Deep learning‐based laser weed control compared to conventional herbicide application across three vegetable production systems

Abstract

BACKGROUND

Herbicides are the primary weed management method for processing vegetable growers, but challenges such as limited chemical options, herbicide resistance, crop injury risks, regulatory changes, and shifting consumer preferences are driving interest in nonchemical alternatives like laser weeding. In 2024, three research trials in New Jersey and New York evaluated the effectiveness of laser weeding using a commercial unit and comparing it with pre‐emergence‐ and postemergence‐applied herbicides on beet (Beta vulgaris L.), spinach (Spinacia oleracea L.), and pea (Pisum sativum L.).

RESULTS

Across all trials, laser weeding was as effective as or superior to S‐metolachlor, bentazon and phenmedipham herbicides applied at label rate in controlling erect annual weeds, including common lambsquarters (Chenopodium album L.) and common ragweed (Ambrosia artemisiifolia L.). However, laser weeding was less effective on purslane (Portulaca oleracea L.) and annual grasses in New York because of sequential emergence patterns and protected meristems, respectively. Compared with untreated controls, laser weeding reduced weed cover by ≥45% and density by ≥66%, resulting in ≥97% less weed biomass by the season's end. In addition, crop stunting did not exceed 1% and crop biomass increased by ≥30% when laser weeding replaced herbicide applications.

CONCLUSIONS

These findings demonstrate that multiple laser passes can control weeds without damaging crops, leading to higher yields than conventional herbicides. Further research is needed to optimize laser weeding across different environments and weed species, and to evaluate commercial units with increased laser capacity and faster processing speeds. © 2025 The Author(s). Pest Management Science published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.

In 2024, trials in New Jersey and New York evaluated the effectiveness of laser weeding with a Carbon Robotics unit compared with herbicides on beets, spinach, and peas. Results demonstrated that laser weeding was as effective or superior to herbicides in controlling weeds, reducing biomass by ≥97% and boosting crop growth by ≥30%.

1. INTRODUCTION

In 2022, more than 1 million ha of vegetables were harvested in the United States (US), generating a production value of nearly US $20 billion. ¹ Growers in the Northeastern and Mid‐Atlantic regions produced fresh and processing vegetables on more than 147 000 ha across 15 700 farms. New York and New Jersey are two of the largest vegetable‐producing states in the region with 44 360 and 17 383 ha, respectively, and a combined crop value of US $650 million.

Vegetable production has traditionally relied on tillage and cultivation before, during, and after crop planting as the primary methods of weed suppression. ² , ³ Although effective at uprooting, cutting, or burying weeds, these techniques require intensive and frequent soil disturbance, which can increase soil erosion and compaction, degrade soil structure, deplete soil organic matter, and reduce water infiltration over time. ⁴ , ⁵ , ⁶ Mechanical control of weeds growing within the crop row can also lead to crop injury. ⁷ Hand weeding faces growing challenges related to rising labor costs and a shrinking labor force, making it increasingly unsustainable, especially for small‐ and medium‐scale producers. ⁸ , ⁹ , ¹⁰ , ¹¹

Polyethylene mulch provides effective weed control and growth benefits in vegetable production but causes soil erosion, pesticide runoff, and disposal problems. ¹² , ¹³ Emerging alternatives include soil‐biodegradable mulches (BDM) and mulches that tighten to the crop growing bed when exposed to sunlight, though concerns remain about BDM degradation efficacy, soil residue accumulation, additive release, and higher costs. ¹⁴ Organic mulches improve soil quality while suppressing weeds but may introduce weed seeds. ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ Newer foam mulches, hydromulches, and agrotextiles show potential but face cost and application challenges. ²¹ , ²² Cover crops can suppress weeds through residue and allelopathic effects but see limited adoption in vegetable systems compared with row crops. ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ Importantly, mulches are impractical for many vegetable production systems, particularly processing vegetables and leafy greens, where inter‐row cultivation and mechanical harvesting operations prevent their implementation.

Chemical control options are more limited for vegetable than agronomic crops because of smaller production areas, potential crop damage, and manufacturer liability concerns; nevertheless, herbicides remain an essential tool for weed control. ²⁸ , ²⁹ , ³⁰ However, overreliance on a few active ingredients, also used in agronomic crops rotated with vegetables, has led to the development of herbicide‐resistant weed biotypes. According to Heap, ³¹ 69 cases of herbicide‐resistant species have been confirmed in vegetable crops, including 30 cases in North America. Although herbicide resistance appears less common in vegetable production than in agronomic crops, Boyd et al. ³² suggest that this may be due to underreporting, likely because limited resources are allocated to resistance monitoring in specialty crops.

High‐performance, image‐based deep learning algorithms have made real‐time weed recognition for precision control possible, performing effectively under typical agricultural field conditions. ³³ , ³⁴ , ³⁵ This technology shows significant promise for advancing sustainable vegetable production systems by enabling targeted interventions that minimize herbicide use and environmental impact. ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ Image‐based, precision weed control can reduce chemical use, labor, and soil disturbance while enabling safe and timely intervention during the early growth stages of vegetables, when they are most vulnerable to weed competition. ³⁷ , ³⁸ Actuators for targeted weed control include precision sprayers, weeding knives, and directed energy implements, like high‐powered lasers. ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ Lasers are effective tools for precision weed management because their coherent beams, where light waves are in phase both spatially and temporally, ensure intense, highly concentrated energy. ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ In early 2022, Carbon Robotics introduced a commercial laser weeder designed specifically for large‐scale vegetable growers. Commercial units are currently in operation, primarily in the western US. ⁵¹ Although Carbon Robotics is a prominent player in the laser weeding industry, the technology has a global history dating back to the early 2000s. Early patents established fundamental approaches to laser‐based weed control, ⁵² with pioneering scientific research by Heisel et al. ⁴⁵ demonstrating the efficacy of CO₂ lasers for cutting weed stems. Today, several companies worldwide offer diverse implementations of this technology. For example, WeedBot, based in Latvia and the United Kingdom, has developed high‐precision laser weeding technology that uses optical sensors to identify weeds and eliminate them with directed laser beams (WeedBot, Riga, Latvia). The field continues to evolve with recent exploration of alternative laser types such as thulium‐doped fiber lasers that may offer efficiency advantages over traditional CO₂ systems. Despite advances in precision weed management, knowledge gaps remain regarding the utility of laser weeding across diverse agricultural regions. Conventional weed control methods continue to face growing challenges, including herbicide resistance, labor shortages, and rising operational costs. Laser weeding systems present a promising alternative, offering nonchemical, automated weed control. However, their adoption is limited by high implementation costs, estimated at a minimum of $500 000 per unit. ⁵¹ Further research is needed to evaluate their practical and economic viability in different cropping systems and regions.

In 2024, field trials were conducted in collaboration with Carbon Robotics company to: (i) quantify the weed control efficacy of laser weeding across diverse vegetable cropping systems representative of Mid‐Atlantic and Northeastern US agriculture; (ii) assess potential impacts on crop development; and (iii) compare laser weeding efficacy to conventional weed management strategies.

2. MATERIALS AND METHODS

2.1. Laser weeding equipment

All laser weeding operations were performed using the Carbon Robotics Autonomous LaserWeeder™ demonstration unit (Carbon Robotics, Seattle, WA, USA) (Fig. 1). According to specifications provided by Carbon Robotics (Perez V, personal communication), the system is equipped with 12 high‐resolution, real‐time cameras to scan the ground and identify vegetation. To enhance plant detection and eliminate shadows, four high‐intensity light bars illuminate the bed tops. Discrimination between crops and weeds relies on Carbon Robotics’ proprietary algorithms, which run on Nvidia graphic‐processing units (GPU) to process high‐resolution imagery (Nvidia Corporation, Santa Clara, CA, USA). These GPUs enable real‐time analysis of plant morphological characteristics for accurate identification during field operations. If a plant is identified as a weed, it is further categorized by type (e.g., broadleaf, grass). The system then evaluates the size of the weed, classifying it as small (one‐ to three‐leaf stage), medium (three‐ to five‐leaf stage), or large (five‐ to nine‐leaf stage). The system then adjusts the laser treatment duration based on the weed type and size, with a preferential focus on weeds at the two‐leaf, or smaller, growth stage, when the meristem is easily identifiable, and the laser exposure time is minimal. Once identified, weeds are targeted and removed by eight independent modules, each equipped with a 150 W carbon dioxide (CO₂) laser that offers targeting precision within 1 mm of the meristem. The lasers are positioned no higher than 1 m above the bed top. In these trials, the area effectively treated was approximately 0.5 m in width, centered over individual crop rows. The lighting, camera, processing, and laser weeding systems are mounted on a self‐propelled, autonomous platform powered by a 74 hp diesel engine and driven by four hydraulic motors. Other technical specifications for the camera systems, processors, laser configuration, and wavelength are proprietary information per the manufacturer (Perez V, personal communication). A Carbon Robotics field engineer manually operated the system to optimize performance, aiming for more than 90% weed control with each pass. In January 2025, the unit, which is representative of the laser technology in the first‐generation commercial systems, was decommissioned; models released in 2025 offer increased processing speed and utilize diode lasers.

Figure 1.

Carbon Robotics Autonomous LaserWeeder™ demonstration unit in operation in Geneva, NY.

2.2. Field trials

Three field trials were conducted in spring and summer 2024 to evaluate the efficacy of laser weed control and its subsequent impact on vegetable productivity at the Rutgers Agricultural Research and Extension Center in Bridgeton, New Jersey (39.5188, −75.2059) and the research north farm at Cornell AgriTech, Geneva, New York (42.8788, −77.0311). Soil type at the New Jersey location was a chillum silt loam (fine‐silty, mixed, semiactive, mesic Typic Hapludults), 15% sand, 68% silts, and 17% clay, with pH values of 5.5 and organic matter of 1.7%; the trial site was planted to soybean [Glycine max (L.) Merr.] in 2023. Soil type in New York was a Honeoye loam (fine‐loamy, mixed, semiactive, mesic Glossic Hapludalfs) with 38% sand, 44% silt, 18% clay, 2.5% organic matter, and a pH of 6.3; the trial site was planted to hemp (Cannabis sativa L.) in 2023. Treatments were replicated four times and arranged in a randomized complete block design in New Jersey and in a split‐plot design (SPD) in New York, with crops assigned to the main plots and weed management strategies assigned to the subplots. Individual plot size was 13.7 m long by 1.8 m wide in New Jersey and 30.5 m long by 0.76 m wide in New York.

2.2.1. New Jersey succulent pea trial

One week prior to sowing, the field was disked to remove any emerged weeds and prepared with a cultipacker to obtain a smooth seeding bed. Fertilization with calcium nitrate (60 kg N ha⁻¹), was performed concurrently with field cultivation. ‘SV0935QF’ succulent pea (Pisum sativum L.) (Seedway, Hall, NY, USA) was seeded on 20 March at a density of 50 seeds m⁻¹ in rows spaced 76 cm apart using a two‐row vacuum seeder (Gaspardo, Campodarsego, Italy). No residual herbicide applications were applied prior to or at the time of crop planting. Treatments included: laser weeding after crop emergence when the height of emerged weeds was less than 2.5 cm; a single postemergence (POST) tank mix of bentazon (Basagran®, UPL, Cary, NC, USA) at 1120 g ai ha⁻¹ plus imazethapyr (Pursuit®, BASF, Research Triangle Park, NC, USA) at 50 g ai ha⁻¹; a single POST application of pyridate at 790 g ai ha⁻¹ (Tough®, Belchim Crop Protection, Wilmington, DE, USA); a weed‐free check that was manually hoed, weekly; and a nontreated control. Bentazon and imazethapyr are standard herbicides frequently used for POST weed control in pea, whereas pyridate safety on pea is currently evaluated across multiple US locations by the IR‐4 Project (IR‐4, 2025). ⁵³ Plots were laser‐weeded on 1 May (the first or ‘A’ pass), when the height of emerged weeds was less than 2.5 cm, and again on 13 May (the second or ‘B’ pass) following a second flush of emerged weeds. The POST herbicide treatments were applied on 13 May using a CO₂‐pressurized backpack sprayer equipped with 11 002 extended range flat‐fan nozzles (TeeJet®, Glendale Heights, IL, USA) calibrated to deliver 140 L ha⁻¹ at 179 kPa. The pea crop was at the eight‐leaf growth stage at the time of the POST herbicide applications. The timing of laser weeding operations was based on Andreasen et al., ⁴³ who showed that laser‐based weed control is most effective when weeds have fewer than four true leaves. Smaller weed sizes are also preferable for POST herbicide applications. Common lambsquarters (Chenopodium album L.) and common ragweed (Ambrosia artemisiifolia L.) dominated the weed population with mean densities of 12.8 and 3.1 plants m⁻², respectively, prior to POST herbicide application. Both species averaged the third‐leaf growth stage during laser weeding operations (Fig. 2). Other weed species, including common evening primrose (Oenothera biennis L.), oakleaf goosefoot (Chenopodium glaucum L.), mouseear cress [Arabidopsis thaliana (L.) Heynh.], spring whitlow (Draba verna L.), and hairy galinsoga (Galinsoga quadriradiata Cav.), were sporadically present. These species were no taller than 2–3 cm and did not exceed the two‐leaf stage at the time of weeding.

Figure 2.

Thermal damage in (a) common ragweed (Ambrosia artemisiifolia L.) and (b) common lambsquarters (Chenopodium album L.) following precision laser treatment with the Carbon Robotics Autonomous LaserWeeder™ in Bridgeton, NJ.

2.2.2. New Jersey spinach trial

The field for the spinach study was prepared following the same protocol as described for the pea study, with the additional step of forming a raised bed measuring 15 cm in height and 91 cm in width. ‘SV2146VB’ spinach (Spinacia oleracea L.) (Seedway) was sown on 9 April at a density of 39 seeds m ⁻¹ with three rows per bed spaced 30 cm apart using a three‐row vacuum seeder (Agricola Italiana, Massanzago‐ca’ Baglioni, Italy). Trial treatments included a single pre‐emergence (PRE) application of S‐metolachlor (Dual Magnum®, Syngenta, Greensboro, NC, USA) at 534 g ai ha⁻¹, S‐metolachlor PRE followed by laser weeding, and S‐metolachlor PRE followed by a POST application of phenmedipham (Spin‐Aid®, Bayer CropScience, Research Triangle Park, NC, USA) at 545 g ai ha⁻¹, clethodim (Shadow®, UPL, King of Prussia, PA, USA) at 120 g ai ha⁻¹, and a nonionic surfactant (Activator 90, Loveland, Greeley, CO, USA) at 0.25% v/v. The laser weeding and POST herbicide treatments were also evaluated without the PRE application of S‐metolachlor. A nontreated control was included for comparison. Laser weeding was carried out on 1 May (the first or ‘A’ pass), 13 May (the second or ‘B’ pass), and 6 June (the third or ‘C’ pass), when the height of the newly emerged weeds did not exceed 3 cm. The POST herbicide treatment was applied on 13 May, concurrent with the second pass of the laser weeder, using the same application equipment as described earlier in the pea study. The phenmedipham application was timed according to label recommendations, which specify applying the herbicide after spinach reaches the six‐true‐leaf stage. ⁵⁴ At the time of the POST application, spinach was at the four‐ to eight‐leaf stage. Except for the additional presence of large crabgrass [Digitaria sanguinalis (L.) Scop.], the density of weed species and their growth stage at the time of POST application were similar to those observed in the previously described pea study.

2.2.3. New York multiple crops trial

The field was disked before planting to eliminate emerged weeds and then cultipacked to produce a flat and even seedbed. The main plot in the SPD included three crop varieties: ‘SV2146VB’ spinach, ‘Little Marvel’ succulent pea, and ‘Ruby Queen’ garden beet (Beta vulgaris L.). Crops were sown in single rows on 12 June using a Monosem vacuum seeder (Largeasse, France), with planting densities of 20 seeds m ⁻¹ for pea and 56 seeds m ⁻¹ for beet and spinach. Four weed management strategies were tested in subplots: a nontreated control, laser weeding as needed following crop and weed emergence, a PRE application of S‐metolachlor at 720 g ai ha⁻¹, and a combination of S‐metolachlor applied PRE followed by laser weeding. S‐Metolachlor was applied on 13 June 13 using a tractor mounted sprayer equipped with 11 002 extended range flat‐fan nozzles (TeeJet®) calibrated to deliver 187 L ha⁻¹ at 206 kPa. Laser weeding operations were conducted on 21 June (the first or ‘A’ pass), 27 and 28 June (the second or ‘B’ pass), and 10 July (the third or ‘C’ pass). Common lambsquarters, common ragweed, and common purslane (Portulaca oleracea L.) were the predominant weed species at the site, with densities averaging 9.4, 7.6 and 3.2 plants m⁻² on 21 June. Mean lambsquarters and ragweed heights did not exceed 2.5–4 cm (two‐ to four‐leaf stages) at each application timing; purslane diameters did not exceed 5 cm. Less‐frequently occurring species, which included Powell amaranth (Amaranthus powellii S. Watson), smartweeds (Polygonum spp.), volunteer hemp, and annual grasses were no taller than 2–3 cm and did not exceed the two‐leaf growth stage at the time of weeding.

2.3. Data collection

In New Jersey, weed control, weed cover, weed density, and crop injury were assessed for each plot 1 and 2 weeks after each laser weeding pass in both trials. Control of the dominant species was determined at the plot level using a scale ranging from 0% (no weed control) to 100% (complete weed control). Weed cover was estimated at the plot level using a scale ranging from 0% (no weed cover) to 100% (complete weed cover). Weed counts were taken from two 0.25‐m² quadrats per plot, which were centered on the crop rows. Crop injury, which was characterized by stunting, was rated at the whole‐plot level on a scale ranging from 0% (no visible injury) to 100% (crop death). Succulent peas and spinach were harvested at commercial maturity on 10 June and 7 June, respectively. For peas, pods were collected from two 1‐m sections of the crop row, sorted into marketable and cull categories, and weighed to determine yields. The pods were then shelled, and the commercial pea yield was recorded. The dry biomass of the whole plant was also measured as an alternative metric to evaluate pea growth responses to treatments. Because of uneven spinach emergence across plots, 20 randomly selected plants per plot were harvested, and head weights were recorded. Weed biomass was estimated by harvesting weeds from two 0.25‐m² quadrats per plot at crop harvest, with dry weights recorded.

In New York, weed control, weed cover, and weed density were assessed for each plot 1 week after each laser weeding treatment. Methods followed the protocols established for the New Jersey trials. Crop injury was evaluated by rating percent emergence and stunting. Emergence was assessed using a scale ranging from 0% (no crop emergence) to 100% (crop completely emerged); crop stunting was also rated on a scale from 0% (no crop stunting) to 100% (crop death). Crop and weed aboveground biomass were harvested from four 1.5 m × 0.5 m (0.75‐m²) quadrats per plot approximately 40 days after planting, corresponding to baby leaf spinach maturity. Pea pod yields were not collected because of suboptimal conditions for flower development and fruit set; therefore, dry plant biomass was used as an alternative measure to assess pea growth responses. Beet leaf yields were harvested alongside spinach and pea biomass to provide a shared data point for cross‐crop comparisons. Beet roots were collected 30 days later to estimate fresh market yield potential. At both locations, the time spent on weed control was recorded for each laser‐treated plot.

2.4. Data analysis

The laser weeding unit employed for these trials was decommissioned by the manufacturer in January 2025, preventing equipment‐identical replication of the 2024 experiment in subsequent years. Although research protocols typically require multi‐year field studies to establish reproducibility, we suggest that our consistent results across multiple experimental sites provided sufficient validation of our findings. Data were analyzed using analysis of variance (ANOVA) with SAS v9.4 (SAS Institute, Cary, NC, USA) and the generalized linear mixed model (GLIMMIX) procedure. Weed management strategy, crops, and their interaction were treated as fixed effects, while replications were considered random factors. Percent crop injury data were arcsine square root transformed to address unequal variance before ANOVA and back‐transformed for presentation. ⁵⁵ When main effect interactions were nonsignificant, data were pooled accordingly. Weed coverage and crop injury were analyzed using a beta distribution, because data were constrained between 0 and 1 (prior to conversion to percentages for presentation). ⁵⁶ , ⁵⁷ Data describing weed density, biomass, and laser weeding time were analyzed using a gamma distribution, as values ranged from 0 to positive infinity. Means comparisons for fixed effects were performed using Tukey's honest significance test (HSD) at α = 0.05.

3. RESULTS

3.1. New Jersey pea weed control

For comparison purposes, only data collected after the second pass of the laser weeder are presented, because this pass occurred simultaneously with application of the POST herbicide treatments. Weed control, weed cover, weed density, and weed dry biomass were significantly (P ≤ 0.05) affected by treatment for all evaluation dates (Table 1). Common lambsquarters control was significantly higher in laser‐treated plots (95%) compared with POST applications of bentazon plus imazethapyr (83%) and pyridate (27%) (Table 1) at 1 week after the second laser weeder pass (WAB). By 2 WAB, control in laser‐treated plots declined to 83% because of new seedling emergence. By contrast, bentazon plus imazethapyr provided 98% control, due to imazethapyr residual soil activity. Pyridate POST provided only 23% control of common lambsquarters at 2 WAB. Common ragweed control was ≥98% at 1 and 2 WAB for both the laser weeder and bentazon plus imazethapyr treatments. Common ragweed emergence had stopped by 2 WAB, which improved the laser weeding performance for the species compared with common lambsquarters. Pyridate was the least effective treatment, providing 85% common ragweed control 1 WAB and 64% 2 WAB. Complete (100%) control was achieved for both species in the weed‐free treatment.

Table 1.

Effect of weed management strategy on common lambsquarters and common ragweed control, total weed cover, total weed density, and total weed dry biomass at harvest in succulent pea at Bridgeton, NJ

Treatment (g ai ha−1)	Weed control				Weed cover			Weed density			Weed dry weight (g m−2)
	CHEAL †		AMBEL
	1 WAB (%)	2 WAB (%)	1 WAB (%)	2 WAB (%)	1 WAB (%)	2 WAB (%)	Harvest (%)	1 WAB (plants m−2)	2 WAB (plants m−2)	Harvest (plants m−2)
Weedy check	—	—	—	—	26 a	32 a	54 a	67 a	56 a	65 a	37 a
Weed‐free control	100 † a	100 a	100 a	100 a	0 b	0 c	0 c	0 d	0 c	0 c	0 b
Laser weeding	95 a	83 b	99 a	98 a	6 b	4 c	5 c	28 c	29 b	22 b	1 b
Bentazon (1120) + imazethapyr (50)	81 b	98 a	99 a	100 a	4 b	4 c	9 c	40 bc	24 b	15 b	2 b
Pyridate (790)	27 c	23 c	85 b	64 b	11 b	15 b	24 b	61 a	57 a	52 a	24 a

AMBEL, common ragweed; CHEAL, common lambsquarters; WAB, weeks after the second laser weeder pass (B).

^† Means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

Weed cover in nontreated controls averaged 26%, 32%, and 54% at 1 WAB, 2 WAB, and harvest, respectively, while remaining below 9% in the laser‐weeded and bentazon plus imazethapyr treatments across all evaluations. Weed cover in the pyridate treatment was 11%, 15%, and 24% at these same intervals. All treatments significantly reduced weed cover compared with nontreated controls, although pyridate was less effective than other treatments at 2 WAB and at harvest. Weed density measurements supported these findings, with 52–67 plants m⁻² observed in the nontreated and pyridate‐treated plots, whereas laser weeding and bentazon plus imazethapyr reduced weed density by 40–77%. At harvest, two passes of laser weeding and the bentazon plus imazethapyr treatment reduced weed biomass by 97% and 95%, respectively, compared with the nontreated check, whereas pyridate did not result in a statistically significant biomass reduction.

3.2. New Jersey pea tolerance

Stunting was the only observed crop injury noted during the trial, with symptoms appearing after the application of POST herbicides (Table 2). Pyridate caused 6% stunting in pea plants at 1 WAB, whereas no injury was observed with bentazon plus imazethapyr. By 2 WAB, stunting increased to 15% and 3% in the pyridate and bentazon plus imazethapyr treatments, respectively. Weed competition in the nontreated weedy control resulted in 9% stunting of the pea crop at 2 WAB. By harvest, pea stunting exceeded 30% in both the nontreated weedy check and pyridate treatments, whereas it remained below 11% in response to the bentazon plus imazethapyr treatment. The laser weeder treatment and weed‐free control exhibited minimal injury (≤1%) throughout the trial.

Table 2.

Effect of weed management strategy on crop stunting, crop dry biomass at harvest and total yield for succulent pea at Bridgeton, NJ

Treatment (g ai ha−1)	Crop stunting			Pod filling (peas pod−1)	Individual pea weight (g pea−1)	Shelled pea yield		Crop fresh weight (×1000 kg ha−1)
	1 WAB (%)	2 WAB (%)	Harvest (%)			Total (×1000 kg ha−1)	Commercial (×1000 kg ha−1)
Weedy check	0 † b	9 b	30 a	5.92	0.64 cd	3.81 c	2.24 c	12.2 c
Weed‐free control	0 b	0 d	1 c	5.92	0.73 ab	5.05 a	3.03 ab	20.2 a
Laser weeding	0 b	0 d	0 c	6.05	0.81 a	5.09 a	3.13 a	19.9 a
Bentazon (1120) + imazethapyr (50)	0 b	3 c	11 b	5.99	0.69 bc	4.58 b	2.72 b	15.6 b
Pyridate (790)	6 a	15 a	35 a	5.95	0.60 d	3.83 c	2.24 c	14.2 bc

WAB, weeks after the second laser weeder pass (B).

^† Means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05). The absence of letters within a column indicates no statistically significant differences between means.

Accumulated crop biomass was 23% to 40% lower for the nontreated weedy control and POST herbicide treatments compared with the weed‐free check (Table 2). Although treatments did not significantly affect the number of peas per pod, individual pea weight decreased by 12% in the weedy controls and by 18% in the pyridate treatment, which reduced both total and commercial yields. Although approved for POST use in succulent peas, bentazon plus imazethapyr reduced plant biomass and total yield by 23% and 9%, respectively compared with the weed‐free control. Only laser‐weeded plots maintained biomass, commercial yield, and pea weight comparable with weed‐free controls, producing 15% higher yields than bentazon plus imazethapyr despite similar weed density and biomass at harvest.

3.3. New Jersey spinach weed control

Laser weeding reduced weed density by 98% at 1 week after the first pass (WAA) and 95% at 2 WAA compared with the nontreated weedy control (data not shown). When a PRE herbicide application was followed by laser weeding, weed density was 91% lower at 1 WAA and 83% lower at 2 WAA. In comparison, the PRE herbicide treatment alone reduced weed density by only 73% at both time points relative to the nontreated control (data not shown). By 2 WAA, weed cover remained below 3% in all laser‐weeded plots, whereas it reached 23% in the nontreated weedy control and 8% in the PRE herbicide treatment (data not shown).

Weed control, cover and density, and weed dry biomass following the second pass of the laser weeder were also significantly (P ≤ 0.05) affected by treatment (Table 3). Laser weeding, used alone or following a PRE application of S‐metolachlor, provided 83% to 96% control of common lambsquarters and common ragweed, respectively, at 1 WAB, and 88% to 96% control at 2 WAB. Conventional herbicide programs were significantly less effective than the laser weeder treatments for managing both species. PRE applications of S‐metolachlor provided 54% and 25% control of common lambsquarters and common ragweed, respectively, at 1 WAB; at 2 WAB, control was 40% and 9%. A single POST herbicide application containing phenmedipham controlled common lambsquarters and common ragweed by 50% and 48%, respectively at 1 WAB and 46% and 25% at 2 WAB. The PRE application of S‐metolachlor followed by a POST application of phenmedipham plus clethodim provided 55% and 71% control of common lambsquarters at 1 WAB and 2 WAB, respectively, but no more than 34% control of common ragweed.

Table 3.

Effect of weed management strategy on common lambsquarters, common ragweed and large crabgrass control, total weed cover, total weed density and total weed dry biomass at harvest in spinach at Bridgeton, NJ

Treatment	Weed control						Weed cover			Weed density (plants m−2)	Weed dry weight (g m−2)
	CHEAL		AMBEL		DIGSA
	1 WAB (%)	2 WAB (%)	1 WAB (%)	2 WAB (%)	1 WAB (%)	2 WAB (%)	1 WAB (%)	2 WAB (%)	Harvest (%)
Weedy check	—	—	—	—	—	—	30 a	56 a	92 a	156 a	203 a
Weed‐free check	100 ‡ a	100 a	100	100 a	100 a	100 a	0 c	0 c	0 d	0 c	0 c
PRE †	54 bc	40 d	25 c	9 b	63 b	44 b	8 bc	22 b	55 ab	87 b	85 b
POST	50 b	46 cd	48 b	25 b	94 a	89 a	13 b	20 b	47 b	80 b	59 b
PRE fb POST	45 c	71 bc	28 c	34 b	82 a	88 a	9 bc	17 c	31 b	57 b	40 b
PRE fb laser weeding	100 a	96 a	83 ab	89 a	98 a	91 a	3 c	5 c	7 d	2 c	0 c
Laser weeding	92 ab	88 ab	96 a	92 a	100 a	90 a	2 c	8 c	14 d	2 c	1 c

AMBEL, common ragweed; CHEAL, common lambsquarters; DIGSA, large crabgrass; fb, followed by; POST, postemergence herbicide; PRE, pre‐emergence herbicide; WAB, weeks after the second laser weeder pass (B).

^† Pre‐emergence herbicide consisted in S‐metolachlor applied at 534 g ha⁻¹ at planting; postemergence herbicide consisted in phenmedipham at 547 g ha⁻¹ plus clethodim at 118 g ha⁻¹ applied 34 days after planting. A nonionic surfactant at 0.25% v/v was added to the postemergence herbicide mix.

^‡ Means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

Large crabgrass was controlled ≥98% at 1 WAB and ≥90% at 2 WAB by all treatments that included laser weeding (Table 3). These results were comparable with treatments that included clethodim applied POST, which provided ≥82% and ≥88% control at 1 WAB and 2 WAB, respectively. By contrast, a single PRE application of S‐metolachlor provided only 63% and 44% large crabgrass control at 1 WAB and 2 WAB, respectively, likely because of dissipation of its residual activity following daily irrigation of the spinach crop. Complete weed control (100%) was achieved for common lambsquarters, common ragweed, and large crabgrass in the hand‐weeded checks.

Weed cover in the nontreated control was 30%, 56% and 92% at 1 WAB, 2 WAB, and harvest, respectively. Programs that relied solely on PRE and/or POST applications of conventional herbicides reduced weed cover by 20% to 46%, relative to the nontreated control, for the same observation dates. Whether used alone or in combination with a PRE application of S‐metolachlor, laser weeding was the most effective strategy for suppressing weed cover, with average reductions of 27%, 50%, and 82% at 1 WAB, 2 WAB, and harvest, respectively, compared with the nontreated control. Weed cover was eliminated in the weed‐free control.

Weed density and dry biomass collected at harvest confirmed the results of the final visual evaluation of weed cover. Laser weeding, whether used alone or in combination with a PRE application of S‐metolachlor, reduced weed density and dry biomass by 99% compared with the nontreated control (Table 3). By contrast, PRE and/or POST applications of conventional herbicides reduced weed density by only 44% to 64% and weed dry biomass by 58% to 80% relative to the nontreated weedy check These results demonstrate that multiple passes of the laser weeder effectively controlled both dense weed patches and weeds of varying sizes, as indicated by the significant reductions in weed numbers and biomass at crop harvest compared to the nontreated weedy check. No weed biomass was recovered from the hand‐weeded control.

3.4. New Jersey spinach tolerance

Crop injury did not exceed 2% following the first and second passes of the laser weeder, with no significant differences between treatments (data not shown). Minimal chlorosis, averaging 3%, was observed at crop harvest only in response to the POST application of phenmedipham plus clethodim (Table 4). By contrast, crop stunting was more severe and variable across treatments. Spinach head size was reduced, as determined by a visual assessment, 10% to 12% in the nontreated weedy check, the solo PRE application of S‐metolachlor, and the PRE application. Spinach stunting was minimal (1%) with laser weeding alone but increased to 8% when combined with a PRE application of S‐metolachlor. Stunting was likely because of inadequate weed control in the PRE and POST herbicide treatments and the phytotoxic effects of S‐metolachlor, as observed in the PRE followed by laser weeding treatment.

Table 4.

Effect of weed management strategy on crop stunting and crop yield at harvest for spinach at Bridgeton, NJ

Treatment	Crop injury		Total yield (×1000 kg ha−1)
	Chlorosis (%)	Stunting (%)
Weedy check †	0 ‡ b	12 a	14.3 cd
Weed‐free check	0 b	1 c	22.3 ab
PRE	0 b	10 ab	15.0 cd
POST	2 a	7 b	16.0 cd
PRE fb POST	3 a	12 a	11.7 d
PRE fb laser weeding	0 b	8 b	17.1 bc
Laser weeding	0 b	1 c	23.7 a

fb, followed by; POST, postemergence herbicide; PRE, pre‐emergence herbicide.

^† Pre‐emergence herbicide consisted in S‐metolachlor applied at 534 g ha⁻¹ at planting; postemergence herbicide consisted in phenmedipham at 547 g ha⁻¹ plus clethodim at 118 g ha⁻¹ applied 34 days after planting. A nonionic surfactant at 0.25% v/v was added to the postemergence herbicide mix.

^‡ Means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

Yield data confirmed the negative impact of S‐metolachlor on spinach growth. Head weight was reduced by 34% across all S‐metolachlor treatments compared with the weed‐free control. For example, spinach head weight was 28% lower when a PRE application of S‐metolachlor was followed by laser weeding compared with laser weeding alone, despite both treatments achieving similar levels of weed control. Laser weeding alone effectively controlled weeds without phytotoxicity, resulting in the highest head weight (71 g head⁻¹), comparable with the weed‐free control, and 65% greater than the untreated weedy control. Treatments with conventional herbicides did not significantly differ from the untreated weedy control (43 g head⁻¹) in head weight.

3.5. New York weed control

Weed control, weed density, weed cover, and weed biomass were significantly (P ≤ 0.05) affected by weed control treatment (Tables 5 and 6). Averaged across crops, a PRE application of S‐metolachlor, used alone, controlled common lambsquarters by 88%, 74%, and 77% at 1 WAA, 1 WAB, and 1 week after the third laser weeder pass (WAC), respectively. Laser weeding was less effective than S‐metolachlor for suppressing common lambsquarters at 1 WAA, providing only 54% control. At 1 WAB and 1 WAC, laser weeding achieved 81% and 95% control of common lambsquarters, respectively, which was comparable with or better than the control achieved with the S‐metolachlor PRE standard. The most effective treatment for managing common lambsquarters was the combination of S‐metolachlor PRE followed by laser weeding, which provided ≥95% control through 1 WAC.

Table 5.

Effect of weed management strategy on common lambsquarters and common ragweed control at Geneva, NY

Treatment	CHEAL			AMBEL			POROL
	1 WAA (%)	1 WAB (%)	1 WAC (%)	1 WAA (%)	1 WAB (%)	1 WAC (%)	1 WAA (%)	1 WAB (%)	1 WAC (%)
PRE †	88 ‡ a	74 b	77 b	69 b	57 b	50 b	100 a	99 a	99 a
Laser weeding	54 b	81 ab	95 a	69 b	82 a	95 a	90 b	80 b	70 b
PRE fb laser weeding	97 a	95 a	99 a	87 a	93 a	98 a	100 a	99 a	99 a

AMBEL, common ragweed; CHEAL, common lambsquarters; POROL, common purslane; fb, followed by; PRE, pre‐emergence; WAA, week after the first laser weeder pass (A); WAB, week after the second laser weeder pass (B); WAC, week after the third laser weeder pass (C).

^† Pre‐emergence herbicide consisted in S‐metolachlor applied at 720 g ha⁻¹ at planting.

^‡ Data were averaged over crops; means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

Table 6.

Effect of weed management strategy on total weed density and total weed cover at Geneva, NY

Treatments	Weed cover			Weed density
	1 WAA (%)	1 WAB (%)	1 WAC (%)	1 WAA (plants m−2)	1WAB (plants m−2)	1 WAC (plants m−2)
Weedy check	18 ‡ a	56 a	80 a	158 a	175 a	553 a
PRE †	2 bc	6 ab	18 ab	17 c	27 b	64 b
Laser weeding	3 b	2 b	5 b	60 b	47 b	52 b
PRE fb laser weeding	1 c	1 c	4 c	5 c	3 c	4 c

fb, followed by; PRE, pre‐emergence; WAA, week after the first laser weeder pass (A); WAB, week after the second laser weeder pass (B);WAC, week after the third laser weeder pass (C).

^† Pre‐emergence herbicide application consisted in S‐metolachlor applied at 720 g ha⁻¹ at planting.

^‡ Data were averaged over crops (except for 1 WAC); means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

A similar trend was observed for common ragweed, with the solo PRE application of S‐metolachlor providing 69%, 57%, and 50% control at 1 WAA, 1 WAB, and 1 WAC, respectively. Although there was no difference between the S‐metolachlor PRE and laser weeding treatments at 1 WAA, laser weeding proved more effective than the herbicide treatment at controlling common ragweed at both 1 WAB (82%) and 1 WAC (95%). As with common lambsquarters, the most effective strategy for managing common ragweed throughout the study was the combination of S‐metolachlor PRE followed by laser weeding, with control estimates ranging from 87% to 98%.

With respect to common purslane, PRE applications of S‐metolachlor, with or without laser weeding, provided almost complete control (>99%) at 1 WAA, 1 WAB, and 1 WAC. However, unlike upright annuals, purslane control with the laser weeder alone declined over time, likely because of continuous seedling emergence and survival of larger plants. At 1 WAA, 1 WAB, and 1 WAC, purslane control was 90%, 80% and 70%, respectively. Grasses, primarily foxtails (Setaria spp.) and barnyardgrass [Echinochloa crus‐galli (L.) P. Beauv.], were rarely observed until the final laser weeding pass. Laser weeding provided only 75% control of monocots at 1 WAC, compared with 95% control when S‐metolachlor was applied (data not shown).

Averaged over crops, weed cover was greatest in the nontreated weedy control, increasing from 18% at 1 WAA to 80% at 1 WAC. A single PRE application of S‐metolachlor suppressed weeds relative to the nontreated check, but did not fully prevent weed emergence and establishment, with estimated cover reaching 2%, 6%, and 18% at 1 WAA, 1 WAB, and 1 WAC, respectively. Laser weeding used alone or following a PRE application of S‐metolachlor reduced weed cover to ≤5% for all observation timings. Weed density was highest in the nontreated check ranging from 158 plants m⁻² at 1 WAA to 553 plants m⁻² at 1 WAC. Compared with the weedy control, the solo application of S‐metolachlor reduced weed densities by 85% to 89% for all evaluation timings. Relative to the nontreated check, weed density was reduced by 62%, 75%, and 91% at 1 WAA, 1 WAB, and 1 WAC, respectively, by laser weeding. Weed numbers were reduced ≥97% compared with the weedy control when a PRE application of S‐metolachlor was followed by laser weeding, regardless of observation date. Weed management strategy also had a significant effect (P ≤ 0.05) on weed dry biomass at 35 days after treatment (DAT) (data not shown). Averaged across crops, weed biomass in the nontreated control was 251 g m⁻², which was reduced by 80% (50 g m⁻²) with a single PRE application of S‐metolachlor. Laser weeding alone and laser weeding following S‐metolachlor PRE lowered weed biomass by 97% (to 8 g m⁻²) and 99% (to 2 g m⁻²), respectively.

3.6. New York multiple crop tolerance

Crop emergence was significantly (P ≤ 0.05) affected by the interaction between crop and weed management strategy at 1 WAA and by weed management strategy, alone, at 1 WAB and 1 WAC (Table 7). At 1 WAA, mean pea emergence was 84%, with no differences observed among treatments. For beet and spinach, the mean crop emergence was greatest (56% to 78%) at 1 WAA in the nontreated check and the solo laser weeding plots. Crop emergence was significantly lower (15% to 24%) for both small‐seeded crops in both treatments where S‐metolachlor was applied PRE. Averaged across all crops, treatment differences remained noticeable at both 1 WAB and 1 WAC; mean crop emergence ranged from 77% to 92% for the nontreated control and the laser weeding plots compared with 65% to 80% in the S‐metolachlor followed or not by laser weeding treatments.

Table 7.

Effect of weed management strategy on crop emergence at Geneva, NY

Treatment	1 WAA			1 WAB (%)	1 WAC (%)
	Beet (%)	Spinach (%)	Pea (%)
Weedy check	57 ‡ ab	56 ab	88	77 ab	83 ab
PRE †	15 c	21 b	83	64 b	71 b
Laser weeding	68 a	78 a	83	91 a	92 a
PRE fb laser weeding	24 bc	21 b	83	65 b	80 b

fb, followed by; PRE, pre‐emergence; WAA, week after the first laser weeder pass; WAB, week after the second laser weeder pass; WAC, week after the third laser weeder pass.

^† Pre‐emergence herbicide consisted in S‐metolachlor applied at 720 g ha⁻¹ at planting.

^‡ Data were averaged over crops (except for emergence 1 WAA); means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

The interaction between crop and weed management strategy was significant (P ≤ 0.05) at all evaluation timings for crop stunting. Therefore, the data were analyzed separately by crop (Table 8). For each crop, ≤3% stunting was observed when the laser weeder was used alone at 1 WAA, 1 WAB, and 1 WAC. In the nontreated check, no stunting was observed for beet and spinach at 1 WAA. However, stunting increased to 7% and 15%, for beet and spinach, respectively, at 1 WAC because of weed competition. Averaged across both treatments that included S‐metolachlor, beet stunting was 38%, 33%, and 21% at 1 WAA, 1 WAB, and 1 WAC, respectively, while spinach stunting was 19%, 53%, and 45% at the same time points.

Table 8.

Effect of weed management strategy on crop stunting and crop dry biomass at Geneva, NY

Treatment	Crop stunting									Crop dry weight (g m−2)
	Beet			Spinach			Pea
	1 WAA (%)	1 WAB (%)	1 WAC (%)	1 WAA (%)	1 WAB (%)	1 WAC (%)	1 WAA (%)	1 WAB (%)	1 WAC (%)
Weedy check	0 ‡ b	3 bc	7 a	0 b	4 bc	15 b	0	0 b	0	67 a
PRE †	39 a	41 a	21 a	24 a	63 a	49 a	4	7 a	1	48 b
Laser weeding	3 b	0 c	0 b	0 b	0 c	0 c	3	0 b	0	75 a
PRE fb laser weeding	37 a	24 ab	20 a	14 a	42 ab	41 ab	2	0 b	0	54 b

fb, followed by; PRE, pre‐emergence; WAA, week after the first laser weeder pass (A); WAB, week after the second laser weeder pass (B); WAC, week after the third laser weeder pass (C).

^† Pre‐emergence herbicide consisted in S‐metolachlor applied at 720 g ha⁻¹ at planting.

^‡ Means followed by the same letter within in a column are not significantly different based on Tukey's HSD test (α = 0.05).

Crop dry biomass harvested approximately 40 days after planting was significantly (P ≤ 0.05) affected by treatment, with the greatest amount of biomass accumulating in the solo laser weeding plots (75 g m⁻²). Despite weed competition, the nontreated control generated greater (67 g m⁻²) crop biomass than S‐metolachlor PRE (48 g m⁻²) and S‐metolachlor PRE followed by laser weeding (54 g m⁻²). Fresh market beet root yield was significantly improved (P ≤ 0.05), relative to the weedy check (2.5 t ha⁻¹), by all weed control treatments (data not shown). The use of S‐metolachlor PRE alone or followed by laser weeding increased beet root weights by 2.7‐fold and 2.1‐fold, respectively. Laser weeding used alone increased beet root weights 9.7‐fold relative to the weedy check.

3.7. Laser weeder processing time

The laser weeder unit was operated by a trained company technician who aimed to achieve a targeting and treatment accuracy of 90%. In the New Jersey pea trial, no significant differences in weeding time were observed between laser passes, averaging 20 s m⁻¹ of row. In the spinach trial, weeding time was significantly (P ≤ 0.05) influenced by the interaction between pass and treatment (Fig. 3). During the first pass, it took the laser weeder 33 s m⁻¹ of row. Conversely, when S‐metolachlor was applied pre‐emergence, the laser weeder took 26 s m⁻¹ of row. For the second and third passes, the time required for weeding did not differ between treatments; it took between 12 and 19 s m⁻¹ of row to complete a pass. In the New York multiple crops trial, weeding time was significantly (P ≤ 0.05) affected by the interaction between pass and treatment (Fig. 4). Weeding speed increased with successive passes; weeding times were slower when the laser weeder was used alone than when S‐metolachlor was applied PRE followed by laser weeding. During the first pass, it took the laser weeder 22 s m⁻¹ of row. When S‐metolachlor was applied PRE, the laser weeder took 13 s to treat the same distance during the first pass. On the second pass, it took 20 s m⁻¹ of row with the laser weeder, whereas the S‐metolachlor PRE followed by laser weeding treatment required 10 s. During the third pass, the laser weeder took 16 s m⁻¹ of row, whereas the S‐metolachlor PRE followed by laser weeding treatment took 8 s. Comparatively, it took between 0.5 and 1 s m⁻¹ of row to apply the PRE and POST herbicides.

Figure 3.

Weeding duration in spinach at Bridgeton, NJ, as affected by operation date and treatment using a Carbon Robotics laser weeding research unit. For comparison purposes, the duration of pre‐emergence (PRE) or postemergence herbicides application was 0.5 to 1 s m⁻¹ of row.

Figure 4.

Weeding duration across pea, spinach and beet crops in Geneva, NY, as affected by operation date and treatment using a Carbon Robotics laser weeding research unit. For comparison purposes, the duration of pre‐emergence (PRE) or postemergence herbicides application was 0.5 to 1 s m⁻¹ of row.

4. DISCUSSION

This novel study, conducted under field conditions in New Jersey and New York using a research unit provided by Carbon Robotics, demonstrated that laser weeding was as effective as, or more effective than, standard herbicide applications in controlling weeds across three vegetable crops. By the end of the season, weed biomass in laser‐weeded plots was reduced by at least 97% compared with nontreated weedy controls across all trials. Performance was, however, impacted by plant size and species’ biological characteristics.

Multiple laboratory‐based studies have demonstrated the impact of weed size on laser weeding effectiveness. ⁴⁴ , ⁴⁹ , ⁵⁸ , ⁵⁹ , ⁶⁰ Wöltjen et al. ⁴⁹ found that higher energy doses were needed to control tobacco (Nicotiana tabacum L.) at the two‐ and three‐leaf growth stages compared with the cotyledon stage, with energy requirements varying by laser type. Using a CO₂ laser, ≤0.1 J mm⁻² was required for 90% control at the cotyledon stage, whereas plants at the two‐ and three‐leaf stages required approximately 0.5 to 2 J mm⁻². By contrast, diode lasers required more energy, ranging from approximately 1 to 13 J mm⁻², likely because of their lower absorption rate compared with CO₂ lasers. Coleman et al., ⁵⁸ using a diode laser, reported that applications of 76.4 J mm⁻² completely controlled annual ryegrass (Lolium rigidum Gaudin) at the three‐ and seven‐leaf stages, whereas the same dose at mid‐tillering resulted in only 13% control (though biomass was reduced by 60%). Applications at late‐tillering had minimal effect. Recent research by Andreasen et al. ⁴³ using a 50 W fiber laser demonstrated that laser irradiation was most effective against both grass and broadleaf weed species at the cotyledon stage, regardless of rate (0.4 to 12.7 J mm⁻²). Although two‐leaf stage applications could also be effective with greater energy inputs, most species regrew following treatments at the four‐leaf stage. In our New Jersey and New York operations, laser weeding was timed to treat dominant weed species when most plants were at the optimal growth stage for control (up to two true leaves). However, some common lambsquarters and common ragweed that exceeded the recommended stages were able to successfully regrow.

Laser weeding success can also be influenced by weed architectural traits. ⁴⁴ , ⁴⁹ Wöltjen et al. ⁴⁹ reported that tobacco was more sensitive to laser weeding than barnyardgrass. Because grass growing points are located at or below the soil surface, laser weeding may sometimes cause only superficial leaf damage. To address this challenge, Andreassen et al. ⁴³ reported adjusting laser angle to better direct the beam towards basal tissue. Conversely, Heisel et al. ⁴⁵ found that perennial ryegrass (Lolium perenne L.) seedling stems were easier to cut than common lambsquarters and wild mustard (Sinapis arvensis L.) at all growth stages and heights, likely because of the relatively thinner grass leaves. In our trials, the two most common grass species exhibited varying responses to laser weeding (Besançon TE and Sosnoskie LM, personal observation). Large crabgrass control was ≥90% using the laser weeder, while foxtail control was 75%. Morphological differences at the one‐ to two‐leaf stage likely contributed to this differential response. Large crabgrass seedlings exhibited relatively flat cotyledons or first leaves, leaving the growing point more exposed, while foxtail seedlings displayed a more erect growth habit with leaves enveloping the meristem, potentially providing greater protection against laser exposure. Andreassen et al. ⁴³ also found that dicot weed responses to laser weeding varied based on leaf morphology and the ability to resprout from lateral meristems. Species with small, thin leaves, like corn speedwell (Veronica arvensis L.), were more sensitive to laser treatments than curly dock (Rumex crispus L.) seedlings with larger leaves. Purple deadnettle (Lamium purpureum L.) and birdseye speedwell (Veronica persica Poir.) regrew from lateral meristems at the base of the cotyledons, which escaped laser damage because the epicotyl grew quickly, moving the apical meristem away from the cotyledons as the first permanent leaves developed. ⁴³ By contrast, Mwitta et al. ⁶⁰ did not observe significant differences in plant mortality between Palmer amaranth and smallflower morningglory [Jaquemontia tamnifolia (L.) Griseb.] seedlings in a greenhouse study. In our trials, differential success was observed across species. At both locations, common lambsquarters and common ragweed treated at cotyledon to two‐leaf stage were effectively controlled. However, in New York, common purslane was less effectively managed, likely because of its continuous emergence throughout the trial, prostrate and spreading growth habit, and ability to root from stems and regrow from lateral buds. Plants that escape initial treatments become more difficult to control in subsequent passes, because larger weeds with multiple meristems require more laser pulses or a wider beam, increasing energy demand and slowing operations.

An important factor contributing to the effectiveness of the laser weeder in this study was precise treatment timing. Operations were conducted when target weed species were small and when seedlings were dry and not subjected to environmental stressors. Excessive plant moisture and adverse weather conditions have been suggested as potential factors that could reduce laser weeding efficacy. ⁴⁴ Various operational variables influence success, including laser type, wavelength, beam width, laser angle, and exposure time. ⁴⁵ , ⁴⁸ , ⁵⁸ , ⁵⁹ The technology's performance also depends on the camera's ability to capture high‐resolution imagery and the image‐based deep learning algorithm that must accurately distinguish weeds from crops and identify meristematic regions for directed laser applications. ⁴⁹ , ⁵⁰ , ⁶¹ In this study, the laser weeder effectively targeted the growing points of both dicotyledonous and monocotyledonous weeds despite substantial morphological differences between species. The system's precision is evidenced by the absence of necrotic damage in the three evaluated crops, indicating minimal or nonexistent laser thermal injury to crops. If the laser unintentionally contacts the crop, injury extent may vary depending on growth habit, age, size, and other biological characteristics. Andreassen et al. ⁴³ suggested that cereal plants, particularly larger ones, are likely to recover following accidental leaf contact with a laser beam because the meristem is not directly exposed. Although sugar beet can be injured at early developmental stages, the relatively thick leaves of mature plants may reduce sensitivity to accidental laser exposure. ⁴³ Laser weeding offers a promising opportunity to reduce off‐target herbicide impacts that can affect nontarget organisms, water quality, and human health. Although Andreassen et al. ⁴² reported that laser treatments could potentially harm insects present on targeted plants, the technology's precise application significantly reduces environmental exposure compared with broadcast herbicide applications. Future refinements in image‐based weed recognition capabilities enabled by deep learning may further enhance selectivity, making laser weeding an increasingly valuable tool for sustainable agricultural systems.

A key observation from this study was the reduced crop seedling emergence, increased crop injury, and decreased crop biomass and yield following pre‐emergence application of S‐metolachlor, particularly in spinach and beet, compared with the control or solo laser weeding treatments. S‐Metolachlor is a selective chloroacetamide herbicide that inhibits long‐chain fatty acid synthesis and is widely used for pre‐emergence weed control in various crops such as corn (Zea mays L.), soybean (Glycine max L. Merr.), potato (Solanum tuberosum L.), table and sugar beet, sunflower (Helianthus annuus L.), and tomato (Solanum lycopersicum L.). It is labeled for use on spinach in New Jersey and both red beet and spinach in New York under a 24(c) Special Local Need registration, allowing applications up to 720 g ha⁻¹, except on coarse‐textured soils with less than 1.5% organic matter. ⁶² , ⁶³ Previous research has generally found S‐metolachlor to be safe for pre‐emergence use in fresh market and processing spinach at rates up to 1060 and 730 g ha⁻¹, respectively, as well as in red beet at 1200 g ha⁻¹. ⁶⁴ , ⁶⁵ , ⁶⁶ However, variable responses have been reported for both species. Umeda ⁶⁷ reported up to 27% injury and 15% stand reduction in spinach treated with 560 g ha⁻¹ followed by furrow irrigation. Robinson and McNaughton ⁶⁸ noted that PRE applications of S‐metolachlor at 1200, 1600, and 3200 g ha⁻¹ reduced red beet aboveground dry weight and marketable root yields. In sugar beet, increasing the S‐metolachlor rate from 540 to 4320 kg ha⁻¹ resulted in greater emergence losses and crop stunting, particularly when cumulative rainfall exceeded 40 mm within 14 days of application. ⁶⁹ These effects were more pronounced in sandy loam soils or soils with organic matter content below 3.5%. Both beet and spinach exhibited significant injury in response to S‐metolachlor under our study conditions. In New York, a 14‐mm downpour 10 days after seeding led to soil crusting, inhibiting seedling emergence and limiting young plants’ ability to metabolize absorbed S‐metolachlor. Metabolism of this herbicide in sugar beet primarily occurs in shoots, so the lack of emerged shoots may have contributed to increased seedling mortality because of herbicide phytotoxicity. Crop tolerance can also be variety‐dependent; less‐tolerant sugar beet varieties require up to twice as long to metabolize S‐metolachlor compared with more‐tolerant cultivars. ⁷⁰ In New York, delayed emergence in ‘Ruby Queen’ beets in the S‐metolachlor treatments led to noticeable crop stunting that persisted for several weeks. By contrast, beets in the nontreated check and solo laser‐weeded plots emerged ahead of the crusting event, establishing dense stands with a significant advantage in early growth and development. This response may warrant future research trials to formally describe beet cultivar responses to labeled soil applied herbicides, much like disease resistance screenings. ⁷¹ Spinach was similarly affected by crusting in the New York trial. In New Jersey, some crusting was also observed in the spinach study, likely because of irrigation gun use, which produced larger droplets than sprinkler systems. The combination of high irrigation frequency and large droplet impact may have exacerbated crust formation, intensifying S‐metolachlor phytotoxicity and ultimately leading to significant yield reductions. The application of POST herbicides also significantly reduced crop biomass accumulation and yield, as observed in the New Jersey succulent pea study. Al‐Khatib et al. ⁷² , ⁷³ reported no herbicide‐induced injury in early‐planted peas following POST bentazon applications at 0.56 to 1.68 kg ha⁻¹. However, the same treatment applied to later‐planted peas caused visible injury and stunting that dissipated by 20 DAT, possibly because of higher temperatures increasing herbicide absorption or altering its metabolism. Similarly, Sikkema et al. ⁷⁴ found that postemergence imazethapyr applications at rates of 45 g ha⁻¹ or higher caused visible injury (≤5%) and delayed pea maturity, with yield reductions under moisture stress conditions. At Bridgeton, NJ, precipitation in May 2024 was 49% below the 30‐year average, which may have exacerbated imazethapyr‐induced injury despite biweekly overhead irrigation.

Pyridate has been linked to crop injury and yield reduction in several vegetable crops, including pepper (Capsicum annuum L.), ⁷⁵ artichoke (Cynara cardunculus L.), ⁷⁶ and sweetpotato [Ipomoea batatas (L.) Lam]. ⁷⁷ However, some vegetable crops, especially those in the Brassicaceae family, have shown acceptable tolerance. Studies have reported transient leaf chlorosis in cabbage (Brassica oleracea L.) and broccoli (Brassica oleracea var. botrytis L.) following pyridate applications, but no significant yield reductions at doses ≤1 kg ha⁻¹. ⁷² , ⁷⁸ , ⁷⁹ Limited data exist on pyridate phytotoxicity in leguminous crops. Lemerle and Hinkley ⁸⁰ reported significant pea yield reductions at 1.35 kg ha⁻¹, whereas Kousta et al. ⁸¹ observed minimal injury (<2%) and no yield impact at 0.9 kg ha⁻¹. Miller and Hopen ⁸² reported that pyridate tolerance in Brassicaceae species depended on plant age and environmental conditions, with greater injury occurring in spring because of smaller seedlings and thinner cuticles. In the New Jersey pea trial, the absence of visible injury following pyridate application may be attributed to the moderate application temperature (19 °C) and advanced seedling stage (eight‐leaf). The observed crop stunting and yield reduction were likely caused by inadequate weed control rather than direct herbicide phytotoxicity.

Similar to POST herbicides, early intervention is crucial for maximizing laser weeding efficacy. Implementation challenges include higher energy requirements for larger weeds with multiple meristems, potentially narrower application windows compared to herbicides due to rapid weed growth, and operational limitations during the Northeastern US's increasingly frequent heavy rainfall events. ⁸³ The dense weed infestations common in the Northeastern US (>100 plants m⁻²) require reduced driving speeds. Our trials showed the laser weeder needed 4–17 min to treat a 30.5‐m row, compared with just 15–30 s for conventional herbicide applications, a speed limitation that restricts treatable acreage during critical weed control windows. Integrating pre‐emergence herbicides improved efficiency by reducing laser weeding time up to 51% in some passes, although this must be balanced with crop safety concerns. Weather or soil conditions delaying field access may require supplemental manual weeding for plants that grow beyond the optimal treatment stage. A combined approach, using herbicides or mechanical weeding between rows with laser weeding within rows, would enhance overall efficiency while reducing crop injury potential. Onion growers in New York who have purchased commercial laser weeding units are taking this approach to meet their weed control needs in a timely manner (Sosnoskie LM, personal observation). The equipment tested in this study is well‐suited for bare ground or processing vegetables but may not be appropriate for certain regionally grown plasticulture staked crops where a nonchemical solution for controlling weeds in the planting hole is still needed. The efficacy of laser weeding in nonstaked plasticulture crops, such as certain cucurbits and strawberries, also warrants further exploration. In February 2025, Carbon Robotics launched a next‐generation diode‐based laser weeder designed for greater modularity and adaptability across diverse cropping systems. ⁵¹ The system's capability to tilt individual laser modules facilitates effective weed control in plasticulture settings, where weeds may not be directly in the drive path. The laser weeding unit used in these trials was decommissioned by the manufacturer in January 2025, preventing replication of the 2024 study with identical equipment. Although multi‐year field studies traditionally establish reproducibility, the consistent results obtained across multiple sites provide sufficient evidence to validate our findings. The robust performance demonstrated across diverse conditions effectively compensates for the equipment limitation and supports the reliability of our conclusions.

This study's purpose was not economic data collection, because we used a research demonstration unit rather than the advanced commercial laser weeders. No comparative studies between commercial laser weeding and conventional herbicide application have been conducted, to date, in the eastern US. However, the Western Growers Center for Innovation and Technology case study (2024) ⁸⁴ provides relevant economic insights from two California Salinas Valley vegetable operations. Their findings showed that laser weeding technology reduced costs from $900/acre (manual labor) to approximately $269/acre on a large‐scale operation. Within 8 months, growers achieved 10%–15% average yield improvements (up to 50% in some cases) through increased recoverable product and reduced crop damage. In addition, the technology diminished the weed seedbank, enhancing long‐term operational efficiency (Western Growers, 2024).

5. CONCLUSION

This study highlights the potential of laser weeding as an effective and sustainable alternative to traditional herbicides for controlling weeds in vegetable crops. Laser weed control technology incorporating deep learning recognition systems offers an innovative solution that combines efficiency with sustainability to address environmental and economic challenges in modern vegetable production. The field trials conducted in New Jersey and New York demonstrated that laser weeding effectively reduced weed biomass by ≥97% when applied to weeds at their optimal growth stages, providing control comparable with or superior to conventional herbicide applications. The precision of the laser weeding system in targeting weed meristems, along with minimal crop injury, further supports its potential as a viable, environmentally friendly weed management tool.

However, the timing of laser applications remains a critical factor for success. Efficacy declines significantly as weeds grow beyond their optimal treatment stages, which underscores the importance of precision in timing and conditions. In addition, challenges such as adverse weather conditions, weed size, and rapid regrowth of certain species need to be considered for optimal performance. Integrating laser weeding with other weed management strategies, such as pre‐emergence herbicides or inter‐row cultivation, could further enhance its effectiveness and operational efficiency.

Although laser weeding offers distinct advantages over chemical herbicide use, such as eliminating the risk of phytotoxicity and reducing chemical inputs, its feasibility may be impacted by environmental variables and dense weed infestations. Future research should focus on refining laser weeding technology, addressing operational constraints, and exploring complementary approaches to optimize its performance across diverse agricultural settings.

CONFLICT OF INTEREST

The authors state that there are no conflicts of interest with any parties.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.