Integrative Data-Driven Approaches to Tornado Wind Field Reconstruction
Advisor: Associate Professor Franklin Lombardo
Abstract
Tornado events are responsible for many fatalities and significant building damage in the United States. Therefore, building design under tornadic loads has become a critical research topic in wind engineering. Furthermore, to enhance the robustness of the tornado design, continued research on tornado event characteristics and induced loading on structures is necessary. Understanding the near-surface tornado wind field and obtaining sufficient data to reconstruct the field is the main challenge for this research. Furthermore, considering the short period, and extreme intensity, in the generation and track of tornadoes, it is dangerous and unlikely to capture in-situ wind data in a tornado event. As a result, this dissertation introduces a novel integrative data-driven approach to recreate the near-surface tornado wind field. In these approaches, the wind field information relies on indirect damage assessment data, such as wind field indicator (FI) discussed in this dissertation, rather than direct measurements. Then, analytical models are developed to simulate the tornado-like vortex, and scale parameters are introduced to govern the numerical wind field based on the FIs. Finally, as the overarching goal, the best-fit numerical near-surface wind field will be generated for the actual scenarios and provide an improved assessment method for tornadoes. This dissertation consists of four main objectives: 1) Development of FIs as an indirect wind field estimation technique; 2) Streamline the indirect damage data collection method in the tornado damage reconnaissance survey; 3) Understanding and developing the analytical tornado-like vortex model; 4) Implementing data-driven methods to evaluate near-surface tornadic wind fields.
This dissertation tackles the challenges inherent in the wind speed estimation of tornado cases utilizing unconventional measurement methodologies. FIs are developed from the tornado damage reconnaissance surveys and applied for constructing a tornadic near-surface wind field. The data retrieving process of the FIs from the post-damage investigations, ground-based survey documentation, aerial photos, and social media data are discussed. Then, through image recognition technology and manual detection, fallen trees, wind-induced debris, and damaged buildings are recognized and tagged with geographic information as FI types.
Collection of FIs via various data sources, including GPS-camera photos, drone photos, and social media videos, the data format and techniques and their documentation are described along with a standardization process. This process will help the survey team to collect data rapidly and exhaustively.
Analytical models for vortex-like flow can be applied in estimating the near-surface tornadic field. This dissertation summarizes famous analytical vortex models and discusses the applicability of these models. In addition, the governing parameters for the analytical models and the optimization process
for fitting the estimation in the real scenario are also discussed. Further, a new full-scale analytical model for tornado-like vortex based on the observed dust-devil measurement data is developed.
Tree-fall information and windborne debris information are considered to be implemented for tornado wind field reconstruction. In this application, an analytical tree-fall model is established for the simulated tree-fall patterns. Further, real scenarios are inputted to adjust the governing parameters for the vortex model and improve the estimated near-surface wind field. As for the wind-induced debris information, the areal distribution map of the debris is implemented for simulating the vortex field. In this simulation, the 6-DOF EOM of the debris is developed to simulate the trajectory, and the estimated fallen points of the debris are compared with the actual distribution map to estimate the near-surface vortex field. In addition, video of the debris trajectories is applied as another method for simulating the time history of a three-dimensional tornado wind field.
In summary, the main findings of this dissertation can be listed as: 1) Numerical vortex models can be optimized with full-scale natural vortex data; 2) Tree-fall pattern and debris distribution maps can be implemented for constructing tornadic wind field; 3) Flying windborne debris can act as sensors for estimation of vortex structure. As for future development, deep learning technology can be applied in the automatic recognition process of FI, and a 3D point cloud is recommended to construct for obtaining detailed information on FI.
