What would it mean for global risk if a hurricane made landfall in the southeast US and a typhoon hit Japan in the same year? At the CGFI Hackathon 2025, we joined researchers and industry partners to explore the global characteristics of tropical cyclones and what they imply for the re/insurance sector. Here, JBA scientists Dr John Ashcroft and Dr Alison Poulston, part of the event’s organising team, share reflections from the hackathon and what it means for modelling correlated extremes.
In June this year, we helped host and support the CGFI Hackathon 2025 at the Leeds Innovation Hub – a two-day event co-designed with the Centre for Greening Finance and Investment (CGFI), Aon, the Met Office, the University of Reading, and Imperial College London. We set out to explore a key question: what is the likelihood of extreme tropical cyclones affecting different ocean basins within the same season or insurance cycle — and what does that mean for catastrophe modelling and reinsurance risk?
The specific focus was a scenario in which a hurricane strikes the southeast US in the same TC season as a typhoon makes landfall in Japan – two high-impact events in two major insurance markets. This hasn’t happened in recorded history, but is that a fluke, or are there climate dynamics at play that make it unlikely? And how should we reflect this in risk models?
Participants tackled these questions from a wide range of angles, and we were pleased to contribute both as members of the steering group and judging panel, as well as in a technical capacity, supporting teams on the day.
Why correlated extremes matter
At JBA, our global flood modelling philosophy is built on understanding the links between climate drivers and extreme events. If tropical cyclones in different basins are influenced by shared climate signals – such as the El Niño–Southern Oscillation (ENSO) – then their impacts may not be independent. This matters for how we structure stochastic event sets, simulate joint risk, and ultimately, support clients making decisions about capital, coverage, and resilience.
The hackathon gave us a unique opportunity to see how others approach this challenge – using different data sources, tools, and assumptions – and to test ideas in a fast-paced, collaborative setting.
What the teams explored
Each team took its own approach, but several consistent themes emerged across the room. Some focused on identifying co-occurrence – looking for years in which both basins experienced high TC activity or landfalls. Others explored the role of ENSO and other climate signals, evaluating whether patterns in one basin could be linked to those in another.
A few groups examined trends over time, asking whether any long-term shifts in cyclone activity or landfall rates might point to systemic changes. One team looked at how exposure – such as urban growth – could amplify impacts, even if hazard remains stable.
The winning team focused on event severity, using accumulated cyclone energy (ACE) to explore whether intense seasons in one basin coincided with activity in the other. Their work surfaced subtle but interesting patterns and offered a different perspective on what it means for two events to be “correlated.”
Our reflections and takeaways
We left the event with a renewed appreciation for just how much the outcome of this kind of analysis depends on the data you use. Some teams worked with relatively short observational records, while others used climate model outputs stretching decades or centuries. These choices come with different strengths and limitations — and that’s worth keeping in mind when interpreting the results.
We were particularly interested in solutions utilising the CHAZ framework, which uses global climate model-driven experiments to explore tropical cyclone rates toward the end of the century. Although it wasn’t directly applied during the hackathon, the ideas it raises showed promising signals — and we’re keen to explore how this might inform future model development.
Another takeaway for us was the gap between wind-based event characterisation and flood impacts. Most of the available datasets were focused on storm tracks and wind intensity, which makes sense in a tropical cyclone context – but as flood risk modellers, we know that precipitation and surge are equally critical. Bridging this gap remains a key part of our modelling work.
Looking ahead
The CGFI Hackathon was a fantastic example of what can happen when academic curiosity meets practical challenge. It brought together an enthusiastic, creative group of participants who shared a common interest in improving our understanding of global hazard.
For us, it reinforced the importance of cross-sector collaboration in tackling complex modelling questions. It also reminded us that modelling correlated extremes isn’t just about probability – it’s about asking the right questions, selecting the right tools, and grounding insights in real-world impacts.
We’re looking forward to taking these insights back into our work at JBA, and to continuing the conversation on how to model global risk more realistically – and more usefully – for those who rely on it.