Please note that this product is experimental and is not operationally supported, meaning that there may be some gaps in the availability on the EFAS web interface due to outages.
The radar-based urban flash flood hazard forecast was designed to highlight the risk of flash flooding within urban areas. Specifically it highlights urban ~1 km grid cells in each 1-hour time step within the next 5 hours where there is a risk of flash flooding. The forecasts have a maximum lead time of 5 hours and are updated every hour using nowcasts of rainfall derived from the EUMETNET OPERA radar rainfall composite, no blending is done with other sources rainfall information, therefore this product is only available within areas of radar coverage, as shown by the Radar Coverage product. Urban grid cells with a risk of flash flooding are categorised as being low (yellow), medium (orange) or high (red) hazard level. An example of the layer is shown in Figure 1.
This layer was originally developed by UPC (Polytechnic University of Catalunya) during the EDERA project.
|
|
Figure 1. Example of the radar-based urban flash flood hazard product at 23:00 UTC on the 10th February 2025 in Belgium.
Methodology
The input data for this product are 1) deterministic nowcasts of the hourly rainfall rate generated on a 1 km grid with up to 5 hours lead time, 2) land cover data to define urban grid cells, and 3) climatological intensity-duration-frequency thresholds in each 1 km urban grid cell.
Deterministic nowcasts of hourly rainfall rate are generated using observations from the pan-European EUMETNET OPERA radar network. A method called SBMcast, which uses Lagrangian extrapolation, (Berenguer et al., 2011) creates 15-minute nowcasts of rainfall which extend up to a lead time of 5 hours ahead.
Urban grid cells were defined where 20% or more of each 1 km grid cell was covered by urban or highly impervious land surfaces according to the CORINE 100 m land cover dataset.
The climatological data, provided by UPC, were generated from OPERA rainfall composite data available every 15-minutes from January 2013 – December 2022. At each 15-minute time step, the hourly rainfall rate in each grid cell was calculated and bias corrected by multiplying a correction factor. The correction factor was computed by accumulating the total precipitation measured by the OPERA composite over the 30 days prior to the time step and dividing it by the total accumulated precipitation over the previous 30 days measured by rain gauges from the European Meteorological Observation dataset (EMO-1), which had been interpolated onto the same grid as the OPERA composite. An extreme value distribution was then fitted to the 13-year time series of bias corrected radar rainfall data to compute the hourly rainfall accumulations associated with the 2, 5 and 20 year return period intervals. These values were interpolated from the 2 km OPERA grid onto a 1 km grid using a nearest neighbour method.
The severity of flash flooding in any urban grid cell at a given time step was calculated by comparing the hourly accumulated rainfall from the nowcast, in each 15-minute forecast time step, against the rainfall accumulations associated with the 2, 5 and 20 year return period thresholds (Figure 2). The 15-minute data are aggregated to 1 hour time steps by taking the maximum urban flash flood hazard level in each grid cell during each hour.
Figure 2. The definitions of the three urban flash flood hazard levels according to the forecast rainfall exceedance of the 2, 5 and 20 year return period (RP) thresholds. For example, if the forecasted rainfall accumulation was equivalent to a 10 year return period, this would be categorised as a Medium urban flash flood hazard level.