Artificial Intelligence in Weed Management

It is encouraging to recognize that the use of artificial intelligence has shown significant global potential in assisting the agricultural industry, particularly in the field of weed management (Amend et al. Reference Amend, Brandt, Di Marco, Dipper, Gässler, Höferlin, Gohlke, Kesenheimer, Lindner, Leidenfrost, Michaels, Mugele, Müller, Riffel, Sampangi and Winkler2019; Costello et al. Reference Costello, Osunkoya, Sandino, Marinic, Trotter, Shi, Gonzalez and Dhileepan2022; Fernandez-Quintanilla et al. Reference Fernandez-Quintanilla, Pena, Andujar, Dorado, Ribeiro and Lopez-Granados2018; Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023). Artificial intelligence can be described as an advanced machine learning system that can emulate the actions of humans, providing alternatives to constant and costly human intervention (Amend et al. Reference Amend, Brandt, Di Marco, Dipper, Gässler, Höferlin, Gohlke, Kesenheimer, Lindner, Leidenfrost, Michaels, Mugele, Müller, Riffel, Sampangi and Winkler2019; Fernandez-Quintanilla et al. Reference Fernandez-Quintanilla, Pena, Andujar, Dorado, Ribeiro and Lopez-Granados2018; Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023; Partel et al. Reference Partel, Charan Kakarla and Ampatzidis2019). This form of technology has already provided several benefits in the area of weed management and is capable of further assisting, or being integrated with, machine learning systems and robotic devices for improved weed detection and control (Partel et al. Reference Partel, Charan Kakarla and Ampatzidis2019). As this area of research is increasingly developing and its full potential has yet to be discovered, the following sections of this review will highlight where artificial intelligence has the capacity to be used alongside a range of emerging technology to assist with improved weed management.

Weed Detection Methods

It is widely agreed that the early detection and control of a weed is a critical step within a weed management program to help reduce its long-term impact on the surrounding environment (Roslim et al. Reference Roslim, Juraimi, Che’Ya, Sulaiman, Manaf, Ramli and Motmainna2021). To assist in this task, recent developments have identified several data and imagery detection methods that can improve the time taken to identify and control a weed within the field (Esposito et al. Reference Esposito, Crimaldi, Cirillo, Sarghini and Maggio2021; Mohidem et al. Reference Mohidem, Che’Ya, Juraimi, IIIahi, Roslim, Sulaiman, Saberioon and Noor2021; Singh et al. Reference Singh, Rana, Bishop, Filippi, Cope, Rajan and Bagavathiannan2020). Of the identified weed detection methods within the literature, the most promising that have been recently developed or that are undergoing further research for improvement include the use of (1) unmanned aerial vehicles, (2) all-terrain vehicles, (3) field robotics, (4) remote sensing, (5) proximal sensing, (6) plant signaling methods, and (7) crop genetic modification (Table 1). A common challenge when using certain detection methods such as unmanned aerial vehicles, all-terrain vehicles, and field robotics is their unintentional movement when capturing data or imagery. This movement, often caused by the wind or vehicle motion, increases the risk of motion blur, which can limit the quality of data and imagery and limit the success of deep learning models analyzing the data (Genze et al. Reference Genze, Wirth, Schreiner, Ajekwe, Grieb and Grimm2023). To address this issue, research by Genze et al. (Reference Genze, Wirth, Schreiner, Ajekwe, Grieb and Grimm2023) has proposed the use of a deblurring segmentation model known as DeBlurWeedSeg, which has shown to successfully mitigate motion blur when detecting weeds such as common lambsquarters (Chenopodium album L.) in grain sorghum [Sorghum bicolor (L.) Moench]. Although this model shows promising signs of reducing motion blur and increasing detection, further factors such as crop and weed height, the type of weed species, and environmental conditions all need to be considered due to their potential influence (Genze et al. Reference Genze, Wirth, Schreiner, Ajekwe, Grieb and Grimm2023). If these factors can be accounted for, or if additional models, data, or imagery can be integrated, then it may be possible to mitigate the influence of motion blur, although further investigation into this combination across a range of crop–weed scenarios would be required.

Table 1. Benefits, challenges and limitations of various weed detection methods

A common aspect of these detection methods is their ability to capture high-quality data and imagery within the field (Coleman et al. Reference Coleman, Macintyre, Walsh and Salter2023; Esposito et al. Reference Esposito, Crimaldi, Cirillo, Sarghini and Maggio2021; Lati et al. Reference Lati, Filin, Aly, Lande, Levin and Eizenberg2014; Mohidem et al. Reference Mohidem, Che’Ya, Juraimi, IIIahi, Roslim, Sulaiman, Saberioon and Noor2021; Pallottino et al. Reference Pallottino, Antonucci, Costa, Bisaglia, Figorilli and Menesatti2019; Su Reference Su2020; Sujaritha et al. Reference Sujaritha, Annadurai, Satheeshkumar, Sharan and Mahesh2017; Weiss et al. Reference Weiss, Jacob and Duveiller2020). To achieve this, the use of specialized cameras systems or sensors are required, which will often include the use of hyperspectral imagery, multispectral imagery, red-green-blue or VIS (visible) imagery, satellite imagery, thermal imagery, and 3D stereo imagery (Esposito et al. Reference Esposito, Crimaldi, Cirillo, Sarghini and Maggio2021; Rosle et al. Reference Rosle, Sulaiman, Che’Ya, Radzi, Omar, Berahim, IIAhi, Shah and Ismail2022; Su Reference Su2020; Xia et al. Reference Xia, Wang, Chung and Lee2015; Table 2). The most commonly used system is the red-green-blue imagery system, as it is a low-cost operational option that can calculate different vegetation indices to distinguish between crop and weed species (Su Reference Su2020; Xia et al. Reference Xia, Wang, Chung and Lee2015). Although this system has the capacity to effectively identify various weed species, it requires significant geometric plant data abilities to confidently recognize a range of species with high-level accuracy (Su Reference Su2020; Xia et al. Reference Xia, Wang, Chung and Lee2015; Xu et al. Reference Xu, Zhu, Cao, Jiang, Jiang, Li and Ni2021). To improve its efficiency in situations where crop and weed species appear geometrically similar, it is suggested that the use of multimodal information that uses red-green-blue imagery with depth information will help to improve weed detection accuracy (Xu et al. Reference Xu, Zhu, Cao, Jiang, Jiang, Li and Ni2021, Reference Xu, Yuen, Xie, Zhu, Cao and Ni2024). This approach has been referred to as the WeedsNet system, and Xu et al. (Reference Xu, Yuen, Xie, Zhu, Cao and Ni2024) demonstrate that the use of multimodal information has the potential to complement red-green-blue imagery in accurately detecting several grass and broadleaf weeds in a wheat (Triticum aestivum L.) crop. In this regard, such technology may have the potential to complement red-green-blue imagery systems in other agricultural settings, although this developing technology would require further detailed investigation in a range of other crop situations to determine its applicability and effectiveness in the field.

Table 2. Benefits, challenges and limitations of various imagery and sensor systems for weed detection

Another scheme known as the multispectral system has proven to be a powerful alternative to the red-green-blue imagery system, as it can capture and calculate a wider range of vegetation indices and spectral band wavelengths (Esposito et al. Reference Esposito, Crimaldi, Cirillo, Sarghini and Maggio2021; Lara et al. Reference Lara, Pedraza and Jamaica-Tenjo2020; Lu et al. Reference Lu, Dao, Liu, He and Shang2020). In a similar way, hyperspectral sensor camera systems are also powerful, being able to record thousands of radiometric narrow-band images from hundreds of spectral band wavelengths (Esposito et al. Reference Esposito, Crimaldi, Cirillo, Sarghini and Maggio2021; Su Reference Su2020). Although hyperspectral imagery has the potential to provide high accuracy in identifying several weeds, research has identified a range of inconsistencies, including insufficient feature extractions and calibration issues that limit its repeated accuracy (Diao et al. Reference Diao, Guo, Zhang, Yan, He, Zhao, Zhao and Zhang2023; Peleg et al. Reference Peleg, Anderson and Yang2005). It has also been noted that multispectral and hyperspectral systems often take a long time to calculate and collect data accumulation, as well as requiring specified [tailored] algorithms for analysis. As a consequence, there is an increase in the challenges and time taken to produce detailed precision weed maps (Zou et al. Reference Zou, Chen, Xhang, Zhou and Zhang2021). To attend to some of these issues, it has been suggested that the use of specialized models or machine learning techniques can provide further enhancement and detection efficiency (Murad et al. Reference Murad, Mahmood, Forkan, Morshed, Jayaraman and Siddiqui2023). In this respect, a range of techniques have been developed to assist in this area, with some examples including the use of (1) artificial intelligence-based image analysis (Aitkenhead et al. Reference Aitkenhead, Dalgetty, Mullins, McDonald and Strachan2003; Haq et al. Reference Haq, Tahir and Lan2023); (2) deep learning systems and algorithms (such as artificial neutral networks, convolutional neutral networks, deep neutral networks) (Hasan et al. Reference Hasan, Sohel, Diepeveen, Laga and Jones2021; Murad et al. Reference Murad, Mahmood, Forkan, Morshed, Jayaraman and Siddiqui2023); (3) image processing techniques (including clustering, generative adversarial networks, Hilbert transformation, histograms of gradients, linear iterative, local binary patterns) (Murad et al. Reference Murad, Mahmood, Forkan, Morshed, Jayaraman and Siddiqui2023; Nixon and Aguado Reference Nixon and Aguado2019); and (4) machine learning systems and algorithms (such as adaptive boosting, artificial neutral networks, decision trees, k-nearest neighbor, and support vector machines) (Murad et al. Reference Murad, Mahmood, Forkan, Morshed, Jayaraman and Siddiqui2023).

Chemical Control

Recent developments in the field of chemical control have shown promising signs in improving the management of a range of weed species around the world (Amend et al. Reference Amend, Brandt, Di Marco, Dipper, Gässler, Höferlin, Gohlke, Kesenheimer, Lindner, Leidenfrost, Michaels, Mugele, Müller, Riffel, Sampangi and Winkler2019; Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023; Roslim et al. Reference Roslim, Juraimi, Che’Ya, Sulaiman, Manaf, Ramli and Motmainna2021; Table 3). One area in particular that has been shown to be of assistance in this field is the use of autonomous chemical control, which allows for a weed to be selectively targeted and sprayed, reducing off-target damage and the need for large-scale or widespread applications (Partel et al. Reference Partel, Charan Kakarla and Ampatzidis2019). These benefits can reduce excess herbicides leaching into the soil or into surrounding waterways, while also having the benefit of reducing the cost of materials and labor needed to treat a field each season (Balafoutis et al. Reference Balafoutis, Beck, Fountas, Vangeyte, Wal, Soto, Gómez-Barbero, Barnes and Eory2017; Partel et al. Reference Partel, Charan Kakarla and Ampatzidis2019). While a range of autonomous chemical control options are currently available or under development for the control of various weed species (Table 1), it has been noted that most of these chemical devices have been tested or designed to be used within cropping systems, all of which generally have heterogeneous weed distributions that occur at a range of levels (Allmendinger et al. Reference Allmendinger, Spaeth, Saile, Peteinatos and Gerhards2022). In this regard, this technology may be difficult to use within natural ecosystems and is problematic in difficult-to-access terrain or regions with varying levels of weed infestations. As such, the use of a range of weed detection methods and imagery or sensor systems discussed earlier in this review may have the potential to assist in this area. Of particular interest, integrating such methods may enable more accurate detection and subsequent control of a weed in areas where it may appear geometrically similar to a crop or native species (Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023; Partel et al. Reference Partel, Charan Kakarla and Ampatzidis2019), although this area of research would require specific investigation relating to each scenario.

Table 3. Examples of autonomous chemical weed control devices reported within the global literature. ^a

^a Sources: Allmendinger et al. (Reference Allmendinger, Spaeth, Saile, Peteinatos and Gerhards2022); Amend et al. (Reference Amend, Brandt, Di Marco, Dipper, Gässler, Höferlin, Gohlke, Kesenheimer, Lindner, Leidenfrost, Michaels, Mugele, Müller, Riffel, Sampangi and Winkler2019); Utstumo et al. (Reference Utstumo, Urdal, Brevik, Dørum, Netland, Overskeid, Berge and Gravdahl2018); Wu et al. (Reference Wu, Aravecchia, Lottes, Stachniss and Pradalier2020); Zhang et al. (Reference Zhang, Miao, Li, He and Sun2022).

Herbicide-Resistance Management

It has been claimed that the use of artificial intelligence and specialized camera systems have the potential to assist in the field of herbicide-resistance management in agroecosystems (Amend et al. Reference Amend, Brandt, Di Marco, Dipper, Gässler, Höferlin, Gohlke, Kesenheimer, Lindner, Leidenfrost, Michaels, Mugele, Müller, Riffel, Sampangi and Winkler2019; Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023; Picoli et al. Reference Picoli, Carbonari, Matos, Rodrigues and Velini2017; Roslim et al. Reference Roslim, Juraimi, Che’Ya, Sulaiman, Manaf, Ramli and Motmainna2021). With the growing concern of more frequent occurrences of herbicide resistance, it is important to quickly identify and manage invasive populations before they reproduce to form a new generation of resistant plants. One method that has shown promise in the identification of herbicide-resistant plants is the use of thermal imagery assisted by machine learning systems (Eide et al. Reference Eide, Koparan, Zhang, Ostlie, Howatt and Sun2021; Picoli et al. Reference Picoli, Carbonari, Matos, Rodrigues and Velini2017). Research has indicated that thermal imagery can assist in the detection of glyphosate-resistant plants, as treated plants often experience stress and an inhibition of stomatal conductance that leads to reduced photosynthetic rates and increased surface temperatures (Eide et al. Reference Eide, Koparan, Zhang, Ostlie, Howatt and Sun2021; Picoli et al. Reference Picoli, Carbonari, Matos, Rodrigues and Velini2017; Shirzadifar et al. Reference Shirzadifar, Bajwa, Nowatzki and Bazrafkan2020). As a consequence, thermal imagery, assisted by machine learning techniques have the capability of identifying these plants within a field, which can then be controlled by other mechanisms (Eide et al. Reference Eide, Koparan, Zhang, Ostlie, Howatt and Sun2021; Shirzadifar et al. Reference Shirzadifar, Bajwa, Nowatzki and Bazrafkan2020). On the other hand, research by Eide et al. (Reference Eide, Koparan, Zhang, Ostlie, Howatt and Sun2021) suggests that when thermal imagery is used in isolation, it is not a completely reliable indicator to predict herbicide resistance, particularly when dealing with herbicides with different modes of action. In this regard, further research into combining thermal imagery with other machine learning techniques may assist in identifying herbicide-resistant weed populations before they can reproduce. It has also been noted that integrating artificial intelligence in real-time image processing may help to assist in identifying herbicide-resistant and susceptible plants by identifying a range of plant and soil characteristics, although this area of research requires further investigation among different herbicides and crop–weed scenarios (Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023). It has also been noted that due to the limited number of new or emerging herbicide modes of action, the integrated use of existing bioherbicides could be another option to increase control efficiency and help reduce potential herbicide resistance in some weed species (Roberts et al. Reference Roberts, Florentine, Fernando and Tennakoon2022).

Mechanical Control

Robotics and machine learning capabilities have shown promising signs as an emerging method to mechanically control a range of invasive weeds (Fennimore et al. Reference Fennimore, Slaughter, Siemens, Leon and Saber2016; Oliveira et al. Reference Oliveira, Moreira and Silva2021). For example, a study by Bakker et al. (Reference Bakker, Asselt, Bontsema, Müller and Straten2010) indicated that the use of intelligent autonomous weeders with real-time kinematics global positioning systems can be used to complete interrow hoeing within corn (Zea mays L.) crops, with minimal to no damage to the surrounding crop plants. Although this method seems promising, the speeds of such devices are often much slower than traditional methods that often operate at a minimum speed of 4 km h⁻¹ or up to 12 km h⁻¹ for methods such as harrowing (Bakker et al. Reference Bakker, Asselt, Bontsema, Müller and Straten2010; Bowman Reference Bowman2002). Due to these slower rates of speed, some robotic devices may not be able to completely bury or uproot certain weed species in a given time, ultimately allowing them to reestablish (Bakker et al. Reference Bakker, Asselt, Bontsema, Müller and Straten2010). This clearly indicates that careful consideration of both the type of weed species and the speed of the mechanical device is needed for the implementation of these methods. Another study by Nørremark et al. (Reference Nørremark, Griepentrog, Nielsen and Søgaard2012) showed 91% success rates in treating various interrow weeds with a robotic cycloid hoe, while research by Kunz et al. (Reference Kunz, Weber and Gerhards2015) showed that the use of a camera-guided interrow hoeing device (Kult Robocrop®) reduced weed density by 89% in soybean [Glycine max (L.) Merr.] crops and 87% in sugar beet (Beta vulgaris L.) crops. It appears that camera-guided devices, fitted with machine learning capabilities, can be utilized to help guide the robotic device in targeting specific weeds within these areas, which can markedly improve these methods (Kunz et al. Reference Kunz, Weber and Gerhards2015). Such technology may also help to identify weeds that are growing very close to a crop plant and specifically target it or identify and map it for an alternative treatment to minimize potential crop damage.

A study by Van Evert et al. (Reference Van Evert, Samsom, Polder, Vijn, Van Dooren, Lamaker, Van Der Heijden, Kempenaar, Van Der Zalm and Lotz2011) was associated with the development and testing of an autonomous robotic device to control broadleaf dock (Rumex obtusifolius L.) on a commercial farm. The device navigated using global navigation satellite system technology and included a downward-facing camera for plant detection and a mechanical tool with rotating blades that was lowered and activated once the plant was identified (Van Evert et al. Reference Van Evert, Samsom, Polder, Vijn, Van Dooren, Lamaker, Van Der Heijden, Kempenaar, Van Der Zalm and Lotz2011). This method was reported to have a 93% detection rate and a 75% success rate in the control of the species. It is anticipated that this method could be used to help decrease the quantity of herbicide needed to treat a field and reduce the costs associated with on-farm labor. Robotic weed control devices can also work in conjunction with other devices, with research by Noguchi et al. (Reference Noguchi, Will, Reid and Zhang2004) creating a primary and secondary system allowing for several devices to work together.

Another device that has been developed as an autonomous mechanical and chemical robot within the agricultural industry is the AgBotII. This device is capable of accurately identifying 90% of the selected species and has shown success in controlling various weeds such as wild oats (Avena fatua L.) and common sowthistle (Sonchus oleraceus L.). It is a powerful alternative, as it can provide both mechanical and chemical control, allowing an integrated control method, and alternately addressing control mechanisms with each pass (Fennimore et al. Reference Fennimore, Slaughter, Siemens, Leon and Saber2016; Oliveira et al. Reference Oliveira, Moreira and Silva2021). Another device that has been developed for the use of robotic weed control is the BoniRob platform produced by Bosch Deepfield Robotics, which has shown up to 94% success in controlling various weeds (Langsenkamp et al. Reference Langsenkamp, Sellmann, Kohlbrecher, Kielhorn, Strothmann, Michaels, Ruckelshausen and Trautz2014). This device is capable of identifying and selectively targeting a weed by means of mechanical control (Langsenkamp et al. Reference Langsenkamp, Sellmann, Kohlbrecher, Kielhorn, Strothmann, Michaels, Ruckelshausen and Trautz2014). Although promising, this technique can be time-consuming and may not be applicable for deeper or heavier clay-type soils, as it is more suitable for sandy light soils (Langsenkamp et al. Reference Langsenkamp, Sellmann, Kohlbrecher, Kielhorn, Strothmann, Michaels, Ruckelshausen and Trautz2014). In this regard, further investigation on a range of autonomous mechanical devices will need to consider more localized conditions to allow adjustment for greater efficacy in control options. Further investigation will also need to consider (1) which weed species can be mechanically controlled and which will require follow up or integrated methods of control; (2) the accuracy of control to limit potential off-target damage or unwanted soil disturbance, particularly when used alongside planted crops or native vegetation; (3) any potential spread of weeds and their propagules into other areas of the field if the devices are not cleaned or treated appropriately; and (4) the likely change to crop patterns or rows to help facilitate automated cultivation in two directions for improved efficiency (Fennimore et al. Reference Fennimore, Slaughter, Siemens, Leon and Saber2016; Sharma et al. Reference Sharma, Tomar and Chakraborty2017).

Electrical Weed Control

A range of research studies have shown that electrical weed control has the potential to successfully control various weeds, with several products already being commercially developed for use around the world (Table 4). This method delivers an electrical current to the targeted plant and can be applied by two main methods: (1) the spark-discharge method or (2) a continuous electrode–plant contact method (Slesarev Reference Slesarev1972; Wilson and Anderson Reference Wilson and Anderson1981). The spark-discharge method transfers a brief high-voltage current directly into a plant, while the continuous electrode–plant contact requires ongoing contact between the electrodes and the plant in order to allow a lethal dose of electricity to pass through the foliage, stems, and roots of the plant (Slesarev Reference Slesarev1972; Wilson and Anderson Reference Wilson and Anderson1981). These methods often cause damage to the plant’s cells and structures by increasing temperature and vaporizing volatile liquids, ultimately damaging cell membranes (Slesarev Reference Slesarev1972; Wilson and Anderson Reference Wilson and Anderson1981). In some cases, electrical treatment may not completely kill a plant and may only cause damage in some areas, allowing the plant to regenerate over time (Slesarev Reference Slesarev1972; Wilson and Anderson Reference Wilson and Anderson1981). To achieve successful control, several factors need to be considered, such as (1) the type of device used and the amount of energy output for a lethal dose; (2) contact time with the plant; (3) surrounding vegetation, to ensure targeted plants are not shielded by other vegetation; (4) surrounding environmental conditions; and (5) the potential risk of fire (Bloomer et al. Reference Bloomer, Harrington, Ghanizadeh and James2024; Landers et al. Reference Landers, Challiol, Vilela and Lanz2016; Lehnhoff et al. Reference Lehnhoff, Neher, Indacochea and Beck2022; Slesarev Reference Slesarev1972; Vigneault et al. Reference Vigneault, Benoit and McLaughlin1990; Wilson and Anderson Reference Wilson and Anderson1981). Of particular interest, research by Schreier et al. (Reference Schreier, Bish and Bradley2022) found a strong correlation between the increase of plant moisture content and the decreased level of weed control using electricity on several weed species such as barnyard grass [Echinochloa crus-galli (L.) P. Beauv.], common ragweed (Ambrosia artemisiifolia L.), giant foxtail (Setaria faberi Herrm.), giant ragweed (Ambrosia trifida L.), horseweed [Conyza canadensis (L.) Cronquist], waterhemp [Amaranthus tuberculatus (Moq.) Sauer], and yellow foxtail [Setaria pumila (Poir.) Roem. & Schult.], in a soybean-cropping system. In this regard, the use of electrical weed control needs to carefully consider any potential influence from the surrounding environment, as this is likely to influence the success of this method (Lati et al. Reference Lati, Rosenfeld, David and Bechar2021; Schreier et al. Reference Schreier, Bish and Bradley2022).

Table 4. Common electrical devices commercially developed around the world for weed control

It is clear that electric weed control can be suitable for a range of cropping systems, with research suggesting that it can be successfully used to control a diverse range of species (Bloomer et al. Reference Bloomer, Harrington, Ghanizadeh and James2024; Landers et al. Reference Landers, Challiol, Vilela and Lanz2016). Research by Landers et al. (Reference Landers, Challiol, Vilela and Lanz2016) in Brazil and Paraguay used a plant–electrode contact machine (16.6 km h⁻¹) and obtained between 94% to 100% control after 28 d for weeds such as high mallow (Malva sylvestris L.), smallflower galinsoga (Galinsoga parviflora Cav.), S. oleraceus, and wild poinsettia (Euphorbia heterophylla L.). On the other hand, only up to 75% success rates were realized when trying to control garden spurge [Chamaesyce hirta (L.) Millsp.], thus showing that certain weed species may respond differently to each treatment (Landers et al. Reference Landers, Challiol, Vilela and Lanz2016). Electrical weed control can also be used to reduce seed development, with an example of the Weed Zapper™ 6R30 (spark-discharge) achieving 54% to 80% reduction in A. artemisiifolia, A. tuberculatus, A. trifida, S. faberi, S. pumila, and common cocklebur (Xanthium strumarium L.) (Schreier et al. Reference Schreier, Bish and Bradley2022).

Although there are several benefits of using electrical weed control methods, there are also several risks associated with its use (Bloomer et al. Reference Bloomer, Harrington, Ghanizadeh and James2024). One drawback of using such methods is the low rate of speed and the considerable time needed to treat large fields (Llewellyn et al. Reference Llewellyn, Ronning, Clarke, Mayfield, Walker and Ouzman2016). A potential solution to this challenge would be to integrate this technology with artificial intelligence or machine learning devices to help identify and then treat weeds autonomously (Llewellyn et al. Reference Llewellyn, Ronning, Clarke, Mayfield, Walker and Ouzman2016; Machleb et al. Reference Machleb, Peteinatos, Kollenda, Andújar and Gerhards2020). It is also important to consider any potential resistance to electrical weed control (Beckie et al. Reference Beckie, Flower and Ashworth2020; Somerville et al. Reference Somerville, Powles, Walsh and Renton2017). In particular, some species with higher levels of cellulose or lignin within their cell walls may have a higher resistance to bursting or becoming vaporized; likewise, plants with a hairy, thicker, or waxy epidermis may also be more protected from electrical control (Bauer et al. Reference Bauer, Marx, Bauer, Flury, Ripken and Streit2020; Vigneault et al. Reference Vigneault, Benoit and McLaughlin1990). Plants that are not completely controlled may regenerate at a later time from their root systems; therefore, careful postcontrol monitoring may be required (Bloomer et al. Reference Bloomer, Harrington, Ghanizadeh and James2024; Lehnhoff et al. Reference Lehnhoff, Neher, Indacochea and Beck2022).

Research has suggested that the use of electrical weed control can also impact soil biota, as it may travel through a plant’s root system and into the soil or adjacent water, thus potentially affecting the cellular constituents of surrounding organisms (Ruf et al. Reference Ruf, Oluwaroye, Leimbrock and Emmerling2023). The effect of electrical weed control on the surrounding organism will depend on the voltage and length of the application (Ruf et al. Reference Ruf, Oluwaroye, Leimbrock and Emmerling2023). For example, research by Lati et al. (Reference Lati, Rosenfeld, David and Bechar2021) used an application of 0.05 Wh, which resulted in an increase of more than 40 C in the shoots and roots of black nightshade (Solanum nigrum L.) and redroot pigweed (Amaranthus retroflexus L.), which is likely to influence the surrounding soil conditions. Ruf et al. (Reference Ruf, Oluwaroye, Leimbrock and Emmerling2023) also found a reduction in earthworm biomass within the top 25 cm of the soil profile after using the Zasso™ XPower XP300 applicator at 3 km h⁻¹ for 2 wk compared with an uncontrolled area. Research has also shown several physical changes to the earthworms from areas that have been treated with electrical weed control, such as a change in skin color, the presence of necrotic tissue, and damage to other cells (Ruf et al. Reference Ruf, Oluwaroye, Leimbrock and Emmerling2023). On the other hand, contrasting results have shown that some macro- and mesofauna can survive in areas treated by electrical weed control and maintain more stable populations compared with other methods such as mechanical control where there are large soil disturbance events (Löbmann et al. Reference Löbmann, Klauk, Lang, Petgen and Petersen2022). Based on the differences in these findings, it is suggested that localized populations, environmental conditions, or soil types may play an important role in the response of soil biota, and therefore would require localized investigations to determine any long-term effects of this method.

Laser Weed Control

The combination of lasers and various weed detection systems has been shown to be a promising method in weed control, particularly in the early life-cycle stage (Heisel et al. Reference Heisel, Andreasen and Christensen2002; Rakhmatulin et al. Reference Rakhmatulin, Kamilaris and Andreasen2021; Wang et al. Reference Wang, Zhang and Wei2019). These devices have the capacity to be integrated with machine learning devices, including classification algorithms and automated devices for efficient weed control within agroecosystems (Ghatrehsamani et al. Reference Ghatrehsamani, Jha, Dutta, Molaei, Nazrul, Fortin, Bansal, Debangshi and Neupane2023; Rakhmatulin et al. Reference Rakhmatulin, Kamilaris and Andreasen2021). The most commonly used lasers include carbon dioxide lasers, diode lasers, and fiber laser devices (Coleman et al. Reference Coleman, Kristiansen, Sindel and Fyfe2021; Gates et al. Reference Gates, Keegan, Schleter and Weidner1965; Heisel et al. Reference Heisel, Andreasen and Christensen2002; Wöltjen et al. Reference Wöltjen, Haferkamp, Rath and Herzog2008). These devices work by emitting an infrared beam that is absorbed by the plant’s cells, consequently burning them (Gates et al. Reference Gates, Keegan, Schleter and Weidner1965; Heisel et al. Reference Heisel, Andreasen and Christensen2002). Several studies have shown that the use of lasers can be effective in controlling various weeds at different rates, with some examples including E. crus-galli at 54 J per plant (Marx et al. Reference Marx, Barcikowski, Hustedt, Haferkamp and Rath2012) and rigid ryegrass (Lolium rigidum Gaudin) at 76.4 J (Coleman et al. Reference Coleman, Kristiansen, Sindel and Fyfe2021). Several other species that have also been severely damaged or controlled using laser weed control include A. fatua (Bayramian et al. Reference Bayramian, Fay and Dyer1992), cereal rye (Secale cereale L.) (Bayramian et al. Reference Bayramian, Fay and Dyer1992), tobacco (Nicotiana tabacum L.) (Wöltjen et al. Reference Wöltjen, Haferkamp, Rath and Herzog2008), and water hyacinth [Eichhornia crassipes (Mart.) Solms; syn.: Pontederia crassipes (Mart.) Solms] (Couch and Gangstad Reference Couch and Gangstad1974). A common trend within the literature shows that each weed species often requires a different contact time or energy dose to be controlled or partially damaged by the use of lasers (Marx et al. Reference Marx, Barcikowski, Hustedt, Haferkamp and Rath2012; Rakhmatulin and Andreasen Reference Rakhmatulin and Andreasen2020). Research has also shown that many of these robotic devices, such as the commercially available LaserWeeder by Carbon Robotics, are often very expensive and very slow, only reaching speeds of 1.6 km h⁻¹ (Vijayakumar et al. Reference Vijayakumar, Ampatzidis, Schueller and Burks2023). For such applications to become more widely available, a time–cost–value analysis may be useful to determine whether this is the most appropriate method for a specific weed situation. Another important consideration regarding laser weed control is its influence on the surrounding soil biota, with long-term data on a range of environments required to examine this influence at a broader range (Khan et al. Reference Khan, Jurburg, He, Brody and Gupta2020).

Thermal Weed Control

The use of thermal weed control has been successfully integrated within the agricultural industry as a pre–crop emergence technique for weed control (Seaman Reference Seaman2016). Thermal weed control works by emitting quantities of intense heat directly to a targeted plant, which can increase its temperature and thus physically disrupt its cells (Brodie et al. Reference Brodie, Khan and Gupta2019; Seaman Reference Seaman2016). Thermal weed control can often be in the form of a flame, hot oil, steam, or radiation (Brodie et al. Reference Brodie, Khan and Gupta2019). Such technology has been implemented across various agroecosystems for the control of various weeds such as bermudagrass [Cynodon dactylon (L.) Pers.], E. crus-galli, hairy beggarticks (Bidens pilosa L.), ragweed parthenium (Parthenium hysterophorus L.), and several Amaranthus species (Mutch et al. Reference Mutch, Thalmann, Martin and Baas2008; Ulloa et al. Reference Ulloa, Datta, Bruening, Gogos, Arkebauer and Knezevic2012). Further research has also been used to trial the use of hot foam as a thermal weed control, which has shown success in controlling more than 75% of various species such as little mallow (Malva parviflora L.) and wild mustard (Sinapis arvensis L.) (Antonopoulos et al. Reference Antonopoulos, Kanatas, Gazoulis, Tataridas, Ntovakos, Ntaoulis, Zavra and Travlos2023). The impact of thermal weed control is claimed to be more successful on annual species, with the effect on perennial species being more variable due to their more developed and structured root systems or underground rhizomes, which can reshoot if they are not completely damaged (Cisneros and Zandstra Reference Cisneros and Zandstra2008). As a consequence, thermal weed control may not be suitable for all weed species and may thus result in surviving weed species creating monospecific stands, all of which can result in the need for further control and resources. For this method to be more widely used with confidence, further investigation on thermal weed control on a range of weed species across different environments would be needed.

The use of radiation-based thermal control, often referred to as microwave radiation, has shown promising results in the control of various weed species (Brodie Reference Brodie2012; Brodie et al. Reference Brodie, Khan and Gupta2019). Under laboratory conditions, Aygun et al. (Reference Aygun, Cakir and Kacan2017) have shown that microwave radiation energy has the capacity to kill various weed species such as C. dactylon, johnsongrass [Sorghum halepense (L.) Pers.], S. nigrum, and X. strumarium. Although this method successful for individual species, each species generally required a different speed and rate of microwave radiation for its effective control (Aygun et al. Reference Aygun, Cakir and Kacan2017). Further research also suggests that different species respond differently to a changing application and rate of microwave radiation (Brodie and Hollins Reference Brodie and Hollins2015; Kaçan et al. Reference Kaçan, Çakir and Aygün2018). In particular, a study in Australia by Brodie and Hollins (Reference Brodie and Hollins2015) identified that wild radish (Raphanus raphanistrum L.) required at least 60 J/cm⁻² for 100% mortality, while L. rigidum required 370 J/cm⁻². Despite these successes, the use of microwave radiation energy for weed control often requires at least 10 times more energy than traditional methods such as chemical control, which therefore has currently limited its commercial use (Brodie Reference Brodie2012). One solution that could improve this method is its potential use alongside artificial intelligence and automated weed control devices. This could allow for plant detection and specific plant control using microwave energy to limit the need for its use across a large scale, although this method would require further investigation and an economic benefit analysis before it could be confidently used as a key weed control method.