The powerful impact of tropical cyclones (TCs) disrupts societies in many coastal regions in the tropics and subtropics. For example, Hurricane Ian (2022) left a trail of destruction as it crossed western Cuba with sustained winds of more than 200 km/h and continued towards the US coast as a Category 4 storm on the Saffir-Simpson scale (1-5). More than 100 people died as Hurricane Ian struck Florida, and millions lost access to electricity in the hours after the storm passed. The storm will likely reach the top of the list of the costliest TCs ever to hit the state, with estimated insured losses of around 50 billion USD. Moreover, the total loss will likely increase further since most of the destruction was due to flooding from the massive storm surges and river flooding that Ian produced with its torrential rain. This recent hurricane shows too well that supporting risk mitigation efforts and increasing societal resilience towards such events with reliable TC risk assessment is crucial. Such assessments, however, are complicated as reliable TC records are temporally and spatially limited.
Various researchers and practitioners in the (re-)insurance industry have recognized this gap and found different modelling solutions to generate larger datasets of TCs. Simulating hurricane risk in the US has been a research endeavor since the 2000s. Early practices, particularly in industry, build on simple statistical analyses of historical events. In 2006, Kerry Emanuel pioneered the physics-based, dynamical approach of synthetic TC modelling. His work inspired other research groups and model developers of the insurance industry to build and improve their own synthetic TC models. By now, a handful of global TC models are available in academia. They are all motivated and appraised with descriptions like: "The STORM dataset can be used for TC hazard assessments and risk modeling in TC-prone regions." or "we develop and assess a complete statistical-dynamical downscaling TC hazard model."
These statements buzz with keywords like "hazard", "risk assessment", "model", "hurricane", and "TC". Rightly so because the models have all been evaluated and used in various contexts to study TC climatology or perform risk assessment in specific contexts. But they have never been directly compared as input hazard datasets in catastrophe models for TC risk assessment and loss estimation. Such catastrophe models are crucial to compute risk and quantify socio-economic impacts by integrating hazard, exposure, and vulnerability.
Hence, this is where this study comes in. We hypothesize that the models which predict TC climatology may not cover the full range of important metrics and views in TC risk assessment and loss estimation. Therefore, we present the first intercomparison of four global-scale synthetic TC datasets in the impact space, comparing various risk and impact metrics. We use the open-source, peer-reviewed CLIMADA (CLIMate ADAptation) platform to simulate direct economic damage in the form of impact on the built environment from a given TC hazard set. After assessing these models at an impact- and risk-level, we use our results to link some of the intermodel differences to key TC model characteristics and provide guidelines for other researchers to determine the applicability of each dataset depending on the research objective.
Weather and climate risk modelling is a relatively young field in academia. Historically, natural catastrophe modelling has been exclusive to the (re-)insurance industry, and academia has focused on the purely physical aspects, the hazard part. The insurance industry mainly uses catastrophe models to set premiums according to the estimated annual damage and to reserve capital for events exceeding the annual probability of 1%. Now the field has started to gain ground in academia, and we have the tools, expertise and curiosity to contrast TC risk modelling practices. In contrast to industry, we have no financial incentive, which would limit us to (data-)rich regions. Besides, industry models are mostly kept behind closed doors, while models or datasets for research are open-access. These two differences are crucial. Generally, we use our models over a wide geographical range for various research questions and foster open collaborations to better understand what drives present TC risk.
Risk and impact modelling in academia is essential to push the frontiers of current TC risk assessments. In an academic setting, we get to discuss different modelling approaches. We question what we view as ground truth and which targets we choose for model validation. We share knowledge and sometimes challenge historically established practices. The industry is very much oriented towards meeting targets prescribed by regulators. These often root in historical observations. However, the historical record may be a limited guide for today's TC risk because the historical record is temporally and spatially limited and sometimes of poor quality. Moreover, the assumption of stationary statistics is flawed for weather-related hazards such as TCs, which are already today influenced by climate change. The last two points are particularly crucial for high-impact events because these tail events are rare by definition. In academia, we have the opportunity to address these challenges and guide the next steps toward robust and reliable TC risk assessment. Finally, our contribution to the emerging field of weather and climate risks in academia may also help this discipline strengthen its scientific position. In the long run, it would be exciting and powerful to join forces with the industry and shape the future of our fields together.
The results, findings and guidance we synthesized in our paper are the product of a curiosity-driven, open and stimulating collaboration between research groups from MIT, Columbia, PIK, Vrije Universiteit Amsterdam, and ETH Zurich. We all felt the need for this model intercomparison to better understand how we model and communicate current global TC risk and loss estimates. Understanding today's TC risk is the foundation for effective decision-support in the space of adaptation measures, mitigation strategies, (re-)insurance practices and physical risk disclosure. As a group, we share a vast pool of knowledge and it was about time we brought this together in a consistent fashion. The different TC risk metrics and the discussion of links to key model characteristics we present in this paper yield an improved understanding of TC impact assessments. Knowing today's risk is ultimately relevant to assess how climate change and socio-economic development drive TC risk in the future.
Please sign in or register for FREE
If you are a registered user on Research Communities by Springer Nature, please sign in