St. Louis-made AI tool will use space tech to identify plants

For years, scientists have learned about distant stars by looking at the wavelengths of light bouncing back from space.

Now, scientists at the Missouri Botanical Garden are using the same technology, called spectral data, to learn about dried plant specimens in its herbarium.

“The physics behind light reflection apply to the entire universe, whether you’re looking at an herbarium specimen or a star or other distant celestial object,” said Curator of Biodiversity Matt Austin.

The Missouri Botanical Garden Herbarium is essentially a library of pressed plants from all over the world. On a recent afternoon, hyperspectral scanning assistant Naeemah Anderson was there flashing a special light at a tree leaf collected in Peru in 1991.

A rainbow of data popped up on her computer. This is a spectral scan — a measurement of the light bouncing back from the leaf.

“The amount of light reflected off of a leaf tells you really important information about the plant because different chemical traits absorb light at different wavelengths across the electromagnetic spectrum,” Austin said.

Each type of plant has a unique rainbow like this, Austin said, which makes this data particularly useful for identifying plants.

“A hyperspectral scan of a plant is, in essence, a spectral fingerprint for that species,” Austin said.

Austin is part of a new working group that is studying how spectral measurements can be used to learn more about herbarium collections, called the International Herbarium Spectral Digitization Working Group.

The group started at Harvard University and is led by Jeannine Cavender-Bares, who also directs the Harvard University Herbaria. She said that while spectroscopic measurements of living plants aren’t a new concept, it’s relatively new to apply the technique to herbarium specimens.

Cavender-Bares said this data could help scientists understand how plant species are responding to global changes.

“Many of the specimens that are in herbaria don’t exist anymore from where they were collected,” Cavender-Bares said. “And so these are records of what biodiversity was and allow us to understand how it’s changing. And spectral information gives us all kinds of new data on what those specimens were like, their chemistry, their function, all kinds of things we never even thought to look at, and allows us to model change through time and evolutionary history.”

In St. Louis, Austin’s team is using these fingerprints to train an artificial intelligence tool for identifying the plant specimens in herbarium collections. The team’s hope is that the tool could speed up identification of the huge backlog of pressed plants sitting in herbaria around the world, waiting to be processed.

But Austin said that even once the AI tool is up and running, humans will always need to be part of the picture and especially taxonomists, the specialized scientists who identify living things.

“I’ll emphasize this model is in no means a replacement for taxonomists,” Austin said. “We will always need taxonomists. We think that this model will simply make their jobs more efficient and their valuable time able to be allocated to the really challenging plants that a computer can’t handle.”

Austin said there’s a shortage of taxonomists, so part of this project is training new ones, including Anderson. She’s been working with a mentor to develop a specialty in African plants.

“Right now it’s African diasporas, which has been really interesting because I’ve been making determinations on different specimens that have been sitting in the collections that really haven’t been touched,” Anderson said. “My next step with the project is kind of narrowing down where in Africa I want to specialize.”

This work is being funded by an anonymous $14.4 million grant to the garden, which is funding both the AI tool and digitization of the herbarium collection.