the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 07 Jul 2026
Natural hazards in La Réunion island (France) since the mid-seventeenth century: production and analysis of a geohistory database (HistoRiskRun)
Abstract. La Réunion is one of the French territories most exposed to natural risks. The island is exposed to seven major natural hazards: cyclones, floods, swells and tsunamis, landslides, earthquakes, wildfires, and volcanic eruptions. Numerous natural disasters have occurred on the island, as well as less significant natural events. This study presents a historical review of the direct impacts of natural events documented in departmental and national archives since human settlement on the island. Thus, in total 2,823 sources, referencing 1,266 events from 1655 to 2025 have been processed. The source's reliability is analyzed using a decision tree method. Based on the impacts, significance scores for each referenced event is established. An analysis of the evolution of impacts and mortality over time is conducted. A detailed geodatabase (HistoRiskRun) is created to capitalize information about natural hazards in La Réunion since its colonization. This database enhances documentation and understanding of past events and the evolution of risk management over time. This work provides elements to discuss the temporal evolution of impacts and mortalities associated with natural hazards and to relate these dynamics to the island’s geo-socio-historical context.
- Preprint
(2672 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3653', Anonymous Referee #1, 20 Sep 2026
-
AC1: 'Reply on RC1', Messie Dupont, 24 Sep 2026
Thank you for your careful reading and for the associated comments. The eight points raised are highly relevant and do indeed require further clarification, which is provided below and will be incorporated into the final manuscript.
1- The total number of events has been the subject of several considerations throughout the study. Indeed, an event considered as a trigger (referred to as “trigger type” in the database) may generate several physical hazards through cascading effects. For example, although a cyclone is considered a trigger type, the floods it generates must also appear in the table dedicated to floods.
The trigger type is therefore specified for each flood event. These events are separated in the database according to the nature of the hazard, the location of the associated impacts, or the information available from the different sources, and each has its own significance index.
The total of 1,266 events therefore corresponds to the inventory of triggering events, without taking cascading effects into account. When these are included, the total number rises to 3,568 events (rather than 3,519), distributed across the different tables according to hazard type.
The Entity Relationship Diagram (ERD) of the database could potentially be added to provide a more visual understanding of the database structure.2- Sources with a reliability score below 60% are retained in the database and included in the analyses. Several points can be clarified in this regard. First, the objective was to preserve, as far as possible, a record of the sources used. Thus, the final reliability score assigned to each source can be directly consulted in the database. Furthermore, the identification of new sources in the future could make it possible to complement the cross-referencing with the sources already identified and thereby increase the reliability score of some of them.
3- We thank the reviewer for highlighting this potential source of bias. We acknowledge that assigning the minimum score to missing information may disproportionately affect older events, for which documentation is generally less complete. We chose this cautious approach because assigning a higher score in the absence of supporting evidence—for example, by assigning a plausible score or an average score relative to the others—could artificially inflate the significance of poorly documented events.
However, we acknowledge that this approach may result in an underestimation of the significance of events for which documentation is incomplete, particularly among the oldest events. This rationale will therefore be clarified in the “Methods” section and explicitly acknowledged as a potential temporal bias and a limitation in the “Discussion” section. We will also refrain from interpreting differences in significance between older and more recent events without taking differences in the completeness of the documentation into account.4- The effect related to source availability does indeed constitute an important limitation of the study presented, as indicated in the discussion in Section 5.1. The final version of the manuscript will place particular emphasis on this limitation, which will also be reiterated in the text, particularly in the analysis presented in Section 5.2.
5- These analyses have not been subjected to formal testing and will therefore be presented as a hypothesis.
6- Regarding the mortality analysis, it is indeed relevant to consider the mortality rate in order to relate the number of deaths to the population present during each period. For this purpose, it is possible to distinguish the period preceding the abolition of slavery (1689–1848), during which the mortality rate averaged 16.08 deaths per 1,000 inhabitants per year. During the subsequent period (1848–1946), this rate decreased markedly, reaching 3.92 deaths per 1,000 inhabitants per year, before reaching 0.8 deaths per 1,000 inhabitants per year during the final period (1946–2025).
7- Indeed, given the level of missing data, this explanation is a hypothesis and will be identified as such in the article.
8- Thank you for this comment. We recognise that this aspect requires further consideration, and this analysis will be carried out using the median, the mean and/or event-level distributions. We are currently reviewing this section of the manuscript and will provide you with a more detailed response once this analysis has been completed. The corresponding changes will be incorporated into the revised manuscript.
Citation: https://doi.org/10.5194/egusphere-2026-3653-AC1 -
AC3: 'Reply on RC1', Messie Dupont, 15 Oct 2026
8. Following
Thank you for this comment. We have generated additional plots using annual means and medians to examine whether the observed patterns depend on the aggregation method. These plots are provided as an attachment. While Figure 9 shows the temporal evolution of the severity, intensity and spatial extension indicators based on annual maxima, the additional plots allow comparison with the patterns obtained using means and medians. Given that annual maxima do not reflect the distribution of events within each year, and that Figure 9 is not essential to the main objectives of the manuscript, we are considering removing it from the article. Alternatively, if retained, its purpose and limitations will be clarified in the revised manuscript.
-
AC1: 'Reply on RC1', Messie Dupont, 24 Sep 2026
-
RC2: 'Comment on egusphere-2026-3653', Anonymous Referee #2, 03 Oct 2026
Dear authors,
Thank you so much for your efforts. I have included my suggestions below.
Best.
The paper shows a substantial archival effort, and a database covering multiple hazards over several centuries would fill an important gap for the island. The mortality data on sailors, enslaved people, and river‑crossing drownings are genuinely interesting. However, the manuscript reads more like a descriptive project or thesis report than a research article. It lacks a research question or hypothesis, and the results are only counts and percentages without any statistical analysis. It would be more suitable as a brief report or review paper.
More detailed major comments;
The core contribution of the paper cannot be evaluated. The database, which the title claims is the main contribution, is under embargo. It has no repository or DOI, yet the conclusion says it has “open source functionality”. The paper does not provide the data model, the 29 source parameters, the decision‑tree questions and weights, or the index components. Consequently, the results cannot be verified or reproduced.
The analysis is not defined, and the main numbers do not match. The abstract and conclusion report 1,266 events. Section 3.3 lists 3,519 recorded events. Section 4.2’s hazard‑specific counts add up to 3,568 events. Floods alone account for 1,612 events, which is more than the total reported. The likely reason, counting cascading events in several tables (p. 7, L166–168), is never explained. Because an event is not clearly defined, all counts, frequencies, and index comparisons are ambiguous. It is also unclear whether the 11.1 % of mentions that fall below the 60 % reliability threshold were excluded, even though the conclusion states that all 2,823 references were validated.
The significance index is biased when comparing events over time. Missing data are given the lowest score. Because documentation drops sharply as we go back in time (Fig. 4), older events are consistently scored lower, so the trends shown in Figs. 5 and 9 partly reflect documentation artifacts. The mortality criterion also saturates at more than 10 deaths (Fig. 3). For example, the 1829 and 1846 cyclones, which caused 260 and 350 deaths, are not among the five most significant events, while the 1913 cyclone, with only 15 deaths, is. The method used to estimate return periods and damage in euros for historical events is not described, and the use of a flood‑specific grid for other hazards has not been validated.
The trends have not been adjusted for source density or exposure. Press coverage accounts for 50.2% of mentions, yet it has not been thoroughly examined since the mid‑20th century, the same period when the decline in impacts is interpreted. Mortality is reported only as absolute deaths per year, even though the population grew from about 1,500 to over 900,000 (L87L71). To interpret any trend, we need rates per capita, uncertainty estimates, and a sensitivity analysis of source coverage. The same requirements apply to the spatial ranking of municipalities.
The conclusions are not supported by the analysis. The authors claim that risk‑management policies, building standards, and administrative status have reduced impacts, but they do not show how these factors relate to the data. Their reasoning is inconsistent. the increase in permanent housing raises the significance in lines 391–393, yet it also lowers impacts in lines 399…. The statement that the results confirm rising economic damage is unsupported by Figure 9, because the figure normalizes the data and hides absolute values. Because the authors do not compare their results with existing catalogues, they cannot prove that the reliability goal has been met.
Minor comments;
Line 42-43: Leroux 2024 is cited, but in the reference list it is 2021.
Line 220: Before 1800 are said to include “telegrams” which is anachronistic please revise.
Line 234-235: It said “wind and lightning exmanined seperately below” but no such analysis appears.
Line 301-303: After 1948…stable is followed by began to decline from 1946… 1848 intended?
Line 342-357: Desarthe 2018, Perotin 1957 are cited but missing in reference list.
Duplicated sections numbers: 3.3 twice, identical titles for 4.3 and 4.4. There are frequent language errors, Thorough English editing is needed.
As I mentioned before, this paper contains important information, but the problems are fundamental and cannot be fixed in a normal revision cycle. I recommend rejecting it. I encourage the authors to submit a new manuscript once the data are ready, perhaps first as a data‑description paper and later as a hypothesis‑driven analysis or review.
Regards.
-
AC2: 'Reply on RC2', Messie Dupont, 12 Oct 2026
We sincerely thank you for your careful assessment of our manuscript and for acknowledging the value of the archival work, the potential contribution of the database, and the scientific interest of the historical mortality records. We appreciate the detailed comments, which have helped us identify several weaknesses in the presentation, methodological transparency, and interpretation of our results.
We acknowledge that some aspects of the manuscript, particularly the formulation of the research questions, the definition of the analytical framework, the consistency of the reported figures, and the treatment of historical documentation biases, require substantial clarification and improvement.
However, we respectfully believe that the manuscript has the potential to make a valuable contribution as a research article. Its purpose is not merely to compile historical events, but to reconstruct and analyse the impacts of natural hazards on the population of Réunion Island from the seventeenth to the twenty-first century. The study combines historical source criticism, the harmonisation of heterogeneous archival records, the classification of natural hazards and their impacts, and an analysis of temporal patterns in mortality and other forms of damage.
We acknowledge that these objectives and their scientific implications were not sufficiently explicit in the original manuscript. We therefore propose a substantial revision to clarify the research questions, improve the reproducibility of the methodology, correct and reconcile the reported figures, address the limitations of the significance index, and strengthen the interpretation of the results. We also recognise that some conclusions in the original manuscript went beyond what could be directly established from the analyses presented.
Our detailed responses to the comments are provided below.
General comment:
We agree that the research question should be stated more explicitly. The overarching question of this study is:
How can the impacts of natural hazards on the population of Réunion Island be reconstructed over the long term, from the seventeenth to the twenty-first century, and what do the resulting records reveal about the temporal distribution of human and material impacts across hazard types and victim categories?
The study therefore has two complementary objectives: (1) to reconstruct a historical record of natural-hazard impacts by critically assessing and cross-referencing heterogeneous archival sources, and (2) to analyse the temporal distribution of recorded impacts, with particular attention to differences between hazard types and victim categories.
The analysis is primarily descriptive and historical. We acknowledge that the original manuscript did not sufficiently distinguish between patterns directly observed in the database and interpretations that would require additional statistical testing. We will revise the manuscript to make this distinction explicit and to avoid presenting descriptive associations as demonstrated causal relationships.
We respectfully consider that a systematic reconstruction of historical impacts, combined with an analysis of their temporal and categorical distributions, can constitute a research contribution. We nevertheless recognise that the scientific contribution must be demonstrated through a clearer analytical framework and a more critical interpretation of the results, rather than through the size of the database alone.
Paragraph 1:
Database availability and embargo
The database was uploaded to the editorial system upon submission of the manuscript and is therefore available to the editor, who can provide it to the reviewers for the purpose of the peer-review process. The dataset is currently under embargo and is not yet publicly accessible. Its public release as an open-access dataset is planned following the revision and publication of the associated article.
We acknowledge that the wording in the original manuscript may have been ambiguous regarding the distinction between the database's current availability for peer review and its planned public release. We hope this clarification addresses the reviewer's concern in the context of the interactive discussion. As the database is intended to be made publicly available upon publication of the revised article, we do not consider it necessary to modify the data availability statement at this stage, provided that the planned public release is implemented as intended.
Database structure and source parameters
We agree that the database structure and the information used to assess the historical sources should be described more explicitly to improve transparency and reproducibility.
The database includes 29 parameters designed to document the provenance, characteristics, historical context, and reliability of the sources. These parameters cover:
Archival provenance and document characteristics: the archive and reference number, source type (primary or secondary), document nature and format, physical condition, language, and title where applicable.
Document production and authorship: the date and historical context of production, authenticity, potential inaccuracies or inconsistencies, author identity and occupation, and the period and place in which the author lived.
Information provenance and perspective: the origin and purpose of the information, its intended audience, the author's perspective and relationship to the event, and the context in which the document was produced.
Event information: event type, year, start and end dates, affected municipalities and localities, and recorded impacts.
Responses and consequences: population responses, actions taken by the authorities, damage assessments, post-disaster developments, assistance, and reconstruction.
Cross-checking: whether information could be corroborated using additional sources.
We will include a more explicit description of these parameters in the revised Methods section. We will also consider adding an entity–relationship diagram (ERD) as a supplementary figure to illustrate the structure of the database and the relationships between sources, triggering events, and physical hazards.
Reliability index and decision trees
The reliability assessment follows the methodology developed by Athimon et al. (2022a, 2022b). We acknowledge that the reference cited as Athimon et al. (2022b) in the original manuscript was incorrect. We have therefore corrected this reference and replaced it with the appropriate publication, which provides a more detailed description of the methodology. This publication explains the evaluation criteria, the closed-ended questions used in the decision tree, the associated weighting procedures, and the calculation of the final reliability score. The methodology was applied without modification in the present study. We therefore refer the reader to Athimon et al. (2022b) for a comprehensive description of the procedure, rather than reproducing the full methodology in the manuscript.
Athimon, E., Gilay, N., Andreevsky, M., Sauzeau, T., Frau, R. : Evaluation of Historical Document Quality: An Interdisciplinary Method, International Journal of Information Research and Management (IJIRM), 9(3), 48–68, 2022b.
Significance index and its components
The significance index used in this study is based on the methodology developed by Boudou (2015). The original methodology is described in detail in this publication, including the decision tree and the components used to assess event significance. We will therefore provide additional explanations in the manuscript only for the components that have been modified or adapted for the present study. The components that are not explicitly discussed will follow the methodology described by Boudou (2015) without modification. This approach avoids unnecessarily duplicating the published methodology while ensuring that the specific adaptations made in the present study are clearly documented.
Paragraph 2:
Definition of an event and reconciliation of the reported totals
We acknowledge that the event totals reported in the original manuscript were inconsistent due to an error in the reported figures. Following verification of the database and the corresponding counts, we confirm that the correct total is 3,568 recorded hazard events, rather than the 3,519 reported in Section 3.3. The total of 1,266 events reported in the abstract and conclusion corresponds to the inventory of triggering events, without counting the associated cascading hazards as additional triggering events. In the database, a triggering event (identified as the “trigger type”) may generate several physical hazards through cascading effects. For example, a cyclone may trigger multiple floods in different locations. These floods are recorded separately in the tables dedicated to floods because their locations, impacts, and supporting documentary evidence may differ. Each recorded hazard event is therefore classified according to its hazard type and may be assigned its own significance index. The distinction between triggering events and recorded hazard events explains the difference between the two totals. We will ensure that this distinction is clearly explained in the revised manuscript and that all figures reported in the abstract, Methods, Results, tables, and conclusions are consistent with the verified database counts.
Sources below the 60% reliability threshold
Sources with a reliability score below 60% have been retained in the database and included in the analyses. This decision was made to preserve the documentary record and to allow future researchers to revisit the assessment as additional sources become available. The reliability score assigned to each source is recorded in the database. Additional sources may enable the information to be cross-checked and the reliability assessment to be updated. We acknowledge that the original manuscript did not clearly distinguish between the validation of a reference as part of the documentary inventory and its achievement of a specified reliability threshold. Consequently, the statement that all 2,823 references were validated could be interpreted as meaning that all of them exceeded the threshold. We will clarify this distinction.
Paragraph 3:
Missing information and historical documentation bias
We acknowledge the reviewer's concern that assigning the minimum score to missing information may systematically underestimate the significance of older events, for which documentary evidence is generally less complete.
We initially adopted this conservative approach because assigning higher scores in the absence of supporting evidence could artificially inflate the significance of poorly documented events. However, we recognise that the approach may produce a temporal bias when the completeness of the historical record varies substantially over time.
We will make this limitation explicit in the Methods and Discussion sections. We will also reconsider the interpretation of temporal differences in significance and avoid attributing observed changes exclusively to variations in the actual impacts of natural hazards.
Mortality criterion and saturation threshold
We acknowledge that the saturation threshold used for the mortality criterion can be discussed, particularly regarding its ability to distinguish between events with substantially different death tolls. The threshold of 10 deaths, together with the maximum damage threshold of €300 million, is based on the criteria used by the French Mission d'inspection spécialisée de l'environnement (1999) to distinguish a very serious accident from a disaster. This threshold was subsequently adopted by Boudou (2015) in the methodology used as the basis for our significance index. The purpose of this classification is to provide a standardised framework for ranking and comparing damaging events according to their impacts, rather than to capture the full extent of the severity of each disaster. Nevertheless, we recognise that the saturation of the mortality criterion limits the ability of the index to differentiate between events causing more than 10 fatalities. Consequently, events with substantially different death tolls may receive the same score for this criterion. Although the five-class scale provides a more detailed assessment than the three-class scale used in our methodology, the possibility of introducing additional classes to better distinguish high-mortality events is a relevant consideration. We acknowledge this limitation and will clarify the rationale for the selected threshold and its implications for interpreting the significance index in the revised manuscript.
Economic damage estimates and historical monetary values
The classification of economic damage into index classes follows the methodology developed by Boudou (2015). The qualitative criteria used to define the different damage classes have therefore been retained from this methodology. The quantitative assessment, however, is specific to Réunion, as it accounts for the currencies used on the island over time and the conversion of reported damage costs into euros at 2025 values, taking monetary inflation into account using the dedicated tool provided by INSEE. In Réunion, the former French franc was used from 1795 to 1945, followed by the Réunion CFA franc until 1975, the French franc until 1998, and the euro thereafter.
For earlier events, particularly when precise damage data were unavailable, a qualitative assessment was considered more appropriate because of the spatiotemporal discontinuity of the available information.
The low damage class (index 1) corresponds to flooding or partial blockage of the road network, brief inundation of buildings or minor landslides, and localised damage. The intermediate class (index 2) corresponds to localised destruction of engineering or coastal defence structures, road or railway infrastructure (between 1878 and 1976), or crops, as well as population evacuations. The highest damage class (index 4) corresponds to destruction affecting a substantial portion of the island, including engineering or coastal defence structures, road and railway networks—resulting in the disruption of transport routes for more than one day—as well as crops, housing and infrastructure. It also includes prolonged inundation lasting several days.
Return periods
The return period is calculated as the final step in the assessment, as it is used to weight the final significance index by assigning greater importance to the most remarkable events. It is calculated by dividing the total time span covered by the data (up to 370 years) by the number of events with the corresponding significance level recorded during that period.
Following the approach proposed by Boudou (2015), which draws on the European Floods Directive to distinguish exceptional flood events, the lowest class (index 1) corresponds to a return period of less than 50 years, the intermediate class (index 2) corresponds to a return period of between 50 and 100 years, and the highest class (index 4) corresponds to a return period exceeding 100 years. This methodology was applied separately to each of the selected hazard types. For each hazard type, the time span considered extends from the first to the last recorded event, and the return-period classification is then weighted according to the significance scores. Only events with significance scores equal to or greater than 14 are considered.
However, this calculation should be understood primarily as a weighting procedure rather than as a statistically robust estimate of the actual return period of natural hazards. By assigning greater weight to events with higher significance scores, it reinforces their contribution to the final results. This approach constitutes a limitation of the methodology, as the resulting index may place disproportionate emphasis on the most remarkable events. The return-period component should therefore be interpreted as a means of weighting event significance within the database, rather than as a direct measure of the probability of occurrence of a given hazard.
Application of a flood-derived framework to other hazards
The significance index used in this study is based on the methodology developed by Boudou (2015), originally designed for the historical analysis of floods at the national scale in France. This framework was adapted to the local and multi-hazard context of Réunion to enable the assessment of the seven major natural hazards using a common evaluation grid.
Several modifications were necessary to account for the island's geographical and administrative characteristics. In particular, the spatial extent criterion was adjusted to reflect the fact that Réunion is a single French department comprising 24 municipalities, rather than a national territory encompassing numerous departments. The original thresholds based on the number of affected departments or municipalities subject to a natural disaster declaration (CatNat) were therefore revised accordingly. The duration of inundation was also replaced by the duration of the event, and the aggravating effects of damage associated with warning-system failures were removed because relevant data were unavailable.
Applying a common framework to all seven hazards provides a consistent basis for comparing their impacts and allows the importance of hazards such as cyclones to be highlighted within a single analytical framework. Furthermore, the adaptation of the index to the geographical and administrative context of Réunion could make it applicable to other French overseas territories.
However, we acknowledge that applying a single index to different hazard types has limitations. Some criteria may not reflect the significance of all hazards equally. For example, the duration of a volcanic eruption or lava flow is not necessarily correlated with the overall significance of the event, although it may contribute to its remarkability. The resulting scores should therefore be interpreted with caution, as the common framework facilitates comparison but cannot fully account for the specific characteristics of each hazard type.
Paragraph 4:
Documentary coverage and media sources
We acknowledge that variations in documentary coverage over time represent an important limitation when interpreting long-term trends in natural hazard impacts. As already discussed in the manuscript, the number and nature of available sources vary considerably across periods. This reflects changes in media coverage, particularly since the mid-twentieth century, as well as differences in archival preservation, the availability of historical records, and the extent to which damage assessments have been conducted.
The manuscript already highlights several limitations affecting the completeness of the historical record, including the absence of a comprehensive press review since the mid-twentieth century, the lack of systematic examination of parish and municipal archives, the incomplete identification of primary sources for some historical events, and the loss or deterioration of archival documents. Access restrictions have also prevented the use of certain sources. These factors may contribute to an apparent increase in recorded events and impacts over time, independently of changes in the actual occurrence or severity of natural hazards.
We agree that the influence of documentary coverage on the observed trends requires further consideration. Although we will not undertake a quantitative sensitivity analysis to measure this effect, we will improve the discussion of source availability by presenting the distribution of sources across the different periods considered. This will help readers assess the temporal variability of the documentary record and better understand the limitations associated with comparisons between periods.
We will also revise the interpretation of the observed trends to distinguish more clearly between the results directly supported by the database and the hypotheses proposed to explain them. In particular, potential explanations for the apparent decrease in impacts, such as improvements in risk management, changes in building standards, and developments in population exposure and vulnerability, will be presented as hypotheses rather than as demonstrated causal relationships. The potential influence of changes in source coverage will also be explicitly acknowledged when interpreting these trends.
The HistoRiskRun database should therefore be regarded as a substantial but non-exhaustive historical inventory. While it provides a valuable basis for documenting and comparing natural hazard impacts in Réunion, its results must be interpreted in light of the uneven availability of historical sources and the resulting uncertainties. These limitations will be made more explicit in the revised manuscript.
Mortality rates and population exposure
Regarding mortality, we have calculated population-adjusted mortality rates to account for demographic changes over the study period. The results show a marked decrease in the annual mortality rate, from an average of 16.08 deaths per 1,000 inhabitants per year during the period 1689–1848, to 3.92 during 1848–1946, and 0.8 during 1946–2025. These results provide a more appropriate basis for comparing mortality across periods than absolute death counts alone, given the substantial population growth in Réunion over the period considered.
However, these results must be interpreted in light of the uneven availability of historical sources, which may affect the completeness of recorded mortality, particularly for earlier periods. Although the calculated rates account for changes in population size, they do not correct for temporal variations in documentary coverage.
We will therefore clarify this limitation in the revised manuscript and distinguish the observed decrease in population-adjusted mortality from the potential explanations for this trend. In particular, the possible contributions of improvements in risk management, building standards, and changes in population exposure and vulnerability have not been formally tested in this study. These factors will consequently be discussed as hypotheses rather than as demonstrated causal relationships.
Spatial distribution of impacts and municipal population
We have also examined the spatial distribution of recorded impacts in relation to population distribution, although this analysis was not included in the original manuscript. The analysis distinguishes three periods: before 1920, 1920–1980, and 1980–2025.
Before 1920, recorded events are concentrated in Saint-Denis, Saint-Paul, and Sainte-Suzanne. Saint-Denis and Saint-Paul were among the most populated areas, while the high number of records for Sainte-Suzanne is partly attributable to the journal of Lescouble, a primary source documenting events in considerable detail between 1812 and 1838, particularly in and around that municipality.
Between 1920 and 1980, recorded events are concentrated in the northern and western parts of the island, with Saint-Denis having the highest number of recorded occurrences. Population data from INSEE indicate that Saint-Denis, Saint-Paul, and Saint-Pierre were among the principal population centres during this period.
Between 1980 and 2025, recorded events are more widely distributed across the island. The relatively high number of records for Sainte-Rose and Saint-Philippe is partly attributable to volcanic eruptions and their effects on road infrastructure.
These patterns suggest that the distribution of recorded impacts reflects not only the occurrence of natural hazards but also population distribution, the concentration of exposed assets, administrative and economic centres, and the geographical distribution of observers and documentary sources. We will consider incorporating this analysis into the revised manuscript, provided that the differences between the limitations of the comparison are made explicit. In particular, we will distinguish the spatial distribution of recorded events from the actual spatial distribution of risk and damage. We will also clarify the historical role of Saint-Paul and Saint-Denis as successive administrative centres, including the transfer of the capital from Saint-Paul to Saint-Denis in 1738, and discuss how this may have influenced the availability of documentary records.
Paragraph 5:
Risk-management policies, building standards, and administrative changes
We acknowledge that the original manuscript did not provide sufficient evidence to establish that changes in risk-management policies, building standards, or administrative status caused a reduction in natural hazard impacts. These factors may nevertheless help interpret long-term changes, as they can have contrasting effects on disaster impacts.
For example, the development of more durable buildings and improvements in infrastructure may increase the resistance of assets to certain hazards and thereby reduce damage. Conversely, urbanisation and the associated increase in soil sealing may exacerbate surface runoff and flood impacts, while changes in population distribution and the concentration of assets may increase exposure to natural hazards. These processes are not necessarily contradictory, as they affect different components of risk, including hazard, exposure, and vulnerability, and their relative importance may vary according to the type of hazard, location, and period considered.
However, the present study does not formally test the relationships between these factors and the observed changes in impacts. We will therefore revise the relevant discussion and conclusions to distinguish findings directly supported by the database from potential explanations that require further investigation. The influence of risk-management policies, building standards, land-use changes, and administrative developments will consequently be presented as a possible interpretation of the observed trends, rather than as a demonstrated causal relationship.
Interpretation of Figure 9 and economic damage trends
We acknowledge that the normalisation used in Figure 9 makes it difficult to assess changes in absolute economic damage. Consequently, the original figure does not, by itself, support the conclusion that economic damage has increased over time.
We will reassess this figure and consider complementary representations of the data, including absolute values where available, medians, means, and event-level distributions. We will also distinguish changes in recorded damage from changes that may result from differences in documentary coverage, population, exposed assets, and monetary valuation methods.
The conclusions concerning increasing economic damage will be retained only to the extent that they are supported by the revised analysis.
Contribution of the HistoRikRun database and comparison with existing catalogues
The main contribution of HistoRikRun is to provide a historical, multi-hazard database of natural events affecting Réunion, covering the period from the island's colonisation to the present day. To our knowledge, it is the only database specifically designed to bring together the island's major natural hazards within a single framework. Its contribution lies in consolidating existing knowledge, critically assessing historical sources, improving existing chronologies, identifying additional events, and documenting their human and material impacts through extensive archival research and a review of the available literature.
The construction of HistoRikRun involved the collection and examination of existing databases produced by different institutions, together with technical reports, scientific publications, and archival sources. Fourteen existing databases were retrieved, six of which provided information on the human or material impacts of natural hazards and were used in the study. Their contribution varied according to the nature, temporal coverage, and level of detail of the available information. Where necessary, existing records were supplemented through additional archival research and cross-referencing with other sources. Conversely, some databases could not be fully exploited because they lacked sufficiently detailed information on impacts. For example, the BRGM's Base de Données des Mouvements de Terrain (BDMVT), which contains 3,028 records for the period 1901–2022, was not systematically incorporated because many entries, particularly minor landslides, did not provide information on specific damage. Given the number of records involved and the objectives of the study, their systematic examination was beyond the scope of this work.
HistoRikRun complements existing hazard-specific catalogues, such as the cyclone chronology established by Desarthe and Moncoulon (2017), which documents tropical cyclones affecting Réunion between 1656 and 2011. While such catalogues provide valuable information on the occurrence and chronology of particular hazards, HistoRikRun adopts a broader approach by bringing together multiple hazard types and documenting their associated impacts, including human mortality and material damage, wherever the available sources permit. Its purpose is therefore not to replace existing catalogues, but to consolidate, enrich, and complement them within a common historical framework.
The mortality inventory illustrates the scientific value of this approach. HistoRikRun documents deaths associated with natural hazards from 1689 to 2025 and provides a basis for more detailed investigations of the circumstances and spatial distribution of these impacts. The database has also contributed to subsequent research on hydrometeorological hazards during the period 1950–2025 (Defossez et al., submitted), which has identified additional information, including more precise locations and circumstances of death.
We acknowledge that incorporating existing catalogues and cross-referencing historical sources do not, in themselves, demonstrate the completeness of the resulting inventory. Nevertheless, the combination of multi-hazard coverage, source assessment, archival research, and the enrichment of existing records constitutes a significant contribution to the historical documentation of natural hazards in Réunion. HistoRikRun provides a substantial, although non-exhaustive, foundation for future research into long-term hazard impacts and historical risk patterns. The limitations associated with incomplete documentation and uneven source availability are acknowledged in the manuscript.
Minor comments
We thank the reviewer for identifying these errors and inconsistencies. We will address them as follows:
Lines 42–43: The correct publication year for Leroux is 2021, not 2024. We will correct the in-text citation and verify its consistency with the reference list.
Line 220: We agree that the reference to “telegrams” before 1800 is anachronistic. We will remove or correct this reference.
Lines 234–235: The statement that wind and lightning are examined separately below refers to the mortality analysis. We will revise the wording to make this connection explicit.
Lines 301–303: We confirm that 1848 is the intended date. We will correct the typographical error and ensure that the chronology is consistent throughout the manuscript.
Lines 342–357: We will add the missing references to Desarthe (2018) and Perotin (1957), after verifying their full bibliographic details.
Section numbering: We will correct the duplicated section numbering, including the duplicated Section 3.3, and ensure that all sections and cross-references are consistent.
Section titles: We will revise the title of Section 4.4 to “Analysis of Mortality Linked to Natural Hazards” and verify the consistency of all section headings.
Language and style: We acknowledge the need for a thorough English-language revision. The revised manuscript will be checked by an English-speaking colleague or professional language editor to improve clarity, grammar, terminology, and consistency.
Final remarks and request for editorial consideration
We thank the reviewer again for the detailed and constructive assessment. We recognise that the original manuscript has significant shortcomings, particularly in the presentation of the research objectives, the definition of the event-counting framework, the explanation and interpretation of the significance index, and the treatment of documentary and population-related biases.
At the same time, we believe that the study provides a potentially valuable contribution through the reconstruction of several centuries of natural-hazard impacts on Réunion Island. The database brings together heterogeneous historical sources, documents the provenance and reliability of these sources, and provides a basis for examining the temporal and spatial distribution of recorded impacts, including mortality across different hazard types and victim categories.
We respectfully ask the editor to consider whether the concerns raised could be addressed through a substantial revision of the manuscript as a research article, rather than requiring its immediate rejection and resubmission as an entirely new contribution. We are prepared to revise the manuscript extensively, including clarifying the research questions, reconciling all event counts, improving the methodological description, reassessing the significance index and its limitations, incorporating the mortality-rate analysis, and revising conclusions that are not adequately supported by the evidence.
We understand that some of these changes may require additional analyses and that the suitability of the manuscript for publication remains the editor's decision. Our intention is not to minimise the reviewer's concerns, but to demonstrate that we take them seriously and are willing to address them systematically.
We hope that the manuscript's contribution to the historical understanding of natural-hazard impacts on Réunion Island can be considered in light of the improvements outlined above.
Regards
Citation: https://doi.org/10.5194/egusphere-2026-3653-AC2
-
AC2: 'Reply on RC2', Messie Dupont, 12 Oct 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 175 | 109 | 38 | 322 | 40 | 27 |
- HTML: 175
- PDF: 109
- XML: 38
- Total: 322
- BibTeX: 40
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This is an interesting and potentially valuable manuscript. The authors have compiled an impressive long-term database of natural hazards and their impacts in La Reunion, covering the period from 1655 to 2025. The historical perspective and integration of multiple hazard types are useful contributions. However, I have several concerns regarding the consistency of the reported event numbers, treatment of source reliability and missing information, construction of the significance index, and interpretation of long-term trends. Some conclusions also appear stronger than what can be supported by the largely descriptive analysis. I therefore think the manuscript would benefit from substantial clarification and revision before publication.
1. The manuscript reports 1,266 events, but later lists 414 storms , 1,612 floods, 1,024 landslides, 241 volcanic eruptions, and other hazards. These numbers clearly exceed 1,266. The authors need to explain whether one physical event can generate several hazard-specific records and distinguish consistently between an event, a hazard occurrence, and a database entry.
2. The reliability threshold of 60 % is important because sources below this threshold are considered insufficiently reliable, yet the Results still report that 11.1 % of mentions score below 60%. Please clarify whether these low-reliability records are excluded from the subsequent event, mortality, and significance analyses or retained in the database and analysis.
3. The significance index may introduce a strong missing-data bias because the authors state that the minimum score is assigned whenever information is missing. Older events naturally have poorer documentation, so this rule may systematically make early events appear less significant. A sensitivity analysis or much stronger justification is needed.
4. Several temporal conclusions appear difficult to separate from the changing availability of historical sources. The manuscript itself acknowledges that recent periods have much more documentation and that the database is not exhaustive. I think the authors should be more cautious when interpreting increases or decreases in event frequency and significance through time.
5. The statement that mortality declined because of improved housing, warning systems, and risk-management policies is plausible, but the analysis presented is mostly descriptive. These variables change simultaneously with population, exposure, reporting practices, health care, transportation, and many other factors. The manuscript should avoid presenting these relationships as causal unless they are formally tested.
6. The mortality analysis needs stronger treatment of population normalization. Reporting average deaths per year before and after 1950 is difficult to interpret because the island population changed enormously over the study period. Mortality per capita, or at least mortality relative to population by historical period, would provide a much more meaningful comparison.
7. I am not convinced by the interpretation of the age data. Age is known for only 13.9% of deaths, yet the manuscript discusses why people under 15 are the most frequently cited group and suggests behavioral and reporting explanations. With this degree of missingness, these explanations are highly speculative and should either be removed or clearly identified as hypotheses.
8. Figure 9 and its normalization procedure are difficult to understand. Taking the annual maximum for each criterion and then normalizing these values may cause a single exceptional record to dominate a year and does not describe the distribution of events occurring that year. The authors should justify this aggregation and ideally show whether the conclusions remain when using median, mean, or event-level distributions