Meet: Dr Lorenzo Nava

Lorenzo Nava is a Senior AI+ Fellow at King's College London. These interdisciplinary fellowships are central to King’s £18 million strategic investment in academic excellence and demonstrate King's commitment to transforming AI research and innovation across all disciplines.

He grew up in a mountain village in northwest Italy, where landslides were simply part of how life worked. Roads got blocked and people found another way around, the same way they worked around rain or snow. He went on to study earth sciences, and during a master’s degree at the University of Florence a conversation with his future supervisor changed his direction. The professor showed him AI being used to detect landslides from radar satellite images. "I immediately got passionate about it," he says.

Earth observations provide different pieces of the puzzle. Radar satellites can see through clouds, optical satellites capture surface changes in remarkable detail, and other measurements reveal how the ground is moving over time. Lorenzo's research brings these complementary observations together with AI and physical models to understand how hazardous processes evolve and whether they can be anticipated before they become disasters. "The advantage of AI," Lorenzo explains, "is that it allows us to combine different Earth observations and relate them to the physical processes driving natural hazards."

Detection is only the beginning. The broader challenge is understanding how hazardous processes evolve, whether they are approaching failure, and how they might unfold. To answer these questions, Lorenzo combines Earth observations, AI and physics-based models. Conventional physics simulations can model these processes, but they're slow, sometimes taking hours to run a single scenario. A trained neural network can emulate these physical processes in seconds. That speed means you can hand something useful to disaster managers before conditions change.

In April 2024, a major earthquake struck Taiwan and triggered thousands of landslides across the island. The sky was completely overcast. Lorenzo's team had produced AI-generated landslide maps for the affected area. With no optical satellite imagery available for validation, they decided to release them openly rather than wait. Subsequent observations confirmed that the mapped landslides were accurate. Experiences like this reinforced Lorenzo's interest in building tools that can support decisions when information is incomplete and time is limited.

Lorenzo makes all of his tools open source, because the communities most exposed to mountain hazards are often in low-income countries, where the agencies responsible for disaster response have no budget for commercial software. He builds his tools to run on free cloud platforms. Getting the model right is only half the work. The other half is making sure it reaches the people who need it, which means being present at the table where decisions about standards and practice get made. He joined a United Nations working group on AI for natural disasters as a master's student, drawn in by curiosity, and has since helped lead the development of educational guidelines on how AI tools should be designed, explained and used in the field.

King's drew him in because of what it has gathered in one place. Computational researchers and geospatial scientists work alongside policy specialists who understand how information reaches the people making decisions. For Lorenzo, forecasting natural hazards is not only a scientific challenge but also a decision-making challenge. A model built without any understanding of how it will be received, communicated or acted upon is a model built in a vacuum. "King‘s has a bit of everything," he says, "and if you really talk to people, you get many angles of the same problem that otherwise you wouldn’t in many other places."

This August, Lorenzo travelled to Tibet for fieldwork. The field trips do something the models alone cannot: they put him back in the landscape, where the problem has a smell and a texture and a scale that no satellite image fully captures. Somewhere underneath the code, the satellite imagery and the UN working groups, the mountains he grew up with are still there.