Assessing Risk and Prioritizing Safety Interventions in Human Settlements A ﬀ ected by Large Wildﬁres

: The large wildﬁres of June 2017 disturbed many communities in central Portugal. The civil parish of Alvares was severely a ﬀ ected, with about 60% of its area burnt. Assessing the risk of large wildﬁres a ﬀ ecting local communities is becoming increasingly important, to reduce potential losses in the future. In this study, we assessed wildﬁre risk for the 36 villages of Alvares parish, by combining hazard, exposure and vulnerability analysis at the settlement scale. Hazard was obtained from ﬁre spread simulations, which integrated exposure together with population and building density within each village. Vulnerability was based on the sociodemographic characteristics of the population, ranked with a hierarchical cluster analysis. Coping capacity was also integrated, considering the distance of each village to the ﬁre station and the time needed for residents to reach a shelter. We simulated 12 di ﬀ erent land management scenarios, regarding the implementation of a fuel-break network and the level of forest management activities. The potential e ﬀ ects of each scenario in the exposure and risk levels of the settlements were evaluated. The results show that, for a business-as-usual scenario, 36% of the villages are at high or very high risk of wildﬁres. Examining each risk component, 28% of the villages are highly exposed, 44% are highly vulnerable, and 22% do not have a potential shelter on-site, calling for di ﬀ erent intervention strategies in each speciﬁc risk dimension. All the land management scenarios, even if designed for other purposes than the protection of settlements, could decrease the proportion of highly exposed villages at di ﬀ erent levels, up to a maximum of 61%. These ﬁndings can contribute to adjust prevention and mitigation strategies to the risk levels and the characteristics of the population and the territory, and to prioritize the protection and emergency actions at the local scale. have shown that risk levels depend on di ﬀ erent combinations of and and therefore actions to reduce risk should be tailored case to case. In addition, wildﬁre risk from and land-use The sharp and aging land it imperative to ﬁnd a suitable territorial vocation, perhaps di ﬀ erent from before, but capable of creating more economic value and promoting the management of the land in a more e ﬀ ective and way, thus also reducing the risk of wildﬁres.


Introduction
In many ecosystems, fire is an inevitable process that coexists with human communities [1,2], but the negative effects of wildfires have increased over the last decades [3]. Wildfire disasters causing huge impacts and human losses occurred recently in different parts of the world, such as Australia, California, Chile, Greece, and Portugal [4][5][6][7]. Threats to human safety and property are a major challenge to wildfire management and, in most countries, their protection is a responsibility that falls primarily on government agencies [8,9]. Often, firefighting options prioritize the protection of people first, and the defence of built-up structures. However, the unprecedented dimension of wildfires and their massive negative impacts highlight the need to enhance the institutional capacity and to reinforce other

Delimitation and Characterization of the Human Settlements in Alvares
We identified 36 settlements in the civil parish (Figures 1 and 2), which are recognized as hamlets or small villages by local population; as such, both terms are here used with equivalent meanings. We delimited the settlements based on the definition of "Lugar" by the National Institute of Statistics (https://smi.ine.pt/Conceito/Detalhes/2990). A settlement is defined as a population cluster with at least 10 residential buildings, with spatial continuity and separated from other groups by more than 500 m. The perimeter of the settlements was established in the line of contact between the furthest buildings of each village and the forested or wildland areas nearby; the settlement can include the agricultural land surrounding the core of the residential cluster, in case it exists. One of the settlements identified is an exception, as it corresponds to a cluster of six non-residential buildings, but the number of people working there surpasses the number of inhabitants in most of the residential clusters. As such, the higher number of exposed people justified its integration as an individual spatial unit (Figure 2, nr. 13). In two other cases, the number of residential buildings is below 10, but they are sufficiently distant from other groups, have a specific name, and both inhabitants and local authorities recognize them as distinct places (Figure 2, nr. 28,36).
The area of the settlements ranges between 0.02 km 2 and 0.8 km 2 , and the number of residents varies between 1 and 208, besides the two villages without permanent inhabitants (Figure 2, nr. 35 In the civil parish of Alvares, there are 812 inhabitants, 53% of whom are women, according to the latest census (2011) [52]. The demographic structure of the population is highly skewed, with only 12% of people under 20 years old, and 47% of people over 65 years old. More than a third of the residents only have elementary school education and 17% are illiterate, whereas less than 2% have an university degree. The economic structure of the parish is marked by a low percentage of active population (28%), the large majority working in the tertiary sector (56%). About 11% of the residents are unemployed and 53% are pensioners or retired.
There are 1662 buildings in Alvares, the large majority with a residential function. Around 76% were built before 1980 and 17% of the housing was vacant in 2011. In the framework of this research, field work was carried out in 2018 in the parish, which allowed to update the number of functional buildings (1634) and the number of people (799), as well as some of the sociodemographic characteristics of the residents.

Delimitation and Characterization of the Human Settlements in Alvares
We identified 36 settlements in the civil parish (Figures 1 and 2), which are recognized as hamlets or small villages by local population; as such, both terms are here used with equivalent meanings. We delimited the settlements based on the definition of "Lugar" by the National Institute of Statistics (https://smi.ine.pt/Conceito/Detalhes/2990). A settlement is defined as a population cluster with at least 10 residential buildings, with spatial continuity and separated from other groups by more than 500 m. The perimeter of the settlements was established in the line of contact between the furthest buildings of each village and the forested or wildland areas nearby; the settlement can include the agricultural land surrounding the core of the residential cluster, in case it exists. One of the settlements identified is an exception, as it corresponds to a cluster of six non-residential buildings, but the number of people working there surpasses the number of inhabitants in most of the residential clusters. As such, the higher number of exposed people justified its integration as an individual spatial unit (Figure 2, nr. 13). In two other cases, the number of residential buildings is below 10, but they are sufficiently distant wildfire hazard is higher [27]. Less than 10% of the settlements (three of them) congregate 54% of the total population of the parish, whereas 31% have less than four residents. About 44% of these settlements do not have young people below 20 years of age, and in 22% of them, ¾ of the inhabitants are over 65 years old. We found people with a university degree only in seven settlements, and in 10 of them, at least half the people have only the elementary education level. Over 60% of the settlements have less than 30% of the economically active population, most residents being retired or pensioners. In 56% of the settlements, most buildings (> 75%) were built before 1980 and in nearly half of these, around one-third of the buildings are made of adobe walls or loose stone masonry with wooden structure.

Data Collection and Risk Analysis Procedure
The risk components of exposure and vulnerability were first assessed separately and aggregated afterwards in a 5-class risk index. Each component represents a specific dimension of the wildfire risk, requiring distinct data and tools. This allows integrating quantifiable variables at different steps, which can be translated into indicators and mapping tools. Each settlement was considered an individual spatial unit and the non-populated area was excluded, since the focus were the places where people and built-up structures spatially coincide within the civil parish.

Exposure
Exposure results from the combination of hazard levels with the exposed elements. Fire hazard was quantified by simulating the spread of wildfires in the landscape, using the Fire UNCertainty SIMulation system (FUNC-SIM) developed in the scope of the project [53] by the Forest Research Centre, School of Agriculture of the University of Lisbon (Portugal). FUNC-SIM simulates thousands of potential wildfires burning under historical meteorological conditions that were coincident with very large wildfires (VLW, > 1000 ha). This threshold was selected based on the fire history of the civil parish, as 90% of the total burned area since 1975 was caused by 10 large fires above 1000 ha. Ignitions are randomly located based on a probability density surface estimated from historical ignitions that led to VLW. Fuel maps were derived from the Portuguese Land Use and Land Cover The area of the settlements ranges between 0.02 km 2 and 0.8 km 2 , and the number of residents varies between 1 and 208, besides the two villages without permanent inhabitants (Figure 2, nr. 35 and 36), but where absent owners may return to for holidays, especially in summertime, when wildfire hazard is higher [27]. Less than 10% of the settlements (three of them) congregate 54% of the total population of the parish, whereas 31% have less than four residents. About 44% of these settlements do not have young people below 20 years of age, and in 22% of them, 3 4 of the inhabitants are over 65 years old. We found people with a university degree only in seven settlements, and in 10 of them, at least half the people have only the elementary education level. Over 60% of the settlements have less than 30% of the economically active population, most residents being retired or pensioners. In 56% of the settlements, most buildings (> 75%) were built before 1980 and in nearly half of these, around one-third of the buildings are made of adobe walls or loose stone masonry with wooden structure.

Data Collection and Risk Analysis Procedure
The risk components of exposure and vulnerability were first assessed separately and aggregated afterwards in a 5-class risk index. Each component represents a specific dimension of the wildfire risk, requiring distinct data and tools. This allows integrating quantifiable variables at different steps, which can be translated into indicators and mapping tools. Each settlement was considered an individual spatial unit and the non-populated area was excluded, since the focus were the places where people and built-up structures spatially coincide within the civil parish.

Exposure
Exposure results from the combination of hazard levels with the exposed elements. Fire hazard was quantified by simulating the spread of wildfires in the landscape, using the Fire UNCertainty SIMulation system (FUNC-SIM) developed in the scope of the project [53] by the Forest Research Centre, School of Agriculture of the University of Lisbon (Portugal). FUNC-SIM simulates thousands of potential wildfires burning under historical meteorological conditions that were coincident with very large wildfires (VLW, >1000 ha). This threshold was selected based on the fire history of the civil parish, as 90% of the total burned area since 1975 was caused by 10 large fires above 1000 ha. Ignitions are randomly located based on a probability density surface estimated from historical ignitions that led to VLW. Fuel maps were derived from the Portuguese Land Use and Land Cover map [54], using the fuel model typology defined by Fernandes (2005) [55] to assign the corresponding fuel class. Uncertainty in the fuel model distribution was integrated in the simulations using a stochastic approach. This was important to estimate the impact of different management scenarios (described in Section 2.4), on the fuel distribution at the landscape-level, and consequently, on the fire hazard. The spread and behavior of each wildfire was simulated using the FARSITE simulation system (United States Department of Agriculture, Forest Service, Rocky Mountain Research Station) [56] at a 100 m spatial resolution. The simulated wildfires were combined into a burn probability map, defined as the percentage of times a given grid cell burned. The probability of a wildfire affecting a settlement was estimated as the percentage of times the cells in its vicinity were burned, assumed as the number of times a fire could reach any part of the settlement's perimeter, i.e., the interface between the built-up structures and forested area.
The exposed elements represent the people and buildings existing in the settlements of the parish, weighted by their importance or need of protection, as given by the municipal plans of defense against forest fires [57]. To calculate the exposed elements, population data were obtained from the latest census survey at the subsection level [52] ( Table 1). The subsections are spatial units defined for statistical purposes that roughly correspond to the smallest populated places ("Lugar"), as defined beforehand. The number of buildings was obtained from a cartographic database at the municipal level and complemented with the statistical data available for the subsection (Table 1). When possible, these data were updated with fieldwork done during 2018. Exposure was estimated as a combination of fire hazard and exposed elements, weighted based on the guidelines of the municipal plans of defense against forest fires currently in place, as follows: Exposure = BP * [norm((pd*1) + (bd*0.75)) * 100] (1) where BP corresponds to the Burn Probability (Hazard level), pd represents the population density of each settlement (weighted by 1), bd represents the buildings density of each settlement (weighted by 0.75), and norm means the normalization according to the minimum and maximum values found within the parish, scaled in the range 0-100. The values of exposure were classified into five categories, using the range of the normalized values to define the thresholds. The classes represent intervals of equivalent size with progressive increments in values.

Vulnerability
Vulnerability is assessed according to the sociodemographic characteristics of the population and their living conditions, which influence the potential for loss of the communities [9,[35][36][37].
We collected data on gender, age groups, education levels, and working sectors, as well as data on building age, construction materials, and arrangement, at the subsection level, from the latest census [52] (Table 1). These variables were integrated in a hierarchical clustering analysis with Ward's method [58], to identify groups based on the (dis)similarities in the variables amongst the settlements. In this method, each settlement is initially considered an individual cluster and compared to the other settlements, one at the time. When another settlement is joined with the group, the mean and variance of the variables in the cluster are recalculated. The association of the new settlement with that cluster is determined by the sum of squared error (SSE) in relation to the group mean. Low SSE values indicate high similarity of the new settlement with the cluster. The process is repeated until all the settlements are tested and integrated in a cluster.
The analysis provided five different clusters that were matched with a specific vulnerability class, according to the values (mean, median, and maximum) of each variable and their relative position in relation to the other clusters. The variables were analysed in view of their influence in vulnerability levels, either positive when it decreases vulnerability, or otherwise negative. The vulnerability class was attributed according to the number of variables with negative and positive influence in each cluster, giving more importance to those representing population.

Coping Capacity
Coping capacity represents the ability to react to a wildfire event. It was incorporated as an additional factor of vulnerability, considering that a lower ability to respond will increase the potential for loss of exposed elements. Coping capacity was measured in two different dimensions: (i) the distance of each village to the fire station, representing the institutional capacity of rescue and emergency services; (ii) the time needed for people in each village to evacuate and reach a safe area (shelter), representing the ability for self-defence of residents.
The distance to the fire station was calculated considering the ground distance via paved roads, measured in metres, between the centroid of each village polygon and the nearest fire station building, which is located in the settlement of Alvares (Figure 2, nr. 2). All other fire stations located in adjacent parishes were farther away.
Regarding the time of evacuation to a shelter, we applied a pedestrian evacuation model based on GIS analysis obtained from the United States Geological Survey (USGS, Reston, Virginia), the Pedestrian Evacuation Analyst Tool (PEAT) [59,60]. Initially implemented in coastal communities potentially affected by tsunamis, this tool has also been recently applied to areas affected by landslides and debris flows [61,62]. Based on land cover types, topographic conditions, road network, and existing structures, this tool allows to define safe meeting spots and evacuation routes within a hazardous territory, and the results are converted to mapping tools that wildfire managers can easily interpret. It uses slope directionality to calculate travel speed, with an anisotropic approach, i.e., walking downhill is faster than walking uphill. To implement this evacuation model, we first had to identify the safe areas within the settlements, in this case, an existing building that could be converted into a wildfire shelter. The shelters are considered the safe zones to where people must evacuate, whereas all the remaining territory, which is mostly forested area in this parish, is the Hazard Zone. We identified potential shelters in each village through field work carried out in 2018, according to the following criteria: (i) possibility to restructure the building in order to comply with fire safety rules; (ii) low fuels loads around the building and ability to implement fuel management over time; (iii) accessibility to the shelter building from all other locations in the village, preferentially public buildings such as community centres or chapels. These conditions are recommended by the Programme "Safe Villages, Safe People", created in 2018 and currently being implemented in selected villages by local authorities and civil protection services. Since population density is low in these villages, the size of the building was not an issue. The other base layers were a Digital Elevation Model and a land use and land cover (LULC) map ( Table 1). The LULC map was aggregated into nine categories, weighted by the speed-conservation value (SCV). The SCV represents the fraction of the maximum speed that could be achieved across a given landcover type [60,63]. Values range from zero (when travel is not possible) to 1 (when pedestrian evacuation speed equals the base travel rate), as shown in Table 2. The paved roads were given the SCV of 1, as they correspond to the preferable path. To estimate the time required to reach one of the safe zones (shelters) identified, the tool calculates the cost path based on a specific travel speed defined by the user. In this case, bearing in mind the high proportion of elderly people in the parish, we opted for a conservative approach and defined a speed of 1.03 m/s (approximately 3.7 km/h), considering the mean walking speed of people above 70 years old [64,65]. The model estimates the time required to reach the closest safe zone for each grid cell (at 5 m resolution) in the civil parish area, in minutes. The estimation for each settlement was retrieved from the maximum values found within the cells of each settlement, assumed as the maximum expected evacuation time (on foot) from the buildings in the village to the nearest shelter. Table 2. Speed-conservation values (SCV) for the nine categories of land cover used in the Pedestrian Evacuation Analysis Tool. Values were based on the tool guidelines [60] and adjusted to the characteristics of the civil parish. Coping capacity was incorporated in vulnerability by adjusting the class initially given to the settlements. The settlements that were distant more than 8 km from the fire station and where people would simultaneously take more than 30 min to reach a shelter, were given the higher vulnerability class that succeeded the original one. On the contrary, the settlements located less than 2 km apart from the fire station and with an evacuation time below 20 min, and those up to 5 km far from the fire station but with an evacuation time under 10 min, would change to the immediately lower vulnerability class, if available. The 20 min corresponds to the maximum time expected for the initial attack by the wildfire suppression services [66].

Wildfire Risk
Risk levels were obtained by multiplying the exposure and vulnerability classes (Figure 3). The final classification of risk followed a five-category scheme and considers that the very high class includes at least one of the components (exposure or vulnerability) in the highest class and the other component in either the highest or the second highest level. An equivalent approach was applied to define the very low class, whereas the intermediate and increasing values were progressively attributed to the other three classes. This risk classification ranks the existing settlements according to the probability that a wildfire may occur and affect them negatively.   Table 1. The 12 management scenarios are described in Section 2.4.

Landscape-Level Fuel Management Scenarios
We analyzed how different landscape-scale fuel management scenarios can change the level of exposure of each village to wildfires and, consequently, of risk. We expect that reductions in fuel loads and the creation of spatial discontinuities in the landscape will decrease the probability of a wildfire affecting a settlement, even when these interventions are defined with the general purpose of decreasing the incidence of large fires in the landscape, and not specifically to protect villages. Two types of scenarios were designed: i) an increase in the level of forest management activities (Mngt) carried out by landowners that would result in lower fuel loads; ii) a fuel break network (FBN) with different priorities and extent in the civil parish.
i) Forest management activities: Overall, 38% of the forest land of Alvares is intensively managed, which includes frequent fuel management (every five years), the use of genetically improved material and fertilization. About 50% of the land in the civil parish has a basic forest management, mainly limited to the tree-cutting process after the rotation period, whereas 10% of the forest land is not managed at all [49]. We considered two different levels of forest management, namely a moderate (~20%) and high (~30%) increase in managed forest area in the parish, besides the 38% already in place. The increase in forest management changes the distribution of fuels, decreasing their loads.
ii) Fuel break network (FBN): the implementation of a fuel break network is defined in the municipal plans of forest defense against wildfires, established by law since 2006 (Decree-Law 124/2006, of 28 June), and follows the technical guidelines of the National Forest Services. The network was designed with the specific function of protecting people, assets, and forested areas against wildfires, considering the topographic and hydrographic conditions, the fire history of Alvares and neighbouring parishes, and the exposure of villages [53]. The creation of a FBN implies the creation of fuel discontinuities in the landscape, reducing fuel density in selected areas. In Alvares, the  Table 1. The 12 management scenarios are described in Section 2.4.

Landscape-Level Fuel Management Scenarios
We analyzed how different landscape-scale fuel management scenarios can change the level of exposure of each village to wildfires and, consequently, of risk. We expect that reductions in fuel loads and the creation of spatial discontinuities in the landscape will decrease the probability of a wildfire affecting a settlement, even when these interventions are defined with the general purpose of decreasing the incidence of large fires in the landscape, and not specifically to protect villages. Two types of scenarios were designed: (i) an increase in the level of forest management activities (Mngt) carried out by landowners that would result in lower fuel loads; (ii) a fuel break network (FBN) with different priorities and extent in the civil parish.
(i) Forest management activities: Overall, 38% of the forest land of Alvares is intensively managed, which includes frequent fuel management (every five years), the use of genetically improved material and fertilization. About 50% of the land in the civil parish has a basic forest management, mainly limited to the tree-cutting process after the rotation period, whereas 10% of the forest land is not managed at all [49]. We considered two different levels of forest management, namely a moderate (~20%) and high (~30%) increase in managed forest area in the parish, besides the 38% already in place. The increase in forest management changes the distribution of fuels, decreasing their loads. wildfires, considering the topographic and hydrographic conditions, the fire history of Alvares and neighbouring parishes, and the exposure of villages [53]. The creation of a FBN implies the creation of fuel discontinuities in the landscape, reducing fuel density in selected areas. In Alvares, the fuel breaks should be located in elevated areas, in the main ridges, to support also firefighting activities, and should be at least 125 m large. We considered three different levels of fuel break network implementation: first priority (1/3 of total extent), second priority (2/3 of total extent), and the entire network ( Figure 4). Each level corresponds roughly to 450 ha of land. The first priority fuel breaks were designed with the purpose to decrease the area traveled by large fires, the second priority to protect roads, infrastructures and buildings, and the third priority to isolate potential fire ignition spots. In the FBN areas, specific land uses are promoted, to ensure low fuel loads: agricultural fields, pastures, shrubs managed every 3 to 5 years, possibly with broadleaved trees planted at least 3 m apart. These fuel break network options influence the distribution of fuel types and loads in the landscape, which is reflected in the burn probability simulations. It was assumed that, in the first year, the FBN areas do not have burnable fuels, progressively increasing the fuel load to low density/low height shrubs, up to the fifth year after their implementation, when they are managed again and the cycle restarts [49].

Exposure
In the BAU scenario, the values of burn probability for the settlements ranged between 12% and 34%. The settlements located in the northeast sector of the parish had a higher burn probability, mostly due to historical ignition patterns and meteorological conditions [53]. The probability that a large fire could reach the perimeter of the villages was above 20% for half of them, whereas 17% of the settlements showed a burn probability above 25%.
When combined with the exposed elements, the three largest settlements show very high exposure, since these are the places where more people and buildings can be affected ( Figure 5A). On the contrary, the settlements located in the southwest sector of the parish predominantly belong to Additionally, we considered a Business-as-Usual (BAU) scenario, equivalent to the pre-2017 settings, without any increase in forest management and/or implementation of fuel breaks. All these types of fuel management options were combined, resulting in a total of 12 scenarios (Table 3). Table 3. Scenarios tested regarding current conditions (Sc0), level of forest management activities (Sc1-Sc2), fuel break network implementation (Sc3-Sc5), and possible combinations between fuel break and forest management (Sc6-Sc11). We first calculated the level of exposure of each settlement for the BAU scenario (Table 3, Sc0). To evaluate the effects of the different management options, we estimated the differences in the proportion of settlements in the exposure classes between each potential scenario ( Table 3, Sc1 to Sc11) and the BAU scenario (Table 3, Sc0). We anticipated that the scenarios with higher level of forest management and more extensive implementation of the fuel break network would reduce the number of settlements in the higher exposure classes, since burn probability would likely decrease. The effect of the different scenarios in risk classes was also assessed, as changes in exposure levels can influence the risk classification, although vulnerability levels remain always the same.

Exposure
In the BAU scenario, the values of burn probability for the settlements ranged between 12% and 34%. The settlements located in the northeast sector of the parish had a higher burn probability, mostly due to historical ignition patterns and meteorological conditions [53]. The probability that a large fire could reach the perimeter of the villages was above 20% for half of them, whereas 17% of the settlements showed a burn probability above 25%.
When combined with the exposed elements, the three largest settlements show very high exposure, since these are the places where more people and buildings can be affected ( Figure 5A). On the contrary, the settlements located in the southwest sector of the parish predominantly belong to the lowest classes of exposure, a pattern resulting from the combination of low density of exposed elements with a low burn probability. Overall, 41% of the settlements are classified with low and very low exposure, 28% of villages are in the high and very high classes and the intermediate class includes 31% of the villages in the parish.
are four settlements (11%) where the evacuation time surpasses 1 h and eight others (22%) where people would take at least 30 min to reach a shelter. In both these situations, no potential shelter was identified amongst the existing buildings and people would have to walk to the nearest village where one exists.
The integration of the clusters analysis and the coping capacity shows a predominance of the high and very high vulnerability classes, with 44% of the settlements in total, whereas the two lowest vulnerability classes include 39% of the settlements. The spatial patterns differ from exposure; overall, the settlements with higher vulnerability are not those with higher exposure, as seen in the three largest villages ( Figure 5B).

Vulnerability and Coping Capacity
The highest vulnerability class was attributed to the cluster of villages with the highest number of variables with a negative influence (increasing vulnerability) and the lowest number of positive ones. This corresponds to the settlements that have the highest mean proportion of elderly, women and illiterate people, with no people with university education and the highest proportion of vacant housing. Conversely, the lowest vulnerability class corresponds to the cluster of villages with the lowest mean proportion of elderly and the highest mean proportion of young people, although still rather low, as well as the highest proportion of active population and of people with secondary education ( Figure 6).
Regarding the coping capacity, 25% of the settlements are more than eight km away from the nearest fire station, and these are located in the north sector of the parish (Figure 7). Evacuation times are short, below 10 min, for the settlements where a potential shelter exists, which were identified in half of the villages (Figure 7). Settlement 8 is an exception, since its buildings spread longitudinally along a road and some are farther away from the shelter identified, although still below 20 min. There are four settlements (11%) where the evacuation time surpasses 1 h and eight others (22%) where people would take at least 30 min to reach a shelter. In both these situations, no potential shelter was identified amongst the existing buildings and people would have to walk to the nearest village where one exists.        Figure 7. Variables regarding response capacity per settlement of the civil parish of Alvares: distance to the nearest fire station, in km (A) and evacuation time, in minutes (B).
The integration of the clusters analysis and the coping capacity shows a predominance of the high and very high vulnerability classes, with 44% of the settlements in total, whereas the two lowest vulnerability classes include 39% of the settlements. The spatial patterns differ from exposure; overall, the settlements with higher vulnerability are not those with higher exposure, as seen in the three largest villages ( Figure 5B).

Wildfire Risk
We found that 11% of the settlements had a very high wildfire risk, adding to the 25% in the second higher risk class. The Low and Very low classes gathered 39% of the villages, most of them located in the southwest area of the parish ( Figure 5C). The three largest villages have either an intermediate or high risk, mainly due to exposure levels, rather than vulnerability, which is low or very low. Five of the settlements (14%) have a high or very high level regarding all three aspects, i.e., exposure, vulnerability, and risk, (Figure 5, villages 7 , 15, 20, 21, 22), indicating problems in all three dimensions.

Effects of Fuel Management Scenarios in Exposure and Risk Levels
The implementation of any of the fuel management scenarios would decrease the hazard and exposure levels. Considering only villages with high and very high exposure, where mitigation measures should be prioritized, moderately increasing the managed forest area ( Figure 8A, Sc1) would reduce the proportion of exposed settlements by 32% when compared with the BAU scenario ( Figure 8A, Sc0). The same effect would be obtained with the implementation of the top priority segments of the FBN without increasing forest management ( Figure 8A, Sc3). If the FBN is implemented by 2/3 ( Figure 8A, Sc3) or completely ( Figure 8A, Sc5), the reduction in exposure is higher, progressively decreasing the highly exposed villages to 14%, which corresponds to a reduction of 50% when compared with the BAU scenario. If the fuel break implementation is combined with an increase in forest management, the decline of exposure is even higher, particularly when the full implementation of the FBN is done jointly with a high increase in forest management ( Figure 8A, Sc11). In this case, the villages with high and very high exposure would decrease to 11%, a reduction of 61% in relation to the current conditions ( Figure 8A, Sc0). located in the southwest area of the parish ( Figure 5C). The three largest villages have either an intermediate or high risk, mainly due to exposure levels, rather than vulnerability, which is low or very low. Five of the settlements (14%) have a high or very high level regarding all three aspects, i.e., exposure, vulnerability, and risk, (Figure 5, villages 7 , 15, 20, 21, 22), indicating problems in all three dimensions.

Effects of Fuel Management Scenarios in Exposure and Risk Levels
The implementation of any of the fuel management scenarios would decrease the hazard and exposure levels. Considering only villages with high and very high exposure, where mitigation measures should be prioritized, moderately increasing the managed forest area ( Figure 8A, Sc1) would reduce the proportion of exposed settlements by 32% when compared with the BAU scenario ( Figure 8A, Sc0). The same effect would be obtained with the implementation of the top priority segments of the FBN without increasing forest management ( Figure 8A, Sc3). If the FBN is implemented by 2/3 ( Figure 8A, Sc3) or completely ( Figure 8A, Sc5), the reduction in exposure is higher, progressively decreasing the highly exposed villages to 14%, which corresponds to a reduction of 50% when compared with the BAU scenario. If the fuel break implementation is combined with an increase in forest management, the decline of exposure is even higher, particularly when the full implementation of the FBN is done jointly with a high increase in forest management ( Figure 8A, Sc11). In this case, the villages with high and very high exposure would decrease to 11%, a reduction of 61% in relation to the current conditions ( Figure 8A, Sc0).
The impact of the different forest management scenarios in wildfire risk is less pronounced. The increase in forest management to any other level and without FBN implementation ( Figure 8B, Sc2 and Sc3), can reduce villages at risk by 8%. If FBN is implemented without a change in forest management, the maximum reduction expected is of 14%, decreasing the proportion of settlements with high and very high risk to 31% when FBN is fully implemented (Figure 8B, Sc5). When FBN full implementation is combined with high forest management ( Figure 8B, Sc11), the proportion of villages at high risk is reduced by 22% in relation to the current conditions.
The classification of the settlements in relation to exposure and risk in each scenario is presented as a colored heatmap in Figure 9. The risk classes are more homogeneous across the several scenarios, since they depend on exposure as much as on vulnerability and this component remains unchanged in all scenarios. Settlements 7 and 22 belong to the high or very high classes in all scenarios, whereas settlements 25, 35, and 36 are the least exposed and vulnerable across all landscape-level fuel management options.  A and B, the x-axis represents the three levels of forest management, and the coloured lines represent the fuel break network implementation with different priorities. A-Exposure, represents the combination of burn probability (hazard) with exposed elements; B-Risk, represents the combination of exposure and vulnerability (with coping capacity).
The impact of the different forest management scenarios in wildfire risk is less pronounced. The increase in forest management to any other level and without FBN implementation ( Figure 8B, Sc2 and Sc3), can reduce villages at risk by 8%. If FBN is implemented without a change in forest management, the maximum reduction expected is of 14%, decreasing the proportion of settlements with high and very high risk to 31% when FBN is fully implemented (Figure 8B, Sc5). When FBN full implementation is combined with high forest management ( Figure 8B, Sc11), the proportion of villages at high risk is reduced by 22% in relation to the current conditions.
The classification of the settlements in relation to exposure and risk in each scenario is presented as a colored heatmap in Figure 9. The risk classes are more homogeneous across the several scenarios, since they depend on exposure as much as on vulnerability and this component remains unchanged in all scenarios. Settlements 7 and 22 belong to the high or very high classes in all scenarios, whereas settlements 25, 35, and 36 are the least exposed and vulnerable across all landscape-level fuel management options. Exposure, represents the combination of burn probability (hazard) with exposed elements; B -Risk, represents the combination of exposure and vulnerability (with coping capacity).

Figure 9.
Heatmap with coloured classes for the risk components (exposure; V-vulnerability; risk) in each settlement, considering the different scenarios (Sc0-business-as-usual, to Sc11-high forest management combined with the full implementation of a fuel break network. See Table 3 for details).

Risk Assessment at the Settlement Scale
In this research, we applied a wildfire risk assessment procedure to a civil parish of central Portugal, focused on the safety of people and assets assembled in 36 small settlements. Our results provide a ranked classification of these small villages, with over a third having high or very high risk. This assessment resulted from the combination of different components, which represent specific dimensions of wildfire occurrence: hazard, exposure, and vulnerability. The stepwise procedure is based on prior studies [67][68][69][70] although it was carried out at a finer spatial scale. Fire simulations were used to estimate the probability that large wildfires, over 1000 ha, could reach the perimeter of each settlement. Simulation approaches have been applied to estimate wildfire exposure of particular assets, such as cultural heritage sites, residential areas, energy infrastructures or natural habitats, specifically in Sardinia, Italy [71]; in Navarra, Spain [41]; and in Chalkidiki, Greece [72]. Similar methods have also been applied to estimate communities exposure in the USA [10,16]. In Alvares parish, large fires are expected to occur mainly in the north and east sectors of the parish and the settlements located in those areas, particularly those with higher density of people and buildings, have high or very high exposure levels, which was found for 28% of the villages.
The settlements with higher exposure do not necessarily have the highest vulnerability. For most villages, the aggregation of these two separate components tended to smooth the level of risk, with one dimension counteracting the other. Regarding vulnerability, nearly half of the settlements had high and very high vulnerability, derived from the sociodemographic conditions of resident Figure 9. Heatmap with coloured classes for the risk components (exposure; V-vulnerability; risk) in each settlement, considering the different scenarios (Sc0-business-as-usual, to Sc11-high forest management combined with the full implementation of a fuel break network. See Table 3 for details).

Risk Assessment at the Settlement Scale
In this research, we applied a wildfire risk assessment procedure to a civil parish of central Portugal, focused on the safety of people and assets assembled in 36 small settlements. Our results provide a ranked classification of these small villages, with over a third having high or very high risk. This assessment resulted from the combination of different components, which represent specific dimensions of wildfire occurrence: hazard, exposure, and vulnerability. The stepwise procedure is based on prior studies [67][68][69][70] although it was carried out at a finer spatial scale. Fire simulations were used to estimate the probability that large wildfires, over 1000 ha, could reach the perimeter of each settlement. Simulation approaches have been applied to estimate wildfire exposure of particular assets, such as cultural heritage sites, residential areas, energy infrastructures or natural habitats, specifically in Sardinia, Italy [71]; in Navarra, Spain [41]; and in Chalkidiki, Greece [72]. Similar methods have also been applied to estimate communities exposure in the USA [10,16]. In Alvares parish, large fires are expected to occur mainly in the north and east sectors of the parish and the settlements located in those areas, particularly those with higher density of people and buildings, have high or very high exposure levels, which was found for 28% of the villages.
The settlements with higher exposure do not necessarily have the highest vulnerability. For most villages, the aggregation of these two separate components tended to smooth the level of risk, with one dimension counteracting the other. Regarding vulnerability, nearly half of the settlements had high and very high vulnerability, derived from the sociodemographic conditions of resident population. Women, who predominate in the parish, are vulnerable elements due to their social role [9,73], a trend particularly relevant in older generations; women's household and family responsibilities kept them away from paid occupations, reducing their financial capability and their access to other opportunities, such as higher education, better jobs, or even a driving license. Overall, gender differences are smoothed in younger generations, but in the aging population of this parish, and across the inner areas of northern and central regions of Portugal, this reality still persists. Age is also a factor of vulnerability, as both the very young and the elderly lack the capability to react to an emergency, have low autonomy, and few resources to implement protective measures [9,18,37,74]. Education levels are reflected in the ability to understand and comply with prevention and safety regulations, in the level of risk perception, and in the access to qualified jobs that could improve their financial capacity [9,30,73]. People who work in the primary sector may be directly and severely impacted by wildfires, losing resources and financial security, and the unemployed lack the financial resources to maintain proper living conditions and to implement protection measures [9,35].
Buildings age and construction materials influence the resistance of the structures to fires [75] and reflect the living conditions of residents. Older buildings can have structural weaknesses and use materials less resistant to fire, as is the case of buildings made of stone masonry and adobe with a wooden structure. In Alvares, buildings constructed before 1980 predominate, when stricter construction rules regarding fire safety were not yet implemented. In addition, nearly one-fifth of the existing buildings are vacant and maintenance can thus be affected, because absent owners are less willing to implement protection and mitigation measures [76].
The use of the settlement as the spatial unit in risk analysis is a novel approach in Portugal, which allowed identifying the villages with higher exposure, vulnerability, and risk in a straightforward manner. Other assets besides human settlements can be included in the analysis, and the procedure can be replicated in other parishes, being particularly useful to local authorities, for the timely planning of safety protocols and mitigation measures adjusted to the circumstances of each settlement. Burn probability simulations can also be adjusted to incorporate weather parameters obtained from climate change scenarios, to obtain future estimations.

Effects of Fuel Management Options in Exposure and Risk Levels of Settlements
Wildfire exposure can be directly modified by the implementation of fuel management measures at the landscape-scale. Our findings show that, in all the scenarios, the proportion of villages with high and very high exposure would decrease with respect to the business-as-usual scenario, although at different levels. If the fuel break network were fully implemented and combined with widespread forest management, the reduction in highly exposed villages could reach 60%. These effects are somewhat expected, considering that any of the options tested would change the fuel distribution in the landscape, and therefore also the probability that a wildfire would reach a given settlement. Previous studies have shown that landscape interventions based on fuel management and different treatment intensities could reduce the average exposure of human assets to wildfires [77,78]. These fuel treatments can have positive effects in reducing communities exposure even when implemented distant from the settlements [79], except in severe wildfires under extreme weather conditions [80][81][82], when clearing the immediate surroundings of built-up structures would likely be more efficient. Fuel treatments can reduce wildfire exposure of human assets even when they are implemented with different purposes [83].
In the case in Alvares, fuel break implementation and forest management scenarios were designed with the general purpose of reducing fuel loads to create spatial discontinuities at the landscape-level. These were expected to reduce fire spread, create fire suppression opportunities and prevent the occurrence of future large fires, and still all of the scenarios would have positive effects in the protection of villages. The choice of the fuel management strategy best suitable to manage large wildfires and reduce impacts to human communities, will also depend on social, legal, economic and land ownership constraints [84,85]. The reduced size of forest properties, which in Alvares is on average 0.5 ha [49], the lack of cadastre records in the country and the high absenteeism of landowners, who migrated to other regions or received the land as legal inheritance and are, therefore, detached from the community, are some of the challenges people have to face in these low-density regions [50]. A balance must be reached between multiple, sometimes competing objectives, since conservation or economic purposes may conflict with fuel treatments options required to reduce fire hazard [86,87]. Our findings indicate that the most efficient approaches will likely result from the combined application of different measures, and these could be further adjusted to the characteristics of the different settlements. Further work is needed to assess if the effectiveness and efficiency of fuel management options, in decreasing wildfire exposure and risk, would increase if they were designed with the specific purpose of protecting the human settlements.

Improving Preparedness and Defining Safety Interventions Within Settlements
The vulnerable circumstances of residents cannot be changed with fuel treatments nor with forest management strategies, but protective measures can be implemented to increase people's coping capacity and the resistance of built-up structures. Wherever old people or residents with low mobility options prevail, a safe shelter on-site is a reasonable possibility [20,24,25,88]. According to the criteria defined, half the villages within the parish did not have a building that might be transformed into a shelter, but a new structure could be built specifically for that purpose, to ensure each human settlement has a safe zone within its boundaries and to reduce evacuation times. This is one of the options advocated in the programme, "Safe villages, safe people", and the decision is made by the local authorities and civil protection services. In the case of Alvares, leaving the village is hindered by the time required to reach other settlements, which can take more than one hour on foot, and most residents lack other mobility means. However, different evacuation options should be investigated, considering other possible paths and transport means. Knowing the time needed to reach a safe spot by different alternatives, while taking into consideration the level of risk of each village, can help defining priorities for emergency intervention, as well as establishing safety and evacuation protocols adapted to each village. When the parish was struck by the wildfire in mid-June 2017, the firefighters and local authorities evacuated some of the villages and gathered people in the Alvares settlement, but since no compulsory evacuation system is in place in Portugal, the efficiency of this strategy depends on the willingness of the residents and on the capacity of the authorities. In severe wildfire conditions, which are expected to increase due to climate change, institutional and suppression capacities will likely be overridden [3,[89][90][91]. For these reasons, fire management approaches must integrate strategies to improve community preparedness and people's coping capacity, tailored to their needs and abilities, to enable a suitable adaptation to fire-prone environments.

Conclusions
The recent large wildfires that affected Portugal in 2017, which caused the highest losses ever recorded in the country, have fostered multidisciplinary initiatives with the purpose of increasing the resilience of affected areas to large wildfires. In this research, we developed one such initiative in a civil parish of central Portugal, focused on the existing human settlements. Risk assessment was applied at the settlement scale and allowed to identify those villages that are either more exposed, vulnerable or at-risk of large wildfires, and that that will require priority interventions. The results of the different components of the risk assessment can be provided as mapping tools that could help the decision-making of fire managers regarding the timely application of prevention, mitigation, and emergency actions. In face of the expected worsening of wildfire risk due to climate change, both the self-defence and the institutional capacities need to be enhanced.
Our results have shown that risk levels depend on different combinations of exposure and vulnerability, and therefore actions to reduce risk should be tailored case to case. In addition, wildfire risk management cannot be disconnected from regional development and land-use planning policies in low-density regions, where Alvares is included. The sharp demographic decline and the aging of population, together with the reduction of farming activities and the subsequent land abandonment, make it imperative to find a suitable territorial vocation, perhaps different from before, but capable of creating more economic value and promoting the management of the land in a more effective and sustainable way, thus also reducing the risk of wildfires.