Environmental Research Letters LETTER • OPEN ACCESS Recent citations Predicting support for flood mitigation based on - Leveraging machine learning for predicting flash flood damage in the Southeast US flood insurance purchase behavior Atieh Alipour et al To cite this article: Wanyun Shao et al 2019 Environ. Res. Lett. 14 054014 View the article online for updates and enhancements. This content was downloaded from IP address 205.156.36.134 on 13/02/2020 at 18:56 Environ. Res. Lett. 14 (2019) 054014 https://doi.org/10.1088/1748-9326/ab195a LETTER Predicting support for flood mitigation based on flood insurance OPEN ACCESS purchase behavior RECEIVED 30 November 2018 Wanyun Shao1,3 , Kairui Feng2,3 and Ning Lin2 REVISED 1 Department of Geography, Alabama Water Institute, Center for Complex Hydrosystems Research, University of Alabama, United States 28 March 2019 of America ACCEPTED FOR PUBLICATION 2 Department of Civil and Environmental Engineering, Princeton University, United States of America 12 April 2019 3 Co-first authors: Wanyun Shao and Kairui Feng have made equal contribution to this manuscript. PUBLISHED 13 May 2019 E-mail: [email protected] Keywords: flood insurance, decision-making mechanism, Bayesian network, flood risk mitigation Original content from this work may be used under Supplementary material for this article is available online the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of Abstract this work must maintain What is the decision-making mechanism people rely upon to mitigate flood risk? Applying Bayesian attribution to the author(s) and the title of Network modeling to a comprehensive survey dataset for the US Gulf Coast, we find that the overall the work, journal citation fl fl ( and DOI. support for ood mitigation can be inferred from ood insurance purchase behavior i.e. without insurance versus with insurance purchased mandatorily, voluntarily, or both). Therefore, we propose a theoretical decision-making mechanism composed of two dimensions including informed flood risk and sense of insecurity. The informed flood risk is the primary determinant on one’s overall support for flood mitigation. Risk mitigation decisions are largely contingent on the level of risk that is effectively conveyed to individuals. Additionally, sense of insecurity plays a moderate role in determining individuals’ overall support for flood mitigation. The sense of insecurity can move people toward overall support for mitigation, but the effect is not as large as the informed risk. Results of this study have fundamental policy implications. The flood risk informed by Federal Emergency Management Agency’s flood maps not only provides the compulsory basis for flood insurance purchase but also determines individuals’ overall support for flood mitigation. Flood map inaccuracy can immensely mislead individuals’ overall risk mitigation decision. Meanwhile, this flood risk mitigation decision-making mechanism inferred from a survey data in the US Gulf Coast needs to be tested and validated elsewhere. Introduction flooding is projected to become more often and intensive in decades to come [7–9]. A recent Intergovernmental Panel on Climate Change Increasing flood risk renders it imperative for special report states that human-induced ‘global coastal residents to take adequate mitigation4 mea- warming is likely to reach 1.5 °C between 2030–2052’ sures. Purchasing flood insurance would be a sensible [1]. The projected temperature rise will pose wide- option for residents who are vulnerable to flooding. spread and serious risk to society, ranging from more Abundant evidence suggests that many individuals do frequent extreme weather events to sea level rise. not have flood insurance [2]. Other flood mitigation fl Among all climate-related hazards, ood has incurred measures include but are not limited to home eleva- the most economic damages and affected the most tion, house modification for flood-proofing, con- [ ] people 2 . The coastal region, due to the coupling struction of sea walls, installation of flood warning effects of climate change and rapid population growth [3], is especially prone to flooding. Recent hurricanes 4 serve as vivid reminders of how devastating hurricane- Mitigation means differently in different contexts. Mitigation in fl [ – ] the context of climate change refers to attempts to curb carbon induced ooding can be to coastal communities 4 6 . emission to limit climate change. Mitigation in this study means In a changing climate, hurricane-induced coastal lowering flood impacts or flood risk reduction. © 2019 The Author(s). Published by IOP Publishing Ltd Environ. Res. Lett. 14 (2019) 054014 system, and relocation [10–12]. Despite the effective- the US [25]. In addition to its exposure to climate ness of flood mitigation measures, few adopt these change [24], this region’s population embodies more measures voluntarily [13, 14]. Fundamentally, what is ethnic diversity, larger income gap, higher poverty the decision-making mechanism people rely upon to rates, lower income, higher percentages of racial min- mitigate flood risk? Furthermore, given various miti- ority and old residents and all suggest greater social gation measures, do they have the proclivity to forgo vulnerability [26, 27]. This region presents an ideal the others because the adoption of one such as pur- natural laboratory to study risk reduction decisions. chasing flood insurance has provided sufficient sense Meanwhile, results of the present study are believed to of security? assist policy makers to reach informed decision. This speculation is based on the concept of moral hazard, referring to the idea that purchasing insurance lowers the incentive for the policyholders to seek self- Data and method protection measures that would increase actual prob- fl abilities of hazardous events [15]. Two previous stu- To investigate how ood insurance purchase behavior dies have provided a theoretical base for the use of influences individuals’ mitigation policy support insurance purchase behavior to predict self-protection among the US Gulf Coast residents, we use a behavior [16, 17]. Moral hazards have been observed comprehensive climate change survey for all coastal in a broad range of insurance markets such as auto- counties in the Gulf Coast in 2012 (SM 1). In addition mobile, long-term care, and health [18]. A limited to flood insurance purchase behaviors, the survey number of empirical studies that have been conducted includes various questions related to socio-demo- in the domain of natural disaster insurance market graphic characteristics, perceptions of local climate however find little evidence for the presence of moral change, and perceptions of flood-related hazards that hazards [19]. Instead, they find that insurance pur- may affect mitigation behavior/intention [28–32].We chases can increase the tendency towards self-protec- construct an interconnected Bayesian Network (BN) tion against flooding across insured individuals model to study how these variables jointly affect the [19–22]. These findings prompt the second specula- support for flood mitigation. tion: do individuals who have adopted one tend to Our model uses a BN [33], which is a statistical adopt the others for more security? The drive behind model to describe probabilistic relationships among a such decisions is risk aversion. A study conducted in set of variables using a directed acyclic graph (DAG). flood-prone areas in Germany indicates insurance The DAG structure of BN enables the Joint probability policy holders adopt more private flood mitigation distribution) of all the modeled parameters to be measures [21]. Another study which was done among expressed in terms of a product of conditional prob- Florida households finds a positive correlation ability distributions (CPD), describing each variable in between insurance coverage and private mitigation terms of its parents, i.e. those variables it depends measures [22]. Similarly, a recent study that uses an upon. We employ BN here for its ability to propagate experiment among Dutch homeowners finds that uncertainty, perform inference and calculate condi- individuals who buy insurance are more likely to self- tional probabilities. BN has better capabilities of hand- protect against flooding [23]. ling nonlinear discrete dataset than other regression- In this study, by using a comprehensive climate based relationship model such as Structural Equation survey among coastal residents in the US Gulf Coast Model [34]. Also, BN shows superiority over other sta- region, we aim to test which tendency is more domi- tistical methods when dealing with incomplete data by nant among flood insurance policy holders. Particu- flexible marginalization5 methods [35]. The sample larly, we classify flood insurance purchase behavior size of this data (∼3800 observations) falls short of the into four categories including non-purchase behavior significant criteria for modern machine learning tools and purchase for mandatory, voluntary, and both (e.g. deep neural network) while BN works well [46]. mandatory and voluntary reasons. The significance of BNs have been applied in previous studies on risk this study is two-fold. First, by classifying insurance communication [36], coastal risk analysis and decision purchase behaviors into four categories as opposed to making [37, 38], and community resilience to coastal ( ) two with versus without insurance , we attempt to hazards [39–41]. The output in a probabilistic form is ’ fl gain deeper insight into individuals ood mitigation well