Due to some predisposing (schist bedrock supplying abundant debris, structural opposite slopes, strong hillslope channel connectivity) and triggering (summer and winter Mediterranean rainstorms) factors, the Upper Guil catchment is particularly prone to hydrogeomorphic hazards such as torrential floods, debris flows, landslides, rockfalls or avalanches (Fort et al., 2002, 2014; Arnaud-Fassetta et al., 2004, 2005, 2014). These hazards frequently impact the local population (fatalities, destruction of buildings and infrastructure, loss of agricultural land, road closures) causing difficulties for local managers, who also have to cope with the legislation and management procedures of the Parc Naturel Régional du Queyras (PNRQ) (Arnaud-Fassetta et al., 2004, 2005). Most catastrophic episodes are related to torrential floods as in 1957, 2000, 2002, 2008 and 2011 (PNRQ, 2016). The two main events described in the literature took place in June 1957 (>100 year RI, EUR 15 million damage) and June 2000 (30 year RI, ≈ EUR 5 million damage) (Arnaud-Fassetta et al., 2004; Tricart, 1958). These catastrophic episodes have severely impacted the mentalities and entailed considerable expenses in terms of risk management and protective structures (dykes, embankments, thresholds, etc.) (Fig. 2). Due to the obsolescence of protective measures and local planner requirements in new studies, it was necessary to assess vulnerability in this area.

Today, the area counts 1770 inhabitants (INSEE, 2012), making it one of France's less densely populated districts (<5 inhabitants by km2). However, during the peak of the touristic season (summer and winter holidays), the resident population can be multiplied by a factor of 10 (INSEE, 2008). Since the second half of the 20th century, Alpine territories have experienced significant changes on their land cover/uses and economic activities (Fuchs et al., 2013). In the Queyras, the progressive decline of agro-pastoralism and the development of skiing tourism activities led to a concentration of human stakeholders in areas that are particularly exposed to several natural hazards (torrential fans and valley bottom). The current land cover/use is the result of a combination of these important changes in human activities together with the impacts of past catastrophic events. Actually, land cover classes are 29 % forest, around 30 % bare rocks and alluvial deposits, 38 % grassland, 3 % agricultural lands and less than 1 % building areas. Apart from houses, major stakeholders are public and administrative services (city hall, schools, hospital, fire station, etc.), industrial and artisanal warehouses and, of course, touristic infrastructure (shops, hotels, museum, ski resorts, etc.). The departmental road (D947) is the most important lifeline ensuring the link with the nearest urban centres (Guillestre, Embrun, Gap). These relatively recent stakeholders are mostly located in areas exposed to natural hazards (Arnaud-Fassetta et al., 2004).

The PDI methodology is indicator oriented. To be used in practice, it is based on the use of commercial databases, aerial imagery and GIS technologies. In the PDI, consequences are expressed in a semi-quantitative way through an index called Total Consequences Index (CTI). CTI is obtained by combining three sub-indices representing the direct and indirect consequences of a hazard on elements at risk (Fig. 3): (1) the PII represents the consequences on people in their physical integrity, (2) the direct SFI expresses the direct and short-term effects on buildings, infrastructure and human activities, and (3) the IFI illustrates the long-term effects on socio-economic activities (Puissant et al., 2013). To obtain these indices and compute the TCI, three steps are required (Puissant et al., 2006). First, the element at risk and its relevant attribute are identified and compiled into a complete database. Then, each modality of the attribute compiled is ranked through an expert weighting (Fig. 3 and Table 1). The value applied is called the Damage Index (di). It is standardised on a scale from zero to one, with higher index values indicating higher potential consequences (Table 1). In the third step, direct (PII and SFI) and indirect (IFI) consequences are modelled using a linear combination. In this step, a coefficient is assigned to each variable with respect to the socio-economic context of the region and the type of consequence assessed (direct or indirect) (Fig. 3 and Table 1). The coefficient, called the local index (Li) varies from 1 to 4. To finish, the three sub-indices are combined to obtain the PDI. In order to be used in a risk analysis, PDI is reclassified in five classes and mapped. With a matrix, PDI map is then combined with a hazard map (reclassified in five classes as well) to obtain a type 1957 flood risk map (Table 2). As support for this work, data sets from Institut National de l'Information Géographique et forestière (IGN BD ORTHO/BD TOPO, 2009) were used. To complete our database, an intensive field investigation was carried out in association with the use of Google Street View^® and OpenStreetMap^® software. Land cover and land uses maps were produced on GIS by combining photo interpretative work with data on natural protected areas (DREAL PACA, 2016), agricultural land (PRG, 2012) and touristic infrastructure (prospectuses, touristic maps, etc.).

In the proposed PCI, PDI methodology has been modified to assess both physical and social consequences. The upgrade consists of the addition of a fourth sub-index in the calculation of the TCI (Fig. 3). This sub-index, called the SCI, is built to represent the social consequences of a hazard on community resilience. The use of an indicator to assess social consequences requires the selection of specific criteria that unequivocally represent the different aspects of social vulnerability (Cutter et al., 2003; Rygel et al., 2006). Literature on vulnerability identifies many elements that contribute to different abilities to cope with hazards (Table 3). Today, the majority of the analyses produced use data from national census to build social vulnerability indices (Cutter et al., 2000, 2008; Wu et al., 2002; Chakraborty et al., 2005; Fekete, 2009; Guillard-Gonçalves et al., 2014, Zhang and You, 2014; Huang et al., 2015; Koks et al., 2015; Nelson et al., 2015; Frigerio et al., 2016; Karagiorgos et al., 2016; Rogelis et al., 2016; Aroca-Jimenez et al., 2017; Heß, 2017). Some indicators repeatedly appear in these analyses such as poverty, age, education or disabilities (Table 3). In agreement with these existing published references, socio-economic data were collected for the six municipalities of the Upper Guil catchment. A set of 21 criteria was first selected (Table 4). Sixteen of them come from the open-access French national statistical database of the Institut National de la Statistique et des Etudes Economiques (INSEE) (INSEE, 2006, 2012, 2015). Five others were selected in a risk-perception survey taken during the SAMCO project. This survey consisted in a questionnaire (38 questions) carried out during autumn 2014 and summer 2015 and 2016 on the six municipalities of the Upper Guil catchment (Fig. 1). It is focused on three main issues: (1) inhabitant perception of the different risks (torrential floods, avalanches, landslides and rockfalls), (2) inhabitant knowledge of preventive and protective measures and (3) inhabitant confidence in stakeholders. A hundred questionnaires were collected (about 5 % of the total population): 8 in Ristolas (10.53 %), 22 in Abriès (6.85 %), 22 in Aiguilles (4.95 %), 16 in Château-Ville-Vieille (4.58 %), 17 in Molines-en-Queyras (5.45 %) and 15 in St-Véran (5.86 %). People were interviewed in-person or by paper questionnaires that were delivered and recovered in person. Special attention was given to a representative view of the socio-economic characteristics of the local population. Indeed, in the second and third campaigns, the surveyed people were selected for their demographic and socio-economic characteristics according to INSEE census data (INSEE, 2012, 2015).

To reduce the number of variables and avoid useless repetition we performed a principal component analysis (PCA) on our data set. We conserved only the criteria containing the highest percentage of information on axis F1 and F2 (Fig. 4). There were six: (1) age, (2) household income, (3) level of education, (4) flood risk perception, (5) level of information on flood risk and (6) confidence in stakeholders (Fig. 4). With respect to PDI methodology, the modalities of the six selected criteria were ranked and a value of 0 to 1 was assigned to them (Table 5). In PCI methodology the term of the consequence index (ci) is preferred to the damage index (id) from PDI. A local index (il) is then assigned to the six criteria with respect to their relative importance in the PCA produced. SCI is calculated using linear combinations on GIS (raster calculator tool on ArcGIS) and applied to each building of the six studied municipalities. Due to the lack of data on a building scale, SCI is equally applied for all the buildings of a same community. The PCI is then calculated by adding the index scores of the four sub-indices (SCI, PII, SFI and IFI) (Fig. 3). PCI is finally reclassified in five classes and mapped. Using a matrix, PCI map is combined with a flood hazard map (in five classes) to obtain a type 1957 flood risk map (Table 2).

Several hazard maps were produced in the SAMCO project. To focus on the new method found to assess physical and social consequences, a single scenario of flooding is considered in this paper. The selected scenario represents a flood type 1957 (RI > 100 years). We voluntarily selected a scenario with the more important spatial extent to highlight the differences between the PDI and the PCI. The type 1957 flood map was made using the hydraulic modelling software HecRAS^®. Fifteen cross sections representing a linear stream of 58.2 km were characterised (Table 6). Due to the lack of accurate data for all the streams of the subcatchment, only eight of them were taken into account in our model (Table 6). Geometry (stream, river banks and flood plains) was extracted from a DEM (digital elevation model) at 1 m resolution. This DEM was produced with lidar data provided by the Regional Natural Park of Queyras (PNRQ). Flooded surfaces (extent, depth, speed) were extrapolated using 371 sections, extracted from our DEM. To take into account the protection along the reaches, dykes and artificial channels were incorporated into the model. The generated flooded surface has an extension of 2.88 km2. This envelope provides a good overview of the water flows and allows a quick and clear visualisation of the potentially flooded areas. The flood map used in this paper was reclassified into five classes considering water elevation (Fig. 5).

The PDI map for flooding is obtained for the Upper Guil catchment by summing the direct PII, SFI and IFI (Fig. 3). CTI scores for buildings ranging between 8.9 and 34.8 (mean: 24.5) (Fig. 7a, b and c). For the sub-indices, the highest scores are generally observed for the PII (mean: 10.9) and the lowest for the Socio-Economic Index (mean: 4.1). SFI scores are the average of them (mean: 9.5). Close-up of Aiguilles, Abriès and Ristolas villages are shown in Fig. 6a, b and c. The produced map displays a majority of buildings with moderate to high scores of total potential consequences for the all studied communities. Buildings with the highest scores are mainly located in the vicinity of the Guil River or one of its main tributaries (Fig. 6a, b and c). Major stakeholders such as rescue centres (hospital, fire station, etc.), town halls and schools are also classified with a high degree of potential consequences. This is due to their important function in local life. Conversely, churches, sheds and warehouses have a low degree of potential consequences. In town centres, buildings with trading or touristic functions are generally in the “high” consequence class, whereas those which only have a housing function are classified as “moderate”. Sparse housing areas (mostly located on the heights) have a high degree of total potential consequences because they were not constructed to resist floods (large opening on ground floor, less resistant building materials, etc.). In most cases, these houses have virtually no chance of being impacted by a flood because they are located away from the torrential streams.

The PCI is obtained by summing the direct PII, the direct SFI, the IFI and the new SCI (Fig. 3). PCI scores calculated for building range from 14.7 to 44 (mean: 31.8) (Fig. 7d, e and f). SCI scores calculated for the six municipalities ranged between 5.2 and 9.2 (mean: 7.2) (Table 7). They are in the same order of magnitude than those of the three other indices used in PCI calculation (PII, SFI and IFI). The PCI map produced for the Upper Guil catchment displays a majority of buildings classified with moderate degree of total potential consequences. (Fig. 6d, e, f). At the community level, buildings classified with high or very high degree of potential consequences are mainly located near the Guil River or one of its main tributaries. Collective housing and major stakeholders (hospital, town halls, schools, etc.) are generally classified with higher potential consequences (Fig. 6d, e, f) than individual housing. In most cases, churches, sheds and warehouses are classified with a low or very low degree of potential consequences. Despite these general tendencies, we observe differences from one community to another. At the Upper Guil catchment level, the studied communities can be divided into three groups (Fig. 6 and Table 7). The first group is made up of communities with a large number of buildings classified with a high and very high degree of total potential consequences: Aiguilles and Saint Véran. A second one is formed by communities with most of their buildings being classified with moderate potential consequences: Château-Ville-Vieille and Molines-en-Queyras. The third group is composed of communities with buildings classified with low to moderate total potential consequences: Abriès and Ristolas. These differences between communities are directly related to SCI scores. The comparison between Ristolas and Aiguilles communities speaks for itself (Fig. 6d, f). The Ristolas community has the lowest SCI score (Table 7). People living here have a good perception of flood-related risks, indicating a high level of preparedness. They have confidence in local managers and there are only a few dependent people (children or elderly people) to care for when an unexpected situation arises. This suggests a high capacity to react when confronted to a catastrophic episode. In addition, they are globally wealthier than the other studied communities. They theoretically have a better ability to quickly recover after a material loss. By contrast, the Aiguilles community has high CTI and SCI scores indicating a lower ability to cope with hazards (Table 7). Compared to other communities, Aiguilles have more dependant people to care for. In addition, people have a lack of information on flood risks and tend to underestimate the danger represented by floods. Aiguilles citizens earn less and have less confidence in their local managers. In the case of Ristolas, CSI tends to reduce the total potential consequences, contrary to Aiguilles. In other words, a community with a resilient population can qualify results obtained for physical consequences.

The PCI is developed as an upgrade of the PDI method. As a consequence, we can observe some similarities between PDI and PCI maps produced (Fig. 6). In most cases, buildings classified with the highest level of potential consequences are buildings considered essential in local life (city hall, hospital, police and fire stations, etc.). In both maps, buildings located in a previously inundated area are also classified with a high degree of total potential consequences (Fig. 6a, b, d, e). Likewise, buildings classified with low or very low potential consequences are generally buildings with no essential function in local life like churches, sheds, warehouses or empty buildings. Moreover, buildings constructed in the last 20 years (mostly individual housing) generally have a higher degree of potential consequences than older buildings. With the PCI method, the influence of the physical consequences indices (PII, SFI and IFI) is thus globally preserved at the community level. The introduction of SCI allows us to qualify the total potential consequences of the elements at risk with regard to the ability of each community to cope with hazards. Ristolas and Abriès have low SCI scores. Floods will have less of an impact on these communities. As a result, elements at risk are classified with lower total potential consequences in comparison with PDI. By contrast Aiguilles and Saint Véran communities have high SCI scores indicating a low ability to cope with hazards. The buildings of these two communities are thus classified with higher total potential consequences in the PCI map and higher potential risk in the risk map produced (Fig. 7). As SCI is equally applied to all the buildings of a community, it tends to homogenise PCI scores at the community level. In comparison with the PDI map, the minimum score values are uplifted, resulting in a partial loss of information. This is particularly true in the communities with the highest SCI scores (Aiguilles and Saint-Véran). This partial loss has, however, a positive impact on the readability of the maps. The global level of potential consequences of each community is evident and allows us to compare each community with one another. This is not so clear with the PDI method. In addition, the smoothing of the results tends to highlight the most vulnerable stakeholders. As a result, the PCI map is easier to understand for local managers than the PDI map.