Abstract:
|
We discuss building a fast statistical surrogate model (emulator) for an expensive storm surge simulator used for understanding the effects of hurricanes on coastal communities. We use a Bayesian approach to multivariate adaptive regression splines (MARS) to model basis coefficients of a principal component basis decomposition of the simulator’s spatial output. We show how we can obtain an exact functional ANOVA decomposition for this model, and how we use that for sensitivity analysis. That sensitivity analysis allows us to understand, for instance, the effect of uncertainty in sea level rise on storm surge.
|