Research and Practice Highlights

In 2024 and 2025, Ohio experienced an unprecedented sequence of severe weather events, including record numbers of tornadoes, damaging windstorms, and hail events, with Franklin County emerging as one of the most affected areas. These repeated disasters disproportionately impacted socioeconomically disadvantaged communities, particularly residents of the 43223 and 43204 ZIP codes, where household incomes are substantially lower than county and national averages and where racial and ethnic minority populations are overrepresented. Building on prior findings that documented elevated levels of anxiety, depression, and post-traumatic stress disorder (PTSD) among low-income residents following the 2024 storms, this study investigates the cumulative mental health impacts of repeated disaster exposure through mid-2025. Using a mixed-methods, longitudinal design, the research examines how recurring severe wind and tornado events exacerbate pre-existing mental health challenges, influence coping strategies, and shape community resilience over time. Quantitative data will be collected through surveys administered to 385 participants, incorporating validated mental health instruments (GAD-7, PHQ, PSS-SR), housing condition assessments, and hazard exposure measures. Qualitative data will be gathered through semi-structured interviews and community workshops to capture lived experiences, adaptive behaviors, and local perspectives on recovery. Particular attention is given to evolving coping mechanisms, including the potential role of marijuana use following its recent legalization in Ohio. By integrating individual-level mental health outcomes with housing conditions, socioeconomic vulnerability, and repeated disaster exposure, this study generates actionable evidence to inform disaster recovery planning, mental health interventions, and resilience-building strategies.

Evidence shows that climatic events are altering existing human mobility patterns. There is growing recognition that climate-related factors do not drive migration in isolation. Rather, environmental change interacts with existing non-climate drivers to shape the nature, direction, volume, and composition of mobility. While existing research has identified common decision-making factors driving migration, two gaps remain. First, we lack understanding of how different factors interact to produce mobility outcomes, especially as it relates to their intersection with increasing climate hazards and vulnerabilities. Second, migration literature exhibits a mobility bias, leaving non-migration understudied. Moreover, research on immobility needs to move beyond the “trapped population” lens, given that the aspirations of people to stay lead to in-situ adaptations. We address these gaps by examining Lower Plaquemines Parish (LPP), Louisiana, a community that has experienced multiple catastrophic hurricanes alongside environmental and economic stressors and is increasingly vulnerable to saltwater intrusion and coastal inundation. Since the 1930s, the state of Louisiana has lost approximately 2000 square miles of land (CPRA), driven by land subsidence, erosion, and relative sea-level rise. Layered on these chronic disturbances are acute shocks such as coastal storms and hurricanes, which severely damage homes, businesses, infrastructure, and the social structure of communities. To understand this dynamic, we are conducting in-depth interviews with 150 residents who have migrated out of LPP over the past 20 years and those who have remained. We will use Qualitative Comparative Analysis (QCA) to identify how multiple factors interact to shape migration and non-migration.

As interconnected hazards increasingly compound impacts on infrastructure, economies, and communities, they are reshaping disaster response and recovery processes. Yet, a comprehensive understanding of the role of social capital, a key mechanism in these processes, across cascading climate disasters and diverse population groups remains limited. We develop an integrative framework to assess the effectiveness of social capital across key response and recovery domains, including economic recovery, resource mobilization, evacuation and relocation, health and well-being, and housing. The framework integrates disaster phases, contextual conditions, and methodological approaches, and introduces a graded scale capturing variation in social capital’s impacts, enabling systematic comparison across studies and outcomes. The study draws on a systematic review and thematic analysis of approximately 40 peer-reviewed articles selected through rigorous inclusion criteria following ROSES guidelines and analyzed using MAXQDA. Findings reveal convergence and divergence in how social capital operates across disaster contexts and among communities with different vulnerabilities. While prior research has largely emphasized positive effects, the evidence also identifies no negative, and modest positive impacts, indicating a more uneven pattern than commonly assumed. We develop policy-relevant recommendations for disaster managers, practitioners, and decision-makers in the United States, emphasizing the need to scale effective forms of social capital while addressing its limitations, systemic constraints, and potential adverse effects. These findings show that social capital is not uniformly beneficial and that disaster policy and practice must distinguish between its positive, limited, and negative dimensions when designing interventions to promote equitable resilience.

Wildfire evacuation outcomes vary due to interactions among fire progression, human behavior, and traffic dynamics. Most models treat these factors separately, leaving a gap in integrated approaches. This study developed a tri-coupled framework linking fire spread, multi-channel communications, and agent-based traffic modeling. The framework integrated a fire model, a communications model capturing varied cognition times, and a spatial-queue traffic model, applied to three Marin County communities. Using a multi-scalar Strategic Action Fields perspective (macro, meso, micro), the analysis showed how asynchronous communication, background traffic, and localized behaviors interacted to shape outcomes. Evacuation efficiency depends on communication timing and coordination, with cognitive delays producing nonlinear congestion. Peak traffic reduced capacity, underscoring the need for governance mechanisms separating emergency from routine flows. At the community level, demographic, infrastructural, and temporal characteristics shaped performance, requiring geographically targeted strategies. Micro-level analyses highlighted the disproportionate impact of everyday responsibilities, such as child pickup, on congestion, emphasizing behavioral realism in planning. The study offered implications for embedding adaptive communication, dynamic traffic management, and behavioral insights into institutionalized crisis information governance. The findings revealed that evacuation capacity functions as a form of access to safety, unevenly distributed by communication access, mobility constraints, and network characteristics. Linking these factors to continuous, data-driven evaluation can enhance wildfire preparedness while creating scalable models for other hazard-prone regions, advancing resilience and equity in evacuation systems. We acknowledge Professor Kenichi Soga’s research group, University of California, Berkeley, Sonoma Tech, Marin Wildfire Protection Authority, and Fehr and Peers Engineering for their support.

To improve disaster risk management in river basins, where most flood-related casualties occur, the local governments in Korea have installed smart technology-based monitoring systems in approximately 10% of the 22,300 rivers nationwide. These systems measure real-time water surface elevation and discharge without direct contact with water, relying exclusively on surface velocity measurements. Accurate discharge estimation therefore requires conversion of surface velocity to mean velocity using an appropriate mean-to-surface velocity conversion coefficient. The conversion coefficient is influenced by various factors, including flow fluctuation, bed conditions, and slope, and is typically determined through direct measurement of depth–velocity profiles. However, because direct measurements are often difficult or unsafe during flood conditions, a constant value of 0.85 or values estimated using empirical equations are commonly applied. Existing studies show that in steep and hydraulically complex rivers, the conversion coefficient is not constant, highlighting the need for a diversified and river-specific estimation approach. The research determines the mean-to-surface velocity conversion coefficient using measured surface velocity distributions at different water depths while accounting for flow characteristics such as velocity distribution, bed condition, and slope. The applicability of existing methods is evaluated using the derived coefficients, and the equation based on Manning’s formula shows the highest accuracy among existing approaches. Furthermore, a new prediction method is developed using hydraulic variables that can be easily measured in rivers. Depth–velocity profiles and hydraulic data are collected at 19 rivers using an RS-5 radar velocimeter. Dimensionless variables are derived through dimensional analysis, and a nonlinear regression model is developed to estimate the conversion coefficient. Validation results show the highest accuracy of 89.47%, indicating excellent reproduction of the mean–surface velocity relationship in rivers. These results demonstrate the practical applicability of the proposed conversion coefficient method to smart technology-based river monitoring systems.

Transition from individual recognition of risk to collective cognition and action in confronting large-scale extreme events is a critical, but problematic process. Could this transition be structured before an extreme event to inform collective action when a sudden hazard strikes? Collective cognition is defined as shared intelligence that emerges from the iterative exchange of knowledge among actors exposed to the same hazard, creating a profile of risk that structures action for the whole community. The experience of two counties in northern California exposed to the same risk of wildfire, Alameda and Marin, is analyzed in a comparative case study to identify how separate organizations become a “system” that translates cognition of risk into action. We conducted a set of 60 semi-structured interviews, using a purposive survey of local managers and experienced personnel representing organizations with responsibilities for wildfire protection in the two counties. The interviews identified information sources and communications processes used in actual practice. Data elicited from the expert interviews were entered into network analyses to compare performance between the two counties. The network analyses identified critical points of connection or disconnection among the organizations involved and were validated by calculating an External/Internal Index to assess the rate of change in performance, by sector and jurisdictional level of authority. Results revealed the emergence of a sociotechnical lattice structure for continuous social learning to enable different components of the counties to align their performance and increase their capacity to act as a system when under threat. This research received support from NSF grant #2230636.

This paper focuses on the experiences, complexities, resiliency, and accomplishments of the 501(c)(3) nonprofit, the Eleventh Street Bridge Community Long-Term Recovery Group (ESBC-LTRG), which is Delaware's first long-term recovery group (LTRG). The ESBC-LTRG was founded in the fall of 2021 in Wilmington, Delaware, following the monumental destruction and flooding caused by Hurricane Ida. Headed by a resident whose home was impacted, the ESBC-LTRG was incorporated as a non-profit organization in the summer of 2022 and has since brought over 250 residents who were affected or displaced by the storm back to a safe and secure means of daily living. There is a significant emphasis on preparation and response within emergency management, leaving much to be desired regarding recovery, especially long-term recovery. Therefore, we conducted a community-based research project through qualitative interviews with board members of the ESBC-LTRG. Additionally, the chair of the ESBC-LTRG, who is a coauthor on this paper, has provided a written reflection to highlight her lived experience of the flooding event, and the formation of the LTRG. This paper sheds light on the impacts of designating responsibility for long-term recovery to community members who may severely lack resources and funding. Despite these challenges, this paper also seeks to share knowledge that has been foundational to the ESBC-LTRG’s success, potentially lessening the burden for future LTRGs to come.

This research advances tornado risk reduction by integrating human behavior into an agent-based modeling (ABM) framework to evaluate shelter access and recommend strategies for improvement. The project is currently finalizing dual-model decision logic that captures both time-sensitive and general sheltering behaviors. Using survey data from over 2,400 households, behavioral decision rules are represented probabilistically, allowing each agent to make independent choices based on its unique attributes. In the time-sensitive model, agents choose among staying at home, leaving for a community shelter, or waiting, with distance to shelter and housing characteristics acting as primary predictors. The modeling sequence follows a structured approach: each household agent is assigned housing attributes from a spatial building inventory—including structural type (wood, steel, or masonry), occupancy type, number of stories, footprint, and basement access— along with key behavioral and socioeconomic factors derived from survey data, including concern, trust, and household and income characteristics—before receiving a simulated tornado warning. Decisions are executed through a hierarchical probabilistic approach, where agents sample actions from survey-derived distributions. For agents choosing to leave for a community shelter, the ABM simulates movement across real-world road networks to assess whether safe occupancy is achieved before tornado arrival. This framework also links building archetypes to a fragility-based simulation to evaluate protection levels. The results feed into evidence-based recommendations for emergency managers to both improve public safety education for modifying sheltering behaviors and inform the strategic siting of new shelter locations for more equitable tornado risk mitigation planning.

We adapted the American Society of Civil Engineers Standard 24: Flood Resistant Design and Construction (ASCE 24-24) building elevation requirements into a practical decision support tool that improves reliability in floodplain management and building design. The tool is intended to reduce the computational burden on planners and engineers who want to implement more protective elevation requirements for improved flood risk mitigation. The tool was designed and implemented using a framework from .NET Core 6 Blazor software to connect the ASCE 24-24 elevation computation to a user-friendly interface. The tool allows users to determine minimum required elevations by incorporating key considerations in building elevations. The computational framework applies to both riverine and coastal flood scenarios. It calculates multiple design relevant metrics and compares them against multiple freeboard standards to identify the governing requirement. Upon testing, the tool accurately reported the minimum required elevation and the height above base flood elevation for both riverine and coastal flood conditions. The results show that the tool reduces ambiguity in interpreting complex elevation tables, improves transparency in determining final elevations, and supports more uniform application of standards across jurisdictions. This work improves risk informed decision-making for planners, engineers, and policymakers by clarifying how multiple flood hazard parameters influence building design elevation and strengthening the implementation of flood resilience strategies.

Streamflow drought—periods of abnormally low river flows—poses significant challenges for water resource management, yet operational forecasting tools remain limited compared to flood or meteorological drought prediction. The U.S. Geological Survey (USGS) has developed River DroughtCast, an experimental web-based platform providing sub-seasonal to seasonal forecasts of streamflow drought across the conterminous United States. Leveraging machine learning (ML) approaches, including Long Short-Term Memory (LSTM) neural networks and Light Gradient-Boosting Machine (LightGBM), the system predicts weekly streamflow percentiles up to 13 weeks ahead for over 3,000 gaged locations. Forecasts target drought intensities defined by percentile thresholds (moderate: 20%, severe: 10%, extreme: 5%) and incorporate seasonally varying baselines to capture hydrologic variability. Evaluation against benchmark models (persistence, ARIMA) demonstrates that ML methods outperform traditional approaches for continuous low-flow prediction and drought onset/termination, particularly at shorter lead times (1–4 weeks). The tool integrates interactive visualizations to support decision-making by water managers, enabling early warning for drought onset, duration, and severity. This work represents a critical advancement in operational hydrological drought forecasting, addressing gaps in national-scale preparedness and resilience planning.

This project employs a mixed-methods approach to document the challenges and nuances broadcast meteorologists face when communicating risk for compound, multi-hazard tropical events to Spanish-speaking populations. The research integrates ethnographic observations at multiple broadcast stations, a national survey, and targeted focus groups to capture the specific difficulties weathercasters encounter when generating accurate translations and consistent naming conventions. By analyzing these diverse datasets, the study identifies critical gaps in how linguistic nuances impact the delivery of life-saving information. The findings will be transitioned into research-to-operations (R2O) products designed to directly inform and improve how broadcast meteorologists shape their communication strategies. Ultimately, this work provides the National Weather Service and its partners with the evidence-based tools necessary to better convey multi-hazard risks to bilingual and Spanish-speaking audiences. By refining these frameworks, the project seeks to enhance community resilience and ensure that high-impact tropical weather information is clear, culturally relevant, and scientifically precise for underserved linguistic populations.

Due to climate change, annual rainfall and the likelihood of extreme precipitation events are projected to increase in the future period (2021–2100), raising concerns about the intensification of flood damage. This research assesses flood risk at the administrative district level based on three key components—Hazard, Exposure, and Vulnerability—using a total of 13 detailed indicators that reflect meteorological factors, historical flood records, population data, and regional characteristics. In particular, future changes in flood risk under climate change were quantitatively evaluated using SSP climate change scenarios. Under current conditions, approximately 25% of administrative regions are classified as high or extremely high flood risk areas; however, under future conditions, this proportion increases to as much as 42%, depending on which SSP (Shared Socioeconomic Pathways) scenario is employed, The flood risk assessment results and high-risk area classifications presented in this study can serve as foundational data for more detailed flood risk analyses and are expected to provide practical evidence for establishing flood risk management policies and prioritizing investments in flood control infrastructure at both national and local government levels in South Korea.

Working with the Pacific Urban Resilience Lab (PURL) and the National Disaster Preparedness Training Center (ndptc.hawaii.edu), our team at the University of Hawaii is conducting research and development of training and education for emergency managers, first/early responders, and community leaders on extreme weather hazards including extreme heat, lightning, tornado, cloudbursts and wildfire/Air quality). In addition to preparedness for events such as the World Cup and Olympics, the research and training focus on interdisciplinary risk management and mitigation and adaptation strategies that can reduce harm to at-risk, vulnerable, and underserved communities. The effort includes improving scenario planning to understand and manage risks, uncertainties, and interventions designed to improve safety, security, resilience, and sustainability. We are interested in collaborating with researchers and practitioners working on positive collective outcomes through research, education, and training.  

This research employs a grounded theory approach to examine the structural efficacy of the South Korean disaster response framework. While formal protocols are well-established, gaps between policy mandates and on-the-ground execution frequently emerge during large-scale crises. By conducting a systematic qualitative case study of recent disaster events, this study analyzes extensive archival data—including government white papers, investigative audit reports, and official documentation—to uncover the underlying mechanisms of institutional performance. Using axial and selective coding techniques, this study identifies a "fragmentation of inter-agency cooperation" as the core phenomenon hindering effective response. The findings reveal that hierarchical rigidities and misaligned communication channels often impede real-time decision-making, leading to critical delays. Conversely, the analysis highlights that resilience is achieved when local command units possess the flexibility to adapt to unfolding environmental conditions. This research proposes a refined theoretical model for a "Resilient Governance Framework," suggesting that shifting from a rigid, top-down structure toward an agile, high-reliability organization (HRO) model is essential for disaster management. By synthesizing empirical data through a grounded theory lens, this study offers actionable insights for policymakers to enhance the functional integration of national safety systems. These findings contribute to a broader understanding of how institutional design influences disaster outcomes in complex, centralized governance systems.

Precipitation-triggered geohazards such as debris flows, landslides, floods, erosion, and scour pose frequent and spatially distributed risks for critical infrastructure. Managing these risks via avoidance or mitigation over large areas calls for an integrated, physics-based probabilistic approach. We are developing spatially distributed models, starting with post-wildfire and non-wildfire, precipitation-induced debris flow hazard, and decision-support tools for asset risk management and emergency preparedness and response. Our Precipitation Risk Management (PRiMa) Program focuses on scalable identification, monitoring, and prioritization of geohazards and timely dissemination of this information to risk managers, asset integrity managers, and emergency preparedness and response teams. Our debris flow models utilize soil, burn-severity, terrain, and precipitation data and output volumetric probability estimates throughout the area of interest. Probabilistic approaches help capture uncertainties and allow for risk-appropriate conservatism while producing actionable results. We developed our probabilistic model using a Monte Carlo approach but to attain operability for rapid modeling and dissemination, we adopted a first order second moment approximation that produces results comparable to Monte Carlo simulations. In addition to developing baseline models for storm scenarios we are developing workflows to utilize these models on a storm-by-storm, near-real time forecast basis for emergency preparedness posture. Furthermore, we are adapting climate change scenarios to produce forward-looking models to gain insight into the effects of changing climate on debris flow hazards and develop long term strategies for infrastructure development and management while increasing safety and reducing costly impacts.

NIH is committed to improving the integration of environmental exposure data into clinical and epidemiological health research, especially in the area of extreme weather and natural disasters. The new NIH Data Accelerator Program builds on work supported by the Department of Health and Human Services Office of the Secretary Patient-Centered Outcomes Research Trust Fund in furthering efforts to facilitate research connections between environmental exposures, such as poor air and water quality and extreme temperatures, and health outcomes. These evolving investments in data infrastructure and resources are intended to address the many challenges of analyzing environmental exposure and health data, such as identifying spatially and temporally relevant data, harmonizing measures, and accessing and linking to protected health information. The tools and resources include a curated data catalog and associated metadata, software and tutorials to assist researchers in handling large geospatial environmental data, and standardized datasets of health-environment linkages. The CHORDS data catalog is available at the CAFE Research Coordinating Center Dataverse collection, along with over 1000 other health and extreme weather datasets. Example use case analyses of wildfire smoke exposures and health using Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) data as well as NIH cohort data demonstrate the practical application and utility of these data resources. By improving the resources and approaches to integrating environmental determinants of health data into health data systems, these NIH efforts aim to reduce the technical barriers that researchers face as they work to enhance patient and population health.

UrbEaT project advances hazards and disasters research by integrating environmental modeling with human perception data to better understand and mitigate urban heat island (UHI) risks in rapidly growing cities. It studies how urban greenery can reduce heat-related hazards while aligning it with residents’ lived experiences. By combining physical and social dimensions of risk, the project contributes actionable insights for climate adaptation and urban planning. The project investigates how urban development influences the intensity and spatial–temporal dynamics of UHIs in Beijing and Warsaw, and evaluates how green infrastructure can mitigate these effects. The research employs a mixed-methods design. First, high-resolution land-use and spatial simulations are used to model UHI patterns and assess vegetation's cooling potential. Second, empirical data on residents’ thermal comfort and perceptions of ecosystem services are collected through a mobile application and surveys conducted during the summer months. By linking environmental data with human-centered data, the project produces planning guidelines and policy recommendations. Its significance lies in addressing heat as an increasingly critical urban hazard under climate change, and in offering scalable strategies to enhance urban resilience and reduce heat-related health risks across diverse socio-environmental contexts.

Extreme hurricanes pose a growing threat to coastal and inland transmission infrastructure, yet system-level resilience assessments often rely on empirical outage models that weakly represent the underlying structural failure mechanisms. This study presents a physics-informed resilience assessment of the IEEE 39-bus East Coast power system under hurricane hazards by explicitly coupling transmission-line fragility with network-level power-loss consequences. A spatially explicit representation of the 39-bus system is constructed using georeferenced tower, bus, and substation locations and integrated with the NHERI SimCenter Regional Resilience Determination Tool (R2D) hurricane scenario module. Hurricane wind fields are simulated across the network, and component-level failure probabilities are evaluated using wind-speed-dependent fragility curves for lattice transmission towers. Monte Carlo simulations propagate these component failures through the network topology to quantify cascading line outages, load shedding, and loss of served demand. System performance is evaluated using resilience curves that capture the temporal evolution of power delivery degradation and recovery as a function of hurricane intensity. The resulting metrics quantify both the magnitude and duration of service loss, enabling comparison across storm scenarios and identification of structurally critical corridors whose failure disproportionately drives system-wide consequences. The results demonstrate that resilience loss is governed not only by storm intensity but by the spatial co-location of vulnerable structures and topologically critical transmission paths. By linking structural reliability directly to power system consequences, this work provides a unified framework for risk-informed planning, targeted hardening, and resilience-oriented decision-making for transmission networks exposed to high-impact, low-probability hurricane events.

Operational flood response requires both timeliness and physical fidelity, yet existing modeling approaches force decision-makers to choose between the two. Physics-based hydrodynamic models predict flood depth and velocity accurately, but can require 6-12 hours to run. State-of-the-art rapid assessment tools sacrifice physical accuracy for speed, typically providing only inundation extent without the velocity data essential for damage assessment and life-preserving decisions. ML-HYDRAS (Machine Learning Hydrodynamic Driven Rapid Assessment of Storms) addresses this operational trade-off through a physics-informed machine learning framework. The approach trains a physically-consistent momentum-conserving neural emulator on synthetic datasets generated from high-fidelity HEC-RAS (Hydrologic Engineering Center's River Analysis System) hydrodynamic simulations spanning diverse storm scenarios and watershed conditions, enabling predictions in a matter of minutes rather than hours. ML-HYDRAS predicts both flood depth and velocity fields, enabling damage forecasts based on the multiple factors that cause infrastructure failure rather than depth alone.ML-HYDRAS will support decision-making throughout the disaster lifecycle. Before events, the system will enable rapid scenario evaluation for infrastructure vulnerability assessment and evacuation logistics optimization. During active flooding, it can provide dynamic hazard assessments that inform resource allocation and infrastructure risk evaluation. After disasters, velocity and sediment transport predictions will support damage assessment and reconstruction prioritization. Initial prototypes will target recently impacted communities including Montpelier, Vermont and Asheville, North Carolina. By delivering physically consistent depth and velocity predictions in operationally relevant timeframes, this work demonstrates how physics-informed machine learning can provide practitioners with immediately deployable decision support while advancing community resilience against escalating climate-driven hazards.

The National Research Institute for Earth Science and Disaster Resilience is developing a disaster resilience index to empirically measure municipal-level indicators for Japan, employing national datasets with a conceptually grounded, top-down approach. Previous models, such as BRIC and ADRI, offer helpful foundations. However, their indicators need to be carefully adjusted to Japan's institutional, social, and legal contexts. This new framework defines disaster resilience as three empirically informed capacities: the ability to maintain essential functions, the ability to recover from disruption, and the ability to transform for better future conditions, as informed by the 2024 systematic literature review by Shiozaki et al. and the 2025 conceptual review paper by Nagamatsu. This approach will allow local authorities to better identify strengths and intervention points. Some key distinctions in the Japanese context include the strong coordination between local governments and civil society, and the role of national legal and institutional frameworks in shaping municipal disaster management. These frameworks further support the sharing of resources across municipal boundaries. By incorporating these features into the index, the project emphasizes a more networked and holistic notion of resilience. Empirical validation is currently in progress. A prototype is scheduled for completion by September 2026.

The Coastal Hazards Economic Prosperity and Resilience (CHEER) Summer Scholars program advances the training of a new generation of researchers in convergent disaster science by combining hands-on fieldwork, interdisciplinary mentorship, and community-engaged research. Now in its fourth year, this six-week residential program selects undergraduate and graduate students to conduct original research within CHEER's study areas in Eastern North Carolina, a region facing the economic, social, and personal costs of natural hazards, damaging hurricanes in particular. The competitive application process attracts students at various levels from multiple universities and disciplinary backgrounds. Selected students scholars are based in Greenville, North Carolina, with East Carolina University's campus as their research base. Situated in one of CHEER's three case study areas, the location enables close collaboration with government partners and community stakeholders across the region. The program's core curriculum includes fieldwork, seminars on the environment and community resilience, and meetings with stakeholders including local officials, resilience professionals, volunteer organizations, emergency responders, and residents. Students identify research topics aligned with CHEER's scope, receive weekly mentorship by CHEER-Hub members, and present findings to peers, stakeholders, and program personnel at the program's conclusion. Students’ projects capture local knowledge critical to developing realistic, applicable hazard models and demonstrate the program's role in bridging the gap between research and community realities. Their findings inform the science of the Cheer Hub, provide researchers with insight into at-risk communities’ challenges, priorities, obstacles, nuances of practice, and lived experiences. This program offers scholars unique opportunities to participate in community research and practice.

In the United States, households participating in a buyout can struggle to find alternate housing that is affordable. In particular, the lack of appropriate comparable affordable housing for low-income rural households who want to stay in the area, albeit outside the floodplain, illustrates the complex and multifaceted mitigation needs of rural coastal communities, with implications for housing policy. In this study, we look at the local housing market in two rural communities in the Carolinas to examine the price gaps that households in a buyout program face when looking to purchase homes outside high flood risk areas. Our study sites are Bennettsville, South Carolina, where the buyout is currently under consideration, and Tarboro, North Carolina, where the buyout occurred following Hurricane Floyd in 1999. The study uses a combination of quantitative (i.e., property transaction and tax value) and qualitative (i.e., focus groups and individual interviews) data and analysis. Findings indicate that home prices outside the floodplain far outstrip the buyout amounts offered to residential property owners participating in the buyout program. Combined with low housing supply and lack of alternate affordable housing units, buyout programs can lead to a population downturn in rural communities. We thus argue for the need to envision buyout programs as a broader housing policy issue and in turn extend the parameters of what is considered flood hazard mitigation and adaptation. This study is funded through a National Oceanic and Atmospheric Administration (NOAA) Climate Program Office’s Adaptation Sciences (AdSci) Research Program.

There are a number of studies that analyze flood preparedness in communities of the Global South. However, only a few studies have investigated flood preparedness in underserved communities with heightened vulnerability. This paper analyzed household preparedness against future floods in Kogi State Last-Mile Communities (LMCs) using 14 indicators and employed chi-square statistics to examine whether households’ socioeconomic characteristics influence their preparedness for future floods. Results show that across the LMCs in the study area, the preparedness against future floods depends on the gender of household heads. While access to flood sensitization resources from responsible agencies is poor, household heads remain a major source of flood education to LMCs. Recommendations and implications for harnessing local-level flood resilience and preparedness are discussed in this paper.

This article explores children’s experiences, knowledge, and altruistic behaviors in the COVID-19 pandemic. Our team sought to answer the following research questions: What did children do to help others during the pandemic? Who did they assist? Why were they motivated to act? We compiled and coded 115 news articles focused on U.S. children’s actions in the pandemic. Our analyses identified eight types of altruistic behaviors that children engaged in, which we grouped into two categories. The first category encompassed children providing material resources such as: making, collecting, or distributing supplies; raising and donating money; and cooking or distributing food. The second category involved children mobilizing to advance well-being by: creating art; offering social and emotional support; providing tutoring or developing other educational services; participating in public health campaigns or vaccination efforts; and conducting or taking part in research. Children sought to help many different people, ranging from family members and friends to at-risk professionals, such as frontline workers and healthcare providers. They were also attuned to the needs of socially or economically marginalized groups, such as older adults, low-income families, and unhoused people. Different factors motivated children to act, including personal experiences, connections to others, and more abstract empathy for those suffering disparate effects in the pandemic. This research found that the pandemic may have enhanced children’s disaster literacy through increasing their recognition of disaster injustice. Children understood the myriad threats associated with the pandemic and acted altruistically in response. Their actions were motivated in part by their recognition of the deeply unequal effects of the pandemic, thus suggesting the potential for a liberatory disaster literacy that is attentive to structural inequalities. Ultimately, this study suggests that experiencing the pandemic may have planted some seeds for growing a more disaster-literate group of young people in the future.

This research advances flood risk modeling by demonstrating that insured housing losses are not determined by rainfall alone and that model performance varies substantially across events and regions. Using National Flood Insurance Program (NFIP) claims data, this study evaluates how well regression and classification models explain census tract-level insured losses and whether models developed for one event can be applied to others. We analyzed 6,065 census tracts affected by three events—the 2016 Tax Day Flood in Texas, Hurricane Harvey in Texas, and Hurricane Irma in Florida—within a unified hazard, exposure, and vulnerability framework. Results show that more than half of the wettest areas are not the most damaged, indicating that exposure and vulnerability factors, particularly population density, play a dominant role in shaping losses. Ensemble classification models outperform regression approaches for predicting loss occurrence, while spatial lag regression improved explanatory power. Model transferability varies across events, with stronger performance observed within similar hazard contexts. These findings highlight that flood models are not one-size-fits-all and that locally calibrated approaches are necessary for decision-making. Ongoing work extends this framework by holding exposure and vulnerability constant while evaluating how changes in extreme precipitation under different global warming levels influence future insured losses at the tract level.

Emergency management agencies are integrating large language models (LLMs) into tabletop exercise facilitation and evaluation, yet institutional governance frameworks have not kept pace with this adoption. This study examines what agency administrators need to require before deploying LLMs in training environments, and proposes a practical policy framework to fill that gap. The study draws on cross-disciplinary analysis of artificial intelligence (AI) safety research and emergency exercise literature to identify conditions under which LLMs break down in simulation contexts. Three failure modes emerge consistently from this evidence base. LLMs deviate from human instructions when task-completion pressure accumulates. They produce misleading outputs when their actions are subject to accountability. Most counterintuitively, more capable models exhibit greater behavioral misalignment, not less, under these conditions. Emergency tabletop exercises create precisely these structural conditions, yet neither federal nor state policy currently addresses any of them in training contexts. In response, this study proposes a three-component governance framework targeting agency administrators. The framework requires (a) mandatory human oversight at critical decision nodes during AI-facilitated simulations, (b) auditable and traceable LLM outputs as a baseline institutional standard, and (c) risk-tiered procurement criteria that align model selection with exercise stakes rather than defaulting to maximum model capability. This framework gives administrators a concrete policy foundation for responsible LLM adoption and contributes to a broader conversation about AI governance in high-stakes public sector training environments.

This work introduces a “diagnosis before repair” framework for housing risk and resilience, grounded in real-world observation of early housing conditions. Drawing from SaferHomes (a housing diagnostics practice) and the Shelter Projekt, which supports participant-led housing decisions and early intervention, it examines how weak signals in housing systems are noticed, understood, and acted on. The approach is based on applied experience across diverse housing contexts, where similar patterns appear: early signs are often present but not recognized, and action is delayed until problems become visible. The framework focuses on understanding underlying conditions before intervening, shifting attention from isolated symptoms to patterns across the system, including how these signals develop over time. Current housing responses tend to be reactive, triggered by visible damage rather than early conditions. This often leads to delayed decisions, fragmented interventions, and higher costs. These challenges are increasing with aging housing stock and climate-related stress, including flood risk. By prioritizing early understanding, this approach supports clearer decision-making, earlier intervention, and better alignment between homeowners, trades, and institutions. It offers a practical way to interpret housing conditions and supports more coordinated, condition-based responses to risk and resilience.

This project advances the disasters field by demonstrating how institutionalized wildlife response systems can strengthen outcomes for both ecosystems and communities. The Wildlife Response Operations Center (WROC), developed by the Oceans and Wildlife Institute (OWI), offers a replicable model for integrating wildlife and community resilience into formal emergency management frameworks through trainings and cross-sector collaboration. This practice-based initiative examines the design and implementation of the WROC in Corpus Christi, Texas, as the nation’s first wildlife-dedicated emergency operations center. Drawing on emergency management principles, including the Incident Command System, this model incorporates coalition-building processes, stakeholder coordination, and operational protocols. Through pre-disaster agreements with regulatory agencies, joint exercises, and establishment of a Training Center, OWI is modeling how to embed wildlife considerations into all-hazards planning frameworks. Preliminary findings indicate that formalizing partnerships among nonprofit responders, government agencies, veterinarians, and researchers improves coordination, reduces response times, and enhances capacity for large-scale wildlife incidents. The WROC model provides a scalable framework to address a persistent gap in emergency management by recognizing wildlife as integral to community resilience, advancing a more holistic, multi-species approach to disaster management.

Together with the American Institute for Avalanche Research and Education (AIARE), we are conducting a multi-phase evaluation of AIARE’s Pro 1 courses using data from the 2024-2026 seasons. The Pro 1 course is a standardized professional training program for operational avalanche risk management that is accredited and overseen by the American Avalanche Association (A3). It serves as an entry point to the avalanche workforce and a prerequisite for teaching public avalanche education, making it a focal point for evaluating professional practice in the avalanche sector. The evaluation examines participant proficiency outcomes within existing A3 standards. The first phase analyzes in-house data from 320 participants across 24 courses over three seasons. Quantitative analyses examine patterns in participants’ proficiencies across individual, course, and environmental variables, complemented by thematic analysis of participant and instructor feedback. Preliminary findings suggest that participants’ proficiency assessments are more strongly associated with course-level risk conditions than with individual attributes or administrative variables. Terrain exposure and avalanche hazard conditions appear to meaningfully shape assessment consistency, with more constrained conditions leading to reduced differentiation in ratings. This work contributes to ongoing efforts to strengthen reliable instruction and assessment of core competencies under varying operational risk conditions. We are particularly interested in connecting with others working on evaluation methods, competency assessment, and the design of professional risk management training.

The NIH Health and Extreme Weather (HEW) Program leads efforts to address knowledge gaps and develop strategies for prevention, preparedness, and solutions to save lives and improve the quality of life for people affected by extreme weather conditions. Integral to this effort is a network of over 20 exploratory research centers across the country focused on fostering transdisciplinary teams to generate new knowledge and action-oriented strategies to protect health and build resilience at the individual, community, national, and global levels. Additionally, four community engagement hubs, part of the Alliance for Community Engagement–Partnership for Action Toward Health (ACE-PATH), are working toward sustainable strategies that address the impacts of extreme weather on communities facing higher risk of environmental harms. NIH is also cultivating a robust network of scientists, health practitioners, students, community members, and educators who are actively engaged in the latest research into the health effects of extreme weather and natural disasters through the CAFE Research Coordinating Center. Key partnerships with the National Science Foundation through two centers known for their disaster response expertise—the University of Colorado Boulder Natural Hazards Center and University of Washing Natural Hazards Reconnaissance (RAPID) Facility—augment the capacity and resources of the HEW Program.

This project introduces a HyperNetwork framework for the rapid wind-resilience assessment of large building portfolios. Emergency planning and insurance mitigation require building-specific response estimates under extreme wind events. However, high-fidelity finite element simulations are computationally prohibitive at the city scale. Even simplified analysis often requires over one minute per building. While conventional neural-network surrogates accelerate predictions, they typically require retraining for every unique structure, limiting their utility across diverse urban inventories. To solve this, our new AI system instantly creates a custom predictive model for any building based on its basic characteristics. This eliminates the time-consuming need to train a new model for every single structure. The system automatically adjusts to provide accurate predictions for buildings of any height and can handle various types of physical stress, such as steady winds, shaking, or sudden impacts. We tested this approach on thousands of buildings, including a Times Square dataset. It proved highly accurate even on buildings it had never analyzed before. Most importantly, it can evaluate large groups of buildings 7,000 times faster than finite element method. By converting structural descriptors into deployable digital twins, this method supports real-time post-event impact assessments, portfolio-scale risk modeling, and smart-infrastructure monitoring workflows. The research team welcomes practitioner partnerships for pilot deployments and real-world testing.

This study has shown how physically possible but statistically rare flood events—the “worst credible cases”—can expose deeper racial, infrastructural, and insurance disparities than conventional probabilistic planning often reveals. Current flood risk planning in the United States relies heavily on probabilistic modeling, which can underestimate the severe consequences of rare rainfall events that exceed statistical expectations. This constraint can obscure the potential for catastrophic inundation in densely populated areas and place extreme but plausible events outside the realm of responsible planning. Using Houston, Texas, as a case study, we examined how a “worst credible case” rainfall scenario would affect two urban watersheds within the same flood control district but with very different histories of development, drainage infrastructure, and flood protection. The study used an integrated possibilistic flood modeling framework that combined high-resolution hydrologic simulations, block group-level sociodemographic data, and National Flood Insurance Program coverage rates. Results showed that extreme but plausible rainfall could dramatically increase flood depths and amplify racial disparities compared with historical flood events. The analysis also revealed the systemic vulnerability of older conveyance-only drainage networks and identified a structural flood insurance gap shaped by watershed-level histories of infrastructural neglect and exclusionary development. These findings suggest that responsible flood planning should account not only for what is statistically likely, but also for what is physically possible and socially consequential.