UC Davis team develops mobile tool, providing farmers with real-time plant health data
By EKATERINA MEDVEDEVA — science@theaggie.org
Imagine you could scan a leaf of a plant like a barcode and instantly know its health status. The Leaf Monitor — a mobile tool in the Digital Ag Lab App developed by a team of researchers from the UC Davis Digital Agriculture Laboratory — does just that: It can analyze leaf spectral data using a machine learning model and returns key nutrient values indicative of plant health within seconds.
The funding for this project came from the U.S. Department of Agriculture and the California Table Grape Commission. Led by Alireza Pourreza, director of the Digital Agriculture Laboratory, and run by Parastoo Farajpoor, a Ph.D. student in the department of biological and agricultural engineering, this project brings technological innovation and agriculture together to address one of the key issues in the field: resource allocation.
“One of the biggest challenges that many growers face is figuring out how many nutrients each plant actually needs,” Farajpoor said. “It is a complicated issue because different plants, even within the same field, can have different nutrient [needs]. On top of that, not all of the fertilizer applied to the soil ends up being used by the plants. This makes nutrient management difficult and often inefficient.”
The traditional approach to evaluating specific nutrient levels in plants is slow, labor-intensive and expensive. Usually farmers need to collect leaf samples from across the field, dry and grind them up, send them to the laboratory and then wait for several weeks until the results come back. In consequence, valuable time is lost when crucial decisions could have been made for improving the yield and quality of the crops.
“[The traditional analysis method] also lacks spatial resolution because the results usually represent an average over a large block or the entire farm, rather than showing the condition of individual plants,” Farajpoor said. “This often leads to overapplication or underapplication of fertilizer in different parts of the field, which increases cost and can harm the environment.”
This is where the Leaf Monitor comes in. For the past five years, the team at Digital Agriculture Laboratory has been building up a database that matches spectral data from leaves of different plants to the nutrient values, biochemical and structural leaf traits that have been retrieved through the traditional chemical and structural analyses. They then used it to train a machine learning model, which is now uploaded to a cloud service, so that it can be accessed to make predictions from new spectra in real time. This enables farmers to make faster and better crop management decisions.
The process of using it is simple: First, spectral data is collected by a hand-held spectrometer that is connected via bluetooth to the mobile app by scanning a leaf. A research paper by Purdue University researcher John J. Couture, titled “Spectroscopic determination of ecologically relevant plant secondary metabolites,” details how on the microscopic level, the light that the spectrometer shines onto the leaf gets absorbed at certain wavelengths and reflected at others.
“[The] vibrational excitation of molecular bonds, primarily C–H, N–H and O–H bonds at specific wavelengths [is] in the visible (400–700 nm), near-infrared (700–1100 nm) and shortwave infrared (1100–2400 nm),” the paper reads.
As a result, a characteristic absorption spectrum is produced, which is sent to the cloud to be processed by the trained model that returns the nitrogen levels, water content and other values reflecting the crop’s condition to be displayed in the app.
In the long run, this technology can help significantly reduce costs for farmers by enabling optimized fertilizer use, as well as support sustainable agricultural practices.
Reflecting on the performance of the Leaf Monitor so far, Farajpoor noted that it received a lot of attention from growers.
“We tested the app on some farms, and the ease of use and the speed of getting results brought a lot of positive feedback,” Farajpoor said. “Growers appreciated that they could simply scan a leaf and get useful information within seconds, without needing any special training […] [It] feels [motivating] when you see your research actually working in the real world.”
The team is currently working on expanding and diversifying the training database for their model in order to raise the accuracy of predictions across various plant types.
“One challenge is that different crops have different spectral responses and trait ranges, so a model trained on one crop may not work well for another,” Farajpoor said. “The early impact has been promising, and we are excited to keep improving it so that it can be used more widely and effectively in real farm conditions.”
As it continues to be developed, the Leaf Monitor is a powerful demonstration of the fruitfulness of cross-disciplinary research: in particular, the projects at the intersection of agriculture, sensing technologies and machine learning.
“I think agriculture is incredibly important because it provides food for people, and this kind of technology can really help us use our resources more wisely,” Farajpoor said. “So, in a way, this project supports both the grower and the environment.”
Written by: Ekaterina Medvedeva — science@theaggie.org
