Physics-based Machine-learning Approach for Tropical Cyclone Risk Analysis

Web Published:

Generating synthetic tropical cyclones is a general approach for assessing tropical cyclone hazards and risk applied by the government, wind engineering consultant labs, catastrophic modeling companies, and insurance companies. However, most methods currently in use can generate only storms under the observed climate conditions, as these models are simple statistical models that do not depend on physical parameters or climate variables. As climate change effects are getting more and more attention, there is an unmet need for a model that can generate tropical cyclones under future climate conditions to estimate future hazards and risk.


Researchers at Princeton University have invented a new climate-dependent probabilistic tropical cyclone (TC) model to generate synthetic tropical cyclones (including their formation, track, and intensity) under observed or projected future climate conditions. This model utilizes advanced machine-learning techniques to establish comprehensive statistical relationship between storm features and most-relevant physical parameters. Given the statistics of the physical parameters in the current and future projected climates, the model can generate synthetic tropical cyclones under different climate conditions. The model consists of three components—a hierarchical Poisson genesis model, an analog-wind track model, and a Markov intensity model. All three model components are dependent on environmental variables including the potential intensity and environmental wind and thus respond to the change of the climate.



• When coupled with TC hazard models, can better quantify TC-related wind, surge, and rainfall hazards under various climate conditions.

• Can be used by ASCE building code development, FEMA flood mapping, NIST, wind engineering consultant labs, catastrophic modeling companies, and insurance companies to estimate future TC risks.



• Better incorporate the estimated impact of climate change compared to present models.

• The model captures observed TC climatology well and is able to reproduce various TC statistics, including the annual genesis rate, interannual genesis variability, and landfall frequency and intensity.

• The model can generate a reasonable fraction of storms that undergo rapid intensification and

can estimate well the distribution of the lifetime maximum intensity.


Stage of Development

The model has been verified against observations using out-of-sample testing. The testing shows that the probabilistic model is able to reproduce various tropical cyclone statistics, such as the annual basin-wide genesis rate, interannual genesis variability, the distribution of lifetime maximum intensity, and landfall frequency and intensity.




Ning Lin is an Associate Professor of Civil and Environmental Engineering at Princeton University. She obtained her PhD in Civil and Environmental Engineering in 2010 from Princeton University, and her BS in Civil Engineering from Huazhong University of Science and Technology, China in 2002. She conducted research in the Department of Earth, Atmospheric and Planetary Sciences at MIT as a NOAA Climate and Global Change Postdoctoral Fellow from 2010-2012. Professor Lin is interested in Natural Hazards and Risk Assessment, Stochastic Modeling, Wind Engineering, Coastal Engineering, Climate Change Impact and Adaptation, and Built Environment and Sustainability.  Specifically, her current research integrates science, engineering, and policy to study tropical cyclones and associated weather extremes (e.g., strong winds, heavy rainfall, and storm surge), how they change with climate, and how their impact on society can be mitigated. She is a recipient of the Lloyd’s Science of Risk Prize in 2014, and the Howard B. Wentz Jr. Faculty Award at Princeton and NSF CAREER Award in 2016.


Renzhi Jing is a Ph.D. candidate in the Civil and Environmental Engineering Department at Princeton University pursuing a certificate in Statistics and Machine Learning. She obtained her BS in Physics/ Atmospheric Science and BA in Chinese from Peking University in 2014. She currently works on hurricane hazard risk assessment, where she develops statistical and machine learning models to generate synthetic hurricanes under different climate conditions. Her research interests include Climate Informatics, Natural Hazards, Risk Analysis and Climate Change Adaptation.


Intellectual Property Status

Patent protection is pending.

Princeton University is currently looking for Industry collaborators to further develop and commercialize this technology.



Prabhpreet Gill

Princeton University Office of Technology Licensing • (609) 258-3653•



Patent Information:
Computers and Software
For Information, Contact:
Prabhpreet Gill
Licensing Associate
Princeton University
Ning Lin
Renzhi Jing