot in your harvest.

The same records feed forward: the bloom counts and set counts, kept weekly or per stage, are what harvest forecast models consume to predict crop load per block per week. Every year you keep them, your block's own ratios replace the textbook's.

Practical rules that keep the data honest

• Count flower clusters, not individual flowers. Units of hundreds are indistinguishable by eye and vary in size; clusters are stable and countable.
• Same tree, both endpoints. Set rate is a per-tree statistic. Block-level bloom counts against block-level set counts hide the trees with problems.
• Photograph within the same window every year (for example, 7-10 days after you first see full bloom on the block). A bloom count taken at 10% bloom cannot be compared with next year's at 90%.
• Watch weather between passes. Chilling hours, heat spikes and rain during bloom explain most set surprises, and a farm-level weather record next to the flowering record is what turns a mystery into a pattern. (Albrigo and Galán's work on floral induction in tropical and subtropical fruit crops is the classic reference on why temperature dominates this.)
• Do not panic-spray from one number. Low set on one tree is a tree problem; low set block-wide is a weather or management problem. They have different fixes.
• Expect the count to fall. A block that sets 50% and holds 35% to harvest is a good block. Tracking is about the shape of the decline, not stopping it.

A checklist you can run each season

1. 1Before bloom: note dormancy/induction weather and any stress treatments applied.
2. 2Peak bloom: 2 photos per tree, flower cluster count per tree, note bloom stage by date.
3. 33-4 weeks post-bloom: fruit count per tree, compute set rate per tree, flag trees under half the block average.
4. 46-8 weeks post-bloom: fruit count per tree again, note the second drop, finalize crop load decisions.
5. 5Pre-harvest: fruit count per tree feeding the yield estimate.
6. 6At harvest: actual fruit and weight per tree, filed against the same tree record.
7. 7Off-season: review the season's ratios per tree; schedule the fixes (pollinators, canopy opening, nutrition) for flagged trees.

How CanopyTwin fits in

CanopyTwin is built around per-tree records, and flower counts are a first-class citizen: the AI counts flower clusters as well as fruit in survey photos, and rolls both up per tracked tree on your map. Walking bloom with your phone gives each tree a dated photo history, so this year's set photo sits next to last year's on the same tree record, and the Arborist chat can flag trees worth a closer look. The Orchard tier covers AI fruit and flower counts from survey photos; every plan starts with a 30-day free trial, no credit card up front.