Search for: All records

Abstract For this study, we present and evaluate an improved agent-based modeling framework, the Forecasting Laboratory for Exploring the Evacuation-system, version 2.0 (FLEE 2.0), designed to investigate relationships between hurricane forecast uncertainty and evacuation outcomes. Presented improvements include doubling its spatial resolution, using a quantitative approach to map real-world data onto the model’s virtual world, and increasing the number of possible risk magnitudes for wind, surge, and rain risk. To assess model realism, we compare FLEE 2.0’s simulated evacuations—specifically its evacuation orders, evacuation rates, and traffic—to available observational data collected during Hurricanes Irma, Dorian, and Ian. FLEE 2.0’s evacuation response is encouraging, given that FLEE 2.0 responds reasonably and differently to all three different types of forecast scenarios. FLEE 2.0 well represents the spatial distribution of observed evacuation rates, and relative to a lower spatial resolution version of the model, FLEE 2.0 better captures sharp gradients in evacuation behaviors across the coastlines and metropolitan areas. Quantitatively evaluating FLEE 2.0’s evacuation rates during Irma establishes model errors, uncertainties, and opportunities for improvement. In summary, this paper increases our confidence in FLEE 2.0, develops a framework for evaluating and improving these types of models, and sets the stage for additional analyses to quantify the impacts of forecast track, intensity, and other positional errors on evacuation. Significance StatementThis paper describes and evaluates an updated version of a modeling system [the Forecasting Laboratory for Exploring the Evacuation-system, version 2.0 (FLEE 2.0)] designed to explore relationships between hurricane forecasts and evacuation impacts. FLEE 2.0’s simulated evacuations compare favorably with different types of observational evacuation data collected during Hurricanes Irma, Dorian, and Ian. A statistical comparison with Irma’s observed evacuation rates highlights uncertainties and opportunities for improvement in FLEE 2.0. In summary, this paper increases our confidence in FLEE 2.0, develops a framework for evaluating these types of models, and provides a foundation for additional work using FLEE 2.0 as a research tool. 

Abstract In addition to measuring forecast accuracy in terms of errors in a tropical system’s forecast track and other meteorological characteristics, it is important to measure the impact of those errors on society. With this in mind, the authors designed a coupled natural–human modeling framework with high-level representations of the natural hazard (hurricane), the human system (information flow, evacuation decisions), the built environment (road infrastructure), and connections between elements (forecasts and warning information, traffic). Using the model, this article begins exploring how tropical cyclone forecast errors impact evacuations and, in doing so, builds toward the development of new verification approaches. Specifically, the authors implement track errors representative of 2007 and 2022, and create situations with unexpected rapid intensification and/or rapid onset, and evaluate their impact on evacuations across real and hypothetical forecast scenarios (e.g., Hurricane Irma, Hurricane Dorian making landfall across east Florida). The results provide first-order evidence that 1) reduced forecast track errors across the 2007–22 period translate to improvements in evacuation outcomes across these cases and 2) unexpected rapid intensification and/or rapid onset scenarios can reduce evacuation rates, and increase traffic, across the most impacted areas. In exploring these relationships, the results demonstrate how experiments with coupled natural–human models can offer a societally relevant complement to traditional metrics of forecast accuracy. In doing so, this work points toward further development of natural–human models and associated methodologies to address these types of questions and improve forecast verification across the weather enterprise. 

In addition to measuring forecast accuracy in terms of errors in a tropical system’s forecast track and other meteorological characteristics, it is important to measure the impact of those errors on society. With this in mind, the authors designed a coupled natural–human modeling framework with high-level representations of the natural hazard (hurricane), the human system (information flow, evacuation decisions), the built environment (road infrastructure), and connec- tions between elements (forecasts and warning information, traffic). Using the model, this article begins exploring how tropical cyclone forecast errors impact evacuations and, in doing so, builds toward the development of new verification approaches. Specifically, the authors implement track errors representative of 2007 and 2022, and create situations with unexpected rapid intensifica- tion and/or rapid onset, and evaluate their impact on evacuations across real and hypothetical forecast scenarios (e.g., Hurricane Irma, Hurricane Dorian making landfall across east Florida). The results provide first-order evidence that 1) reduced forecast track errors across the 2007–22 period translate to improvements in evacuation outcomes across these cases and 2) unexpected rapid intensification and/or rapid onset scenarios can reduce evacuation rates, and increase traffic, across the most impacted areas. In exploring these relationships, the results demonstrate how experiments with coupled natural–human models can offer a societally relevant complement to traditional metrics of forecast accuracy. In doing so, this work points toward further development of natural–human models and associated methodologies to address these types of questions and improve forecast verification across the weather enterprise. 

Abstract Demands to manage the risks of artificial intelligence (AI) are growing. These demands and the government standards arising from them both call for trustworthy AI. In response, we adopt a convergent approach to review, evaluate, and synthesize research on the trust and trustworthiness of AI in the environmental sciences and propose a research agenda. Evidential and conceptual histories of research on trust and trustworthiness reveal persisting ambiguities and measurement shortcomings related to inconsistent attention to the contextual and social dependencies and dynamics of trust. Potentially underappreciated in the development of trustworthy AI for environmental sciences is the importance of engaging AI users and other stakeholders, which human–AI teaming perspectives on AI development similarly underscore. Co‐development strategies may also help reconcile efforts to develop performance‐based trustworthiness standards with dynamic and contextual notions of trust. We illustrate the importance of these themes with applied examples and show how insights from research on trust and the communication of risk and uncertainty can help advance the understanding of trust and trustworthiness of AI in the environmental sciences.