CEMS-Floods has generated a global and pan-European 3 arc second (~90 m) resolution dataset of flood inundation for different return period scenarios. These datasets are used to produce the Rapid Flood Mapping layer from the EFAS and GloFAS forecasts from EFAS v5 and GloFAS v4.

The maps have been generated for the following return periods: 10, 20, 50, 75, 100, 200, 500 years.

The data in these maps represents the flooding from river processes for rivers greater than 500 km2 (GloFAS version 4 onwards) and 150 km2 (EFAS from version 5 onwards). The data do not account for flooding caused by pluvial or coastal processes.

Flood Inundation Map Generation Methodology

The flood inundation maps are created by using the GloFAS and EFAS re-analyses to generate flood event hydrographs for different return period events which are then used as inputs to a 2D hydraulic flood inundation model. More details about the original methodology are given in Alfieri et al., 2014 and Dottori et al., 2017, some modifications to the methodology have been made since these papers, a new manuscript it being prepared to describe these.

Flood Event Hydrograph Generation

Flood event hydrographs refer to the rise and fall of river discharge during an event. These were generated in every grid cell with an upstream area above a predefined threshold (150 km2 from EFAS version 5 and 500 km2 from GloFAS version 4) using historical river discharge data from the EFAS and GloFAS re-analyses. Firstly, in each grid cell a gumbel extreme value distribution was fitted to the maximum annual river discharge values, this was used to compute the river discharge associated with the 10, 20, 50, 75, 100, 200 and 500 year return periods. Next, a flood event hydrograph was generated for each of these return period scenarios, the hydrograph peak equalled the discharge associated with the return period, the rate of the hydrograph rise and fall was defined from the time of concentration (Tc) which was computed using the Giandotti 1934 method (see Grimaldi et al., 2012 for a description):

Tc = (4* SQRT(A) + 1.5L) / 0.8*SQRT(H)

A = upstream area (km2), L = length of the main river channel (km), H = difference between the mean basin elevation and the elevation at the grid cell being analysed (m)

These parameters for the Tc calculation were obtained from the MERIT-Hydro Digital Elevation Model (DEM) (Yamazaki et al., 2019) which has been resampled to the resolution of EFAS (1 arc minute from version 5) and GloFAS (3 arc minute from version 4). The time between the start of the hydrograph and the peak was set to the maximum of either Tc or 5 days, this allowed sufficient time for the water to pass to the downstream end of the 2D hydraulic simulation domain. The time between the hydrograph peak and the end of the simulation was set to the maximum of 2*Tc or 10 days, the longer time for the falling limb was to allow sufficient time for the peak discharge to travel downstream within the simulation domain. The flood event hydrographs were used as inputs to a 2D hydraulic inundation model simulation, described in the next step.

Figure 2. Procedure to generate flood event hydrographs at each grid cell using the EFAS/GloFAS historical re-analysis.

Flood Inundation Simulation

The LISFLOOD-FP 2D hydraulic inundation model (Bates & De Roo, 2000) was used to simulate flood inundation for EFAS and from version 4 for GloFAS (previously CA2D was used). As well as the flood event hydrographs generated in the previous step, LISFLOOD-FP required the following inputs:

  • Topographic data for the floodplain: obtained from the MERIT-Hydro DEM (Yamazaki et al., 2019) at 3 arc second spatial resolution (~90 m)
  • Floodplain roughness defined by the Manning's n parameter: these were defined from the Copernicus global landcover dataset version 3 at ~90 m resolution (Buchhorn et al., 2020), a lookup table was used to translate each landcover category into a Manning's n parameter value
  • Channel width: obtained from the MERIT-Hydro dataset (Yamazaki et al., 2019) at 3 arc second spatial resolution (~90 m)
  • Channel depth: no global dataset on channel depth was available, therefore 0 m channel depth was used. To compensate for this, the discharge associated with the 1.5 year return period, which represents the bankfull discharge, was subtracted from each of the flood event hydrographs so that they represented the floodplain flow only. 

LISFLOOD-FP simulations were carried out every 9 km (GloFAS) and 5 km (EFAS) along the channel network defined by the 3 arc second MERIT-Hydro DEM, for grid cells with an upstream area greater than 500 km2 (GloFAS) and 150 km2 (EFAS). At each simulation location, the size of the domain was set to be a maximum of either 18 km or twice the maximum distance from the channel network to the edge of the floodplain defined from Nardi et al., 2019. However, in rivers with a very low slope (<10-6 m/m) it was necessary to define a larger domain in order to account for backwater conditions, in this case a domain size of 100 km was set. These domain sizes were applied to the downstream and left and right bank directions relative to the location of the inflow.

Downstream boundary conditions were also defined in the simulation, which allowed water to exit the simulation domain. These were defined at the downstream, left and right bank boundaries from the location of the inflow flood event hydrograph. A free water surface slope boundary condition was applied, this was calculated as the average channel slope between the location of the inflow the channel elevation 45 km downstream, however if this slope was less than 1 x10-4 m/m, it was recomputed along a greater distance of 135 km.

After each simulation, a series of quality checks are performed:

  • Volume errors are less then 1%
  • Maximum outflow > 0
  • The outflow is decreasing by the end of the simulation

If one or more of these checks is not satisfied, the simulation is re-run, for example with a larger initial timestep to prevent large volume errors. The checks are performed on the results of the re-run simulation, if any of the checks are still not satisfied, the simulation is re-run for a maximum of another two attempts, if after this the checks are not satisfied the simulation is discarded and a note of this is stored in a logfile.

From each simulation, the maximum flood depth in each grid cell over the entire simulation period was saved (see figure below for an example).

Flood depth (m)

Figure 3. Example output of the maximum simulated flood depth from the LISFLOOD-FP hydraulic model.

Creation of a Single Global/Pan-European Flood Inundation Dataset

The above procedure resulted in many individual small domains with the simulated maximum flood depth. For each return period (10, 20, 50, 75, 100, 200 and 500 years) all the smaller domains were mosaicked together to create a single gridded dataset. Where multiple smaller domains overlapped each other, the maximum depth value from the overlapping datasets was selected. These mosaicked datasets will soon be made available for download through the JRC website, the previous versions of the datasets are already available to download.

Global 'Flood hazard 100 year return period' layer on GloFAS webviewer

The mosaicked flood inundation data for the 100 year return period scenario is available to view on the GloFAS website as a static layer, it shows the simulated flood extent and depth associated with the 100 year return period scenario. The layer presents the flood depths in one of four categories and has a separate category for permanent water bodies.

Flood depth (m)

Permanent water bodies are defined as:

  • Being defined as 'Open sea' or 'Permanent water bodies' in the Copernicus Global land cover version 3 dataset (Buchhorn et al., 2020)
  • Having an upstream area >= 500 km2 according to the MERIT-Hydro DEM (Yamazaki et al., 2019). Values exceeding this threshold were assumed to be river channels.


Alfieri, L., Salamon, P., Bianchi, A., Neal, J., Bates, P. and Feyen, L. 2014. Advances in pan-European flood hazard mapping. Hydrological Processes, 28: 4067-4077. https://doi.org/10.1002/hyp.9947

Bates, P.D. & De Roo, A.P.J. 2000. A simple raster-based model for flood inundation simulation. Journal of Hydrology, 236(1-2), pp 54-77. https://doi.org/10.1016/S0022-1694(00)00278-X

Buchhorn, M.; Smets, B.; Bertels, L.; De Roo, B.; Lesiv, M.; Tsendbazar, N.E., Linlin, L., Tarko, A. 2020. Copernicus Global Land Service: Land Cover 100m: Version 3 Globe 2015-2019: Product User Manual; Zenodo, Geneve, Switzerland, September 2020; https://doi.org/10.5281/zenodo.3938963

Dottori, F., Kalas, M., Salamon, P., Bianchi, A., Alfieri, L., and Feyen, L. 2017. An operational procedure for rapid flood risk assessment in Europe, Nat. Hazards Earth Syst. Sci., 17, 1111–1126, https://doi.org/10.5194/nhess-17-1111-2017 

Grimaldi, S., Petroselli, A., Tauro, F. and Porfiri, M., 2012. Time of concentration: a paradox in modern hydrology. Hydrological Sciences Journal, 57 (2), 217–228. https://doi.org/10.1080/02626667.2011.644244

Nardi, F., Annis, A., Di Baldassarre, G., Vivioni, E.R., Grimaldi, S. 2019. GFPLAIN250m, a global high-resolution dataset of Earth’s floodplains. Sci Data 6, 180309. https://doi.org/10.1038/sdata.2018.309

Yamazaki, D.Ikeshima, D.Sosa, J.Bates, P. D.Allen, G. H., & Pavelsky, T. M. 2019MERIT Hydro: a high-resolution global hydrography map based on latest topography datasetWater Resources Research555053– 5073https://doi.org/10.1029/2019WR024873