The IMA is located in north-eastern Romania, in proximity to the Republic of Moldova (Fig. 1), and includes 18 communes (ATU3) surrounding the municipality of Iași. To have a more unitary image from the point of view of floods and landslides, we decided to add another five ATU3 areas (Costuleni, Golăiești, Horlești, Tigănași, and Voinești) to the 18 communes of the metropolitan area (Fig. 1). As part of the Moldavian Plateau, the study area is a monoclinic hilly region, with altitudes ranging from 30 to 400 m a.s.l. (Niculiță et al., 2018), developed in a Miocene mudstone–marlstone lithology, with sand, sandstone, and limestone intercalations, which favoured a dense distribution of landslides (Mărgărint and Niculiță, 2017; Niculiță et al., 2019, Bălteanu et al., 2020). According to the Köppen–Geiger classification of the world climate (Kottek et al., 2006), the analysed area is characteristic of the dry continental climate (Minea, 2013; Mărgărint and Niculiță, 2017). At the Iași meteorological station (102 m a.s.l.), the mean annual temperature and the mean annual precipitation were 9.6 ∘C and 559.7 mm, respectively, for the period 1950 to 2006 (Croitoru and Minea, 2015). The Iași metropolitan area is particularly vulnerable not only to anthropogenic hazards (Dicu and Stângă, 2013) but also to natural ones, as a direct consequence of dramatic changes in population dynamic and sprawl of built-up area in the settlements surrounding the city of Iași in recent decades. After the period of socio-political adjustments following the events of 1989, with ambiguous legislation, economic stagnation, and a lack of territorial planning, after 2000, Iași again became one of the main poles of urban and economic growth in Romania (Benedek and Cristea, 2014). In recent decades, there was a noticeable tendency for built-up areas to sprawl along main roads, regardless of the low suitability of the land for construction (Stoleriu, 2008). The old agricultural activities were gradually replaced with new construction and industrial and storage spaces. Residential areas appeared more and more on land with erosive risk, without coherent territorial development plans, and in neighbourhoods with inadequate infrastructure: an undersized utility network and an unmodernized road network that constantly generates traffic problems. Traditional occupations of the inhabitants (agriculture, vineyards, orchards, vegetable farming, and livestock) were gradually moving further and further away from the central urban pole, thus creating a permanent readjustment of the land cover and labour force (Cîmpianu and Corodescu, 2013).

A new peri-urban area is developing spontaneously around the city of Iași, which is growing rapidly but chaotically, generating severe problems related to environmental quality and the future possibilities of landscape planning (Stoleriu, 2008). These complex changes in the recent past have created a greater degree of vulnerability of the population to natural hazards. A synthesis (Rotaru and Răileanu, 2009) of the damages caused in the 2000–2005 period by rain, hail, strong winds, and landslides in Iași County (NUTS 3 level, which includes IMA; nomenclature des unités territoriales statistiques, nomenclature of territorial units for statistics) revealed losses estimated at RON 37 million (around EUR 11.5 million at the time). Also, earthquakes are a constant threat to the life of people and to dwellings: in terms of the total affected dwelling stock, Iași County was affected most by the 7.1 MW subcrustal earthquake in 1977 (Georgescu and Pomonis, 2008), and it remains one of the most vulnerable to seismic hazard in Romania (Bunea and Atanasiu, 2014; Dutu et al., 2018)

To differentiate the risk perceptions of interviewees based on geographic location in the major landforms of the study area, the communes in which the present study was carried out have been split into two categories: (i) floodplain communes, located mainly on the major floodplains in the area (the Prut, Jijia, and Bahlui floodplains), and (ii) hilly communes, with a large development of hillslopes and associated geomorphological processes: landslides and soil erosion (Fig. 1).

Natural hazards considered in our study are droughts, rainstorms, heavy snowfall, floods, landslides, soil erosion, and earthquakes.

Droughts in north-eastern Romania are associated with anticyclone conditions in summer and autumn, characterized by high temperature and low precipitation. The most frequent periods with drought appear in August, while the lengthiest appear in October, and the shortest are in June (Mihăilă, 2006; Pelin, 2015). The impact of droughts on rural communities is high in north-eastern Romania, and it can affect a wide range of activities (agriculture, forestry, livestock, water supply, and industry). The quality of public health is also affected by drought and is considered one of the main factors of rural poverty (Chiriac et al., 2005). Considering the intensity and multi-annual variability of droughts on the Moldavian Plateau, Cismaru et al. (2000) found that for the 1981–1998 period the percentage losses of crops have a logarithmic relationship with drought intensity at the end of the vegetation period (usually October). In some parts of the Moldavian Plateau these losses reached up to 41 %–50 % in the mentioned period for corn crops and 40 %–43 % for sugar beet or alfalfa. The historical trends (over the last 50 years) of droughts in north-eastern Romania are of increasing frequency but decreasing magnitude (Minea and Croitoru, 2015, 2017; Minea et al., 2016; Spinoni et al., 2015), while the forecast has a slight increase (Stagge et al., 2015).

The majority of annual precipitation comes from rainstorms, which are highest during the summer and frequent in late spring and at the beginning of autumn (Mihăilă, 2006). In Iași, the frequency of rainstorms is up to 40 per year, the maximum 24 h precipitation was 136.7 mm in June 1985, when in 3 d Iași received 193.8 mm of rainfall, and the monthly cumulated value almost reached 300 mm (Mihăilă, 2006; Niculiță, 2020). In the proximity of Iași, toward the contact with the Central Moldavian Plateau, the 24 h maximum value was even higher: at Sinești (30 km east-south-east) 185.3 mm fell in 12 h; at Mogoșești (15 km south-east) 154.4 mm fell; and at Bârnova (10 km south) 167.9 mm fell (Minea, 2013). Hail is a common phenomenon, associated not only with rainstorms, with an aleatory distribution in space and time, but also with important events in 1950 and 1984, which produced important damage to agriculture (Mihăilă, 2006).

The mean yearly number of snowfall days is 45 in Iași, but the annual variation is between 16 and 70 (Mihăilă, 2006). Heavy snowfall can negatively affect agriculture and society when they happen very late, in April or even May, or when the intensity is extreme during winter (Mihăilă, 2006). Blizzards usually manifest from December to February (in February being the most frequent), but early (November) or late (April) events can appear (Mihăilă, 2006; Niacșu et al., 2019). In Iași, there is a mean of 9 d/yr, but the variation is between 0 and 22 d/yr. During blizzards the wind has a mean speed of 50–75 km/h, with a predominant direction from the north-west and north, the maximum speed registered being 200 km/h in 1966 (Mihăilă, 2006).

Floods are widespread on the Prut River, where the two remarkable ones occurred in 2008 and 2010 when thousands of hectares were covered by water and many settlements were threatened and partially evacuated (Romanescu et al., 2011a, b; Romanescu, 2015). Much earlier, a 1991 event produced significant damages in the Jijia River's floodplain (Romanescu et al., 2017). In the Bahlui catchment, the hydro-technical infrastructure has diminished the frequency and severity of floods (Minea, 2013), which had critical negative impacts on the population of the city of Iași before 1960 (Tufescu, 1935). The effect of major floods in the last century on settlements in north-eastern Romania was recently depicted using detailed topographic maps: dozens of villages were partially or entirely displaced in the Moldavian Plateau (Văculișteanu et al., 2019) in the last 100 years. In north-eastern Romania, climate change is expected to increase precipitation extremes in both wet and dry regions as has happened in the past (Donat et al., 2016, 2017; Ingram, 2016; Jacob et al., 2014; Kurnik et al., 2017). It is predicted that the flood magnitude will increase (Alfieri et al., 2015; Reker et al., 2017), so probably the number of flood-related deaths in Romania will continue to be one of the biggest in Europe (Vanneuville et al., 2017).

Landslides and soil erosion are common natural hazards in the study area. In the recent decades, landslides have been slow-movement reactivations that generated household displacement and infrastructure destruction (Niculiță et al., 2017, 2018). One of the most destructive recent events near our study area was reactivation of the Pârcovaci landslide in December 1996, triggered by heavy rains and snowmelt: 97 houses were destroyed or heavily damaged, affecting up to 400 inhabitants (Cioacă and Dinu, 2002; Rotaru and Răileanu, 2009). In a recent study, Niculiță et al. (2018) have identified and mapped a total number of 518 landslides that happened in the last century in the Iași metropolitan area. They are usually reactivations of old landslides and present an obvious temporal pattern in a strong relationship with precipitation variability. Their low magnitude and the fact that almost all the identified landslides happened outside populated areas show that landslides could be perceived as not so dangerous by the inhabitants. But the situation could change in the future, considering the continuing expansion of the built-up area (Cîmpianu and Corodescu, 2013; Iațu and Eva, 2016) and future changes in climate (Niculiță, 2020). Soil erosion is favoured by fragmented topography, the increased tendency for extreme meteorological events, and the land use. These characteristics make our study area one of the most critical hotspots of soil erosion in Romania (Prăvălie et al., 2020).

The city of Iași is located about 200 km from the Vrancea region, one of Europe's seismic hotspots. Since 1800, seven earthquakes with moment magnitudes (MW) above 7 were registered there, while four major events marked the last 120 years, measuring 7.1 MW (1908 and 1986), 7.4 MW (1977), and 7.7 MW (1940) (Lungu et al., 2007; Mărmureanu et al., 2011). The last strong earthquake (4 March 1977, 7.4 MW, with 109 km depth of the hypocentre) was the cause of much socio-economic damage in Romania (exceeding USD 2 billion at that time), claiming the lives of 1578 people and injuring another 11 300 persons. At a national scale, the impact was huge: 32 897 collapsed or demolished dwellings, 34 582 homeless families, 763 industrial units affected, and many other damages in all sectors of the economy (Georgescu and Pomonis, 2008). Although located relatively far from the epicentral zone, Iași County was the most affected in Romania in terms of percentage of dwelling stock affected: 47 % were affected, including 11 % destroyed, 13 % requiring strengthening, and 23 % requiring repair (Georgescu and Pomonis, 2008). Since then, minor damages were reported from earthquakes of over 6 MW in 1986, 1990, and 2004.

The statistical analyses was performed in R stat (R Core Team, 2020).

There is an ongoing debate if Likert data is fit to be transformed to an interval scale by considering that the distance between ordinal-scale elements is the same (Cliff, 1996). Some argue that Likert scale data typically do not meet the assumptions of the parametric tests (Baker et al., 1966; Stevens, 1968; Gaito, 1980; Knapp, 1990; Jamieson, 2004; Gardner and Martin, 2007; Mangiafico, 2016; Kero and Lee, 2016). Others argue (Armstrong, 1981; Kanpp, 1990; Pell, 2005; Norman, 2010) and prove with study cases (Carifio and Perla, 2007, 2008; de Winter and Dodou, 2010; Mircioiu and Atkinson, 2017) that while conceptually parametric statistics is not appropriate, in practice the differences are not important, and in this regard using parametric statistics brings their robustness and sensitivity into the analysis.

While this issue as to what methods, parametric or non-parametric, are better for Likert scale data is still disputed, we have chosen to comply with the standard statistical assumptions, especially regarding the failure of parametric statistics in the case of extreme values of ordinal data and unequal interval scales (Baker et al., 1966; Armstrong, 1981).

The statistical analysis was performed in three main steps (Openheim, 2000): (i) univariate analysis, (ii) bivariate analysis, and (iii) multivariate analysis.

The univariate analysis was performed by plotting the relative frequencies on the Likert scales to provide descriptive statistics for a first overview of the data. This approach is straightforward in identifying the overall perception of stakeholders toward a particular risk or factor and in ranking it by the majority of data (the likert R stat package plots the Top 2 Box score percentages, which is another measure used for Likert scale data).

Further, the bivariate analysis consisted of applying various measures of association and independence between the variables to the cross-tabulations. First of all, we tested the association of perceptions with risks/factors and stakeholder characteristics (stakeholder type, village, commune, flood vs. hilly, gender, and education) in two-way tables.

We used the asymptotic generalized Pearson chi-squared test (chisq_test() function) from the R stat coin package (Agresti, 2002; Hothorn, 2008) to test the association of the observations of two variables in a contingency table, one ordinal and the other categorical (two-way cross-tabulation with the ordinal variable in the column). The null hypothesis is that the variables are not associated with each other, so they are independent. If the null hypothesis is rejected, then the variables have a certain degree of association: they are not independent. Vice versa, the presence of dependence means that the perception of the stakeholders about a certain risk/factor is different from the other risk/factors or that the perception of the stakeholder is influenced by its appurtenance to certain groups/typologies. This test can be applied to categorical and ordinal data, but the ordering is not considered, and the strength of association is not available.

The Kruskal–Wallis rank sum test is more powerful because it uses the mean of the rank to assess if there are differences in the responses of different groups (Agresti, 2002; Mangiafico, 2016), not requiring further assumptions about the distribution of the data. However, the test is fit for small samples in which there are not normal distributions. The null hypothesis states that the groups represent populations stochastically equal (if the shape of the distribution is not considered to be known and is of a similar shape and spread), while the alternative hypothesis is that at least one sample stochastically dominates another sample. Post hoc analysis can pinpoint which groups are different from other groups (Mangiafico, 2016). In the case of our research questions, this test is able to show if the perception of stakeholders is different by risk/factor (RQ1) or if the perception toward a certain risk/factor is significantly different as a function of stakeholder characteristics. The test was performed using the kruskal.test() function from R stat (R Core Team, 2020).

When the difference exists (the null hypothesis is rejected), Freeman's epsilon-squared statistic was used to assess the strength of the difference of one ordinal variable over one nominal variable (Mangiafico, 2016). This statistic ranges from 0 to 1, with 0 indicating no association and 1 indicating perfect association. Values bigger than 0.26 were regarded in our case as a measure of powerful association in the presence of dependence (considering the values proposed by Mangiafico, 2016). This measure was computed using the epsilonSquared() function from the rcompanion R stat package (Mangiafico, 2021). A post hoc analysis was performed in the cases where the Kruskal–Wallis test shows significant differences in the groups to show which groups are different from each other group. The post hoc analysis uses pairwise Mann–Whitney U tests, based on the p value, that allow for the identification of significantly different items (Mangiafico, 2016).

Finally, we applied a multivariate method for those questions and risks that were found conclusive in the bivariate analysis step. CA (correspondence analysis) is a graphical method for exploring the relationships between variables in contingency tables (Greenacre, 2007) by assessing the interaction (Jobson, 1992). The theory behind the method is straightforward, based on the singular-value decomposition of the matrix data structure of the contingency table. We have chosen this method because it describes our data graphically to show the differences between stakeholder types or other categorical variables, especially for those with big Freeman's epsilon-squared values. The Likert scale with the answer to the question is considered the dependent variable, and the variants of the response or the categories of stakeholders or other associated categorical data (flooded or non-flooded communes) are the independent data.

We used mainly ordinal vs. categorical cross-tabulation tables and CA contribution biplots (with the ca R stat package; Nenadic and Greenacre, 2007; Greenacre, 2013), which display the data in a two-dimensional space using the first two extracted principal coordinates (and which should contribute to the majority of the variance) from both rows and columns, in order to get an idea of the association between rows and columns variables of the two dimensions. The plot is asymmetric, with the values of the axes corresponding to the standardized residuals and the points that are contributing very little to the components being located close to the centre of the biplot. The column variables (e.g. stakeholder type) are displayed as oriented vectors, while the Likert scale counts are displayed as dots with size proportional to the count. The orientation of the stakeholder type vector toward one of the axes shows its contribution to the variance of that axis. If the angle between the vector and the lines is 45∘, then the contributions to the two axes are the same, while if the angle is smaller toward a certain axis, the greater the contribution to the variance of that axis is. The length of the arrow vectors is proportional to their contribution to the two-dimensional solution. Since we have an ordered variable and the distances between the categories are not the same, there is no logic in taking into account the distances along the axes of the CA plot and making comparisons (although this type of plot allows that, in the sense that the axes are scaled to a common scale). The points that are close to the centre of the biplot contribute very little to the solution, while those which are too far might be considered outliers.

Usually, the differences between the responses of different stakeholder types are either striking and showing the overall importance of every stakeholder type or non-significant, so we have chosen the CA plots because they easily show the associations graphically. The circles' colour intensity and diameter depend on the relative frequency, while only the arrows' colour intensity is proportional to relative frequency. In this way, low-frequency categories located on the periphery that give a false impression of importance can be identified because they are pulled toward the centre of the biplot (Greenacre, 2013). These can also be seen on other types of plots (Likert plots, bubble plots, mosaic plots, etc.) but often require more attention to be spotted.

The first question addressed to the interviewees was designed to investigate which main socio-economic and environmental factors could affect the communities' quality of life (Fig. 2). The majority of stakeholders (61 %) consider that the level of development is the main factor that can threaten the quality of life in their territory (Fig. 2). It is closely followed by risks induced by natural hazards (57 % of responses), then climate change (40 %), criminality (37 %), and environmental pollution (27 %), with technological risks (8 %) of much less concern.

The level of development and natural risks are similarly perceived as important threats, while criminality, environmental pollution, and climate changes are likewise lower as is shown by the post hoc analysis of pairwise Mann–Whitney U tests (Table 3).

Generally, the stakeholders who participated in the present survey consider droughts as the most threatening natural hazard for their communities and personally (Figs. 3 and 4).

Water scarcity is a direct consequence of the continental climate of the region: it has affected the agricultural economy of north-eastern Romania for centuries (Mărgărint et al., 2021; Niculiță, 2020). Many stakeholders reported a drastic reduction in the number of cattle, which, in the driest years, can reach 80 % of the total animals:

In the past 10 years, I had serious problems every year. I purchased a special car tanker to get water for livestock. And very little remains for vegetable crops. I get water from the reservoir (5 km away), and I do not know what will happen when it disappears. (a farmer, 35 years old, managing 300 ha of agricultural land and 35 cows)

Earthquakes represent the second most threatening hazard. The memory of the 1977 Vrancea earthquake, when Iași County registered the highest number of buildings affected in Romania (Georgescu and Pomonis, 2008), is still vivid in many stakeholders' memory. Although the norms in construction were strongly upgraded after this event, the discipline in building decreased suddenly due to the lack of legislation after 1989. How many dwellings have been built in recent years is not far from the interviewees' knowledge, and, from this point of view, many raised serious questions regarding the resistance of the new construction:

Many who bought new homes think they are new and strong, but at the next big earthquake, they will find that they were built just to be sold. (a mayor, 58 years old, personally affected by the 1977 earthquake)

Regarding differences in perception of the threat to the community vs. that to themselves (Fig. 3), the stakeholders' perception is similar (Top 2 Box score percentages very close), except for snowstorms, rainstorms, and earthquakes (Top 2 Box score percentages different), where the community threat is perceived as higher than the personal threat.

A middle position is occupied by the hazards that registered a higher frequency: rainstorms and snowstorms had an increasing trend in the last decade in the study area. Consequently, their impact on communities is quite essential. During the year, the strongest storms occur in late spring and summer. In some cases accompanied by hail, the most significant damages are recorded in agriculture and in newly built areas with insufficient drainage infrastructure. When these phenomena occur in large areas, they can affect transport, trigger soil erosion, and generate high flows along fluvial channels, leading to the destruction of the bridges, the erosion, siltation of drainage and fluvial channels, etc. These issues were invoked as the most pressing by farmers, mayors, and police chiefs:

I have been here for few years. In the centre of the locality, there are no problems – there is asphalt on the street – but towards the valley, those who have moved to a house in the last 4 years live a nightmare every time it rains. The road is muddy and becomes impassable. (a police officer, 34 years old, in a settlement with many new dwellings)

Climate-related hazards that have a relatively low temporal frequency, like floods, landslides, and soil erosion, are perceived as imposing a low threat in general. The landslide risk is high in hilly regions of north-eastern Romania (Micu et al., 2017; Mărgărint and Niculiță, 2017; Bălteanu et al., 2020). In the last century, one of the most significant events took place 50 years ago in a succession of years with high precipitation (Pujină, 2008). With few exceptions, the memory of those events seems to have been erased. But the risk is still high, and people will again face landslide reactivations in years with an increased pattern of precipitation (Niculiță, 2020). There is a lack of prevention behaviour in terms of recent expansions of built areas due to several factors: investors' desire to build and sell and lack of knowledge and awareness of the danger by those who buy and by those who should make decisions regarding the expansion of built-up areas.

In our commune, the landslide risk has been solved: we have the study regarding landslide hazard and risk in an updated form, so we are in line with the legislation.” (the mayor of a commune affected by landslides in 1969–1972, 66 years old)

Outputs of the Kruskal–Wallis rank sum test and Freeman's epsilon-squared statistics show correlations between every category of natural risks and a set of socio-economic and geographic variables (for further results, see Tables A7, A8, and A9 of Appendix A). The most significant differences are in stakeholder type (answering RQ2), gender, age, spatial localization, and geomorphological context. The division between education types is far too uneven for any effect to be tested. The results indicate that the risk perception is dependent on stakeholder type, affirming RQ2. Also, it has been found that the age of the respondents is an essential factor regarding certain risks (Table 4) because some of them were born after certain important hazard events such as the 1977 earthquake, 43 years ago, or landslide events such as those between the 1970 and 1980 (Niculiță et al., 2017, 2018). For floods, climatic hazards, and soil erosion, it seems that younger respondents are more aware.

The CA contribution biplot for Q1 from Fig. 5 shows the correspondence between the perceived role of natural hazards as threats to the local community by different stakeholder types, considering the first two dimensions that sum 96.8 % of the variance. The plot shows striking differences in the stakeholder type perception toward natural hazards (which overall are considered as threats to the quality of life – there is a significantly strong association of stakeholders' type perception as is shown in Table A3 in Appendix A for Q1 item d), by their different contributions to variance axis; if no difference would be present, the arrows would point to one main axis and would be very close to the centre. Police chiefs and priests perceive natural hazards as low and medium threats; mayors and farmers perceive them as high threats; and school directors perceive them as very high threats. The explanation of the low perception of hazards as threats to the community's quality of life in the case of priests and police chiefs is due to their relatively low knowledge of natural hazards provided by their professions. School directors, mayors, and farmers have a high level of awareness associated with the threats to the quality of life regarding the following factors: level of development (91 % of school directors), natural risks (82 % of school directors and 81 % of farmers), and climatic change (78 % of farmers and 55 % of school directors). The exception is related to technological risks, given the predominantly rural background of the communities. Priests and police chiefs generally expressed a low level of perception regarding the threats to local communities, with some exceptions, e.g. police chiefs regarding criminality, which is their duty. The same threat is seen by school directors, in association with their high level of childcare.

The highest values of the perceived threat associated with droughts (Fig. 4) have been registered for school directors (95 %) and farmers (93 %) who expressed great concern compared to the other stakeholders. Also, earthquakes are seen as a significant threat by school directors (77 %), farmers (56 %), and priests (52 %). By interpreting the enlarged discussions during the interview, this could be considered as a consequence of still lively memories of the 1977 Vrancea earthquake (Armaș, 2006), a social trauma affecting the Romanian people, but also other factors are present: (i) a high vulnerability characterizes the majority of institutional buildings (especially schools and churches) to earthquakes (Mosoarca and Gioncu, 2013; Albulescu et al., 2020), and (ii) there are frequent exercises aimed at improving earthquake preparedness (in schools these exercises usually take place annually). The problem of the vulnerability of old buildings in Romania represents a constant public and scientific debate (Armaș, 2012; Banica et al., 2017), and, in this sense, we also raise on this occasion an alarm signal regarding the need for essential investments in the modernization of public spaces in urban and rural areas in Romania.

From these general results, significant differences have been recorded among the two geomorphological types of the administrative units (Figs. 1 and 6): floodplain administrative units (FAUs) and hilly administrative units (HAUs).

The results highlight that stakeholders have different levels of perception related to different hazards, according to the main events that have been recorded in recent decades: in the floodplain administrative units (FAUs in Fig. 6), there is a significantly higher degree of awareness concerning flood risk and possible threats, while in the hilly administrative units (HAUs) the level of threat associated with landslides and soil erosion is higher than in the FAUs.

Again, droughts are the most life-changing natural hazards with the highest likelihood of occurrence. Rainstorms, snowstorms, and earthquakes follow them. A lower level of probability was assigned to soil erosion, landslides, and floods (Fig. 6). But here, there are essential differences, depending on the geomorphological type of the locality. The stakeholders who come from floodplain settlements have indicated a higher probability for floods than the others (HAU stakeholders) and a lower probability for landslides and soil erosion. This finding responds affirmatively to RQ3.

The main geomorphological characteristics which can influence different hazardous processes and the distance to the potential risk areas constitute essential factors of how different people perceive different risks (Bickerstaff and Walker, 2001; Heitz et al., 2009; Gao et al., 2020). Some natural hazards affect large areas (droughts, earthquakes, or snowstorms), while others (e.g. landslides and floods) are spatially concentrated in direct relation to topographic characteristics at the local scale. From this point of view, the settlements from the study area, as part of the Moldavian Plateau, have been constantly affected by landslides and floods (Văculișteanu et al., 2019), and their consequences are found in the answers given by the interviewees. Tables 2 and A3 of Appendix A and Fig. 6 show that the geomorphological context of the area where the stakeholder works is important in their perception regarding floods and landslide risk (although the investigation of the Likert plot is much more intuitive than the statistical tests), responding affirmatively to RQ3. These results are seen in the context of the social trauma of inhabitants managed by the stakeholders during the evacuations of some settlements along Prut Valley in 2008 and 2010. Due to the risk of flooding of inhabited areas, in July 2008, over 3000 inhabitants from Iași County, including the communes of Victoria, Ungheni, and Țuțora (Fig. 1), were evacuated (Ziarul de Iași, 2008).

Concerning the likelihood of occurrence of natural hazards (Fig. 7), some types of natural hazards are perceived to increase in the near future, especially climatic-induced hazards: droughts (86 %), rainstorms (68 %), and snowstorms (64 %). Landslides and soil erosion are perceived as not increasing, while for earthquakes, the results are balanced.

Personal experience is one of the most critical factors influencing risk perception (Weber, 2006; van der Linden, 2014; Knuth et al., 2015; Öhman, 2017). The study participants indicated that they were affected mainly by droughts, rainstorms, and snowstorms, with farmers bearing high costs (Fig. 8). A large proportion of them was affected by droughts (93 %), rainstorms (78 %), snowstorms, and soil erosion (48 %). According to their activities and responsibilities, stakeholders are affected by natural hazards in their daily life, exposing them to different vulnerabilities. Also, knowledge about the community's past events makes them aware of the natural hazard threat at the community level but not at a personal level, especially for natural hazards that are not related to certain physiographic conditions (earthquakes, rainstorms, and snowstorms; see Fig. 9).

The other stakeholders were affected in a smaller measure by soil erosion. This process can generally pose problems only to those who directly connect with the land, which affects the built-up areas less. It is shown that experience is higher with age (Table 4), not only especially for the analysis with earthquake occurrence (the mean age is lower for those who reported no damage due to earthquakes; Table 4, Q6 item b) but also for landslides (Table 4, Q6 item c). These are disasters that, for their high magnitude, can be impressed vividly in people's memory. Stakeholders' role in disaster risk management and coordination allows them to remember the most significant events where they served the community. In contrast, memory of slow-onset events (e.g. droughts or soil erosion) can disappear quickly.

The knowledge of participants about natural hazards has been investigated through several sub-questions. Stakeholders get information differently about the probability of occurrence and the severity of these events. The majority get information from the TV/radio (82 %), friends/family and community peers (60 %), and social networks on the internet (53 %). The more official channels are the least represented with national information initiatives (47 %), school (44 %), local administration (41 %), and volunteer associations (40 %). Looking at the triggering factors of those events, stakeholders mentioned all sub-sections from the questionnaire (Table A1 from Appendix A) that they consider to have an important influence on the negative impact of natural hazards. Some exceptions have been registered for 57 % of mayors who responded that uncontrolled urbanization and unmanaged land use planning are not influencing the occurrence of any hazard. Local administration is controlling the land use planning, and, in any case, this might be the cause of negative consequences derived by climate extremes and geological movements.

The majority of priests and mayors do not consider that climate change can exacerbate the negative consequences of natural hazards (56 % and 22 % of them indicated “low” and “very low”, respectively). Among solutions to avoid the negative consequences of natural hazards, results showed a uniform answer among all stakeholders, except that the victims' compensation scheme was noted especially by mayors. Financial-compensation schemes represent a particularly sensitive issue in the post-communist society of Romania. Many interviewees highlighted that these compensations could encourage non-compliance with the law, especially regarding unauthorized construction on land at risk of floods and landslides.

The level of preparedness was investigated individually and regarding the community. Overall, the results indicate a low level of preparedness in the case of all the natural hazards discussed. The lowest ranks were given to soil erosion (64 % unprepared), droughts (58 %), earthquakes and landslides (55 %), floods (52 %), rainstorms (50 %), and snowstorms (35 %). It seems that, despite a low level of readiness, stakeholders feel a bit more prepared to withstand the consequences of storms and floods. Snowstorms affect the communities in winter (and exceptionally in spring, for example in April 2018), and agriculture does not suffer. Life in rural areas can be more comfortable compared with urban areas. In Romania, after recent intense snowstorms such as those from January 2008 (Georgescu et al., 2009) or January–February 2012 (Bălteanu et al., 2013), rural settlements were endowed with specialized equipment in rapid intervention, especially in the case of roads, and these endowments seem to improve the respondents' concerns.

Similarly, the existing embankments along rivers (Prut, Jijia, and Bahlui) have often been invoked during discussions as ensuring a relatively good level of protection, especially of built-up areas. The lower level of preparedness is associated with soil erosion and landslides, for which many stakeholders declared their lack of knowledge concerning the processes themselves and related protective measures. The survey results made us respond affirmatively to RQ2, which states that the level of preparedness depends on the risk type.

The same pattern of the answers was registered in the assessment of communities' preparedness level. However, preparedness was low, and stakeholders affirmed strongly that their preparedness and that of their community could be increased by good training and knowledge of natural hazard occurrence and mitigation practices. Asking how much they think that their personal knowledge might increase the level of preparedness of their community (Q11 from Table A1 from Appendix A) reveals significant differences among stakeholders. Whereas for school directors, “high” and “very high” responses reached 95 %, for police chiefs, the percentage dropped to 14 %. Intermediate values were recorded for other stakeholders: “high” and “very high” answers were given by 67 % of farmers, 56 % of priests, and 39 % of mayors. Police chiefs and mayors are responsible for risk management during an emergency, and preparedness is at the base of their training. But surprisingly they do not see their level of preparedness as increasing that of the community, probably because they are aware of their inability to fully control individual decisions (for example the old people who refuse to leave the property when the flooding is imminent). School directors who have the duty of caring for small children feel that individual preparedness is the key to successful disaster management, evacuation, and recovery. In this regard, participation in simulation evacuations is a crucial step for a positive response to potential disaster. Most of the stakeholders declared that they had participated, especially in the simulations concerning earthquakes, and few of them indicated other specific hazards (e.g. fires). Seventy-two stakeholders (61 %) declared that they participated in simulations in the recent years, most of them in earthquake simulations (especially school directors and mayors). Stakeholders from floodplain communes declared participation in flood simulations. In a particular case (Aroneanu settlement, located close to Iași International Airport), stakeholders participated in a technological disaster exercise (aircraft crash). The period elapsed since the last simulation varies from a few months to over 10 years, the most recent being declared mostly by school directors.

The same differentiated pattern of stakeholder responses was recorded in respect of the level of their communities' preparedness.

Several factors have been listed (Fig. 10) and discussed as representing long-term solutions to improve current risk management plans.

Most of the participants agreed with all the items proposed. On the other hand, priests seemed to be the most pessimistic, especially in terms of predictability, people's preparedness, intervention, and recovery capacity. Again, the role of trust depicts a negative situation in which stakeholders showed low trust of mitigation and management measures (Fig. 11). As mayors followed the same trend, it is plausible to think that they delegate the responsibility during emergencies to other institutions, imputing ineffective planning and organization.

Question 16 (“In your judgement, how much are the opinions of the following actors taken into account when deciding about measures to adopt for preventing or reducing damage from natural hazard phenomena?”) presents a grouping of “high” and “very high” responses around 70 % for the following sub-sections: local communities, technicians/engineers, and elected representatives at local and national levels. A lower percentage (34 % of “high” and “very high” responses) has been registered for the sub-section “environmental organizations.” Among stakeholder types, we should highlight the higher percentages of “low” and “very low” responses in the following cases: priests for “elected representatives at the local level” (16 %) and “technicians/engineers” (16 %), school directors (50 %) and mayors (43 %) for “environmental organizations”, farmers for “local communities” (16 %), and “state elected representatives” (26 %).

The stakeholders' roles as leaders of their institution during events generated by natural hazards is critical. They refer to direct intervention in the affected areas and management and communication with the community's entire population. These issues were addressed in the following question (Q17 from Table A1 from Appendix A: “According to your position in society, how much do you think that your institution could help in the communication/management of people during the events associated with natural hazards?”). The gathered answers are not only generally in line with the level of social responsibility of the institutions that stakeholders represent according to legislation but also with the moral leadership in the community. “High” and “very high” responses were acquired as follows: from priests (88 %), police chiefs (86 %), mayors (74 %), school directors (64 %), and farmers (52 %). There are interesting absences of “low” and “very low” responses in the case of mayors, school directors, and priests, and there is a low proportion of these responses in the case of police chiefs (5 %) and farmers (7 %).