Emulating Arbitrary Tornado Debris Fluxes Using “SimRadar”

Funded by the National Science Foundation (NSF), “SimRadar” was recently developed as a virtual means to study tornadic debris and associated polarimetric signatures under a controlled environment. To emulate polarimetric tornado debris signatures (TDSs), SimRadar ingests realistic tornado wind fields generated through a large eddy simulation (LES), air drag model (ADM) with six degrees of freedom for trajectory calculations, and a library of tabulated polarimetric radar cross section (RCS) of different objects for electromagnetic backscattering calculations. In the original design, debris objects were populated in the simulation domain uniformly. In this work, a new technique was developed that allows SimRadar to populate the debris objects as a function of the near-surface velocities. This is motivated by the observations that suggest greater debris lofting in more intense regions of the tornado as well as wind tunnel studies that have revealed a power-law relationship between the wind and debris flux. One of the novel aspects of this work is the debris flux method that properly emulates an arbitrary two-dimensional distribution, with the support of non-uniform grid spacing. To that end, the technique allows us to use any two-dimensional debris distribution or flux map, determined by the locations of debris sources and the intensity of near-surface winds. For example, specific locations of debris sources can be simulated, such as isolated buildings that generate particular debris types (e.g., wood framing), or the distribution map could represent uniform sources of debris (e.g., leaves from a forest). From our literature search, the commonly used technique to generate random samples with an arbitrary distribution is the so-called inverse sampling of the cumulative distribution function (CDF). This method, however, is limited to 1-D applications and the extension of such technique to a 2-D version is still an open research problem. Inverse sampling of a 2-D CDF is inherently problematic since it could result in multiple solutions given an inverse lookup CDF value. We developed practical technique that can be simplified into a two-stage method: first use a random number to pick a cell of a 2-D map, then uniformly distribute 2-D random samples within the cell, compensated by the cell area. That is, the number of samples should be proportional to the area of the cell. A naïve method requires three random number generators, but we are able to reduce the method into just two random number generators. Significant effort was also devoted in this work into making it GPU acceleration compatible. At the conference, a detailed technical overview will be provided and preliminary results from the study of debris release as a function of LES wind field will be presented.