the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 May 2026
Living amidst Mazuku (dry gas vents): linking hazard and impact perceptions to improve risk communication in Goma, Virunga volcanic province
Abstract. Communication of diffuse carbon dioxide (CO₂) degassing hazards remains less explored than more visible volcanic hazards such as eruptions, despite their persistent and sometimes fatal impacts on exposed populations. This study examines how residents living amidst mazuku (CO₂-rich dry gas vents) in western Goma, Virunga volcanic province, perceive both the hazard and its impacts, and how these perceptions are interrelated to inform risk communication. A structured household survey was conducted among 573 household heads to assess perceived proximity, likelihood, and magnitude of mazuku across different environments of daily life, as well as experienced impacts, their perceived likelihood, and expected severity. Measured distance from each surveyed household to the nearest mapped mazuku vent was also calculated using geographic information system tools. Descriptive analyses, Spearman correlations, linear regression, and multinomial logistic regression were used to explore relationships between perception dimensions, socio-economic factors, and actual spatial exposure.
The results reveal clear spatial differentiation in hazard perception. Mazuku is often perceived as spatially close in domestic environments but less likely and less intense, whereas public and less more intense. Levels of experienced impacts are strongly associated with the perceived likelihood controllable spaces, such as streets, markets, and sanitation areas, are perceived as more likely and of impacts, while perceived severity remains high even for rarely experienced events. Hazard perception is positively related to the perceived likelihood of impacts but shows weaker associations with perceived severity. Measured distance also significantly influences hazard perception within domestic environments.
These findings highlight that mazuku risk perception is shaped by collective experience, local environmental knowledge, and the differentiation of everyday spaces. They support the need for risk communication strategies that better connect local understandings with scientific information, particularly by making invisible and dynamic CO₂ hazards more tangible. In the context of rapid urban expansion in Goma and the limited availability of safe land for settlement, relocation strategies are often unrealistic. As a result, in situ mitigation approaches appear more effective than displacement-based strategies. Strengthening continuous monitoring, regularly updating hazard maps, and integrating community-based knowledge into communication approaches are therefore essential to improve risk awareness and support context-sensitive risk reduction in mazuku-prone urban environments.
- Preprint
(3481 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- AC1: 'Comment on egusphere-2026-1776', Blaise Mafuko Nyandwi, 12 May 2026
-
RC1: 'Comment on egusphere-2026-1776', Anonymous Referee #1, 28 May 2026
The submitted article focuses on a very interesting topic that deserves further attention and that relates with the permanent hazard associated with the mazuku. This social study aims to evaluate the perceptions of the population to improve the risk mitigation strategies, highlighting the communication of risk as an important strategy to reduce the risk. The study is worth of publication, and I make some considerations that I believe can improve the document, especially in what concerns the methodological approach and some clarifications on the options.
Some comments/suggestions:
Line 65 – It is the first time the reference Smets et al. (2010) appears, and I wonder why it appears as “b” and not as “a”. I suggest authors to check this along the manuscript, since the same happens with other references,
In what concerns the Introduction (or in the Characterization of the area), and for the readers that never heard about mazuku, it could be welcome to mention the dimensions of the mazuku (general information), as well as how do they look like (different vegetation, confined areas,…). The “mazuku” inside home (for example, line 387) should also be mentioned, as this appears afterwards in the manuscript, as evaluation of the person’s perception, but it is not so clear if it relates with a mazuku in the home’s garden, or to buildings that are directly built over these diffuse degassing areas. Some additional information about this would be welcome.
Line 86 – The sentence is repeated.
In the point 2.1, which is included in the Methodology, I think it should be or a separate section as “Characterization of the study area”, or a subsection of the Introduction, since it is not Methodology.
Lines 156 and 157 – Do you have any statistics about the number of fatalities per year? Previous studies reported some statistics, but I guess that unfortunately they are not updated. Any additional numbers would be welcome.
Lines 162 to 165 – Check the format of the text, and in this case the reference Mafuko-Nywandi (2026) misses in the list, but I guess is the reference that appears as 2026a. So, in this case, the “a” is not needed in the references list.
On the Data collection, it is not very clear who answers on the case of the public buildings, since if I well understood they were also sampled.
On the Data analysis section, several statistical approaches are mentioned (examples in lines 196, 207, 216, just to mention some) but no references added. Authors should add references to support this statistical methodologies, and should mention what was the software or statistical packages used to perform the data treatment.
Considering the demographic profile, authors sampled 573 buildings. It would be interesting to have the number of the total individuals involved (considering the inhabitants of each space, which were accounted, if I well understood).
Lines 335 to 340 – This sentence seems contradictory with what is observed in the Table. Can you please check if this is what you meant?
Table 2 – I have a major concern with some variables on Table 2: “type of resident” and “gender”. What is the positive and negative effect. How these variables were categorized to quantify with a positive vs. negative effect? What authors mean with “gender also shows a negative effect…”. What is the positive and negative gender?
Still in what concerns Table 2, I wonder why other parameters such as the age of the household members, or their academic degree/type of work, were not considered as Variables to include in the Table? Another question about the sampling is the fact that the study is dominated by a higher percentage of female. Is this explained by the fact that more female stay at home, and for this reason, were the ones that were able to respond to the questionnaires? Some sentence about this should be added, as well as eventually mentioned on the limitations of the study.
Lines 821 and 824 – Check the format of the CO2. The “2” should appear as subscript.
Citation: https://doi.org/10.5194/egusphere-2026-1776-RC1 -
AC2: 'Reply on RC1', Blaise Mafuko Nyandwi, 12 Aug 2026
We sincerely thank the referee for the careful and constructive review of our manuscript and for recognising the relevance of the study. The comments and suggestions were highly valuable and helped us improve the manuscript’s clarity, methodological description, interpretation of the results and overall presentation. We have carefully addressed all the points raised and revised the manuscript accordingly. Our detailed responses are provided below, and all corresponding changes have been incorporated into the revised manuscript.
Comment – References and citation format (lines 65 and 162–165)
Response: Thank you for identifying these citation inconsistencies. Several references were incorrectly labelled in the preprint. We have systematically checked the in-text citations and reference list and corrected the use of the suffixes “a” and “b”, including the references to Smets et al. (2010) and Mafuko-Nyandwi (2026).Comment – Description of mazuku and location of the study-area section
Response: We agree that the former Section 2.1 does not form part of the methodological approach. We have therefore moved it and made it a separate section entitled “Study-area characterisation”, immediately following the Introduction. We have also added two paragraphs describing the formation, dimensions and physical appearance of mazuku, as well as clarifying what is meant by mazuku occurring inside homes:“Locally, the term mazuku refers both to the gas-emission area and to the cold, CO₂-rich gas itself. Mazuku typically form where CO₂ from a deep magmatic source migrates towards the surface through fractures in highly permeable lava flows and, because it is denser than air, accumulates in low-lying depressions or confined spaces (Wauthier et al., 2018). In the Virunga Volcanic Province, they are particularly common near Goma, between Lake Kivu and the western parts of the Nyiragongo and Nyamulagira lava fields (Smets et al., 2010b; Wauthier et al., 2018). Comparable phenomena occur in other volcanic regions, including Mammoth Mountain in the United States, Royat in France and the Siena Graben in Italy, although their gas sources and patterns of human exposure differ (Hansell and Oppenheimer, 2004). Despite their long-standing recognition, their formation mechanisms remain incompletely understood and debated (Williams-Jones and Rymer, 2015).
Mazuku vary considerably in appearance, ranging from circular or elliptical hollows to fissure-like structures, with surface areas of approximately 5–4,500 m² (Smets et al., 2010b). They may be identified by papyrus, unusually vigorous light-green grass or khaki-coloured moss surrounding bare patches where CO₂ concentrations are highest (Smets et al., 2010b; Wauthier et al., 2018). In strongly affected areas, vegetation is stunted or absent, exposing weathered black or dark-grey lava. The gas can form a shallow, invisible “pond” extending from a few tens of centimetres to approximately one metre above the ground. It may also accumulate in lava caves, tunnels, cellars and buildings constructed over diffuse emission zones (Smets et al., 2010b). Therefore, “mazuku inside the home” refers to CO₂ accumulating within buildings or enclosed domestic spaces located over or near diffuse gas-emission zones, rather than only to visible depressions in gardens or yards. As Goma has expanded into lakeshore areas where mazuku are concentrated (Pech et al., 2018; Pech and Lakes, 2017), human and livestock exposure has increased (Mafuko-Nyandwi, 2026b). Although official mitigation measures include mapping gas-emission zones and installing warning signs, human fatalities and livestock asphyxiation continue to occur; however, these incidents are not systematically documented in continuous official records (Mafuko-Nyandwi, 2026b).”
Comment – Repeated sentence (line 86)
Response: Thank you for identifying this repetition. The duplicated sentence has been removed.Comment – Annual fatality statistics (lines 156–157)
Response: We agree that updated annual fatality statistics would strengthen the study context. However, apart from the figures reported in the studies already cited, we could not identify reliable, consolidated official records of annual mazuku-related fatalities. We have added a sentence to the study-area section acknowledging this important data limitation.Comment – Sampling of public buildings and number of individuals involved
Response: Thank you for highlighting the need for clarification. Public buildings were not included as sampling units. The survey covered 573 households, with one adult household representative interviewed in each household, giving a total of 573 respondents. References to nearby public buildings, such as schools and hospitals, described their presence within the surveyed neighbourhoods and did not indicate that these buildings or their occupants were independently sampled. We have clarified the sampling unit and the number of respondents in the data-collection section.Comment – Statistical references, software and packages
Response: We have revised the data-analysis section to include appropriate references supporting the statistical methods and to specify the software and packages used for data processing, statistical modelling and visualisation. We have also clarified the coding of categorical variables, the initial variables considered and the procedure used to obtain the final models. We plan to add these paragraphs:“Finally, the influence of actual spatial exposure was examined by analysing the relationship between the measured distance from each surveyed household to the nearest mapped mazuku vent and hazard perception within domestic environments. Multinomial logistic regression models were used to estimate the probability of respondents belonging to different perception categories (low, moderate and high) for both perceived likelihood and perceived magnitude of mazuku hazards. Separate models were fitted for different domestic environments—inside the home, the yard or property, and sanitation areas such as toilets or septic tanks—to account for variations in exposure and environmental characteristics within the same household compound.
In addition to measured distance, the initial models included respondents’ sociodemographic characteristics, including sex, residence status, household size, educational attainment and occupation. Binary categorical variables were coded as follows: female = 1 and male = 0; internally displaced person = 1 and established inhabitant = 0. Variables that did not contribute significantly to the models—including household size, educational attainment and occupation—were progressively removed through backward elimination until parsimonious, best-fitting models were obtained. Predicted probabilities from the final models were plotted against measured distance to visualise changes in perceived hazard levels with increasing distance from mazuku emission points. Model fit and statistical significance were evaluated using likelihood-ratio tests, the Akaike Information Criterion and McFadden’s pseudo-R².”
Comment – Interpretation of type of resident and gender in Table 2
Response: Thank you for highlighting this ambiguity. We have clarified the reference categories used to code these variables: female respondents were coded as 1 and male respondents as 0, while internally displaced respondents were coded as 1 and established inhabitants as 0. Consequently, a positive coefficient indicates a higher predicted value for women relative to men or for internally displaced respondents relative to established inhabitants. A negative coefficient indicates a lower predicted value relative to the corresponding reference category. We have revised the table notes to clarify this interpretation.Comment – Apparent contradiction concerning the number of rooms (lines 335–340)
Response: Thank you for highlighting this apparent inconsistency. The coefficients for the number of rooms were reported correctly, but our interpretation of dwelling size as a direct indicator of higher household income was insufficiently supported and potentially misleading, particularly because household income was negatively associated with perceived magnitude. We have therefore revised the interpretation to distinguish the effect of dwelling size from that of household income and to acknowledge its negative association with perceived impact severity:“Holding the other variables constant, the number of rooms was positively associated with perceived magnitude of mazuku (β = 0.245, p < 0.001) and perceived likelihood of mazuku occurrence (β = 0.089, p < 0.01), but negatively associated with the perceived severity of its impacts (β = −0.101, p < 0.05). Respondents living in dwellings with more rooms therefore tended to perceive the occurrence and magnitude of mazuku as greater while anticipating less severe impacts. Although the number of rooms may partly reflect household economic circumstances, these contrasting associations indicate that dwelling size should not be interpreted simply as a proxy for income.”
Comment – Exclusion of other sociodemographic variables and predominance of women
Response: Age, household size, educational attainment and occupation were considered in the initial models. However, variables that did not contribute significantly to model performance were progressively removed through backward elimination to obtain parsimonious final models. We have clarified this variable-selection procedure in the data-analysis section.“Regarding the higher proportion of women, this was not necessarily caused by men being absent for work. Surveys were conducted mainly during weekends, early mornings and evenings, when both men and women were more likely to be at home. We have added the following clarification to the demographic description:
Finally, women were more represented than men (61.8% versus 38.2%). This was not necessarily related to men’s absence for work, as surveys were conducted mainly during weekends, early mornings and evenings, when both men and women were more likely to be at home, but may partly reflect the town’s general demographic profile, in which women outnumber men.”
We also acknowledge in the limitations section that the unequal representation of women and men may have influenced the overall distribution of reported perceptions.
Comment – Formatting of CO₂ (lines 821 and 824)
Response: Thank you for identifying this formatting issue. We have checked the entire manuscript and corrected the formatting of CO₂ so that the numeral 2 consistently appears as a subscript.Citation: https://doi.org/10.5194/egusphere-2026-1776-AC2 -
RC2: 'Reply on AC2', Anonymous Referee #1, 13 Aug 2026
I checked the authors's replies and I agree that improvements were done to reply to the suggestions.
Citation: https://doi.org/10.5194/egusphere-2026-1776-RC2
-
RC2: 'Reply on AC2', Anonymous Referee #1, 13 Aug 2026
-
AC2: 'Reply on RC1', Blaise Mafuko Nyandwi, 12 Aug 2026
-
RC3: 'Comment on egusphere-2026-1776', Anonymous Referee #2, 02 Sep 2026
This manuscript addresses an important but insufficiently studied disaster-risk problem: persistent exposure to mazuku in densely populated, rapidly urbanising neighbourhoods of Goma. Its principal strength is the attempt to connect physical proximity, perceptions of hazard occurrence and magnitude, experienced impacts, and anticipated consequences across different everyday environments. The study has clear potential to contribute to social volcanology and locally grounded disaster risk reduction. The survey size is substantial, the spatially differentiated treatment of domestic and public environments is valuable, and the manuscript appropriately recognises that wholesale relocation is generally unrealistic in Goma.
However, several central interpretations currently extend beyond what the study design and statistical analyses can demonstrate. In particular:
- Distance to a mapped vent is treated too readily as a measure of actual physical exposure.
- Low explanatory power of selected socio-economic variables is interpreted as evidence of homogeneous, collectively constructed perception.
- Perception patterns are interpreted as proof that residents understand CO2 occurrence and diffusion.
- The proposed risk-communication implications remain general and are not yet developed into an actionable framework.
- Important information about sampling, variable construction, model specification, spatial dependence, and survey ethics is missing.
- Acute lethal exposure, recurrent sublethal exposure, and possible chronic effects are not adequately distinguished.
The most important methodological omission is the absence of a clear sampling strategy. The manuscript states that 573 household representatives were surveyed, but does not explain:
- How the study area was delineated.
- How neighbourhoods, streets, or households were selected.
- Whether sampling was random, systematic, stratified, clustered, or convenience-based.
- Whether households were selected according to distance from previously mapped mazuku.
- How many households or residents live in the sampling area.
- How many eligible households declined or could not be contacted.
- Whether interviews were conducted at different times of day.
- How the respondent within each household was selected.
- Whether enumerators avoided substituting more accessible households for unavailable ones.
These details are essential. A large sample does not ensure representativeness if households were recruited through accessibility, local contacts, or proximity to recognised vents.The chi-square goodness-of-fit tests against equal category frequencies in Section 3.1 add little value. There is no theoretical reason to expect equal numbers of respondents in age, income, household-size, gender, or eruption-experience categories. Comparisons should instead use census, municipal, humanitarian, or other defensible population benchmarks where available. If reliable reference data do not exist, the authors should present the distributions descriptively and acknowledge that population representativeness cannot be established.
I suggest to add a dedicated subsection covering sampling frame, recruitment, household and respondent selection, response rate, geographic coverage, and possible selection bias. A map showing surveyed households, mapped mazuku points, neighbourhood boundaries, and the sampling boundary would be particularly valuable. Exact household coordinates should not be published if they create privacy or safety risks.
The manuscript primarily presents mazuku as an acute asphyxiation hazard, but it combines fatal events with nonspecific symptoms that may reflect recurrent sublethal exposure. The distinction is not sufficiently developed. The statement that 10% CO2 is a “lethal threshold” should be refined. Health effects depend on concentration, exposure duration, oxygen displacement, physical activity, individual vulnerability, and the confined or open nature of the space. A single concentration should not be communicated as a universal boundary between safe and lethal conditions.
There is also a conceptual inconsistency in estimating the distance between a household coordinate and a vent while modelling perception “inside the home,” in the yard, and at the toilet. Unless the coordinates of these micro-environments were measured separately, the physical distance predictor is identical for all three. Differences among models therefore reflect differences in reported perception, not separately measured physical exposure at the three locations. I suggest to replace “actual exposure” with language such as “distance-based spatial proxy” throughout. Also, the limitations section should explicitly state that no contemporaneous CO2 concentration, flux, oxygen, meteorological, or topographic measurements were collected.
The statistical workflow requires stronger justification and fuller reporting. The decision to use the mean when the coefficient of variation is below 25% and the median when it is above 25% appears ad hoc. The coefficient of variation is problematic for bounded ordinal scales and depends on arbitrary numeric coding and the location of zero. Ordinary least squares may be acceptable for a well-constructed multi-item scale, but the manuscript needs to demonstrate that the aggregated dependent variables behave approximately continuously and that model assumptions are reasonable. Diagnostics for residuals, heteroscedasticity, influential observations, and collinearity are not reported. The very small R2 values in Table 2 (0.013–0.047) indicate weak predictive models. Statistically significant coefficients in a sample of 573 may still have limited substantive importance. Confidence intervals and predicted contrasts would help readers judge practical relevance.
Essential revisions
1. Fully describe household sampling, recruitment, response rate, geographic coverage, and possible selection biases.
2 .Supply the complete questionnaire and clarify the wording, reference period, coding, and interpretation of all scales.
3. Reanalyse or justify the construction of aggregated indices. Avoid the mean-versus-median rule based on coefficient of variation.
4. Treat the ordinal nature of the data explicitly and justify the choice of OLS and multinomial models.
5. Report reference categories, confidence intervals, diagnostics, category cut-offs, missing-data handling, and multiple-testing control.
6. Account for neighbourhood or spatial clustering.
7. Replace “actual exposure” with “distance-based spatial proxy” and clearly explain its physical limitations.
8. Moderate claims of homogeneous collective perception and proof of local understanding.
9. Clearly separate findings from this survey from findings taken from the companion mitigation study.
10. Distinguish acute, recurrent sublethal, possible chronic, and indirect impacts.
11 Expand the limitations to address symptom attribution, fatality attribution, common-method bias, cross-sectional design, vent-map completeness, and absence of CO2 measurements.
12. Develop a concrete, audience-specific, affordable risk-communication framework.
13. Provide fuller ethical-review and geolocation-protection information.Minor comments:
21 capitol M in mazuku
85 Remove duplicated Fischhoff et al. (1978)
176 Remove duplicated Table1: Table1:
339 Replace “number of rooms is generally associated with higher household income” with an empirical test or more cautious interpretation.
397 Correct the reference to “Fig. 8C” in Section 3.5.1, which appears to mean Fig. 9C.
412 Section 3.5.2 is titled “Perceived magnitude of impacts,” but the model appears to concern perceived mazuku magnitude; revise for consistency.
438 Change Discussions in Discussion
- Use consistent terminology for “vent,” “emission point,” “hotspot,” and “diffuse degassing area.” A point representation may oversimplify a spatially diffuse source.
- Resolve the apparent mismatch between the Zenodo DOI cited in the Data Availability section and the DOI listed in the references.
- Make the questionnaire openly available with the dataset rather than only “upon request,” subject to ethical and privacy constraints.
- Consider making analysis code available to support reproducibility.Citation: https://doi.org/10.5194/egusphere-2026-1776-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 270 | 44 | 32 | 346 | 19 | 25 |
- HTML: 270
- PDF: 44
- XML: 32
- Total: 346
- BibTeX: 19
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Dear readers and reviewers,
We would kindly like to inform you that, due to a formatting error in the submitted manuscript, Figure 6 was inadvertently repeated again as Figure 7. We therefore provide here the correct version of Figure 7 as attached. We sincerely apologize for this oversight and thank you for your understanding.