The most common approach used to understand why there are individual differences in risk perception is the psychometric paradigm developed by Fischhoff et al. (1978) and modified by Slovic et al. (1986) and Sjöberg (2003). In contrast to the cultural approach, which is a qualitative understanding of risk perception (Douglas and Wildavsky, 1982), the psychometric approach seeks to quantify people's subjective assessment of risk and risk-related impacts. It argues that people make a quantitative appraisal about the current and likely risk of various hazards and the desired level of regulation of each risk (Lechowska, 2022). Therefore, the psychometric approach, used in this study, is an appropriate way to characterise factors to which risk perception is related.

Wachinger et al. (2013) reviewed the main factors of risk perception, particularly in connection with natural hazards. They highlighted the influence of personal factors related to the demographic, cognitive and psychological characteristics of the individual, as well as contextual factors related to the family, community and society in which they live.

Personal factors are demographic, such as age (Knoll et al., 2017; Useche et al., 2019), gender (Bee, 2016), educational level (Carlino et al., 2008), disaster experience (Bronfman et al., 2020; Paton et al., 2000) or ownership of transport (Chauvin, 2018). In addition, personal factors can be cognitive, such as understanding of the risk processes (Sim et al., 2018) or interest in seeking risk information (Donovan et al., 2018). The perceived availability and predictive power of environmental cues (sights and sounds that are considered to indicate a hazard onset) are also cognitive factors influencing risk perception (Lindell and Perry, 2012; Perry and Lindell, 2008). In addition, personal factors are psychological, including anxiety (Lemée et al., 2019) or trust in authorities (Bronfman et al., 2016; Siegrist et al., 2005).

Contextual factors are economic, such as household income (Barclay et al., 2019, 2015), or family related, like family status or household size (Donovan, 2010; Barclay et al., 2015). Religion or other cultural dimensions are also key contextual factors shaping risk perception (Gaillard and Texier, 2010; Chester et al., 2008).

Goma, the capital city of the North Kivu province, is built in the lava field of the Nyiragongo volcano along the northern shore of Lake Kivu in eastern DRC (Fig. 1). It shares a border with the town of Gisenyi in Rwanda. It is an important humanitarian hub (Büscher et al., 2010) and an economic centre for regional trade (Vlassenroot and Büscher, 2013, 2011). Small business is one of the main sources of income, forcing the population to spread out along the roads by doing odd jobs for day-to-day survival (Syavulisembo et al., 2021; Oldenburg, 2020). Over the past 3 decades, Goma and its surroundings have been affected by several armed conflicts (Pech et al., 2018; Vlassenroot and Büscher, 2011). People from the nearby villages and towns have sought refuge in Goma for safety and comfort, resulting in the growth of the population (Van Praag et al., 2021). Therefore, the city is constantly expanding, but it is bounded (Fig. 1a) to the south by Lake Kivu, to the northwest by the Virunga National Park and to the east by the Rwandan border, forcing the expansion of the urbanised area northwards, up to the foot of the Nyiragongo volcano (Büscher et al., 2010; Pech et al., 2018; Michellier et al., 2020). From 2002 to 2020, the population of the city had doubled from half a million to more than 1 million inhabitants (INS, 2021). Urban growth is associated with an increase in a population's exposure to volcanic hazards, especially to lava flows emitted on the southern flank of the volcano.

Nyiragongo is a stratovolcano in the Virunga volcanic province (Poppe et al., 2013). Its main crater is surrounded by two main adventive cones: Baruta and Shaheru on the northern and southern flanks respectively. The volcanic field of Nyamuragira surrounds that of Nyiragongo, and both undergo permanent CO2 degassing (Smets et al., 2010, 2015a). Since the early 1900s, an active lava lake has characterised almost continuously the activity of Nyiragongo, interrupted by three effusive flank eruptions in 1977, 2002 and 2021 (Barrière et al., 2022). Some of these eruptions were preceded by seismic swarms (Oth et al., 2017; Barrière et al., 2022), and each caused long and fast lava flows (i.e. speed of the order of 6 to 10 to 20 km h-1 in 1977 and less than 10 km h-1 in 2002) (Muhindo Syavulisembo, 2019) that came out from eruptive fissures and headed south towards the city of Goma (Favalli et al., 2009) (Fig. 1).

Two historical eruptions impacted the city before our survey in 2020. On 10 January 1977, the first one poured 20 million m3 of lava flows over 15 km2 (including 4.9 km2 within the Virunga National Park) on the northern, southern and western flanks of Nyiragongo, destroying several villages and roads north of Goma. Tazieff (1977) reported less than 100 deaths. After a relative calm period, Nyiragongo erupted on 17 January 2002 while the city was under rebel occupation (Komorowski et al., 2002). This new flank eruption, which generated lava flows, was larger (25 million m3 over 14 km2) and more destructive than that of 1977 (Wisner, 2017; Wauthier et al., 2012; Smets et al., 2015a). In less than 24 h, Goma was crossed by two lava flows, one of which reached Lake Kivu (Schmid et al., 2002). Komorowski et al. (2002) estimate that 40 people died and that 120 000 people had their homes destroyed. In addition, they note that several infrastructures were lost and evaluate the devastated part of the city at 13 %.

For this risk perception study, data were collected through a questionnaire survey developed on the KoBoToolbox application installed on tablets. All questions related to perception used a five-level Likert scale. The specific questions on risk perception were constructed according to PMT (Mertens et al., 2018), as mentioned in the theoretical background of this study. The following questionnaire sections were used:

Demographic profile of participants. This includes gender, age, family status, religion, household size, household monthly income, education level, prior experience of a volcanic eruption and possession of a means of transport;

The survey was conducted in 7 out of 18 neighbourhoods of the city of Goma and in a part of the urbanised area of the Nyiragongo territory as an eighth neighbourhood (Fig. 1a). These eight representative neighbourhoods were selected based on the contrasted social vulnerability assessed in 2017 by Michellier et al. (2020a) and other criteria such as their existence in 2002 (year of last eruption at the time of survey); their spatial distribution relative to potential hazards and evacuation routes; and the existing contrasts in population density, average income, and level of education. One neighbourhood was selected to represent two or more neighbourhoods having similar characteristics.

A total of 2224 adults from the general population were surveyed (Mafuko Nyandwi, 2023a). The size of sampling was calculated from the following statistical formula (Krejcie and Morgan, 1970): 1n=tp2×P(1-P)×Ntp2×P1-P+(N-1)×y2, where n is the sample size, N is the population of the entire city, P is the population proportion (assumed to be 0.50 since this would provide the maximum sample size), tp2 is the table value of chi square for 1∘ of freedom at a confidence level (3.841) and y is the degree of accuracy expressed as a proportion (0.05).

According to the 2020 report of the National Institute for Statistics (INS) of the North Kivu province, the population of Goma exceeded 1 million inhabitants in 2020 (INS, 2021). With a 50 % variance of population, 3 % margin of error and 99 % confidence level, our survey's sample size should be 1831 individuals. The 2224 inhabitants surveyed is a larger sample than the minimum sample size required to be representative of the population of Goma, even considering the Nyiragongo neighbourhood. We worked with an almost equal number of participants per neighbourhood (almost 280 people per neighbourhood). This sample is also representative for each neighbourhood within a confidence interval ranging between 0.01 to 0.05.

The data were collected between September and October 2020. In every surveyed neighbourhood, around 280 points were randomly distributed and plotted with a defined minimum distance between points using a geographical information system (Fig. 1b) on a 2017 very high resolution orthomosaic picture of Goma (Smets et al., 2018). Data were collected in one of the four households located closest to the point. We undertook the survey with a team of 16 trained enumerators. The interviews were conducted face-to-face, with a questionnaire in French. Each enumerator had a notebook with the translation of the questions into Swahili, the common local language. The interviews were conducted with people aged 18 years or above, living in the selected household. Verbal informed consent was obtained from the survey participants before the survey. A survey day started early in the morning (07:00 local time) and was also conducted during weekends, to meet parents and working adults. Each interview lasted about 35 min.

Descriptive statistics were used for categorical variables, such as demographic and risk perception (Harpe, 2015). A non-parametrical test of Wilcoxon–Mann–Whitney (for two-group variables) or Kruskal–Wallis (for multi-group variables) was used to determine the variation in risk perception according to demographic, contextual, cognitive and psychological variables. Statistically significant variations were represented on boxplots. Pearson (for binomial variables) or Spearman's (for Likert-scale variables or ordinal demographic variables) correlations were used to measure the correlations between potential risk perception factors and the risk perception indicator. To analyse the spatial contrast of the risk perception, a geographic information system was used.

Table 1 describes the demographic profile of the survey participants. There are fewer men than women among the participants, and most of them are parents. The majority lives in large households: half of the households surveyed counts 4 to 7 persons, and 30 % have 8 to 11 persons. Despite the large household size, the average monthly income is very low. More than half of the households live on less than USD 250 per month, and another significant proportion (29 %) live on a monthly income of USD 250–500, thereby limiting access to certain services such as transport. Nevertheless, 34.2 % of the participants have a university degree, and 47.3 % achieved their secondary school completion. The high rate of participants who did not experience the 2002 eruption is an indication of the high migration reported in Goma. Table A1 in the Appendix shows the differences in demographic characteristics of participants between neighbourhoods. In general, households with a very low income live mainly in the Karisimbi municipality and the territory of Nyiragongo. In the Mugunga neighbourhood, one-third of participants are not educated, and this proportion falls to 1.7 % in Katindo or 4 % in the Les Volcans neighbourhood. To summarise, there are strong economic contrasts, but sampled respondents in the different neighbourhoods are homogenous in terms of demographic characteristic (age, gender, household size).

When asked to rate their perception of a range of threats, the population does not mention natural hazards as the main source of danger (Fig. 3) but rank it among its top five threats, after physical insecurity, at the same level as personal economic insecurity and above other environmental or health threats.

When evaluating perceived severity, there is no major variation in the levels of the perceived likelihood of different hazards (Fig. 4a), as well as in the perceived severity of their impacts on the city (Fig. 4b). This similar level of perception is surprising, as several of the hazards mentioned had not occurred (i.e. release of gas from Lake Kivu, explosive eruption at the shoreline of Lake Kivu, explosive ash from Nyiragongo) in recent history at the time of the survey and thus nor their potential impacts. Although all the listed hazards are possible scenarios at Nyiragongo, their homogeneous perception is interpreted to reflect a poor understanding of the contrast between these hazard processes, rather than a proper understanding of all eruption scenarios. Regarding perceived vulnerability, most respondents have a high to very high perception of the damaging impacts on infrastructure and functioning of the society. When considering the potential impact on their own life, participants have a lower perception of the risk of loss of life and family disruption than the perception of other impacts (Fig. 4c). When indicators of perceived likelihood of hazards and the perceived severity of impacts on the city are aggregated as the perceived severity, it is higher than the perceived vulnerability (Fig. 4d), suggesting that volcanic hazards are perceived to be more threatening to the city and its functioning than the individuals themselves. In general, the perception of volcanic risk by the population of Goma was high (mean = 3.7) before the May 2021 eruption of Nyiragongo.

Table 2 shows the results of the tests of differences in the means of risk perception according to the different potential risk perception factors. Figure 5 presents the variation of the risk perception indicator according to factors for which the Wilcoxon and Mann–Whitney or Kruskal–Wallis's test highlighted a significant contrast between the factors' categories (Table 2). The level of risk perception varies less with demographic and contextual factors than with cognitive and psychological factors. Indeed, there is a limited variation in risk perception by age group i.e. the older age group having a slightly higher risk perception, family status and prior experience of a volcanic eruption (Fig. 5a, b, c). The results interestingly highlight that respondents from households with a lower income tend to have a higher risk perception than respondents from wealthier households. Moreover, the positive relationship between risk perception and anxiety suggests that the high risk perception among the population of Goma induces fear of the impacts from volcanic hazards. The risk perception is directly proportional to the perception of availability and the predictive power of environmental cues, as well as the comprehension and interest in seeking risk information (Fig. 5d, e, f, g, i). This means, as expected, that feeling exposed to the signs and sounds that indicate an onset eruption leads to a perception of a likely occurrence of a hazard and its impacts.

The spatial differences in risk perception indicators were assessed at two levels: between neighbourhoods and between the western and the eastern parts of the city (Fig. 6). We used a Kruskal–Wallis rank sum test for analysis between neighbourhoods and a Wilcoxon test for contrast between the western and the eastern parts of Goma. Results in Table 5 indicate that there are significant risk perception differences between neighbourhoods due to variations in perceived severity and in perceived vulnerability. In addition, a contrast was observed between the western and the eastern parts of the city. Participants living in the eastern neighbourhoods, affected by the 2002 lava flows, demonstrate a higher level of perceived risk than respondents from the western neighbourhoods. In addition, there are significant differences in both perceived severity and perceived vulnerability between participants from these two areas.

The maps in Fig. 5 illustrate the differences in risk perception indicators per neighbourhood. The lowest levels of perceived vulnerability or severity are observed in the extreme west (Mugunga and Kyeshero), while the highest levels of these two risk perception indicators are observed in the neighbourhoods that were severely impacted in 2002 (Majengo and Virunga) and in Kahembe (the neighbourhood that hosted the Virunga and Majengo disaster victims in 2002). The risk perception as a derivative of the perceived severity and vulnerability follows the same pattern.

According to Chauvin (2018), Barclay et al. (2015) and Haynes et al. (2008), several sociodemographic factors (gender, age, level of education, level of income, etc.) have been shown to influence risk perception. However, in Goma, prior to the May 2021 eruption, only age (Fig. 5a), family status (Fig. 5b) and monthly household income (Fig. 5d) were associated, to a limited extent, with variation of risk perception. Younger people and those who do not belong directly to the close family have a lower perception than older people and close family members (Fig. 5b). The sense of responsibility for the well-being and security of the household seems to be one of the determinants of risk perception in Goma as documented in several other case studies (Gaillard and Dibben, 2007; Gaillard and Mercer, 2013). In addition, a high household income reduces the level of risk perception (Fig. 5d). Indeed, the perceived risk of asset loss or impact on livelihoods is higher compared to the perceived impact on lives (Figs. 3, 4b and c). This can be interpreted by the fact that, although poor households have little to lose, they would experience a relatively large impact of such loss, whereas rich people having many more assets would be relatively less affected by the loss. Blake et al. (2017) argue that people who are labelled as vulnerable, especially the poor, typically find it more challenging to reconstruct their lives after a disaster strikes.

Considering the demographic factors that control risk perception in other volcanic environments around the world, mostly assessed in western countries (Barrett, 2020), family considerations do play a role in Goma. Reviewing sociodemographic factors of risk perception, Chauvin (2018) notes that gender is a determining demographic factor in controlling of risk perception in several cases, with women having a higher level of perception than men. However, in Goma, it is the economic context of the family, the position of the respondent in the household and their age that control the perception of risk. Considering these three parameters, it can be deduced that a parent or a responsible person in the household (usually the eldest of the household) with limited resources is more concerned by the household vulnerability to external hazards, and their risk perception level is higher than other family members'. Wu and Zhong (2022) highlight that people in collectivist cultures, as is the case to some extent in Goma, are better insured and supported by their nuclear and extended family members, as well as friends in their communities. Consequently, collectivist culture acts as a form of implicit mutual insurance to protect people from catastrophic losses, which leads to fewer perceived risks by family members who are not directly responsible for the household or community. Thereby, risk perception is influenced by the household's sense of responsibility and desire for well-being. Risk assessment and development of DRR strategies at the household level should be prioritised over those at the individual level.

The sub-permanent lava lake hosted in the Nyiragongo crater emits a gas plume (Arellano et al., 2017; Michellier et al., 2020), and in some inhabited neighbourhoods, there is localised emission of dry volcanic gas through fractures, called mazuku (Smets et al., 2010). Moreover, in January 2002, before the eruption, strong detonations were heard from the volcano (Komorowski et al., 2002). These are environmental evidence that most of the respondents consider warning signs as a good predictor of an imminent or starting eruption. Indeed, the predictive power of these processes is considered very high for respondents that have a high risk perception. However, Lindell and Perry (2012) warn that the perception of these environmental cues can bias interpretations of a hazard prediction. For the individual, a good knowledge of the mechanisms related to the hazard is required, as well as an understanding of the uncertainty associated with predictions of the natural event.

Our study also highlights a logical link between the level of interest in seeking information related to volcanic phenomena and the level of their understanding. It is however unclear whether the understanding is higher because people actively look for information on the volcano or whether a good understanding of the threat encourages inhabitants to further inform themselves on the volcanic activity. Both elements are associated with a high level of risk perception. Moreover, confidence in the actors involved in DRR does not influence the perception of risk (r = -0.06), but it influences the interest in seeking information (r = 0.22). This means that the population considers that it is possible to find reliable information from those actors. Finally, as advocated by Gaillard and Mercer (2013), increasing knowledge about volcanic phenomena could have a real impact on the level of risk perception.

This study is affected by several limitations, one of which is the demographic characterisation of respondents that did not consider the housing tenure of respondents (ownership vs. rental) and the duration of residence in a specific neighbourhood. A qualitative approach through focus groups and interviews could help to capture local interpretations of the volcanic risk depending on culture. Our survey formulation of perceived vulnerability might have led to misinterpretation between the likelihood or the impact. Thus, multiple phrasing should be tested for the same concept.

Future research on risk perception in Goma should also consider (1) the impact of the population growth by highlighting differences of risk perception according to migration status, (2) the impact of false alarms spread by social media on risk perception, (3) the relationship between perceived vulnerability and scientifically assessed social vulnerability, and (4) the influence of risk experiences in general (vicarious, life difficulties, disaster experience, experience of insecurity related to civil wars or criminality) on volcanic risk perception. As our survey was conducted prior to the 2021 eruption crisis, it would be needed to study how this eruption and the associated evacuation have affected the risk perception of inhabitants. Finally, it would be relevant to further analyse how the highlighted contrasts in risk perception impact a population's preparedness and reaction during a volcanic crisis.