AI-Radar Irrigation Used Less Water in Kansas Field Trials

A working field system used materially less irrigation

A center-pivot system in central Kansas did more than turn on the water. As it moved, ground-penetrating radar estimated soil moisture. A camera helped estimate crop development. Weather data informed crop water demand. Software converted those signals into prescriptions for eight irrigation zones.

Across two peer-reviewed on-farm comparisons, fields using this AI-radar system received materially less irrigation than nearby fields using fixed-schedule subsurface drip irrigation. In corn, the AI-radar field also showed lower satellite-derived crop water stress during critical reproductive stages. In soybeans, it recorded higher estimated water-use efficiency and a larger, more uniform harvest.

That combination matters in a region that depends on the declining High Plains aquifer. The studies do not prove that AI alone caused the differences: the adjacent fields were not randomized and used different irrigation hardware. They do show that a sensor-and-control system worked at field scale across corn and soybeans with substantially less applied water.

Research references: [3] [2] [1]

From soil scans to irrigation decisions

Kansas State University researchers evaluated a system that combined several technologies in one loop. Ground-penetrating radar operating from 400 to 1,200 megahertz estimated soil moisture below the surface and recorded about 300 data points every 30 minutes as the pivot moved. A convolutional neural network analyzed crop images to estimate growth stage and canopy cover. Weather information and crop coefficients informed evapotranspiration—the water lost from soil and plants.

The pivot field was divided into eight management zones. Instead of applying one fixed amount across the field, the system varied irrigation according to the estimated deficit in each zone.

That is the practical promise of agricultural AI: not a chatbot telling a farmer what to do, but a sensor-and-control system turning frequent local measurements into physical action under real weather, soil and equipment conditions.

What the corn comparison found

The first peer-reviewed report followed two adjacent corn fields in 2021 and 2023. The AI-radar field covered about 11 hectares and used a center pivot. The roughly 7-hectare comparison field used subsurface drip irrigation on a fixed schedule.

During the reproductive stages, the AI-radar field had a water-deficit index of 0.15 to 0.16, compared with 0.20 to 0.24 in the drip-irrigated field. The authors reported these reproductive-stage differences at p < 0.07; their table marked field differences at p < 0.1. The AI-radar field received 23.5% to 25.1% less irrigation across the two seasons.

One late-vegetative comparison in 2023 was not statistically distinguishable between fields. The study therefore does not show that every indicator improved at every moment.

Research references: [2]

The soybean comparison

A 2026 study compared a soybean field using the AI-radar system with a nearby soybean field using subsurface drip irrigation. Researchers estimated evapotranspiration with both high-resolution drone imagery and OpenET satellite data.

Both methods pointed in the same direction. Drone estimates put water-use efficiency at 7.17 ± 2.41 kilograms per hectare per millimeter in the AI-radar field, versus 5.70 ± 2.13 in the drip field. OpenET produced 6.66 ± 2.24 versus 5.06 ± 1.89. The field differences were statistically significant at p < 0.05.

Grower records indicated that the AI-radar field received 15% to 27% less seasonal irrigation. Its mean soybean yield was 3,189 kilograms per hectare, compared with 2,208 kilograms per hectare in the nearby field—a 30.8% field difference—and yield varied less across the field.

Agreement between the drone and OpenET estimates was moderate rather than perfect: R² values ranged from 0.65 to 0.69, with normalized root-mean-square error of 21.8% to 23.4%. Open satellite-based monitoring could make field-scale evaluation more repeatable, but it is not a substitute for ground truth.

Research references: [1]

What the evidence does—and does not—show

The evaluated outcome is encouraging: in these Kansas comparisons, AI-radar fields received less irrigation and had favorable stress, efficiency and yield measures.

The causal claim must remain narrower. The fields were adjacent, not randomized. They used different irrigation hardware, and their management histories may have differed. The soybean result came from a single on-farm comparison. The studies cannot establish how much of the measured difference came from AI, radar sensing, variable-rate center-pivot application, field conditions, management or other factors.

Some technical descriptions of the commercial system in the corn paper trace to a vendor product guide. Those details explain how the system was intended to work; they are not independent evidence of performance.

The next scientific step is a multi-site, multi-year evaluation that holds more variables constant and compares the system across crops, soils, climates and irrigation equipment. Useful measures would include groundwater withdrawals, energy use, yield and quality, maintenance, missing sensor readings, labor and total cost per hectare.

Why this matters beyond one farm

Irrigation is often managed with incomplete information: a few soil probes, broad weather estimates, fixed calendars or the farmer’s accumulated judgment. The opportunity is to make that judgment more spatially precise.

A moving radar system can sample more of a field than a handful of stationary probes. Computer vision can adjust estimated crop demand as the canopy develops. Variable-rate control can respond zone by zone. Together, those capabilities could help farmers reserve water for the days and places where it protects the most crop value.

That remains a design target, not a demonstrated regional outcome. These Kansas studies do not tell us whether widespread adoption would extend the useful life of the High Plains aquifer, pay for itself on a typical farm or work reliably outside the tested fields. They do show a credible path worth testing: measure more of the soil, infer plant demand more frequently and turn those estimates into smaller, better-timed applications.

The breakthrough is not that AI has solved agricultural water scarcity. It is that an integrated system has produced repeatable field-scale signals, across corn and soybeans, that smarter irrigation may preserve production with materially less water. The question for the next trials is whether those gains travel.