Background

When wildfire reaches a city, it stops being a wildfire problem. The January 2025 Los Angeles fires and London’s fires of 19 July 2022, its busiest day since 1945, overwhelmed two of the best-prepared fire services in the world because the hazard, once inside the urban fabric, is governed by buildings, materials and street layout rather than vegetation. In August 2026 a wildfire Emergency Alert reached phones across England and Wales: the hazard is no longer distant. Urban climate science can now simulate the local weather that drives such fires (wind, heat and dryness at building scale) and insurers can price the losses. But between the two sits a poorly understood chain: how uncertainty in climate-driven hazard propagates through a city’s physical and institutional response, and which adaptations actually contain it. This is now an open question in climate risk science, with immediate consequences for adaptation policy, fire-service planning and the insurability of entire districts.

PhD project

The January 2025 LA fires showed that even well-protected buildings burn when their neighbourhood does not adapt: protection works like immunity, at community level. But we cannot yet say which patterns of adaptation work, how confident we can be in any prediction, or precisely when insurers withdraw. This PhD tackles that uncertainty directly. Starting from probabilistic wildfire hazard supplied by Climate X (stochastic burn-probability at 30 m resolution, present and future climates), you will build a fast scenario engine coupling urban climate modelling (SUEWS) with building-by-building fire spread simulation, and run thousands of weather scenarios (wind, temperature, humidity) crossed with adaptation scenarios (which buildings are hardened, in which spatial patterns). You will quantify where uncertainty grows and where it shrinks along the chain from hazard to loss; compare traditional heuristics, machine-learning and knowledge-graph decision rules inside the same fire-spread model as a novel probe of structural uncertainty; and test whether today’s best adaptation measures survive future climates. A capstone formalises community “fire immunity” as a percolation-style threshold whose distribution, not single value, is the decision-relevant quantity. The project connects all three UNRISK themes: climate science (urban microclimate and hazard), data science (ensembles, ML, uncertainty quantification) and decisions (adaptation, insurance). Ortec Finance (climate risk economics; industrial supervision agreed in principle) links results to loss and insurability, with access to fire-service practitioners in London; Climate X provides hazard data and model insight, and both partners offer placement opportunities. Outcomes: an open scenario-engine framework, and evidence for cities, fire services and insurers on which adaptations reduce not just risk but uncertainty about risk.

Applicant Profile

Students with a strong background in physics, mathematics, engineering, computer science or a quantitative environmental science who want to apply modelling, simulation and machine learning to a problem where the answers will change real decisions: how cities adapt, how fire services plan, and whether homes remain insurable. An active interest in artificial intelligence, including AI agents both as research tools and as representations of human behaviour, is welcome. The common thread is curiosity about uncertainty itself, not just prediction, and the imagination to ask what a city could do differently; prior fire science knowledge is not expected.

Other Information

Ting Sun, UCL profile (urban climate modelling, SUEWS): https://profiles.ucl.ac.uk/88606-ting-sun
Climate X wildfire model methodology: https://www.climate-x.com/articles/press-releases/climate-x-european-wildfire-analysis
Zhao (2011), cellular-automata simulation of urban fire spread, Fire Technology: https://link.springer.com/article/10.1007/s10694-010-0187-4