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

Source: https://www.youtube.com/watch?v=hioZNUbzGUQ
Recap page: https://rapidrecap.app/video/hioZNUbzGUQ
Generated: 2026-02-10T22:34:04.108+00:00

---
## Quick Overview

Dr. Steve Shirtliffe advocates for integrating Artificial Intelligence (AI) and satellite imagery into modern precision agriculture practices to improve crop phenotyping, manage variable field conditions like salinity and weed patches (like Kochia), and ultimately increase yield efficiency while protecting farmers' livelihoods and the environment.

**Key Points:**
- Dr. Shirtliffe, a digital agronomist, uses historical spy satellite imagery from 1982 to illustrate the long-term variability in his family's Southern Manitoba farm, which was impossible to control then.
- The core focus is using AI and freely available satellite imagery (like Google Earth Engine data) to perform high-resolution phenotyping, identifying crop characteristics like growth rate and head count.
- A key application demonstrated is the detection of Kochia (a weed resistant to glyphosate) using satellite imagery, allowing farmers to target weed patches before they spread seed.
- The research also involves creating productivity maps (four-year average) classifying fields as 'Stable high,' 'Stable low,' or 'Unstable' based on factors like soil salinity and wetness, enabling variable management prescriptions.
- The ultimate goal is to increase farming efficiency by tailoring inputs (fertilizer, etc.) to specific management zones within a field, rather than treating the entire field uniformly.
- Shirtliffe emphasizes that this technology must be used ethically, benefiting farmers and food security, not exploiting them or negatively impacting commodity prices or national security interests.

![Screenshot at 08:09: The speaker discusses using the GAIG \(Geospatial Agroecosystem Inference Generator\) model, which analyzes publicly available satellite data, to address major agricultural questions like yield prediction and input efficiency.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-08-09.jpg)

**Context:** Dr. Steve Shirtliffe, a digital agronomist from the University of Saskatchewan, presents his work exploring the application of Artificial Intelligence (AI) and remote sensing, specifically satellite imagery, to revolutionize precision agriculture. He grounds his modern research in personal history, referencing aerial photos of his family farm in Southern Manitoba taken during the Cold War era (1982) to establish the long-term challenge of managing inherent field variability.

## Detailed Analysis

Dr. Steve Shirtliffe details how AI and satellite imagery are transforming agriculture by enabling precision management based on granular data. He begins by showing a 1982 spy satellite image of his family farm in Manitoba, illustrating that variability (like weed patches) has always existed but was uncontrollable. His current research focuses on using tools like the Geospatial Agroecosystem Inference Generator (GAIG) to analyze massive amounts of publicly available Earth Observation data, including historical imagery, to perform detailed phenotyping—measuring traits like crop growth rate and flower/head count. A successful application shown is the detection of Kochia weed patches, allowing farmers to spot and treat infestations before they seed, which is crucial for herbicide-resistant weeds. Furthermore, the research uses four-year average productivity maps to identify areas within fields that are consistently high-yielding (stable) versus low-yielding or unstable due to factors like soil salinity or wetness. The goal is to move from uniform application to variable management prescriptions across millions of acres, thereby increasing efficiency, saving money on inputs, and protecting farmers' livelihoods ethically. Shirtliffe notes that while the technology is powerful, its application must adhere to an ethical framework that benefits farmers rather than exploits them or destabilizes markets.

### Introduction and Historical Context

- Speaker is Dr. Steve Shirtliffe, Digital Agronomist
- Discusses using 1982 spy satellite imagery of his Manitoba farm to illustrate historical field variability
- He left farming to pursue work benefiting farmers.

### The Role of AI in Agriculture

- Established a research program focused on creating solutions for weed resistance and managing new crops
- Utilizes drones (like Dragonfly) to capture high-resolution crop images
- The University of Saskatchewan received a grant to design and breed better, safer crops.

### GAIG Model and Data Utilization

- Introduced the GAIG (Geospatial Agroecosystem Inference Generator) model
- Uses publicly available satellite imagery (Google Earth Engine) to analyze productivity and variability (e.g., soil wetness, salinity) across large areas
- The model predicts yield and helps farmers manage specific zones within fields.

### Application and Ethical Considerations

- Successfully detected Kochia weed patches using imagery to guide targeted spraying
- Productivity assessments identify marginal/unstable land
- The ultimate goal is increasing efficiency and protecting farmer livelihoods, necessitating ethical guidelines to avoid exploitation or market disruption.

![Screenshot at 00:07: An early USGS EROS Data Center satellite image from 1982 showing the speaker's family farm layout, illustrating historical land use and variability.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-00-07.jpg)
![Screenshot at 01:44: A drone \(DJI Matrice series\) is shown landing on a black mat in a tilled field with young crop rows visible, demonstrating the current data collection method.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-01-44.jpg)
![Screenshot at 04:56: A slide titled 'Satellite imagery related to productivity' displays two different satellite views, one showing broad regional fields \(green/purple/yellow\) and another zoomed-in view highlighting productivity variations.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-04-56.jpg)
![Screenshot at 05:59: A slide titled 'Productivity assessment: 4-year average' shows a map coded red \(low\), yellow \(unstable\), and green \(stable high\), illustrating yield variability across farm parcels.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-05-59.jpg)
![Screenshot at 07:32: A slide introducing the GAIG \(Geospatial Agroecosystem Inference Generator\) model with a colorful map of Western Canada showing productivity inferences.](https://ss.rapidrecap.app/screens/hioZNUbzGUQ/00-07-32.jpg)