Introduction

Tropical cyclones are one of the more frequent natural disasters globally. The observed record is limited, so it is difficult to evaluate what the ‘worst possible’ cyclone would look like. Based on these historic records, we can statistically ‘extrapolate’ the record and use a stochastic model to generate plausible, synthetic TC events.

The Tropical Cyclone Risk Model is a stochastic tropical cyclone model developed by Geoscience Australia for estimating the wind hazard from tropical cyclones.

Due to the relatively short record of quality-controlled, consistent tropical cyclone observations, it is difficult to estimate average recurrence interval wind speeds due to tropical cyclones. To overcome the restriction of observed data, TCRM uses an autoregressive model to generate thousands of years of events that are statistically similar to the historical record. To translate these events to estimated wind speeds, TCRM applies a parametric windfield and boundary layer model to each event. Finally an extreme value distribution is fitted to the aggregated windfields at each grid point in the model domain to provide ARI wind speed estimates.

TCRM was borne out of a desire to explore the epistemic uncertainty around parameterisations used in tropical cyclone hazard models. There are a wide range of idealised radial profiles and boundary layer models available in the literature, and the choice of profile or boundary layer model will influence the resulting hazard (and risk) produced the model. TCRM allows users to run with a range of profiles or boundary layer models and evaluate the resulting differences, within a single modelling system. Users can also implement their own models into the code base (which can then be shared with the user community).

Model outline

TCRM has 6 main modules to generate hazard information: DataProcess, StatInterface, TrackGenerator, wind, hazard and database. Supporting Utilities and plotting routines (PlotInterface) are included.

The DataProcess module reads the input track database and extracts parameters from the data, namely intensity and location information, speed, bearings and genesis locations. The data are stored in text files to allow users to examine the data in another analysis tool. There are also a number of plotting routines to display some of the basic relationships in the data.

StatInterface uses the processed files generated by DataProcess to evaluate the statistical relations between the parameters, including the means, variance and autocorrelations on a coarse grid across the model domain. PDFs of the initial values of parameters (initial speed, bearing and intensity) and genesis probability are also calculated.

TrackGenerator is the stochastic engine of the model. This module samples from the distributions of genesis location and initial parameter values to start a new TC event, then steps forward in time using the autoregressive nature of TC behaviour. This allows users to generate thousands of random events that share the same statistical properties as the input track dataset.

The wind module calculates the maximum wind speed around each of the synthetic TC events generated by the TrackGenerator. A 2-dimensional parametric profile is used to calculate the wind field to enable large numbers of events to be processed efficiently, but at high spatial resolution (e.g. a grid spacing of around 2 km). Users can select from a number of radial profiles and combine with one of three boundary layer models to incorporate asymmetries associated with forward motion of the TC vortex.

hazard uses the wind fields calculated in the wind module to calculate average recurrence interval (ARI) wind speeds. There are a number of approaches that can be used:

#. empirical recurrence intervals
#. fitting a generalised pareto distribution (GPD)
#. fitting a generalised extreme value distribution (GEV)

The first option calculates ARI wind speeds up to the limit of the number of simulated years. The second and third options allow for calculating ARI wind speeds beyond the limit of the number of simulated years.

The database module builds a SQLite database that holds information about the simulation. The wind speeds associated with all events in the simulation catalouge are extracted for locations in the model domain, along with recurrence interval wind speeds. Details (maximum wind, minimun central pressure, lifetime, etc.) of each storm are recorded, as well as the closest distance of approach for storms to locations.

Software

The software is made available as an open-source package. Users can add new components to the model and are encouraged to submit them back to the project. Where possible, the code has been modularised to ease the process of adding new methods (such as radial profiles or boundary layer models).

TCRM can be downloaded from our GitHub repository: http://github.com/GeoscienceAustralia/tcrm

Features

  • Multi-platform: TCRM can run on desktop machines through to massively-parallel systems (tested on Windows XP/Vista/7, *NIX);

  • Multiple options for wind field & boundary layer models: A number of radial profiles and simple boundary layer models have been included to allow users to test sensitivity to these options.

  • Globally applicable: Users can set up a domain in any TC basin in the globe. The model is not tuned to any one region of the globe. Rather, the model is designed to draw sufficient information from best-track archives or TC databases;

  • Evaluation metrics: Offers capability to run objective evaluation of track model metrics (e.g. landfall rates);

  • Single scenarios: Users can run a single TC event (e.g. using a b-deck format track file) at high temporal resolution and extract time series data at chosen locations;

References

1

Hosking, J. R. M. (1990): L-moments: Analysis and Estimation of Distributions using Linear Combinations of Order Statistics. Journal of the Royal Statistical Society, 52, 105–124.