Data-driven probabilistic scenarios simulation model and visualization for hurricane impact

The 2017 Atlantic hurricane season was horrendous and became one of the costliest hurricanes in the United States. Understanding its highly erratic movement with changing speed direction and precipitation, access to storm forecast models is essential for local and state officials and emergency responders as these models provide critical storm information (e.g., possible storm trajectories and potential impacts) that acts as basis in decision making under uncertainty for preparedness and response operations. Unfortunately, access to these sophisticated models is limited or come at substantial costs. We developed a data-driven simulation model that takes possible hurricane trajectories, referred here as scenarios, as inputs, coupled with publicly available rolling horizon forecasts of overall weather while incorporating hurricane characteristics (wind speeds, precipitation and storm surges) to generate spatial-temporal storm predicted impacts at scenario level. The model also estimates the scenario probabilities, not directly provided in the public forecasts. The computational results on hurricane Irma case study illustrates the scenario-level storm impacts are generated with fairly accurately within reasonably short runtimes. In the next phase of research not reported here, these predictions impacts can drive models for disruptions in transportation, supply chains and power grids