Objectives
1. To identify kochia using image based methodologies.
2. To determine the degree in which soil salinity influences kochia density, plant size, and abundance within a field.
3.To determine if the identification of kochia using UAV imagery can be scaled up using high resolution satellite imagery, enabling this technology to have much broader use for the industry.
4. Identify the potential to use salinity-based data to develop management systems for spatially explicit weed management systems for kochia and eliminate “seed nurseries”.
Project Description
Kochia is a herbicide-resistant weed posing an increasing threat to agricultural production across the Canadian Prairies. Its ability to survive in drought and saline conditions allows it to flourish in areas where crops often struggle—especially in saline zones within fields that act as “seed nurseries” supporting early kochia growth and reproduction. With widespread resistance to Groups 2, 4, and 9 herbicides, management has become more complex and costly. In response, this project developed precision, remote sensing–based methods for detecting kochia and mapping soil salinity, with the goal of enabling targeted, site-specific weed control and slowing resistance development.
Between 2015 and 2024, the team compiled extensive datasets, including 621 documented kochia patches, UAV imagery across 1,025 hectares, aerial photography covering 14,354 hectares, and high-resolution satellite imagery spanning 722 square kilometers. Repeated UAV flights throughout the growing season identified late July to early August as the optimal time for detecting kochia, due to heightened NDVI contrast with surrounding crops. Results showed high accuracy in UAV-based detection, with models achieving 95% accuracy in wheat and 97.5% in lentil fields.
These findings confirm the effectiveness of UAV imagery when collected during the right time window. To support broader applications, satellite-based methods were also developed. Using 30 cm PNEO satellite imagery and a Random Forest classifier, researchers mapped kochia with 96% accuracy. For large-scale scalability, a deep learning model was applied to create super-resolution Sentinel-2 imagery at 1-meter resolution. When processed using machine learning in Google Earth Engine, this approach achieved over 98% classification accuracy. Key variables included RGB bands,
Sentinel-2’s Leaf Area Index (LAI), the Modified Terrestrial Chlorophyll Index, and a custom “Kochia Index” created during the project.
In parallel, the project investigated the role of soil salinity in shaping kochia distribution. Using data from 166 field survey sites, a stacked ensemble model predicted soil electrical conductivity (EC) with an R2 of 0.74 under spatial validation and 0.69 under space–time validation, demonstrating solid generalizability. These results support the use of satellite-derived environmental data to model salinity across broad regions.
Grower Benefits
A major deliverable from this work is the development of two interactive tools: a public soil salinity map covering over 601,000 hectares, and a scalable kochia mapping workflow using enhanced Sentinel-2 imagery. These tools are web-accessible and designed to support real-time monitoring and site-specific weed control for producers, agronomists, and researchers. This project met all four objectives and provides immediately usable, cost-effective tools for precision weed management. The integration of kochia and salinity mapping enables producers to focus interventions in high-risk zones, improving control, reducing herbicide use, and extending the life of resistance-sensitive products. These methods offer a solid foundation for advancing precision agriculture across Western Canada.
