1 Introduction

Hydrological hazards, such as floods and landslides, pose a significant threat worldwide, inflicting extensive damage. In 2022 alone, these disasters caused over 7000 fatalities and economic losses exceeding US$44 billion (CRED 2023). With climate and land use change, the magnitude and frequency of such events are likely to intensify (IPCC 2022; McDermott 2022), amplifying risk management challenges.

Traditionally, the mitigation of hydrological disaster risk has followed a technocratic approach, relying on structural interventions (Opperman et al. 1979; Blöschl et al. 2013; Roos et al. 2017). However, this approach often neglects societal aspects and feedback between human and water systems, which can lead to unintended outcomes (Sivapalan et al. 2012; Di Baldassarre et al. 2019). For instance, issuing temporary water abstraction licenses during droughts may exacerbate underlying water scarcity. In fact, revoking these licenses often proves difficult due to dependence on the new water source, ultimately reducing long-term water availability (Di Baldassarre et al. 2018b). Similarly, constructing levees and dams can encourage the occupation of flood-prone areas and increase the number of people and assets exposed to residual risk (White 1945; Tobin 1995; Di Baldassarre et al. 2013). Non-structural measures, such as land use policies and building codes, can also inadvertently escalate risks by classifying hazardous areas as safe for development (Burby 2006; Cutter et al. 2018). These situations are referred to as socio-hydrological phenomena caused by the complex and interconnected nature of water and society and the unintended consequences of short-sighted policies/measures.

Among the various socio-hydrological phenomena documented in the literature (Online resources 1), the safe development paradox (SDP) and the levee effect (LE) are particularly noteworthy. Both involve counterintuitive outcomes resulting from risk mitigation efforts, challenging traditional risk reduction strategies and emphasizing the need for a more holistic approach to risk management. Consequently, these phenomena have garnered increasing research attention (Breen et al. 2022).

White (1945) was the first to document the LE in a comprehensive assessment of flood adaptation. The author found that while structural protection reduces minor floods, it paradoxically encourages floodplain development, amplifying long-term damages as more people and assets become exposed to floods that exceed protection structures (i.e., residual risk). Furthermore, the absence of floods over time may lull residents into a ‘false sense of security’, impeding preparedness (White 1945). Tobin (1995) later coined the term LE, arguing that reliance on structural measures can lead to a lock-in situation, where increasing development necessitates heightened levels of protection to respond to the rising hazards. Within this context, Ferdous et al. (2019, 2020) observed that areas with levees had higher asset accumulation and population growth. Consequently, these locations grew more dependent on ever-higher levees to maintain protection against intensifying flood hazards.

The SDP was conceptualized by Burby (2006) as a phenomenon arising from government policies designed to enhance safety in hazardous areas but inadvertently escalating disaster risk. For instance, Burby (2006) and Cutter et al. (2018) documented cases where insurance, building codes, and structural flood protection classified hazardous areas as safe to build and occupy, increasing overall risk. While Burby’s (2006) study partially exemplifies the LE—focused on structural protection measures like levees—the SDP extends to non-structural approaches to disaster risk reduction. Both phenomena stem from a disconnection between adaptation intentions and outcomes and behavioral aspects of exposed individuals, such as a false sense of security. In this context, the SDP can be considered a general phenomenon that includes the LE as a subcategory (Baldassarre et al. 2019).

Despite efforts to understand the SDP and LE and prevent adverse consequences of adaptation, research on socio-hydrological phenomena is still fragmented. In this regard, Di Baldassarre et al. (2018a) proposed a research agenda for the LE, wherein the authors attribute this fragmentation in research to the lack of standardized methods and comparability between case studies. In an effort to summarize existing research, Breen et al. (2022) analyzed 42 studies investigating the SDP and LE, focusing on their temporal and spatial distribution. While this study shed light on geographical biases and offered recommendations for overcoming the SDP, the authors did not address the methodological approaches, variables, or critically examine the findings of these studies. Moreover, Breen et al. (2022) restricted their analysis solely to flood hazards.

Consequently, knowledge gaps persist in understanding the SDP and its sub-phenomenon, the LE. Indeed, the lack of a comprehensive evidence synthesis hampers our ability to identify how these phenomena manifest and what effective mitigation strategies are. Furthermore, the absence of a synthesized methodology and variable set for investigating the SDP impedes comparisons across studies (e.g., meta-analysis) and hinders our understanding of the conditions that trigger these phenomena.

To address these gaps and enhance comparability, we comprehensively analyzed the methods and variables used in SDP and LE case studies covering various hydrological hazards. Specifically, our goals were to (1) assess the geographic distribution, hazard types, and spatial scales of these studies; (2) examine the methods used to investigate the SDP and the LE; (3) identify the variables considered; (4) evaluate the empirical evidence confirming these phenomena; and (5) provide evidence on the non-structural measures and individual adaptation measures in generating or mitigating the SDP. By addressing these goals, we highlight the most and the least assessed variables and methods and contribute to understanding the different factors triggering the SDP and LE across various social, technical, and hydrological contexts. With this, we aim to provide a foundation for researchers and practitioners, facilitating comparative analyses.

2 Methods

To ensure transparency and replicability in our methodology, we conducted a systematic review following the Reporting of Strategies in Systematic Evidence Syntheses standards—ROSES (Haddaway et al. 2018). We considered English articles available up to December 31st, 2023, sourced from the Web of Science (WoS) and Scopus databases. The search terms (Table 1) were selected based on relevant keywords identified by Ferdous et al. (2019) and Haer et al. (2020). The search was performed based on the title, abstract, and both author and index keywords on January 11th, 2024.

Table 1 Search string used for each database
Full size table

The search elicited 158 documents. After removing duplicates, we screened the unique articles (n = 94) first at the title and abstract level and then at the full-text level by considering the following inclusion criteria: (1) it addresses directly or indirectly the general phenomena SDP or the sub-phenomena LE; (2) it is available in English; (3) it does not focus exclusively on hydraulic aspects of levees or dikes; (4) it analyses a hydrological disaster, (5) it is a case study. We then performed a full-text analysis of the studies identified as relevant (n = 40). A summary of the review process is presented in Fig. 1.

Fig. 1
figure 1

Systematic review procedure following the ROSES guidelines (n = number of articles)

Full size image

To address the research goals, we categorized the selected articles according to several topics (see Table 2). For goal 1, we assessed the country where each case study was conducted, the spatial scale (based on Fischer et al. 2021), and the hydrological hazard typology based on Below et al. (2009). Fischer’s et al. (2021) spatial scales ranged from international to local, encompassing ‘social network’ studies focusing on specific social constructs as (e.g., leveed areas, development sites), ‘water bodies network’ area delimited based on a water body feature (e.g., floodplains, and deltaic regions).

Table 2 Topics used to classify the reviewed articles and their relation to the study goals
Full size table

For goals 2 and 3, we extracted information on the research design, the applied methods, and the variables considered to assess the SDP and the LE. Given their variety, we classified the variables according to their similarities into four distinct groups: human system (individual characteristics and behaviors), interaction (interplay between human and natural systems; these variables capture exposure and risk reduction measures), natural system (characteristics of natural elements that may contribute to hazards), and outcome (consequences of disasters). This classification acknowledges that variables can often encompass multiple hazard, vulnerability, exposure, or response dimensions. For example, adaptation can be considered within vulnerability and response groups (Rufat et al. 2015; Simpson et al. 2021), while risk perception reflects the interaction of vulnerability, exposure, and hazard.

To assess goal 4, we verified the presence of empirical evidence supporting SDP or LE and classified each article as confirming, refuting, or providing inconclusive evidence for the phenomena. Case studies with ongoing mitigation efforts were considered confirmation of the phenomenon as they demonstrate the phenomenon’s persistence rather than its absence – refutation of the phenomena. Finally, for Goal 5, we identified the non-structural protection measures explored in each study. These could be either policies (e.g., risk management, risk insurance) or individual adaptation measures (e.g., installing pumps or avoiding ground-floor living spaces). We acknowledge that some classes may overlap due to their interconnectedness. For instance, risk insurance and compensation policy classes can also be considered as a risk management policy.

3 Results

3.1 Spatio-temporal scope and scale of analysis of the reviewed articles

The reviewed case studies examining the SDP and the LE span the years 2001–2023, with a notable increase in publications on the LE over the last 6 years (Fig. 2). The publications include both direct and indirect assessments. For instance, Blanchard-Boehm et al. (2001) indirectly evaluated the SDP by analyzing the acquisition of flood insurance in generating risk, though not explicitly introducing the concept itself. The studies were conducted in 18 different countries, with the majority in the US (n = 12), followed by Italy (n = 5), Australia (n = 4), New Zealand (n = 3) and Bangladesh (n = 3). Single studies were found in Austria, Canada, China, France, India, Ireland, Japan, Nepal, Netherlands, Niger, Poland, Taiwan, Vietnam and the European Union. This distribution highlights the disparity in research conducted in the Global North (n = 32) and South (n = 9) (Fig. 3), a pattern also observed in previous socio-hydrology (e.g., Vanelli et al. 2021; Fischer et al. 2021; Breen et al. 2022) and natural hazard literature reviews (e.g., Emmer 2018).

Fig. 2
figure 2

Temporal distribution of the reviewed articles according to the socio-hydrological phenomena investigated from 2001 to 2023

Full size image
Fig. 3
figure 3

Hazards assessed according to global north and global south countries. Division based on Brandt (1980)

Full size image

Floods (n = 37) were the main hydrological hazard type investigated, followed by flash floods (n = 3) (Fig. 3). Despite the increasing call for multi-hazard research (de Ruiter et al. 2020), only ten studies addressed multi-hazards. Exceptions include studies that investigated floods associated with hurricanes (n = 3) (Burby 2006; Fox-Rogers et al. 2016; Malecha et al. 2021), flood, subsidence, and sea level rise (Smits et al. 2006) (n = 1), and floods, landslides, and debris flows (n = 1) (De Marchi and Scolobig 2012). No single study addressed landslides alone.

Concerning the spatial scale, the reviewed studies were developed predominantly on city scales (n = 16, 39%), followed by waterbody networks (n = 8, 19%) (areas delimited by hydrological features such as the middle and lower floodplains of the Pó River watershed as assessed by Domeneghetti et al. (2015)), neighborhood and national scales (n = 4, 9.8%), social network (n = 3, 7.3%) (areas defined by social connections and interactions, such as communities within levee-protected zones e.g., Dufty et al. (2022)) (Fig. 4). No study was conducted on a global scale. A qualitative examination showed a lack of explicit justification for the chosen spatial scales. The prevalence of city spatial scale is likely due to practical considerations, including its alignment with territorial management units, data availability, and the scale of the disasters under investigation. Within this context, Fischer et al. (2021) suggest that city-level studies often focus on water resource management (stormwater, wastewater, and human consumption) and prioritize social processes. Conversely, natural delimitations like watersheds emphasize natural processes. This helps explain the observed prevalence of city-scale studies, as assessing SDP and LE involves understanding social dynamics within human-water systems. Furthermore, most studies (n = 33, 82.5%) focused on medium-small spatial scales, including neighborhoods, cities, waterbody networks, and social networks. Interestingly, waterbody networks (e.g., delimited floodplains) were more common than the traditional larger-scale watershed focus typically found in socio-hydrology studies (Fischer et al. 2021).

Fig. 4
figure 4

Spatial scale considered by the studies when assessing the SDP and LE. Some studies encompassed multiple scales, counting in both scales’ classes

Full size image

Most studies (n = 33, 82.5%) explicitly assessed the LE or SDP, while the remaining mentioned it implicitly, using related terms such as the false sense of safety (e.g., Blanchard-Boehm et al. 2001; Smits et al. 2006; López-Marrero 2010; Dahal and Hagelman 2011; Lawrence et al. 2013; Rohr 2013; Salman and Hurlbert 2022). Typically, assessing these phenomena was a secondary result, not the studies` primary goal. Amongst the assessed phenomena, the LE emerged as the most prevalent (n = 32, 80.0%). Conversely, studies on the SDP were less common (n = 8, 20.0%) and primarily concentrated in North America. Seven studies acknowledge the SDP or LE but did not directly assess them (Glavovic (2014a), Glavovic (2014b), Lawrence et al. (2013), Rohr (2013), Salman and Hurlbert (2022), Smits et al. (2006), Wasson et al. (2019). Other phenomena, such as the local government paradox (Burby 2006; Cutter et al. 2018), adaptation effect (Richert et al. 2019; Michaelis et al. 2020; Mazzoleni et al. 2021; Chang et al. 2022; Salman and Hurlbert 2022), pendulum swing (Yu et al. 2020; Luu et al. 2022), and cry-wolf syndrome (Dahal and Hagelman 2011), were also identified in the reviewed articles. However, they are not presented in detail here, as they are outside the scope of this study.

3.2 Research design and methods used

We found a noticeable variation in how studies conceptualized risk, which, in turn, influenced how the phenomena were assessed. Overall, studies adopting the UNDRR (2022) definition of risk (risk = vulnerability × exposure × hazard) focused on individual behavior, adaptation, and the sense of safety aspects. Conversely, engineering-focused studies equated vulnerability to expected damage or embedded exposure within vulnerability. They often assessed the increase in assets and population (i.e., changes in exposure) within areas protected by hazard protection measures (e.g., floodplains behind levees).

The research design employed to address the SDP and LE was primarily quantitative (n = 16), followed by qualitative (n = 13) and mixed methods (n = 11). The trend toward quantitative research can be attributed to the call by Sivapalan et al. (2012) to maintain socio-hydrology as a quantitative science. A large portion of both quantitative (n = 6, 37.5%) and qualitative studies (n = 5, 41.7%) assessed the phenomena by solely addressing exposure changes, neglecting vulnerability aspects. By doing so, they fail to account for vulnerability aspects that are crucial in determining hazard impacts (Fuchs and Glade 2016; de Brito et al. 2017). In contrast, most mixed research design studies (n = 6, 54.5%) assessed both exposure changes and vulnerability among community members and stakeholders. Mixed studies provide comprehensive assessments by combining the strengths of both qualitative and quantitative methods (Vanelli et al. 2022) to assess exposure, hazard, and vulnerability changes.

The reviewed studies utilized a diverse range of methods (Online Resources 2 and Fig. 5). Surveys (n = 17) were the most common one, followed by spatial and demographic analysis (n = 16), document review (n = 15), and hydrological analysis/flood modeling (n = 8). Notably, 27 studies employed a combination of these. For instance, Ferdous et al. (2020) integrated surveys and demographic and spatial analysis to examine economic changes (investments, family wealth) and population shifts (urban/rural) within and outside leveed-protected areas. This wide variety of methods hampered a meta-analysis or a quantitative comparison. Therefore, we present and discuss the methods used in the context of their strengths and weaknesses in assessing the SDP and LE.

Fig. 5
figure 5

Methods employed in the reviewed studies. The “Others” class comprises methods employed by only one article (e.g., agent-based modeling). Some studies applied more than one method. A description of the methods used in each study is provided in Online Resources 2

Full size image

Surveys were primarily used to observe individuals’ behavior, focusing on assessing a false sense of safety and non-protective behavior. They were also used to gather information from government stakeholders about government decisions and their disaster risk reduction activities. Most survey studies evaluated changes in vulnerability due to mitigation measures using a cross-sectional design. An exception is the study by Gissing et al. (2018), who employed a sequential design across multiple stages. Surveys were conducted through interviews (semi-structured) or questionnaires via mail, telephone, online, or face-to-face. Due to their flexibility, surveys allowed exploring in-depth risk perception and risk preparedness concepts. However, these studies often lack transparency, as many fail to describe the sample size and sampling techniques adequately. Furthermore, surveys restricted to single communities or with small sample sizes (i.e., less than 20 participants) may have limited generalizability and representativeness (Fox Rogers et al. 2016; Salman and Hurlbert 2022). Survey studies face challenges such as restricted researcher access to study areas, the COVID-19 pandemic, and individual willingness to participate (Stevens et al. 2010; Dahal and Hagelman 2011; Fox-Rogers et al. 2016; Dufty et al. 2022; Chang et al. 2022; Salman and Hurlbert 2022). Additionally, participation bias may occur, as individuals with less disaster experience or living in areas with fewer disaster records may be less interested in participating. This could lead to underestimating the prevalence of a false sense of safety.

Studies using spatial and demographic analysis (n = 16) focused on assessing exposure changes (e.g., increasing urbanization, occupation, asset accumulation, and population) within areas protected by hazard mitigation measures. To achieve this, remote sensing techniques, historical maps of human occupation, urban development modeling, and cross-comparisons have been employed. For instance, researchers have compared protected and unprotected flood-prone areas across different spatial scales to verify the relationship between mitigation structures and heightened hazard exposure (e.g. Ding et al. 2023; Ferdous et al. 2019, 2020; Collenteur et al. 2015). This group of methods faces limitations, such as the lack of detailed building type information (commercial, residential, etc.), hampering accurate flood exposure assessments (e.g., Domeneghetti et al. 2015). Remote sensing data also suffers from low spatio-temporal resolution, which hinders capturing landscape changes like building construction and levee breaches (e.g., Mandarino et al. 2023). Additionally, issues such as unavailable or inaccessible socioeconomic data limit the research scope (e.g., Ferdous et al. 2020), potentially introducing selection bias towards areas with better data quality (Chang et al. 2022). A further drawback lies in the tendency to overlook the dynamic nature of urban development and human occupation. Urban growth is often assessed solely through remote sensing data or building databases, neglecting socioeconomic factors (e.g., migration, alternative settlement locations) and physical limitations influencing land use (e.g., limited space for construction) (Ferdous et al. 2019, 2020). This can lead to inaccurate assessments of the impacts of structural or non-structural measures on occupation patterns.

Document review studies examined historical documents and policies to understand their impact on vulnerability and exposure (Malecha et al. 2021). This method was also used to elucidate the contextual background contributing to the SDP or LE. Document reviews often provide a rich historical context, allowing researchers to understand how socio-hydrological phenomena lead to increased exposure. However, limitations arise due to document availability, quality, and selection bias. These factors can hamper the comprehensiveness of the assessment and exclude relevant areas. In addition, document review studies often lack transparency regarding document selection (Bowen et al. 2009).

Lastly, we highlight hydrological analysis and flood modeling, which are employed to comprehend hydrological patterns (Massazza et al. 2021). This approach facilitates understanding changes in hazard levels and residual risks. When coupled with damage estimation and modeling, it allows for a comparative analysis between past disaster events and the current occupation status (Domeneghetti et al. 2015). Conversely, they do not directly account for changes in vulnerability and exposure, which are the phenomena’ key drivers. Therefore, they should be complemented with methodologies explicitly targeting these factors for a comprehensive assessment.

3.3 Variables employed

Only 32 studies explicitly described the variables used to assess the LE and SDP (see Table 3 and Online Resource 1). Most of them were classified as interaction (n = 24), followed by human system (n = 20), outcome (n = 15), and natural system (n = 9) variables. Interaction variables represent exposure and mitigation measures and their characteristics, such as the level of protection conferred by mitigation measures and their attributes (e.g., implementation year, improvements, operation, protected area) (Domeneghetti et al. 2015; Hutton et al. 2019; Mazzoleni et al. 2021; Ding et al. 2023). They quantify and track populations and assets over specific spatial–temporal spaces. They often include the position of exposed elements relative to levees and water bodies, population density, policy control on land occupation, and land-cover evolution. Studies primarily employed interaction variables as independent or control variables. For instance, Ferdous et al. (2019) observed that urban development (independent variable) was higher in areas protected by levees compared to unprotected or low-maintenance levee-protected areas (control variables), thereby inducing the LE (dependent variable).

Table 3 Variables used in the reviewed studies according to different groups. A list of the variables used in each study is presented in Online Resource 1
Full size table

Human system variables encompass risk perception, risk awareness, preparedness, adaptation, motivation to protect, coping appraisal, gender, trust in government, and prior disaster experience variables. They are often used as independent or control variables. An example is the study by Fox-Rogers et al. (2016), who observed a high flash flood threat appraisal among individuals (independent variable) accompanied by a low coping appraisal (independent variable), leading to a non-protective response (dependent variable). This finding aligns with De Marchi and Scolobig (2012), indicating that risk perception and awareness alone are insufficient to prompt protective behavior. In addition, Fox-Rogers et al. (2016) indicated that a non-protective response can be attributed to a false sense of safety due to the presence of a mitigation measure.

Outcome variables comprise attributes of past events, such as economic and human losses, as well as disaster frequency, intensity, and the affected area. These variables are often used as independent variables to represent changes in the observed or modeled disaster impact due to the implementation of mitigation measures. Therefore, they directly result from the vulnerability and exposure relationship (Damage = vulnerability × exposure) (D’Angelo et al. 2020).

The last group, natural system variables, relates to hazard variables such as hazardous event maps and hydrological model outcomes representing hazards or hazard changes. These variables were mostly employed as independent variables and used to assess how hazards evolved over time, identifying their contribution to increasing risks in protected areas. For instance, Massazza et al. (2021) identified an increasing hazard level for the Niger River basin, which heightened the flood potential for Niamey’s current protection system and urban development patterns.

3.4 Empirical evidence for the SDP and Le phenomena

Most of the reviewed (SDP, n = 6, 75%; LE, n = 24, 75%) case studies provided conclusive evidence to support the occurrence of the studied phenomena (Table 4). In general, the empirical evidence can be categorized into: (a) increased development in protected areas, (b) reduced preparedness and a false sense of safety, (c) increased damage from rare disaster events, and (d) contradictory and inconclusive evidence.

Table 4 Evidence on the phenomena summary
Full size table

3.4.1 Increased development in protected areas

About 40% of the body of research (SDP: n = 5; LE: n = 14) demonstrated that areas protected by hazard mitigation measures experienced increased urban development. Researchers have observed increased exposure, population and/or assets in areas that are susceptible to hydrological hazards, due to the building of dams and levees (Stevens et al. 2010; Collenteur et al. 2015; Hutton et al. 2019; Jiao et al. 2022; Luu et al. 2022; Fu et al. 2023; Mandarino et al. 2023). Others found a higher rate of urbanization in areas protected by hazard protection measures after implementing a mitigation measure than in unprotected areas (Georgic and Klaiber 2022; Ding et al. 2023; Ferdous et al. 2019). Land use and insurance policies were also found to facilitate the increased urbanization of protected areas (Burby 2006; Cutter et al. 2018; Malecha et al. 2021).

3.4.2 Reduced preparedness and false sense of safety

The second set of evidence supporting the SDP and the LE comprises studies that show reduced preparedness and a false sense of safety in areas with hazard protection measures (SDP n = 1; LE n = 10). These studies observed that structural measures can represent a sign of safety for residents (Babcicky and Seebauer 2019), reducing their risk perception and creating a false sense of safety (Dahal and Hagelman 2011; De Marchi and Scolobig 2012; Chang et al. 2022; Krasiewicz and Wierzbicki 2023) even before the construction is completed (López-Marrero 2010; Fox-Rogers et al. 2016). Additionally, non-structural measures, such as flood insurance, may induce the adoption of less protective measures (Blanchard-Boehm et al. 2001). As a result, mitigation measures may reduce the individual motivation to protect and prepare, as individuals in presumably protected areas tend to take fewer adaptative measures than those in unprotected flood-prone areas (Gissing et al. 2018; Richert et al. 2019).

3.4.3 Increased damage from rare disaster events

The last group of supporting evidence comprises studies demonstrating increased damage in areas with a hazard protection measure (SDP n = 2; LE n = 7). Notably, studies have observed heightened damage and/or fatalities resulting from rare disaster events in protected areas compared to unprotected areas (Ferdous et al. 2020). This trend is consistent with findings in historical records of past disasters (Massazza et al. 2021) and estimates of damage caused by extreme hydrological events in different periods (Toshiharu and Narantsetseg 2019; Jiao et al. 2022).

3.4.4 Refuting and Inconclusive evidence

Only one study presented evidence refuting the LE (Starominski-Uehara 2021). Residents living in an area protected by a dam in Brisbane, Australia, did not develop a false sense of safety due to the protective structure. The study found that householders in the protected area did not feel safe despite the dam’s presence due to their past disaster experiences and continued to take precautions. In addition, the occupation of the dam’s downstream area was attributed to the security assurance provided by the authorities.

Two studies yielded inconclusive results. Dufty et al. (2022) observed low-risk perceptions among individuals living in an area protected by a hazard protection measure. However, they did not identify reduced preparedness, with preparedness being measured through disaster insurance and the existence of flood emergency plans. Similarly, Domeneghetti et al. (2015) observed increased exposure to protected floodplains due to urbanization. However, the authors could not establish a causal link.

3.5 Relationship between the SDP and non-structural and individual measures

The relationship between the SDP and non-structural (e.g., policies) and individual adaptation and preparedness measures emerged as a focal point in most studies (n = 29, 74.4%). However, the simultaneous consideration of structural (e.g., dams, levees) and non-structural or individual measures made it challenging to isolate the impact of each measure on the SDP. To clarify this relationship, we categorized the evidence found based on the types of measures considered. Overall, studies often focused on risk management (n = 10), land use and urban development (n = 9), risk insurance or compensation (n = 9), building codes (n = 5), environmental management (n = 5), and individual adaptation and preparedness measures (n = 12) (Table 5).

Table 5 Assessed non-structural and individual adaptation measures
Full size table

The ‘risk management’ class includes disaster risk mitigation policies, preparedness, response, and recovery policies. Overall, researchers found that the effects of the SDP can be mitigated by adopting appropriate policies. For instance, Richert et al. (2019) found that risk mapping and building bans could potentially increase the residents’ risk awareness and thus reduce risk. Conversely, other studies (e.g., Stevens et al. 2010; Lawrence et al. 2013; Malecha et al. 2021; Chang et al. 2022) indicated that these policies might worsen risks by allowing further development in hazard-prone areas.

‘Land use and urban development’ policies focused on urban development were found to enable and indirectly encourage development in hazard-prone areas, while ‘environmental management’ policies can counteract this trend (Malecha et al. 2021). It was observed that ‘land-use and urban development’ policies (e.g.,land-use control, planning) and ‘building codes’ for new development in protected areas are often less stringent than in non-protected areas. For instance, individual measures such as elevating houses were not legally required in levee-protected areas (Burby 2006; Stevens et al. 2010). These studies suggest an overreliance on hazard maps, allowing unrestricted development just beyond the designated hazard zones (Lawrence et al. 2013; Malecha et al. 2021).

The studies analyzing ‘risk insurance and compensation’ policies addressed factors influencing the acquisition of insurance (Blanchard-Boehm et al. 2001) and the impact on risk perception and development (Burby 2006; Cutter et al. 2018). Research conducted in the US suggests these policies may create a false sense of security, hindering individual adaptation (Burby 2006; Cutter et al. 2018). However, other studies have not observed insurance effects, possibly due to mitigation factors from other policies (e.g., Richert et al. 2019).

Finally, one-fourth of the studies (n = 12) assessed ‘individual adaptation and preparedness’ measures. Adaptation measures, such as raising the house floor, dry or wet proofing, and taking insurance, are used to reduce human and asset exposure. Preparedness measures, such as placing sandbags and raising furniture, enhance response capabilities. The presence of such practices indicates attempts to mitigate risk, as individuals do not present a false sense of safety or unprotective behavior. However, López-Marrero (2010) noted that individuals who performed individual measures might present a reduced risk perception as they might be less exposed to hazards.

In summary, non-structural measures present a dual nature and can both enhance and mitigate the SDP. Malecha et al. (2021) found that smaller spatial scales of policy application, such as district and municipal plans, are more prone to urban development and resistant to risk reduction compared to regional or state-level policies. Also, flood insurance schemes and land use policies may stimulate exposure by removing the requirement for elevating houses in areas protected by hazard protection measures as a condition for insurance coverage (Burby 2006; Cutter et al. 2018). Therefore, targeted policies are necessary to prevent occupation and increased exposure in undeveloped flood-prone areas (Burby 2006). In addition, given the dynamic nature of hazards, especially in the context of climate change, adaptable policies that are not overly reliant on static hazard maps are essential (Lawrence et al. 2013; Malecha et al. 2021).

4 Discussion and recommendations for further research

This study has presented a systematic review of 40 studies that assessed the SDP and LE, aiming to provide an overall picture of existing evidence confirming or refuting their occurrence, as well as the methods and variables used to investigate them. One of the new insights of this study relates to the complex dynamics between structural, non-structural, and individual mitigation measures in influencing the SDP and LE. While previous research has primarily focused on structural measures like dams and levees, our study reviewed the impact of non-structural measures, such as land-use policies. For instance, policies restricting development in flood-prone areas, such as zoning regulations and building codes, can reduce the SDP. Conversely, policies encouraging development in these zones, such as subsidies or tax incentives, may inadvertently exacerbate the risk. In general, we found that the impacts of non-structural and individual measures remain less understood. This limitation calls for a deeper investigation into how they might trigger the SDP and LE.

Another key finding is the intricate interplay of variables like risk perception, preparedness, and individual adaptation in shaping the SDP and LE. These factors, often sidelined, play a pivotal role in the effectiveness of mitigation measures. These variables are key drivers of SDP and LE, but their complexity and interaction with various social factors present a challenge. For example, individuals with high-risk perception may be more likely to engage in preparedness activities and adopt individual adaptation measures, which can mitigate the SDP and LE (Richert et al. 2019). However, the presence of structural mitigation measures may lead to a false sense of safety, reducing risk perception and preparedness (Di Baldassarre et al. 2018a). Hence, further research is needed to determine how these variables interact with the effectiveness of disaster risk reduction interventions.

Due to their complex nature, we found that 40% of the studies provide a fragmented understanding of the SDP and LE, concentrating only on exposure or hazard as evidence to confirm their occurrence. However, this narrow focus may lead to overestimations, as it overlooks crucial vulnerability factors such as coping capacity, risk perception, individual adaptation, and the false sense of safety. Consequently, the mere urbanization or population increase of protected areas cannot definitively affirm the occurrence of these phenomena, as individual adaptation and preparedness measures may still be in place. Similarly, stagnant or declining populations within protected areas do not definitively demonstrate the absence or mitigation of these phenomena. Householders may still exhibit a false sense of safety and be inhibited from adopting protective behavior. Hence, we argue that studies centered exclusively on exposure and/or hazard need a complementary vulnerability assessment to better understand the actual risk and provide evidence on the phenomena.

Our review highlights the need for interdisciplinary approaches to understand the SDP and LE. However, the diversity in methods to investigate the SDP and LE poses challenges for broader analysis, as also observed by other researchers investigating adaptation-related concepts (Rufat et al. 2022; Kuhlicke et al. 2023). In fact, the interplay between structural and non-structural measures and urban development in hazard-prone areas remains poorly quantified (Ding et al. 2023), obscuring an objective assessment. While spatial and demographic analysis is one of the primary methods used, it sometimes offers misleading evidence about increased development in hazard-protected areas. This biased perspective overlooks cases where less populous or urbanized regions might still face SDP and LE due to increased vulnerability from mitigation measures (Ferdous et al. 2019).

Most of the reviewed studies concentrated on flood-related hazards. However, disaster mitigation strategies targeting landslides and multi-hazards can also trigger the SDP and LE. Also, similar to other reviews, we found that a key limitation in SDP and LE research is its narrow geographical focus, particularly the underrepresentation of the Global South (Breen et al. 2022). The relative scarcity of research in these regions should not be interpreted as an absence of SDP and LE phenomena. Instead, it should be seen as an opportunity to promote further research in these geographical contexts.

Despite the advancements presented, some limitations need to be acknowledged. Firstly, our focus on English literature and the search terms used limited the comprehensiveness of our findings. We primarily analyzed studies that directly assessed the SDP and LE, but many other socio-hydrological phenomena exist (Table 1). Future studies could explore indirect assessments of SDP and LE, incorporating research of fields other than socio-hydrology (e.g., economics) and considering other languages. Additionally, examining the relationship of these phenomena with broader concepts like moral hazard Hudson and Berghäuser (2023) could offer valuable insights. Secondly, the heterogeneity of variables and methodologies across studies hampered a meta-analysis. In fact, most studies did not provide quantitative assessments due to the complexity of doing so. Despite this limitation, our synthesis of methods and variables still provides relevant insights for promoting increased comparability in future research. Lastly, our review focused solely on SDP and LE in the context of hydrological hazards (e.g., floods, landslides). However, we did encounter studies investigating these phenomena in relation to other natural hazards, such as hurricanes, droughts, and tsunamis. This suggests that future research could explore whether and how the SDP and LE are addressed for different hazard typologies, potentially identifying transferable methods and best practices.

5 Conclusion

In this systematic review, we have critically examined 40 case studies investigating the safe development paradox and the levee effect phenomena within the broader context of socio-hydrological systems. By analyzing the methods and variables across the reviewed studies, we aimed to facilitate the standardization of approaches and pave the way for future meta-analyses, currently hindered by methodological and variable heterogeneity.

We found that existing research predominantly focuses on flood risk, structural mitigation measures, and Global North studies, overlooking the influence of non-structural and individual measures alone on the SDP. Hence, future research should examine how these measures can also trigger the SDP. Relying solely on exposure variables as a single metric to investigate the SDP and LE proved insufficient as it obscures the complex interplay of policies and individual measures for mitigating risk. We thus advocate for integrating exposure and vulnerability assessments to deliver more robust evidence in understanding these socio-hydrological phenomena.

In conclusion, our study adds to the growing body of knowledge on socio-hydrological systems and disaster risk management. We emphasize the need for holistic, integrated approaches that consider structural, non-structural, and individual measures and call for expanded research in underrepresented regions. As hydrological hazards are expected to intensify worldwide, understanding and addressing the complexities of the SDP and LE becomes ever more crucial in building resilient communities. Within this context, a more comprehensive analysis will deepen our understanding of these phenomena, leading to insights into effective mitigation and adaptation strategies.