The Kakhovka Dam in Ukraine once towered nearly 30 meters high, stretching more than three kilometers across the Dnieper River. It created one of the nation’s largest reservoirs, storing billions of cubic meters of water that nourished Kherson’s farmlands and cooled the Zaporizhzhia nuclear plant.
When it collapsed in June 2023, destruction rippled through the region. Floods ravaged towns downstream, while fields upstream turned dry and barren. Canals ran empty, and new forests of willow and poplar began rising from the mud. Toxic sediments reached the Black Sea, fueling massive algal blooms that choked marine life.
Amid this chaos, researcher Sarah Hartman from the University of California, Berkeley, sought to understand the crisis through technology. Her team wondered whether AI could reveal how agriculture adapts when nature and infrastructure fail.
With ground access impossible, Hartman’s group relied on satellites and AI models to examine crop health before and after the collapse. Using public data from Landsat and private imagery from Planet Labs, they trained unsupervised machine learning algorithms to detect changes in vegetation. Decision-tree analysis then separated active farmlands from abandoned plots and emerging regrowth.
This study, finalized after Hartman joined Australia’s CSIRO as a postdoctoral fellow, exposed what she called a “pattern of loss and endurance.” She explained that the research uncovered zones of failure and areas of revival—evidence that nature, aided by data, can reveal its own story.
This approach illustrates the core of AI’s value in agriculture—its ability to observe what human eyes cannot and interpret change from a distance.
The Power to See Beyond the Surface
AI combined with satellite imagery can now identify drying irrigation systems, newly built greenhouses, and crops that deviate from expected growth. Even in regions with missing or corrupted data, AI can reconstruct decades of agricultural history. Hartman cautions, however, that the accuracy of AI depends on the quality of its data sources.
She noted that artificial intelligence mirrors the information it consumes. Poor, outdated, or biased inputs produce misleading insights. When trained with current, local, and diverse datasets, however, AI can deliver precise and practical results for agriculture.
This idea inspired what she and her peers call “responsible, application-driven AI,” or RAD-AI. The philosophy emphasizes building tools rooted in context—technology designed by and for those who work the land, guided by reliable information.
Building Trust in a Changing Sector
Hartman describes RAD-AI as an approach grounded in collaboration and real-world purpose. It involves designing systems with the participation of agronomists, growers, and agricultural technology developers to ensure AI reflects practical needs.
True trust, she believes, arises from inclusivity. Integrating Indigenous knowledge systems, honoring data sovereignty, and prioritizing participatory design allow AI to align with the complexity of Australian farming. She stresses that such collaboration ensures that AI serves both technical advancement and social fairness.
Australia’s agriculture faces unrelenting pressures: extreme weather, workforce shortages, productivity demands, and global market volatility. Without thoughtful investment, the sector risks missing AI’s benefits at a pivotal time. Hartman warns that innovation must be intentional, not rushed, to ensure that AI strengthens rather than destabilizes agriculture.
As progress accelerates, she insists that responsibility must be embedded from the start. Preventing misuse, bias, and inequity isn’t just about ethics—it’s about ensuring that AI becomes a trusted partner in the field.
Innovation Taking Root
Across Australia, AI in agriculture is already yielding results. On vast cattle stations, drones equipped with predictive algorithms are being tested for automated mustering. Created by the local firm SkyKelpie, these drones anticipate herd behavior and optimize flight paths, reducing costs and emissions compared to helicopters or vehicles.
Meanwhile, robotics innovators like Swarmfarm are developing autonomous tractors capable of performing multiple tasks through integrated AI systems. Although challenges in scaling and reliability remain, these machines mark a turning point in how farms operate.
Weather prediction is another frontier. Google’s GraphCast and Microsoft’s Aurora are providing ultra-accurate forecasts that guide decisions on planting, irrigation, and drought preparation. Generative AI is also beginning to reshape agriculture—tools such as Microsoft’s AgPilot and other emerging platforms offer tailored recommendations based on farm-specific data.
Each of these innovations showcases AI’s flexibility across livestock management, logistics, and forecasting. Yet they also expose a crucial vulnerability. Generative AI, known for producing new content, can sometimes fabricate false but convincing information. In agriculture, such errors could translate into financial and environmental damage.
Preparing For Tomorrow’s Farms
Hartman’s “Trusted AI Agronomist” initiative demonstrates how AI can translate advanced crop science into practical insights. By training a feed-forward neural network to emulate CSIRO’s Agricultural Production Systems Simulator, her model generates forecasts for crop growth under varying conditions. What sets it apart is its inclusion of uncertainty ranges, giving farmers both predictions and confidence levels to inform their decisions.
She observed that most agricultural decisions rely on managing risk rather than certainty. Having insight into prediction confidence helps farmers choose strategies that align with their risk tolerance. Seeing a spectrum of possible outcomes, rather than a single forecast, nurtures both trust and resilience in decision-making.
Under the CSIRO Ag2050 program, Hartman has also identified three key AI breakthroughs shaping future agriculture: integrated data analytics, improved climate forecasting, and generative AI tools. These are not futuristic ideas—they are already transforming how Australia plans its agricultural future.
Dr. Rose Roche, who leads Ag2050, emphasized that AI could redefine decision-making in farming. She explained that boosting productivity amid climate challenges and shifting markets requires adopting tools that enhance adaptability. CSIRO’s research shows how AI can help farmers remain resilient despite uncertainty.
Frank Sperling, co-editor of Ag2050’s special edition of the Farm Policy Journal, highlighted the need to explore diverse futures for Australian agriculture. He noted that facing intertwined global and local risks requires scenario planning to identify opportunities while mitigating harm. Sperling stressed that Hartman’s focus on responsible AI provides the foundation for this balanced approach.
Choosing the Path Forward
Hartman’s experience in Ukraine continues to influence her work. The disaster revealed how AI can uncover both devastation and recovery—an insight now applied to Australian conditions. From AI-guided drones and enhanced weather forecasting to generative chatbots assisting farmers, artificial intelligence is steadily moving from experimental projects to daily agricultural practice.
For Hartman, the lesson is clear: the direction AI takes depends on how humans design and deploy it. The technology can become a tool for sustainability and trust—or a source of new vulnerabilities. Her leadership in this space earned her the Women in AI Asia-Pacific Award for Agribusiness and Rural Development, recognizing her commitment to inclusive innovation.
She reminds fellow researchers that AI should never be treated as a magic solution. When developed responsibly and trained with quality data, however, it becomes a transformative ally for agriculture—one capable of strengthening global food systems and helping farmers adapt to a changing planet.
