A Penn State App Uses AI to Score How Much Pollinators Like a Flower, and All It Needs Is 10 Minutes and a Cardboard Square

Plant tags at a garden center often promise that a flower is pollinator friendly. They rarely say how anyone checked. Researchers at Penn State have built a way to check that fits on a phone: an app called FloraCount, which lets a gardener with no training in entomology rate how attractive a plant is to bees, butterflies and flies in about 10 minutes.

FloraCount comes from Penn State’s Center for Pollinator Research and is available in the Apple App Store and Google Play. The center describes it as an AI-enabled tool meant for researchers, conservationists, landscapers and home gardeners alike. Its makers include entomologists Christina Grozinger, Harland Patch and Edward Amoah, along with graduate student Hannah Thomas, according to Penn State’s Insect Net research page.

With fall planting season underway in the low desert, the app is a timely look at a question many East Valley gardeners ask when they shop: which of these flowers will actually feed the bees?

How an observation works

The app’s instructions are short. After creating an account and logging in, a user starts a new observation and types in the plant’s genus, species and cultivar if known. The app then asks for four landscape photos taken at eye level, one each facing forward, right, behind and left. Those pictures capture the habitat around the plant.

Next comes the one piece of equipment the app requires: a 3-by-3-inch square that serves as a size reference. The center says brown cardboard works best, and that taping the square to a stick makes it easier to hold beside taller plants. With the square next to the flowers, the user takes a photo from directly above the plant, about 3 feet up, with the phone held level with the ground.

Then the timer starts. For 10 minutes, the observer watches the plant and taps an icon each time a bee, butterfly or fly touches it. Other insects are not counted. The rules are specific: an insect that leaves the field of view and comes back counts as a new visit, while an insect that works its way around one clump of flowers counts once. When the timer ends, the app produces a floral attractiveness score. Users can review past records in a web browser and export them as a spreadsheet file.

Where the AI comes in

The overhead photo and the cardboard square feed a piece of software the same Penn State group described in the journal PLOS ONE in September 2025. Called FloralArea, it uses two image-recognition models, one that picks out the flowers in a photo and one that finds the reference square. Because the square’s real size is known, the software can compare pixels and estimate how many square centimeters of bloom the plant is showing.

Floral area matters because, as the researchers write, it is a major predictor of how attractive a plant is to pollinators, but measuring it by hand is slow and inconsistent from one study to the next. In their test of 75 images of flowering plants, FloralArea’s estimates lined up closely with careful manual measurements, with a coefficient of determination of 0.93. The team reported that it cut the time needed to calculate floral area from an image by more than 99 percent compared with doing the work in a standard image-analysis program.

The paper is candid about the weak spots. The model that finds the reference square performed very well, but the flower-finding model’s accuracy was described by the authors as moderate. The app pairs that photo with a human count rather than relying on the image alone.

Why 10 minutes is enough

The short observation window comes from earlier research. In a 2022 study in the Journal of Economic Entomology, Grozinger, Patch and lead author E. Erickson analyzed pollinator observations collected at Penn State Extension’s flower trials. They concluded that 10 minutes is long enough, for most plant types, to estimate how many visitors a plant draws, which groups of pollinators it attracts and how it ranks against other varieties, without giving up much accuracy.

The same study found that properly trained non-experts could collect accurate data, which is what makes a phone app plausible in the first place. It also found that attractiveness is not fixed. Climate and the stage of a plant’s bloom affected the results, and while some cultivars drew pollinators consistently, others did so only under certain conditions. The researchers suggested that protocols be designed to keep results consistent across many different observers.

That is why FloraCount gathers more than a bug count. Penn State’s Insect Net page describes it as part of a suite of tools that also accounts for floral area, landscape quality and weather.

Volunteers are helping refine it

The app is still being improved with help from gardeners. Penn State reported in June that Thomas was partnering with the Penn State Extension Master Gardener program to organize data collection events at Master Gardener demonstration gardens across Pennsylvania this summer. The university said the volunteers’ observations would be used to refine the tool and improve its ability to evaluate pollinator-friendly plants.

What other research says about cultivars

FloraCount was built in part for the ornamental plant industry, which releases hundreds of new flower varieties every year, according to the center. A study published this year in the journal Ecosphere shows why that kind of testing is useful.

Researchers led by Nicholas Dorian at the Chicago Botanic Garden compared four native wild-type flowering plants with 13 cultivars, using a two-year common garden trial and a five-year community science project across eastern North America. Wild-type plants were consistently the most attractive to pollinators. Some cultivars matched their wild relatives, others drew fewer visitors, and none did better. The authors recommended wild-type plants and cultivars shown to match them, and cautioned against assuming that a cultivar is ecologically equivalent to the native plant it was bred from.

The study also found that results from the controlled garden trial lined up closely with the community science data. The authors wrote that this suggests garden trial results may hold across varied landscapes, and that observations by volunteers can add to the science at a large scale.

A note for desert gardeners

FloraCount was developed and tested in Pennsylvania, and the research behind it comes from gardens in the eastern United States. Nothing published so far shows how its scores compare across Sonoran Desert plants or Arizona’s pollinators, and the 2022 study is a reminder that climate and timing change the results. A plant that scores well in a Pennsylvania flower trial will not necessarily do the same in a San Tan Valley yard in October.

That does not make the method less useful here. The core habit it teaches, standing still for 10 minutes and counting who visits, works in any garden. A gardener comparing two varieties of penstemon or salvia side by side can learn a lot from a few careful counts on calm, sunny mornings at the same time of day, with or without an app.

For anyone who wants to try it, the step-by-step guide and account registration link are on the Center for Pollinator Research’s FloraCount page at pollinators.psu.edu. The only supplies needed are a phone, a square of cardboard and 10 uninterrupted minutes beside a flower.