Weeds pose the most persistent and costly threat to Canadian crop production, with widespread herbicide use and an accelerating rise in herbicide-resistant species. This article explores how new AI and trait-based decision-making tools can transform weed management and usher in a new era of accurate and sustainable herbicide management.
Weeds are the most consistently recurring and detrimental biological threat to crop production in Canada. Without the use of herbicides, yield losses due to weeds can be well over 50%. (1,2). Herbicides are used on nearly every field in Canada and account for more than 75% of all pesticides used in Canada. Intensive preventive use of herbicides has selected for herbicide-resistant (HR) weed biotypes that currently infest approximately 24 million acres of the Canadian prairies and cost producers more than CAD 600 million annually. (3). Herbicide resistance continues to rise. This problem is not unique to Canada.
As the commercialization of new herbicide modes of action has stagnated over the past 40 years, continued effective weed management requires improved management of existing herbicides. Improving on-farm herbicide management requires more accurate field- and context-specific decision-making tools, as not all weeds are equally competitive or evenly distributed across the field. Not all weeds require herbicide treatment.
An underutilized decision-making tool: thresholds and critical weed-free periods.
Two underutilized weed management tools to improve herbicide management are weed control thresholds and critical periods for weed control. The threshold is the weed density at which economic damage occurs to the crop, while the critical period for weed control defines the portion of the crop life cycle where the crop must be cleared of weeds to prevent unacceptable yield losses. (4). Both tools have had limited adoption. Threshold densities are weed species specific, and yield losses are influenced by the relative emergence times of weeds and crop management practices.
Accurately quantifying threshold density in large fields is not practical using traditional means.
A shift from traditional density-based approaches to functional traits (i.e., ground cover, height) that capture real-time crop-weed interference regardless of species or time of occurrence appears to be a more robust and accurate approach suitable for remote sensing. Combining thresholds and critical weed-free periods into one weed management decision-making tool, allowing more accurate site-by-site estimates of crop-weed interference, is a technological achievement that will revolutionize herbicide management.
Technology enabling next generation weed management
Technological advances such as (i) the ability to precisely control the dosage of herbicide at each sprayer nozzle, (ii) precise GPS-guided machines and drones, and (iii) high-resolution image capture with remote sensing combined with computer vision for plant identification and phenotyping provide the backbone for developing field-specific weed management decision support systems that can take herbicide management to a new level while maintaining crop productivity.
A University of Manitoba research team is working with partners to develop a smart AI-based decision support system that recommends treating weeds with herbicides based on canopy development rather than density. This, combined with estimates of yield loss, will improve herbicide application and reduce selection pressure on HR weeds.
The proposed smart decision-based system can also monitor weed growth after treatment, thus enabling rapid detection of HR weed biotypes after application using remote sensing. Traditional pre- and post-spray weed searches are labor-intensive and resistant individuals or small patches are rarely detected before they drop their seeds.
Improved accuracy and site-specific weed management
Precise, or site-specific, weed management is an approach that can be used in conjunction with a number of management options. Accurate weed management relies on spatially specific weed detection and geolocation, classification of weeds by category, species, or biotype, and management based on best practices. This is a true revolution in weed management, as weed management at the field scale was based on observations at a few isolated locations within the field. Traditional weed surveys are labor-intensive and time-consuming.
Key to these systems are computer vision and trained machine learning models that can identify and segment weeds within crop canopies by category or species. (5). Models for weed segmentation for soybean (Figure 1), canola, and wheat have been developed and continue to be refined by the UM team. Computationally, it is more efficient to classify weeds into broad categories than by species.
Towards a future of comprehensive and sustainable weed management
Once operational, this comprehensive tool will provide superior weed control, reduce pesticide costs for producers, and provide a safer way to produce food that meets the demands of discerning export markets while protecting the environment. This decision-making tool can also be used in conjunction with non-herbicidal weed management tools such as in-crop tillage, laser weeding, and other site-specific weed management approaches. Moreover, computer vision algorithms are highly effective in early detection and rapid response of existing and new HR weeds through high-density remote sensing, and detection of herbicide leakage growing in the understory.
References
- Soltani N et al. 2017. Weed Technology 31:148-154
- Soltani N et al. 2018. Weed Technology 32:342-346
- Becky HJ et al. 2020. Weed Technology. 34:461-474
- Swanton CJ et al. 1999. Canadian Journal of Plant Science. 79:165-167
- Gomloki M et al. 2025. smart farming technology. DOI: https://doi.org/110.1016/j.atech.2025.101472
