Numerical Method for a Stochastic Inverse Problem with Application to Wind Drag Parameterization

Flooding from storm surge and rainfall poses great risk to life and property during severe hurricanes and extratropical cyclones. Predicting such flooding events is an important task, complicated by the complex underlying physical processes. Critical input in computational models of storm surge include the meteorological fields and physical parameters, among them wind speeds, sea ice concentration (in polar regions), bottom and wind drag parameterizations, etc., which are expensive to obtain and quantify, are uncertain, and evolve in time.

In this presentation, we focus on wind drag, of particular importance as it controls the momentum transfer from winds to the water column. In high latitudes, the presence of sea ice can enhance wind momentum transfer under fragmented ice conditions or reduce it under packed ice conditions, thus complicating the choice of wind drag parameterization.

First, we will formulate the task of inference and prediction of wind drag parameterization in the setting of a stochastic inverse problem. Then, we will present recent work on the construction, analysis, and implementation of a numerical method for this inverse problem. The proposed method combines feature extraction techniques, to learn the quantities of interest from time series data, with high-fidelity mathematical and computational models, including the ADvanced CIRCulation (ADCIRC) model.