Flood risk models are frequently used to analyse the climate- and socio-economic-driven impact of flooding hazards. However, model validation is rarely done adequately due to the rare occurrence of floods and even less frequent reporting of corresponding damages.

In this research, validation is defined as the process of ensuring that a model performs within a range of accuracy and precision, satisfactory for its intended use. To guide experts in their validation efforts, a four-phased framework is developed to validate flood-event damage estimations, created with hazard x exposure x vulnerability models.

The framework was applied to two damage estimates created by the Global Flood Risk Tool (GFRT). 1) For damage caused by the Limburg 2021 river flood (The Netherlands - Europe) and 2), for damage caused by a 2019 hurricane-induced coastal flood in Beira (Mozambique - Africa). For the Limburg case, total direct damage was determined at 349,4 million euro. An initial model overestimation of 34% was caused primarily by a large exaggeration of exposed agricultural surface area, and significant modelling errors of linear infrastructure. Furthermore, an uncertainty range was quantified between 271,8 (-23%) and 388,2 million euro (+11%) due to uncertainty in residential assets (across all three model parameters) and an uncertain exposure parameter of agricultural assets.

To create additional damage estimates for verification, a Structured Expert Judgement (SEJ) experiment was executed with ten flood-damage experts. Due to the high experiment cost and low expert-informativeness, the method is currently not advised as a validation approach. In situations with limited data, experts may still be a relevant information source.

For Beira, damage was determined at 8,1 million US dollar. The model underestimated damage by 82% due to errors in infrastructure, industrial, and commercial assets. Besides, overestimations were found for informal residential- and agricultural assets. The estimate ranges between 5,2 (-36%) and 13,2 million US dollar (+62%). This range excludes uncertainties at port and industrial assets, as insufficient information was available. Contrary to the Limburg case study, insights from the plausibility assessment were too uncertain for quantification, thus the validated estimate is based on damage- and construction cost data. Novel techniques were used to disaggregate the compound damage data, such as comparing wind and flood vulnerability curves and applying employee-based estimations.


The significantly altered damage estimate for both case studies demonstrates the usefulness of the framework. However two main limitations remain: first, lacking information on direct damage to critical infrastructure hinders validation.
Second, additional detail in data is required to allow parameter calibration that increases accuracy across multiple flooding scenarios. Therefore, the main recommendation for future research is to increase the detail in damage data reporting so that parameter calibration is supported. This may be done by increasing the spatial resolution of reported damages or adding additional variables such as inundation depth in reports.