I know what you grew last summer | Steve Shirtliffe | TEDxUniversityofSaskatchewan

Key Concepts

• Phenotyping: The process of measuring and analyzing plant characteristics.
• Precision Agriculture: Managing crops and fields based on localized data and variability.
• Google Earth Engine: A cloud-based platform for accessing and analyzing geospatial data, including satellite imagery.
• GAGE (Geospatial Agroecosystem Inference Engine): The model developed to map and analyze crop land in Western Canada.
• Marginal Land: Areas within fields with consistently low productivity, often due to salinity or dryness.
• Glyphosate: A widely used herbicide.
• Kosha (Kochia): A herbicide-resistant weed.
• Yield Maps: Data collected by combines showing grain yield variations across a field.

The Evolution of Remote Sensing in Agriculture: From Cold War Spying to Farmer Empowerment

The speaker, a professor with roots in a Manitoba farming community, details his journey from observing a childhood farm captured in a Cold War-era spy satellite image to leading a cutting-edge research program utilizing satellite imagery and AI to revolutionize agricultural practices. His narrative highlights the progression from basic observation to sophisticated data analysis and the ethical considerations surrounding its application.

Early Observations & Academic Pursuit

The presentation begins with a personal anecdote: a 1982 image of the speaker’s family farm taken by a US Keyhole spy satellite. This image, declassified in 1995, sparked an early fascination with remote observation. Despite leaving the farm for university to study agronomy – the science of crop management – the speaker maintained a desire to contribute to the well-being of farmers. He emphasizes the importance of developing solutions for challenges like herbicide-resistant weeds and efficient crop management.

The Drone & the Genesis of Phenotyping

Around 2015, the acquisition of a drone from Dragonfly in Saskatoon marked a turning point. This allowed for direct crop imagery and led to involvement in a University of Saskatchewan research grant focused on food security and crop breeding. The speaker became a lead researcher, focusing on phenotyping – the precise measurement of plant characteristics like development timing, growth rate, and flower/head count – using drone imagery to aid breeders in selecting superior crop varieties.

The Power of the Google Earth Engine

The real breakthrough came with the introduction to Google Earth Engine by researcher Bruno Basso of Michigan State University. The speaker describes the Earth Engine as a “turbocharged” version of Google Earth, providing free access to a vast archive of publicly available satellite imagery and Earth observation data for academic use. This enabled analysis of millions of acres, a scale impossible with drones alone. The team began integrating satellite imagery with weather, soil, and topographic data to understand field variability. They discovered that crop performance is significantly influenced by factors like slope (hillsides yield less) and proximity to wetlands (saline soils reduce yield).

Precision Agriculture & Identifying Marginal Land

This research falls under the umbrella of precision agriculture, which aims to divide fields into management zones with tailored crop input prescriptions. The goal is increased efficiency by adapting management practices to specific field conditions. The team identified marginal land – areas with consistently low productivity due to salinity or dryness – as potential targets for reduced input costs or conversion to perennial grasses.

Mapping & Modeling: The GAGE Engine

The team’s work expanded to mapping the productivity of land using satellite imagery and machine learning. They collaborated with farmers who shared yield maps (data from combines logging grain yields) to train a machine learning model to predict yields for canola and wheat across 14 million acres. This proved successful, identifying yield variations at a 10-meter resolution. This success led to an ambitious project: mapping all 70 million acres of crop land in Western Canada. This initiative resulted in the development of GAGE (Geospatial Agroecosystem Inference Engine), an AI-powered model for analyzing agroecosystems.

Real-World Applications: Weed Detection & Beyond

Specific applications of this technology include identifying patches of the herbicide-resistant weed kosha using AI-trained satellites, allowing farmers to target mowing efforts before seed dispersal. The team is also exploring applications like identifying areas prone to root disease, optimizing crop selection, and improving fertilizer efficiency. They are investigating how satellite data can reveal where greenhouse gases are emitted and how to enhance soil biodiversity.

The Ethical Dilemma & Farmer Concerns

Presenting this work to farmers in Manitoba revealed a significant ethical challenge. Farmers expressed concerns about data privacy, potential exploitation, and the possibility of their data being used against them (affecting land values or commodity prices). One farmer bluntly stated, “This is terrible. You are spying on us.” The speaker acknowledged the validity of these concerns, recognizing that the models were accurate enough to potentially betray the trust of their farmer collaborators.

A New Ethical Framework: "How Will the Farmers Benefit?"

This feedback prompted a shift in approach. The speaker realized that simply releasing the models would be irresponsible. Instead, they are adopting a policy driven by the question, “How will the farmers benefit?” They will focus on developing applications that directly assist farmers – such as identifying weed locations or predicting disease risk – while not selling or releasing detailed yield predictions. The speaker emphasizes that farmers own their fields (their “factories”) and that while satellite imagery is publicly available, the value lies in the interpretation provided by yield maps and sophisticated models.

Future Vision & Conclusion

The speaker envisions a future where this technology empowers farmers, protecting them from exploitation and enabling them to thrive. He stresses the importance of integrating AI and modern methods into farming to ensure the long-term sustainability of agriculture, particularly for farmers like his cousins on Shirtlift Road. The University of Saskatchewan’s new research center is dedicated to becoming a global leader in digital agriculture, guided by a commitment to ethical data practices and farmer well-being. The ultimate goal is to produce more food with fewer resources and minimize environmental impact.