The first choice in Figure 2 is which water/sea surface level to use. There are several relevant indicators for the height of the sea.

A systematic analysis by Hauer et al. (2020) shows that common sea-level indicators include (from low to high): Mean Sea Level (MSL), Mean High High Water (MHHW), or more general High Sea Level (HSL), a water level with a specific annual return period (e.g., 100 years, see Section 3.1.3 for a detailed discussion), and a fixed level of 10 m referred to as LECZ (originating from McGranahan et al., 2007).

In meso- and macro-tidal areas, tidal high water is an important precondition for flooding. That is why taking into account tide is important to evaluate the flood proneness. We refer to the sea-level including tides as HSL. There are different HSL’s. Common in the context of coastal flooding are Mean High High Water (MHHW) and Highest Astronomical Tide (HAT).

MHHW is the average height of the highest tide recorded each day over a certain observed period. The Highest Astronomical Tide (HAT) is the highest predicted astronomical tide expected to occur (in a 18.6 years cycle), which can be significantly higher than MHHW due to the amplitude magnification of the nodal tide (Baart et al., 2012).

In areas with a non-reflective gentle sloped coast the combination of storm surge and wave run-up is also an important factor for flooding. Including storm surge levels (without waves) can be referred to as Storm Surge Level (SSL). This typically includes both the inverse barometer effect (low pressure at the coast and high pressure offshore pushing the water towards the coast) and storm surges due to the wind forcing water towards the coast. Since 2020s studies (Vousdoukas et al., 2020b; Kirezci et al., 2020) have opted also to include waves. This can include wave set-up and wave run-up (see discussion Melet et al., 2018; Aucan et al., 2019). The sum of tides, storm surge and wave setup is commonly referred to as the eXtreme Sea Level (XSL). Both for surge and wave height one needs to define which surge/wave height to take into account. This is addressed in Section 3.1.3.

The motivations for selecting between different water levels can vary. Kulp and Strauss (2019) use MHHW and add a water height of 2 m as a representative water level corresponding to a “bad flood in the nearer term or an extreme sea-level scenario for 2,100”. Kirezci et al. (2020) aim to quantify the relative importance of tidal variations to potential episodic coastal flooding by 2,100. To do that they sum up surge, wave setup, and sea-level rise with event probabilities of 1 100 per year.

Topography selection is crucial for coastal flooding studies. A topographic dataset is referred to as a Digital Elevation Model (DEM). Two relevant subtypes are Digital Surface Model (DSM), which includes objects such as trees, buildings, and cars, and Digital Terrain Model (DTM), where these objects are filtered out (Hirt, 2014; Guth et al., 2021). Estimating bathymetry, sometimes called Digital Bathymetric Model (DBM), is also important for calculating how much energy reaches the coast. The intertidal bathymetry often combines topographic and bathymetric information. Many bathymetric survey efforts are part of the General Bathymetric Chart of the Oceans (GEBCO) initiative (Hall, 2006). This section focuses on the terrain aspect of elevation modelling, where there is less convergence.

The comparison of two cases with similar water levels but different flooding levels, shown in Figure 1, illustrates that next to sea surface height and topography, ignoring features that impede the influx of water can have a significant impact on the end result. The scenario based on CoastalDEM (Kulp and Strauss, 2019; Central, 2024) did not include coastal protection measures. In countries with robust coastal protection policies (clear policy targets, strong accompanying funding arrangements, effective maintenance organisation and emergency response), low-lying areas can still have very high levels of protection. These protection levels may be the result of hard infrastructural measures, like dikes, dams, storm surge barriers, etc., or well-maintained natural protective features, such as dune areas, tidal flats, salt marshes, mangroves, reefs, etc.

Natural dunes defend a large part of the Dutch coast. Dunes are “soft” coastal protections. Den Heijer et al. (2012) showed that the Dutch dune coast mostly offers very high levels of protection. The weak sections of the coast have since been upgraded. An intricate system of dike rings provides further protection and compartmentalisation of the low-lying areas, reducing the risk even further (see for example, Eijgenraam, 2007). The peaks of coastal dunes do not show up in digital terrain models, because of the limited resolution.

The Dutch coast is also protected by “hard” coastal protection. Examples include the new hard-fixed sea wall on the Afsluitdijk and the parking garage in a dune near Katwijk that provides extra coastal protection (Al, 2022). Breakwaters near Rotterdam and IJmuiden shield the inlets from the waves. Some of these coastal protections can be seen in the topography or bathymetry, but only with sufficient resolution. Some hard protections are movable. A very visible, but only during extreme water levels, example is the Maeslantkering (Mooyaart and Jonkman, 2017). Other countries have also used movable coastal defense structures, the MOSE barriers at Venice Italy (delle Acque, 1997), the Thames barrier (Kendrick, 1988), and the concept of the Ike Dike (Torres et al., 2017) are examples. These storm surge barriers provide temporary protection during high water, but this function is not directly discernible from a DTM. Smaller scale, movable protection measures include coupures in levees and openings that can be closed by a gate or by placing segments when high water is imminent.

These two types of protection measures (soft, hard) in combination with dynamic flooding routines make all the difference in the area considered actually prone to flooding a country like the Netherlands. When you do include these it also makes sense to include the potential failure of these defense systems. Hurricane Katrina showed that dike systems are only as strong as their weakest link (Jonkman et al., 2009). Despite the presence of dyke systems in the US, and the availability of a highly skilled organisation like the United States Army Corps of Engineers, an extreme event like a hurricane still caused a large part of the region to be flooded. In the aftermath, coastal protection policies were reviewed and strengthened. There are many studies that focus on the potential of protection using structures, such as sea walls and dikes (Hinkel et al., 2014; Tiggeloven et al., 2020) and on using nature-based solutions (Beck et al., 2018; Menéndez et al., 2020; Van Coppenolle and Temmerman, 2020; de Vries et al., 2021; van Zelst et al., 2021).

Some coastal flood risk models are specifically designed to support climate adaptation decisions. These studies often compute the needed disaster risk reduction as an economic decision problem [as proposed by van Dantzig (1956)]. Due to the limited empirical data on actual implemented defence and adaptation (Hinkel et al., 2014), assuming a ‘protection level’ is a common approach. Lincke and Hinkel (2018) give an overview of where coastal adaption is a robust investment. Using a similar approach Vousdoukas et al. (2020a) show that economically efficient adaptation can reduce 83% of the coastal flooding.

For flooding assessment a range of methods is available, ranging from pure Geographic Information System (GIS) operations to highly detailed numerical flood modelling. The earliest global analysis of flood risk (Nicholls et al., 1999) used a simple approach in 192 coastal zones globally. They assumed a coastal plane with a constant slope per assessment zone, and used that to calculate inundation. The development of global covering high-resolution topography and bathymetry (as discussed in Section 3.2) allows for a wider repertoire of flooding methods (Vousdoukas et al., 2016b).

The simplest approach to flood models are the so-called bathtub models (Williams and Lück-Vogel, 2020). They are also known as static, equilibrium or planar surface projection models and are sometimes referred to as “GIS-based”. In this context, GIS based refers to that the maps can be made using standard GIS software (for example, using GDAL, SAGA in QGIS, Google Earth Engine, or ArcGIS). These methods are based on spatial operations on vector or raster maps. The term static refers to that the flooding occurs instantaneously, and the duration is not taken into account. The advantage of these models is that they do not take dependencies between neighbouring areas into account and scale up very well, as they are suitable for map-reduce operations commonly used in the cloud. They can be considered worst-case scenarios.

The bathtub approach can also include more information on the local roughness. An example of including roughness to limit flooding is given by Vafeidis et al. (2019); Ward et al. (2020). They artificially steepened the DEM land slopes to mimic roughness that limits water fluxes. The land use determined by satellite data is used to make a spatially varying change to the local slope. The Euclidean distance of each DEM point to the sea is shortened with this factor. After this mapping, calculating the flood exposure is in the GIS domain again. They report 44% less inundation exposure, using the method sometimes referred to as the “sloped coast” method.

Because the pure “bathtub” approach does not consider connectivity between adjacent cells, a cell can flood even if not connected to the flood source. In this approach, polders and other low-lying areas near the coast get flooded. Some studies correct this by removing unconnected flood cells in a post-processing step (Muis et al., 2016), which brings us to the following approach.

To overcome the issue that water floods the area behind coastal protection (as seen in the two lower left figures in Figure 6) methods that we refer to as “connectivity” methods can be used, sometimes referred to as semi-dynamic. An example application was presented by Poulter and Halpin (2008). They compared three flooding methods, all based on geometric operations on a raster. The static bathtub method, in the connectivity world referred to as the zero-side rule, was compared to two methods that can take into account obstacles. Poulter and Halpin (2008) considered the 4-sided and 8-sided rule. In these connectivity methods, a cell can only flood if the extreme water level can reach the cell via the 4 lateral sides or also via its diagonal connections. The latter methods are said to take care of hydrological connectivity. This calculation is harder to split up as subdomains need to communicate to their neighbour if their boundary is flooded (see Breilh et al., 2013; Williams and Lück-Vogel, 2020, for discussion).

A downside of both these static methods is that they can lead to an overestimation of the flooded area of up to 200% (Vousdoukas et al., 2016b), see also Ramirez et al. (2016) for a broader discussion. Figure 6 also makes this clear. Water reaches areas that are not flood-prone (top left two). The connectivity (middle bottom) result can be reached. Here, the problem is that this ‘flood fill’ method does not take into account the limited duration of a storm and high tide. If it finds a ‘hole’ in the coastal defence, it will instantly fill the entire flood-prone area, while in reality, the peak of the event under which flooding occurs might be over in 6 hours, and to be able to reach far inland, water needs to overcome roughness (vegetation, buildings).

An example of a simple approach to including the concept of time is given by Dottori et al. (2018). They apply the assumption that the discharge of water over a dyke/weir is limited by laws of physics, described by the excess water level compared to the dyke height in an empirical formula. Using the storm duration and its water level variations therein - the design hydrograph - this leads to a cap on the volume that can flood the area. The hydrograph limits flood duration, and so the total flood water volume is limited. The second factor leading to attenuation of the flooding is resistance experienced by the flood wave, via bed roughness or obstructions of the flow path. We refer to methods that assume connectivity and limit the flood extent as “convolution” methods, as they tend to require an iteration of a convolution step.

The most realistic approach, but also the most computationally intensive and requiring the most input data, is to use a reduced complexity “physical” models based on hydrodynamics, such as LISFLOOD-FP (Bates et al., 2010). Also promising for global flooding applications is the SFINCS model Leijnse et al. (2021), which is fast, can deal with compound flooding (river + coastal) and also includes wave run-up over dunes or dykes. One can also use traditional 2D overland flooding models, such as Mike 2D overland flooding, 3Di or Delft3D Flexible Mesh. Still, these are harder to set up and require more computation hours because they also contain physics, such as advection, that are less relevant during a flood. New approaches promise to speed up computation time (van den Bout et al., 2023).

More advanced models do exist that also take into account erosion, the possibility of dune and dike failure, and flooding due to groundwater (sunny day flooding). These have been applied on a national scale but not yet on a global scale. There have been recent developments towards machine learning-based methods to compute flood extends (see Jones et al., 2023, for an overview). The creation of flood-specific datasets can help to estimate flood extents without physics-based models and without the need for oversimplification (e.g., Bonafilia et al., 2020).

Despite efforts to establish benchmarks for coastal flooding (Nederhoff et al., 2024; Néelz and Pender, 2013) and pluvial flooding (Aerts et al., 2020), a need remains for a structured, community-driven intercomparison benchmark specifically for coastal flooding. Jafarzadegan et al. (2023) also highlighted the weak and selective validation of flood inundation models and the lack of sufficient validation data. They also discuss the balance between accuracy, reliability versus computationally efficiency.

Finding the balance between applicability, scalability, explainability, computation time, and model setup is a reason why there is not one dominant method. The baththub based approach is also still evolving, Kasmalkar et al. (2024) try to address the overestimation of flooding by baththub models by including hydraulic connectivity and path-based attenuation. In Artificial intelligence (AI) based approaches the connectivity models are also used as one of the input features or as a reference (e.g., Jones et al., 2023).

Flood risk is defined as the product of hazard, exposure, and vulnerability (e.g., Aitsi-Selmi et al., 2015). Following the definitions of United Nations Office for Disaster Risk Reduction (UNDRR) hazard is “A process, phenomenon or human activity that may cause loss of life, injury or other health impacts, property damage, social and economic disruption or environmental degradation.” (Author, 2016). Exposure is “The situation of people, infrastructure, housing, production capacities and other tangible human assets located in hazard-prone areas.”. Vulnerability is “The conditions determined by physical, social, economic and environmental factors or processes which increase the susceptibility of an individual, a community, assets or systems to the impacts of hazards.”.

The hazard in coastal flooding entails the seawater over-flooding the shore, typically quantified with a probability of a certain flood extent. Exposure entails the economic assets and people in that in the flood-prone area. Vulnerability refers to the predisposition of a community to suffer harm when exposed to a hazard, based on aspects like social, economic, and environmental factors that influence the ability to cope with and recover from a flood. The hazard component is a result of the flood computations described in the previous sections. This section focuses on the societal impact of a flood and the result: exposure times vulnerability. De Moel et al. (2015) and Botzen et al. (2019) provide an extended overview of aspects to take into account while assessing exposure and vulnerability. Staupe-Delgado (2019) gives an overview on how the terminology has evolved over time and the challenges around ambiguity.

Flood exposure is typically quantified with metrics of the population, the urban area, or economic assets that are exposed (see Figure 7) to a flood with a certain probability (e.g. Muis et al., 2017). The area at risk is the easiest to determine. It is determined by the results of the flooding model defined in Section 3.4. Areas already underwater (rivers, streams, lakes, reservoirs) should be excluded from the potentially flooded area. Population maps are needed to estimate the population living in flood-prone areas (see for example, Edmonds et al., 2020). Estimating the value that is flood prone is the most laborious. This requires the express cost of the property. Some use a more simple approach by assuming that Gross Domestic Product (GDP) is uniformly distributed over the population. The population maps can then be used as a measure of asset value (e.g., Kirezci et al., 2020). One can also refine the distribution of house prices. For example, properties close to the coast tend to be more valuable, referred to as the “coastal premium” (see e.g., Conroy and Milosch, 2011; Ling, 2021).

Flood vulnerability is a step further. Here one should consider how the flooded areas are affected by a flood. Coastal regions may become saltier after coastal flooding. That “loss of function” type of evaluation is mainly applied on a local scale (e.g., Storlazzi et al., 2015), rarely in global studies. Not all regions that get flooded are vulnerable. Many people choose to live in flood-prone areas because the land is more fertile, it is closer to economic activity (e.g., near ports), et cetera. People tend to be aware of a potential flood and are prepared, thus their vulnerability is low.

A flood can also have indirect consequences. The consequences of the consequences, such as the Fukushima disaster (Hollnagel and Fujita, 2013) show us that the indirect effects of a flood can sometimes be more severe than the direct effects. To do this on a global covering scale is not realistic at the moment, because it requires integrated information on complex systems. Rising sea levels and flooding can lead to increased groundwater levels and salinity. This rise in salinity can significantly reduce the suitability of water for agriculture and drinking purposes (as discussed by e.g., Youssef et al., 2021).

The “future” label in the WEPFIF notation outlines the incorporation of future changes into the coastal flooding workflow. Typically, this process involves duplicating a workflow and adjusting the input parameters to account for various scenarios. The section comprehensively breaks down how to include future scenarios in each component.

Most of the technical challenges to generate global hyper-resolution coastal-flooding impact studies have been addressed over the last decade. It is now possible to generate maps with resolutions of 30 m or higher by combining several publicly available datasets. Thus, the new challenge is not only achieving global coverage and high resolution but also ensuring local relevance (see e.g. Gianinazzi, 2018). To be locally actionable, one needs to be able to compare results with local studies. For this, it is important to know which methodological choices have been made.

To facilitate the task of defining and determining the details of these methodological choices, this study presents a workflow with an accompanying semantic coding scheme for coastal flooding applications. This coding scheme can help to do a comparison between global and local studies or between different studies of the same spatial extent. The notation assigns codes to each of the six steps in the workflow of a coastal flood risk calculation: the land elevation (Section 3.2) that will flood (Section 3.4) from the sea (Section 3.1) if not properly protected (Section 3.3), resulting in societal impact (Section 3.5), now or in the future (Section 3.6).

For each of the steps in the workflow methodological choices can be noted using the coding scheme.

Waterlevel

The application of this coding scheme to a collection of coastal studies illustrates how it can be used. Table 4 shows the wide variation of choices that studies make in their methods.

The research objective of each study in Table 4, particularly in relation to quantifying current and future impacts, elucidates the contextual differences in the applied methodologies. Hallegatte et al. (2013) quantify potential future economic loss due to climate change to provide advice on adaption policy. Hinkel et al. (2014) estimate future damages and adaptation costs to support efforts to reduce emissions. Vergouwe (2014) estimates current flood risk in the Netherlands. Muis et al. (2017) estimate the current risk of coastal flooding to assess how much risk could be avoided if flood protection standards were increased. Kulp and Strauss (2019) improve exposure estimates to inform coastal communities that the future is more difficult than previously thought. Kirezci et al. (2020) estimate future episodic floods to show the massive environmental and socio-economic impacts.

The goals of these studies and their typical methodological considerations can be summarized in a general recommendation. In all cases, employing a DTM appears more advantageous than a DSM, as reflected in the preference for a E = T choice.

The state of the world

This overview shows different considerations for the choices based on the different objectives of the global coastal flooding studies. This large variation corresponds to the large variation in methods, which points out how important it is to elaborate and notate these choices but also the corresponding goal clearly.

Not only does the research goal determine the choices made, but also the availability of open and Findable Accessible Interoperable Reusable (FAIR) datasets and computing resources influences which choices can be made within the time available for a study. Unfortunately, many of the tools, models, and model schematizations are still not available under an open-source license. This makes it difficult to build upon the work of others. Examples from the studies above include the CoastalDEM dataset, the tidal model TPXO, the model schematization of the GTSM, and the Dynamic Interactive Vulnerability Assessment (DIVA) impact model; none of these are available under an open-source license. When data is open, it is not always easy to use. For example. For example, the GTSM reanalysis data is available through the Copernicus portal, but it can not be downloaded as a whole. One has to make several small requests to access the data, which violates the FAIR principles.

Physics-based flooding models that can be used for global applications are available, but there is a lack of a public global schematization that can be used. This is why most studies fall back to the bathtub or connectivity methods; they need large amounts of data to set up an alternative approach. Some of the data just has not been collected on a global scale. There is no high resolution ( O ( 30 ) m) global covering reproducible DTM available. There have been some efforts to make a dataset of protection levels, but no dataset that covers hard and soft coastal defence is available. There is no dataset available on current and future protection investment strategies. This makes it impossible to create future scenarios based on the realistic assumption that people will defend their coasts as the sea level rises.

This paper provides an overview of methodological approaches in recent global coastal flooding impact studies. To enhance comparability, it introduces the WEPFIF coding scheme. Applying this scheme, the study demonstrates the reasons for inconsistencies between results. Achieving consistent and explainable results is crucial as we stand on the brink of shifting from theoretical global evaluations to actionable local-level information.

This is the first approach to a coding scheme for coastal flooding studies. It encourages others to improve upon it, create variations, or extend it based on studies not included in this analysis or when new methodological choices are introduced. Like other coding schemes, it will need to adapt as methodology progresses. Scientists are encouraged to extend or simplifythe suggested approach.