AI for Cotton Leaf Area Measurement: How Petiole Pro Matched the LI-COR 3100 in Independent Research

The study at a glance

The paper — “Use of mobile applications for determining the leaf area of cotton plants” by Duarte and colleagues — appeared in Nativa: Pesquisas Agrárias e Ambientais (v. 14, n. 1, e20143, 2026). It was submitted in July 2025 and published in March 2026 under an open DOI. The authors set out to answer a practical question: can a phone-based app measure cotton leaf area as reliably as a dedicated laboratory instrument?

The experiment ran in a greenhouse at the Federal University of Rondonópolis (Mato Grosso, Brazil) from June to October 2023. Four commercial cotton cultivars were grown, and leaf area was measured weekly from 30 days after planting — 85 leaves per cultivar, each measured by four different methods, giving a large, like-for-like dataset rather than a handful of cherry-picked leaves.

Why cotton leaf area is worth measuring

Leaf area sits close to the centre of how a plant lives. It governs how much sunlight a canopy intercepts, drives photosynthetic capacity and the production of photoassimilates, and is one of the first traits to respond when a plant is stressed by water, nutrients, heat, pests or disease. For a commodity crop like cotton, grown at scale and under tight economic margins, tracking leaf area is a direct window into plant performance.

The catch is how you measure it. The traditional route is destructive: cut the leaf, carry it to a laboratory, and run it through a bench instrument such as the LI-COR 3100. That gives an accurate number — but it kills the leaf, so you can never measure that same leaf again, and you disrupt the plant's carbohydrate balance in the process. A non-destructive method that leaves the plant intact lets researchers follow the same leaf through an entire growing cycle.

This is exactly why the study matters. If a phone app is accurate enough to trust, leaf-area monitoring stops being a one-shot, destructive lab task and becomes a repeatable field measurement — the kind you can run every week on the living plant.

How the researchers tested each method

The team compared four ways of arriving at a cotton leaf's area, using the same leaves for each:

• Petiole Pro app — a photo-based measurement taken with an Android phone, calibrated using the app's reference images.
• LeafArea app — a second mobile application, which requires the user to measure and enter each leaf's length to set the scale.
• Manual dimensions — leaf area estimated from length × width × a correction factor (LA = W × L × f).
• LI-COR 3100 — the benchtop laboratory leaf-area meter, adopted as the reference standard against which the other three were validated.

For the Petiole Pro measurements the researchers followed the app's own workflow: they calibrated the area scale first, positioned the camera perpendicular to a fixed-height bench, and placed each leaf on a white background film to sharpen the contrast between leaf and surface before the app quantified the area automatically.

That white background and fixed geometry are not fussy details — they are the whole basis of a trustworthy reading. The same discipline applies whether you are in a greenhouse or a field, which is why we have written a separate guide on how to photograph leaves for accurate leaf-area measurement.

Every leaf was then cut and sent to the laboratory for the reference measurement on the LI-COR 3100. With four numbers for each of 340 leaves, the researchers could score every method with a battery of statistical indices — correlation, error, efficiency, agreement and a composite performance rating.

What the results showed

The headline is simple: Petiole Pro gave the most accurate leaf-area estimates for three of the four cultivars — TMG44B2RF, IMA5801B2RF and FM985GLTP — and was classified as “excellent” on every one. For the fourth cultivar, BASF – FM 944GL, the manual length-and-width method edged ahead, though Petiole still scored an excellent rating. The competing LeafArea app was consistently the weakest of the methods.

*For BASF – FM 944GL, the manual length × width × correction-factor method was the single most effective, with Petiole close behind. Values from Duarte et al. (2026), Tables 1–4 and Figures 1–4.

Read the scatter plots and the pattern is easy to see. Where a method tracks the LI-COR reference perfectly, its points fall on the diagonal. Petiole's points hug that line tightly; LeafArea's scatter far more widely.

What “excellent” means in the statistics

A single correlation number is easy to over-read, so the study leaned on a whole panel of indices — and it is worth knowing what each one is telling you, because together they are what turns “the app looks good” into “the app is validated.”

• Pearson's correlation (r): how tightly the app's numbers move with the reference. Petiole reached r = 0.988 for TMG44B2RF — very close to a perfect 1.0.
• Coefficient of determination (r²): the share of variation the app explains. Petiole hit r² = 0.976; LeafArea fell as low as 0.437.
• RMSE and MAE (error): how far off the estimates are on average. Petiole's were the smallest of the tested methods.
• Method efficiency (ME) and agreement index (d): whether the method reproduces the reference's pattern, not just its trend. Petiole reached d = 0.999.
• Confidence index (c): a composite of correlation and agreement, mapped onto a labelled scale by Camargo & Sentelhas (1997). Petiole's landed in the “excellent” band; LeafArea's ranged from “average” to “very good.”

No single index carried the verdict. Petiole earned the top classification because it was strong across all of them at once — high correlation, low error, high efficiency and high agreement — which is a far harder bar to clear than a good-looking r on its own.

From the laboratory to the field

The most exciting part of this result is not the leaderboard. It is what an accurate, non-destructive method unlocks. Because you never cut the leaf, you can measure the same cotton leaf again next week, and the week after — turning a one-off snapshot into a growth curve.

That is precisely the kind of repeated measurement that lets researchers and agronomists see how a cotton plant responds to the things that actually drive a season:

• fertilisation and nutrient management;
• irrigation and water availability;
• climate and heat stress;
• pests and disease pressure;
• and the management decisions made in response to all of the above.

It also lowers the barrier to entry. A LI-COR 3100 is accurate but expensive and lab-bound; a phone is already in the researcher's pocket. In regions where access to laboratory equipment is limited — much of the world's cotton belt included — a validated app is not a compromise, it is the difference between measuring and not measuring at all. The same shift toward measuring plant health directly on the living plant runs through our work on AI assessment of necrotic leaf area damage, and it underpins Petiole Pro's wider role as an on-device plant-phenotyping tool.

Measure cotton leaf area yourself

You do not need a greenhouse trial to reproduce the method the researchers used. The free Petiole Pro app runs the same on-device measurement — and the workflow is deliberately simple.

1. Calibrate. Place your leaf on a Petiole Pro calibration plate, or set the area scale using the app's reference images, so pixels can be converted into real cm².
2. Photograph. Shoot perpendicular to the leaf, flattened against a plain, high-contrast background — the white film the researchers used works well.
3. Analyse. The app segments the leaf from the background and returns its area and perimeter in real units, on the phone, in seconds.
4. Repeat on a timeframe. Because nothing is destroyed, measure the same leaf again every week to build a curve rather than a single point.

Here is that Leaf Analysis output on real leaves — each one segmented cleanly from its background, with area and perimeter reported directly:

Whether you need the calibration plate at all depends on what you are measuring and how; we cover that decision in detail in do I need a calibration plate for Petiole Pro?

Honest limits and good practice

A validated tool is not a magic one, and the study is refreshingly candid about it. Non-destructive digital methods still need calibration and validation for each species, and some images require manual adjustment before the segmentation is right. The result for the cultivar where the app placed second, BASF – FM 944GL, is a reminder that leaf morphology varies and no single method is best for everything.

The practical rules that follow from the paper — and from years of measuring leaves with the app — are worth keeping in view:

• Lighting, background and leaf flatness matter. A plain, high-contrast background and even light are what make the segmentation clean.
• Calibrate for absolute numbers. Percentages need only a photo; real cm² needs a known scale in frame.
• Validate for your crop. An app proven on cotton is not automatically proven on your species — check it against a reference the first time.
• Expect the occasional manual touch-up. Torn edges, overlapping lobes and awkward shapes sometimes need a corrected outline.

This is how practical AI in agriculture should grow: not as a black box, but as a tested tool, compared openly against trusted reference methods — and used with judgement.

The story behind the study

Petiole Pro's founder, Dr Maryna Kuzmenko, shared the result on LinkedIn — not to celebrate a scoreboard, but because independent academic validation against a laboratory reference is the strongest kind of trust a measurement tool can earn. “Accuracy first” has been the app's motto since the beginning.

It is part of a broader conversation the team runs on AI in agriculture — from leaf phenotyping to the AI agents that will one day act on this kind of data automatically.