Data and AI for Decision-Support and Policy May 30, 2025 Image: Whitehall Street Station Flooding LEAP Wallerstein Panel Series: AI + Extreme Weather Preparedness Based on panel presentations and discussions by Dr. Terri Adams-Fuller, Dr. Equisha Glenn, and Dr. Miguel-Ángel Fernández-Torres. Machine learning (ML) and artificial intelligence (AI) offer new avenues to provide data and build decision support tools that can facilitate extreme weather preparedness and life-saving decision-making in the case of disasters. However, despite these technological advances, in many instances, there are still persistent challenges in creating usable data for decision makers in practice. At the LEAP Wallerstein Panel on AI + Extreme Weather Preparedness, experts from academia and public planning came together to discuss the use of AI/ML for real-world decision-making for disaster management and climate resilience. Panelists included Dr. Terri Adams-Fuller, a Professor of Sociology at Howard University specializing in emergency management and crisis decision-making; Dr. Equisha Glenn, a Senior Planner on the Climate Resilience Planning team at the New York City Metropolitan Transportation Authority; and Dr. Miguel-Ángel Fernández-Torres, an Assistant Professor at the Signal Theory and Communications Department at the Universidad Carlos III de Madrid and co-lead of the Working Group on Data at the Global Initiative on Resilience to Natural Hazards through AI Solutions. The following is a synthesis of themes and ideas from their discussion. AI, Data, and Disasters “In disasters, data is fragmented. You need to think about how these tools will be used in a fast-moving crisis environment”. ~ Dr. Terri Adams-Fuller Disaster management involves complex, often life-saving, decisions to be made in uncertain, evolving, and resource-limited circumstances–all while navigating public perceptions of risk and trust (AI Institute for Societal Decision Making, 2025). AI’s superpower lies in its ability to process large volumes of multi-modal information to support human reasoning under time constraints while also accommodating uncertainty. Each phase of the disaster management cycle–mitigation, preparedness, response, and recovery–requires different tools, datasets, and approaches, all of which can benefit from the application of AI (Table 1). Table 1. AI and the Disaster Management Cycle Bridging the Gap between Data, Tools, and Decision-making Despite AI-backed advances in data, modelling, and analysis, the practical application of decision support tools during emergencies remains inconsistent. Tools are not always designed with integration and field usability in mind, and there is often a lack of centralized systems for disseminating new tools to disaster teams, highlighting communication and coordination issues. Many decision support tools fail to meet the needs of end users—often because they are developed in isolation from the people who will ultimately rely on them. Engaging with end users throughout the development process is critical in building effective tools and building trust in the decision-making process. Fortunately, there are examples that demonstrate the benefits of integrating end users into the development process. The New York City Panel on Climate Change (NPCC) The New York City Panel on Climate Change (NPCC) serves as a model for how scientists and decision-makers can work together to produce actionable, policy-relevant climate data. The NPCC, created in 2009 and codified in Local Law 42 of 2012, has the mandate to provide an authoritative and actionable source of scientific information on future climate change in New York City and its potential impacts. It is an independent advisory board that provides regular assessments, facilitates interagency communication, and shares knowledge to help policymakers develop resilience strategies. The NPCC is instrumental for climate resilience planning in New York City. The dedicated science advisory board focuses on local characteristics and produces downscaled climate models at a spatial resolution relevant for municipal-level decision-making. Its outputs directly informed the Metropolitan Transportation Authority’s (MTA) first Climate Resilience Roadmap in April 2024. The NPCC formal integration into law illustrates the potential for institutionalizing localized, data-driven processes and scaling research-informed policy. Figure 1. MTA Climate Resilience Roadmap Coastal Surge Map using NPCC analyses. NOAA Testbeds Another example is NOAA’s Testbeds, where new tools, models, and technologies are tested for practical application. Collaborative research-to-operations environments where researchers and forecasters work together to enhance weather prediction systems, these testbeds facilitate the integration of new observing systems into models, streamline data assimilation methods, and test model improvements to benefit public weather services. Translators and Boundary Spanners An interdisciplinary project manager—or translator—who focuses on the “big picture” of linking data and tools to decision-making and policy can play a vital role in bridging the gap between developers and end users. Often referred to as boundary spanners, these individuals facilitate communication between information producers and users, straddling the divide between the two groups (Safford et al., 2017). Their work ensures that developers are connected with the needs of on-the-ground users, integrating socio-cultural perspectives and decision-making contexts alongside technical inputs to improve the relevance, usability, and effectiveness of decision support tools. How AI itself can be used to improve accessibility Once a tool or database is built, end users may have trouble understanding how to effectively use it. Much in the same way as a house can be “staged” to give future tenants an idea of how to use the space, AI can help “stage” possible use cases of disaster risk reduction or resilience tools with interactive prompts, guided walkthroughs, or real-world scenarios. For example, a city planner trying to reduce future heat stress in their community could prompt an AI-backed use case generator to “show me how this map can inform heatwave planning for 2050” in order to better understand how a tool can be used. Trust in AI Solutions for Disaster Risk Reduction Trust is a barrier to the use of AI in disaster management and long-term resilience planning. Tools must not only perform technically but also be perceived as reliable and ethically sound by the users deploying them. Emergency managers, for instance, handle sensitive information under public and institutional scrutiny. Any failure or misjudgment can carry serious consequences, making the adoption of new tools, particularly those powered by AI, contingent on transparency and confidence in their outputs. Ethical and privacy concerns around data use further complicate deployment, especially in communities with a historical distrust of institutions. Carnegie Mellon University’s AI Institute for Societal Decision-Making is advancing AI systems that offer data-driven recommendations to enable socially accepted, effective, and ethical interventions for emergency managers. Emerging technologies require guidelines that integrate science, technology, and expertise to enable interoperability, regulation, and capacity-building (International Telecommunication Union, 2022). A joint report by the International Telecommunication Union, World Meteorological Organization, and the UN Environment Programme Focus Group on AI for Natural Disaster Management (now the Global Initiative on Resilience to Natural Hazards through AI Solutions) highlights the need for international standards to ensure reliability, transparency, and acceptability of AI solutions for managing natural hazards. This standardization effort further aims to increase collaboration between stakeholders to reduce potential duplication of efforts, leverage synergies, identify trends, and address important gaps where standardization is required to advance effective and trustworthy AI solutions. Key Takeaways To improve the efficacy of AI-backed decision support tools for disaster management: IDENTIFY the users (i.e., decision makers) INTEGRATE the users early and often through an ITERATIVE process COMMUNICATE across producers and users via interdisciplinary translators Communication and trust remain among the biggest challenges in disaster management today. The design and deployment of decision support systems must be a collaborative process, integrating end users into the development of disaster tools to improve usability. To scale effective, data-driven decision-making, interagency co-production processes are also essential. The trustworthiness in these systems is built not only through performance but through inclusive, iterative design processes that foreground human judgment, institutional context, and local knowledge. This was produced by the National Science Foundation’s Center for Learning the Earth with Artificial Intelligence and Physics (LEAP) (Award #2019625) in collaboration with the National Center for Disaster Preparedness (NCDP). References AI Institute for Societal Decision-making. (2025). Accessed 20 May 2025, https://www.cmu.edu/ai-sdm/index.html Global Initiative on Resilience to Natural Hazards through AI Solutions. (2025). Accessed 20 May 2025 https://www.itu.int/en/ITU-T/extcoop/ai4resilience/Pages/default.aspx/ International Telecommunications Union. (2024). Disaster management: The standards perspective. Focus Group on AI for Natural Disaster Management (FG-AI4NDM). Accessed 20 May 2025 https://www.itu.int/net/epub/TSB/2024-T24-T-TUT-Disaster%20Management/index.html#p=1 International Telecommunications Union. (2022). Standardization Roadmap on Natural Disaster Management: Trends and Gaps in Standardization. Technical Report ITU-T FG-AI4NDM-Roadmap. Accessed 20 May 2025 https://www.itu.int/dms_pub/itu-t/opb/fg/T-FG-AI4NDM-2022-PDF-E.pdf Metropolitan Transportation Authority. (2024). Climate Resilience Roadmap. Accessed 20 May 2025 https://www.mta.info/document/136871 NOAA Weather Program Office. (n.d.) Testbeds: Research Meets Operations in the Test Environment. Accessed 20 May 2025 https://wpo.noaa.gov/testbeds/ NYC Mayor’s Office of Climate and Environmental Justice. (2025). New York City Panel on Climate Change. Accessed 20 May 2025 https://climate.cityofnewyork.us/initiatives/nyc-panel-on-climate-change-npcc/ Safford, H. D., Sawyer, S. C., Kocher, S. D., Hiers, J. K., & Cross, M. (2017). Linking knowledge to action: the role of boundary spanners in translating ecology. Frontiers in Ecology and the Environment, 15(10), 560-568. Geneva List Senior Staff Associate Sumana Palle Tags: AI, early warning systems, climate disasters Event: Coastal Surge Map Climate Change Disasters Systems Readiness