Robots and shocks: emerging non-herbicide weed control options for vegetable and arable cropping

ABSTRACT

For decades, herbicides have provided easy-to-use, cost-effective weed management, but alternatives are desired. Consumer preference for chemical-free food, awareness of environmental impacts, regulation increasingly restricting agrichemical use, and increasing prevalence of herbicide resistance are forcing changes to weed management strategies. New Zealand farming must remain sustainable and profitable while responding to changes in its overseas markets, among which are increasing demands for regeneratively grown, safe, high-quality produce. Current reliance on herbicides should be reduced, with more emphasis on preventative management by cultural means, and weed suppression by alternative technologies. The emergence of agritechnologies incorporating automation, machine vision and artificial intelligence, and development of new techniques for weed destruction, offer alternatives that minimise or avoid the requirement for herbicides, avoid soil disturbance and can work effectively in high crop or crop-residue conditions. We have identified electric weeding as a feasible alternative and pulsed electric microshocks as a very low-energy option requiring a fraction of the energy of any other system. Pulsed microshocks enable an integrated weed management system for vegetable and arable crop production combining cultural controls and inexpensive pre-planting treatments with automated application of chemical-free in-crop weed control. Open-source software enables community development of autonomous deployment for niche crops.

KEYWORDS:

• Weed control
• integrated weed management
• regenerative agriculture
• alternative weeding
• automation
• electric weeding
• machine vision
• artificial intelligence

Introduction

Weed control is essential for reliable yields of quality produce, as weeds are the major cause of yield loss in key crops such as maize, wheat, potato, rice, soybean (Oerke 2006), and maize (James et al. 2000). They impact product quality and make farm operations more difficult (McErlich and Boydston 2014). For decades, chemical herbicides have provided cost-effective weed management. Their ease of use has enabled chemical technology to become the dominant weed control technique, with an array of choices for broad-spectrum or selective control of weeds, even from within standing crops. Weeds of any size can be killed, from newly germinating seedlings to large trees. Intra-row weeds can be effectively removed with minimal if any crop damage. However, the dominance of herbicides is challenged.

A desire for healthier food, soil, and ecosystems, consumer preference to avoid the use of chemicals, and increasing prevalence of herbicide resistance, are driving efforts to find alternatives. At the same time, social challenges to current production systems stemming from an awareness of carbon balance impacts to climate change, and increased interest in conservation and regenerative agriculture are compounded by labour shortages and increased energy costs. The cost of weed control in New Zealand seed and grain crops – arable including maize - was estimated in 2017 to be c. $18.4 million per annum or about $94 ha⁻¹ (Saunders et al. 2017). For specialty and vegetable crops the costs can be even higher, especially if hand-labour is required (Fennimore and Cutulle 2019). Political and market requirements impacting weed control include more stringent regulation, restrictions on the use of herbicides, restrictions on burning crop residues, the introduction of carbon accounting, and the promotion of alternative farming systems including regenerative agriculture (Ministry for Primary Industries 2022). An integrated weed management system must evolve that recognises such requirements and suits production from farms that place more emphasis on regenerative practices, energy efficiency, reduced use of chemicals and maintenance of soil cover.

Recent years have seen many weed control approaches revisited. As a result of research, integrated weed management is recommended, although there is a suggestion that on-farm uptake of this is low (Moss 2008). Accelerating development of automation and vision systems coupled with artificial intelligence is allowing old and new weed control technologies to be used in hitherto impossible ways. In 2019, we reviewed and conducted preliminary tests (unpublished) with a commercial hot foam/hot water weeding system, and a custom-manufactured air-blast grit system based on the work of Forcella (2009b, 2017). The hot foam/hot water system was rejected on the basis of the extremely large volume of water requiring heating and transport, the energy required (De Cauwer et al. 2015; De Cauwer et al. 2016), the impracticality of the system, as demonstrated, to be effectively deployed in a cropping system, and its inability to achieve selective weeding. The abrasion system provided was tested using a range of grit types, including macadamia shell, oyster shell, coffee grounds and fine sand. None proved totally effective. Fine sharp sand caused most damage, but generally just pock-marked the treated leaf tissue rather than destroying it. Our experience contrasted to the US research that has demonstrated good control in many trials over many years (Forcella 2009a; Forcella et al. 2011; Forcella 2019). However, we discounted the approach as it requires heavy equipment and materials, and air compression demands high energy.

We also investigated electrical weeding technologies. Literature searches identified work demonstrating that very low energy electrical weeding is possible (Mizuno and Hori 1988; Mizuno et al. 1990; Blasco et al. 2002) and a chance meeting with a group of engineering students at our university offered the possibility to develop and test such a system. Our first results showed the system can be successful in greenhouse trials (Bloomer et al. 2022) and indicated that field deployment is viable as part of a manual or robotic system.

This paper reviews vegetable and arable crop weed management in a New Zealand context, reviews the range of new and not-so-new technologies available and considers how weed management can adapt to the new context in which vegetable and arable production operates. One concept, pulsed electric microshocks, is an example of a novel approach for selective weeding within a multimethod, integrated weed management system. This ultralow-energy weeding system can be integrated with automation technologies and artificial intelligence to create an autonomous ultralow-energy, selective, non-chemical weeding system.

New Zealand farming environment in 2023

Consumer preferences for organic or chemical-free food (Magnusson et al. 2003; Devcich et al. 2007; Forbes et al. 2009; Rutledge 2009; Crinnion 2010; Wooliscroft et al. 2014; Hoek et al. 2017; Koch et al. 2017; Galati et al. 2019), public concern about residues in water (Bajwa et al. 2015; Hageman et al. 2019) including public perceptions in Twitter ‘heavily skewed toward negative sentiments’ (Jun et al. 2023), awareness of potential impacts of chemical treatments on soil microflora (Helander et al. 2018) and fauna (Salminen et al. 1996) and restrictions or bans on products including glyphosate (Beckie et al. 2020; Alcántara-de la Cruz et al. 2021), further promote herbicide alternatives. Labour shortages (Kitchin 2021; Giovannetti 2022; Ministry for Business Innovation and Employment 2023; Statistics New Zealand 2023) force a reduction in manual weed management and drive demand for technology to replace it. Spikes in the costs of diesel, electricity, fertilisers, chemicals and compliance are forcing farmers to ensure all costs are tightly managed (Murphy 2022).

Regenerative agriculture, which places emphasis on sequestering carbon and reducing the use of chemicals (Grelet and Lang 2021; Ministry for Primary Industries 2022; Schlesinger 2022; Thomas 2022; Danley 2023; McCain Foods 2023), has re-emerged as a trend supported by governments and industry (Thomas 2022). In New Zealand, it is seen as a production system focused on reducing the impacts of food production on the environment, a way to shift to low-emissions and a sustainable economy, and is part of the Government and primary sector’s ‘Fit for a Better World’ roadmap (Ministry for Primary Industries 2022). A bundle of principles rather than a set of prescribed rules, and variously described (Grelet and Lang 2021; Ministry for Primary Industries 2022; Thomas 2022; McCain Foods 2023), regenerative agriculture advocates minimising soil disturbance, keeping living roots in the soil, enhancing biodiversity, reducing the use and impact of agrochemicals and artificial fertilisers, and integrating animals into the system. Regenerative agriculture seeks to keep the soil covered with living crops or crop residues covering the soil as much as possible (Thomas 2022; McCain Foods 2023). However, in a study of USA corn production, LaCanne and Lundgren (2018) found regenerative agriculture systems had 29% lower grain production but 78% higher profitability than conventional production systems. The movement towards conservation agriculture, systems that are typically reliant on herbicides (Melander et al. 2013; Reiser et al. 2019), is also accompanied by greater quantities of ground cover in the form of crops, cover crops or crop residues, and possibly living mulches within main crops (Westbrook et al. 2022). This makes mechanical weed control by tillage difficult, so non-contact methods are preferable (Bauer et al. 2020) and equipment used should operate effectively in high residue conditions.

Herbicide resistance, ‘the inherited ability of an individual plant to survive a herbicide application that would kill a normal population of the same species’ (Peltzer 2019), is increasing globally (Wilson et al. 2011; Owen et al. 2014; Powles 2014; Walsh and Powles 2014; Lamichhane et al. 2017; Moss 2017; Heap 2022). Herbicide resistance has been increasingly reported in New Zealand (Rahman et al. 1983; Buddenhagen et al. 2019; Ghanizadeh and Harrington 2021). Increased identification in New Zealand may in part, be due to proactive investigation through surveys (Harrington et al. 2016; Buddenhagen et al. 2019; Ghanizadeh et al. 2019; Ngow et al. 2020; Buddenhagen et al. 2021; Ghanizadeh et al. 2022a). Knowledge of the nature of resistance in New Zealand has been enhanced through molecular and genetic studies (Ghanizadeh 2015; Ghanizadeh and Harrington 2017; Ghanizadeh et al. 2019; New Zealand Plant Protection Society 2019; Ghanizadeh et al. 2020; Ghanizadeh et al. 2022b). Herbicide resistance removes a key management tool. While all potential weeds are unlikely to be resistant to all potential herbicides, the loss of one or more critical control options for serious weeds has significant implications for crop managers. Effective, affordable alternatives are required.

Robotics and weed management

Smart agricultural robots are now being released to market, and more are at the early development stages (Petrovic 2020; Pardell 2021; Petrovic 2022; Rispens 2022; Papadopoulos 2023). Oliveira et al. (2021) conducted a review of more than 60 agricultural robots of which 81% were in the research stage, and 22% were designed for weeding. Merfield (2016b) noted that many autonomous systems are ‘self-guided vehicles carrying weeding tools’ using mechanical weeding technology little different to that used for centuries. However, ongoing advances in mechatronics, vehicle electrification and information technologies such as artificial intelligence are combining in powerful ways to offer robotic systems with automated weed recognition and precision application (Fennimore et al. 2016; Reiser et al. 2019; Ruigrok et al. 2020; Oliveira et al. 2021) and new technologies such as laser weeders are being employed (Andreasen et al. 2022; Asscheman 2023; Silverberg 2023). Automation enables site-specific weed management; a system of weed identification, a weed control model, and the precision application of a control method (Christensen et al. 2009). With respect to autonomous weeding in sugar beet, eight evaluation criteria and target values were proposed for a concept selection matrix: weed control efficiency (90% target), ability to target weeds (including under crop leaves), resolution of action (control border <5 mm), work rate (10 weeds sec⁻¹ per row or 0.5 m s⁻¹ operational speed), auxiliary rate (labour to assist the tool ∼18 min per 2 h), energy (∼2 kJ m⁻¹ row), applicability to autonomous vehicles (< 150 kg, < 1 m width and height), and material costs (€11,250) (Nørremark et al. 2006). A target of 2 kJ m⁻¹ of row for inter-row weeding of crops in 50 cm rows equates to 40 MJ ha⁻¹, which is about twice the estimates of Coleman et al. (2019) for electric spot-weeding of 19 MJ ha⁻¹ assuming five plants m⁻².

Numerous agricultural robots and weed recognition algorithms are in development and are being brought together (Bloomer 2017a; Lottes et al. 2018; Sabzi et al. 2018; Ruigrok et al. 2020; Anken and Latsch 2022), and some are already in the market (Bakker et al. 2010; Frasconi et al. 2014; Chen et al. 2015; Bawden et al. 2017; Reiser et al. 2019; Carrington 2021; Verdant Robotics 2022). Technology to automatically locate crop rows and enable inter-row weeding was developed in the 1990s (Hague and Tillett 2001; Tillett et al. 2002). Further developments that recognise plants and enable intra-row weeding by machine followed (Blasco et al. 2002; Tillett et al. 2002; Slaughter et al. 2008; Xiong et al. 2017; Reiser et al. 2019). Some systems can now discriminate between common crop plants and weeds and make selective decisions (Bloomer 2017b; Ahmad et al. 2018; Gerhards 2018; Wang et al. 2019; Wu et al. 2020; Li et al. 2021; Coleman et al. 2022). Robotic systems deploy a range of plant destruction methods, including cultivation (Tillett et al. 2002; Reiser et al. 2019), crushing (Akerman 2015), spot-spraying (Petrovic 2020; Ecorobotix 2023; SwarmFarm Robotics 2023), electric shocks (Blasco et al. 2002; Rootwave 2019; Malewar 2021; Baxter 2022), laser (Heisel et al. 2001; Xiong et al. 2017; Andreasen et al. 2021; Baxter 2022) and high-intensity light (Pardell 2021; Koerhuis 2022). Robots such as University of Sydney’s ‘Ladybird’ and ‘Rippa’ (Bogue 2016) and Ecorobotix ‘AVO’ (Petrovic 2020) operate almost 24 h a day entirely on batteries with solar power. The robotic arms can carry electrodes instead of spray nozzles. A similar prototype research robot demonstrated in 2001 (Blasco et al. 2002) applied 90 J shocks to each seedling treated, equating to 4.5 MJ ha⁻¹. It reached the target of one weed s⁻¹ including recognition, movement and treatment, a rate of > 3,600 weeds h⁻¹, suggesting a work rate of 13.9 h ha⁻¹.

Agricultural technology in New Zealand has historically focused on increasing production and profitability across the New Zealand supply chain (Ministry for Business Innovation and Employment 2020), with notable examples including refrigerated shipping (Stringleman and Peden 2015), electric fencing (Peden 2008) and grass genetics (Galbreath 2023). Realisation that the technologies themselves are export products and services has led to the government investing in the sector and encouraging its growth (Agritech New Zealand 2020; Ministry for Business Innovation and Employment 2020; Invest Auckland 2023). While some express disappointment at the rate of agritechnology development and adoption, there is also evidence that New Zealand farmers rapidly adopt those technologies that add value (Agritech New Zealand 2020; Invest Auckland 2023). Between 2000 and 2019, the availability of technology enabled agriculture to increase labour productivity in New Zealand at a rate 27% higher than the total industry average (Ministry for Business Innovation and Employment 2023). Adoption is moderated by the technology’s cost, complexity, and convenience as well as the end-user’s capacity and capability to integrate the technology into the existing farming system (Bloomer and Posthuma 2020). A simple change in method, ideally by swapping one practice for another, can be easier to implement. A requirement to change a whole system is more difficult and less likely to occur. Both Carbon Robotics (Carbon Robotics 2023) and Ecorobotix (Anken and Latsch 2022) have launched tractor-mounted versions of what were initially autonomous-robot-carried weeding systems.

Evolution of weed management

Weed control thinking constantly changes. The publication of ‘Silent Spring’ in 1962 (Carson 2002) stimulated awareness of the undesirable effects of chemicals (McErlich and Boydston 2014). Recommendations to develop alternatives to over-dependence on herbicides can be influenced by economics, public concern or government regulation (Burnside 1993). By the 1990s, weed science was moving more to integrated sustainable weed management systems (Zimdahl 1995). Integrated weed management (IWM) blends a range of control methodologies and technologies together. Negative perceptions of pesticide use and residues continue (Koch et al. 2017), with consumers often believing that naturalness, including ‘less chemicals’ in food, is more healthy and environmentally friendly (Hoek et al. 2017). In New Zealand’s global markets, especially Europe, action plans and EU mandates are mandating a reduction in pesticide use and promoting non-chemical methods (Melander et al. 2013; Ministry for Primary Industries 2022).

Cultural weed control includes crop rotation, stale seedbeds and avoiding seed set. Crop rotation can simplify the application of a range of weed control measures, offers different treatment windows, and diversifies weed selection pressures, reducing the build-up of species better suited to survive any particular treatment method (Lamichhane et al. 2017). The use of fallow or stale-seedbed techniques removes any weeds that germinate, providing a competition-free start for a crop. The first Europeans to arrive in New Zealand observed that Māori gardens were weed-free. It is postulated high levels of care and long fallow periods discouraged native species, and the cooler temperatures helped eliminate imported tropical weeds (Leach 2005). However, long fallow periods that replace crop production necessitate more land for the same overall yield, are associated with increased nitrogen leaching (Francis 1995) and reduced soil organic matter which can further reduce crop yields in the long term (Oldfield et al. 2022). Regardless of the control system, the benefits of ensuring a critical weed-free period to protect yield are well documented (Welsh et al. 1999; Keller et al. 2014; Knezevic and Datta 2017; Annu et al. 2023; Kumari et al. 2023) as is the need to understand the lethal or effective dose of any treatment in a given situation (Ascard 1995). Problem weeds that germinate and successfully establish require management.

Awareness that weed population dynamics are highly driven by agronomic practices has seen a shift from total weed destruction to rational suppression (Froud-Williams 2017). Indeed, some suggest the notion of a weed is merely a value judgement based on ethical and social constructs rather than a scientific one. In a post-herbicide era with integrated weed management, we might recognise crop plants, weeds and other non-crop plants that do not cause harm ‘either immediately or in the longer term’ (Merfield 2022). A multifaceted systems-approach to management of herbicide resistant weeds such as annual ryegrass is increasingly being promoted, including reducing the weed seed bank by seed catching, delayed sowing, and pre-emergent and post-emergent herbicides (Matthews et al. 1996). The objective is to avoid resistant gene spread using ‘all cultural, mechanical and herbicidal options’ within multiyear management plans (Norsworthy et al. 2012). Automation technologies now allow IWM to move from ‘broadcast applications’ of different techniques to one where each plant in a field can be categorised and given custom treatments based on biology (Young et al. 2017) including herbicide resistance. For example, robotic mechanical weed control might be combined with pre- and post-emergent herbicides (Saile et al. 2022).

Methods of weed destruction

Chemical weed control is relatively modern in the context of human crop production. Prior to the development of modern herbicides in the 1940s, cultural and physical weed control technologies were the only options available (Froud-Williams 2017). The efficiency and efficacy of herbicides rapidly led to their adoption as the primary control technique, becoming increasingly dominant from about the 1960s (McErlich and Boydston 2014). Not currently permitted in New Zealand, genetically introduced herbicide resistance in crops in countries such as the USA has increased reliance on a narrow range of herbicides in those countries using such crops, notably glyphosate, depleting farmland biodiversity and increasing the rate of herbicide resistance development (Lamichhane et al. 2017). A wide range of herbicide types are available for weed control, including selective and non-selective products, contact herbicides, some with translocated and/or residual action, that may be applied pre-sowing, pre-emergence, or post-emergence of the crop (Ross 1995). The herbicide selected will depend on the species present and the growth stage of plants treated, the climatic or soil conditions and other factors. Further classification of herbicides is based on their mode of action, ‘the overall manner in which the herbicide affects a plant at the tissue or cellular level’ (Christensen et al. 2009). The wide variety and potential for selectivity makes herbicides a powerful management tool. But while herbicides have been a proven and economically viable weed control tool, the effectiveness and public acceptance of herbicides are being challenged (Datta and Knezevic 2013). The amount of herbicide applied can be reduced using automated systems with machine vision and artificial intelligence to apply the best chemical at the optimal dose only to individual weeds (Christensen et al. 2009). However, a desire for healthier food (Wooliscroft et al. 2014), soil and ecosystems, a preference to avoid the use of chemicals, and the increasing prevalence of herbicide resistance, are driving efforts to find alternatives (Peruzzi et al. 2017; Grelet and Lang 2021).

Non-chemical weed management strategies have been well documented (Parish 1990; Nørremark et al. 2006; Melander et al. 2017; Peruzzi et al. 2017; Jabran and Chauhan 2018; Merfield 2018). They may be included in a multifaceted weed control programme that still includes herbicides. A wide range of tools has been developed since agriculture began 10,000 years ago, and variations continue to appear. Physical weed control methods that can be applied post-emergence rely on separation of shoot from root, uprooting and subsequent desiccation, burial, and/or above-ground tissue rupture (Nørremark et al. 2006). Cultivation by ploughing, discing, harrows, tines, or hoes to separate, uproot or bury weeds also damages soil structure and may have a high labour and energy cost (Kaufman and Schaffner 1980; Laguë and Khelifi 2001; Ascard et al. 2007; Balzhaeuser et al. 2012; Coleman et al. 2019). The disturbance of soil may disperse and bury seed, induce dormancy and promote additional weed germination (Vleeshouwers and Kropff 2000; Tørresen et al. 2017). Non-soil engaging methods rely on the destruction of above-ground parts by mowing (William 2007; An et al. 2020), abrasion (Forcella 2009a; Forcella et al. 2011; Perez-Ruiz et al. 2018; Wortman et al. 2018; Forcella 2019), crushing (Akerman 2015) or seed destruction at harvest (Korres et al. 2019). Removing above-ground parts may not kill all weeds but can reduce competition and prevent seeding.

Thermal controls include the use of flame, hot air, steam, hot water, radiation, and electrothermal equipment. Microwaves (Whatley et al. 1973; Hoschle 1984; Sartorato et al. 2006; Brodie et al. 2007; Kacan et al. 2018), ultraviolet light (Khalilov and Akhmedov 1992; Andreasen et al. 1999; Knezevic et al. 2016), laser pyrolysis (Heisel et al. 2001; Griepentrog et al. 2006) and high-intensity light (Johnson et al. 1989; Rakhmatulin and Andreasen 2020; Koerhuis 2022) are alternative thermal technologies. Thermal weeding technologies can control weeds without mechanical contact and can operate in the presence of high crop or mulch residues. Flame and high-energy electric or laser systems have a high fire risk, especially in dry organic residue environments. For autonomous systems with no human observer in the field to respond, this is particularly important.

Electric weeding

As a discipline, electric weeding has a long history, with the first patents in the 1890s (Sharp 1893; Scheible 1895), and it has subsequently been the subject of ongoing development and patent applications (Opp and Opp 1952; Pluenneke and Dykes 1975; Dykes 1977; Carr 1994; Diprose 2016; Rona et al. 2019). Development in the 1970s and 1980s (Dykes 1977; Diprose and Benson 1984) competed with the advent of cost-effective translocated herbicides such as glyphosate and ‘weed wiper’ technology (Diprose et al. 1985). With increasing objections to glyphosate and increasing herbicide resistance, electrical weeding is again the subject of research and development (Lati et al. 2021; Lysakov et al. 2021; Bloomer et al. 2022; Lehnhoff et al. 2022). Many technology firms are active in the field, some reviving the 1980s systems, others moving to smaller robotic-mounted systems (Vigneault and Benoit 2001; McCool et al. 2018; Rootwave 2019; Schneider 2020). The commercial electrical systems available today (Kaufman and Schaffner 1980; Moretti 2021) have high energy demands and are suited only to certain weed/crop scenarios. Electric weeding plant selectivity has been achieved by physical separation, notably treating taller weeds in low-growing crops, such as bolters in sugar beet (Vigneault et al. 1990).

Understanding of electro-impulse weeding is lacking in many areas, including requirements and mechanisms of plant destruction, the modes and safety of electrical equipment and optimum forms of electrodes (Judaev and Brenina 2008; Korres et al. 2019).

Relative to continuous alternating current (AC), Judaev and Brenina (2008) reported weeding with high voltage (DC) impulses requires less energy, less-bulky equipment, and increases work safety. They sought to determine ‘electric energy lethal doses’ that caused irreversible tissue damage in a range of weeds and noted plants at the end of flowering/beginning of fruit set require more energy than at other development phases. Using 18 kHz AC, Rootwave claims effective weed control using 50 MJ ha⁻¹, stating that the very high-frequency electricity is safer than the usual 50 Hz frequencies or DC systems (Claver 2022).

There are two main electric weeding technology categories: continuous-contact high-energy systems and pulsed-discharge low-energy systems. Diprose et al. (1978) and Diprose and Benson (1984) investigated electrothermal weed control using continuous contact alternating high-voltage electrodes and suggested that through the plant’s resistance, electrical energy is converted to heat and the cell membranes are disrupted by rapid heating and volatilisation. In field trials with large beets, destruction was achieved with 100–200 kJ per plant applied energy. Effectiveness varied according to plant size, species, age, and soil moisture status. Extensive or large in-ground root systems may not be damaged if the electric current earths before passing through the tissue (Diprose et al. 1980). Vigneault and Benoit (2001) noted the rhizomes of couch (Elytrigia repens) will survive several treatments with continuous electric current. High-voltage, short-pulse, low-energy weed control is possible. Electric weeding systems that use low-energy pulsed electric shocks for selective weed management have been researched and prototypes developed (Mizuno et al. 1990; Blasco et al. 2002; Yudaev et al. 2019) but not commercialised. A 5 kV pulsed system targeting individual weeds is described by Bennett (2019), but online images suggest it may be continuous contact and the reported tractor power take-off shaft energy source implies high energy is employed. An alternative is to use high-voltage, short-duration pulsed systems, which require much less energy. Slesarev et al. (1970) published research on controlling weeds with microsecond 25 kV electric pulses. Three days after treatment, 3-4 cm tall fathen (Chenopodium album L.) had completely stopped respiring, transpiring, and photosynthesising. Other broadleaf plants treated also died, and the researchers reported that perennial sow thistle (Sonchus arvensis L.) roots were killed to a depth of 23 cm. Mizuno et al. (1990) report Russian techniques using 30–50 kV DC discharges or repeated 30–80 kV pulses of over 100 J. For safety and energy efficiency, they recommended reducing the discharge energy by pulsing charges.

Mizuno et al. (1990) developed apparatus for laboratory use and showed small plants (40–60 mm height, 1–3 mm stem diameter) could be destroyed by one spark discharge of 0.14 J energy and very large plants (800–1,200 mm height, 10–15 mm stem diameter) with repeated pulses totalling only 2 J. A low-energy spot-weeder incorporating a number of safety features was reported by (Mizuno et al. 1993) to kill Poa annua L. in golf courses. Powered by a 12 V battery, the 3 kV 200 W rated system used high-frequency discharge. Blasco et al. (2002) describe a robotic weeder applying 15 kV at 30 mA for about 200 ms ( =  90 J) and reported successful trials. They stated electrical discharge control removes the need for weed species discrimination but provided no evidence. They reported that lettuce crop plants with more than ten leaves were unaffected, with only directly affected leaves showing any damage. Harvey et al. (2019) reviewed electric weeding and suggested that an exponential increase in required energy as plants get larger makes small plant management desirable. Their unpublished report described a prototype electric system able to supply very short high-voltage pulsed discharges. Applying 6.5 J or 20 J to both broadleaf weeds and grasses (∼ 6 tillers, leaf length < 10 cm), they observed no immediate signs of impact but recorded plant deaths several days later. While effective in laboratory testing, the system was not further developed.

Our own recent studies applying pulsed electric microshocks showed small broadleaf weeds, and to a lesser extent grasses can be controlled with very small energy requirements (Bloomer et al. 2022). Laboratory trial results showed 5 J was sufficient to kill or severely limit the growth of many seedlings up to 15 cm height. This is as little as 1% of the energy of, and more effective than, ultralow energy treatments reported in recent research (Lati et al. 2021; Lehnhoff et al. 2022). To control herbicide-resistant weeds at five plants m⁻², the required energy would be about 0.25 MJ ha⁻¹ plus transport and actuation energy for weed destruction, as compared to an optimum target of about 20–40 MJ ha⁻¹ including transport suggested by Nørremark et al. (2006). Our system uses pulsed DC electricity which is safer than AC because the ‘let go of parts gripped’ is less difficult and the threshold of ventricular fibrillation is considerably higher for shock durations longer than the period of the cardiac cycle (Standards Australia/Standards New Zealand 2022). In reviewing the safety of conducted electrical weapons such as TASER®, Panescu et al. (2017) noted that relevant international standards that specify safety requirements for electrical medical devices and electrical fences ‘give very relevant guidance’. However, like our pulsed electric shock weeder, conducted electrical weapons use much higher frequency pulses, so an alternative is needed. Our research apparatus has pauses between pulses so the capacitors can fully recharge. We typically have < 20 pulses s⁻¹ at < 0.2 J pulse⁻¹, so the energy is below the 5 J limit. A commercial device should seek to increase pulse frequency for work-rate efficiency and may increase pulse energy, so the safety factor must be addressed.

While our studies were mostly undertaken in a greenhouse, the field application of pulsed electric microshocks could be practical as a manual option, or with advances in automation, robotics and image analysis, would be a viable precision agritech opportunity as described by Blasco et al. (2002). The energy demand is potentially highly competitive with any existing or proposed weed control option, but knowledge of the correct dose and treatment point is very limited. Experimental devices were built in Soviet countries (Yudaev et al. 2019) and Japan (Mizuno et al. 1990) in the 1970s, but no evidence of commercialisation has been found.

Combining a targeted low-energy system with the automation and vision technologies now available promises an energy and cost-efficient method of non-herbicide weed management, particularly important with the increasing emergence of herbicide resistance. Such systems can be modular, with small units combinable to cover wide swaths as required. A system killing weeds with 2 J electric ‘doses’ Mizuno et al. (1990) could use perhaps 4 MJ ha⁻¹ at 100 weed seedlings m⁻², equivalent to the lowest energy requirement of any system reported by Coleman et al. (2019) and an order of magnitude lower than their estimates for site-specific electric weeding at five plants m⁻². Energy demand would be several orders of magnitude lower than thermal weeding techniques.

Open-source software and hardware designs such as Open Weed Locator and open image-libraries empower community development of intelligent systems (Coleman et al. 2022). This is important for vegetable and arable producers who grow niche crops, usually missed by proprietary commercial products.

Energy demands of weeding systems

The energy requirements of different weeding methods have been variously determined and reported (Kaufman and Schaffner 1982; Barber and Lucock 2006; Barber 2010; Coleman et al. 2019). Comparisons of total energy requirements for broadcast weed control suggest values for light mechanical techniques (e.g. flexible tines, basket weeders) of 4–17 MJ ha⁻¹. Mowing which ranges from 30–285 MJ ha⁻¹, is about an order of magnitude higher, and ploughing which ranges from 614–768 MJ ha⁻¹ is higher again (Coleman et al. 2019). Herbicide control is variously reported as 15 MJ ha⁻¹ (Coleman et al. 2019), 38–115 MJ ha⁻¹ (Audsley 2000), and 127 MJ ha⁻¹ (Alluvione et al. 2011). However Helsel and Pimentel (2007) note that herbicides have a very high energy demand in manufacture which adds considerably to the overall energy demand. They provide examples ranging from 9 MJ ha⁻¹ for chlorsulfuron, 567 MJ ha⁻¹ for glyphosate, to 880 MJ ha⁻¹ for propanil at recommended application rates. Systems involving heat from flame, air, steam, or radiation typically have very high energy demands in a range of 1,000–4,000 MJ ha⁻¹. De Cauwer et al. (2015) tested hot water weed control in bagged plants and reported applied energy intensities as high as 39,000 MJ ha⁻¹ without achieving full control. Microwave systems were reported to use between 10,000 and 75,000 MJ ha⁻¹ (Coleman et al. 2019) as a broadcast application.

Highly targeted laser weeding of very small plants at the cotyledon stage may require 50–125 J plant⁻¹ (2.5–6 MJ ha⁻¹), although much higher energy requirements are reported depending on wavelength and plant size (Wöltjen et al. 2008; Kaierle et al. 2013; Andreasen et al. 2022). Coleman et al. (2021) laser treated ryegrass (Lolium rigidum Gaudin) and found 93% of three-leaf plants were controlled at 75 J plant⁻¹ with 300 J plant⁻¹ sufficient to control plants with up to seven leaves. However, the electro-optical efficiency of CO₂ lasers is only about 10% (Wöltjen et al. 2008) so the input energy is ten times the dose delivered. This suggests energy demands of about 25–150 MJ ha⁻¹ for small plants at five plants m⁻² or 100–600 MJ ha⁻¹ applying a broadcast treatment with 200,000 plants ha⁻¹. Weeding systems that involve weed and crop recognition and autonomous control and application also have a high energy cost for processing and actuation. System specific details are not available, although the Carbon Robotics system uses 21 NVidia GPUs (Ward 2023), 12 high resolution cameras, 9 LED lightbars and controls for 30 lasers (Carbon Robotics 2023). The system is carried and powered by a large tractor so the total energy may be in the vicinity of 900 MJ ha⁻¹. This corresponds to about 26 L diesel ha⁻¹, or 15.5 L hr⁻¹ which is reasonable.

Electrothermal weed control is much more efficient than flame, air, or steam, which have very low heat transfer efficiencies (Merfield 2016a). Even so, the energy requirements reported for electrothermal weed control have a very wide range, with values of 0.04 to > 200 kJ plant⁻¹, depending on species and age (Vigneault et al. 1990). Treating individual plants using a scenario of five plants m⁻², a reasonable assumption for weed populations surviving other control attempts (Kaufman and Schaffner 1982; Vigneault et al. 1990), this equates to anywhere between 2 and 10,000 MJ ha⁻¹ for the direct energy cost, to which must be added the transport of the weeding equipment around the field. A newly developed system using very high-frequency AC was reported to give effective broadcast control using 50–100 MJ ha⁻¹ (No-Till Farmer 2022). Coleman et al. (2019) report continuous contact electrocution of two-leaf-stage broadleaf weeds at five plants m⁻² requires 19 MJ ha⁻¹ and spark electrocution 14.5 MJ ha⁻¹, values at the lower end of their calculations. Estimates for grasses are not given. Our own studies applying pulsed microshocks showed small broadleaf weeds, and to a lesser extent grasses, can be controlled with very small energy requirements of about 5 J per weed (Bloomer et al. 2022) which equates to about 0.25 MJ ha⁻¹ nett of transport and actuation energy. By comparison, Blasco et al. (2002) reported using 4.5 MJ ha⁻¹ nett of transport and actuation. A solar powered system requires virtually no extra external energy, so our system could operate at about 1% of the 20–40 MJ ha⁻¹ target for autonomous weeding proposed by Nørremark et al. (2006).

Relative energy requirements of some weeding systems are summarised in Table 1. Where conversion from diesel to MJ equivalents was required, diesel consumer energy of 38.4 MJ L⁻¹ was used rather than the primary energy value of 46.3 MJ L⁻¹ reported by Barber and Stenning (2022).

Table 1. Estimated energy requirement ranges of selected selective and broadacre weeding methods.

Weeding Method	Low Energy Estimate MJ ha^−1	High Energy Estimate MJ ha^−1
Selective (Site Specific)
Pulsed microshocks^1	0.25
Pulsed electric^2	4.5
Herbicides – site specific^3	15
Electrocution – site specific^3	15	19
Hoeing – site specific^3	17
Laser pyrolysis – site specific^3	15	249
Broadacre (Whole field)
Flextine harrow^3	4	6
Spoon weeder^3	10	11
Herbicide^4	38	115
Herbicide including chemical manufacture^3	46	712
Herbicide including chemical manufacture^5	47	1,007
Laser pyrolysis^6	100	900
Electrocution^7	32	77
Power harrow^8	230	384
Hot water^3	1,500	7,600
Hot water^9	3,840	19,150
Microwave^10, 11	9,600 ^10	75,000 ^11

¹ Bloomer et al. (2022)

² Blasco et al. (2002)

³ Coleman et al. (2019)

⁴ Kaufman and Schaffner (1982); Audsley (2000)

⁵ Calculated from Audsley (2000), Helsel and Pimentel (2007) and Alluvione et al. (2011)

⁶ Estimated from Wöltjen et al. (2008), Coleman et al. (2021) and Carbon Robotics (2023)

⁷ Kaufman and Schaffner (1982)

⁸ Calculated from Barber (2010)

⁹ Estimated from Martelloni et al. (2021)

¹⁰ Brodie (2012)

¹¹ Calculated from Wayland et al. (1975), Helsel and Pimentel (2007) and Alluvione et al. (2011)

Where to from here?

Cultural weed control, including crop rotation, stale seedbeds and avoiding seed set can form the foundation of an integrated weed management system. Crop rotation can simplify the application of a range of weed control measures, offers different treatment windows, and diversifies weed selection pressures, reducing the build-up of species better suited to survive any particular treatment method. Within a multi-year, multifaceted systems approach, the overall operations mix for effective, sustainable weed suppression can allow for occasional higher-cost systems, particularly if they effectively manage the seed bank and enable other cheaper methods to be continued. If problem weeds do germinate and successfully establish, they must be managed. High weed densities are best managed using methods such as stale seedbeds, perhaps using broadcast herbicides or light cultivation. Dealing with lower weed populations, especially herbicide-resistant escape weeds, those that emerge late or that emerge through cover mulches, and populations in sensitive crops, may be more efficiently managed using automated systems and precise application.

Preference to avoid applying agrichemicals to food and the increasing prevalence of herbicide resistance can remove herbicides as tools. Precision guidance allows for inter-row cultivation in bare-soil situations but can stimulate more weed germination. Conservation or regenerative vegetable and arable systems with an emphasis on retaining soil cover require weeding equipment that operates effectively in high residue conditions. Mechanical weed control is difficult, and hand-labour is expensive and hard to obtain. To manage costs and reduce climate impacts, low-energy systems are preferred. As a component of integrated weed management, pulsed electric microshocks as we have researched meet these criteria with a fraction of the energy required by any comparable system. Deployed as an automated weed identification and control system, our method offers an ultralow-energy, non-contact option well suited to higher residue, non-herbicide vegetable or arable production systems such as organics, conservation agriculture, or regenerative agriculture. The system can be deployed by fully automated, solar-powered, field-robots or, to cover wide swaths rapidly, as modular units on booms carried by conventional tractors. Appropriate dose rates, delivery specifications, and real-time assessment of effective treatment of weeds need further investigation, and best application within an integrated weed management system requires refinement. Different species are not equally susceptible to microshocks, so a shift in weed species spectrum could be expected if this was the only control technique used.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Ministry for Business Innovation and Employment: [grant no. C10X1806].