The AI that sprays only where a weed is present

Precision agriculture is not simply about collecting more data. It works when a decision changes something concrete in a field. Targeted spraying is a clear example: instead of treating an entire area with herbicide, a camera and a model identify where a weed is present and open only the nozzle that is needed.

In 2024, Iowa State University Extension evaluated that approach with a John Deere See & Spray Ultimate machine on five conventionally managed soybean fields. The demonstration covered 415 acres in Boone and Story counties. It was not a laboratory experiment or a catalogue promise: it was field-scale work using application maps, drone imagery, and crop follow-up.

How the application works

The system mounts cameras along the sprayer boom. As it moves, computer vision distinguishes the target vegetation and sends a command to onboard processing modules. Those modules activate individual nozzles to apply herbicide only where a weed is detected; areas without that signal do not receive the post-emergence dose.

The logic sounds simple, but it requires coordination between vision, stable boom height, vehicle speed, individual nozzle control, and product recirculation. Iowa State notes that today’s precision-spraying systems typically operate between 10 and 15 miles per hour, and that accuracy depends on keeping those components in working order. The farmer still defines the weed-management programme; AI handles one bounded task within it: deciding which area to treat during a pass.

What Iowa measured

The five fields received similar pre-emergence residual programmes, but entered the post-emergence pass with different weed pressure. That difference showed directly in the results. Product savings were 87.2%, 43.9%, 71.2%, 90.6%, and 87.6% by field. The average was 76%, equal to 4,700 gallons of tank mix and roughly US$6,500, or US$15.7 per acre, in product savings across the 415 acres.

Not every field produced the same figure, and that is part of the case’s value. The field with the heaviest weed pressure saved only 43.9%; fields with stronger residual control saved more than 87%. Imagery and post-application follow-up confirmed that the system identified and treated emerged weeds and that fields stayed clean through canopy closure.

The growing-season scale also provides context. John Deere said its See & Spray technology was used on more than one million acres in 2024 and estimated average herbicide savings of 59% in corn, soybean, and cotton compared with broadcast spraying. The company estimated eight million gallons of tank mix saved. Those are manufacturer figures, while the Iowa work offers a smaller, independent, and more detailed measurement.

What it improves, and what it does not promise

The immediate benefit is less product applied where there is no weed to treat. That can lower input costs, transport burden, and unnecessary exposure of crops or the environment to an application. “Where applied” maps add useful information as well: they reveal higher-pressure areas that can inform later management.

But spraying less does not mean abandoning weed management. The university demonstration explains that savings are a function of initial weed pressure: a heavily infested field comes closer to a full application. It also does not by itself demonstrate higher yields or make every chemical mixture sustainable. Crop type, weed resistance, residual strategy, weather, and machine calibration still matter.

John Deere likewise warns that results vary with field conditions, weed pressure, settings, software version, and operating conditions. That caution is necessary. A model can misclassify a plant, a camera can read poorly, and a misconfigured nozzle can compromise the application. Agronomic inspection and map review are therefore not optional tasks.

The transferable lesson

This case illustrates a useful agricultural-AI application: computer vision turns one “entire field” decision into many small, checkable decisions. The system does not replace the farmer or determine the whole crop strategy. It reduces one concrete physical action, spraying, when the data indicate that it is not needed.

To adopt this approach rigorously, a farm should measure product saved, effective weed control, cost, yield, and detection errors in its own fields. The promise is not automated agriculture. It is a more precise application, monitored and adapted to each field’s real variability.