As AI is increasingly applied to fields studying how climate change affects the conditions beneath the Earth's surface, Faculty of Environmental and Urban Change (EUC) postdoctoral fellow Adedibu Sunny Akingboye is helping ensure the technology is used responsibly and reliably.
Scientists in environmental geophysics and geotechnical engineering study underground conditions that influence groundwater resources, landslide risks and the stability of critical infrastructure.
As climate change increases flooding and other extreme weather events, understanding what goes on beneath the Earth's surface is becoming more important.
To do that, researchers collect large volumes of information using seismic surveys, ground-penetrating radar, environmental sensors, satellite observations and borehole testing. Each source provides different information. Integrating and interpreting those datasets, however, can be a complex and time-consuming challenge.
"We now collect large amounts of geophysical, geotechnical and environmental data, but these datasets are often analyzed separately," says Akingboye, whose research in EUC focuses on climate-related hazards and infrastructure resilience.
That is one reason AI is steadily being applied in the field, he says. The technology improves how researchers merge datasets, analyze results and identify patterns. This provides a clearer understanding of subsurface conditions to assess landslide risks, monitor infrastructure and evaluate groundwater systems.
With AI expanding into environmental geophysics and geotechnical engineering, Akingboye paused to assess how the technology was being used.
"A question I had been thinking about for some time was, with the rapid growth of artificial intelligence, are we actually using it in the most meaningful way in environmental geophysics and geotechnical engineering?" he says.
That question became the starting point for a new co-authored by Akingboye and Professor Adeyemi Oludapo Olusola, who supervises his work in EUC. The paper examined hundreds of recent studies to assess how AI is being used in environmental geophysics and geotechnical engineering, evaluating both the progress being made and the challenges that remain.
鈥淭he sheer range of AI applications was surprising. We found AI being used for subsurface imaging, groundwater assessment, soil-rock characterization, landslide and slope instability prediction, and infrastructure monitoring, among many other applications,鈥 says Akingboye.
The review shows AI improves subsurface research by integrating multiple data streams to deliver detailed underground maps and efficient estimates of soil, rock and groundwater properties.
For Akingboye, documenting AI's growth was only part of the story; it was also about whether the technology could be trusted in environmental and engineering decision-making.
鈥淎 high-performing model is not automatically a useful or trustworthy one. We wanted to ask broader questions: Does the result agree with what we know about the subsurface? Can we explain it? Can we account for the uncertainty?鈥 says Akingboye.
The answer, the review found, is not always clear-cut. Across studies from around the world, researchers encountered recurring challenges including limited data, uncertainty, difficulty applying models across different geological settings and understanding how AI systems arrived at their conclusions.
Akingboye notes the challenges identified in the review are not simply technical problems. Inaccurate or poorly understood predictions can influence decisions about environmental hazards and critical infrastructure, making reliability as important as performance.
鈥淭hese are important gaps because a wrong interpretation can have serious consequences for environmental and engineering decision-making,鈥 says Akingboye. 鈥淭he impacts can be catastrophic and are often felt most severely by vulnerable and marginalized communities with fewer resources to prepare for, respond to and recover from environmental or infrastructure failures.鈥
As well as identifying challenges, the authors also sought to outline a path forward. They advocate for future AI-driven tools that merge diverse environmental and geological datasets, account for uncertainty and respect established scientific laws governing geological systems.
To advance AI tools, the authors recommend developing systems with more robust prediction rationale and stronger validation benchmarks. They also call for closer collaboration between researchers, industry and policymakers to ensure future models combine sensor and monitoring data to provide near real-time pictures of changing underground conditions.
Ultimately, the review argues that the next step lies in engineering complex AI models capable responding beyond the datasets and conditions on which they were trained.
鈥淭here is still a significant gap between a model that performs well in a research study and one that can be trusted and applied in real-world practice,鈥 says Akingboye. 鈥淚 think closing that gap, while ensuring that these technologies ultimately support safer, more resilient and equitable communities, is one of the most important areas for future work.鈥