A social-ecological approach to estimate fisher resilience: a case study from Brazil

Social-ecological systems (SESs), such as fishing communities, are human and biophysical subsystems that are intrinsically connected to one another and strongly depend on natural resources. That is why these human groups are usually the first to feel the effects of policies concerning fisheries and ocean governance and the most affected by them. These policies can potentially build or erode social-ecological resilience (SER), especially if they are coupled with environmental changes. SER assessments offer a valuable tool to identify human-nature linkages, and the implications and feedbacks in SESs when facing human-induced or natural changes. We created a SER index by combining interviews with fishers with environmental datasets on a fine scale that has never been presented for the Brazilian coast. This scale was then tested in marine protected areas that allow sustainable use. Our approach estimated SER from information on fisheries ecosystem services and adaptive capacity at the local scale, considering the individual and community levels. We synthesized blocks of critical indicators of an individual or community’s ability to build and maintain resilience in SESs, such as flexibility, ability to learn, ability to organize, assets, social capital, and ecological characteristics. We identified that fishers’ ability to learn and to organize, as well as the biological sensitivity of an ecosystem are determinant to enhancing SER in the studied coastal communities. A Bayesian model also showed that the fishers’ SER was related to socioeconomic factors, thereby indicating that older fishers, fishers who consistently catch more fish, and fishers with a higher reliance on fishing for their income presented lower index values. By knowing the variables that influence the ability of fishers to cope with changes to their SESs, we can devise smarter management approaches that may include compensatory mechanisms for more fragile fishers. Our findings can also inform decision making about where fisheries management strategies are likely to be more participative and effective in order to minimize the social impacts of policy decisions and increase SER in coastal communities.


INTRODUCTION
Global environmental changes have led to an era characterized by rapid human-induced biodiversity loss (Corlett 2015) associated directly or indirectly to habitat destruction, climate change, and biodiversity overexploitation. This environmental crisis has been hitting the human groups that depend on natural resources the hardest because of its effects on social-ecological systems (SES) (He and Silliman 2019). Thus, human and environmental systems must be considered in unison, as an integrated system, when proposing approaches to cope with and mitigate global and local changes (Colding and Barthel 2019). It is especially important to consider that both environmental changes and the social mechanisms used to deal with changes may impact the resilience of a system, which is, in turn, also formed by environmental and social components. Resilience is defined as the ability of a system to cope with disturbances while maintaining its functions (Folke et al. 2004). For instance, the ecological resilience of coral reefs has declined because of chronic overfishing, acidification, rising temperatures, and water quality, which has led to radical regime shifts that make them more vulnerable to natural disasters (Hughes et al. 2003).
In the social context, resilience is the capacity of human communities to absorb changes and adversities while maintaining their livelihoods (Adger 2000). For example, fishing communities may maintain their social resilience despite transformative changes, e.g., technological or cultural, that involve regime shifts, such as a shift from a fishing economy to a tourism-based economy, if such transformations maintain or improve their livelihoods (Folke et al. 2010). When social systems are nested in ecological systems through mutual feedbacks, i.e., SES, the term social-ecological resilience (hereafter SER) is adopted (Berkes and Folke 1998, Adger et al. 2005). An example of this is topdown biodiversity conservation initiatives (Diegues 2008), such as the establishment of marine protected areas (MPAs), that are imposed upon a community and compromise the use of natural resources (Bueno and Schiavetti 2019). Although these measures can have a positive effect on ecological resilience, they can negatively impact social resilience if those affected cannot find alternative ways to cope with the limitations imposed on the use of natural resources. In turn, decreased social resilience can have a negative feedback on ecological resilience, e.g., by sabotaging conservation, thereby affecting the SER in the end (Carpenter et al. 2001). By better understanding SER within a system, better management policies can be devised and implemented accordingly, thus reducing the risk of SER backfire and negative impacts on local livelihoods. Furthermore, this would allow the simultaneous increase in the sustainability of an ecosystem's functions and services .
Fish are an important marine ecosystem service and support human food security, well-being, and economic and cultural livelihoods, especially where fisheries are primarily small-scale (Béné and Friend 2011). However, global marine fisheries are currently underperforming because of the combined impacts of overfishing, degradation of ecosystems, and climate change https://www.ecologyandsociety.org/vol25/iss1/art23/ (Cheung et al. 2009). This has demanded increasing efforts to restore marine ecosystems and rebuild fish stocks (Worm et al. 2009), which usually require a reduction of fishing exploitation rates. Despite the overall long-term ecological and social benefits (Sumaila et al. 2012), rebuilding small-scale fisheries in developing countries involves an overlooked challenge: fishers do not always have the means to ensure their income, food, and employment during the rebuilding process.
Coastal communities that depend on fisheries can respond to social and ecological changes differently, with coping and adaptation strategies dependent on the interactions between these changes and the SES they are in (Faraco et al. 2016). Local fishers can be affected by both the local ecological conditions, such as fluctuations of marine stocks, and their social surroundings, such as how resource governance is enacted at different scales (Broderstad and Eythórsson 2014). The severity of the consequences will depend on a combination of ecological shifts and institutional reforms, such as changes to fisheries management regimes with the establishment of MPAs (Broderstad and Eythórsson 2014). Although MPAs are a strategy to protect biodiversity and the flow of marine services (McCook et al. 2010), their effectiveness can be stunted when the social factors of SER, such as the local socioeconomic context and governance and livelihood aspects, are ignored in their design (Fox et al. 2012).
Coastal fisheries-dependent communities in Brazil represent an important case study for SER because they exploit a great diversity of fish, use multiple types of gear, and are under different management regimes (WWF 2016), in addition to being affected by complex socioeconomic mechanisms (Ruffino 2016). Here, we had two main goals: (1) to develop a broadly applicable index to measure SER and (2) to use case studies to identify the factors that raise or lower SER at an individual level. We first created an index of different factors at the individual, community, and ecosystem levels that were identified in the literature to contribute to SER, e.g., flexibility, ability to learn and to organize, assets, social capital, and ecological characteristics of the local marine ecosystem. We tested this index by estimating the individual SER of Brazilian fishers who live in MPAs that permit the sustainable use of natural resources. Finally, we analyzed the socioeconomic aspects that influence fisher SERs. The developed index can be applied worldwide, but it is especially proposed as a tool to provide information to support management initiatives that do not compromise, but rather enhance, SER in coastal systems in the tropical developing world in order to support both fish conservation and human well-being.

METHODS
The social-ecological resilience assessment lens Studies addressing SER have increased over the last decade because of an interest in understanding, anticipating, and mitigating the increased risks of social-ecological disasters in coastal areas (Colding andBarthel 2019, Ferro-Azcona et al. 2019). A recent systematic review of the literature on adaptive capacity and SER highlighted that most studies are geographically biased toward developed countries and tend to address biophysical stressors related to climate change (Ferro-Azcona et al. 2019). Moreover, most studies use a single impact pathway, e.g., climate change (Cai et al. 2016), or, when focused on climate change, include only the socioeconomic aspects affecting SER (Ahmed et al. 2016). Additionally, and despite the advertised relevance of MPAs for conservation and the maintenance of ecosystem services (Eriksson et al. 2017), MPA SER assessments are scarce and inconclusive (Ferro-Azcona et al. 2019 identified only two studies on this topic). Thus, by recognizing that MPAs are crucial institutions to enhancing SER (IPCC 2014) and that they are poorly studied in developing countries, we created an approach to estimate fisher SER that combined the biophysical and socioeconomic aspects of coastal communities affected by MPAs.
In Brazil, specifically, the few existing SER studies have either focused on fisheries or adopted an exclusively qualitative social lens. For example, Prado et al. (2015) and Leite et al. (2019) used the SER concept to investigate coastal communities through a more qualitative approach, whereas Lopes et al. (2011) and Jones et al. (2013) analyzed different governance systems in protected areas by assessing flexibility, ability to organize, and diversity of those people living in these areas.
Considering that one of the objectives of measuring resilience is to inform policy and decision makers, we adopted a local-level approach (Hinkel 2011), especially because resilience is context specific (Tolentino-Arévalo et al. 2019). At the local level, there are better chances of understanding the interactions between humans and nature and, therefore, there is less room for misrepresentation of the vulnerabilities of a system (Tschakert et al. 2013). Our integrative and operational approach helps untangle the roots of anthropic stressors in MPAs. The previous attempts to operationalize SER have been bumpy, mostly because of the use of an extensive and not necessarily practical framework that makes implementation difficult (e.g., DasGupta and Shaw 2015). Our index, on the other hand, narrows the resilience indicators to a more localized level to capture the niceties that affect SER within fishing communities, with an easily replicable framework. This index provides rapid systematic measurement of SER across similar systems, facilitates comparative analyses, and enables the identification of interactions where supplementary analyses should be conducted to better plan priorities for the MPA to be effective.
Measuring resilience is context-specific and highly dependent on the type of threats posed (Quinlan et al. 2016). For the developing world, resilience should be understood according to its particular conditions, value sets, and livelihoods, instead of blindly adopting the findings from developed countries with different socialecological needs and limitations.

A framework to build the social-ecological resilience index (SERI)
Despite important advances in studies addressing SER worldwide (Ferro-Azcona et al. 2019), the operationalization of SER still lacks quantitative approaches and there is much to learn about what makes SES resilient or not. An SES model should include the entities of common-pool resources, the links between these entities, and aspects related to resource dependencies, such as resource users, public infrastructure, infrastructure providers, institutional rules, and the external environment (Özerol 2013). Here, we consider the SES of fishing communities located within MPAs by integrating biophysical and social processes, community https://www.ecologyandsociety.org/vol25/iss1/art23/ self-organization, and fisher behavioral aspects. Our framework is strongly focused on the needs of those affected by stresses because they tend to be ignored. We consider that altering the relationship between fishers and fisheries resources could influence fisher resilience, which is likely to depend on how management strategies are delivered and perceived, how environmental changes are felt, and on the intensity of resource dependency (Fig. 1).

Fig. 1.
A conceptual framework for assessing the socialecological resilience (SER) of small-scale fishers as a function of policy, environmental, and socioeconomic changes in the social-ecological system. SRi = Social Resilience at the individual level; SRc = Social Resilience at the community level; ER = Ecological Resilience. Adapted from Marshall et al. 2009.
One way to assess SER is by evaluating the capacity of resource dependent communities to cope with changes and uncertainty, and to evaluate whether they know how to nurture learning and adaptation, and create opportunities for self-organization (Berkes and Seixas 2005). Resource dependency is the relationship that people have with the environment that they depend upon (for example, fishers rely on fish) and is influenced by social, economic, and environmental factors at different scales (Marshall et al. 2009). These factors can interact and vary among resourcedependent communities and individuals, thereby resulting in different levels of dependency and resilience . For instance, a resource user that is socially resilient could improve his/her resilience if he/she is inserted in a community that is also socially resilient.
Our variables are assigned to the individual, community, and ecosystem levels of the fisheries system, because fisher resilience is influenced by fishers themselves, fisheries resources, and aspects related to resource dependency at the community level. In the fisher element, we included personal characteristics known to promote or erode social resilience at the individual level. The resource element aggregates ecological aspects of the resilience of marine ecosystems. Finally, we considered preexisting conditions within communities, such as socioeconomic and political trends, that can affect impacts and a community's adaptive capacity (Cutter et al. 2008).
The framework is broken down into three components: social resilience at the individual level (SRi), social resilience at the community level (SRc), and ecological resilience (ER) of marine ecosystems (Fig. 1). Each component is formed by indicators commonly used in the literature (Adger 2000, Cinner et al. 2009, Ruiz-Ballesteros 2011, such as flexibility, ability to learn and to organize, social capital, assets, and ecological variables (Fig. 2).
The SERI ranges from 0 to 1, after standardization of indicators and subindicators, which were equally weighted. A simple average approach was used to aggregate the indicators (IND) to form a component (C; OECD 2008). The value for each component was calculated as the average of these indicators: SERI = (1/3 * C + 1/3 * C + 1/3 * C ) / 3 (1) An equal weighting approach was used because it is assumed that each component is equally important to building SER in coastal communities: SERI = (1/3 * C + 1/3 * C + 1/3 * C ) / 3 (1) Although the number of indicators in each component is different, we normalized the data to vary from 0 to 1 and we used the mean values of each indicator to reduce possible bias from the number of variables. Because each fisher has his/her own score, we suggested a scale where a high SERI comprises values between 0.66 and 1, a moderate SERI exhibits values between 0.34 and 0.65, and a low SERI comprises values between 0 and 0.33.

Assessing the components of SERI
Social-ecological resilience at the individual level (SRi) was measured from four indicators related to the personal characteristics of fishers: flexibility (FLEi), ability to learn (AL), ability to organize (AO), and personal assets (ASi). These indicators show how flexible a fisher is to changes in the community and/or to new rules defining use and access to resources , Adger 2000. They also involve aspects that show how fishers can learn from environmental disturbances and local social-environmental actions (Adger et al. 2005), and the self-organizing ability of fishers both within the community and related to the fishing activities (Carpenter et al. 2001). Furthermore, communities with better infrastructure and better equipped households are considered more socially resilient because they can promote better conditions to learn and organize (Cinner et al. 2009).
Social-ecological resilience at the community level (SRc) focuses on indicators that represent proxies of a community's social resilience, such as institutional change, economic structure, and collective participation of the community (Adger 2000).
Communities that are more flexible tend to be more resilient because they can more easily circumvent problems that arise with changes (Marschke and Berkes 2006). We selected three indicators to measure SRc: community flexibility (FLEc), social capital (SC), and community assets (ASc). https://www.ecologyandsociety.org/vol25/iss1/art23/ Communities that directly depend on natural resources are highly vulnerable to the effects of environmental management, climate change, overfishing, and environmental degradation, which can potentially undermine SER (Adger 2000). Therefore, the measure of resilience should consider the factors that contribute to ecosystem health and ecological adaptive capacity. Here, estimates of biological sensitivity (BS) and fish species exposure (FSE) were used as input indicators to determine the ER of the coastal ecosystem.  Fig. 3). The reserves share similar climates (humid tropical, Köppen Climate Classification System) and encompass beaches, dunes, dry Atlantic forest, mangroves, lagoons, and some shrubland (part of the semiarid biome known as Caatinga). There are multiple villages located inside the reserves, but we used the reserve as our sample unit. These reserves were established after intense local demand and are theoretically comanaged with the participation of the local people, but do not yet have formal management plans that define access rules. Given the latter, the rules in place are based on general federal laws, such as temporal bans on lobster fishing.   (Dias et al. 2007). This reserve includes a marine area (847 ha) and coastal mangroves, dunes, and semiarid Table 1. Description of the indicators and subindicators of the three components of the social-ecological resilience index (SERI).

Description of the indicators and subindicators
Social resilience at individual level (SRi) • Flexibility indicator (FLEi) The FLEi indicator shows how flexible a fisher is to changes in the community and/or to new rules defining use and access to the resources. The FLEi subindicators seek to determine whether fishers would be open to changing their economic activity. Improving flexibility by including additional income sources can be a key component to building resilience (Cinner et al. 2009). Perception of the marine protected area (MPA) reflects a fisher's flexibility when faced with change and uncertainty of resource policies or conservation initiatives that affect resilience in response to these changes ). Dependence on a narrow range of resources can lead to social and economic stress because a smaller variety of economic resources to pool from makes people more susceptible to market variations (Adger 2000) and to natural fluctuations in resources. Greater diversification in fishing gear implies a more resilient fisher (Cinner et al. 2009). Subindicators: flexibility to change, resource use diversification, gear diversification, and perception toward MPA • Ability to learn indicator (AL) The AL indicator relates to how fishers learn from environmental disturbances and social-environmental actions that occur in their areas. Social or environmental changes can teach fishers how to behave when faced with future disturbances, thus making them more resilient (Adger et al. 2005). Further, fisher memories and the local ecological knowledge they hold are crucial to innovation in resource monitoring and management, and to building SER (Berkes and Seixas 2005). Subindicators: years of schooling, fishing experience, and awareness of fishing impacts • Ability to organize indicator (AO) Societies that have a higher level of social organization are considered to be more resilient because organization, cooperation, and collective action in a community decrease transaction costs and, therefore, make it easier to cope with and adapt to imposed changes (Carpenter et al. 2001). Here, elements of self-organizing ability include financial security, fishing investment, association involvement, and migration. Migration can either be understood as a risk to societies, given that it disrupts social and economic systems, or as an effective adaption strategy to environmental changes given that it represents an alternative during harsh periods in their place of origin (Adger et al. 2015).
Here, migration shows a fisher's ability to deal with emerging risks in their communities, such as the loss of ecosystem services. Subindicators: financial security, fishing investment, association involvement, and migration

• Assets indicator (ASi)
Quantitative data on quality of life was used to estimate fisher assets (ASi) through the proxy "number of home appliances" . We considered that fishers with a greater number of appliances were more resilient. Better household infrastructure is also an indicator of wealth, and wealthier people, within a given community context, are more likely to holistically understand the conservation practices of resource management (Cinner et al. 2012). Subindicator: material style of life Social resilience at community level (SRc) • Flexibility indicator (FLEc) The FLEc, which followed the same reasoning applied to individual flexibility, was measured from economic diversification, which was established by the maximum number of economic activities carried out in a specific reserve. The most resilient reserve was defined by the reserve with the greatest number of activities. Subindicator: economic diversification • Social capital indicator (SC) At the community level, we used the SC indicator instead of ability to learn and to organize. Although both social capital and ability to learn and to organize focus on similar characteristics related to enhancing social resilience, social capital is measured from the collective aspects of resource users and their communities. Social capital, e. g., social cohesion, effective local governance, and capacity for collective actions, contributes to building resilience because it can create an appropriate social environment to sustain changes and unlock the capacities of communities to adapt to changes (Béné et al. 2016). We understand that social capital is a more appropriate nomenclature than ability to learn and to organize at the community level because it covers collective aspects related to fisher behavior and attitudes that influence fisher involvement in local management, assuming that high involvement can influence the promotion of SER. Subindicators: fisher engagement, Knowledge of management rules, collective action, social organizations, and fisher participation • Assets indicator (ASc) Infrastructure provision can be a first step toward involving communities in the conservation process, given that when basic needs are cared for people are free to focus on exterior problems. A good level of infrastructure, with quality schools, for example, can contribute to other indicators of social resilience, such as the ability to learn and flexibility . Better infrastructure leads to higher flexibility and learning capacity because it generates economic alternatives, new sources of information, and opportunities (Adger 2000). ASc was measured by the presence or absence of 12 community-level types of infrastructure: schools, pharmacies, electrical services, banking access, sewage collection and treatment, access roads, food markets, phone services, post office service, police service, health centers, and hotels. Subindicator: community infrastructure Ecological resilience (ER) • Biological sensitivity indicator (BS) Predictable and unpredictable impacts of climate change are expected to affect marine systems and have direct effects on fish species because of the impacts on marine habitats, e.g., coral reefs and mangroves (Scheffers et al. 2016). For instance, temperature has a fundamental effect on biological processes, and the sea surface temperature (SST) has already increased by about 0.7 ºC over the last century (NOAA 2016). Increases in SST cause, for example, coral reef bleaching (Ojea et al. 2017), results in diversity and structural changes that, in turn, lead to reduced catches for fishers who target reef species (Pratchett et al. 2008), and undermine reef resilience . The measures of climate exposure, e.g., changes to SST over time, and coral bleaching risk, e.g., occurrence of coral bleaching events, aim to capture such changes at the community level. Biological information on fish species, such as resilience and vulnerability, can be a proxy for marine ecosystem health because it contains aspects related to species vulnerabilities in fishing . Biological traits were extracted from the FishBase dataset (http://www.fishbase.org/). Subindicators: climate exposure, coral bleaching risk, resilience of fish species, and vulnerability of fish species • Fish species exposure indicator (FSE) The FSE indicator describes the exposure to fishing pressure and captures economic demand and threat level in a single measure, because economic demand can increase the threat level of a species over time. The FSE variables were also extracted from the FishBase dataset. Subindicators: price category of fish species and threat level of fish species https://www.ecologyandsociety.org/vol25/iss1/art23/ areas. The fishers mostly target sardines (Opisthonema oglinum), flying fish (Hirundichthys affinis), and clams (Anomalocardia brasiliana; Dias et al. 2007). The reserve was established after an intensive process led by the local people against destructive environmental developments (shrimp farms and land encroachment) that threatened their land tenure and the use of space (Dias et al. 2007). The RDSE Ponta do Tubarão has its own office where meetings with the villages and partner institutions, such as NGOs and universities, are held. Financial restrictions were identified by the community and reserve staff to be the main hindrance to effective management actions.
RESEX Batoque was created in 2003 and is exclusively marine (601 ha). About 320 families live in this reserve and depend mostly on community-based tourism, local commerce, and fishing (Vidal and Silva 2007), especially lobster fishing (Panulirus argus and P. laevicauda). This reserve also faces problems related to land use and occupation, such as irregular construction near water bodies and dunes, which interferes with local environmental dynamics (Vidal and Silva 2007).
The RESEX Canto Verde is mostly a marine reserve (29.216 ha).
There are approximately 300 families who live in its small land area (578 ha), who depend on finfish fishing (Scomberomorus cavalla, S. brasiliensis, Katsuwonus pelamis), lobster harvesting (P. argus and P. laevicauda ;Schärer 2003), agriculture, crafts, and community tourism for their livelihoods. Canto Verde was created in 2009 after an intense struggle, dating back to the 1970s, among the local residents against land encroachment for tourism (Almeida and Pinheiro 2004). This reserve has a comparatively good level of organization. For example, they count on a local management council, a deliberative council (composed of leaders with the power to plan and execute environmental and social actions in the community), a neighborhood association that has been operating for over 20 years, regulation of land use, and a fisheries agreement that defines local fishing rules (Carvalho et al. 2010).
These coastal reserves were selected because of their high dependence on small-scale fisheries. Notwithstanding the importance of small-scale fisheries in coastal developing countries, increasing evidence indicates that coastal ecosystems have been deeply altered by overfishing, climate change, and changes to fishery governance, thus reducing their productivity and resilience (Ojea et al. 2017). Such effects have intensified conflicts over resources (Prestrelo and Vianna 2016) and, consequently, affected SER and the overall vulnerability of fishing villages. Therefore, understanding the factors that enhance the SER of fishers living within MPAs could help promote sustainability and effective conservation of these areas.

Data survey
In 2010 and 2011, 100 artisanal fishers were interviewed with the use of a semistructured questionnaire (Appendix 3): RDSE Ponta do Tubarão (N = 40), RESEX Batoque (N = 30), and RESEX Canto Verde (N = 30). From a list of names provided by the fisher associations in the region, we selected active full-time and parttime fishers who had been fishing in the area for at least five years. The minimum fishing experience in the area, although arbitrary, aimed to include people who could have felt changes brought about by the MPA. We also used the snowball method to reach the most experienced fishers in each village, by asking interviewees to indicate other experienced fishers who fulfilled the established criteria (Biernarcki and Waldorf 1981). Fishers were free to join the study or leave it at any point after being verbally informed of the research goals. The survey included questions related to the indicators used to estimate the index and fisher socioeconomic backgrounds, such as personal characteristics and fishery aspects. Secondary information about the historical climate data and coral bleaching events, target fish species traits, and socioeconomic aspects of the communities were collected using available online datasets. The necessary information and source data used to calculate the fisher SERI are in the supplementary materials (Appendices 1, 2, and 4).

Data analysis
Kruskal-Wallis tests were used to identify differences among indicators and among the average SERI values of the three MPAs. Pairwise comparisons were computed using Bonferroni post hoc tests to determine which indicators and SERI values were statistically significant (Neter et al. 1996).
The Pearson correlation ratio was used to check the relative importance of each indicator to the construction of the SERI (response variable). This analysis is a variance-based measure to examine dependence between input variables and the composite index (Becker et al. 2017). A Pearson correlation matrix was obtained using the "corrplot" package (Wei and Simko 2017) of the R software (R Development Core Team 2017).
A hierarchical Bayesian generalized linear mixed model (BGLMM) was used to analyze the influence of independent individual socioeconomic factors (fisher's age, household economic dependence on fishing, and mean catch per unit effort (CPUE) on the fisher SERI. Data for these explanatory variables were also collected from the surveys with fishers and the calculations used to define these variables are explained in Appendix 5. A remaining potential source of variation on the SERI could be caused by idiosyncrasies between villages, which we dealt with by including a random village effect. We used the SERI for each fisher (Y i ,) as a response variable of the BGLMM, which can take any continuous value between 0 and 1. We opted for a beta distribution given the interval that bounds the SERI and its asymmetric shape [0, 1] . We assigned a noninformative zero-mean Gaussian prior distribution to the fixed effects with a variance of 100 (Held et al. 2010). In order to compare the goodness-of-fit between each model, three different measures were computed: (1) the Watanabe-Akaike information criterion (WAIC), (2) the root mean square error (RMSE), and (3) the adjusted coefficient of determination (R²). Finally, for each estimated parameter of the final selected model we computed the probability (Pr) that they differed from 0 (Pr > or < 0) using the Bayes theorem property (Faraway et al. 2018). The BGLMs were performed using the R-INLA package (Rue et al. 2009). We also ran a sensitivity analysis to improve transparency in the construction of the SERI and to check the robustness of our findings (Appendix 6).

Fishing communities
The three reserves showed similar characteristics (Table 2). Fishers have been living in the reserves for more than 30 years, long before the areas became MPAs, but not all fishers were born on site (60% born on site). Most fishers were older than 40 (61%) and studied for at least five years (63%). Around 50% of the fishers in each reserve owned boats and had been fishing for more than 20 years. Years of schooling, birthplace, and age of the interviewees were slightly different in RESEX Canto Verde, where fishers studied for more years, were mostly born in the community, and were younger than the other fishers ( Table 2). The economic activities practiced by the fishers, in addition to fishing, were similar among the three reserves, with construction, commerce, and tourism being the most common. The fishers from RDSE Ponta do Tubarão had more motorboats and motorized canoes than the others, who were more artisanal. Additionally, fishers from the three reserves targeted similar species, especially snapper and mackerel (Lutjanidae and Scombridae families).

Social-ecological resilience index (SERI)
The average SERI of fishers, considering all MPAs together, was 0.59 on a scale 0-1 (with a range of 0.35 to 0.78). Most fishers (80%) showed moderate SERI values, followed by 20% high SERI values (highest SERI was 0.78 from RDSE Ponta do Tubarão). The ER component had the lowest average value (0.47, with a range of 0.22 to 0.69), followed by SRi (0.55, with a range of 0.01 to 0.90) and SRc (0.69, ranging from 0.59 to 0.86).
RESEX Batoque had the lowest median SERI value (p < 0.05, Bonferroni post hoc). The MPAs also differed in some of their components (Table 3). Specifically, RESEX Batoque performed worst in the SRc component, followed by RESEX Canto Verde and RDSE Ponta do Tubarão, respectively (p < 0.05, post hoc). RESEX Batoque again performed worse than RDSE Ponta do Tubarão and RESEX Canto Verde in the ER component (p < 0.05, post hoc).
RESEX Batoque was revealed to have lower economic diversification at the community level, lower fisher engagement in environmental monitoring, and precarious public infrastructure. However, fishers from this reserve had more knowledge of management rules and were more actively involved in the fishing association. Collective action and social organization variables were similar among the reserves. All three reserves presented worrisome ER results regarding fish resilience, fish vulnerability, and fish price category. Fishers from RESEX Batoque performed the worst because they targeted fish with lower resilience and higher vulnerability compared to the other reserves. The climate exposure and coral bleaching risks were similar among the reserves, as well as the fish price category. The SRi component was similar among the reserves, with all of them showing low gear diversification and low levels of individual financial security (Table 4).
Indicators from the three components showed around 50% correlations among themselves (p < 0.05). The biological sensitivity indicator was the most important to build the SERI. A correlation of 61% was found for this indicator (p < 0.05) in the ER component (Fig. 4).

Socioeconomic aspects defining fisher resilience
The final Bayesian GLMM included fishing dependency, fisher's age, mean CPUE, and the reserve random effect as relevant variables to explain the individual SERI (Table 5). The best model (based on the lowest WAIC, RMSE, and highest R²) showed that the random effect of the reserve is an important factor in explaining data variability, meaning that the peculiarities of each reserve are important. Specifically, there was an 89% probability that older fishers have lower SERI values and an 83% probability that the higher the mean CPUE of the fisher, the lower their SERI (Fig. 5). Likewise, there was an almost 100% probability (97%) that fishers who live in households with a higher dependence on fishing tended to be less resilient (Fig. 5). Percentages represent the posterior probability that the parameter is different from 0 (Pr > or < 0).

Analyzing the SERI
There is a lack of consensus when it comes to understanding adaptation and resilience processes on a regional scale (Ferro-Azcona et al. 2019), but the findings of this study may provide a key to understanding the most important aspects of SER. By analyzing the resilience of fishing SES, we found low adaptive capacity and high environmental degradation among coastal villages to be critical factors affecting SER, thereby corroborating previous empirical studies (Thiault et al. 2018). For the case studies considered here, we found that the social resilience at both individual and community levels and the ecological resilience were important to fisher SER, with almost all indicators showing around 50% of correlation with the SERI. Thus, for further applications, we strongly recommend keeping the main structure of the SERI because it reflects the main aspects of SER worldwide. Still, for the MPAs considered here, biological sensitivity of the marine ecosystem, which includes climate exposure and traits of fish species, was the most important factor, despite the fact that the MPAs presented different levels of resilience among them.  Notwithstanding the geographic and climate similarities of the region, ecological resilience was different among the reserves, which could be related to the different target species, which have different biological traits. In general, the fish species caught in the reserves were revealed to have low resilience and high vulnerability. Additionally, the studied region presented a moderate increase in sea surface temperature, which can alter the productivity of marine ecosystems (Cheung et al. 2009) and affect fisher livelihoods due to a decline of fishing resources. On the other hand, mangroves have been strongly affected in the study area, as a result of coastal development, pollution, deforestation, and climate change (MMA 2012). Mangrove protection and restoration are urgently needed, considering that they play a crucial role in coastal areas by reducing exposure to hazards and providing resources (Elfes et al. 2014) and, in turn, they reduce social vulnerability and improve resilience (Beck 2014).
At the individual level, fisher social resilience differed little among the reserves. Despite the similarity and high resilience at the community level, infrastructure could be improved overall, especially because our measure was based on the simple presenceabsence of a given infrastructure and not on its quality. In agreement with a previous study (Faraco et al. 2016), the infrastructure of fishing communities were deficient in several aspects: the public health system, security, cultural options, and job opportunities. Improving these conditions could minimize the social impacts of extreme events (McDaniels et al. 2008). Studies of coastal fishing communities highlight that the socioeconomic dimension, such as community public infrastructure, constitutes the most binding dimension of building adaptive capacity (del Pilar Moreno-Sánchez and Maldonado 2013, Cinner et al. 2015.
In general, our findings suggest that some indicators are more important than others and should be considered accordingly.
Given the high contribution of the biological sensitivity indicator to form SERI, together with the lower ER values found in the three reserves, working toward the protection, restoration, and maintenance of biodiversity may enhance SER. Individual and community resilience, especially through capacity building and leadership, should also be promoted.

Socioeconomic aspects affecting SERI
Human actions dominate SES and, in turn, humans influence their own resilience (Berkes et al. 2003). Here, we found that some fisher socioeconomic aspects, namely age, fishing dependency, and mean CPUE, are crucial to defining individual resilience among small-scale fishers in the considered MPAs. Older fishers, fishers who depend solely on fishing, and fishers who had higher mean CPUEs were found to be less resilient. Age is a crucial factor of a fisher's resilience in general , and influences flexibility and adaptability (Teh et al. 2012), and fisher attitudes and perceptions toward fisheries management (Leleu et al. 2012). Older fishers, for example, are more resistant to the idea of learning something new, such as a new profession. The relationship observed between dependency and resilience was expected because the more dependent a fisher is on one specific activity, the less economically flexible s/he will be (Silva and Lopes 2015). Furthermore, user dependence on natural resources and access to markets strongly influence overexploitation and act as a barrier to conservation (Marshall et al. 2010). These findings support the idea that some fisher groups, such as older individuals and those highly dependent on fishing, may require additional incentives to accept any restrictions imposed by conservation policies. Moreover, unmeasured idiosyncrasies among the reserves are another important source of variation in the SERI and could be investigated further, to determine whether they represent measurable aspects of resilience yet to be identified.

SERI caveats and blind spots
Composite indexes are sensible to variable selection. Despite the careful choice of variables and carrying out a sensitivity analysis of the SERI to double check (see Appendix 6, which shows that the SERI is robust to changes in variables), there can still be some bias in their choice and use. For instance, migration can either be understood as a risk to the receiving society, given the disruption to existing socioeconomic systems, or as an effective adaptive individual response to changes, given that it allows emigrants to seek better opportunities somewhere else during harsh periods in their original place (Adger et al. 2015). In this study migration was estimated by analyzing the number of years immigrant fishers have been living in the community and the number of years they have been fishing there. Although some fishers could have been living in a community that never faced issues that forced them to emigrate and, therefore, never needed to show more resilience, we assumed that all fishers live under relatively similar conditions and face the same problems. However, considering that this index can be used in different socioeconomic scenarios around the world, we think that this variable must be adapted to the fisher context, by either increasing or decreasing fisher resilience.
Additionally, some of the indicators were estimated based on a comparative analysis among reserves, which can bring some bias to their operationalization. For instance, the community assets variable, which varied from 0 to 1, assigns a value of 1 to the reserve with the most infrastructure. It does not necessarily mean that this reserve has infrastructure that is good enough to enhance resilience, but simply that infrastructure is relatively better in one reserve than the others. Overall, both households and villages lacked assets and quality of the infrastructure was never assessed. Some of the biological variables are also subject to some uncertainty. For example, improvements to the global coral bleaching events dataset, such as percentage affected, bleaching duration, and mortality and recovery rates, could help refine the ecological resilience component. Although the Brazilian coastline is geographically located where there is not a high occurrence of Ecology and Society 25(1): 23 https://www.ecologyandsociety.org/vol25/iss1/art23/ climatic events, considering the current climate change scenario, the ability to measure SER should be seen as a key step toward disaster risk reduction in the future.
Thus, we suggest that anyone wanting to apply this index to another context make changes to the variables used to estimate indicators, provided that the three components used to build fisher SERs are maintained: individual, community, and ecosystem. Nevertheless, whenever possible, indicator selection should consider the aspects of SER used to build the SERI, given that they are widely accepted in the literature. We understand that fisher resilience is context-dependent, therefore some of the subindicators of our index may be adapted to consider local idiosyncrasies to better reflect the important aspects of local communities. Future analyses should also include variables to account for access to assets (e.g., cultural memory and natural capital), quality of existing infrastructure, and current governance and institutions (e.g., gender relations and levels of trust and cooperation; see Whitney et al. 2017). Despite these caveats, the index seems robust enough and could be used to support fisheries management in MPAs and their surroundings.

SER in Brazilian coastal areas: management implications
Building resilience in coastal regions is imperative, given current trends of urban settlement, resource use, global environmental change (Adger et al. 2005), and the political marginalization of fishing communities (Kalikoski et al. 2009). In Brazil, environmental and socioeconomic changes in coastal areas have affected both fish stocks and local livelihoods (MMA 2012), and although many such changes are common throughout the country, communities vary in their degrees of resilience (Hanazaki et al. 2013). Therefore, it is worth analyzing the case studies to provide specific directions for Brazilian fisheries management.
For example, fishers from RESEX Canto Verde were more flexible overall when it came to substituting fish and fishing with alternative resources and activities, such as collecting crustaceans and shellfish, gathering native fruits, and practicing small-scale agriculture. Yet, in the three reserves, crustaceans were the second most commonly exploited resource, indicating the same dependence on the coastal environment, although not necessarily on the same habitats. This dependence on a few livelihood strategies can lead to social and economic stress because a lower variety of economic resources makes people more susceptible to both market (Adger 2000) and resource fluctuations. Furthermore, fishers do not use a lot of gear which, therefore, limits their ability to exploit other resources, makes them more vulnerable to resource fluctuations (Aguilera et al. 2015), and simultaneously increases pressure on specific resources (Roos et al. 2016). Even under fluctuating environmental and economic conditions, catch diversification can reduce economic risks and be a potential management option for achieving SER (Matsuzaki et al. 2019). Management strategies should support sustainable diversification, whether it be of gear, fisheries, or alternative sources of income to improve fisher flexibility under vulnerable conditions.
Fisher ability to learn and to organize suggests a worrisome low level of fisher engagement in the decision-making process, despite the fact that these reserves were created by local demand. Less than half of the fishers in the three reserves said they were engaged in meetings. Low involvement in community organizations can increase the transaction costs associated with collective decision making (Castello et al. 2009). Moreover, it can affect participatory management because of the lack of trust between communities and other organizations that are also part of the decision-making process (de Vos and van Tatenhove 2011). Furthermore, the low schooling among fishers is an issue likely reflected in fisher plasticity and their low knowledge of the issues that affect fish stocks. MPAs would likely do better if they adopted strategies that integrate different institutions and create horizontal networks of civil society for social learning to facilitate the adaptive capacity of communities (Berkes 2009). Social organization could fill some of the gaps in education and enable fishers to better understand cause and effect connections, as observed in communities in the Brazilian Amazon (Kalikoski et al. 2009) and in Madagascar (Cinner et al. 2009).
The social resilience component at the community level (SRc) performed better than the other index components. However, the overall community resilience in RESEX Batoque should be given more attention, especially regarding social capital, economic diversification, and community infrastructure. A strong social context can enhance learning and adaptive capacity (Maldonado and del Pilar Moreno-Sánchez 2014), which together can also foster adaptive management, which is a useful tool for building resilience in SES (Folke et al. 2002). Resilience can also be improved if there are more flexible and income generating options beyond those offered by fishing (Hanazaki et al. 2013). Improving economic flexibility is not only important for building resilience (Cinner et al. 2009), but also for decreasing noncompliance (Karper and Lopes 2014) because fishers who have greater flexibility can turn to other economic activities when interventions limit fish extraction.
Even though the RDSE reserve presented a better SERI, it also faces challenges that can compromise its effectiveness, such as the lack of environmental awareness, ignorance about management rules, and disagreement with the MPA. Lack of environmental awareness and conservation support are among the global causes of environmental impacts, e.g., pollution, deforestation, and overfishing, and threaten conservation efforts (Easman et al. 2018). As such, urgent environmental education is needed, both in Brazil and elsewhere, to boost support among resource users (Ramírez and Santana 2019). This does not mean that all communities need to be told what to do or learn. In fact, in response to the inefficient management of coastal resources some coastal communities have self-organized and requested changes in the management regimes of Brazilian MPAs (Faraco et al. 2016). Investing in building and strengthening social capital in general can be a much more powerful tool with which to equip people to develop other necessary capacities.
The low ER average is worrisome because it could be suggesting alterations in the ecosystem health, e.g., changes in the trophic chain structure or equilibrium of the community, and this could undermine both fisher and ecosystem resilience. For example, the collapse of a fish stock, depending on the species, could hit select fishers harder. Here, the RESEXs presented the highest number of specialist fishers, whereas fishers from RESEX Batoque showed lower SRi and SRc, thus compromising their adaptive capacity if they were to face a disturbance to this system. https://www.ecologyandsociety.org/vol25/iss1/art23/ Investment in coastal management in Brazil is insufficient, but part of it could be spent on building social capital and increasing user involvement in management, especially in places where comanagement is mandatory, such as the studied MPAs. The presence of community leaders, strong social cohesion, and community-based management are important attributes that lead to successful fisheries comanagement (Gutiérrez et al. 2011) and should be promoted. Overall, for the MPAs studied here, none can be said to be social-ecologically resilient. The fact that two of them (RDSE Ponta do Tubarão and RESEX Canto Verde) perform slightly better than the third by no means suggests a satisfactory resilience level.

CONCLUSION
Our main goal with this study was to develop an index to measure fisher SER based on an integrative ecosystem approach. The analyses performed in this study reiterate important aspects that should be considered when designing management strategies for artisanal fisheries. The results suggest that individual, community, and ecological resilience are all strong indicators of SER. Furthermore, the results highlight areas in need of improvement to enhance fisher SER. The index has some limitations and some improvements could be applied to future global calculations. The proposed index could also easily be adapted to communities outside MPAs if some changes are made to the conservation perception variable, given that in its present form it only related to MPAs.
Overall, we believe that the SERI index proposed here can be of great value to marine conservation planning and to broaden the resilience discussion. It represents a useful framework that can assist governments and environmental organizations to better understand SER among fishing communities. It is a practical and feasible approach that comprises easily collected variables from structured surveys and datasets available online. In a changing and uncertain world, building resilience can sustain SES in the face of unexpected changes. Specifically, in poor fishing communities, where financial security and public services are limited, and the most basic infrastructure is inadequate, unexpected events that disrupt access to resources are often felt the hardest by fishers. Moreover, small-scale fisheries need to be managed by addressing the problems related not only to fish stocks but also to the welfare and satisfaction of the people who depend on them. By knowing the aspects that affect fisher coping strategies in response to changes to their SES, we can devise smarter management approaches that inform where conservation strategies and policy decisions are likely to be most effective.

Social resilience at the individual level (SRi)
This component includes aspects regarding social resilience at the individual level, considering the personal characteristics of individual fishers. SRi was measured through four indicators: flexibility (FLEi), ability to learn (AL), ability to organize (AO) and personal assets (ASi).
The FLEi indicator was based on four sub-indicators: flexibility to change, perception of MPA, resource use diversification, and gear diversification. Flexibility to change seeks to determine whether fishers would be open to changing their economic activity. Fishers who did not want to change their economic activity scored zero, whereas those that said they would be willing to change scored 1 point. The perception of the MPA reflects fishers' flexibility in a context of change and uncertainty of resource policies or conservation initiatives which affect the resilience of people in response to these changes , Marshall 2007. We used the fishers' acceptance of the planning or implementation of the MPA as a sub-indicator of the flexibility at the individual level. Fishers who agreed that a protected area was needed scored 1 point and those who disagreed were attributed 0. The resource use diversification sub-indicator was estimated from the number and types of natural resources exploited by an individual fisher. Fishers who exploited marine and terrestrial resources received 1 point; those who only exploited marine resources received 0.5 points, and those who only exploit fish received zero points (0 points). The gear diversification sub-indicator was measured by the ratio of the number of fishing gears used by a fisher and the variety of fishing gears represented in the sample, i.e., the number of fishing gears used by all the fishers interviewed.
The AL indicator refers to the years of schooling, fishing experience, and awareness of fishing impacts. The level of schooling, specifically, was calculated by the years of formal education achieved by a fisher and the relationship between the Brazilian average formal education and the average of all fishers in the sample. Fishers with a score below the average Brazilian schooling level did not receive points, those whose level was between the Brazilian and the sample average received 0.5 points and those who had a higher level than the sample average received 1 point. Fishing experience was measured by the ratio between fishing experience in years of an individual fisher and his age. Awareness of fishing impacts was estimated by a fisher's perception of the causes of stock reduction (e.g., overfishing, sea pollution, lack of governance, religion, predatory fishing and illegal fishing). Fishers who were unaware of the causes of stock reductions or quoted religion as a reason did not receive any points, those who quoted only one documented cause (e.g., overfishing, sea pollution, lack of governance, and illegal fishing) received 0.5 points, and those who quoted more than one documented cause received 1 point.
The ability to organize (AO) indicator covers aspects related to a fisher's ability to selforganize to prepare for changes to the socio-ecological system, either within the community or in their fishing activities. We understand that such an ability could include financial security, fishing investment, involvement in an association, and migration (but see below regarding how migration can be interpreted either way, depending on the situation). Financial security was assessed by the existence of alternative sources of income (e.g., savings or property), whereas fishing investment was measured by ownership of at least one boat. Fishers that had some sort of financial security or owned a boat scored 1 point in each of these sub-indicators, respectively. The association involvement sub-indicator is estimated by the involvement and frequency that a fisher participates in any community organization (e.g., fisher association, neighborhood organization). Fishers who were not involved with any organization did not score, those who were involved but rarely participated in meetings (maximum once a year) scored 0.5 points and those who showed a higher level of involvement (two or more meetings a year, which were generally held once a month) scored 1 point. Migration can be either understood to be a risk to societies, given its disruption to social and economic systems, or as an effective adaption strategy to environmental changes, given that it provides and alternative to people during harsh periods in their original place (Adger et al. 2015). Here, we used the second concept of migration to show a fisher's ability to deal with emerging risks in the communities, such as the loss of ecosystem services. Emigration may be the most effective way to allow people to diversify their income and build resilience when environmental changes threaten livelihoods (Adger 2000), and thus can function as an adaptive response to these changes. Considering that the communities examined here have similar socioeconomic and environmental characteristics, and there are overfishing concerns with some fish species, we expected equally emerging risks among fishers. Considering that the communities examined here have similar socioeconomic and environmental characteristics, and some of the target species may be overfished or overexploited, we expected equally emerging risks among fishers. Migration was estimated based on a fisher's relationship to the community they live in, by analyzing the number of years the outside fishers have been living in the community and the number of years they have been fishing there. Native fishers or fishers who migrated to the community in their childhood are not considered migrants and they did not receive points.
However, if a fisher had migrated after they had turned 18 (adulthood in Brazil), and only then began to fish there, then they were considered to be a migrant and received 1 point.
Quantitative data on quality of life was used to estimate a fisher's' assets through the proxy "number of home appliances" . Our sub-indicator, assets at the individual level (ASi), measured an individual's material style of life based on the presence of household possessions from a list of 21 appliances, such as television, radio, gas stove, car, and refrigerator. After a normalization process was carried out to create a range between 0 and 1, we considered that those who had a greater number of appliances were more resilient.

Social resilience at the community level (SRc)
We selected three indicators to measure SRc: community flexibility (FLEc), social capital (SC) and community assets (ASc).
FLEc was measured based on economic diversification, which was established by the maximum number of economic activities for a specific reserve mentioned by interviewees and the manager of the reserve. The most resilient reserve was determined to be the one that had a greatest number of activities, whereas in the least resilient reserve the only economic activity carried out was fishing. We also carried out a normalization process to place the variable in a range between 0 (least resilient) and 1 (most resilient). SC was estimated by fisher engagement, knowledge of management rules, collective action, social organizations and participation. Fisher engagement takes on different values depending on a fisher's level of engagement in environmental monitoring actions in the community. For example, a fisher who was engaged in any type of environmental monitoring (e.g., helping with government, university or non-governmental organization environmental projects) received 1 point and those who were not engaged did not score. Knowledge of management rules was based on a fisher's knowledge about local management rules in their communities: fishers who did not know the rules (0 points), fishers who knew one rule (0.5 points) and those who knew more than one rule (1 point). In turn, collective action was measured by demand for the reserve creation. Local demand (1 point) was considered more resilient in opposition to a top-down initiative (0 point), as local demand is assumed to promote community involvement in the management and promotion of SER (Gunderson, 2000). The social organizations sub-indicator was measured by the presence of social and fishery organizations in the MPA's region. Only reserves with social organizations were attributed 1 point and they were considered more resilient than the others. The fisher participation sub-indicator assessed the level of engagement that fishers have in social organization meetings and events (s/he acts only as a spectator, issues opinions, proposes ideas, etc.); essentially, it assessed whether a fisher participation is active or passive. The active fishers were attributed 1 point and the passive ones did not score. Fisher engagement, knowledge of management rules, and fisher participation sub-indicators were measured as the percentage of fishers who scored 0 or 1 in each reserve.
ASc was measured from fieldwork observations of the presence of 12 community-level aspects of infrastructure: schools, pharmacies, electrical services, banking access, sewage collection and treatment, access roads, food markets, phone services, post office service, police service, health centers, and hotels. The total number of infrastructure aspects by reserve was normalized to attribute a proportional score to each community which ranged from 0 to 1, whereby 1 was assigned to the reserve with most infrastructure.

Ecological resilience (ER)
Here, estimates of biological sensitivity (BS) and fish species exposure (FSE) were used as input indicators to determine the ER of the coastal ecosystem. Change (IPCC), the RCP8.5 scenario is the worst for the climate. It ranges from 0.71 to 2.73 ° C in the 2090s compared to the 1990s (Bopp et al. 2013). Here, we considered an increase of up to 0.5ºC to be low (1 point), between 0.6 ºC and 2ºC to be medium (0.5 points) and over 2ºC to be high (0 points).
Coral bleaching risk was estimated by the presence of coral bleaching events in the study area, available from the Global Information System for Coral Reefs (http://www.reefbase.org).
Based on the ReefBase categories of bleaching events (no bleaching, low bleaching, medium bleaching, and high bleaching), we attributed a score for coral bleaching risk: presence of one high bleaching event did not score, presence of low or medium bleaching events scored 0.5 points and absence of bleaching events scored 1 point.
Biological information on target fish can be extracted from scientific datasets available online. We used the indexes of fish species vulnerability (VUL) and resilience available in FishBase (httt://fishbase.org, Froese and Pauly 2017). The vulnerability index integrates the fish species characteristics related to their ecology and life history using fuzzy logic . Similarly, the resilience index combines biological parameters of a species' life history with the intrinsic rate of population growth as the main determinant of resilience because it is the most complete parameter. Resilience is expressed on a scale that varies from very low, low, medium, and high and vulnerability is expressed by low, moderate, high, and very high. Following the FishBase classification, we created three categories for each indicator: low resilience -0 points (including the very low and low categories), moderate resilience -0.5 points (including the medium category), and high resilience -(including the high and very high categories); and low vulnerability -1 point (including the low category), moderate vulnerability -0.5 points (including the moderate category), and high vulnerability -0 points (including the high and very high categories). Once we had those, we searched this information for all fish species caught by each fisher, as cited in the interviews, and used the mean as our sub-indicator value.
The economic demand sub-indicator follows the same estimates of scores and averages used in the VUL sub-indicator. For the threat level sub-indicator, we estimated the score by critically endangered or endangered (0 points), vulnerable (0.5 points) and near threatened, least concern or data deficient (1 point). We double-checked the status presented on FishBase with the status on the Brazilian Red List (Decree nº 445; Brazil's Red List 2014) (MMA 2014) and, whenever there was some divergence, we assumed the latter to be more accurate as it was more recent. No involvement= 0 Low frequency = 0.5 High frequency = 1.0

Assets (ASi)
Material style of life Measured through the proxy "number of home appliances" by household.
Normalization process a was used to assign a proportional score to each fisher.

Flexibility (FLEc)
Economic diversification The number of economic activities mentioned by interviewees and managers. The more resilient reserve was the one that had the greatest number of activities, whereas the only economic activity in the least resilient reserve was fishing.  to approximate normality. The HD variable was measured based on the proportion of people who depend on fishing as an economic activity by household. Table A5.1 below shows these explanatory variables for each of the 100 fishers in the three marine protected areas.

The beta distribution and the goodness-of-fit measures
Beta distributions have long been used in a wide range of applications involving proportions and probabilities . However, only recently have they been applied to linear regression modelling (Ferrari and Cribari-Neto 2004, Smithson and Verkuilen 2006, Liu and Kong 2015, Paradinas 2016 and time-series analyses (Da- Silva and Migon 2016), to allow for bounded estimates and intervals with model parameters that are directly interpretable in terms of the response mean. For these reasons, the SERI was modeled with a beta distribution Yj ~ Be(µj,ȹj) to avoid non-sensical predictions outside the index limits (between 0 and 1) (Paradinas et al. 2016).
In order to compare the goodness-of-fit between each model, three different measures were computed: (1) the Watanabe-Akaike information criterion (WAIC), (2) the Root Mean Square Error (RMSE), and (3) the adjusted coefficient of determination (R 2 ). WAIC can be viewed as an improvement to the Deviance Information Criterion (DIC) that is traditionally used in Bayesian models, and is better suited than the Akaike Information Criterion (AIC), which is usually used with frequentist modelling procedures (Spiegelhalter et al. 2002). Unlike DIC, which is conditioned on a point estimate and is not fully Bayesian, WAIC is a fully Bayesian measure and uses the entire posterior distribution to make inference about the parameters; hence, estimations are more precise (Watanabe 2010). RMSE consists in the standard deviation of the residuals and, thus, measures how much the observed values deviate from the predicted values. The R 2 expresses the percentage of variability in the response variable that was explained by the model.
Given that the decisions taken in the process of weighing the indicators equally could be unclear, we ran a sensitivity analysis to check the robustness of our findings. A sensitivity analysis is a repeat of the primary analysis but uses alternative decisions (or weights) to check uncertainty in the output of a mathematical model (Deeks et al. 2008, Nardo et al. 2008. It is also used to prove that the findings are not dependent on arbitrary decisions. We chose four weighting schemes to check how each component contributes to estimating index values. In addition to assigning the same weight to the three components (equal weight), we also calculated the index by emphasizing one dimension at a time. We did that by assigning a ½ weight to the emphasized component and ¼ to the remaining two. This alternative was run three times: once for each emphasized component. Although the weight variations changed SERI values (Table A6.1), the ordination of the three reserves was similar in the four weighting schemes. In other words, regardless of the weight assigned to the components, the RESEX Batoque was always the least resilient, followed by RESEX Canto Verde and RDSE Ponta do Tubarão (the most resilient).
Additionally, we checked if the SERI average values were statistically different among the three reserves in the four weighting schemes. For that, we ran Kruskal-Wallis tests with pairwise comparisons (Bonferroni post-hoc tests) (Neter et al. 1996) (See table A6.2).