Long-term regional simulation of tropical cyclones using a Generalized Stochastic Empirical Storm Model

Abstract

In coastal areas, Tropical Cyclones (TCs) are one of the greatest threats to humanity. Unfortunately, current risk reduction measures are not completely successful in lessening TC's consequences due to the remaining uncertainties in the estimates of key parameters, on which the designs of these measures rely. Because reliable observations of TCs, having affected many regions, are restricted to quite a small number, it is not feasible to derive accurate TC statistics solely based on historical records, without producing large errors. This research presents a comprehensive methodology to effectively overcome the observed data scarcity problem. TCs are stochastically simulated over a period of thousands of years by a numerical model, which results in a long-term database of synthetic TCs, with specifications of the central track and intensity as well as the wind field at each time step. Because TC evolution is heavily dependent on local conditions, the simulation is carried out at a regional scale to maintain relative homogeneity within both the input and outcome, and to reduce computational demand. Since the model has a generalized theoretical framework and contains the worldwide historical weather data, it can be applied to any case study. Once users define the Area Of Interest (AOI), a stepwise calibration procedure is automatically performed by a computer program to achieve the most suitable approach and to specifically determine every single detail of the model for this user-defined AOI. The method is validated though comparisons of observed and simulated TC statistics in the AOI. For a case study of Vietnam in the Western North Pacific, this evaluation proves the model's ability to reproduce the actual TC characteristics and to generate a useable long-term database with an acceptable accuracy for practical projects. Finally, the wind speed maps and the annual exceedance probability maps are provided as possible applications of the model results.