Exploring the sensitivity of probabilistic surge estimates to forecast errors

Statistical predictions of storm surge are critical for guiding evacuation and emergency response/preparedness decisions during landfalling storms. The probabilistic characteristics these formulated by utilizing historical forecast errors to quantify relevant uncertainties in the National Hurricane Center advisories. This ultimately leads description probability distributions quantifying deviation from nominal advisory four different features: intensity, size, cross-track variability along-track variability. Propagation uncertainty features, serving as input a numerical model calculating surge, definition statistical estimates. work investigates application variance-based global sensitivity analysis (GSA), quantified through estimation Sobol' indices, explore importance peak predictions. GSA can assist better understanding impact typical storms, also offer important insights specific storm, regarding that influence across its advisories, comes closer landfall. An efficient implementation is presented here address two key challenges problem: (i) need perform multi-dimensional output, corresponding multiple locations within geographic domain interest will be affected (ii) restriction use only small number hydrodynamic simulations, since associated computational burden such simulations significant. For addressing challenges, dimensionality reduction Principal Component Analysis (PCA) probability-based variance conditional expectations combined provide necessary efficiency proposed framework. development aggregated indices entire discussed, incorporating each separate location (within this domain) using weighting. formulation compared with an alternative, computationally efficient, importance, based on readily available PCA information. A demonstration framework’s utility considering storms (using Weather Service advisories past events) provided, establishing comparisons them storm.