A hydrated leaf is a healthy leaf. That’s true for the leaves of crop plants in a farmer’s field and for the leaves of trees in an area vulnerable to forest fires.

But the traditional techniques to monitor leaf hydration require cutting them from their plants, which is time-consuming and cannot give live measurements. That’s why many researchers are building sensors that measure a plant’s health in real time.

Now, researchers in Texas have developed a graphene “tattoo” that can be stuck directly onto a leaf to provide real-time moisture readings. The researchers also believe it could one day be the building block for a new kind of plant monitoring, by turning the patches into a neural network that computes on the plants themselves.

“Not only are we just sensing the moisture level, but we can have that sensor act as this artificial synapse, which then we can put into a neural network,” says Jean Anne Incorvia, an associate professor of electrical and computer engineering at the University of Texas at Austin. Incorvia and colleagues (including her graduate student Utkarsh Misra) published their work in Nano Letters in February.

A forest of the future, Incorvia and colleagues think, might hold a whole grove of sensors networked to gauge the risk of fire or drought in real time.

A Graphene Leaf “Tattoo” as a Moisture Sensor

The sensor is a graphene patch that can be pasted onto the leaf of a plant (the researchers used Monstera) like a stick-on tattoo. It’s functionally a sort of three-terminal transistor, with a graphene channel, gold strips as its electrodes, and the leaf itself as its dielectric insulator.

The sensor can gauge a leaf’s hydration level in real time by sending an electric pulse into the leaf, which moves around ions within the leaf and changes the graphene’s conductance. The magnitudes of these conductance changes depend on moisture inside the leaf, so the researchers can read out a leaf’s hydration without a need for external processing.

Graphene is a good material for a leaf tattoo. It’s nearly transparent, so it won’t block light and disrupt photosynthesis. It can stretch and squeeze as the leaf grows, shrinks, or twists.

This isn’t the first graphene sensor of its kind, but real-time hydration sensors aren’t common in the field. The researchers hope this new sensor can change that by fitting into a neural network, because it also acts like a synapse in a brain.

In particular, just as neural activity can strengthen or weaken a synapse, the researchers could use particular electric pulses to slightly adjust their sensor’s conductance up or down. Moreover, after a pulse ended, the sensor returned to its original conductance slowly, over about 90 seconds. In that time, their sensor could act as a sort of short-term memory.

The researchers imagine they could one day use such artificial synaptic qualities to tune and store a neural network’s weights.

Researchers study a Monstera plant in the lab, with sensors pasted on each leaf. Andrew Carr/UT Austin

Neuromorphic Plant Computing

For several years, Incorvia’s group has designed non-leaf-based devices for these kinds of neuromorphic computing. They’ve typically crafted transistors with graphene and Nafion, a polymer that’s a good proton conductor. With a current pulse, the transistors can control how many protons migrate inside the Nafion—and, in turn, how many electrons cross the graphene channel. Thus, their devices can take on different weights inside a network.

Maya Borowicz, an undergraduate visiting Incorvia’s lab for one summer, noted that a plant leaf could conduct protons too. Why not swap out the Nafion, Borowicz suggested, to make a device that’s part-leaf?

“We actually did it a few years ago, and it worked, and we were like, ‘This is cool,’” Incorvia says. At first, they weren’t sure how to use it. “We kind of just tabled it.”

Months passed before Incorvia had an encounter with Ashley Matheny, a geologist working on better ways to monitor moisture levels in forests. “Through talking to her, I realized…there actually is a good value proposition for needing these types of sensors,” Incorvia says.

Now, in their work, Incorvia and colleagues demonstrated one possible future. They trained a relatively simple neural network called a single-layer perceptron to examine their sensor’s readings and classify the leaf as hydrated, normal, or in drought conditions.

This model ran on external hardware, but the researchers hope the sensor’s quality as an artificial synapse can help it run similar networks on plants themselves.

Incorvia envisions a network that ties leaf-mounted sensors together with others in soil and tree sap. Farmers could use it to monitor their fields in the face of climate change-induced drought; forest rangers could receive live updates on the numbers of dry leaves that could kindle a fire.

“You could imagine a neural network of trees, where we could be sensing across the forest,” Incorvia says.