Session Abstracts:
Assistant Professor
University of Rhode Island
Model-to-data validation of coastal flood damage models for buildings impacted by Hurricane Ian (2022) at Fort Myers Beach, Florida
Co-Authors: Daniel Cox (Oregon State University) and Andre Barbosa (Oregon State University)
Abstract: Data with a high spatial resolution are required to properly validate existing flood damage models. This study aims to leverage a hindcast study of the impact by Hurricane Ian (2022) on buildings in Fort Myers Beach, Florida. A comprehensive virtual damage assessment (VDA) methodology is proposed to estimate building exterior damage states (DS) ranging from DS0 (no damage) to DS6 (complete damage), which includes undergraduate students as part of the VDA process. A second methodology is proposed to measure the first-floor elevation (FFE) using street-level and aerial imagery. The dataset created includes DS and FFE for over 3,000 buildings, and validation/uncertainty quantification for a subset of the data based on student and expert opinions. For the damage assessment, a hazard layer was developed based on observed high-water marks with/without wave effects. Using a subset of the dataset, two simulation platforms are adopted to predict structural damage to buildings due to surge and waves: (1) regional resilience determination (RD2) simulation software developed by SimCenter and (2) open-source Interdependent Networked Community Resilience Modeling Environment (IN-CORE) developed by the Center for Risk-Based Community Resilience Planning. Simulation results are compared with reconnaissance observations at the parcel level to provide an understanding of the accuracy and uncertainty associated with existing flood models, which is critical for engineers, stakeholders, and emergency planners. The results indicate the success of the VDA methodology and the potential for future model-to-data and model-to-model comparisons.
Graduate Student Researcher
University of Missouri
Post-Disaster Damage Assessment from High-Resolution Satellite Imagery using D-LinkNet with Hierarchical Transformer Difference Blocks
Co-Authors: Hyeong Suk Na (University of Missouri)
Abstract: There is an increasing demand for a robust and comprehensive approach to assess the severity of damage caused by natural hazards and disasters. To achieve this goal, a well-classified damage level segmentation map is required at large scale. The process involves three key steps: high-precision feature extraction, change detection, and classification of damaged areas. However, many existing post-disaster damage assessment models struggle to integrate these processes seamlessly in an end-to-end manner. In this study, we propose a large-scale post-disaster assessment model using the D-LinkNet Hierarchical Transformer network. This model seamlessly integrates all three steps—feature extraction, change detection, and classification—into a unified framework, enabling accurate large-scale classification of damaged buildings and roads. For the image extraction, our network leverages a dilation module at the center of the U-Net architecture, following the approach used in D-LinkNet. This dilation module preserves high-level features in low-resolution images during downsampling. By combining this with hierarchically structured difference blocks based on transformers, the network effectively performs change detection and damage classification. Our model was trained using satellite imagery from the Haiti earthquake dataset, available through DesignSafe’s Data Depot, as well as the xBD dataset from the xView2 Challenge. The network takes pre- and post-disaster images as inputs and generates a four-level damage classification, ranging from level 1 (least damage) to level 4 (most severe damage). Additionally, our model outperforms state-of-the-art methods, including DAHiTrA and BiTransUNet, by being trained on two different datasets and effectively preserving spatial relationships and high-level features among the detected objects.
Graduate Student Researcher
University of Delaware
Hurricane Wind Loss Modeling Using Insurance Claims Data
Co-Authors: Rachel Davidson (University of Delaware), Ertugrul Taciroglu (University of California, Los Angeles), Patrick Hadinata (University of California, Los Angeles), and Mohammad Askari (University of California, Los Angeles)
Abstract: Hurricane hazard poses a major risk to properties in coastal regions, causing significant damage and substantial losses for homeowners and insurance companies. The accurate prediction of hurricane wind-induced loss is crucial for risk assessment, resilience planning, and proper insurance premium pricing. Insurance claims data, although challenging to obtain due to its proprietary nature, plays an essential role in hurricane loss model development and validation. This study presents a predictive modeling for hurricane wind-induced property damage and loss using insurance claims data from four recent hurricanes [Matthew (2016), Florence (2018), Dorian (2019), and Isaias (2020)] that affected Eastern North Carolina. Predictors such as wind speed, building attributes, land cover type, rainfall, and insurance policy attributes are considered. By leveraging machine-learning, a two-step approach is adopted. This comprises of (1) loss occurrence prediction model to determine the probability of a building experiencing loss, and (2) loss severity prediction model to estimate the loss amount for affected buildings. The performance of multiple machine learning models used in each step is evaluated using different evaluation metrics at three spatial scales—total study area, county level, and individual building level. The final outputs of the two steps are then integrated to assess the expected loss at the spatial scales.
Assistant Professor
University of Wyoming
Capturing Wood Rot Implications for Community Resilience through Multi-scale Modeling
Co-Authors: Natassia Brenkus (The Ohio State University)
Abstract: Hurricanes cause not only significant immediate damage to homes and infrastructure but also render significant, long-term, and far-reaching impacts on communities. Often these are discussed in terms of the impacts on people, such as through population displacements. However, the built environment also bears scars from previous hazard events. One under-studied long-term consequence of moisture intrusion into structures through hurricane wind-driven rain, storm surge, and flooding is the potential for wood rot to degrade the capacity of structural members in residential structures. This presentation will discuss recent efforts by the authors to quantify the impacts of brown rot decay on wood mechanical properties and how decay may impact the performance of structural members. At the center of this effort is a multi-scale computational material model that implements decay effects from the polymer scale to the scale of structural assemblies. The material model utilizes continuum mechanics and finite element approaches to study decay impacts on wood composition and cell structure and how these influence the behavior of wood structural materials. We will discuss the model formulation and the computational approaches utilized to assess the sensitivity of the model. Finally, we will present a framework for integrating wood degradation effects into our understanding of community resilience.
Postdoctoral Fellow
National Institute of Standards and Technology
NED: A refactored Nonstructural Element Database for performance-based engineering
Co-Authors: Jitendra Bhatta (NIST), Amanpreet Singh (NIST), Mohammed Eladly (JHU), and Siamak Sattar (NIST)
Abstract: Performance-based earthquake engineering, which probabilistically quantifies building performance from the fragility of individual building elements, is foundational to the adoption of functional recovery-based seismic design. However, currently available nonstructural element fragility databases only represent a small portion of the building stock and experimental research, thus limiting the robustness of current assessments.
This project addresses key data gaps in the seismic performance of nonstructural components through the introduction of NED, an opensource Nonstructural Element relational Database. Currently in beta form, NED updates, expands, and refactors the collection of nonstructural element experimental test data and seismic fragilities into a robust SQL relational database. Through extensive literature review, we have collected over 1500 data points on experimental testing of nonstructural building elements and have complied this alongside seismic fragilities produced by the FEMA P-58 report to better represent the damage and failure of nonstructural systems due to earthquakes.
NED represents a new era in data representation for performance-based earthquake engineering, where observations of element damage can be continuously uploaded by researchers and practitioners into an opensource database, which then in turn directly feeds into new fragility models for use in performance based assessments. Now building assessments can benefit from the most up to date element performance models through a thoughtful and shared data structure that promotes model transparency and data reuse.
Graduate Student Researcher
UC Berkeley
Collective Behaviors and Phase Transitions in Regional Seismic Risk Metrics
Co-Authors: Raul Garcia (Rice University), Jamie Padgett (Rice University), and Ziqi Wang (UC Berkeley)
Abstract: As the Performance-Based Earthquake Engineering framework extends to a regional scale, significant research has focused on regional seismic response analysis, primarily through scenario-based simulations. While these simulations allow for computing decision variables for specific earthquake scenarios, assumptions made at the component level may lead to significant errors and variabilities in global quantity predictions. For instance, according to the Central Limit Theorem (CLT), the substantial independence assumption among structural damage states can result in Gaussian distributions for aggregated decision metrics. However, correlations between damage states and the uncertainty in models and damage data can lead to emergent patterns in collective behavior. Hence, the global risk metric becomes complex, and the CLT no longer applies. This study investigates such collective behaviors, particularly first- and second-order phase transitions in global risk metrics as we introduce and systematically treat various sources of uncertainty and correlations in the risk assessment pipeline. The first-order phase transition involves an abrupt shift in the mode of the global risk metric, while the second- denotes a fundamental change in the distribution shape of the metric. We demonstrate these phenomena by considering uncertainties in the pairwise damage state interaction and the fragility model. Two illustrative examples are presented: the Pacific Heights neighborhood in San Francisco and the Jonesboro transportation network, leveraging computational tools and capabilities within the SimCenter and the DesignSafe Cyberinfrastructure. Additionally, we present the Ising model from statistical physics to describe these phase transitions and explain the engineering significance of such emergent properties.