PhD Student Stanford University
Improving regional wildfire risk assessment frameworks by including the probability of ignition sources
Co-Authors: Neetesh Sharma (Florida State University) and Jack Baker (Stanford University)
Abstract: Wildfires represent a growing threat to communities in the wildland-urban interface (WUI), driven by climate change, urban development, and inadequate fuel management. Together, these factors increase both environmental impacts and community vulnerability. A significant portion of damaging fires has been linked to power-system equipment failures (e.g., transmission and distribution lines); in California, electrical infrastructure has been implicated in about 40% of the state’s most destructive wildfires. Despite this risk, regional ignition models that explicitly incorporate utility infrastructure remain limited, and existing risk frameworks rarely integrate ignition, spread, damage, and loss in a unified workflow. Therefore, this work aims to develop a comprehensive regional wildfire risk framework that begins by developing ignition models for different sources of ignition using parametric Poisson Point Process analysis to quantify the rate of ignitions across power utility networks and space as a function of spatial variables, as well as considering weather variables such as temperature and wind as weight factors. Then, using the results from the ignition model, this study employs state-of-the-art models to conduct wildfire spread simulations under various weather and simulation time parameters. Next, the damage is assessed in each wildfire simulation and quantified across different levels of destruction in the community. Lastly, the quantified risk represents the expected losses the community could foresee. Beyond quantifying the direct financial impact of these wildfire scenarios, estimating the likelihood of ignitions from power infrastructure can inform mitigation actions and potential liability costs in future wildfire events.
Assistant Professor The University of Texas at Austin
Hydro-UQ–Enabled GPU Simulations of Wave–Debris–Structure Interactions in Inland Coastal Forests During Tsunamis
Co-Authors: Youngchul Choi (The University of Texas at Austin), Justin Bonus (University of California, Berkeley), and Che-Wei Chang (University of Rhode Island)
Abstract: Tsunamis generate complex hydrodynamic and debris-loading processes that pose significant risks to coastal communities. Although coastal forests are often regarded as natural defenses that dissipate wave energy, observations from historical events indicate that inland coastal forests—those located landward of buildings—can sometimes intensify damage. This occurs when debris becomes trapped within the forest, forming temporary dams that reflect and redirect tsunami flows toward structures. Despite its importance, this debris-induced damming mechanism and the associated wave–debris–structure interactions remain poorly quantified. This study uses the SimCenter’s Hydro-UQ platform to conduct high-fidelity simulations based on the Material Point Method (MPM) on the Texas Advanced Computing Center’s (TACC) system. The model was calibrated and validated using large-scale tsunami flume experiments conducted at NHERI’s O.H. Hinsdale Wave Research Laboratory. We then performed 138 simulations under idealized flume conditions, systematically varying forest–structure spacing, forest density, and debris quantity. Each simulation used three NVIDIA A100 graphics processing units (GPUs), employed a 2.5 cm spatial resolution, and tracked approximately 95 million particles. Results show that debris damming within inland forests increases reflected wave heights, significantly raises local water levels near structures, and amplifies structural loading. Sensitivity analyses identify forest–structure distance and debris volume as dominant contributors to wave amplification. Using the full simulation dataset, we developed an empirical relationship linking forest spacing, debris quantity, and forest density to reflected water levels. This work demonstrates how high-fidelity, GPU-accelerated MPM modeling within Hydro-UQ can advance computational understanding of tsunami-induced hazards.
Postdoctoral Scholar University of Kansas
From Census Blocks to Buildings: Downscaling Workforce Data for Natural Hazard Impact Modeling
Co-Author: Elaina Sutley (University of Kansas)
Abstract: This work presents a computational simulation pipeline to downscale the U.S. Census Bureau’s LEHD Origin–Destination Employment Statistics (LODES) data for natural hazard impact and community resilience modeling. LODES datasets provide detailed employment and commuting information at the Census Block level and are widely used to characterize social and economic dimensions of urban systems. As such, they offer valuable indicators for assessing disaster-induced workforce disruption and recovery.
However, LODES and similar areal datasets represent spatial averages, whereas hazard impacts, such as building damage, utility outages, and transportation network disruptions, occur at the level of individual structures and infrastructure links. This scale mismatch limits the ability to directly map physical damage to affected workers and commuting patterns. To address this challenge, the proposed framework employs stochastic modeling to generate a pseudo household-level employment inventory that preserves aggregate block-level statistics while respecting data privacy constraints.
Worker and employment attributes are algorithmically assigned to synthetic households and linked to residential buildings, revealing intra-block variability in worker–home linkages that is not captured in existing datasets. This enables finer-grained simulations of disaster-induced economic disruption, including the impacts on commuting flows and job accessibility due to building and infrastructure damage. The proposed pipeline supports multi-resolution modeling of urban systems and improves the integration of workforce dynamics with critical infrastructure performance, providing a scalable approach for computational resilience and recovery analyses.
Postdoctoral Scholar Stanford University
Extending tropical cyclone risk assessment through recovery simulations
Co-Authors: Nikola Blagojevic, Meredith Lochhead, Avantika Gori, and Jack W. Baker
Abstract: Extreme weather events such as tropical cyclones increasingly threaten societies as climate change amplifies their impacts. While climate risk assessments have traditionally focused on direct impacts, such as economic losses, population exposure, or mortality, post-disaster recovery remains largely absent from these frameworks, limiting our ability to assess long-term resilience.
We present an approach to integrating recovery modeling into tropical cyclone risk assessment using open-source, regional disaster recovery simulations. The framework leverages building permits as a proxy for local construction capacity, enabling the explicit representation of spatial heterogeneity in recovery potential all along the US North Atlantic coast.
Results reveal spatial disparities in rebuilding capacity relative to climate risks, highlighting regions where recovery may be constrained despite comparable hazard exposure. These findings illustrate how incorporating recovery processes into risk assessments can inform targeted policy and planning interventions aimed at accelerating recovery and strengthening long-term resilience.
Assistant Professor University of Nebraska–Lincoln
A Probabilistic Digital Twin Framework for Multi-scale Post-Earthquake Assessment of Instrumented Buildings
Abstract: Post-earthquake assessment of instrumented buildings is traditionally framed as either a model-updating problem or a fragility-based loss estimation task. More recent approaches recast this problem as one of structural state estimation and dynamic response reconstruction. This study advances the latter by proposing a probabilistic Digital Twin (PDT) framework that integrates physics-based simulation, sensing data, and performance-based engineering (PBE) within a unified computational workflow. The framework couples a nonlinear structural model with high-rate sensor measurements through a probabilistic nonlinear model-data fusion strategy. An extended model-based observer (EMBO) serves as the core synchronization engine, enabling real-time estimation of structural states, response quantities, and associated uncertainties. The estimated demand parameters are subsequently propagated into a PBE-based damage and loss assessment module, enabling the direct translation of observed structural responses into decision-relevant metrics, such as damage states, expected losses, and re-occupancy suitability. By emphasizing continuous data-model interaction rather than offline simulation or calibration, the proposed PDT advances beyond conventional post-event analysis, enabling uncertainty-aware and computationally efficient assessments under evolving structural conditions. The framework is validated using full-scale seismic response data from the NEESWood Capstone building tests conducted at the E-Defense facility in Japan, demonstrating accurate real-time reconstruction of nonlinear response and consistent damage-to-loss inference within a single probabilistic environment. This work highlights the role of OpenSees-based state estimation combined with PBE loss modeling as a scalable pathway toward operational digital twins for post-earthquake decision support.
Research Associate Professor Texas A&M University
Bridging the People Gap in Community Resilience Modeling: Introducing pyncoda
Co-Author: Adam Zsarnoczay (Standford University)
Abstract: In community resilience modeling, people are too often an afterthought—added only after building and infrastructure damage has been calculated. This approach risks systematically missing or underrepresenting vulnerable populations, leading to biased results and less effective disaster planning.
INtegrate COmmunity DAta (pyncoda) is an open-source tool designed to help put people first in community resilience planning workflows by quantifying key characteristics of households and linking them to buildings and infrastructure within the study area. This process helps ensure that building and infrastructure data are comprehensive and fully cover the entire population. pyncoda enables more accurate, equitable, and actionable resilience assessments by making it easy to identify and address data gaps and inaccuracies. Originally developed as part of the IN-CORE project, pyncoda has been integrated with the NEHRI SimCenter Framework, including the Regional Resilience Determination (R2D) application. This presentation highlights the key features of pyncoda and demonstrates how it enables propagation of uncertainties in community features and fosters reproducible research in any study location across the United States.
Associate Research Scientist Johns Hopkins University /
National Institute of Standards and Technology
Simulation of hurricane surge and waves through the built environment at regional scales
Co-Authors: Don Slinn (National Institute of Standards and Technology)
Abstract: Coastal models of hurricane storm surge typically neglect buildings despite their importance in influencing overland waves and currents. This presentation describes a novel approach to consider the built environment in hurricane modeling of surge and waves using XBeach. Hurricane Ian and Estero Island in Lee County, Florida are used as a test case to demonstrate the approach. Forcing for XBeach is provided from ADCIRC/SWAN simulations. The actual locations, sizes, and shapes of buildings are obtained from building footprint data available on DesignSafe and contain structure information regarding building foundation height and type. Multiple simulations are performed with: (1) no buildings in the model domain, (2) all buildings are located on the ground assuming no information is available regarding building elevation, (3) elevated buildings are removed from the model domain assuming that waves can freely pass under elevated buildings, and (4) buildings are sequentially removed from the model domain assuming structural collapse. Results show that including buildings significantly influence modeled wave heights. The presence of buildings tend to decrease overland wave heights, particularly behind the first row of houses; however, some areas also show that including buildings leads to increases in wave heights. By simulating sequential failure of structures, we are able to closely reproduce structural collapse that was observed during Hurricane Ian. The approach presented in this talk can lead to more accurate regional risk and resilience models that consider hurricane impacts at the building-scale.
PhD Student University of Michigan
CommunityWindSim: An open-source framework to simulate wind flow around a community using computational fluid dynamics
Co-Authors: Jieling Jiang (University of Michigan), Seymour M.J. Spence (University of Michigan), Ann Jeffers (University of Michigan), Abiy Melaku (UC Berkeley), and Frank McKenna (UC Berkeley)
Abstract: Accurate predictions of wind flow in an urban environment are critical for various environmental and structural applications. Obtaining this is an extremely complex and tedious process that includes several sub-processes. A framework – CommunityWindSim – that provides a user interface to simulate wind flow in a community is proposed here. This framework is added as an advanced computational fluid dynamics (CFD) module in the WE-UQ software of the NHERI SimCenter. User-inputs include information regarding the total region to be modeled, the region of interest (ROI), mesh refinement regions, boundary conditions, turbulence modeling, and time controls, and optionally a CSV file if the user chooses to use the turbulent inflow (TINF) tool. The framework uses BRAILS++ to generate building inventory and then extracts the building footprints along with their heights and creates a computational domain. Turbulence models include Reynolds-averaged Navier-Stokes (RANS), unsteady RANS (URANS), and large eddy simulations. Two GeoJSON files are provided as outputs, one covering the total region, while the other covers the ROI, along with a case folder which can directly be run in OpenFOAM. The wind simulation results can be viewed in Paraview. A URANS k-ω shear stress transport wind simulation example is considered in a community based in Altadena, California to illustrate the use of this tool. Results from the simulation are presented. This framework thus provides an interface to directly run a community-scale wind simulation which can be used for various wind-hazards assessment scenarios.
PhD Student New York University
RoofNet: A Vision–Language Model–Enabled Dataset for Scalable Building Exposure Characterization Across Natural Hazards
Co-Authors: Noelle Law (New York University) and Yuki Miura (New York University)
Abstract: Reliable building exposure data remain a central limitation in regional natural hazard risk assessment, particularly for attributes such as roof material that strongly influence vulnerability to wind, flood, and wildfire hazards. Outside well-instrumented regions, such information is often missing or inconsistently represented, constraining the fidelity and transferability of computational simulation and uncertainty quantification.
This talk presents RoofNet, a large-scale, multimodal dataset that enables scalable roof material classification using high-resolution Earth observation imagery and vision–language models (VLMs). By combining visual feature extraction with prompt-based semantic reasoning, the RoofNet pipeline infers roof material attributes across diverse geographic contexts without reliance on parcel-level records or manual annotation workflows. The resulting dataset is designed to integrate seamlessly into computational natural hazard workflows by providing a critical exposure attribute that directly supports fragility modeling, probabilistic loss estimation, and uncertainty-aware simulation.
We describe the RoofNet data generation and evaluation framework and demonstrate how consistent roof material information improves exposure characterization across wind, flooding, and wildfire contexts. By enabling globally transferable roof material classification, RoofNet advances data-driven exposure identification and measurement, supporting more robust, scalable, and equitable multi-hazard risk assessment and decision-making.