Mapping uncertainty in flood impacts: a multi-source assessment and catalogue of simulated, remotely sensed and reported floods in Europe

Scarpellini, Lorenzo; Ficchì, Andrea; D'Angelo, Claudia; Betterle, Andrea; Salamon, Peter; Castelletti, Andrea

doi:10.5194/egusphere-2026-2807

Preprints

https://doi.org/10.5194/egusphere-2026-2807

Preprints

01 Jun 2026

| 01 Jun 2026

Mapping uncertainty in flood impacts: a multi-source assessment and catalogue of simulated, remotely sensed and reported floods in Europe

Lorenzo Scarpellini, Andrea Ficchì, Claudia D'Angelo, Andrea Betterle, Peter Salamon, and Andrea Castelletti

Abstract. Flooding is Europe’s most costly natural hazard, with riverine floods alone accounting for over a third of disaster related damages, which have increased nearly tenfold since the 1990s. Despite advances in large-scale hydrological modeling and flood hazard mapping, continental-scale flood risk assessments remain highly uncertain, also due to simplified representations of flood protections and limitations in observational datasets and reported impacts. In this study, we assess how flood protection assumptions and data sources influence flood impact and risk estimates across Europe. We compile and provide a catalogue of simulated flood events based on the European Flood Awareness System (EFAS, version 5.0) modeling chain and quantify differences across three flood protection scenarios. Simulated events are then systematically compared with satellite-derived flood extents and impacts from the Copernicus Global Flood Monitoring (GFM) system, as well as with reported impacts from the HANZE database, using an automated event-matching procedure. Results indicate that flood protections are a major source of uncertainty, strongly affecting simulated flood frequency, spatial distribution, and damages. Moreover, substantial inconsistencies are found between simulations, observations, and reported data. Simulations tend to overestimate flood extents and damages, especially for large events, while satellite observations underestimate small and short-lived floods, particularly in urban environments. Reported datasets capture only a subset of impactful events and are biased toward populated regions, resulting in limited agreement across datasets and in differing regional impact patterns. Overall, results show that no single dataset can be currently considered a ground truth for flood impacts at continental scale, highlighting the need for multi-dataset approaches in large-scale flood risk assessment. These findings contribute to a more robust interpretation of flood risk estimates and support the development of more reliable and uncertainty-aware multi-dataset approaches for continental-scale flood risk assessments.

Received: 15 May 2026 – Discussion started: 01 Jun 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Lorenzo Scarpellini, Andrea Ficchì, Claudia D'Angelo, Andrea Betterle, Peter Salamon, and Andrea Castelletti

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-2807', Anonymous Referee #1, 03 Jul 2026
The manuscript presents a comprehensive assessment of the uncertainties associated with continental-scale flood risk estimation in Europe. By comparing EFAS-LISFLOOD simulations under three flood protection scenarios (FLOPROS, P25, and No Protection) with satellite-derived observations (Copernicus GFM) and historical impact records (Hanze database), the authors examine the characteristics and limitations of each data source.
The analysis shows that flood protection assumptions are a major source of uncertainty, substantially influencing estimated losses and the relative risk ranking of European countries. The manuscript also examines the limitations of different reference datasets, including the tendency of the model to overestimate inundation due to static hazard map aggregation and the tendency of satellite observations to underestimate flooding in urban areas because of SAR-related limitations.
The manuscript is well organized, and the conclusions are generally supported by the presented analyses. I recommend this manuscript for Moderate/Minor Revisions. The following comments are intended to further clarify the discussion and strengthen the manuscript.
Spatial mismatch in flood protection (limitations of FLOPROS): While the manuscript demonstrates the influence of using FLOPROS versus P25, the discussion could further address the implications of applying polygon-based protection standards at the NUTS3 level to localized hydrological processes. Applying a uniform protection standard across an entire NUTS3 region may overrepresent protection in rural and agricultural areas where such infrastructure is limited. Discussing how this spatial mismatch contributes to differences in agricultural damage estimates between the model and GFM would improve the interpretation of the results.

Addressing the low return-period gap: The manuscript identifies that the absence of hazard maps for return periods between 2.5 and 10 years accounts for nearly 20% of the unmatched events. Because EFAS is an operational system, the authors may consider emphasizing this finding in the Abstract or Conclusions and suggesting the development of hazard maps for return periods below 10 years in future Copernicus/EFAS releases.

Satellite limitations in urban areas: The manuscript attributes the underdetection of flooding in residential and industrial areas by GFM to SAR backscatter limitations and the 0.1 m water-depth threshold. The discussion could briefly mention potential directions for improving urban flood mapping, such as integrating multiple remote sensing data sources (e.g., SAR and optical imagery) or data assimilation and AI-based approaches.

Integration of multiple data sources: The manuscript demonstrates that each dataset (hydrodynamic simulations, SAR observations, and reported impacts) has different limitations, suggesting that combining multiple information sources could improve flood risk assessment. The discussion could mention recent developments in Earth system foundation models or flood foundation models as a possible research direction. Such models could potentially integrate hydrodynamic simulations, remote sensing observations, and historical impact records to improve consistency among different data sources and reduce the uncertainties identified in this study.

The use of a 90-day window is justified based on economic recovery considerations. However, from a hydrological perspective, this criterion may merge independent flood events occurring within the same period. A brief discussion of the potential influence of this assumption on event matching with the Hanze and GFM datasets, or a short sensitivity statement, would help clarify its implications.

In Section 3.2, the large number of events detected by both GFM and the model but absent from the Hanze database in Scandinavia is attributed to sparse population and limited reporting. It may be useful to further characterize Hanze as an exposure-dependent dataset, whose spatial coverage is influenced by population distribution and reporting practices in addition to hydrological hazard. This distinction would provide a more precise interpretation of the observed discrepancies.
Citation: https://doi.org/10.5194/egusphere-2026-2807-RC1
RC2:
'Comment on egusphere-2026-2807', Anonymous Referee #2, 09 Aug 2026
This manuscript compares simulated flood occurrence and damage under different flood-protection assumptions with satellite observations and reported impacts across Europe. Among the tested protection scenarios, uncertainty is primarily associated with flood protection, and model agreement improves in larger catchments. My main concerns involve the temporal inconsistency among input datasets, the rationale for the continental scale, and several claims in the Conclusions.
Major comments
The input datasets refer to different years or periods. These temporal mismatches introduce uncertainty into the exposure and damage estimates, but the manuscript does not assess or discuss their effects. The authors should explain how differences in temporal coverage and reference years affect the comparison. In addition, the flood analysis covers 1992–2024, whereas the authors use the mean protection value for 2015–2020. Flood-protection standards and infrastructure may have changed during the study period. State whether the analysis accounts for these changes. If it uses a fixed protection level, justify this choice and discuss how it may affect estimates for earlier events. A summary table would make these differences easier to evaluate by reporting each dataset’s source, reference year or period, spatial resolution, and other relevant information.

The manuscript presents continental coverage as a key contribution but does not establish why this scale is necessary for the research question. The authors should explain what the European-scale analysis reveals that studies of selected countries or river basins could not show. National differences in flood protection, exposure, reporting practices, and data availability may affect the comparisons. Discuss how these differences influence the results and whether they limit comparisons across countries. The continental scope also creates an opportunity to discuss transboundary flood-risk management, including implications for communities located near national borders.

The flood-hazard maps cover return periods up to 500 years. The Methods do not explain how the authors handle simulated discharges with return periods above this threshold.

At the end of the Introduction, the authors state that the study will “provide recommendations for improving large-scale flood risk assessments.” The Discussion and Conclusions do not present a defined set of recommendations. List the recommendations and connect each one to evidence from the analysis. Relevant categories may include flood-protection data, satellite observations, event-matching methods, hazard maps, and reported-impact databases. The authors should also explain how practitioners could apply these recommendations in transboundary assessments. Moreover, the Conclusions mention AI as a potential way to address the limitations identified in the study. The claim remains too broad because the manuscript does not specify the tasks, data requirements, or validation strategy. Identify which parts of the workflow could benefit from AI, what training and validation data those applications would require, and which uncertainties AI would leave unresolved. Otherwise, remove the statement or narrow its scope.

Minor comments
Line 11: Replace “observations” with “satellite observations” to distinguish these data from reported flood-impact records.

Lines 33–34: “Thanks to” sounds informal in this context. Consider “due to” or “because of.”

Line 39: Define the “simplifying assumptions and incomplete information.” One or two examples would make the research motivation more specific.

Lines 48–49 and line 157: The references appear to use inconsistent formatting. Check the citation format throughout the manuscript.

Line 96: Capitalize the figure reference as “Figure 1.”

Figure 1: Expand the caption to describe the main datasets, processing components, and outputs. Readers should be able to understand the workflow from the figure and caption.

Lines 114–116: Define NUTS level 3.

Line 120: Add a web address or citation for the phrase “documentation available online.”

Lines 121–122: Insert spaces between values and units, including “km” and “km².” Check unit formatting throughout the manuscript.

Lines 155–165: Explain why the population and land-cover datasets represent different years. Discuss how this mismatch may affect the exposure and damage estimates.

Lines 166–167: Specify the spatial resolutions of NUTS 2 and NUTS 3.

Line 214: Specify the reference point for the radius.

Line 216: Define convex hull.

Figure 2: The labels describe processing stages rather than time steps. Replace “time steps” with “processing steps” or “workflow stages” to avoid confusion with the temporal resolution of the flood data.

Lines 237–239: The sentence combines several points. Divide it into shorter sentences.

Line 377: Introduce the No Protections scenario and the rationale for comparing it with the two protection datasets in the Methods before presenting the results.

Figures 6 and 14: Increase the font size of the axis labels and annotations. Check all other figures as well.
Citation: https://doi.org/10.5194/egusphere-2026-2807-RC2

Lorenzo Scarpellini, Andrea Ficchì, Claudia D'Angelo, Andrea Betterle, Peter Salamon, and Andrea Castelletti

Data sets

[Dataset] Mapping uncertainty in flood impacts: a multi-dataset assessment and catalog of simulated, remote sensed and reported floods in Europe Lorenzo Scarpellini, Andrea Ficchì, Claudia D'Angelo, Andrea Betterle, Peter Salamon, and Andrea Castelletti https://doi.org/10.5281/zenodo.19666073

Model code and software

Interactive computing environment

Lorenzo Scarpellini, Andrea Ficchì, Claudia D'Angelo, Andrea Betterle, Peter Salamon, and Andrea Castelletti

Viewed

Total article views: 199 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
108	77	14	199	10	9

HTML: 108
PDF: 77
XML: 14
Total: 199
BibTeX: 10
EndNote: 9

Views and downloads (calculated since 01 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	67	26	8	101
Jul 2026	35	37	5	77
Aug 2026	6	14	1	21

Cumulative views and downloads (calculated since 01 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	67	26	8	101
Jul 2026	35	37	5	77
Aug 2026	6	14	1	21

Viewed (geographical distribution)

Total article views: 174 (including HTML, PDF, and XML) Thereof 174 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 24 Aug 2026

Short summary

This study shows that flood protection assumptions strongly affect risk estimates, and that flood model simulations, satellite observations, and historical records often disagree. Simulations overestimate large floods, satellites miss small ones, and historical records favour populated areas. No single data source provides a complete picture, highlighting the need to combine multiple datasets for more reliable continental-scale flood risk assessments.


Total:	0
HTML:	0
PDF:	0
XML:	0