Polynomial chaos analysis of hurricane wind parameters in the Gulf of Mexico from 2008-08-28 to 2008-08-31

The focus of this project is uncertainty quantification (UQ) of storm surge impacts due to uncertainty in wind predictions, focusing on hurricanes in the Gulf of Mexico, with a specific focus on the New Orleans region. The original input data, which is the best track of Hurricane Gustav, is obtained from NOAA. We use a perturbation technique to generate a series of ensembles as model inputs for our UQ analysis. Then a hydrological shallow water model ADCIRC has been used to obtain model outputs (i.e. water elevation time series on specified stations & maximum water surface elevation over time), corresponding to the ensemble inputs. The dataset has been organized in the following way: 2 scenarios represent different sources of uncertainty input. In the first scenario, the sources of uncertainty stem from the track (parametrized by 2 random variables) and the maximum wind speed (parametrized by 1 random variable) of the hurricane. In the second scenario, the sources of uncertainty come from the maximum wind speed (parametrized by 1 random variable) and the Manning’s n coefficient (parametrized by 2 random variables). For each scenario, we split the dataset (150 ensembles in total) into two parts: training (100 ensembles) and validating (50 ensembles). For both training and validating set, we provide inputs as well as model output data, respectively. The training set is used to build the (polynomial chaos) surrogate model of the maximum water surface elevation. The validation set, independent from the training set, allows us to validate the accuracy of the surrogate before conducting the statistical analysis.