Livelihood vulnerability and human-wildlife interactions across protected areas

. Protected Areas (PAs) are important wildlife refuges and act as climate change buffers, but they may impact human livelihoods, particularly engendering a high risk of negative human-wildlife interactions (HWI). Understanding synergies and trade-offs among the drivers of overall human vulnerability within PAs is needed to ensure good outcomes for conservation and human well-being. We examined how climate variability, HWI, and socio-demographics affect livelihood vulnerability across three PAs in Mozambique, Southeast Africa. We used structured questionnaires to obtain information on livelihood vulnerability and social-ecological context-specific variables. We applied principal component analysis to understand synergies and trade-offs between the dimensions of vulnerability and linear models to test the effect of social-ecological drivers on vulnerability. We show that households are mostly vulnerable within PAs due to exposure to climate variability and to HWI, and their low capacity to employ livelihood strategies or to have a strong social network. Furthermore, we show that vulnerability to HWI and climate variability increases with distance to strict protection areas within the PAs and distance to rivers, which implies that proximity to strict protection areas and rivers within PAs still promotes better livelihood conditions than elsewhere. On the other hand, we also found that lower access to infrastructure and other livelihood assets enhances vulnerability, which reflects a trade-off within PAs that potentially limits the benefits of socially inclusive conservation. Our results show that the impacts of PAs, HWI, and climate on community vulnerability should not be viewed in isolation, but instead, conservation and livelihood improvement strategies should reflect their interconnectedness. Although livelihood vulnerability appears to be shaped by these general effects of PAs, it is important also to consider the local PA context when addressing or mitigating livelihood vulnerability in and around them


INTRODUCTION
Protected areas (PAs) are the most common strategy worldwide for protecting biodiversity and maintaining ecological services such as clean water, carbon storage, genetic reservoirs, climate mitigation, and soil stabilization (e.g., Watson et al. 2014, De Vos et al. 2017, Wei et al. 2020).However, PAs are complex adaptive systems embedded within a larger context and therefore novel considerations require that PAs' goals go beyond conserving iconic landscapes and protecting endangered species, to also include societal and equity goals that contribute to human wellbeing (Watson et al. 2014, Dudley et al. 2017, Dawson et al. 2018, Rai et al. 2021).Much debate has been put forward on whether conservation goals are compatible with global goals of poverty alleviation and human health, and whether PA conservation actions benefit people living near PAs (Gurney et al. 2014, Naidoo et al. 2019, Abukari and Mwalyosi 2020, Loos 2021).On one hand, PAs have been touted as contributing to poverty alleviation through the provision of economic benefits (e.g., benefit-sharing schemes and tourism revenues), improved infrastructure, and increased supplies of valuable ecosystem services (Gurney et al. 2014).For instance, in Thailand, PAs were found to reduce poverty among human inhabitants by 30% (Fevrraro et al. 2011).On the other hand, PAs may reinforce the vulnerability of communities by perpetuating poverty traps (Ferraro et al. 2011), limit land use and access to natural resources (Joppa et al. 2009, Doumenge et al. 2021), and increase exposure to negative impacts of human-wildlife interactions (HWI) on humans or on wildlife, or both (Conover 2001).The conceptualization of PAs as complex social-ecological systems (SES) is therefore required to advance our understanding on how to manage both conservation and social goals because feedbacks between ecosystems and socioeconomic systems need to be managed in tandem (Palomo et al. 2014, Cumming and Allen 2017, Loos 2021).Such SES sustainable management of PAs is not only important now, but also as new and larger PAs are expected to meet the goals of the Convention of Biological Diversity (CBD) and of the Intergovernmental Panel on Climate Change (IPCC).Also, climate change is likely to redistribute global biodiversity and, together with projected human population growth, this will elicit stronger demand for land that in turn may conflict with PAs (Heller and Zavaleta 2009, Skogen et al. 2018, Abrahms 2021).Today, approximately 50% of global PAs are established in areas where humans live (Dudley et al. 2010), many of whom have nature-dependent livelihoods, high levels of poverty (Ferraro et al. 2011, Miranda et al. 2016), and high overall vulnerability to environmental, political, social, and economic risks (Schaafsma et al. 2021).
We use the term "livelihood" to denote "a means to a living" encompassing people and their capabilities.Livelihoods are considered dynamic, dependent on emerging opportunities, and context specific (Kasperson et al. 2001, Turner 2017).Livelihood vulnerability is defined as the degree to which livelihoods are susceptible and unable to cope with adverse pressures (e.g., climate change, market instability; Adger 2006, Parry et al. 2007, Oberlack et al. 2016).Livelihood vulnerability may be seen as multidimensional (Fisher et al. 2013, Gerlitz et al. 2017) encompassing: (1) exposure, i.e., the nature and degree to which livelihoods are in contact with, or subject to, a stressor or driver of change (McCarthy et al. 2001, Gallopín 2006, Thiault et al. 2021); (2) sensitivity, i.e., the set of conditions determining the degree to which livelihoods are directly or indirectly altered or modified in the short term by exposure to the stressor (McCarthy et al. 2001, Parry et al. 2007, Thiault et al. 2021); and (3) adaptive capacity, i.e., the ability to implement effective responses to changes by minimizing, coping with, or recovering from the potential impacts of a stressor (McCarthy et al. 2001, Whitney et al. 2017, Cinner et al. 2018, Thiault et al. 2021).The dimensions of livelihood vulnerability depend on several factors.First, exposure is determined by the variation in biophysical stressors, such as climate change and climate-related events (e.g., floods, droughts), natural resource degradation, pollution, and pests (Reed et al. 2013).Further, exposure is also a function of social stressors such as agricultural markets, governance, and globalization, likely to make human populations more vulnerable (Räsänen et al. 2016).Second, sensitivity is typically more related to factors intrinsic to households (Brooks 2003).A previous study has shown that households suffering from poor health or lacking secure food and water supplies are less capable of mitigating the impacts of external shocks or adapting to change (Maru et al. 2014).Third, vulnerability in general may decrease in cases where communities have adaptive capacity.As such, livelihood diversification, i.e., a diverse portfolio of activities and social support capabilities to survive and improve life standards, has been proposed as a key strategy to enhance adaptive capacity (Moench 2008, Proag 2014) and reduce risk exposure (Baird and Gray 2014) because it provides households with an array of options to deploy under different conditions.Another such strategy for managing risk exposure is social support and reciprocity between households (Baird and Gray 2014), e.g., exchange of goods, sharing risk information, and collective action.
Livelihood stressors do not act independently.Because stressors might interact, a given livelihood strategy (or alternative livelihood) may alleviate one dimension of vulnerability while simultaneously increasing exposure to HWI.For example, the harvest of non-timber forest products (NTFP) for commercial purposes may alleviate poverty and decrease households' sensitivity (Paumgarten 2005, Paumgarten and Shackleton 2009, Nguyen et al. 2019), but during this activity people may use forests and increase the likelihood of wildlife encounters (Duffy et al. 2016).Food insecurity may prompt individuals to poach threatened species (Borgerson et al. 2019) or to expand subsistence agriculture or livestock herding into areas inhabited by wildlife, and water insecurity may result in more dangerous routes being used to access water (Ratnayeke et al. 2014, Pereira et al. 2021).Such risk-prone activities enhance the likelihood of crop-raiding or livestock predation (Seoraj-Pillai and Pillay 2017) and increase the risk of fatal encounters with wildlife (Barua et al. 2013), but also other HWI impacts such as disease transmission (Blair and Meredith 2018), property damage (Lamichhane et al. 2018), increased child labor and poor school attendance (Barua et al. 2013), and/or effects on social and psychological states, e.g., fear, well-being (Barua et al. 2013, Jacobsen et al. 2020).This is particularly problematic in the context of PAs because it might create a conflict between mitigating livelihood vulnerability and improving socioeconomic development with conservation goals (Sampson et al. 2021, Meyer andBörner 2022), and this has been shown by many studies (Colchester 2004, Lockwood et al. 2006, Maynard et al. 2021, Pereira et al. 2021).Socio-demographic characteristics (e.g., gender, age, education, and wealth) can influence households' abilities to cope with or prevent HWI impacts or other external stressors (Barua et al. 2013, Khumalo andYung 2015).Also, these factors may vary locally, regionally, or globally (Qiu et al. 2021), thus it is important to compare PAs in different contexts to fully understand their impact on livelihood vulnerability and to provide insights on how to manage the vulnerability without compromising PAs conservation goals (i.e., without increasing HWI).To our understanding, there remains a gap in comprehending the interconnectedness of vulnerability dimensions and the consequential impacts of these interactions on the management of both biodiversity and livelihoods.
We examined whether climate variability, HWI, and sociodemographics affect livelihood vulnerability in African PAs.Africa is strongly affected by global change, and it's considered one of the most vulnerable continents to climate change and climate variability (Boko et al. 2007).Africa in general exhibits high soil degradation, incidences of pests and diseases, food and water scarcity, and socioeconomic conflicts, all of which negatively affect local communities (Ado et al. 2018).Exposure to these stressors may force both human and wildlife communities to move and migrate at the risk of increased HWI (Abrahms 2021).Currently, 14% of the land in Africa is designated as PAs (UNEP-WCMC, IUCN, and NGS 2021), and its abundant megafauna and nature-dependent livelihoods render it alone as accounting for 66% of the negative HWI reported globally (Seoraj-Pillai and Pillay 2017).Within Africa, we focused our study on Mozambique where many local communities live within PAs, and those communities rely on subsistence agriculture, are highly vulnerable to climate change, and encounter wildlife daily (Anderson andPariela 2005, Notelid andEkblom 2021).We sampled three protected areas: Quirimbas National Park (QNP), Niassa Special Reserve (NSR), and Limpopo National Park (LNP).Our objectives were to determine: (1) how livelihood vulnerability and its dimensions vary across and within the three PAs, (2) the synergies and trade-offs between the various dimensions of livelihood vulnerability and if they vary between the three PAs; and (3) if the drivers of livelihood vulnerability in the PAs are similar.We expected that the importance of vulnerability drivers would depend on the social and environmental context of each PA.Therefore, the higher abundance of large mammals (WCS 2021) in the north would likely increase HWI in NSR, while the lack of water sources for communities (INE 2017, Pereira et al. 2021) would likely increase risky behavior for water retrieval in rivers thus increasing HWI in both QNP and LNP (see Table 1 for study hypotheses in detail).To test these hypotheses, we built on our previous work (Pereira et al. 2021) and we applied an adapted vulnerability framework (from Hahn et al. 2009) that includes both climate variability and HWI exposure to three PAs in Mozambique using data on livelihoods and exposure from a standard survey data collection protocol to rural communities.

Case studies
Protected areas cover 26% of the terrestrial area of Mozambique (Sitoé et al. 2015).We studied three PAs located in different regions of the country: Quirimbas National Park (QNP; IUCN category VI; Northeast), Niassa Special Reserve (NSR; IUCN category VI; Northwest), and Limpopo National Park (LNP; Table 1.Socio-demographic and village attribute variables used to model livelihood vulnerability in Mozambican protected areas (PA), as well as hypothesized associations between variables and livelihood vulnerability index (LVI).Note: (-) we expect the independent variable to decrease vulnerability; (+) we expect the independent variable to increase vulnerability; (?) we expect the independent variable to either decrease or increase vulnerability.(+female) We expected women to be more vulnerable than men because they are likely to be less empowered, less educated, and have lower incomes, and their rigidly defined gender roles mean they are more highly exposed to human-wildlife interactions (HWI) and climate change (Khumalo and Yung 2015).education proportion of household members with education (continuous) (-) We expected higher vulnerability for households with lower education levels because educated households might have more options for livelihood diversification (Cardona et al. 2012).household_assets number of assets owned by the household (to estimate household purchasing power).Score ranging from 0 to 7.
(-) We expected higher vulnerability for households with fewer physical assets because fewer possessions are linked to a lack of purchasing power (Moser and Satterthwaite 2008).household_size number of household members (continuous) (+) We expected higher vulnerability for larger families because their costs of living are higher and the probability of any member of the family being exposed to HWI is also greater (Pereira et al. 2021) , and distance to the nearest primary road (< 5 km, 5-10 km, > 10 km).For road class, we also considered a sub-category of road types, i.e., national (N, major intercity roads) or regional (R, connecting towns/localities).
(?) We expected two potential outcomes: (-) reduced vulnerability linked to greater village accessibility because it increases mobility, access to markets, resources, and livelihood opportunities (Salerno 2016); (+) increased vulnerability linked to greater accessibility because it might exacerbate resource depletion (Mwangi et al. 2016).https://www.ecologyandsociety.org/vol29/iss1/art13/Combined, the three PAs encompass various socialenvironmental conditions in Mozambique namely, the areas have different levels of poverty and access to infrastructures, opportunities of livelihoods, distinct governance and management systems, and wildlife populations (Table 2).

Sampling design
We sampled households in each protected area independently at three time points: 2013, 2019, 2021 (Table 3), adopting a standard data collection protocol to obtain information on the vulnerability indicators according to recommendations specified by Hahn et al. (2009) and Pereira et al. (2021).This study is an extension of the previous study of Pereira et al. (2021) done in QNP, and we used the same protocols and the same dataset for the case of QNP.
We used data from a total of 481 questionnaires covering 31 villages from 3 studies, 1 in each of the 3 PAs (Table 3).We randomly selected 15-16 households per village in QNP (Pereira et al. 2021) and NSR (this study).In LNP, we chose 25% of the total number of households in each village, ranging from 16 to 23 households depending on village size (Notelid and Ekblom 2021).Because we cannot correct the use of random or stratified household sampling methods in the different studies, we tested the effect of household size as a random variable in our models (see Appendix 4 for model descriptions).The surveys were targeted at the head of the household in QNP and NSR, but wives/ husbands or sons were also interviewed in LNP households (Notelid and Ekblom 2021).The overall sex ratio of interviewees was balanced, with 256 men (53%) and 225 women (47%).The surveys were conducted in the local language (e.g., Makua, Shangaan), either by local researchers or translators from within the communities.All surveys started with a description of the purpose of the study, followed by an explanation about participation (i.e., that participation in the survey was voluntary, that the answers were anonymous, and that participants could withdraw from the study at any time and for any reason).Finally, we asked for oral or fingerprinted consent from each participant.We limited identifying information to the name of the village and the number of the questionnaire.The QNP study was approved by the ethics committee of the "Comissão de Ética para Recolha e Protecção de Dados de Ciências" (CERPDC), and the NSR study was approved by the Wildlife Conservation Society (WCS) under its Collaborative Institutional Training Initiative (CITI) program.The university of the leading author of the LNP study (Notelid and Ekblom 2021) did not require an ethics protocol by the time the study was conducted in LNP.

Survey structure
We used structured questionnaires that included both closed-and open-ended questions to obtain information on livelihood vulnerability and context-specific information (see Appendix 1 for the questionnaire).We used two types of surveys: (1) household surveys to collect information on vulnerability indicators and socio-demographics, and (2) interviews with community leaders to obtain information on village attributes.Socio-demographic data included gender, religion, ethnicity, household size, number of family members, level of education, and number of assets.Village attributes included population size, number of built features in the village (infrastructure, see "development" in Table 1), and road type and road condition, which were used as predictors of vulnerability in the linear models.

Local communities' participation and benefits
Low participation from local communities (Mucova et al. 2018) and generation of revenues is insufficient to benefit communities because of the steep decline of tourism opportunities in the park (tourism is restricted to the coast and islands; MITADER 2012).
The involvement of the private sector created more opportunities for revenue generation in the reserve, and some investments have been made in communitybased approaches and in the involvement of communities in management decisions (Gross 2021).
The park is currently receiving high donor attention, which could potentiate benefits to local communities.The 20% revenue scheme's structure is still underutilized by communities as an opportunity for mitigating sustainability risks ( Givá and Raitio 2017).

Social and political instability
Islamic terrorism; customary land being threaten by the expansion of mining and logging concessions from international companies (Sakota 2020).

Islamic terrorism (Sakota 2020).
Community displacement (Milgroom andSpierenburg 2008, Witter 2013); conflicts between the park and communities (Hübschle 2016, Givá and Raitio 2017, Lunstrom and Givá 2017).Human-wildlife conflicts Crop-damage by baboons and bush pigs is the most common incidence (Pereira et al. 2021) The livelihood indicators data collected with the questionnaires are described in detail in Appendix 3 and summarized in Figure 2. The indicators were used to calculate the livelihood vulnerability index (LVI).

Data processing
We used geographic data from the Department of National Conservation Areas of Mozambique (ANAC) and from the World Database on Protected Planet (UNEP-WCMC, IUCN, and NGS 2021) to measure each village's geographic characteristics, namely location (i.e., within or outside the park or in the buffer zone) and distance to roads, rivers, and nearest strict protection areas within the PA (i.e., highest protection of biodiversity where extraction activities are not allowed).We produced maps and calculated distance metrics in QGIS 3.6.1 (QGIS Development Team 2018).We counted the number of built features in each village to calculate the infrastructure development index (Sahn and Stifel et al. 2003; see "development" in Table 1) and used road type, road condition and distance to roads to calculate the village accessibility index (Pereira et al. 2021; see "accessibility" in Table 1).

Livelihood vulnerability
To assess how climate variability, HWI, and socio-demographics may affect livelihood vulnerability, we calculated the livelihood vulnerability index (Fig. 2; see Appendix 2 for the LVI equation; Hahn et al. 2009).The version of the LVI we used was the one developed in our previous work (Pereira et al. 2021), which includes HWI as an element of exposure.In this formulation, LVI is calculated as the sum of eight components deemed to influence  (Karger et al. 2017), which provides bioclimatic data (BIO4, BIO10, BIO11, BIO14, BIO15, BIO16, BIO17) at a resolution of 30 arc sec, ~1 km² for the period 2010-2040.We selected the database under the simulation of GFDL Earth System Model (ESM4), which is the current model with the highest priority (Karger et al. 2017, Krasting et al. 2018), and under the SSP5-RCP8.5climate scenario following the current business-as-usual situation (Lange 2019).

Data analysis
First, we examined whether the dimensions of livelihood vulnerability vary between PAs using a Kruskal-Wallis analysis of variance (Zar 2010) and the Dunn-Bonferroni method to perform a post-hoc comparison of the pair-wise mean between groups.We compared the LVI and the components across the three PAs, but also within each PA.
Second, we determined if there were synergies and trade-offs between livelihood dimensions using a principal component analysis (PCA; Jolliffe 2002) to identify groups of components based on their patterns of covariance (Hegre and Petrova 2020).
We define synergies as livelihood dimensions located in the same quadrant of the PCA plot (i.e., contribute to segregating groups in the same direction), whereas we define trade-offs as livelihood dimensions located in different quadrants (i.e., having opposing influences).
Finally, to assess the main drivers of LVI within and across PAs, we adopted a multi-stage modeling approach on a set of linear models (LMs) with LVI as the response variable (Zuur et al. 2009).First, we built two models, one with a set of predictors reflecting social variables (e.g., religion, ethnicity, gender, education), and another set of predictors reflecting ecological variables (e.g., distance to PA, distance to rivers; see Table 1 for the specific predictors).Then, we built a combined model that included variables with reliably estimated significant effects (Morin et al. 2020) from the previous models.All variables used in the models were standardized using maximum and minimum values (Vincent 2007).We assessed if collinearity between variables influenced model performance by determining the variance inflation factor (VIF) with a threshold of 3 (Zuur et al. 2007), which indicates that variables with VIF < 3 do not contribute much to collinearity.We selected the best-performing models from each set using the Akaike's information criterion corrected for small sample size (AICc; Akaike 1974).We chose ΔAICc < 2 to identify the best models, and then calculated the final parameter and error estimates by a conditional model averaging of the best models (Burnham and Anderson 2002).Statistical analysis was performed in R (R Core Team 2019) using the packages Hmisc (Harrel 2015), lme4 (Bates et al. 2015), MuMIn (Bartón 2019), FactoMiner (Le et al. 2008), Factoextra (Kassambara and Mundt 2016), and missMDA (Josse and Husson 2016).

Livelihood vulnerability among and within protected areas (PAs)
We found that the average livelihood vulnerability for the three PAs is low-to-medium (LVI= 0.34 on a scale ranging from 0 to 1; SD = 0.07, range = 0.15-0.5),with the most contributing components being social network, HWI, livelihood strategies, and climate variability (Appendix 3, Table A3.1).We found no significant differences between the average PAs' LVI (Z = 0.62, p < 0.73; Fig. 3A) However, the LVI components varied across PAs, with livelihood strategies and climate variability being both significantly higher in LNP than NSR or QNP, respectively (Fig. 3B; Appendix 3, Table A3.2).Interestingly, QNP and NSR were similar across the majority of components (socio-demographic profile, social network, food, water, and climate variability), whereas HWI was similar for LNP and QNP, and health was similar between LNP and NSR (Appendix 3, Table A3.2).We also uncovered strong differences in the importance of components within PAs.In QNP, social network contributed significantly more to vulnerability whereas food had the lowest contribution (Z = 383, p < 0.001; Fig. 3B; Appendix 3, Table A3.3).In NSR,  HWI and social network were the most contributing components (Z = 635, p < 0.001; Appendix 4, Table A4.4).In contrast, in LNP, climate variability and livelihood strategies had the highest contribution to LVI (Z = 179, p < 0.001; Appendix 3, Table A3.5).

Synergies and trade-offs between livelihood components
Our PCA (Fig. 4) explained 37.7% of the variances in our data.Principal component analysis 1 (PCA1; 19.5%) mainly grouped livelihood strategies and climate variability on the positive side of the axis, and HWI and social network on the negative side.Principal component analysis 2 (PCA2; 18.2%) highlighted potential trade-offs between food and climate variability, with these component pairs being located on opposite sides of the axis.We also uncovered some distinctiveness among the three PAs, with LNP being clearly separated from the other two PAs.The vector for climate variability in PCA2 mostly contributed by separating LNP from the other two PAs.In contrast, the QNP and NSR clusters largely overlapped in PCA2, and such overlap corresponds to the components HWI and social network (Fig. 4).

Drivers of livelihood vulnerability across protected areas (PAs)
Overall, we found that LVI increased with greater distance to PA (i.e., distance to a strict protection area within the PA), distance to river, limited access to household assets (e.g., motorcycle, bicycle, radio), or lower village development (e.g., school, health center, electricity, or other infrastructure; Fig. 5; Appendix 4).

Drivers of vulnerability within protected areas (PAs)
We detected variation in the most influential drivers of vulnerability among the PAs we studied.The specific outcomes for each PA are described separately.
Quirimbas National Park: QNP followed the same trend identified for all PAs combined, with households having lower access to assets (household assets), better access to main roads (village accessibility), and those further away from rivers (distance to river) were more vulnerable.We also found that households headed by men were less vulnerable than those headed by women (gender; Fig. 5; Appendix 4).
Niassa Special Reserve: Only two drivers exerted significant effects on livelihood vulnerability in NSR, with households living far from a strict protection area within the PA (distance to PA) and those in villages with more infrastructure (village development) being less vulnerable (Fig. 5; Appendix 4).
Limpopo National Park: Although the distance to river, development, village area, distance to PA, household assets, and gender variables were all included in the best average combined model for this PA, we could not reliably estimate their coefficients because the 95% confidence interval of the coefficients encompassed 0. Consequently, their influence on LVI could not be confidently inferred (Fig. 5; Appendix 4).

DISCUSSION
We set out to examine how climate variability, HWI, and sociodemographics may affect livelihood vulnerability in PAs.We show that the main sources of such vulnerability for communities inhabiting the Mozambican PAs were, in order of importance: . social network .
The PAs did not differ in their overall vulnerability, however, LVI components were more different in LNP than between the other two PAs, primarily because of climate variability.In contrast, QNP and NSR exhibited similar vulnerability patterns due to concurrent influence of HWI and social network.Therefore, the dimensions of vulnerability changed between PAs in accordance with our hypothesis.Human-wildlife interactions were the main contributor to vulnerability in NSR as we had hypothesized, whereas water access was mainly relevant for the LNP context and not so much for QNP.The dimensions driving vulnerability also varied in terms of their interactions.The social network and HWI presented a positive interaction between LNP and QNP, but a negative interaction with both climate variability and livelihood strategies.Overall, we showed that vulnerability increases with distance to a strict protection area within PAs and rivers, and with lower access to household assets and infrastructure.However, at a local scale, gender (for QNP and LNP) and village accessibility (for LNP) also proved important.To our knowledge, no previous study has calculated the LVI of communities living inside Mozambican PAs.We found that our LVI values are, on average, higher for the communities we studied than other communities in Mozambique (Hahn et al. 2009), but less vulnerable than communities living, for example, in the Northeastern Highlands The dimensions that contributed most to vulnerability were exposure and adaptive capacity.In terms of exposure, both climate variability and HWI contributed significantly to livelihood vulnerability in the PAs we analyzed, and both components could be linked to the sensitivity dimension.Indeed, we identified strong synergies between climate variability and water security, as well as trade-offs between food security and both exposure to climate variability and HWI.These results are intuitive given that, in Mozambique, droughts and heavy rains have been frequent over the past 45 years (with an overall average of 1.17 natural disasters per year; Mondlane 2004), with increasing frequency more recently.These disasters have had wellreported effects on Mozambique's communities (Hilemelekot et al. 2021) Hilemelekot et al. 2021), however, we found the opposite.It is possible that the communities we studied have devised livelihood alternatives and are diversifying their food production, rendering them less dependent on agriculture (Hilemelekot et al. 2021).For example, as drought events have become more frequent in LNP, its farmers have shifted from growing maize to more drought-resistant crops (e.g., sorghum), modified the duration of the growing season, and acquired more cattle (Milgroom andGiller 2013, Notelid andEkblom 2021).Such measures may guarantee food security under more extreme climates, but they are not necessarily transferable to other areas.Further, it could also be that our failure to detect relationships in line with previous studies could be because we used an indicator of crop damage (% of households whose agricultural fields were damaged by wildlife) that is not sufficiently sensitive to capture the effects of climate variability on food supply (Pereira et al. 2021).Given the projected changes in climate and growing population size, it will be important that future studies further examine this relationship in detail.
In terms of adaptive capacity, the risks of HWI to communities (Ogra andBadola 2008, Parker et al. 2014, Seoraj-Pillai and Pillay 2017) can be mitigated through social cohesion as reflected in the synergy we identified between HWI and social network (e.g., help and money lending between households) in line with what we found in our previous work Pereira et al. (2021).Strong social networks likely enable greater reciprocal assistance, information flow (Province et al. 2003), trust in community leaders to solve problems (Charles 2021), which combined can ensure greater household security (Baird and Gray 2014).Previous studies have shown that cooperation between households diminishes exposure to HWI because guarding of fields or livestock can be shared, which is a common practice in many African countries (Gross et al. 2019), or through better information exchange on crop varieties, as reported in Tanzania (Muange et al. 2014).
Our results show that the risk of livelihood vulnerability is influenced by the ecological and social context of PAs.Overall, we found that distance to strict protection zones within a PA and distance to a river increased vulnerability.Communities closer to a strict protection area may benefit (by spillover effects or illegal activities) from natural resources that are better preserved there, making them less vulnerable than more distant communities (Naidoo et al. 2019).For example, in QNP and NSR, logging and mining occur further from the strict protection areas (Naidoo et al. 2019).The importance of distance to a river is not surprising because communities in this region use rivers as their main water source (Boer et al. 2019), so they are particularly sensitive to fluctuations in water availability/quality although they are more exposed to wildlife that share the same water sources (Gillingham and Lee 2003).We also uncovered that infrastructure and household assets influence livelihood vulnerability the most.
Having more infrastructure in a community has previously been associated with greater access to education, electricity, health, and water facilities (Boer et al. 2019), which is suggested to decrease vulnerability.At the level of individual households, the number of assets owned (e.g., car, motorcycle, tv, radio), which reflects the purchasing power of the family, is also predictive of livelihood vulnerability.These results highlight the challenges of compromise between socioeconomic development and conservation solutions.On one hand, solutions that increase infrastructure access and economic power of communities in PAs guaranty wellbeing and improve life conditions, while potentially increasing the influx of people within PAs and increases direct and indirect pressures on conservation (Joppa et al. 2009, Caro et al. 2014, de la Fuente et al. 2020).
Although the livelihood vulnerability index values exhibit similarity across the protected areas (PAs), our analysis revealed that distinct environmental and socioeconomic attributes inherent to each PA play significant roles in driving variations in vulnerability within them.Most noticeably, LNP was quite well separated from overlapping QNP and NSR in our PCA, driven mostly by climate variability and livelihood strategies.Limpopo National Park is situated in Southern Mozambique, which suffers from more droughts than the northern regions (Ribeiro and Chaúque 2010), and livelihood strategies are also slightly different from the other two PAs, with more livestock herding and historical labor migration (Notelid and Ekblom 2021).Moreover, we found that HWI was significantly more influential in NSR, which is one of the largest continuous wilderness areas in Africa and hosts Mozambique's largest mammals (WCS 2021).Furthermore, in NSR, the distance to strict protection areas within the PA increased vulnerability, probably due to the benefits communities gain from spillover effect of natural resources preservation and the potential benefits of community-based management that operate in the area (Gross 2021).Village infrastructure emerged as an important driver of vulnerability within the NSR (i.e., reducing vulnerability), aligning with the overarching trend observed across all three protected areas.Notably, a generalized lack of healthcare and health facilities in QNP meant that we identified health as contributing to vulnerability, exacerbated by a paucity of transport links (Asafo-Adjei et al. 2017).Our results also pointed to some of the same important socio-demographic factors identified in QNP (i.e., gender and household assets) being influential in LNP, but the respective statistical results were inconclusive.
Our study represents a first effort to determine vulnerability drivers in three remote areas of Mozambique, therefore it is important to consider potential limitations.First, there might be some bias associated with using data collected at different time periods, especially in the case of LNP, and by different researchers.A2.Tell me about the threats or concerns that are affecting your community.These concerns can be big problems or small problems, which can be affecting your village.(it will be assumed that the order by which the interviewee mentions a problem corresponds to its importance for that interviewee) D2: Does your family always have enough food throughout the year or are there times of the year when food is not enough?(if the answer is "it depends on the year", ask about the last year)  Yes, we always have enough food (move to D4)  No, there are times when it is not enough  In the last year, we have always had enough food  In the last year, we haven't had enough food D3: For how many months a year, did your family struggle to get enough food?

Fig. 1 .
Fig. 1. A. Location of Mozambique within the African continent.B. Location of the three protected areas (PA) studied in Mozambique and their sampled villages: a) Niassa Special Reserve, b) Quirimbas National Park, and c) Limpopo National Park.

Fig. 2 .
Fig. 2. Components and indicators of the livelihood vulnerability index (LVI) we used to assess the vulnerability of rural communities to climate variability and human-wildlife interactions (HWI) in Mozambique.The LVI was adapted from Pereira et al. (2021) and assumes that vulnerability is a function of the dimensions Exposure, Sensitivity, and Adaptive capacity (gray bars), each measured by 2-3 components (black bars), with 3 indicators for each component.

Fig. 3 .
Fig. 3. A. Livelihood vulnerability index (LVI) for the three protected areas (LNP = Limpopo National Park; NSR = Niassa Special Reserve; QNP = Quirimbas National Park).B. Radar plot of the average values for LVI components (HWI = human-wildlife interactions, socio.demo.profile= sociodemographic profile) in the three study areas.

Fig. 4 .
Fig. 4. Principal component analysis (PCA) plot in which the components contributing to the livelihood vulnerability index (LVI) are displayed as vectors.Colored symbols represent the scores per household, grouped according to protected areas, with ellipses representing the 95% confidence interval.Note: HWI = human-wildlife interactions; socio.demo.profile= socio-demographic profile; LNP = Limpopo National Park; NSR = Niassa Special Reserve; QNP = Quirimbas National Park.

Fig. 5 .
Fig. 5. Model results for social-ecological drivers of vulnerability present the best average models for all protected areas (PAs) combined and for each PA.Supporting statistical results are presented in Appendix 4.

D4:
Do you own one or more machambas  Yes  No D5: What is the size of the machamba (in ha)?D6: How far is the machamba from your home (in km)?D7: For how long have you had the machamba?D8: For how long do you plan to keep that machamba (or when do you plan to look for another machamba)?D9: The machamba was given to you by:  Community head (go to D11)  Family (go to D11)  Found it alone D10: How do you choose the place where you will make the machamba?D11: With whom are you going to the machamba?D12: Which type of crop types do you produce?D13: What are some problems that jeopardize the yield of your machamba(s)?(If conflicts with wild animals are mentioned, move to D15) D14: Could the presence of wild animals have compromised the yield of your machamba(s)? Yes.Can you give examples of animals? No (go to D16) D15: How often do animals attack the machamba? Every day  1 or 2 times a week  1 or 2 times a month  A few times a year D16: Does your family keep part of the harvest to eat during the year? Yes  No D17: Does your family keep seeds for next year? Yes  No D18: Do you have domestic animals? Yes.Which and how many of each?

Table 2 .
Characteristics of the three studied protected areas (PAs), socioeconomic characteristics of the communities living within or nearby the PA, and human-wildlife interactions (HWI) and climate change impacts.

Table 3 .
(Gerlitz et al. 2017na 2017)ree studied protected areas, including year, sampled villages, sample size, methodology, and source.Note: LNP = Limpopo National Park, QNP = Quirimbas National Park, and NSR = Niassa Special Reserve.Fig.2).Every component always contributes to increased vulnerability.The LVI builds on the vulnerability framework used by the Intergovernmental Panel on Climate Change (IPCC), which considers vulnerability as a function of exposure, sensitivity, and adaptive capacity(Parry et al. 2007).The LVI was developed and tested for Mozambique byHahn et al. (2009)and is designed to measure livelihood vulnerability in regions that are predominantly rural, poor, and remote.It is a comprehensive, multidimensional aggregate index that is sufficiently flexible to enable the study of different target or geographic contexts according to selective inclusion of components and/or indicators(Hahn et al. 2009).We considered exposure the combination of the components: climate variability and HWI; sensitivity as the combination of health, food, water; and adaptive capacity as the combination of socio-demographic profile, livelihood strategies, and social networks.Each of the components is measured by a set of indicators; in total, our version of the LVI includes 24 indicators selected over 2 steps.First, we pre-selected 36 indicators based onHahn et al. (2009)andPereira et al. (2021).Second, we used correspondence analysis (CA) to reduce the dimensionality of the indicators to obtain a balanced representation of indicators per component, thereby circumventing the issue of components with more indicators having greater weight(Greyling and Tregenna 2017).For components with > 3 indicators, we selected the 3 most important indicators from our CA(Gerlitz et al. 2017; Appendix 2).Values for the selected indicators were obtained from the answers to the survey questions (Appendix 1), with the exception of climate variability.For this indicator, we used data from the global database CHELSA version 2.1 Previous studies have shown https://www.ecologyandsociety.org/vol29/iss1/art13/that climate instability is likely to reduce food production (Seoraj-Pillai and Pillay 2017, Hilemelekot et al. 2021)bility has an impact on water availability for Mozambican communities(Hilemelekot et al. 2021) that use natural sources of water (Nordström 2019,Hilemelekot et al. 2021).Unexpectedly, we found negative associations between climate variability and food security.

2. Questionnaire to households Village number Interview number
Different times of data collection may represent different exposures and potentially differences in the development of livelihood strategies.For instance, we could not account for the influence of natural disasters (e.g., drought of 2015-2016 and Cyclone Kenneth in 2019) on our earlier data collection.The impacts of these natural disasters in the different PAs, which occurred during the survey periods, limit our ability to directly compare LVI across the PAs.As for different researchers, because data were already collected with the goal of being transformed into objective indicators, this might reduce the researcher-bias associated with different data collectors.It is also important to note that the LVI has been criticized for its oversimplification and overgeneralization of the complex concept of livelihoods into indicators that are difficult to replicate(Hanh et al. 2009), so the interpretation of the results should take this into consideration.Lastly, the choice of indicators for calculating LVI is contextspecific and transferring such an index to different areas entails a trade-off between generality and location-specific information.https://www.ecologyandsociety.org/vol29/iss1/art13/Tell me about the threats or concerns that are affecting your community.These concerns can be big problems or small problems, whatever you think is worrying your village.(it will be assumed that the order by which the interviewee mentions a problem reflects the importance of the threat) Do you communicate/interact with any organization/institution or group of individuals regarding issues of management and use of the QNP?If so, which one? (e.g.regarding problems with wildlife, land conflicts, etc.) D2: How often did you communicate/interact, in the last year, with each organization/ institution or group Open and exploratory questions -Community needs and problems Understand which issues are a priority for the communities and if conflicts with wildlife are voluntarily mentioned.If these are mentioned, the interviewer will try to steer the conversation in that direction.
E5: In your opinion, what measures could be implemented to reduce the negative interactions?A1.A1: Describe what a normal day is like for you.What are you doing?Where do you go?Who do you interact with?Encourage the interviewee to tell a story about themselves or something that has happened in the past.
How long does it take you to reach the water source?/At which distance is the water source?E3: How many times a day do you fetch water?You or someone in your household is a member of any decision-making committee, council or political party (for example, natural resource management committees, farmers' committees, Mozambican Women Organization, Mozambican Youth Organization, scouts, associations of parents and teachers, etc.)? Yes.Which?No F2: In the past year, how many times have you asked for assistance from the village structure to resolve any problems or conflicts related to subsistence activities, use or resource management?(e.g.land conflict, problems with land for construction or cultivation of machambas, problems with wildlife).F3:In the past month, has a friend or family member lent you food or money, or helped you with medical care, looking after children, products, or in some other way?In the last month, did you lend a friend or family member food or money, or did you help them with medical care, take care of children, sell products, or in some other way?How many people live with you in the same house?G7: Of the people living with you, how many are under 18 years old?______And over 64 years old?___ G8: Do you live with:  Only family members  Family and friends  _____________________________________ G9: Did the head of the family attend school? Yes.Until which level?___________________________  No G10: Of the people who live with you, how many currently attend or have attended school (including the interviewee and his partner)?_____________________ Did anyone leave the village to study?What has improved or worsened?(interviewer should try to contextualize with landmark historical events -park establishment, civil war, storm Kenneth) H2: How would you like your life to be in the future?What are your wishes for the future?H3: Do you feel you have the capacity to improve your own future? Yes  No  Don´t know H4: In the past 6 years, has there been a natural disaster in your village (cyclone, flood, drought, etc.)?What were the damages?