the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 May 2026
Impact attribution of compound flooding from Tropical Cyclone Idai: Assessing the influence of land cover change and underlying socio-economic drivers using a mixed-methods approach
Abstract. In this study, we investigate the influence of socio-economic drivers on the impacts of compound flooding induced by tropical cyclone (TC) Idai. Making landfall close to the city of Beira in Mozambique in 2019, TC Idai was one of the most devastating TC’s to have hit the Southern Hemisphere. Attribution studies generally quantify the contribution of climate change to extreme events and their societal impacts; however, few studies assess how socio-economic drivers amplify or attenuate those impacts. We develop a mixed-methods approach, combining qualitative data from Key Informant Interviews (KIIs) and Causal Loop Diagrams (CLDs) with quantitative data from a physics-based modelling chain to assess how land use and land cover (LULC) changes over 20 years prior to TC Idai plausibly influenced the compound flooding impacts from TC Idai. Results from the quantitative approach show that land use changes (irrespective of climate change) potentially worsened the flood hazard from TC Idai. Results from the qualitative approach explain the underlying drivers of these land use changes such as deforestation driven by charcoal production and informal urban expansion. By integrating two methodologies, we find that the impacts of TC Idai were not only the result of intense climatic hazards but were amplified by complex, deeply rooted socio-economic processes that create reinforcing cycles of vulnerability and exposure. This research demonstrates the value of an interdisciplinary, mixed-methods approach, using localised contextual information to advance impact attribution in data-scarce settings.
- Preprint
(2356 KB) - Metadata XML
-
Supplement
(953 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-2454', Anonymous Referee #1, 28 Aug 2026
-
AC1: 'Reply on RC1', Poppy Webb, 01 Oct 2026
Dear Editor,
We sincerely thank the Editor and the two reviewers for taking the time to carefully review our manuscript and for providing thoughtful and constructive feedback. We have carefully considered all comments from all two reviewers and have responded to each point to the best of our abilities. We believe that the revisions have strengthened the manuscript and improved its clarity. A detailed point-by-point response to each comment is provided below.
Kind regards,
Poppy Webb on behalf of all co-authors
Reviewer #1
The paper investigates compound flooding associated with Tropical Cyclone Idai in Mozambique, combining qualitative research based on eight key informant interviews on which are constructed causal-loop diagrams. The quantitative scenarios compare observed land cover (in 2000 and 2020) and climate conditions (with and without anthropogenic climate change), while the qualitative component is used in parallel to identify socio-economic drivers such as charcoal-related deforestation, informal urban expansion, poverty, and limited government capacity, as drivers of the flood risk.
The paper currently conflates flood-hazard attribution with impact attribution. The quantitative model estimates changes in flood depth, extent, volume, and discharge under selected counterfactuals; it does not quantify changes in damages, displacement, affected population, livelihoods, or other societal impacts.
Separately, the qualitative analysis provides interesting (if unsurprising) interpretation, but it is not integrated at all into the quantitative estimates; the two parts are not integrated but are conducted in parallel, ostensibly to shed light on each other, but with little evidence of that.
Maybe most importantly, the paper compares the actual impact of the storm against three counterfactuals: (1) without climate change; (2) with 2000 LULC rather than 2020 LULC; (3) combining (1) and (2). Counterfactual (1) was already investigated in a previous paper by a some of the same authors; (2) is possibly interesting but has nothing to do with attribution or climate change; and the relevance of this is not very clear, as we don’t really know much about impacts (see comment below); and it is not entirely clear why (3) is an interesting question. Clearly, if the changes in LULC made the flooding more severe, they would have made the attributable flooding more severe as well. Why the question which one is more important is a worthwhile question to ask is not made clear at all. I think the reader deserves to know that. This is not, as far as I can tell, a policy-relevant question, so what is it useful for?
Response: We thank the reviewer for their thoughtful comments. We agree that our analysis attributes changes in flood hazard rather than flood impacts. While event attribution studies typically do not extend beyond meteorological drivers of extreme weather events, in our study we analyze how the compound flood, which was the primary driver of impacts, was affected by both climate change and land cover change. We agree that it is important to clearly distinguish between climate, hazard, and impact attribution, and we will therefore revise the title and rephrase sections of the introduction, discussion, and conclusion to make these distinctions more explicit:
- We have revised the title by removing the ‘Impact’ of the attribution. We acknowledge this paper is framed at making advances towards a broader concept of impact attribution but given that we do not actually quantify the impacts, the title simply refers to a ‘Attribution’, to avoid confusion.
- In the second paragraph of the Introduction, we define the differences between impact, climate and, hazard attribution- more appropriate to our study yet not formally defined. We have added: “Impact attribution, an evolving concept within attribution science, aims to quantify the extent to which observed social or economic impacts from an extreme event are attributable to climate change (Hope et al., 2022). This differs to climate attribution which quantifies the effect of climate change on the likelihood or intensity of the meteorological drivers of an extreme event (Hergel et al., 2010). Hazard attribution, although not yet formally defined, represents an intermediate step between climate and impact attribution by extending attribution to the resulting hazard, but not to exposure or vulnerability.” (L47-51).
- In the discussion we have tightened the clarity of these terms. For example, in Section 4.2 ‘Implications towards mixed-method storyline attribution’ we clearly define the limits and scope of our analyses that we do not quantify the impacts from the compound flood hazard. In this paragraph it is now clearly stated for example that “We demonstrate a proof of concept, that in data scarce regions, where it is difficult to fully quantity impacts, it is possible and highly feasible to integrate qualitative methods to gain a greater understanding of drivers of impacts and produce more policy-relevant insights (Jack., 2025; Grant et al., 2015). Within the rapidly evolving field of impact attribution (Perkins-Kirkpatrick et al., 2024), this research shows how integrating qualitative causal mapping with quantitative modelling provides a practical pathway for advancing impact attribution where observational data are limited.” (L420-422).
- In the conclusion, we focus our narrative on the feasibility and benefits of a mixed method approach towards impact attribution, and the added value of integrating qualitative methods to contextualise quantitative modelled results. “Our study exposed under-discussed but crucial drivers of the impacts of TC Idai such as that of charcoal driven deforestation and informal urban land conversions. Whilst the hydrodynamic modelling showed that land-use changes can amplify the flood hazard, the qualitative analysis contextualised the model results and revealed important socioeconomic drivers of this same hazard. By employing a mixed-method approach, our analysis highlights that TC Idai’s impacts were not only the result of the intense climatic hazards but also amplified by socio-economic drivers.” (L480-485).
The reviewer is also correct that the quantitative and qualitative components of the study are not fully integrated. The main purpose of the qualitative analysis is to provide contextual information about tropical cyclone Idai and its drivers and consequences. This information informed the development of counterfactual scenarios and provides an important basis for interpreting the attribution results. In response to this comment, we have revised Figure 1 (see below) to more clearly illustrate the link between the quantitative and qualitative components of this study, specifically how the KIIs informed the construction of scenarios and modelling choices, and how the CLD helps placing the attribution statement in the appropriate context. We will clarify these links throughout the manuscript.
Finally, we would like to further elaborate on the relevance of attributing changes in LULC. Understanding the relative contributions of climatic and non-climatic drivers is essential for placing attribution results in a decision-making context. Climate change does not occur in isolation. While climate change may alter the probability or magnitude of hazards, land cover change and other societal factors can substantially influence the resulting hazard and associated impacts. To date, these dynamics are often much more important than the impact of climate change (Sauer et al., 2021; Rogers et al., 2025). Explicitly considering land cover change therefore helps identify opportunities for risk reduction that are within the control of local and national decision-making. In the case of Idai, this broader perspective helps demonstrate that the impact of Idai, but also future risks, will depend not only on climate change, but also on societal and environmental changes that shape exposure and vulnerability.
Major Comments 1. Reframe or substantiate the claim of impact attribution
The title, abstract, and discussion frame the paper as an impact-attribution study. Yet the quantitative results are flood-hazard results: maximum flood depth, flood volume, flood extent, and discharge responses under climate and LULC counterfactuals. These are important hazard metrics, but they are not socio-economic, cultural, or environmental impact metrics. The qualitative interviews identify mechanisms of vulnerability and exposure, but the manuscript does not propagate the modelled hazard changes through exposed population, exposed assets, displacement caused, livelihoods lost, or damage experienced.
This matters because the strongest conclusion in the manuscript is that LULC changes amplified the impacts of TC Idai. The evidence supports the more cautious claim that LULC changes plausibly affected vulnerability/exposure (though this is not investigated) and that LULC changes modified flood hazard locally. We still don’t know if impacts were amplified.
I would further recommend to use the terms hazard, exposure, vulnerability, risk, and impact consistently throughout the manuscript, following the IPCC or UNDRR definitions of these terms (both have glossaries where these are defined).
We agree with the reviewer for this comment and have revised our manuscript accordingly. Although we can extend the framework and compute exposed population or economic damages, for simplicity we decided to limit the analysis to flood hazard. However, as the reviewer points out, we could have made the implication of this choice clearer, and we could have been more cautious with our conclusions. In the revised manuscript, we now make a clear distinction between climate, hazard, and impact attribution, and we have used the risk terminology more consistently. For example:
- We now clearly defined the scope of our research that we limited our analysis to the flood hazard, and we do not make unjustified claims towards ‘impact’ attribution. In the Introduction, it now clearly defines the differences between climate, hazard and impact attribution and frame our analysis as hazard attribution:
- In the Introduction we explain that “Impact attribution, an evolving concept within attribution science, aims to quantify the extent to which observed social or economic impacts from an extreme event are attributable to climate change (Hope et al., 2022). This differs to climate attribution which quantifies the effect of climate change on the likelihood or intensity of the meteorological drivers of an extreme event (Hergel et al., 2010). Hazard attribution, although not yet formally defined, represents an intermediate step between climate and impact attribution by extending attribution to the resulting hazard, but not to exposure or vulnerability” (L46-51).
- In the methods we explain in greater detail storyline attribution and explain: “Storyline attribution examines the causality chain through conditional explanations (Shepherd, 2016; Sillmann et al., 2021). This method uses KIIs to examine the causality chain beyond LULC, where the quantitative conditional approach (conditioning on TC track) examines the causal chain from driver to hazard. Our objective is not to provide a statistically representative assessment, but rather to demonstrate, as a proof of concept, how key informant interviews (KIIs) combined with causal loop diagrams (CLDs) can strengthen the development and interpretation of modelling results. (L117-123).
- In the discussion we have clearly stated the implications of limiting our analysis to the flood hazard. For example, “We recognise that within the scope of this paper, we do not quantify the realized impacts from the compound flood hazard, which would require include exposure and vulnerability data. We demonstrate a proof of concept, that in data scarce regions, where it is difficult to fully quantity impacts, it is possible and highly feasible to integrate qualitative methods to gain a greater understanding of drivers of impacts and produce more policy-relevant insights (Jack., 2025; Grant et al., 2015). Within the rapidly evolving field of impact attribution (Perkins-Kirkpatrick et al., 2024), this research shows how integrating qualitative causal mapping with quantitative modelling provides a practical pathway for advancing impact attribution where observational data are limited.” (L420-427).
- The LULC counterfactual requires stronger justification and sensitivity testing
The study uses a 2020 LULC map to represent factual conditions during TC Idai, which occurred in March 2019, and a 2000 LULC map as the counterfactual. This choice is driven by data availability, but it is a potential weakness of the paper. The 2020 map may include post-Idai land-cover changes, storm damage, reconstruction, or recovery signals; it is difficult to believe that such a big event did not lead to significant LULC changes. The 2000 map represents a historical landscape in an arbitrary point in time rather than a clearly defined counterfactual based on policy changes, demographic shifts, or any other change related to this exact timing.
The manuscript also attributes observed LULC changes to socio-economic drivers identified in interviews. That is plausible, but the causal link between the mapped 2000-2020 land-cover changes and the interview-derived drivers is not established (other than by the claims from the 8 KIIs). The higher resolution is used to justify the 2000-2020 choice, but this decision involves a significant trade-off. Is it possible to use remote sensing products to construct LULC maps that would be better timed (and many higher frequency), even if they will be less spatially detailed.
We acknowledge the potential weakness of using the 2020 LULC map and possibility that this map may include post-Idai land-cover changes.
It is possible to construct LULC maps from 2019 Landsat imagery using remote sensing products, but this was beyond the scope and purpose of this study. For the highest resolution and accuracy maps available, we chose to use the existing maps created by Lisboa et al., (2024). It is possible that the 2020 LULC from Lisboa et al. (2024) dataset includes post-Idai LULC changes. Charrua et al. (2021) found a decrease in dense vegetation by 59% following the cyclone in the most damaged areas, and therefore it is likely that the 2020 dataset includes some of these changes.
All different LULC datasets classify satellite imagery using different methods and classify land cover into different classes, and therefore it is difficult to compare pre-existing LULC maps from 2019 and 2020 (both world cover LULC datasets and localized studies such as Charrua et al. (2021) and Lisboa et al. 2024). It was beyond the scope and purpose of the study to conduct our own LULC classification. In addition, Charrua et al. (2021) found that the intensity of damage to different vegetation classes is closely related to their physiognomy and distance to Idai’s trajectory. An intense cyclone like Idai caused significant damage to dense vegetation, wetland vegetation due to great defoliation, branch stripping, affecting the condition of the vegetation (Charrua et al., 2021), but may have not changed the broad LULC classes used in the model. We therefore assume that the 2020 classification is unlikely to misrepresent the broader hydrological land-cover distribution at the time of the event, and that the use of the 2020 dataset as our factual is valid for the main purpose of our study. There assumption have now been made explicit in the Methods:
- “We use the LULC map of 2020 to represent the factual conditions during TC Idai, which occurred in 2019. We recognize that using the 2020 LULC map may include post-Idai land cover changes. Whilst it is possible to construct LULC maps from 2019 Landsat imagery using remote sensing products, this was beyond the scope and purpose of this study. With a 30m resolution, we chose to use the existing maps created by Lisboa et al., (2024), that have a higher resolution and accuracy than global datasets. An intense cyclone like Idai caused significant damage to dense vegetation, wetland vegetation due to great defoliation, branch stripping, affecting the condition of the vegetation (Charrua et al., 2021), but may have not changed the broad LULC classes used in the model. We therefore assume that the 2020 classification is unlikely to misrepresent the broader hydrological land-cover distribution at the time of the event, and that the use of the 2020 dataset as our factual is valid for the main purpose of our study. The choice to use the Lisboa et al. (2024) dataset is intended to represent the broad scale, accumulated land-cover changes over the last two decades, not the one-year affect post TC Idai. We use the LULC map of 2000 as a counterfactual scenario and represents the situation before extensive socio-economic development of the region in the last two decades, as revealed through a literature review at the beginning of this research. We recognize the limitation of this assumption and highlight future research possibilities in the Section 4.3.” (L187-199).
In addition, it is likely that both direct and indirect LULC changes from Idai (longer term vegetation decline or human activities like forest degradation from charcoal burning) may not be as prominent in the 2020 dataset but have continued to have implications on LULC change since 2020. Post-Idai land cover changes would be interesting to investigate, especially following the results from our qualitative analysis, but was not the purpose of this study. We have added this limitation explicitly and future research suggestion in the Discussion as follows:
Following the results from our qualitative analysis, it would be interesting to quantify post-Idai LULC changes, both direct and indirect (such as longer-term vegetation decline or human activities like forest degradation from charcoal burning) and their implications on LULC change since 2020.
- “A major limitation is using the Lisboa et al. (2024) 2020 landcover dataset, as the ‘factual’ dataset for 2019. This landcover map therefore is likely to include LULC changes from TC Idai itself or in the recovery period after (Charrua et al., 2021)”. (L437-439)
- “Following the results from our qualitative analysis, it would be interesting to quantify post-Idai LULC changes, both direct and indirect (such as longer-term vegetation decline or human activities like forest degradation from charcoal burning) and their implications on LULC change since 2020.” (L466-468)
The choice to use the Lisboa et al. (2024) dataset is intended to represent accumulated land-cover changes over the last two decades, not the one-year affect post TC Idai. The year 2000 may to some degree represent an arbitrary point in time, but we used this Lisboa et al. (2024) dataset as the best available dataset to represent broad scale changes. A thorough literature review at the beginning of this research revealed extensive demographic shifts in this region since 2000, and therefore this dataset captures this wider pattern.
The major LULC transitions identified between 2000 and 2020 represent longer-term changes, particularly the conversion from forest to cropland. We interpret the 2020 LULC map as an approximation to the pre-Idai LULC state and acknowledge this temporal mismatch as a source of uncertainty. We have highlighted this limitation within our study in both the methods and discussion.
- For example, in the discussion it now states: “We interpret the 2020 LULC map as an approximation to the pre-Idai LULC state and acknowledge this temporal mismatch as a source of uncertainty” (L439-441-).
Reference:
Charrua, A. B., Padmanaban, R., Cabral, P., Bandeira, S., & Romeiras, M. M. (2021). Impacts of the Tropical Cyclone Idai in Mozambique: A Multi-Temporal Landsat Satellite Imagery Analysis. Remote Sensing, 13(2), 201. https://doi.org/10.3390/rs13020201
3. The hydrological interpretation of LULC effects needs more explanation
I wasn’t able to follow the hydrological modelling (I am not a hydrologist), but since this is based on a previous paper, and there is nothing new here, maybe this is not an issue.
We indeed use the same modelling framework as in related work (Vertegaal et al., 2026). As such, we focus on the parts of the hydrological modelling that are not covered by other studies.
References:
Vertegaal, D. M., van den Hurk, B. J. J. M., Couasnon, A., Aleksandrova, N., Bovenschen, T., Gradiyanto, F., Leijnse, T. W. B., Goulart, H. M. D., and Muis, S.: Climate and impact attribution of compound flooding induced by tropical cyclone Idai in Mozambique, Nat. Hazards Earth Syst. Sci., 26, 1417–1433, https://doi.org/10.5194/nhess-26-1417-2026, 2026.
4. The qualitative component needs more methodological transparency
The qualitative strand is central to the claimed contribution of the paper (since the quantitative part is very similar to the previous paper published in the same journal). The authors acknowledge the limited number of key informants and the lack of local community and government perspectives. That limitation is important because the paper makes claims about poverty, charcoal livelihoods, informal settlement dynamics, urban planning, and humanitarian access. The paper should make clearer which claims are based on direct interview evidence, which are supported by the literature, and which are interpretive syntheses by the authors.
I would probably also argue that using only 8 KII, neither of them, as far as I can tell, a resident of the affected area, and all of them with limited experience there (and possibly even limited experience in Mozambique) is a significant weakness, even if it is acknowledged. Why not increase the number and variety of KIIs?
The text of the paper should also mention whether the research received ethics approval (or did the authors decide one is not required for the KIIs interviews).
In response to this comment, we more carefully phrased our results, clarifying how certain claims are derived. Specifically, we rephrased in Section 3.2 to ensure that all claims and sentences made are referenced by either the KII or literature sources. In Section 3.2, literature references and corresponding interview references have been added to ensure clarity on how all claims are derived Changes such as the following were made throughout this section (modified text is underscored):
- 3.2: “Figure 3 shows the CLD that synthesizes the dominant causal mechanisms and feedback loops, interpreted by the author from direct KII evidence and supported by available literature.” (L242-245)
- “This rural feedback is further intensified by urban demand (see the blue lines in Fig. 3). As found by Sedano et al. (2016), and explained by KI8, growing urban populations, such as cities such as Beira, sustains the demand for charcoal that drives continued deforestation upstream.” (L326-328)
- “The displacement of urban agricultural practices also reduces food security, and livelihood diversification (Shannon et al., 2021), thereby as KI1 made the link increasing socio-economic vulnerabilities, and heighten future flooding impacts”. (L335-337)
While the inclusion of eight key informants (KIIs) may appear limited, we consider this number sufficient for the purpose of this study. The KIIs were carefully selected based on their expertise and knowledge of the region. Our objective is not to provide a statistically representative assessment, but rather to demonstrate, as a proof of concept, how key informant interviews (KIIs) combined with causal loop diagrams (CLDs) can strengthen the development and interpretation of modelling results. We acknowledge that none of the KIIs were residents in the area affected. This was a limitation of conducting this analysis remotely without field research. The 8 KIIs selected all had considerable experience and insights for the context of this analysis. The aim is to develop and showcase a proof of concept and the value of integrating qualitative information in storyline attribution analysis, which has been made explicit as follows:
- Introduction: “We use eight KIIs, as a proof of concept, ranging from disaster risk reduction practitioners, researchers to humanitarian aid workers, all with experience in Beira, pre- and/or post-TC Idai.” (L104-106)
- Abstract: “By combining the attribution of hazard drivers with qualitative localized contextual information of impact drivers, this paper demonstrates a proof of concept to mixed-method attribution in data-scarce settings.” (L24-26)
- Discussion: ‘In our study, we use a storyline attribution approach to generate climate change and LULC counterfactuals, that are supported through a qualitative analysis.’(L375-277).
- Discussion: “We demonstrate a proof of concept, that in data scarce regions, where it is difficult to fully quantify impacts, it is possible and highly feasible to integrate qualitative methods to gain a greater understanding of drivers of impacts and produce more policy-relevant insights (Jack., 2025; Grant et al., 2015).” (L420-424)
- Discussion: ‘The integration of local contextual insights with quantitative modelling, enabled us to interpret the quantitative results and identifysocio-economic drivers and their interactions that likely contributed to the experienced cascade of impacts associated with TC Idai. (L386-388).
- Conclusion: “By combining quantitative hydrodynamic modelling with qualitative methods, we demonstrate a proof of concept for how storyline attribution analyses can use a mixed-methods approach.” (L474--477)
The impacts of disasters are strongly shaped by their social context, including pre-existing vulnerabilities, governance structures, and risk management practices. These factors and their dynamic interactions are difficult to capture using quantitative methods alone. Despite the clear potential of mixed-method approaches to address these complexities, such approaches remain uncommon in peer-reviewed attribution studies. This study provides an example of how quantitative modelling can be complemented with qualitative insights in a data-poor context. We do not expect that including additional KIIs would substantially change the main conclusions of the study.
In response to the comment regarding an ethics approval, the research was undertaken within the VU Amsterdam research ethical guidelines.
-
AC1: 'Reply on RC1', Poppy Webb, 01 Oct 2026
-
RC2: 'Comment on egusphere-2026-2454', Anonymous Referee #2, 04 Sep 2026
Review of manuscript egusphere-2026-2454
Impact attribution of compound flooding from Tropical Cyclone Idai: Assessing the influence of land cover change and underlying socioeconomic drivers using a mixed-methods approach.
by Poppy E. Webb et al.
This manuscript investigates how climate change, land-use/land-cover change, and underlying socio-economic processes influenced compound flooding during TC Idai in Mozambique using a mixed-method approach. The manuscript would benefit from a clearer distinction between flood-hazard attribution and impact attribution, as the quantitative analysis estimates changes in flood depth, and extent. However, it does not quantify impacts through exposure and vulnerability analysis. The integration between the qualitative and quantitative components should also be explained more explicitly. These revisions would strengthen the methodological part and interpretation of the results. I therefore recommend major revisions before the paper can be considered for publication.
Introduction
1. The definition of compound flooding in terms of fluvial, pluvial and coastal flooding is appropriate for the focus of this study. However, I wonder whether it would be useful to briefly clarify at this point that these represent the flooding-related drivers considered in the study, rather than the complete set of physical hazards associated with a tropical cyclone. TC Idai also involved strong winds, as the authors themselves acknowledge later in the Introduction (lines 74-76). Previous studies on tropical cyclones in Mozambique, including TC Idai, have shown that wind and flooding may contribute differently to direct and indirect impacts (e.g. Mühlhofer et al., 2023; Espejo et al., 2025). A short clarification here would help distinguish the broader physical drivers of TC impacts from the specific compound-flooding processes investigated in this study, without changing the scope of the analysis.
References
Espejo GG, Stalhandske Z, Mühlhofer E, Röösli T, Brönnimann S, Bresch DN and Zischg AP (2025) From hazard to disruption: forecasting direct and indirect tropical cyclone impacts on infrastructure in Mozambique. Front. Clim. 7:1666586. doi: 10.3389/fclim.2025.1666586
Link: https://www.frontiersin.org/journals/climate/articles/10.3389/fclim.2025.1666586/full
Link: https://scholar.google.com/citations?view_op=view_citation&hl=en&user=2TrzxOQAAAAJ&citation_for_view=2TrzxOQAAAAJ:eQOLeE2rZwMC
2. In line 69 in the introduction. There is a typo error probably. “This can be achieved by using a mixed-method approach that combines qualitative and qualitative methods”. It is maybe quantitative?
Methods
The quantitative framework itself is clear: The combination of KIIs/causal mapping with a SFINCS- workflow compound-flood chain and four factual/counterfactual scenarios. But I see the following points worth checking.
3. In Section 2.2.2, the authors describe the use of the 30 m Lisboa et al. (2024) LULC dataset and use the 2020 map to represent factual land-cover conditions for TC Idai, which occurred in 2019. I assume this choice is related to the lack of a comparable regional dataset at that resolution for 2019. However, the current description raises some questions, since the objective is to represent land-cover conditions before the TC Idai 2019. Could the authors clarify why the 2020 dataset is considered representative of conditions during TC Idai, and whether land-cover changes between March 2019 and 2020, including changes caused by Idai itself or post-event recovery/reconstruction, could affect the factual scenario?
The manuscript also states that the Lisboa et al. dataset was “minorly” merged with the 2019 Buchhorn et al. global LULC product to cover the full model domain. Could the authors clarify which parts of the domain required this additional dataset and approximately what proportion of the model area it represents? If feasible, a brief sensitivity check comparing the results with and without this merged area would help demonstrate whether the merging has a significant influence on the results and would support the statement that the merge is minor.
4. After line 105 the Figure1; I suggest reviewing it and modified
Figure 1 may be slightly misleading in the way the qualitative and quantitative components are linked. The arrow from the causal loop diagram directly to “Flood hazard (extent & volume)” suggests that the qualitative analysis is used as an input to the flood-hazard calculation. However, as I understand the methodology, the quantitative modelling chain assesses how the LULC and climate scenarios influence flood hazard, while the KIIs and CLD are used to identify and interpret the socio-economic drivers underlying the observed LULC changes.
I therefore suggest revising Figure 1, or adding a complementary schematic, to make the sequence of the mixed-method approach clearer. For example, were the KIIs conducted and analysed first to develop the CLD, and was the CLD then used to interpret the observed differences between the 2000 and 2020 LULC maps? Or were the LULC changes identified first and then used to guide the qualitative analysis? Clarifying this sequence would help the reader understand how the two components are integrated and avoid the impression that the CLD directly feeds into the quantitative flood model.
Explain it for steps maybe to understand the sequence
5. Figure 4 effectively shows the spatial differences in maximum flood depth between the factual and counterfactual scenarios. However, the interpretation of the positive and negative differences, particularly in the LULC-change panel, could be explained more explicitly in the text. For example, it would be useful to clarify why some areas experience increased flood depth while others show decreases under the factual LULC scenario, and to relate these spatial patterns to the main LULC transitions identified earlier (e.g. deforestation, cropland expansion, urbanization or reforestation). This would help connect the mapped differences more directly to the physical mechanisms discussed in the study.
For example:
Please clarify the physical mechanism proposed for the local decrease in flood depth northwest of Beira. The statement that reforestation reduces land-cover roughness and therefore decreases water velocity appears counterintuitive. Could you please review it. Is right the description?
6. In line 260 there is a part to edit maybe: of “socioeconomic deprivation … was. Those points .........
7. Section: Impacts of changes in climate and LULC on flood hazard.
The combined climate + LULC scenario results in a 12% increase in flood volume, compared with 9% and 2% for the individual effects. Could the authors clarify whether this difference is due to rounding?
8. In lines 377–382, the manuscript refers to assessing impacts and integrating exposure and vulnerability. However, these components are identified qualitatively rather than quantitatively assessed. I suggest using more cautious wording and clearly distinguishing flood-hazard attribution from qualitative interpretation of exposure, vulnerability, and socio-economic drivers. If the authors wish to claim impact assessment more explicitly, this could be strengthened in future work by combining the hazard results with exposure data such as buildings, population, or crops.
I am referring to this part: “We combine a quantitative attribution framework with KIIs visualised in a CLD to assess cascading drivers, impacts, their interactions and consequently reinforcing socio-economic feedback cycles in rural and urban areas. We propose a mixed-methods approach to truly understand real-word impacts and identify drivers of vulnerability and exposure for compound flooding after TC Idai. By doing so, we consider all risk components (Simpson et al., 2021), particularly integrating exposure and vulnerability as they drive the experienced impacts”
10. The dataset referenced at https://doi.org/10.5281/zenodo.19328562 could not be accessed. Please check that all the data is available and code. You could add also add the repository in GitHub for the modelling chain, as it is open source so far I understand.
-
AC2: 'Reply on RC2', Poppy Webb, 01 Oct 2026
Reviewer #2
This manuscript investigates how climate change, land-use/land-cover change, and underlying socio-economic processes influenced compound flooding during TC Idai in Mozambique using a mixed-method approach. The manuscript would benefit from a clearer distinction between flood-hazard attribution and impact attribution, as the quantitative analysis estimates changes in flood depth, and extent. However, it does not quantify impacts through exposure and vulnerability analysis. The integration between the qualitative and quantitative components should also be explained more explicitly. These revisions would strengthen the methodological part and interpretation of the results. I therefore recommend major revisions before the paper can be considered for publication.
We thank the reviewer for their useful and thoughtful comments. We have made changes throughout the manuscript to distinguish with greater clarity the difference between flood-hazard attribution and impact attribution, particularly in Introduction and the Discussion. For example, in the Introduction we have now clearly made the distinction between impact attribution and flood-hazard attribution: ‘Impact attribution, an evolving concept within attribution science, aims to quantify the extent to which observed social or economic impacts from an extreme event are attributable to climate change (Hope et al., 2022). This differs to climate attribution which quantifies the effect of climate change on the likelihood or intensity of the meteorological drivers of an extreme event (Hegerl et al., 2010). Hazard attribution, although not yet formally defined, represents an intermediate step between climate and impact attribution by extending attribution to the resulting hazard, but not to exposure or vulnerability.” (L46-51)
We chose in our analysis to attribute changes in flood hazard, rather than changes in flood impacts, and therefore we have explained this with greater clarity for example in the Discussion section:
- “We recognize that within the scope of this paper, we do not quantify the realized impacts from the compound flood hazard, which would require exposure and vulnerability data. We demonstrate a proof of concept, that in data scarce regions, where it is difficult to fully quantify impacts, it is possible and highly feasible to integrate qualitative methods to gain a greater understanding of drivers of impacts and produce more policy-relevant insights (Jack., 2025; Grant et al., 2015).”(L419-422)
We have updated Figure 1 (Page 5) and have explained more explicitly the link between the qualitative and quantitative components. Specifically, we explain how the KIIs informed the construction of scenarios and modelling choices, and how the CLD helps place the attribution statement in the appropriate context. We have clarified these links throughout the manuscript.
- “This methodology illustrates how the qualitative approach using KIIs and causal mapping (Section 2.1) informed the construction of the scenarios used in the compound flood modelling chain (Section 2.2). The causal loop diagram also helped place the attribution analysis in context, contributing to the overall storyline attribution analysis.” (L114-116)
We hope that these revisions will provide greater clarity to the interpretation of the results.
Introduction
- 1. The definition of compound flooding in terms of fluvial, pluvial and coastal flooding is appropriate for the focus of this study. However, I wonder whether it would be useful to briefly clarify at this point that these represent the flooding-related drivers considered in the study, rather than the complete set of physical hazards associated with a tropical cyclone. TC Idai also involved strong winds, as the authors themselves acknowledge later in the Introduction (lines 74-76). Previous studies on tropical cyclones in Mozambique, including TC Idai, have shown that wind and flooding may contribute differently to direct and indirect impacts (e.g. Mühlhofer et al., 2023; Espejo et al., 2025). A short clarification here would help distinguish the broader physical drivers of TC impacts from the specific compound-flooding processes investigated in this study, without changing the scope of the analysis.
References:
Espejo GG, Stalhandske Z, Mühlhofer E, Röösli T, Brönnimann S, Bresch DN and Zischg AP (2025) From hazard to disruption: forecasting direct and indirect tropical cyclone impacts on infrastructure in Mozambique. Front. Clim. 7:1666586. doi: 10.3389/fclim.2025.1666586
Link: https://www.frontiersin.org/journals/climate/articles/10.3389/fclim.2025.1666586/full
Link: https://scholar.google.com/citations?view_op=view_citation&hl=en&user=2TrzxOQAAAAJ&citation_for_view=2TrzxOQAAAAJ:eQOLeE2rZwMC
We agree with this comment and made revisions accordingly. Specifically, we have adjusted the introduction (lines 102) and have added these suggested references:
“Whilst TC Idai also involved very strong winds (Espejo et al., 2025; Mühlhofer et al., 2023), we focus specifically the compound flooding processes.” (L96-97)
- In line 69 in the introduction. There is a typo error probably. “This can be achieved by using a mixed-method approach that combines qualitative and qualitative methods”. Is it maybe quantitative?
Indeed, yes, this is amended.
Methods
The quantitative framework itself is clear: The combination of KIIs/causal mapping with a SFINCS- workflow compound-flood chain and four factual/counterfactual scenarios. But I see the following points worth checking.
- In Section 2.2.2, the authors describe the use of the 30 m Lisboa et al. (2024) LULC dataset and use the 2020 map to represent factual land-cover conditions for TC Idai, which occurred in 2019. I assume this choice is related to the lack of a comparable regional dataset at that resolution for 2019. However, the current description raises some questions, since the objective is to represent land-cover conditions before the TC Idai 2019. Could the authors clarify why the 2020 dataset is considered representative of conditions during TC Idai, and whether land-cover changes between March 2019 and 2020, including changes caused by Idai itself or post-event recovery/reconstruction, could affect the factual scenario?
The manuscript also states that the Lisboa et al. dataset was “minorly” merged with the 2019 Buchhorn et al. global LULC product to cover the full model domain. Could the authors clarify which parts of the domain required for this additional dataset and approximately what proportion of the model area it represents? If feasible, a brief sensitivity check comparing the results with and without this merged area would help demonstrate whether the merging has a significant influence on the results and would support the statement that the merge is minor.
The 2020 dataset was considered the closest available representation of 2019 conditions. We acknowledge there is a mismatch and the possibility the 2020 dataset may include changes by Idai itself or post-event activities. The purpose of using the Lisboa et al. (2024) dataset was to highlight the broad-scale LULC changes over the 20 years 2000 to 2020, for example the conversion of forest land to cropland. It was beyond the scope of this analysis to conduct our own LULC classification of satellite imagery, and therefore we chose the dataset to represent with greatest accuracy broad-scale LULC changes.
Land-changes occurring post-Idai were identified by Charrua et al. (2021) and therefore are likely to affect the factual scenario. We do not dismiss this claim and have highlighted these limitations with greater clarity in Discussion. However, due to differences in classification methods of different datasets, it is difficult to quantify and truly identify the extent to which the 2020 dataset we used may include post-Idai LULC changes. It is highly probable that damage to vegetation from TC Idai will have damaged the vegetation condition but not necessarily changed the classification of the LULC class used in the maps. We acknowledge these limitations in the data resolution and highlight the purpose of this analysis is to use the 2020 dataset as an approximation of the 2019 conditions.
- In the methods we now explicitly state ‘’We use the LULC map of 2020 to represent the factual conditions during TC Idai, which occurred in 2019. We recognize the limitation that using the 2020 LULC map may include post-Idai land cover changes. Whilst it is possible to construct LULC maps from 2019 Landsat imagery using remote sensing products, this was beyond the scope and purpose of this study. With a 30m resolution, we chose to use the existing maps created by Lisboa et al., (2024), that have a higher resolution and accuracy than global datasets. An intense cyclone like Idai caused significant damage to dense vegetation, wetland vegetation due to great defoliation, branch stripping, affecting the condition of the vegetation (Charrua et al., 2021), but may have not changed the broad LULC classes used in the model. We therefore assume that the 2020 classification is unlikely to misrepresent the broader hydrological land-cover distribution at the time of the event, and that the use of the 2020 dataset as our factual is valid for the main purpose of our study. The choice to use the Lisboa et al. (2024) dataset is intended to represent the broad scale, accumulated land-cover changes over the last two decades, not the one-year affect post TC Idai. We use the LULC map of 2000 as a counterfactual scenario and represents the situation before extensive socio-economic development of the region in the last two decades, as revealed through a literature review at the beginning of this research. We recognize the limitation of this assumption and highlight future research possibilities in the Section 4.3.” (L184-196).
In response to the comment regarding the merging of the Lisboa et al. land use land cover (LULC) dataset with the 2019 global LULC product (here referred to as Vito), we have performed an additional sensitivity analysis using the European Space Agency (ESA) Worldcover global land cover dataset (Zanaga et al., 2021) instead (Figures R1-4 below). Merging with an additional LULC product is necessary to ensure complete coverage of the hydrological model domain and to correctly represent the catchment boundaries (Figure R1a). The area covered by the supplementary dataset represents 24.8% of the model domain and is indicated by the hatched region in Figure 2 of the submitted manuscript. Figures R1b-c show the land use maps resulting from merging the Lisboa dataset with the Vito and ESA products, respectively.
The sensitivity analysis indicates that the choice of supplementary LULC dataset has only a minor influence on the simulated discharge (Gauge 1 and 3, most affected by the merged area, and flood depths (Figures R2-R3), with the largest differences observed at gauges 1 and 3. These differences are substantially smaller than those associated with the LULC change between 2000 and 2020. We therefore conclude that the model results are largely insensitive to the choice of supplementary dataset and that the merging procedure is justified. Moreover, any residual differences are effectively cancelled when comparing the factual and counterfactual scenarios, as both use the same supplementary LULC dataset. Using the Vito dataset, which is adopted in the manuscript, produces slightly lower flood depths than the ESA-based merge (Figure R3c), making our results marginally more conservative. The locations of the wflow gauges used to force the flood model are shown in Figure R4.
Reference:
- Zanaga, D., Van De Kerchove, R., De Keersmaecker, W., Souverijns, N., Brockmann, C., Quast, R., Wevers, J., Grosu, A., Paccini, A., Vergnaud, S., Cartus, O., Santoro, M., Fritz, S., Georgieva, I., Lesiv, M., Carter, S., Herold, M., Li, L., Tsendbazar, N.-E., … Arino, O. (2021). ESA WorldCover 10 m 2020 v100 (Version v100) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.5571936
Figure R1: Maps of the Factual Lisboa land use map for 2020 (a) merged with Vito (b) and Esa Worldcover (c) for the missing land use data in the wflow domain that aligns with the upstream catchment of our flood model area. The Lisboa dataset coverage is outlined in a black dashed line and the wflow basin boundary in a blue dashed line (showing all basins in panel a).
(Please see Supplement for referred figures)
Figure R2: Modelled discharge during TC Idai for different wflow gauges, which flow into the flood model. The modelled discharge uses Lisboa 2020 merged with Vito (blue) or Lisboa 2020 merged with ESA Woldcover as land use dataset (orange). Location of the different gauges can be found in Figure R4.
(Please see Supplement for referred figures)
Figure R3: Maps showing the difference in maximum flood depth between the factual and counterfactual land use and climate change scenarios, between the factual (Lisboa merged with vito) and factual merged with esa worldcover land use dataset.
(Please see Supplement for referred figures)
Figure R4: Overview of the wflow and flood (SFINCS) model boundaries in dashed blue and grey, respectively, including the river network and gauge locations.
- After line 105 the Figure1; I suggest reviewing it and modified
Figure 1 may be slightly misleading in the way the qualitative and quantitative components are linked. The arrow from the causal loop diagram directly to “Flood hazard (extent & volume)” suggests that the qualitative analysis is used as an input to the flood-hazard calculation. However, as I understand the methodology, the quantitative modelling chain assesses how the LULC and climate scenarios influence flood hazard, while the KIIs and CLD are used to identify and interpret the socio-economic drivers underlying the observed LULC changes.
I therefore suggest revising Figure 1 or adding a complementary schematic to make the sequence of the mixed-method approach clearer. For example, were the KIIs conducted and analyzed first to develop the CLD, and was the CLD then used to interpret the observed differences between the 2000 and 2020 LULC maps? Or were the LULC changes identified first and then used to guide the qualitative analysis? Clarifying this sequence would help the reader understand how the two components are integrated and avoid the impression that the CLD directly feeds into the quantitative flood model.
Explain it for steps maybe to understand the sequence.
Thank you for these constructive comments. We have revised Figure 1 by removing the inaccurate line that went directly from the ‘Causal loop diagram’ to the ‘Flood hazard’ box. We have added two dashed lines to represent the integration of the qualitative information both in the construction of the scenarios, and in complimenting the analysis of the storyline attribution of the flood hazard. The dashed lines are used to represent the more interpreted links between the qualitative and quantitative methods. The box ‘Storyline attribution analysis’ was added to around the ‘flood hazard’ to represent the integration of the qualitative insights from the causal loop diagram with the attributed flood hazard information.
- Figure 4 effectively shows the spatial differences in maximum flood depth between the factual and counterfactual scenarios. However, the interpretation of the positive and negative differences, particularly in the LULC-change panel, could be explained more explicitly in the text. For example, it would be useful to clarify why some areas experience increased flood depth while others show decreases under the factual LULC scenario, and to relate these spatial patterns to the main LULC transitions identified earlier (e.g. deforestation, cropland expansion, urbanization or reforestation). This would help connect the mapped differences more directly to the physical mechanisms discussed in the study.
For example:
Please clarify the physical mechanism proposed for the local decrease in flood depth northwest of Beira. The statement that reforestation reduces land-cover roughness and therefore decreases water velocity appears counterintuitive. Could you please review it. Is right the description?
We thank the reviewer for pointing out this counterintuitive statement and agree this was an error. We have reviewed the text associated with Figure 4, and made the following revision:
- “In the area Northeast of Beira, Figure 4 (panels a and b) shows a local decrease in flood depth (shown in blue). This can be linked to local reforestation (Figure 2), resulting in increased surface roughness, and therefore decreased water velocity that potentially corresponds to the localised reduction in flood depth.” (L360-365)
- In line 260 there is a part to edit maybe: of “socioeconomic deprivation … was. Those points .........
Amended.
- Section: Impacts of changes in climate and LULC on flood hazard.
The combined climate + LULC scenario results in a 12% increase in flood volume, compared with 9% and 2% for the individual effects. Could the authors clarify whether this difference is due to rounding?
The effect is not due to rounding but due to non-linear effects.
- In lines 377–382, the manuscript refers to assessing impacts and integrating exposure and vulnerability. However, these components are identified qualitatively rather than quantitatively assessed. I suggest using more cautious wording and clearly distinguishing flood-hazard attribution from qualitative interpretation of exposure, vulnerability, and socio-economic drivers. If the authors wish to claim impact assessment more explicitly, this could be strengthened in future work by combining the hazard results with exposure data such as buildings, population, or crops.
I am referring to this part: “We combine a quantitative attribution framework with KIIs visualised in a CLD to assess cascading drivers, impacts, their interactions and consequently reinforcing socio-economic feedback cycles in rural and urban areas. We propose a mixed-methods approach to truly understand real-word impacts and identify drivers of vulnerability and exposure for compound flooding after TC Idai. By doing so, we consider all risk components (Simpson et al., 2021), particularly integrating exposure and vulnerability as they drive the experienced impacts”
We agree with the reviewer that our attribution analysis does not fully include exposure and vulnerability. Therefore, we have made the following revisions:
- The dataset referenced at https://doi.org/10.5281/zenodo.19328562 could not be accessed. Please check that all the data is available and code. You could add also add the repository in GitHub for the modelling chain, as it is open source so far I understand.
We find it important that our work is reproducible and have checked the link and ensures all data and code is openly available. The zenodo link actually links to a specific release on GitHUb of our modelling chain. The correct Zenodo link has now been added: https://doi.org/10.5281/zenodo.20505361.
-
AC2: 'Reply on RC2', Poppy Webb, 01 Oct 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 169 | 80 | 27 | 276 | 34 | 16 | 20 |
- HTML: 169
- PDF: 80
- XML: 27
- Total: 276
- Supplement: 34
- BibTeX: 16
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The paper investigates compound flooding associated with Tropical Cyclone Idai in Mozambique, combining a qualitative research based on eight key informant interviews on which are constructed causal-loop diagrams. The quantitative scenarios compare observed land cover (in 2000 and 2020) and climate conditions (with and without anthropogenic climate change), while the qualitative component is used in parallel to identify socio-economic drivers such as charcoal-related deforestation, informal urban expansion, poverty, and limited government capacity, as drivers of the flood risk.
The paper currently conflates flood-hazard attribution with impact attribution. The quantitative model estimates changes in flood depth, extent, volume, and discharge under selected counterfactuals; it does not quantify changes in damages, displacement, affected population, livelihoods, or other societal impacts.
Separately, the qualitative analysis provides interesting (if unsurprising) interpretation, but it is not integrated at all into the quantitative estimates; the two parts are not integrated but are conducted in parallel, ostensibly to shed light on each other, but with little evidence of that.
Maybe most importantly, the paper compares the actual impact of the storm against three counterfactuals: (1) without climate change; (2) with 2000 LULC rather than 2020 LULC; (3) combining (1) and (2). Counterfactual (1) was already investigated in a previous paper by a some of the same authors; (2) is possibly interesting but has nothing to do with attribution or climate change; and the relevance of this is not very clear, as we don’t really know much about impacts (see comment below); and it is not entirely clear why (3) is an interesting question. Clearly, if the changes in LULC made the flooding more severe, they would have made the attributable flooding more severe as well. Why the question which one is more important is a worthwhile question to ask is not made clear at all. I think the reader deserves to know that. This is not, as far as I can tell, a policy-relevant question, so what is it useful for?
Major Comments 1. Reframe or substantiate the claim of impact attribution
The title, abstract, and discussion frame the paper as an impact-attribution study. Yet the quantitative results are flood-hazard results: maximum flood depth, flood volume, flood extent, and discharge responses under climate and LULC counterfactuals. These are important hazard metrics, but they are not socio-economic, cultural, or environmental impact metrics. The qualitative interviews identify mechanisms of vulnerability and exposure, but the manuscript does not propagate the modelled hazard changes through exposed population, exposed assets, displacement caused, livelihoods lost, or damage experienced.
This matters because the strongest conclusion in the manuscript is that LULC changes amplified the impacts of TC Idai. The evidence supports the more cautious claim that LULC changes plausibly affected vulnerability/exposure (though this is not investigated) and that LULC changes modified flood hazard locally. We still don’t know if impacts were amplified.
I would further recommend to use the terms hazard, exposure, vulnerability, risk, and impact consistently throughout the manuscript, following the IPCC or UNDRR definitions of these terms (both have glossaries where these are defined).
2. The LULC counterfactual requires stronger justification and sensitivity testing
The study uses a 2020 LULC map to represent factual conditions during TC Idai, which occurred in March 2019, and a 2000 LULC map as the counterfactual. This choice is driven by data availability, but it is a potential weakness of the paper. The 2020 map may include post-Idai land-cover changes, storm damage, reconstruction, or recovery signals; it is difficult to believe that such a big event did not lead to significant LULC changes. The 2000 map represents a historical landscape in an arbitrary point in time rather than a clearly defined counterfactual based on policy changes, demographic shifts, or any other change related to this exact timing.
The manuscript also attributes observed LULC changes to socio-economic drivers identified in interviews. That is plausible, but the causal link between the mapped 2000-2020 land-cover changes and the interview-derived drivers is not established (other than by the claims from the 8 KIIs). The higher resolution is used to justify the 2000-2020 choice, but this decision involves a significant trade-off. Is it possible to use remote sensing products to construct LULC maps that would be better timed (and many higher frequency), even if they will be less spatially detail
3. The hydrological interpretation of LULC effects needs more explanation
I wasn’t able to follow the hydrological modelling (I am not a hydrologist), but since this is based on a previous paper, and there is nothing new here, maybe this is not an issue.
4. The qualitative component needs more methodological transparency
The qualitative strand is central to the claimed contribution of the paper (since the quantitative part is very similar to the previous paper published in the same journal). The authors acknowledge the limited number of key informants and the lack of local community and government perspectives. That limitation is important because the paper makes claims about poverty, charcoal livelihoods, informal settlement dynamics, urban planning, and humanitarian access. The paper should make clearer which claims are based on direct interview evidence, which are supported by the literature, and which are interpretive syntheses by the authors.
I would probably also argue that using only 8 KII, neither of them, as far as I can tell, a resident of the affected area, and all of them with limited experience there (and possibly even limited experience in Mozambique) is a significant weakness, even if it is acknowledged. Why not enlarge the number and variety of KIIs?
The text of the paper should also mention whether the research received ethics approval (or did the authors decide one is not required for the KIIs interviews).