Artificial intelligence for risk analysis-A risk characterization perspective on advances, opportunities, and limitations.

Artificial intelligence (AI) has seen numerous applications for risk analysis and provides ample opportunities for developing new and improved methods and models for this purpose. In the present article, we conceptualize the use of AI for risk analysis by framing it as an input-algorithm-output process and linking such a setup to three tasks in establishing a risk description: consequence characterization, uncertainty characterization, and knowledge management. We then give an overview of currently used concepts and methods for AI-based risk analysis and outline potential future uses by extrapolating beyond currently produced types of output. We end with a discussion of the limits of automation, both near-term limitations and a more fundamental question related to allowing AI to automatically prescribe risk management decisions. We conclude that there are opportunities for using AI for risk analysis to a greater extent than is commonly the case today; however, critical concerns about proper uncertainty representation and the need for risk-informed rather than risk-based decision-making also lead us to conclude that risk analysis and decision-making processes cannot be fully automated.

A data-driven method for identifying the locations of hurricane evacuations from mobile phone location data.

How evacuations are managed can substantially impact the risks faced by affected communities. Having a better understanding of the mobility patterns of evacuees can improve the planning and management of these evacuations. Although mobility patterns during evacuations have traditionally been studied through surveys, mobile phone location data can be used to capture these movements for a greater number of evacuees over a larger geographic area. Several approaches have been used to identify hurricane evacuation patterns from location data; however, each approach relies on researcher judgment to first determine the areas from which evacuations occurred and then identify evacuations by determining when an individual spends a specified number of nights away from home. This approach runs the risk of detecting non-evacuation behaviors (e.g., work trips, vacations, etc.) and incorrectly labeling them as evacuations where none occurred. In this article, we developed a data-driven method to determine which areas experienced evacuations. With this approach, we inferred home locations of mobile phone users, calculated their departure times, and determined if an evacuation may have occurred by comparing the number of departures around the time of the hurricane against historical trends. As a case study, we applied this method to location data from Hurricanes Matthew and Irma to identify areas that experienced evacuations and illustrate how this method can be used to detect changes in departure behavior leading up to and following a hurricane. We validated and examined the inferred homes for representativeness and validated observed evacuation trends against past studies.

Power outage prediction using data streams: An adaptive ensemble learning approach with a feature- and performance-based weighting mechanism.

A wide variety of weather conditions, from windstorms to prolonged heat events, can substantially impact power systems, posing many risks and inconveniences due to power outages. Accurately estimating the probability distribution of the number of customers without power using data about the power utility system and environmental and weather conditions can help utilities restore power more quickly and efficiently. However, the critical shortcoming of current models lies in the difficulties of handling (i) data streams and (ii) model uncertainty due to combining data from various weather events. Accordingly, this article proposes an adaptive ensemble learning algorithm for data streams, which deploys a feature- and performance-based weighting mechanism to adaptively combine outputs from multiple competitive base learners. As a proof of concept, we use a large, real data set of daily customer interruptions to develop the first adaptive all-weather outage prediction model using data streams. We benchmark several approaches to demonstrate the advantage of our approach in offering more accurate probabilistic predictions. The results show that the proposed algorithm reduces the probabilistic predictions' error of the base learners between 4% and 22% with an average of 8%, which also result in substantially more accurate point predictions. The improvement made by our algorithm is enhanced as we exchange base learners with simpler models.

Using mobile phone data to evaluate access to essential services following natural hazards.

Natural hazards bring about changes in the access to essential services such as grocery stores, healthcare, schools, and day care because of facility closures, transportation system disruption, evacuation orders, power outages, and other barriers to access. Understanding changes in access to essential services following a disruption is critical to ensure equitable recovery and more resilient communities. However, past approaches to understanding facility closures and inaccessibility such as surveys and interviews are labor-intensive and of limited geographic scope. In this article, we develop an approach to understanding facility-level inaccessibility across a broad geographic area based on location-based services data collected from cell phones. This approach supplements current approaches and helps both researchers and emergency response planners better understand which communities lose access to essential services and for how long. We demonstrate our approach by analyzing loss of access to supermarkets, schools, healthcare facilities, and home improvement stores in Southwest Florida leading up to and following the landfall of Hurricane Irma in 2017.

Trust and trustworthy artificial intelligence: A research agenda for AI in the environmental sciences.

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.

Impacts of climate change on cassava yield and lifecycle energy and greenhouse gas performance of cassava ethanol systems: An example from Guangxi Province, China.

Significant shock of climate change on crop yield will challenge the performance of bio-crop on substituting fossil energy to mitigate climate change. Taking cassava-to-ethanol system in Guangxi Province of South China as an example, we coupled a random forest (RF) model with 10 Global climate models (GCMs) outputs to predict the future cassava yields. Subsequently, the net energy value (NEV) and greenhouse gas (GHG) emissions of the cassava-to-ethanol system across varied topographies are assessed using a life cycle analysis. We demonstrate that the abrupt increases in temperatures are the primary contributors to declining yields. Notably, cassava yields in hilly regions decline more than those in plains and display greater variability among concentration pathway scenarios over time. Future NEV and GHG performance of cassava-to-ethanol will undergo significant decreases over time, especially within the high concentration pathway scenario (NEV decrease 28%, GHG increase 3.4% from 2006 to 2100). The performance reductions in hilly area are exacerbated by more harvest loss and labor and material inputs during the "field-to-wheel", negating its energy advantage over fossil fuels. Therefore, adopting a lower concentration pathway and favoring plantation in plains could maintain cassava-to-ethanol as a viable climate mitigation strategy. Our research also advances the methodological approach to climate change adaptation within the domain of life cycle assessment.

Data-driven predictive modeling in risk assessment: Challenges and directions for proper uncertainty representation.

Data-driven predictive modeling is increasingly being used in risk assessments. While such modeling may provide improved consequence predictions and probability estimates, it also comes with challenges. One is that the modeling and its output does not measure and represent uncertainty due to lack of knowledge, that is, "epistemic uncertainty." In this article, we demonstrate this point by conceptually linking the main elements and output of data-driven predictive models with the main elements of a general risk description, thereby placing data-driven predictive modeling on a risk science foundation. This allows for an evaluation of such modeling with reference to risk science recommendations for what constitutes a complete risk description. The evaluation leads us to conclude that, as a minimum, to cover all elements of a complete risk description a risk assessment using data-driven predictive modeling needs to be supported by assessments of the uncertainty and risk related to the assumptions underlying the modeling. In response to this need, we discuss an approach for assessing assumptions in data-driven predictive modeling.

Individual learning as a driver of changes in community vulnerability under repeated hurricanes and changing climate.

The risks from singular natural hazards such as a hurricane have been extensively investigated in the literature. However, little is understood about how individual and collective responses to repeated hazards change communities and impact their preparation for future events. Individual mitigation actions may drive how a community's resilience evolves under repeated hazards. In this paper, we investigate the effect that learning by homeowners can have on household mitigation decisions and on how this influences a region's vulnerability to natural hazards over time, using hurricanes along the east coast of the United States as our case study. To do this, we build an agent-based model (ABM) to simulate homeowners' adaptation to repeated hurricanes and how this affects the vulnerability of the regional housing stock. Through a case study, we explore how different initial beliefs about the hurricane hazard and how the memory of recent hurricanes could change a community's vulnerability both under current and potential future hurricane scenarios under climate change. In some future hurricane environments, different initial beliefs can result in large differences in the region's long-term vulnerability to hurricanes. We find that when some homeowners mitigate soon after a hurricane-when their memory of the event is the strongest-it can help to substantially decrease the vulnerability of a community.

Improved Decision-Making: A Sociotechnical Utility-Based Framework for Drinking Water Investment.

To achieve the goals of the Safe Drinking Water Act, state and local water authorities need to make decisions about where to direct limited funding for infrastructure improvements and currently do so in the absence of adequate evaluative metrics. We developed a framework grounded in utility theory that compares trade-offs explicitly and broadens the factors considered in prioritizing resource allocations. Relevant existing indices were reviewed to identify data applicable to drinking water decision-making. A utility-theory-based decision analysis framework was developed and applied to evaluate how different objectives affect funding decisions for lead service line replacement (LSLR) programs in Pennsylvania and Michigan, United States. The decision framework incorporates drinking water quality characteristics with community and environmental quality attributes. We compare additive and multiplicative model structures, different weights, and spatial scales. Our decision framework showed that the inclusion of additional data beyond what is usually considered in LSLR decisions could change the top 10 counties or public water systems prioritized. Further, the counties or water systems in the top 10 were influenced by the model structure and weights. Prioritization changed based on which data were included, and has implications for the use of evaluative metrics beyond traditional water system data.

COVID-19 aerosol transmission simulation-based risk analysis for in-person learning.

As educational institutions begin a school year following a year and a half of disruption from the COVID-19 pandemic, risk analysis can help to support decision-making for resuming in-person instructional operation by providing estimates of the relative risk reduction due to different interventions. In particular, a simulation-based risk analysis approach enables scenario evaluation and comparison to guide decision making and action prioritization under uncertainty. We develop a simulation model to characterize the risks and uncertainties associated with infections resulting from aerosol exposure in in-person classes. We demonstrate this approach by applying it to model a semester of courses in a real college with approximately 11,000 students embedded within a larger university. To have practical impact, risk cannot focus on only infections as the end point of interest, we estimate the risks of infection, hospitalizations, and deaths of students and faculty in the college. We incorporate uncertainties in disease transmission, the impact of policies such as masking and facility interventions, and variables outside of the college's control such as population-level disease and immunity prevalence. We show in our example application that universal use of masks that block 40% of aerosols and the installation of near-ceiling, fan-mounted UVC systems both have the potential to lead to substantial risk reductions and that these effects can be modeled at the individual room level. These results exemplify how such simulation-based risk analysis can inform decision making and prioritization under great uncertainty.

A machine learning approach for the prediction of overall deceased donor organ yield.

BACKGROUND: Optimizing organ yield (number of organs transplanted per donor) is a potentially modifiable way to increase the number of organs available for transplant. Models to predict the expected deceased donor organ yield have been developed based on ordinary least squares regression and logistic regression. However, alternative modeling methodologies incorporating machine learning may have superior performance compared with conventional approaches. METHODS: We evaluated the predictive accuracy of 14 machine learning models for predicting overall organ yield in a cross-validation procedure. The models were parameterized using data from the Organ Procurement and Transplantation Network database from 2000 to 2018. The inclusion criteria for the study were adult deceased donors between 18 and 84 years of age that had at least 1 organ procured for transplantation. RESULTS: A total of 89,520 donors met the inclusion criteria. Their mean (standard deviation) age was 44 (15) years, and approximately 58% were male. Our cross-validation analysis showed that a tree-based gradient boosting model outperformed the remaining 13 models. Compared with the currently used prediction models, the gradient boosting model improves prediction accuracy by reducing the mean absolute error between 3 and 11 organs per 100 donors. CONCLUSION: Our analysis demonstrated that the gradient boosting methodology had the best performance in predicting overall deceased donor organ yield and can potentially serve as an aid to assess organ procurement organization performance.

Interdisciplinary Theory, Methods, and Approaches for Hazards and Disaster Research: An Introduction to the Special Issue.

What is interdisciplinary research? Why is it vital to the advancement of the field of hazards and disaster research? What theory, methods, and approaches are fundamental to interdisciplinary research projects and their applications? This article addresses these and other pressing questions by taking stock of recent advancements in interdisciplinary studies of hazards and disasters. It also introduces the special issue of Risk Analysis, which includes this introductory article and 25 original perspectives papers meant to highlight new trends and applications in the field. The papers were written following two National Science Foundation-supported workshops that were organized in response to the growing interest in interdisciplinary hazards and disaster research, the increasing number of interdisciplinary funding opportunities and collaborations in the field, and the need for more rigorous guidance for interdisciplinary researchers and research teams. This introductory article and the special collection are organized around the cross-cutting themes of theory, methods, approaches, interdisciplinary research projects, and applications to advance interdisciplinarity in hazards and disaster research.

The Role of Time in Risk and Risk Analysis: Implications for Resilience, Sustainability, and Management.

There is a persistent misconception that risk analysis is only suited for considering the immediate consequences of an event. Such a limitation would make risk analysis unsuitable for many challenges, including resilience, sustainability, and adaptation. Fortunately, there is no such limitation. However, this notion has stemmed from a lack of clarity regarding how time is considered in risk analysis and risk characterization. In this article, we discuss this issue and show that risk science provides concepts and frameworks that can appropriately address time. Ultimately, we propose an adjusted nomenclature for explicitly reflecting time in risk conceptualization and characterizations.

Protective impacts of household-based tuberculosis contact tracing are robust across endemic incidence levels and community contact patterns.

There is an emerging consensus that achieving global tuberculosis control targets will require more proactive case finding approaches than are currently used in high-incidence settings. Household contact tracing (HHCT), for which households of newly diagnosed cases are actively screened for additional infected individuals is a potentially efficient approach to finding new cases of tuberculosis, however randomized trials assessing the population-level effects of such interventions in settings with sustained community transmission have shown mixed results. One potential explanation for this is that household transmission is responsible for a variable proportion of population-level tuberculosis burden between settings. For example, transmission is more likely to occur in households in settings with a lower tuberculosis burden and where individuals mix preferentially in local areas, compared with settings with higher disease burden and more dispersed mixing. To better understand the relationship between endemic incidence levels, social mixing, and the impact of HHCT, we developed a spatially explicit model of coupled household and community transmission. We found that the impact of HHCT was robust across settings of varied incidence and community contact patterns. In contrast, we found that the effects of community contact tracing interventions were sensitive to community contact patterns. Our results suggest that the protective benefits of HHCT are robust and the benefits of this intervention are likely to be maintained across epidemiological settings.

Tropical Cyclone Exposures and Risks of Emergency Medicare Hospital Admission for Cardiorespiratory Diseases in 175 Urban United States Counties, 1999-2010.

BACKGROUND: Although injuries experienced during hurricanes and other tropical cyclones have been relatively well-characterized through traditional surveillance, less is known about tropical cyclones' impacts on noninjury morbidity, which can be triggered through pathways that include psychosocial stress or interruption in medical treatment. METHODS: We investigated daily emergency Medicare hospitalizations (1999-2010) in 180 US counties, drawing on an existing cohort of high-population counties. We classified counties as exposed to tropical cyclones when storm-associated peak sustained winds were ≥21 m/s at the county center; secondary analyses considered other wind thresholds and hazards. We matched storm-exposed days to unexposed days by county and seasonality. We estimated change in tropical cyclone-associated hospitalizations over a storm period from 2 days before to 7 days after the storm's closest approach, compared to unexposed days, using generalized linear mixed-effect models. RESULTS: For 1999-2010, 175 study counties had at least one tropical cyclone exposure. Cardiovascular hospitalizations decreased on the storm day, then increased following the storm, while respiratory hospitalizations were elevated throughout the storm period. Over the 10-day storm period, cardiovascular hospitalizations increased 3% (95% confidence interval = 2%, 5%) and respiratory hospitalizations increased 16% (95% confidence interval = 13%, 20%) compared to matched unexposed periods. Relative risks varied across tropical cyclone exposures, with strongest association for the most restrictive wind-based exposure metric. CONCLUSIONS: In this study, tropical cyclone exposures were associated with a short-term increase in cardiorespiratory hospitalization risk among the elderly, based on a multi-year/multi-site investigation of US Medicare beneficiaries ≥65 years.

On the Use of Standards and Guidelines as a Tool to Fulfil Regulatory Requirements.

Over the years, industrial safety regulation has shifted from a "hard" command and control regime to a "soft" regime. A "hard" regime includes the use of strict prescriptive requirements which explain how industry should solve particular issues. A "soft" regime, uses more functional requirements, pointing out what goals are to be achieved. In a "soft" regime, prescriptive standards might still exist, but they are considered suggested solutions, with alternative solutions also being considered if they achieve the overall regulatory goals. The purpose of such a shift is to create regulations that are more flexible, meaning that they are more open for the use of novel technology and for the use of risk assessments as a basis for decision making. However, it is not clear that the shift from a hard to a soft regime has made it easier to use risk assessments for such a purpose in practice. In the present article, we discuss the limitations caused by strict adherence to prescriptive requirements presented in standards or regulations and present our perspective on why and how these can limit risk management in practice. The article aims to discuss the strengths and weaknesses, with regard to risk management, when regulations are strictly dependent on prescriptive or specification-based standards and guidelines. Several examples are used to illustrate some of the main challenges related to the use of specification-based technical standards and how the regulatory shift from "hard" to "soft" has not necessarily made it easier to implement technological solutions based on risk assessments.

Risk Factors for 100-Year Flood Events in the Mid-Atlantic Region of the United States.

Anecdotal information indicates that streams in the Mid-Atlantic region of the United States experience more extreme flood events than might be expected. This leads to the question of whether this is an unfounded perception or if these extreme events are actually occurring more than should be expected. If the latter is true, is this due solely to randomness, or alternately to characteristics that make certain watersheds more prone to repeated events that may be defined as 100-year or greater floods? These questions are investigated through analysis of flood events based on standard flood frequency analysis. 100-year streamflow rates for stream gages were estimated using Bulletin 17B flood frequency analysis methods, and the probability of the annual peak flow record for each gage was calculated. These probabilities were compared to a set of synthetic probabilities to evaluate their distribution. This comparison indicates that for the Mid-Atlantic region as a whole, the Bulletin 17B method does not systematically over or underestimate flood frequency. A Random Forest model of probability of actual flood record (PAFR) versus watershed and stream gage characteristics was developed and used to understand if certain characteristics are associated with PAFR. This analysis indicated that unexpected numbers of large flood events in a stream gage period of record can be attributed primarily to randomness, but there is some correlation with watershed and gage characteristics including weighted skew, drainage area, and mean annual peak discharge. The results indicate that watersheds with high values of these characteristics may warrant advanced flood frequency methods.

Agent-Based Models as an Integrating Boundary Object for Interdisciplinary Research.

Many of the most complicated and pressing problems in hazards research require the integration of numerous disciplines. The lack of a common knowledge base, however, often prohibits clear communication and interaction among interdisciplinary researchers, sometimes leading to unsuccessful outcomes. Drawing on experience with several projects and collective expertise that spans multiple disciplines, the authors argue that a promising way to enhance participation and enable communication is to have a common model, or boundary object, that can integrate knowledge from different disciplines. The result is that researchers from different disciplines who use different research methods and approaches can work together toward a shared goal. This article offers four requirements for boundary objects that may enhance hazards research. Based on these requirements, agent-based models have the necessary characteristics to be a boundary object. The article concludes by examining both the value of and the challenges from using agent-based models as the boundary object in interdisciplinary projects.

Assessing United States County-Level Exposure for Research on Tropical Cyclones and Human Health.

BACKGROUND: Tropical cyclone epidemiology can be advanced through exposure assessment methods that are comprehensive and consistent across space and time, as these facilitate multiyear, multistorm studies. Further, an understanding of patterns in and between exposure metrics that are based on specific hazards of the storm can help in designing tropical cyclone epidemiological research. OBJECTIVES: a) Provide an open-source data set for tropical cyclone exposure assessment for epidemiological research; and b) investigate patterns and agreement between county-level assessments of tropical cyclone exposure based on different storm hazards. METHODS: We created an open-source data set with data at the county level on exposure to four tropical cyclone hazards: peak sustained wind, rainfall, flooding, and tornadoes. The data cover all eastern U.S. counties for all land-falling or near-land Atlantic basin storms, covering 1996-2011 for all metrics and up to 1988-2018 for specific metrics. We validated measurements against other data sources and investigated patterns and agreement among binary exposure classifications based on these metrics, as well as compared them to use of distance from the storm's track, which has been used as a proxy for exposure in some epidemiological studies. RESULTS: Our open-source data set was typically consistent with data from other sources, and we present and discuss areas of disagreement and other caveats. Over the study period and area, tropical cyclones typically brought different hazards to different counties. Therefore, when comparing exposure assessment between different hazard-specific metrics, agreement was usually low, as it also was when comparing exposure assessment based on a distance-based proxy measurement and any of the hazard-specific metrics. DISCUSSION: Our results provide a multihazard data set that can be leveraged for epidemiological research on tropical cyclones, as well as insights that can inform the design and analysis for tropical cyclone epidemiological research. https://doi.org/10.1289/EHP6976.

Reframing Resilience: Equitable Access to Essential Services.

We urgently need to put the concept of resilience into practice if we are to prepare our communities for climate change and exacerbated natural hazards. Yet, despite the extensive discussion surrounding community resilience, operationalizing the concept remains challenging. The dominant approaches for assessing resilience focus on either evaluating community characteristics or infrastructure functionality. While both remain useful, they have several limitations to their ability to provide actionable insight. More importantly, the current conceptualizations do not consider essential services or how access is impaired by hazards. We argue that people need access to services such as food, education, health care, and cultural amenities, in addition to water, power, sanitation, and communications, to get back some semblance of normal life. Providing equitable access to these types of services and quickly restoring that access following a disruption are paramount to community resilience. We propose a new conceptualization of community resilience that is based on access to essential services. This reframing of resilience facilitates a new measure of resilience that is spatially explicit and operational. Using two illustrative examples from the impacts of Hurricanes Florence and Michael, we demonstrate how decisionmakers and planners can use this framework to visualize the effect of a hazard and quantify resilience-enhancing interventions. This "equitable access to essentials" approach to community resilience integrates with spatial planning, and will enable communities not only to "bounce back" from a disruption, but to "bound forward" and improve the resilience and quality of life for all residents.