Stochastic Simulation Of Tropical Cyclones For Risk Assessment At One Go: A Multivariate Functional Pca Approach

A multivariate functional principal component analysis approach to the full-track simulation of tropical cyclones (TCs) is developed for risk assessment. Elemental variables of TC along the track necessary for risk assessment, such as center coordinates, maximum wind speed, minimum central pressure and ordinal dates, can be simulated simultaneously at one go, using solely the best-track data with no data supplemented from any other sources. The simulation model is optimally determined by means of the ladle estimator. A TC occurrence model using the Conway-Maxwell-Poisson distribution is proposed as well, by which different dispersion features of annual occurrence can be represented in a unified manner. With the occurrence model, TCs can be simulated on an annual basis. The modeling and simulation process is programmed and fully automated such that little manual intervention is required, which greatly improves the modeling efficiency and reduces the turnaround time, especially when newly available TC data are incorporated periodically into the model. Comprehensive evaluation shows that this approach is capable of generating high-performance synthetic TCs in terms of distributional and extreme value features, which can be used in conjunction with wind field and engineering vulnerability models to estimate economic and insurance losses for governments and insurance/reinsurance industry.Plain Language Summary Tropical cyclones (TCs) are one of the biggest threats to life and property around the world. However, the infrequent nature of catastrophic TCs invalidates the standard actuarial loss estimation approaches. TC risk assessment requires estimation of catastrophic TCs having a very low occurrence probability, or equivalently a very long return period spanning up to thousands of years. Since reliable TC data are available only for recently decades, stochastic modeling and simulation turned out to be an effective approach to achieve more stable TC risk estimates for regions where little or no historical TC records exist. Here we present a novel model for the full-track simulation of TCs for risk assessment, via a machine learning approach called multivariate functional principal component analysis. Using this model, high-performance synthetic TCs can be generated in a fully automated manner such that little manual intervention is required, which greatly improves the modeling efficiency and reduces the turnaround time, especially when newly available TC data are incorporated periodically into the model. These synthetic TCs can be used in conjunction with wind field and engineering vulnerability models to estimate economic and insurance losses for governments and insurance/reinsurance industry.