Camera-Based Targeted Spraying in US Row Crops: Measured Savings, Costs and Limits

The use case

A targeted sprayer replaces a uniform pass with one in which boom-mounted cameras and onboard processors decide, nozzle by nozzle, whether a weed is present, and spray only then⁶⁷. Green-on-brown systems find weeds on bare soil or stubble and have existed since the 1990s. Several use colour or chlorophyll-fluorescence sensors rather than learned models⁴¹⁰. Green-on-green systems find weeds inside a growing crop and depend on machine learning⁴¹⁰.

Three sources of savings must be kept apart. Machine-learning detection is one. Older tools are another: automatic boom section control cut sprayed area by 2.3% to 14.5% in a University of Kentucky study cited in a Kansas State thesis¹². The third is farming practice, chiefly broadcast residual herbicides early in the season, which lowers weed pressure and raises later savings⁴⁵. Laser and mechanical robotic weeders exist but are outside this paper⁶.

Where it is running

John Deere's See & Spray, launched in 2021, is the largest deployment. Deere reports more than 5 million acres in 2025 and an average cut in non-residual herbicide of nearly 50%, about 31 million gallons of spray mix (vendor-reported, 2025 season)⁷. Ultimate is factory-fitted with two tanks. Premium is a single-tank retrofit for model-year 2018 and newer Deere sprayers³⁷. Other systems include Greeneye (retrofit), Bosch-BASF One Smart Spray, Trimble Bilberry, Carbon Bee and Ecorobotix ARA¹⁰. Greeneye and Deere machines were trial-tested in Nebraska in 2022 and 2023².

Ecorobotix, in a peer-reviewed article that uses its own machine data, reports 57 ARA machines making 1,682 onion missions over 8,266 ha in 2023, and 33 machines covering about 2,000 ha of sugar beet⁶. Five million acres is roughly 3% of the 177.6 million US corn and soybean acres used in the Kansas State thesis (editor's calculation; Deere's total includes fallow)⁷¹². A 2024 Government Accountability Office report, as summarised by GROW, found only 27% of US farmers use any precision-agriculture technology⁹. No European deployment data were found.

What the evidence shows

Measured in trials. In Arkansas, a three-year soybean trial (2022–2024, one site, conventional tillage) sprayed on average 41% to 58% of the area at early post-emergence and about 48% to 49% at mid post-emergence. It saved $43.22 to $129.19 per hectare against a broadcast post-emergence cost of $227.22¹. In Nebraska corn in 2022, a Greeneye sprayer saved 94% and 87% of non-residual herbicide at two timings. Weed density before the second pass was 0.1 to 0.2 plants per square metre. Palmer amaranth control matched the double-broadcast comparison at 99%, while giant foxtail reached 90% to 94% against 99%². Iowa State extension tested See & Spray Ultimate on five conventionally tilled fields in 2024. Savings ranged from 43.9% to 90.6%, averaging 76%, or $15.70 per acre⁴. In a one-season Swiss trial, Ecorobotix's ARA gave 92% control in onions against 99% for broadcast, a difference its authors report as not significant (manufacturer-linked)⁶.

Self-reported (2024): a Wisconsin grower estimated savings of 65% in post-emergence corn, 60% in 30-inch soybeans and 36% to 40% in 15-inch soybeans⁴. Vendor-reported: Deere's 2025 fleet average of nearly 50%⁷, and Ecorobotix's 78.9% volume saving in onions and 79.8% in sugar beet in 2023, drawn from machine dashboards⁶.

Yield effects are inconsistent. Arkansas recorded about 130 kg/ha more soybean at the lowest sensitivity setting, which the authors called difficult to explain¹. Nebraska found no yield difference in 2023 soybean². Deere-commissioned trials in seven states reported +2.0 bushels per acre on average (vendor-reported)⁷. On speed, the Arkansas machine ran at 19.3 km/h against about 24.2 km/h typical for broadcast sprayers¹.

All of these figures come from machine-learning systems. Even so, the size of the saving is set largely by the residual programme: the better the upfront residual work, the more targeted spraying saves⁴⁵.

What it takes: cost, integration, governance

Equipment. The Arkansas economics assumed an upgrade of $40,000 where a sprayer already has GPS and nozzle bodies, or $80,000 for a base machine, on a $600,000 sprayer. These are the authors' assumptions¹. Premium needs Deere's ExactApply nozzle control, BoomTrac Pro 2 and a G5 CommandCenter display¹⁰. No published retrofit price for non-Deere systems was found.

Fees. Deere charges $5 per in-crop acre and $1 per fallow acre, only on acres not sprayed (vendor-reported, 2025)⁷⁸. It states that "savings" means volume savings only, not other costs⁸. An unlimited annual licence is offered from 2026, with no price given⁷. Arkansas break-even was 218 to 1,273 hectares a year, about 540 to 3,150 acres¹.

Operation. Operators choose a sensitivity setting that trades savings against missed weeds. Calibration, lighting and sensor height matter, and staff need training¹¹⁰. Both Nebraska machines were labelled for 76 cm rows². Premium cannot apply a residual and a targeted product in one pass⁵.

Rules. EPA's Mitigation Menu (updated April 2025) awards runoff points for the share of a field left untreated, naming smart sprayers: 2 points for 10–30% untreated, 3 for 30–60%, and 4 at the highest band. It applies only where a label requires points¹¹. EPA staff said labels have not yet accounted for targeted application, and an ASABE committee is drafting test standards⁹. The EU's Directive 2009/128/EC sets the pesticide-use framework. No EU rule specific to targeted spraying was found¹³.

Where it falls short

Baselines. The headline 50% to 90% figures cover non-residual post-emergence products, usually applied after a broadcast residual, not the whole herbicide programme. Arkansas broadcast its pre-emergence herbicide on every plot¹. In 2023 Nebraska soybean with no residual and kochia up to 91 cm tall, savings were 1.1% to 5.4% at the first pass and zero at the second, although weed control matched broadcast². Evidence covers mainly corn and soybean, one to three seasons, and one or a few sites. The Arkansas trial was funded by Deere's Blue River and the United Soybean Board. Nebraska was partly funded by Deere and Greeneye, the seven-state yield work by Deere, and the onion data come from Ecorobotix¹²⁶⁷. No wholly independent multi-season trial was found. The Iowa State and Wisconsin material is closest, and it was read only through trade press and extension commentary. Deere's average is unaudited, and its fee is charged on the same unsprayed acres⁷⁸.

Does control hold with fewer weeds sprayed? At the highest sensitivity, yes. At the lowest, no. Arkansas weed density rose from 867 to 2,870 to 11,300 plants per hectare over three years. Harvest escapes averaged 1,380 per hectare against 268 for broadcast. The chance of treating a 1.9 cm amaranth fell from over 99% to about 55%¹. The lowest setting sprayed less only at the first pass (41.3% against 57.9% of area)¹. One Arkansas Palmer amaranth population resists seven post-emergence herbicide sites of action, and whether targeting speeds resistance is unmeasured¹. Dense canopies make green-on-green detection harder and may need an extra pass¹⁰. Savings were lower in 15-inch soybeans and late passes⁴⁵, and grass control trailed broadleaf control².

Typical farms. Arkansas break-even needs hundreds to thousands of acres and assumes a compatible sprayer¹. Whether the Iowa $15.70 per acre is net of fees is not stated⁴. No source reports adoption by farm size. Indirect evidence is the cost barrier⁹ and a Syngenta expectation that large-acreage Midwest fields will adopt first⁹.

Stage: In early use — why

The hypothesis holds. Commercial use is real: one vendor reports over 5 million acres in 2025 under a fee model, and several competitors sell machines⁷¹⁰. But that is roughly 3% of US corn and soybean acres. Labels do not yet reflect the technology, test standards are still being drafted, and published trials span at most three seasons¹⁹¹². The evidence is too thin for "In wide use" and the deployment too large for "Pilot with results".

What to watch next

• Arkansas follow-up. The authors say further work is needed on whether the yield gain persists and whether resistance shifts. The trigger is a published follow-up¹.
• ASABE X665 committee. It is drafting test standards and proposed label wording. The trigger is a published standard or EPA label language⁹.
• EPA Mitigation Menu. EPA intends to update it each fall, so check whether the autumn 2026 update changes points for smart sprayers¹¹.
• Deere 2026 season. Watch its reported results and uptake of the unlimited annual licence, new for 2026⁷. Also watch for night spraying, which a Deere speaker described as close at a February 2026 symposium⁹.

Questions to ask before adopting

• What share of our herbicide spend is non-residual post-emergence product, and what residual programme will precede the sprayer?
• Which sensitivity setting will operators use, and who checks end-of-season escapes each year?
• Does our sprayer accept the system, and what is the all-in cost per acre, fees included, against our acreage?
• Are our crops, row spacings and weed species supported and tested independently in our region?
• How will savings be measured and audited when fees are charged on unsprayed acres?