the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
An emergency-oriented risk assessment method for typhoon storm surge disasters
Abstract. Typhoon storm surge disaster (TSSD) is one of the most severe marine disasters affecting China's coastal areas, and risk assessment plays a crucial role in disaster prevention and mitigation. Existing studies primarily focus on risk zoning and long-term disaster reduction strategies, whereas risk assessment for an imminent TSSD, which is essential for pre-disaster emergency response, remains insufficient. The main challenges lie in quantifying hazard uncertainty and in representing the multiple dimensions of vulnerability. To address these challenges, this study develops an emergency-oriented risk assessment method for an imminent TSSD that integrates ensemble forecasting into a complete risk assessment framework. For hazard assessment, ensemble forecasting techniques are employed to address hazard uncertainty. A controlled scenario is generated by integrating multi-source forecast data, and perturbed scenarios are derived based on historical forecast errors. These scenarios form an ensemble of 15 scenarios for hydrodynamic simulations, with the mean and maximum water depths characterizing the hazard intensity grade. For vulnerability assessment, land use types and densely populated facilities are selected as two categories of vulnerability. The water depth-loss relative grade relationship for 11 land use types is established, and Point of Interest (POI) density is introduced to represent potential casualty vulnerability. Comprehensive risk is calculated by combining the results of hazard and vulnerability assessments. The method is validated through a case study of the Hato TSSD (2017) in Zhuhai, with the forecast initialization time set at approximately 24 hours before landfall. The proposed method can be re-executed to provide dynamic risk assessment updates as forecast data evolve, providing timely support for emergency decision-making.
- Preprint
(9460 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 19 Nov 2026)
- RC1: 'Comment on egusphere-2026-3929', Anonymous Referee #1, 19 Oct 2026 reply
-
RC2: 'Comment on egusphere-2026-3929', Anonymous Referee #2, 30 Oct 2026
reply
This manuscript proposes an emergency-oriented framework that combines multi-agency tropical cyclone forecasts, MIKE 21 inundation simulations, and land-use/POI indicators to map storm-surge risk before landfall. The topic is relevant, and the Hato case illustrates a potentially useful workflow. The following issues should be addressed before the principal conclusions can be accepted.
1. The 15-member ensemble is generated by crossing five tracks with three intensities, with perturbation magnitudes set to historical mean forecast errors (Sect. 2.2). A mean error is neither a probabilistic bound nor a description of joint track–intensity uncertainty, so the coverage of plausible Hato outcomes is not established. For comparison, Davis et al. (2010, Monthly Weather Review, doi:10.1175/2010MWR3136.1) derived surge ensembles from forecast-error probability distributions and assessed inundation coverage across forecast lead times; the fixed mean-error perturbations used here do not provide an equivalent probabilistic interpretation. Please report the underlying error distributions and assumptions, show whether the observed track, intensity, and water levels fall within the simulated range, and test sensitivity to perturbation size. The grid-cell mean and maximum of 15 equally weighted simulations (Eqs. 10–11) should not be called ‘expected’ and ‘worst-case’ inundation without probability calibration; the maximum map may also combine peaks from different storms and thus correspond to no single event. Relabel these outputs or justify their probabilistic interpretation.
2. The hydrodynamic model is evaluated mainly against water levels at Chiwan and Hengmen stations (Sect. 3.2; Fig. 10), while the stated agreement of simulated and published inundation patterns is qualitative. The manuscript later acknowledges that the mapped flooding differs from the actual affected area and that Hato's seawall breaches and wave overtopping are omitted (Sect. 4.4). Please provide quantitative water-level statistics, including peak magnitude and timing errors, and a spatial comparison with available observed inundation or damage data. Distinguish forecast-input error from hydrodynamic-model error. Because breaches reportedly flooded protected areas, a simplified breach/overtopping sensitivity case, or a clear restriction on where these maps can guide evacuation, is needed.
3. The vulnerability construction needs stronger support. The land-use loss curves combine mean losses from the Z study with maximum losses from the H study, and normalizing each source separately (Eq. 12) does not by itself make the resulting grades comparable. Table 5 extends categories beyond the depth ranges covered by the source curves and assigns cropland, forest land, grassland, water body, and bare land grade V at every depth. Meanwhile, POI density is treated as a proxy for potential casualties without information on exposed population, occupancy, or facility capacity. This differs from the flood-fatality framework of Jonkman et al. (2008, Natural Hazards, doi:10.1007/s11069-008-9227-5), which explicitly considers flood characteristics, exposed population, and evacuation; POI density alone cannot substantiate a casualty-vulnerability estimate. Please show the full source-to-table derivation and extrapolation, assess sensitivity to source selection and normalization, and compare the economic and human-exposure proxies with available Hato observations. Qualify the casualty-related claim if only facility locations are available.
4. The claimed emergency utility and dynamic updating are not demonstrated by one retrospective event initialized only once, at 14:00 on 22 August 2017 (Sects. 3–5). The two lead-time comparisons in Table 6 assess only the controlled track and intensity for Hato; they do not validate inundation/risk performance or show a benefit over a single forecast. Please report the complete time required to obtain inputs, construct and run all 15 simulations, and produce risk maps; compare ensemble-based priorities with a control-only map; and, if feasible, repeat the analysis for successive forecast initializations or an independent typhoon. Otherwise, limit the conclusion to proof of concept and state that operational timeliness and generalizability remain untested.
Minor comments
1. The definitions below Eqs. (4) and (6) refer to annual ‘total’ track/intensity errors, whereas the text calls them annual means. Clarify the denominator and whether unequal annual forecast counts were accounted for.
2. In Table 1, ‘controlled high intensity’ and ‘controlled low intensity’ appear to describe perturbed intensities; revise these labels for consistency with Eqs. (8)–(9).
3. Tables 2 and 5 use opposite open/closed endpoints for depth intervals. Harmonize the conventions so values exactly at 0.5, 1.0, and 2.0 m are assigned consistently.
4. Sect. 2.3 lists ‘hotels and restaurants’ as a facility category but its examples include hotels, inns, and guesthouses only. State whether restaurants are included in the POI dataset.
5. For Eqs. (10)–(11), specify whether each D_n is the instantaneous depth at a common time or the maximum inundation depth over the event; these operations produce different maps.
6. For Figs. 14, 17, and 18, report the inundated area or cell count in each grade and the change between the mean and maximum maps, rather than relying only on visual description.
7. The Code and data availability statement should identify dataset versions and acquisition dates and provide the scenario tracks, model settings, and grading inputs needed to reproduce the maps. Clarify which software performed the post-processing if ‘no custom code’ was developed.
8. The final paragraph of Sect. 5 repeats the claim that the assessment can be dynamically updated; remove the duplication and distinguish demonstrated results from future capability.
Citation: https://doi.org/10.5194/egusphere-2026-3929-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 346 | 144 | 42 | 532 | 43 | 41 |
- HTML: 346
- PDF: 144
- XML: 42
- Total: 532
- BibTeX: 43
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors use MIKE21 to simulate storm surge hazard, and apply land use based loss functions to calculate economic damage from storms. They also apply this to specific urban centers to develop a proxy for casualties. They then combine these into a mixed metric. The storm surge simulations use variance between simulations from different agencies to estimate an uncertainty in typhoon tracks.
They propose this as a way to quickly estimate spatial risk a day or two before landfall. This is a potentially useful and novel contribution.
My main confusion about the paper is what the various hazard/risk/vulnerability levels I-V mean. These seem to be defined based on water levels, but it is not clear how this relates to risk or vulnerability. Also, it is not clear if these are levels defined here for the first time, or if they are in reference to more common values for hazard levels used elsewhere. This should be clarified. I am also confused by the use of the terms hazard, vulnerability, and risk here. Hazard is water depth, which makes sense. Usually vulnerability will mean the depth-damage function itself, while risk is the integral of hazard x vulnerability (if assuming 100% exposure) over all hazard periods. A clearer definition of each of these terms is needed here in order to understand what the authors are doing.
Line 208. Since you neglect waves, you should justify that by stating in what situations the presence of waves is not important. That will be important to determine where your method is applicable and where it should not be applied.
Table 2. Is this something you made up? Or do the hazard intensities reference an existing scale? This is not clear. Also, is this mean or max water depth?
Fig 5 and Table 4. Does "other construction land" mean "other developed land"?
Table 5. Are these loss grades referenced from somewhere? Or your invention?
Line 376. Would be helpful to have a bit more info about the MIKE21 global tide model. Please at least refer to papers or reports that are publicly available and explain it.
Line 377. Again, state where it is reasonable to neglect waves, and where the result of your simulation does not apply due to this assumption of negligible wave setup and other effects.
Fig 10. There seems to be an offset in the astronomical tide of about 4 hours. Is this due to a time zone problem in the data used in the tidal model? Or is it due to complicated shallow water tide behavior.
Line 439. You state that model performance is consistent with Du, 2020. If you state this, you need to back this up with a figure or data.
Fig 15. The inset map does not appear to show any useful information, as it is identical to the main map. You can delete the inset in this case. Fig 16 also.