Flood Vulnerability Analysis in Urban Context: A Socioeconomic Sub-Indicators Overview

: Despite indicators-based assessment models for ﬂood vulnerability being a well-established methodology, a speciﬁc set of indicators that are universally or widely accepted has not been recognized yet. This work aims to review previous studies in the ﬁeld of vulnerability analysis in order to overcome this knowledge gap identifying the most accepted sub-indicators of exposure, sensitivity and adaptive capacity. Moreover, this review aims to clarify the use of the terms of vulnerability and risk in vulnerability assessment. Throughout a three-phase process, a matrix containing all the sub-indicators encountered during the review process was constructed. Then, based on an adaptation of the Pareto diagram, a set of the most relevant sub-indicators was identiﬁed. According to the citation count of each sub-indicator, indeed, 33 sub-indicators were chosen to represent the most universally or widely accepted sub-indicators.


Introduction
The prevention of disastrous flood events is supported by models and quantitative flood risk assessments that provide the evidence for risk-based decision making [1]. Nevertheless, the understanding of present and future flood risk is still a challenge [2], since both contextual characteristics and climate changes influence flood risk with uncertain and unclear impacts [3,4]. Contextual characteristics include both the geographical dimension (hazard formation and its interaction with the territory) and the socio-economic aspects of the society (social, economic and political conditions, cultural and institutional norms, societal networks, governance and historical processes) [5]. Consequently, a modern approach relies more on flood risk management-involving not only purely technical measuresrather than only on flood protection [6][7][8]. Thereby, the information on the vulnerability of a territory to flood events support risk-informed decision-making approaches.
The IPCC Fifth Assessment Report [3] defines the vulnerability as "the propensity or predisposition to be adversely affected [by hazards]. Vulnerability encompasses a variety of concepts and elements including sensitivity or susceptibility to harm and lack of capacity to cope and adapt". Moreover, it stresses that "vulnerability and exposure are dynamic, varying across temporal and spatial scales, and depend on economic, social, geographic, demographic, cultural, institutional, governance, and environmental factors". Thus, a territory may be more or less prone to natural hazards risk based on social conditions [9]. For example, [10] proposes the definition of social vulnerability in terms of individual and social groups capacity to respond to any external stress threatening their well-being, with a special attention to socioeconomic and institutional constraints. This framework proposes a proactive approach and the importance of reducing the vulnerability before a hazardous Climate 2021, 9,12 3 of 18 The remainder of the paper is organized as follows. Section 2 clarifies the definition of vulnerability and risk used in this paper. Section 3 provides information on the methods adopted to conduct the literature review, to screen and select the papers as well as to identify the thresholds for sub-indicators retention. Section 4 provides and describes the results concerning the most widely used sub-indicators, along with the discussion of the results. Conclusions are provided in Section 5.

The Definition of Vulnerability and Risk
The vulnerability is generally set out as [20][21][22][23]: where S is the physical predisposition of human beings, infrastructure and environment to be affected by a dangerous phenomenon due to lack of resistance and predisposition to suffer harm as a consequence of intrinsic and context conditions; E refers to the inventory of elements in an area in which hazardous events may occur. Exposure is a necessary, but not sufficient, determinant of risk; AC is the ability of an individual, family, community, or other social group to adjust to changes in the environment guaranteeing survival and sustainability.
The risk is usually set out as [24][25][26]: where H refers to the possible, future occurrence of natural or human-induced physical events that may have adverse effects on vulnerable and exposed elements; E refers to the inventory of elements in an area in which hazardous events may occur; V refers to the propensity of exposed elements such as human beings, their livelihoods and assets to suffer adverse effects when impacted by hazard events.

Terms of Search and Literatura Database Used
The proposed review was completed in a three steps process: a general review of the outcomes of the searches, a brief review of a narrow set of papers and the content analysis in order to establish the most used sub-indicators. Considering the topic of interest, the authors selected some keywords that are intentionally general although fitting for the purpose: TITLE: ["vulnerability OR risk"] TITLE, ABSTRACT AND KEYWORDS: ["indicator OR index", "flood", "urban", NOT "coast"] Since the existence of alternative words, some terms were searched using the instruction OR. In particular, the terms "vulnerability OR risk" were searched in the title field, while the terms "indicator OR index" were searched in the title-abstract-keywords fields along with the terms "flood" and "urban". Considering that many studies involve coastal flooding vulnerability and risk, the authors decided to exclude the term "coast" and its alternative forms. Other options of search used were (1) finalized publications (2) in the type form of article or book chapter (3) contained in sources such as journals and books. Reviews were purposely left out from the research in order to retain only original use of sub-indicators in the vulnerability assessment applications.
The search was performed in three different literature databases: Scopus, Science Direct and Web of Science.

Paper Selection Processes
The selection process was then realized eliminating irrelevant results, most of which focused on different natural or non-natural hazards related to flood events, e.g., contamina-Climate 2021, 9, 12 4 of 18 tion processes, health consequences, as well as applications to rural contexts that were not of interest for this review.
Since the focus of this review is the socio-economic conditions, another significant selection criterion was the presence of socio-economic sub-indicators demonstrating that social vulnerability factors are taken into consideration.
However, since the vulnerability concept is investigated in a variety of disciplines, the selected papers belong to different subject areas as well ( Figure 1). Consequently, it would be a challenge to perform a cross-check review of the references contained in the eligible papers. A common characteristic of the papers is the publication date after 2009. In 2009, the publication of [27,28], indeed, marked a division. Even if the attention to the disaster risk management emerged in the 1990s when the United Nations General Assembly recognized the need for reducing the impact of natural disasters [29] generating a worthwhile debate on disaster risk components. These were discussed and investigated based on the pioneering studies on the social component of vulnerability of Blaikie et al. [9] and Cutter et al. [12,30], furtherly reviewed by [16]. However, the holistic approaches of [27,28] contributed greatly to the creation of indicator-based assessment models literature. All the territories and societies are vulnerable to floods, with impacts that differ from cases and situations, additionally to the occurrence of the hazard. This consideration of systemic impacts is then embraced in water resources decision making.

Paper Selection Processes
The selection process was then realized eliminating irrelevant results, most of which focused on different natural or non-natural hazards related to flood events, e.g., contamination processes, health consequences, as well as applications to rural contexts that were not of interest for this review.
Since the focus of this review is the socio-economic conditions, another significant selection criterion was the presence of socio-economic sub-indicators demonstrating that social vulnerability factors are taken into consideration.
However, since the vulnerability concept is investigated in a variety of disciplines, the selected papers belong to different subject areas as well ( Figure 1). Consequently, it would be a challenge to perform a cross-check review of the references contained in the eligible papers. A common characteristic of the papers is the publication date after 2009. In 2009, the publication of [27,28], indeed, marked a division. Even if the attention to the disaster risk management emerged in the 1990s when the United Nations General Assembly recognized the need for reducing the impact of natural disasters [29] generating a worthwhile debate on disaster risk components. These were discussed and investigated based on the pioneering studies on the social component of vulnerability of Blaikie et al. [9] and Cutter et al. [12,30], furtherly reviewed by [16]. However, the holistic approaches of [27,28] contributed greatly to the creation of indicator-based assessment models literature. All the territories and societies are vulnerable to floods, with impacts that differ from cases and situations, additionally to the occurrence of the hazard. This consideration of systemic impacts is then embraced in water resources decision making.

Threshold Identification for Sub-Indicators Retention
At the end of the selection process, the sub-indicators were screened using an adaptation of the Pareto diagram. The Pareto diagram, indeed, is used when it is necessary to evaluate the importance of the elements, i.e., priorities, decisions [31]. In this case, it helps in identifying the threshold above which we decide to keep a sub-indicator using a specific threshold for the indicators of exposure, sensitivity and adaptive capacity. Based on the matrix results, the sub-indicators were divided in citation count classes then separated

Threshold Identification for Sub-Indicators Retention
At the end of the selection process, the sub-indicators were screened using an adaptation of the Pareto diagram. The Pareto diagram, indeed, is used when it is necessary to evaluate the importance of the elements, i.e., priorities, decisions [31]. In this case, it helps in identifying the threshold above which we decide to keep a sub-indicator using a specific threshold for the indicators of exposure, sensitivity and adaptive capacity. Based on the matrix results, the sub-indicators were divided in citation count classes then separated in two different groups, one containing the 80% of the cumulative relative frequency of the classes that contain the least cited sub-indicators and the other containing the remaining 20% that contains the most cited sub-indicators. The higher frequency value contained in the cut-off class is used as the sub-indicators' acceptance threshold. In this way, we select the sub-indicators that appear more frequently in the papers reviewed.

Study Selection
The search in Web of Science obtained 176 results, in Scopus 196 results, in Science Direct 50 results. The review was then conducted in a three steps process ( Figure 2). Another publication was added to the collection through the service Mendeley Suggest that provides recommendations for papers to read according to the most recent query performed. After all duplicates were removed, the collection resulted in a selection of 227 papers. This selection contained also irrelevant results that were not of interest for this review. The results not relevant for this review were applications to different natural disasters (landslide, earthquake, storm surge, hurricane or tsunamis) or other type of disasters (water contamination, heavy metal contamination) or focused on rural contexts (for a total amount of 45 papers). Then, after the selection based on the presence of socioeconomic sub-indicators, which constituted the main reason based on which the papers were discarded, this process of general review ended with the selection of a narrow set of 60 papers.
in two different groups, one containing the 80% of the cumulative relative frequency of the classes that contain the least cited sub-indicators and the other containing the remaining 20% that contains the most cited sub-indicators. The higher frequency value contained in the cut-off class is used as the sub-indicators' acceptance threshold. In this way, we select the sub-indicators that appear more frequently in the papers reviewed.

Study Selection
The search in Web of Science obtained 176 results, in Scopus 196 results, in Science Direct 50 results. The review was then conducted in a three steps process ( Figure 2). Another publication was added to the collection through the service Mendeley Suggest that provides recommendations for papers to read according to the most recent query performed. After all duplicates were removed, the collection resulted in a selection of 227 papers. This selection contained also irrelevant results that were not of interest for this review. The results not relevant for this review were applications to different natural disasters (landslide, earthquake, storm surge, hurricane or tsunamis) or other type of disasters (water contamination, heavy metal contamination) or focused on rural contexts (for a total amount of 45 papers). Then, after the selection based on the presence of socio-economic sub-indicators, which constituted the main reason based on which the papers were discarded, this process of general review ended with the selection of a narrow set of 60 papers. This set of 60 papers contained both flood vulnerability analysis and risk analysis in an urban context. In fact, even if this review focuses on vulnerability analysis, the misuse of the term risk also determines the investigation of the so-called risk analysis that are vulnerability analysis. At the end of the second selection process, the collection was composed of 35 papers focusing on vulnerability analysis and 25 papers focusing on risk analysis. This set of 60 papers contained both flood vulnerability analysis and risk analysis in an urban context. In fact, even if this review focuses on vulnerability analysis, the misuse of the term risk also determines the investigation of the so-called risk analysis that are vulnerability analysis. At the end of the second selection process, the collection was composed of 35 papers focusing on vulnerability analysis and 25 papers focusing on risk analysis.

Brief Review: Risk Analysis Studies Selection Process
In the set of 25 papers focusing on risk analysis, different combinations of indicators are found although the usual definition of risk should be H × E × V as suggested by [23,24]. In fact, only 10 out of 25 apply the definition as above suggested. The remaining studies apply the vulnerability definition (10 results) either an extension of the risk definition that includes sensitivity sub-indicators (five results) (references without formatting in Table 1).
Climate 2021, 9, 12 6 of 18 The difference between the definitions of vulnerability and risk stands in the estimation of the adaptive capacity of a system. Cai et al. [33] x x x 2019 Chen et al. [34] x x x x 2020 Chen et al. [35] x Domeneghetti et al. [36] x x x 2014 Edjossan-Sossou et al. [37] 2018 Elboshy et al. [38] x x x x 2020 Ellena et al. [39] x Geng et al. [40] x x x 2020 Hossain and Meng [41] x x x 2011 Kaźmierczak and Cavan [42] x x x x 2020 Koc and Is , ik [43] x Kubal et al. [44] x x x 2013 Li et al. [45] x Lin et al. [46] x x 2020 Lv et al. [47] x Maantay et al. [48] x x x 2014 Muller [49] x x x x 2018 Rana and Routray [50] x Ronco et al. [51] x x x 2019 Shi et al. [52] x Sun et al. [53] x x x x 2020 Wang et al. [54] x x x 2014 Yoon et al. [55] x x 2019 Yu et al. [56] x x x Therefore, only the papers containing sub-indicators of adaptive capacity (10 results) were retained for the subsequent content analysis, as they are completely comparable to vulnerability analysis (references in bold italics in Table 1).

Content Analysis
A matrix was constructed gradually integrating the sub-indicators encountered during the reading of the papers. Therefore, as the review progressed, additional sub-indicators were included.
The final matrix, reported in Table A1 in the Appendix A, contains 165 sub-indicators: 40 sub-indicators of exposure, 97 sub-indicators of sensitivity and 28 sub-indicators of adaptive capacity/resilience.
To establish the sub-indicators' acceptance threshold, an adaptation of the Pareto diagram is used. Based on the matrix results, the sub-indicators were divided in citation count classes (each class represent the number of citations received for each sub-indicator). In the diagram in Figure 3, these classes are arranged, depending on their absolute frequency, with a decreasing order. The vertical dashed bar identifies the cut-off class that divides the 80% of the cumulative relative frequency of the classes that contain the least cited sub-indicators from the remaining 20% that contains the most cited sub-indicators. The higher frequency value contained in the cut-off class is used as sub-indicators' acceptance threshold. In this way, we selected the sub-indicators that appeared more frequently in the papers reviewed. In fact, even if the absolute frequency of the remaining classes is higher, these are not relevant citation count classes, as they are referring to the least cited sub-indicators. Thus, we will use the thresholds of six citation counts for exposure, sensitivity and adaptive capacity. Accordingly, the total number of sub-indicators retained as

Discussion on the Sub-Indicators Selected
The most used exposure's sub-indicators are related to population characteristics, above all population density along with population specification like inhabitants age groups (0-4/5, 5-15, >65 years) and household size. Population density and its dynamical variation in time and space along with age stratification measures the different exposition of the territory in terms of people that may be affected when inundation occurs [57]. Other sub-indicators concern the contextual environment, in particular, the urbanized area/builtup area, topography and the presence of green spaces or urban green coverage. Those characteristics influences the severity of a flood event, since they all modify the surface run-off and the soil permeability/imperviousness [58]. These sub-indicators belong to the category of geomorphological and physical sub-indicators that are taken into consideration in this analysis, as they influence socioeconomic aspects.
Sensitivity is more subjected to variability in terms of sub-indicators retained. However, most of them can be attributed to an "entrapment situation". In fact, relevant subindicators belong to social characteristics that reduce the ability to cope with flood events, such as having a low education level or different form of disabilities as well as being foreigners or dependent on others. For example, minorities and foreigners are considered more sensitive targets, because they may have different languages and cultural barriers impeding the penetration of warnings and risk awareness. Moreover, since they occupy lower classes in society, they are prone to live in hazardous areas [59] and may experience difficulties in receiving disaster recovery funds. Being a female is equally considered a sensitivity sub-indicator; however, this application should be limited to context where women have a subordinated role compared to men in terms of education and economic dependence, i.e., less developed countries. The housing type, age of construction, stories above or below ground level, renters or owners also play a role in the sensitivity indicator, since the house may be more or less prone to be damaged and, consequently, exposing the occupiers to injuries or fatalities [49]. It is demonstrated that home owners are more likely to take measures to reinforce the building as prevention measure, and, at the same time, they are more connected to the neighborhood tissue due to a more stable residence [30,60,61]. Being unemployed or having an irregular occupation, indeed, increase the sensitivity as those people have lower saving capacity for house protection [60] and, at the same time, spend most of their time at home [30]. In the case of urban areas, the sub-indicator identifying industries and other important economic activities give a measure of the economic damages if inundation occurs [57].
The sub-indicators retained for the adaptive capacity concern past experience and preparedness to flood events in terms of individual abilities to contrast adverse situations along with institutional ability to communicate good practices, but also infrastructural preparedness. Therefore, the adaptive capacity includes road density and existence of evacuation routes sub-indicators, as proxies for the ability to escape [62], and drainage network density that measures the capacity to collect rainstorm water, preventing flooding [38]. Sub-indicators assessing personal and communities' preparedness and coordination such as preparedness and awareness, past experience and, to a certain extent, also the performance of warning system are usually investigated through the support of techniques that require people and/or experts' involvement, i.e., focus groups, structured interviews and questionnaires [38,50,63].
These results contribute to answer to the issue of which sub-indicators should be used to assess the vulnerability of a territory to flood events. However, the lack of agreement on core sub-indicators pairs with the difficulty of placing the sub-indicators detected in the correct indicator of exposure or sensitivity. In fact, there is neither agreement among the researchers on the collocation of the sub-indicators in one or another category. For example, the age class of the inhabitants has no certain collocation. Several studies [24,[64][65][66][67][68] place the age groups in the sensitivity category, while others [69,70] place them in the physical exposure and finally, only one [71] places the youngest and the oldest age classes in the sensitivity indicator and the people between 15 and 64 years in the resilience class. This is because the age group classes are a specification of the demographic density and, therefore, a sub-indicator of exposure. However, it can also be considered a characteristic of the sensitivity to floods since both children and old people are care-dependent. Moreover, [71] distinguishes these categories from the adult category, since they can help evacuate people during a flash flood event, and, thus, it can be considered a sub-indicator of adaptive capacity.
Similarly, in the exposure indicator, very different forms of the same sub-indicators are used. For example, the group of rainfall contains the following variety of forms: annual maximum precipitation, average annual rainfall, flood seasonal rainfall, monthly average precipitation, monthly total precipitation, continuous rainfall day, maximum rainfall in 24 h and heavy rainfall. However, these different forms could be counted as the same subindicator, since they are different rainfall specifications, and the choice of a specific form is just context-related. Therefore, even if they appear in their exact form only 1-3 times, their cumulative count suggests that the rainfall sub-indicator should be taken into consideration as an important sub-indicator when performing vulnerability analysis. Likewise, in the group of geography and topography, the sub-indicators of urbanized area, rural area, degraded area, vegetation cover, green spaces, forested area and land use could be grouped together in the land use and land cover class (LULC) and researchers can choose to use the form according to context characteristics and data availability instead of necessarily using the sub-indicators of urbanized/built-up area and green spaces/urban green coverage resulted from the Pareto diagram. Therefore, we here identify a set of 33 most used sub-indicators; however, following the reasons abovementioned, the set should be expanded including indicators of rainfall and LULC as well.

Conclusions
This review aimed at identifying the most accepted sub-indicators of exposure, sensitivity and adaptive capacity, filling a gap in vulnerability assessment literature. To date, in fact, researchers have not agreed on the sub-indicators that should be used for assessing the vulnerability of a territory to flood events. Therefore, a systematic review was performed in order to construct a matrix containing all the sub-indicators encountered during the review of the papers. Throughout the Pareto diagram the most used sub-indicators in flood vulnerability analysis were identified. The Pareto diagram, however, is not able to capture similarities among sub-indicators. Therefore, other considerations on sub-indicators were evaluated suggesting the additional use of sub-indicators of rainfall and LULC.
The review contributed also to shed light on the definition of vulnerability and risk that are often use interchangeably. Throughout the revision process, papers conducting a vulnerability analysis instead of a risk assessment-evaluated based on the use of the indicators and the presence of adaptive capacity/resilience sub-indicators-were integrated in the set of papers analyzed.
The set of 33 sub-indicators should be seen a starting point to construct an indicatorsbased assessment model for flood vulnerability with the due limits above discussed.
More work will be needed, however, to determine the sensitivity of the results both to the variables selected and to the index construction. In fact, results could be affected by aggregation techniques and weight attribution and by minor changes in sub-indicators used as well. Sensitivity analysis on an aggregation and weights issue is investigated by several studies with different outcomes. Works on weighting and aggregation sensitivity were carried by [72][73][74][75][76] for flood-specific index and by [77,78] for broader natural-disasters. Further sensitivity analysis are sometimes applied limited to social vulnerability sub-indicators, as in [79][80][81][82], or with an hazard-oriented view, as in [83][84][85]. Sensitivity analysis on minor changes in sub-indicators used for the index construction are performed by [86][87][88] focusing on the comparison of the performance of the Social Vulnerability Index (SoVI) [30]-mostly applied in the U.S.A.-with other flood vulnerability indices or with sub-indicators subset of the SoVI itself. The paper of [88] finds that the employed subset of sub-indicators provided similar results to that derived using the full set of SoVI. However, [87] provides different results when compared to another widely used index (Center for Disease Control Social Vulnerability Index or CDC SVI). The SoVI is more equipped to find age related vulnerability, while the CDC SVI is better at finding socioeconomic related vulnerability. The results of [86] argue that, besides the internal validity of the SoVI, the index fails to explain empirical disaster outcomes. The explanatory sensitivity analyses conducted for the SoVI demonstrates the need for further investigation to obtain indices characterized by internal validity and empirical efficiency. Therefore, a fruitful area of research could start from a set of universally recognized indicators, as here proposed, to weigh and validate them in different contexts in order to achieve satisficing comparable results, rather than a maximizing result for which there is still a long way off.
This review contributes to shorten the sub-indicators selection process, as this work examined the most recent and relevant studies in the field, even if the variety of disciplines to which it refers may constitute a limit. At the same time the multidisciplinarity considered by this work opens up the boundary of the research on socioeconomic vulnerability giving the opportunity to apply the sub-indicators selected in analysis of different fields. Moreover, the choice of the sub-indicators is based on an innovative use of the Pareto diagram for citation counting that facilitates the sub-indicators selection process. The results of this work can be used in future applications as is, or else, researchers could apply the proposed methodology to other natural hazards.

Conflicts of Interest:
The authors declare no conflict of interest.