CSIRO’s Hyperspectral Drone Cuts Wheat Trait Testing From Weeks to Minutes
16 November 2026, Canberra, AU: Scientists at Australia’s national science agency, CSIRO, have built a drone-mounted camera system that can estimate hidden physiological traits inside wheat plants, such as how fast grain is filling or how efficiently a plant is photosynthesizing, from the air, work that could eventually let breeders screen thousands of plots for traits that currently require weeks of manual measurement.
The system, developed under a project called AcceleTraitPlus, a partnership between CSIRO, the Grains Research and Development Corporation (GRDC) and breeding company Australian Grain Technologies (AGT), was detailed in a CSIRO article published November 9. The project uses a drone nicknamed “Mjolnir” fitted with a hyperspectral camera, a sensor that splits reflected light into hundreds of narrow wavelength bands spanning 350 to 2,500 nanometers, more than seven times the range of light visible to the human eye.
Explaining how the hyperspectral data translates into plant trait information, lead CSIRO researchers Aswin Singaram Natarajan and Ryan Lagerstrom, along with team leader Dr. Gonzalo Estavillo and project leader Dr. Anton Wasson, describe each leaf’s internal physiological state as giving off a distinct spectral “fingerprint.” Changes in a plant’s photosynthetic activity, respiration or grain-filling dynamics subtly alter how its leaves reflect light across those hundreds of wavelength bands, even when nothing is visible to the naked eye. By capturing and analyzing that full spectral fingerprint, the system aims to estimate internal physiological processes without needing to physically sample the plant.
From hand-held sensors to whole-field mapping
To build the system, researchers first used a hand-held leaf-clip spectrometer, a device that measures the same type of spectral data from a single leaf at a time, to create reference models linking specific spectral signatures to physiological measurements taken by hand. Those reference models were then scaled up using imagery collected from the Mjolnir drone, allowing the same relationships identified in individual leaves to be applied across entire trial plots. Machine learning models, which the team said are progressing from simpler linear regression methods toward more advanced deep-learning approaches, process the resulting hyperspectral image data, while cloud computing handles the large data volumes involved and allows results to be shared with breeders close to the time of capture.
(function() { var grid = document.getElementById(‘hln-grid’); if (!grid) return; fetch(‘https://www.global-agriculture.com/wp-json/wp/v2/posts?per_page=4&_fields=title,link,date’) .then(function(res) { return res.json(); }) .then(function(posts) { if (!Array.isArray(posts) || posts.length === 0) { grid.innerHTML = ”; return; } grid.innerHTML = posts.map(function(p) { var d = new Date(p.date); var dateStr = d.toLocaleDateString(‘en-GB’, { day: ‘numeric’, month: ‘short’ }); var title = p.title && p.title.rendered ? p.title.rendered : ”; return ‘
‘; }).join(”); }) .catch(function() { grid.innerHTML = ”; }); })();
The project ran field trials across the 2024 and 2025 growing seasons at breeding sites in Western Australia and New South Wales, with results presented at the Australian Crop Breeders Week conference. To validate the drone-based estimates, the researchers compared them against a conventional, labor-intensive measurement process: manually tracking grain fill typically requires 10 to 12 separate site visits across a season, from flowering through harvest, with researchers hand-collecting and weighing thousands of individual wheat heads. By contrast, a single drone flight can image an entire trial site in minutes. CSIRO described the system’s ability to predict grain-filling rate, a trait closely linked to final grain yield and quality, as possibly the project’s strongest result to date, though the organization has not yet published specific accuracy figures, such as error rates or correlation statistics, for that or other traits, saying only that the approach works reliably at scale based on the trial comparisons so far.
Dr. Estavillo cautioned that the technology still has a long way to go before it is ready for everyday use in commercial breeding programs, and the team is now testing its practicality with an industry breeding partner to see how it performs outside a research setting. No product name, pricing or commercial release date has been set, reflecting the project’s current stage as applied research rather than a market-ready tool.
If the approach continues to validate well, it would address a long-standing bottleneck in wheat breeding: traits tied to internal plant physiology, such as photosynthetic efficiency or the pace of grain filling, are difficult and slow to measure directly across the thousands of plots a breeding program needs to evaluate each season, which limits how quickly breeders can select for them. Replacing weeks of manual sampling with a single aerial flight would let breeding programs screen far more material for these traits within a season, potentially speeding up the release of new wheat varieties with improved yield potential or grain quality.
For seed companies and grain breeding programs elsewhere, including those breeding wheat and other cereals in similar environments, the underlying approach, pairing hand-held reference measurements with scalable drone-based hyperspectral imaging, offers a template that could extend beyond wheat to other field crops where internal physiological traits are currently too labor-intensive to measure at scale.
Also Read: Crystal Crop Protection to Launch Rs 455 Crore IPO on November 18
Global Agriculture is an independent international media platform covering agri-business, policy, technology, and sustainability. For editorial collaborations, thought leadership, and strategic communications, write to pr@global-agriculture.com
