| Summary: | Tropical cyclones (TC) in the western North Pacific (WNP) pose a serious threat to the coastal regions of Eastern Asia when they make landfall. The limited amount of observational data and the high computational cost of running TC-permitting dynamical models indicate a need for statistical models that can simulate large ensembles of TCs in order to cover the full range of possible activity that results from a given climate change. I construct and apply a statistical track model from the 1945-2007 observed "best tracks" in the IBTrACS database for the WNP. The lifecycle components--genesis, track propagation, and death--of each simulated track is determined stochastically based on the statistics of historical occurrences. The length scale that dictates what historical data to consider as "local" for each lifecycle component is calculated objectively through optimization. Overall, I demonstrate how a statistical model can be used as a tool to translate climate-induced changes in TC activity into implications for risk.
In contrast to other studies, I show that the El Niño/Southern Oscillation (ENSO) has an effect on track propagation separate from the genesis effect. The ENSO effect on genesis results in higher landfall rates during La Niña years due to the shift in genesis location to the northeast. The effect on tracks is more geographically and seasonally varied due to local changes in the mid-level winds. I use local regression techniques to capture the relationship between ENSO, cyclogenesis, and track propagation. Stationary climate simulations are run for extreme ENSO states in order to better understand changes in TC activity and their implication for regional landfall. Additionally, Several diagnostics are performed on model realizations of the historical period, confirming the model's ability to capture the geographical distribution and interannual variability of observed TCs.
Lastly, as a step to connect synthetic TC track simulations to economic damage risk assessment, I use a Damage Index and total damage data for U.S. landfalling hurricanes and fit generalized Pareto distributions (GPD) to them. The Damage Index uniquely separates out the effects of the physical damage capacity of a TC and the local economic vulnerability of a coastal region. GPD fits are also performed using covariates in the scale parameter, where bathymetric slope and landfall intensity are found to be useful covariates for the Damage Index. Using the Damage Index with covariates model, two examples are shown of assessing damage risk for different climates. The first takes landfall data input from a statistical-deterministic TC model downscaled from GFDL and ECHAM model current and future climates. The second takes landfall data from a fully statistical track model with different values of relative sea surface temperature given as a statistical predictor.
|
|---|
