Poster Presentations:
Associate Professor Oregon State University
Multi-disciplinary framework for wildfire risk research
Co-Authors: Jenna Tilt (Oregon State University), Erica Fleishman (Oregon State University), Chris Dunn (Oregon State University), Heidi Huber-Sterns (University of Oregon), Eduardo Cotila-Sanchez (Oregon State University), Nicole Errett (University of Washington), and Jamie Trammel (Southern Oregon University)
Abstract: Wildfire research has been predominately performed within the forestry community. However, as fires continue to impact the wildland urban interface (WUI), research is needed to explore the impacts of fire on the built environment as well as the social and economic impact of infrastructure failure. This poster will summarize an ongoing effort to integrate research across civil engineering, forestry, electrical engineering, climate science, social science, and public health. This research has exposed where our research does not align temporally as well as spatially and how we can deepen our understanding of wildfire risk within communities by performing interdisciplinary research across these boundaries. The team used Ashland, Oregon as a case study community performing a scenario analysis within the community to identify social and economic assets within the community through a survey, modeling 213 fires through the community to calculate fire exposure to these assets, modeling the water system demands to identify areas of vulnerability as well as when the water system would depressurize, and exploring electrical grid resilience. This poster will summarize the methodology as well as the research results.
PhD Student University of Southern California
Evaluating Frequency Contributions to Pseudo-Spectral Acceleration Using Observations and Physics-Based Earthquake Simulations
Co-Author: Chukwuebuka Nweke (University of Southern California)
Abstract: Pseudo-spectral acceleration (PSA) is a fundamental metric in seismic hazard analysis and engineering applications, yet its relationship to the underlying frequency content of ground motions remains incompletely understood. In this study, we develop a methodology to identify and quantify the frequency bands that contribute most strongly to PSA at a given oscillator period, explicitly linking frequency-domain characteristics of ground motion to oscillator response.
The proposed framework is applied to observed earthquake recordings to investigate how frequency contributions to PSA vary with geological site conditions and source-to-site distance. By examining recordings from basin and non-basin sites across a range of distances, we characterize how the dominant contributing frequency bands shift with site properties and wave propagation effect. We then apply this methodology to evaluate and understand the PSA residuals in physics-based ground-motion simulation validation providing insight into how frequency-dependent misfits between observations and simulations contribute to PSA bias.
Overall, this study shows that the response of an oscillator at a given period is not controlled by a single frequency component of the ground motion, but rather by a broadband range of frequencies whose relative contributions depend on the oscillator period and damping. The proposed methodology provides a physically interpretable basis for understanding observed PSA trends and diagnosing PSA residuals in simulation-based ground-motion validation.
PhD Student New York University
Estimating First Floor Height Above Ground Using Google Street View and Machine Learning
Co-Authors: Gabriela Nino Herrera (UC Berkeley), Qirui Su (NYU), Lishun Liu Gaara (NYU), Danni Yang (NYU), and Luis Ceferino (UC Berkeley)
Abstract: First Floor Height Above Ground (FFHAG) is the vertical distance between a building’s lowest occupied floor and the adjacent ground level. Accurate FFHAG estimates are essential for flood risk analysis, as damage depends on how high the first floor lies relative to floodwaters. Worldwide, FFHAG data remain incomplete: some jurisdictions have conducted field surveys, but many still rely on coarse estimates. The Hazus loss estimation software assigns default first-floor heights based on building characteristics like construction year and location relative to the Special Flood Hazard Area, achieving broad coverage but limited accuracy. Recent methods using street-view imagery and machine learning offer a scalable alternative. They typically follow two steps: (1) image acquisition, retrieving street-view images (façades showing doorways, stairs, and ground lines); and (2) FFHAG estimation, applying machine learning models to infer the lowest floor elevation relative to grade. Recent papers adopt this pipeline, yet no study has systematically compared their feasibility, accuracy, and implications for flood risk analysis. We review FFHAG estimation methods using street-view imagery and machine learning, documenting methodological choices, reported performance, and gaps. Building on this, we develop a quantitative scoring framework to compare methods by accuracy, scalability, and data needs. We then benchmark selected methods against Hazus defaults and evaluate how FFHAG uncertainty propagates into flood risk estimates in a New York City neighborhood using ground-truth elevations from the NYC Building Elevation and Subgrade (BES) dataset, with implications for flood risk assessment, insurance modeling, and mitigation planning.
PhD Student Texas A&M University
Integrating Agent-Based Modeling and Microscopic Traffic Simulation for Port and Intermodal Network Resilience under Hurricanes
Co-Author: Maria Koliou (Texas A&M University)
Abstract: Ports are critical components of transportation networks, particularly intermodal networks, as they serve as indispensable hubs that connect water- and land-side traffic. However, their coastal and waterside locations make them highly vulnerable to extreme weather events such as hurricanes and storm surge, and disruptions can have severe consequences, including cargo delays, economic and supply chain losses, delayed emergency response, and slower local and inland community recovery.
This research develops a comprehensive resilience assessment framework for port systems and their interconnected networks under hazard conditions. The port is treated as a complex system whose behavior emerges from multiple interdependent infrastructures (wharves, internal roads, storage yards, and buildings), equipment (cranes, trucks), and operating subsystems. Internal traffic within the port is explicitly modeled as a system of its own, essential to terminal functionality and tightly linked to external road and rail networks. To capture these interactions, the framework couples an agent-based model implemented in SimPy with a microscopic traffic simulator in SUMO, synchronized at each time step to represent both operational decisions and traffic dynamics under hurricane-induced infrastructure damage. The resulting disruption patterns are then propagated to the external road network to provide a more holistic resilience assessment.
The framework is applied to two major container terminals at the Port of Houston (Bayport Terminal and Barbours Cut Terminal). Key entities, including ships, trucks, yard blocks, and terminal operating systems, are modeled as agents, enabling realistic, detailed resilience metrics and supporting the design of effective mitigation strategies based on quantitative resilience indices.
PhD Student University of California Berkeley
A Physics-Inspired Deep Learning Framework for Seismic Site Response Analyses
Co-Author: Mohamad M. Hallal (University of California Berkeley)
Abstract: The Federal Emergency Management Agency (FEMA) estimates that earthquakes cost the United States approximately $14.7 billion annually. Indeed, our improved understanding of the physics of earthquakes, the increased number of case histories, and the rapid growth in computational resources have advanced hazard modeling and mitigation. Despite these advancements, hazard modeling remains constrained by inherent subsurface uncertainties and computations costs. Currently, most seismic site response studies rely on simplified 1D analyses, overlooking critical subsurface variations and reducing prediction accuracy. While 2D and 3D simulations offer higher fidelity, their complexity and high computational costs hinder broader adoption in engineering practice.
This study proposes a framework based on Physics-Inspired Deep Learning to enable AI-accelerated seismic site response simulations in earthquake engineering. We introduce the Spatially-Variable-Informed Neural Operator (SVINO), which is designed to capture the effect of subsurface variability on seismic site response. SVINO is designed to extract frequency-domain information and directly predict the site-specific transfer function. Crucially, this approach integrates neural operator discretization-agnostic property, moving beyond purely data-driven methods and guaranteeing physically consistent, frequency-informed results. The proposed framework delivers predictions that are of comparable accuracy to high-fidelity numerical simulations but at significantly faster speeds. By overcoming the computational barriers of traditional 2D/3D methods, the proposed approach enables more accurate, efficient site response predictions, directly enhancing the efficacy of seismic risk mitigation strategies.
PhD Student Texas A&M University
Vessel-risk-aware: a decision support model for vessel routing based on multicriteria decision analysis and an advanced Dijkstra algorithm
Abstract: Maritime weather routing research has largely prioritized minimizing operating costs, fuel consumption, and estimated time of arrival (ETA). However, existing models often neglect accident avoidance and environmental risks. Fishing vessels particularly face many challenges from dynamic coastal hazards, frequent operational stops, and diverse safety thresholds. This study proposes a multicriteria spatial decision support system (SDSS) that integrates an enhanced Dijkstra algorithm with Weighted Sum Model (WSM) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to address these gaps. The framework advances maritime navigation through three major contributions. First, it incorporates eight-directional pathfinding into the Dijkstra algorithm to better account for coastal navigation constraints. Second, it integrates dynamic vulnerability indices derived from bathymetry, wave conditions, and vessel speed, calibrated against historical accident data to reflect real-world risk. Third, it provides an interactive interface that enables stakeholders to assign relative importance to safety, efficiency, and environmental criteria, thereby fostering transparent and collaborative decision-making. Together, these innovations generate optimized routes that balance multiple objectives under varying oceanic conditions such as wind, waves, and currents. By bridging theoretical routing models with the practical demands of fisheries management, this framework offers a scalable tool for safer maritime navigation in weather-dependent contexts.
PhD Student Florida State University
Decoding Urban Drainage Resilience: Linking Network Topology and Spectral Resilience to Dynamic Flood Vulnerability
Co-Author: Neetesh Sharma (Florida State University)
Abstract: Urban Drainage Networks (UDNs) are critical infrastructure systems whose resilience and vulnerability assessment is often limited by the high computational cost of hydrodynamic models such as PCSWMM. To address this limitation, this study employs a Graph Network Analysis (GNA) framework for the Teaneck, New Jersey, urban drainage network. The drainage system is represented as a directed graph, and key topological and spectral metrics are used to characterize its inherent structural behavior. Results reveal a highly vulnerable network with extremely low redundancy, evidenced by a meshedness coefficient of approximately 1% and near-zero algebraic connectivity. The analysis establishes a direct quantitative link between static network structure and dynamic flood vulnerability, demonstrating how flooding increases existing topological weaknesses and causes cascading failure. By utilizing graph-based methods, this approach helps in rapid, system-wide resilience assessment and multi-component failure screening, which are impractical using hydrodynamic models alone. The proposed framework provides an efficient decision-support tool for prioritizing targeted infrastructure reinforcement and enhancing flood resilience in urban drainage systems.
PhD Student Stanford University
Large-Eddy Simulations of Downburst Flow and Wind Loading using a Cooling Source Term
Co-Authors: Jianyu Wang (Stanford University) and Catherine Gorle (Stanford University)
Abstract: Downburst outflows generate highly transient and spatially non-uniform wind fields that can produce peak wind loads on structures with very different characteristics than stationary atmospheric boundary-layer flow. Experimental research has provided important insights into non-stationary downburst flow and its interaction with structures, but a generally accepted approach to determine the design loads for structures exposed to downbursts is still lacking. Two main challenges can be identified. First, wind loads depend on many parameters that determine the specific wind flow and resulting loads. Second, the short-duration, non-stationary character implies that the peak loads observed might be different for two similar events. This study presents a computational framework for large-eddy simulations of downbursts and the resulting wind loads. A non-stationary downburst-like flow is created by specifying a cooling source term that induces an impinging jet, followed by outward radial flow that generates a non-log-law velocity profile with high wind speeds near the surface. In this initial study we exploit the quasi-axisymmetric property of a downburst event with zero translational velocity generated with this framework to explore the second challenge identified above. By sampling across multiple radial locations that can be considered statistically independent, the effective sample size of one simulation is increased, enabling an assessment of the statistical convergence of peak load predictions. Future work will extend the LES framework to investigate the variability of the near-ground wind field and wind loads associated with other parameters, including downburst characteristics, structure locations within the downburst, and the presence of neighboring structures.