Global disaster risk assessment from Emergency Events Database (2013–2023)

Kong, Qingzhao; Zhu, Erqi

doi:10.5194/nhess-26-4881-2026

Articles | Volume 26, issue 10

https://doi.org/10.5194/nhess-26-4881-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/nhess-26-4881-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 26, issue 10

Research article

|

07 Oct 2026

Research article |

| 07 Oct 2026

Global disaster risk assessment from Emergency Events Database (2013–2023)

Qingzhao Kong and Erqi Zhu

Download

Final revised paper (published on 07 Oct 2026)
Preprint (discussion started on 01 Jul 2025)

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-2706', Anonymous Referee #1, 29 Jul 2025

The authors are to be commended for the substantial effort to undertake their study, using an events database from 2013 to 2023. However, this is a narrow time window for a proposed global multi-disaster risk assessment index using 5°×5° latitude-longitude grids.
The authors state that this narrow time window is for preliminary research. Using a threshold of 50 fatalities (Total Deaths), their study identifies 344 major disaster events for detailed statistical analysis. However, death count is an unreliable risk measure, since it fluctuates according to a number of factors such as time of day of event occurrence, and the effectiveness of hazard warnings.
For more meaningful results, the authors should consider a much longer time window, with a lower threshold for fatalities.

Citation: https://doi.org/10.5194/egusphere-2025-2706-RC1
- AC1: 'Reply on RC1', Erqi Zhu, 06 Aug 2025
  
  We are grateful for the constructive and insightful comments provided by the referee. These comments helped us clarify key methodological aspects of our manuscript and inspired meaningful revisions. In response, we have carefully addressed each point in the attached response letter. Specifically, we provided detailed clarifications regarding the time window and the risk measure adopted in the study. In addition, we also revised the manuscript by adding a forward-looking statement at the end of the Conclusions section, outlining potential improvements in temporal coverage and indicator design. We hope the revised version meets your expectations and respectfully submit it for your further consideration.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2706-AC1
RC2:
'Comment on egusphere-2025-2706', Anonymous Referee #2, 19 Aug 2025

The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-2706/egusphere-2025-2706-RC2-supplement.pdf

Citation: https://doi.org/10.5194/egusphere-2025-2706-RC2
- AC2: 'Reply on RC2', Erqi Zhu, 25 Aug 2025
  
  We sincerely thank the referee for the careful review and constructive feedback. The comments have been invaluable in improving the methodological rigor, clarity, and overall contribution of our manuscript. In the revised version, we have substantially clarified the objectives and practical relevance of the ARI, provided a coherent methodological workflow, and added justifications for threshold and grid-size choices. We also expanded the discussion to address uncertainties, robustness, and implications for disaster risk management, while refining the integration and supplement of remote sensing data to better align with the ARI framework. Detailed responses to each comment are provided in the attached reply letter. We are grateful for the referee’s guidance and hope that the revisions meet the expectations.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2706-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (30 Oct 2025) by Animesh Gain

AR by Erqi Zhu on behalf of the Authors (15 Dec 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (16 Dec 2025) by Animesh Gain

RR by Anonymous Referee #3 (22 Mar 2026)

Suggestions for revision or reasons for rejection

This manuscript presents a novel and well-structured approach to disaster risk assessment. The authors propose an Accumulated Risk Index (ARI) that combines recent disaster event data (from EM-DAT) with the established World Risk Index (WRI) to identify spatial hotspots.
Some issues that need further clarification
The index is defined effectively as a weighted count of recent high-fatality events within a grid cell, where the weight is the country-level WRI. The authors refer to his as accumulated risk, but this definition is single dimensional and as they acknowledge in Equation 1do not account the elements hazard, exposure, and vulnerability. ARI does not measure the magnitude or severity of each event, nor does it account for the spatial extent of the hazard footprint. The WRI is used as a constant weight per country, not as a dynamic measure of vulnerability.
The spatial processing of EM-DAT data is a major methodological weakness. The authors state that for events with textual location descriptions, they select a single representative location from the affected area. For events spanning multiple administrative units, they select one representative coordinate to avoid double-counting. This approach introduces several unresolved challenges:
- The loss of spatial information: e.g. a flood that affects an entire river basin (e.g., the 2022 Pakistan floods) is reduced to a single point.
- The criteria for selecting the representative location are not sufficiently detailed. Which site is chosen when multiple locations are affected? How is the location of an affected area determined for a spatially extensive event?
- By artificially assigning one coordinate, the true spatial footprint is suppressed, and neighboring high-impact areas may be incorrectly assigned zero events.
the examples show that for the 2019 Bahamas storm, the selected coordinate is for Great Abaco Island, but the text description includes both Great Abaco and Grand Bahama. This illustrates the problem of using a single point.
The paper lacks any form of validation. While the sensitivity analyses (Section 3.3) are a solid point, they do not answer whether the ARI actually identifies meaningful "risk" or "accumulated impacts." The authors do not compare ARI hotspots to any independent measure of disaster impact or risk.
The remote sensing dataset (3H) is a valuable contribution, but its connection to ARI is not fully justified. The dataset includes imagery for all areas within the 14 high-risk grids, not specifically for the locations where major disasters occurred. For a grid cell spanning hundreds of kilometers, this includes many areas that may not have been directly affected. The supplementary imagery (Grids 3 and 7) is centered on the location of the highest-fatality event, which is a good approach, but this is not done for the other grids.
The dataset is presented as a tool for future research. However, the manuscript does not demonstrate how the imagery correlates to the ARI or to the disaster events themselves.
The abstract is quite dense. Consider shortening it to focus more clearly on the ARI methodology, the key findings and the intended contribution.
The intended utilization of ARI is listed for policymakers, practitioners, and researchers, but the manuscript does not later return to these specific purposes.
Please ensure the source citations are correctly formatted in Table 1 and that the "statistic year" for the Natural Disaster Hotspots Project is accurate
Add a note in the text acknowledging this limitation explicitly.
Several typos throughout the text should be double checked, as for example the ARI formula
Figure 2 Please explain what "N-Na" means and how overlap degree is calculated.
Figure 3 and 4 The color scales for ARI and Equivalent WRI are different (ARI uses a continuous color bar, Equivalent WRI uses discrete categories). For direct comparison (especially in Figure 4), it would be helpful to use a consistent color scale, or to clearly explain why different scales are used.
Figure 5 This figure is not present in the provided PDF. It is referenced in the text (page 11) but the figure file is missing.
Table 4 shows that the number of events changes with threshold, but the authors conclude that the main spatial clusters remain stable.
The authors should clarify how the tiling was performed (e.g., was there any overlap? Were tiles that contain only water or non-vegetated areas included? Was there any filtering for cloud cover beyond visual inspection?).
The references are generally well-formatted, but there are a few issues that need attention.

Hide

RR by Mohammad Mokhtari (24 Mar 2026)

RR by Anonymous Referee #5 (27 Mar 2026)

Suggestions for revision or reasons for rejection

Dear authors,
Thank you for your efforts. I have included my suggestions below.
Best
For Abstract
The ARI formulation is not described explicitly. Please briefly include the core mathematical idea (example: accumulation of WRI-weighted events).
Please soften the language “valuable resource” and “precise analyses” to “potentially useful for” like phrases.
Introduction
The review is descriptive but not critical enough. For example, explicitly compare methodological paradigms (indicator-based vs statistical vs hybrid). Clarify why existing indices fail quantitatively.
The paper implies a gap but does not state a clear research question. Add a statement like “This study addresses whether a grid-based accumulation framework improves global multi-hazard risk differentiation.
Data acquisition
Over 50 deaths, this threshold introduces selection bias toward high-mortality events.
It ignores economic loss from disasters and slow-onset hazards.
Please provide a sensitivity analysis or a justification for why 50? What happens at 10 or 100?
You dismiss other indicators too strongly. Please acknowledge limitations, like “This study prioritizes mortality but does not capture economic or indirect impacts.”
You exclude some subtypes because there are no major events. This may distort hazard representativeness. Please clarify whether exclusion is methodological or data-driven, and the potential bias introduced.
Mathematical clarity
Equation (3) is poorly formatted and unclear, and variables are not clearly defined. Please rewrite the equation cleanly and define all variables explicitly.
Remote Sensing Dataset
The dataset feels somewhat detached from the risk model. Please explicitly link: how ARI guides dataset selection and potential ML applications.
Image filtering criteria are vague, for instance, “removing damaged or irrelevant images.”
Please define quality criteria and a preprocessing pipeline.
Also, please discuss geographic imbalance, not only focus on Asia.
For Conclusion
There is an overgeneralization. Claims about “bias in traditional methods” are not rigorously proven. Please tone down or support with a quantitative comparison.
For the limitation, please add data bias (EM-DAT), mortality threshold bias, spatial simplification, and ARI assumptions.
Your future work is too generic. Please add concrete directions, dynamic ARI, integration with climate projections, and machine learning applications.
Regards.

Referee Report: PDF

Hide

ED: Reconsider after major revisions (further review by editor and referees) (31 Mar 2026) by Animesh Gain

AR by Erqi Zhu on behalf of the Authors (27 Apr 2026) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (27 Apr 2026) by Animesh Gain

RR by Anonymous Referee #3 (16 May 2026)

Suggestions for revision or reasons for rejection

This manuscript introduces the Accumulated Risk Index (ARI) and a remote sensing dataset (3H) for disaster hotspot identification. While improved from the original version, significant problems remain. The ARI is conceptually circular, the point-based geocoding for floods is a fatal flaw, validation is inadequate, and the 3H dataset is oversold.
- The ARI remains conceptually circular because it sums WRI values for events already selected by a mortality threshold, meaning high-WRI countries contribute more to ARI and high-ARI grids are therefore partly a reflection of WRI itself rather than an independent measure of "accumulated pressure."
- "Accumulated risk" is a misleading label because the index measures the number of major events in a grid multiplied by national WRI, with no recovery period, no temporal decay, and no resilience erosion, so it is not "accumulated risk" in any dynamic sense.
- The mortality threshold of 50 deaths is arbitrary, and sensitivity analysis shows the top 10 grids change rank under thresholds of 40, 60, and 70 deaths, so the claim of "stability" is overstated.
- Point-based geocoding remains a fatal flaw for floods, as demonstrated by the 2022 Pakistan floods, which were basin-wide covering thousands of square kilometers but were reduced to a single coordinate, meaning ARI cannot distinguish between a localized landslide and a catastrophic basin flood.
- The representative-point selection logic is not reproducible because criteria such as "proximity to geographic center" and "spatial coincidence with most severe impacts" are subjective, with no algorithm, no inter-rater reliability, and no validation against true footprints.
- The 2024 out-of-sample check is not validation but merely a consistency check, because ARI is built on 2013-2023 events and showing that 2024 events also occur in high-ARI grids does not test whether ARI outperforms a null model such as random assignment or population-weighted assignment.
- No comparison against any existing index is provided, so it is unknown whether ARI identifies different hotspots than a simple event-count grid, population-weighted event density, or the WRI itself.
- Sensitivity analysis is incomplete because only thresholds of 40, 60, and 70 are tested, and the authors do not justify the "reasonable neighborhood" claim or quantify what percentage of high-ARI grids at 5° resolution remain high-ARI at 2.5° resolution.
- The 3H dataset is largely detached from ARI because ARI identifies 14 high-risk grids and the authors then compile imagery within those grids from pre-existing RAMP and DigitalGlobe datasets, meaning ARI merely selected which grids to look at rather than generating new data.
- The supplementary acquisitions for Grids 3 and 7 total only 100 km², which is vanishingly small relative to 5° × 5° grids that cover approximately 500,000 km² at the equator, and the authors do not explain why this tiny sample is adequate.
- The abstract still overclaims, and the conclusion states that ARI "shows meaningful external plausibility and internal robustness as a comparative hotspot-identification framework," but this is true only in the weakest sense because it is plausible by construction that high-ARI grids have had more disasters.
- Figure 5 is missing from the PDF, the reference list has formatting inconsistencies, Table 4 uses "" to indicate no prior ARI but the footnote says "-" is used, and supplementary material is not provided despite the authors stating the 3H dataset will be on Zenodo.

Hide

RR by Anonymous Referee #5 (20 May 2026)

ED: Reconsider after major revisions (further review by editor and referees) (05 Jul 2026) by Animesh Gain

AR by Erqi Zhu on behalf of the Authors (05 Aug 2026) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (27 Aug 2026) by Animesh Gain

RR by Anonymous Referee #3 (06 Sep 2026)

ED: Publish subject to technical corrections (22 Sep 2026) by Animesh Gain

AR by Erqi Zhu on behalf of the Authors (25 Sep 2026) Manuscript

Short summary

This study develops the Accumulated-Event Risk Indicator to identify where major disaster events accumulated during 2013–2023 under country-level risk conditions. It identifies 14 principal hotspots, concentrated mainly in South and Southeast Asia and along the Himalayan belt. The study also presents the 3H Dataset, providing standardized high-resolution remote-sensing imagery for finer-scale disaster research.