Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150514

150514 - Bayesian Network Framework for Wind Risk Assessment: Modeling Spatial Correlation of Tropical Cyclone Wind Fields 

Risk analysis of a spatially distributed asset exposed to winds from tropical cyclones (TC) requires consideration of the spatial correlation of wind intensities at multiple sites. This correlation is associated with windspeed residuals (i.e., the differences between estimated windspeeds and predicted ones). It is assumed that windspeed residuals at multiple sites are spatially correlated with each other, which is due to the fact that all sites experience the same TC with specific characteristics and possibly other reasons, such as the type of land cover in which they are located. The residual has two main components: within-event residuals (i.e., variability of windspeeds in multiple sites due to different storms) and within-event residuals (i.e., variability of windspeed among multiple sites due to a single storm).

Previous research shows that neglecting the spatial correlation of wind speeds at multiple sites could overestimate or underestimate the calculated risk for a spatially distributed asset. Also, the literature review indicates a gap in assessing wind correlation models' uncertainty and sensitivity to specific parameters. The current study aims to develop a robust wind correlation model using a rich dataset of different TCs landing across different locations across the US. We also intend to assess the uncertainties associated with the statistical approaches used to estimate the correlation model and how they affect the risk estimates. We also attempt to investigate the effect of TC's physical characteristics and the land cover of the distributed infrastructure on the correlation structure and estimated risk. We use windspeed data from sixteen TCs that made landfall in various parts of the United States. The observed dataset is part of the H*wind dataset created by NOAA's Hurricane Research Division. The HAZUS Hurricane Model was used to estimate wind speeds at the center of census tracks impacted by these sixteen TCs.

We used empirical variograms, a widely used geospatial tool, to capture the dissimilarity of wind speed residuals at spatially separated sites, as well as the correlation structure between them. The correlation model is constructed using three standard models: exponential, Gaussian, and spherical. To fit these models to the experimental variograms, the ordinary and weighted least squares techniques, with two simple and exponential weighting functions, are used. The weighted least squares technique prioritizes the close distances that are more important in correlation and risk analysis. Furthermore, the directional variogram is used to capture the anisotropy correlation in wind fields, which is crucial for the risk assessment of spatially distributed assets.

Preliminary results show a clear correlation between windspeed residuals. Also, results indicate that the correlation structure is significantly dependent on the fitting techniques, and different candidate fitting models characterize the correlation differently. Furthermore, the anisotropy analysis using directional variograms reveals that the correlation in specific directions is stronger, leading to a higher wind risk in those directions.

In conclusion, the preliminary results of this study reveal important characteristics of the windspeed correlation and some impactful parameters on this correlation. These outcomes could be significantly helpful in the robust wind risk assessment of spatially distributed buildings and lifelines. We intended to leverage the outcomes of this study to develop a Bayesian network-based framework for wind risk assessment of spatially distributed assets.

Presenting Author: Amirreza Mohammadi University of Maryland

Presenting Author Biography: Amirreza Mohammadi is a first-year PhD student in civil engineering at the University of Maryland, where he earned his bachelor's degree in civil engineering and a master's degree in earthquake engineering. With a solid foundation built on three years of research experience in probabilistic hazard analysis, focusing on seismic hazards, he has demonstrated a commitment to advancing safety in civil infrastructure.

Amirreza's previous research includes contributions to understanding and mitigating regional seismic hazards through innovative probabilistic concepts and their engineering applications. During his doctoral studies, he aims to develop a comprehensive multi-hazard risk assessment framework based on Bayesian networks, integrating his expertise in earthquake engineering with broader environmental and societal risks.

Authors:

Amirreza Mohammadi University of Maryland
Michelle Bensi University of Maryland