Soils store carbon, sustain ecosystems, and underpin global food and water systems. A new Frontiers in Science paper details how AI tools can help us adapt soils—and the systems they nurture—to a changing climate.
Soil science affects how we respond to the world's most urgent challenges, from food security to climate change. Yet soil systems, affected by climate, weather patterns, and agricultural practices, are highly complex and difficult to predict, especially as climate pressures and land use intensify. The authors say the field needs tools that can help researchers make sense of that complexity.
Soil science currently uses machine learning approaches such as digital soil mapping and spectroscopy. AI systems could enhance this by creating digital soil twins with data from sensors, enhancing soil microbiome monitoring, and trialing climate adaptation strategies in computer models before testing them in the field for faster results.
The paper outlines how AI tools can accelerate soil science by speeding up early-stage work, improving predictions to support decisions on land-use, carbon, and climate adaptation, handling complex data, and freeing scientists to focus on questions that require expert judgment.
Senior author Prof Alex McBratney from The University of Sydney, Australia, said: "In partnership with experts, AI could help us better match the complexity and ever-changing nature of soil ecosystems. Unlike current machine learning tools that focus on isolated tasks, these systems can mimic scientific collaboration to a highly sophisticated degree—combining reasoning, planning, and interdisciplinary insight to support researchers and drive significant progress.
"Perception of the vital importance of soil in planetary functioning is increasing, and soil science will continue to grow and flourish under scientist-led AI."
The earth that sustains us
To illustrate such a tool, the research team tasked a multi-agent AI system with reviewing relevant scientific literature and generating ideas about how soils store carbon and what controls their storage limits.
The agents successfully generated five hypotheses, including climate influence, saturation thresholds, biological and chemical controls, interdisciplinary feedback, and management strategies.
Each hypothesis was then evaluated through expert opinion and simulated peer review. The system successfully mimicked key parts of the scientific process, with outputs beyond what's currently being used that strongly align with expert research.
Lead author Prof Budiman Minasny, also from The University of Sydney, said: "Our findings indicate the opportunity for AI to accelerate soil research—the understanding of which can benefit our food and climate systems. Improving our understanding of soils could support more sustainable agriculture, better soil management, and stronger climate adaptation by helping land managers detect nutrient loss, water stress, compaction, and erosion earlier.
"We assessed the system's ability to perform perceptual processing, strategic planning, and scientific reasoning. Our findings highlight the promise that multi-agent AI systems hold, with important global implications for soil—a precious but perhaps undervalued resource."
Artificial intelligence, human expertise
Despite AI's potential, challenges remain, particularly around data quality, model transparency, trust, and maintaining foundational scientific knowledge. The paper also points to further considerations around computational cost and the ethical dimensions of such tools.
Co-author Dr Mercedes Román Dobarco from the Basque Institute for Agricultural Research and Development (NEIKER), Spain said: "While the use cases are clearly persuasive, and though AI can emulate some aspects of expert reasoning, it cannot replace the contextual judgment, creativity, and critical interpretation scientists bring to research. AI agents also pose challenges around data quality, interpretability, creativity, and dataset bias, particularly without human oversight and domain expertise.
"Given these limitations, we should treat AI as an augmentative tool that enhances, not replaces, human scientific work."
The paper also underscores AI's ability to accelerate both 'fast' and 'slow' science. For example, by automating time-intensive preparatory tasks such as literature review and scenario development, AI could free soil researchers' time to focus on deeper foundational understanding and field work while maintaining scientific rigor and accountability.
Prof McBratney said: "Soils are among our planet's most vital and existential resources. To fully benefit from AI-enhanced soil science, we must embrace interdisciplinary collaboration, ensure equitable access to AI tools, and thoughtfully address the ethical challenges we have outlined.
"By bridging digital innovation with real-world application, as well as non-negotiable human oversight, AI can supercharge soil science—but only if human knowledge keeps pace. Striking that balance can help us unlock new levels of stewardship and security for soil."
