A saturated stochastic simulator: synthetic US Gulf coast tropical cyclone precipitation fields

Abstract

The space–time fields of rainfall during a hurricane and tropical storm (TC) landfall are critical for coastal flood risk preparedness, assessment, and mitigation. We present an approach for the stochastic simulation of rainfall fields that leverages observed, high-resolution spatial fields of historical landfalling TCs rainfall that is derived from multiple instrumental and remote sensing sources, and key variables recorded for historical TCs. Spatial realizations of rainfall at each time step are simulated conditional on the variables representing the ambient conditions. We use 6 hourly precipitation fields of tropical cyclones from 1983 to 2019 that made landfall on the Gulf coast of the US, starting from 24 h before landfall until the end of the track. A conditional K-nearest neighbor method is used to generate the simulations. The TC attributes used for conditioning are the preseason large-scale climate indices, the storm maximum wind speed, minimum central pressure, the latitude and speed of movement of the storm center, and the proportion of storm area over land or ocean. Simulation of rainfall for three hurricanes that are kept out of the sample: Katrina [2005], Rita [2005], and Harvey [2017] are used to evaluate the method. The utility of coupling the approach to a hurricane track simulator applied for a full season is demonstrated by an out-of-sample simulation of the 2020 season.