Integrated high-resolution dataset of high-intensity European and Mediterranean flash floods

This paper describes an integrated, high-resolution dataset of hydro-meteorological variables (rainfall and discharge) concerning a number of high-intensity flash floods that occurred in Europe and in the Mediterranean region from 1991 to 2015. This type of dataset is rare in the scientific literature because flash floods are typically poorly observed hydrological extremes. Valuable features of the dataset (hereinafter referred to as the EuroMedeFF database) include (i) its coverage of varied hydro-climatic regions, ranging from Continental Europe through the Mediterranean to Arid climates, (ii) the high space–time resolution radar rainfall estimates, and (iii) the dense spatial sampling of the flood response, by observed hydrographs and/or flood peak estimates from post-flood surveys. Flash floods included in the database are selected based on the limited upstream catchment areas (up to 3000 km2), the limited storm durations (up to 2 days), and the unit peak flood magnitude. The EuroMedeFF database comprises 49 events that occurred in France, Israel, Italy, Romania, Germany and Slovenia, and constitutes a sample of rainfall and flood discharge extremes in different climates. The dataset may be of help to hydrologists as well as other scientific communities because it offers benchmark data for the identification and analysis of the hydro-meteorological causative processes, evaluation of flash flood hydrological models and for hydro-meteorological forecast systems. The dataset also provides a template for the analysis of the space–time variability of flash flood triggering rainfall fields and of the effects of their estimation on the flood response modelling. The dataset is made available to the public with the following DOI: https://doi.org/10.6096/MISTRALS-HyMeX.1493. Published by Copernicus Publications. 1784 W. Amponsah et al.: High-intensity European and Mediterranean flash floods


Introduction
Flash floods are triggered by high-intensity and relatively short-duration (up to 1-2 days) rainfall, often of a spatially confined convective origin (Gaume et al., 2009;Smith and Smith, 2015;Saharia et al., 2017).Due to the relatively small temporal scales, catchment scales impacted by flash floods are generally less than 2000-3000 km 2 in size (Marchi et al., 2010;Braud et al., 2016).Given the large rainfall rates and the rapid concentration of streamflow promoted by the topographic relief, flash floods often shape the upper tail of the flood frequency distribution of small-to medium-size catchments.Understanding the hydro-meteorological processes that control flash flooding is therefore important from both scientific and societal perspectives.On the one hand, elucidating flash flood processes may reveal aspects of flood response that either were unexpected on the basis of less intense rainfall input or that highlight anticipated but previously undocumented characteristics.On the other hand, improved understanding of flash floods is required to better forecast these events and manage the relevant risks (Hardy et al., 2016), because knowledge based on the analysis of moderate floods may be questioned when used for forecasting the response to local extreme storms (Collier, 2007;Yatheendradas et al., 2008).
However, the small spatial and temporal scales of flash floods, relative to the sampling characteristics of typical hydro-meteorological networks, make these events particularly difficult to monitor and document.In most of the cases, the spatial scales of the events are generally much smaller than the sampling potential offered by even supposedly dense raingauge networks (Borga et al., 2008;Amponsah et al., 2016).Similar considerations apply to streamflow monitoring: often the flood responses are simply ungauged.In the few cases where a stream gauge is in place, streamflow monitoring is affected by major limitations.For instance, peak water levels may exceed the range of available direct discharge measurements in rating curves, causing major uncertainties in the conversion of flood stage data to discharge data.In other cases, stream gauges are damaged or even wiped out by the flood current: in these cases, only part of the hydrograph (usually a segment of the rising limb) is recorded.
The call for better observations of flash flood response has stimulated the development of a focused monitoring methodology in the last 15 years over Europe and the Mediterranean region (Gaume et al., 2004;Marchi et al., 2009;Bouilloud et al., 2009;Calianno et al., 2013;Amponsah et al., 2016).This methodology is built on the use of post-flood surveys, where observations of traces left by water and sediments during a flood are combined with accurate topographic river section survey to provide spatially detailed estimates of peak discharges along the stream network.However, the important thing to note here is that the survey needs to capture not only the maxima of peak discharges: less intense responses within the flood-impacted region are important as well.These can be contrasted with the corresponding generating rainfall intensities and depths obtained by weather radar re-analysis, thus permitting identification of the catchment properties controlling the rate-limiting processes (Zanon et al., 2010).The large uncertainty affecting indirect peak discharge estimates may be constrained and reduced by comparison with peak discharges obtained from hydrological models fed with rainfall estimates from weather radar and raingauge data (Amponsah et al., 2016).Post-flood surveys typically start immediately after the event and are carried out in the following weeks and months (Gaume and Borga, 2008), during the so-called Intensive Post-Event Campaigns (IPEC, in the following), before possible obliteration of field evidence from restoration works or subsequent floods.
The aim of this paper is to outline the development of the EuroMedeFF dataset, which organises flash flood hydrometeorological and geographical data from 49 high-intensity flash floods, whose location stretches from the western and central Mediterranean, through the Alps and into Continental Europe.The database includes high-resolution radar rainfall estimates, flood hydrographs and/or flood peak estimates through IPEC, and digital terrain models (DTMs) of the concerned catchments.Collation of the EuroMedeFF dataset is a challenging task (Borga et al., 2014), due (i) to the lack of conventional hydro-meteorological data which characterises these events (owing to the small spatio-temporal scales at which these events occur), and (ii) to the fact that extreme events are, by definition, rare.Collecting rainfall and flood data by means of opportunistic post-flood surveys required the mobilisation of a group of researchers (ranging in size from 5 to more than 20 persons) for an extended period of time (ranging from a few days to some weeks).In addition to this, high-quality weather radar estimates of extreme events such as the ones triggering flash floods are not easy to gather, due to the number of sources of error affecting radar estimation under heavy precipitation and in rough topography environments (Germann et al., 2006;Villarini and Krajevski, 2010).Owing to these reasons, the EuroMedeFF dataset of 49 flash flood events comprising high-quality radar rainfall estimates, flood hydrographs, surveyed flood peaks at ungauged sites, and digital terrain models is simply unprecedented in size in Europe and in the Mediterranean in terms of (i) number of events, (ii) variety of provided data, and (iii) the degree of integration.Given the quality and resolution of the rainfall input, the archive provides unprecedented data to examine the impact of space-time resolution in the modelling of high-intensity flash floods under different climate and environmental controls.Since results from previous modelling studies are quite mixed, much of the knowledge being either site-specific or expressed qualitatively, the availability of the EuroMedeFF data archive may open new avenues to synthesise this knowledge and transfer it to new situations.
The criteria for the EuroMedeFF database development and a summary table and spatial locations of the collected flash floods are presented in Sect. 2. Section 3 describes the components of the flash flood datasets, whereas the methods used to generate the rainfall and discharge datasets are presented in Sect. 4. Section 5 discusses the main features of the dataset, based on climatic regions and the two methodologies for discharge data collection (stream gauges and indirect estimates from post-flood analysis).General remarks on the scientific importance of the EuroMedeFF database are provided in the Conclusions section, whereas a link to the freely accessible EuroMedeFF database is provided in the Data Availability section.

Criteria for EuroMedeFF database development
The EuroMedeFF database includes data from high-intensity flash flood events from different hydro-climatic regions in the Euro-Mediterranean area.To be included in the dataset, the following data availability was ensured: (i) digital terrain model (DTM) of resolutions 5-90 m of the impacted region/catchment; (ii) weather radar rainfall estimation with high spatial and temporal resolutions, and (iii) discharge data from stream gauges and/or post-flood analyses.Rainfall data are provided at a time resolution of 60 min or less and as "best available rainfall products" (i.e.estimates which include the merging of radar and raingauge estimates).
Three criteria have been considered for the development of the EuroMedeFF database.
i. Flood magnitude.A unit peak discharge of 0.5 m 3 s −1 km −2 (this parameter is termed F th ) is considered as the lowest value for defining a flash flood event.This means that, for an event to be included in the database, at least one measured flood peak should exceed the value of F th .The authors are aware that, depending on climate and catchment size, a unit peak discharge of 0.5 m 3 s −1 km −2 can correspond to a severe flash flood (for instance, in the inner sector of the alpine range) or a moderate flash flood (for instance, in many Mediterranean basins).A value of 0.5 m 3 s −1 km −2 can be considered as a lower threshold for flash floods across a variety of climates and studies (Gaume et al., 2009;Marchi et al., 2010;Tarolli et al., 2012;Braud et al., 2014).For the sake of simplicity, we adopted the same value of F th in all the studied regions.
Since the identification of the flash floods included in the database is primarily driven by the local observed impact, for most floods the lowest unit peak discharge is much higher than F th .
ii. Spatial extent.The upper limit for a catchment impacted by the flood is 3000 km 2 (this parameter is termed A th ).The same meteorological event may have triggered multiple floods (e.g.September and October 2014 floods in France which have affected several catchments of about 2000 km 2 -Ardèche, Cèze, Gard, and Hérault).In this case, we report several events for the same date, corre-sponding to different specific catchments with areas less than A th .
iii.Storm duration.The upper limit for the duration of the flood-triggering storm is up to 48 h (this parameter is termed D th ).The rainfall duration is identified by defining a minimum period duration with basin-averaged hourly rainfall intensity less than 1 mm h −1 over the impacted catchment to separate the time series in consistent events.The methodology is similar to Marchi et al. (2010) and Tarolli et al. (2012), where the duration is defined as "the time duration of the flood-generating rainfall episodes which are separated by less than 6 h of rainfall hiatus".We made this threshold explicit to reduce subjectivity.Here, the minimum duration depends subjectively on hydro-climatic settings and basin size.
The reported D th is the duration of the rainfall responsible for each event flood peak, separated from other rainfall events that may have occurred before or after the main event depending on the characteristics of the largest involved catchment.In a number of cases in which the features of the flash flood response were specifically affected by wet initial soil moisture conditions, rainfall data are provided for a longer period than the storm duration.This enables us to account for antecedent rainfall in the analyses.
In general, the preliminary selection of flash floods was based on rainfall data (amount, intensity) from meteorological agencies and qualitative field recognition of flood response.This led to the exclusion of a number of low-intensity events.Post-flood reconstruction of peak discharge was carried out for events that passed this preliminary screening.Several of these events were not included in the dataset because they failed to meet the requirements in terms of flood magnitude, spatial extent and storm duration.Given these constraints, the EuroMedeFF database includes 49 highintensity flash floods: 30 events in France, 7 events each in Israel and in Italy, 3 events in Romania, and 1 event each in Germany and in Slovenia.
Figure 1 shows the location of the basins impacted by the flash floods included in the data archive and provides information on the basic features, such as timing of occurrence over the year, size of the largest affected river basin and highest unit peak discharge.The figure shows that the timing of the floods varies gradually from the south-west, where the floods occur mainly in the September to November season, to the east, where the floods occur mainly in the period from autumn to late spring.The shift in seasonality is paralleled by a decreasing basin size and unit peak discharge from south-west to east.These findings are supported by the work of Parajka et al. (2010), who analysed the differences in the long-term regimes of extreme precipitation and floods across the Alpine-Carpathian range, and of Dayan et al. (2015), who analysed the seasonality signal of atmospheric deep convection in the Mediterranean area.Table 1 reports summary information of the EuroMedeFF database.In the table, each event is labelled as an "EventID", which comprises the impacted catchment/region and the year of occurrence, e.g.ORBIEL1999 (cf.event 1 in Table 1).The "EventID" is used in the archive to uniquely identify the event.The table is ordered first on a country basis, followed by the date of flood peak for each country, from past to most recent events.For each of the 49 events, the table reports the river basin and the country, the date of the flood peak, the climatic region, the number of river sections for which discharge data are available (in terms of both indirect post-flood estimates and streamgauge-based data), with indications of the sections with streamgauge information, the range of basin area for the catchments closed at the studied river sections, the storm duration, the range of unit peak discharges and the indication of earlier works on the event.In a few cases, more than one flash flood event is reported for the same river basin.
We used the Budyko diagram (Budyko, 1974) to characterise the climatic context of the catchments included in the EuroMedeFF database (Fig. 2).The Budyko framework plots the evaporation index (i.e. the ratio of mean annual actual evaporation to mean annual precipitation, AET / P ) versus the aridity index (i.e. the ratio of mean annual potential evapotranspiration to mean annual precipitation PET / P ).The mean values of these variables were calculated for each river basin, so the number of points plotted in Fig. 2 is smaller than the total number of flash floods in the database.Figure 2 also reports the empirical Budyko curve (dotted curve; Budyko, 1974), which fits well with the upper envelope (continuous curve) of the data included in the data archive.Not sur-   The EuroMedeFF dataset consists of high-resolution data on rainfall, discharge, and topography.The information in the data archive is categorised into three main groups: generic, spatial, and discharge data.

Generic data
The "Readme" text file contains generic data on the date of the flash flood occurrence, the name of the impacted catchment and the country and administrative region of the catchment.Detailed generic information on the spatial data (DTM and radar) and discharge data (flood hydrographs and IPECs) are also elaborated in the file.Also, the coordinate systems and grid sizes of the spatial data, and the time resolutions and reference of the radar and flood hydrographs, are summarised.

Spatial data
i. Topographic data.Digital terrain model (DTM) with a grid size of 5-90 m.For each event, DTM data are provided in compressed ASCII raster files, with label "EventID_DTMXX", where XX is the grid size in metres.The DTM is provided in the local country coordinate system, with a file (DT-MXX_WGS84_LowLeft_corner) reporting the coordinates of the lower left corner in the WGS84 coordinate system.All the data relative to one country are in the same coordinate system.
ii. Radar rainfall data.Corrected and raingauge-adjusted radar rainfall data are provided with a 1 km or less grid size and temporal resolution appropriate for the flood (typically 60 min or less).For each event, radar data are provided in compressed ASCII raster files, with label "EventID_RADAR".Radar data are provided, consistent with the DTM data, in the local country coordinate system with a file (Radar_WGS84_LowLeft_corner) reporting the coordinates of the lower-left corner in the WGS84 coordinate system.At least, all the data relative to one country are in the same coordinate system.The time reference for the radar data is provided as yymmddH b M b − yymmddH e M e , with H b , M b referring to the beginning and H e , M e to the end of the considered time period.
The spatial data (DTM and radar) are provided in ASCII format.The coordinates for radar and DTM data as well as locations of streamgauge and IPEC sections are consistently provided in both local (country-specific) and WGS84 systems.The main advantage of WGS84 is that it avoids possible conversion problems from local coordinate systems while providing a homogeneous coordinate system throughout the database.

Discharge data
i. Flood hydrographs.For each event, the location of the available streamgauge stations, upstream area of the basin draining to the station and observed hydrographs are provided in the Excel file "Even-tID_HYDROGRAPHS".The coordinates are consistent with the local country coordinate system given for the spatial data, and are also provided in the WGS84 coordinate system.The time reference system for the hydrograph data are consistent with that used for the radar data.
ii. Post-flood data.Comprehensive data on post-flood surveys through IPEC are provided in the excel file "Even-tID_IPEC".For each section, the location of the surveyed cross section, the area of the basin, the indirect estimation method used and peak discharge estimates are provided.When possible, the following further parameters are reported: flood peak time, wet area, slope, roughness parameter, mean flow velocity, Froude number, geomorphic impacts (in three classes - Marchi et al., 2016), and the estimated peak discharge uncertainty range (Amponsah et al., 2016).Coordinates of the surveyed sections are consistent with the local country coordinate system given for the spatial data, and are also provided in the WGS84 coordinate system.

Rainfall estimation methods
Raw radar data were provided by several sources and elaborated following different procedures depending on the quality and type of available radar and raingauge data, in order to obtain the best spatially distributed precipitation estimate for each event.In general, original reflectivity data in polar coordinates have been used as raw radar data.A set of correction procedures, taking into account the highly nonlinear physics of radar detection of precipitation, and procedures for the raingauge-based adjustment, were used.The procedures include the correction of errors due to antenna pointing, ground echoes, partial beam blockage, beam attenuation in heavy rain, vertical profile of reflectivity and wet radome attenuation, and a two-step bias adjustment that considers the range-dependent bias at yearly scale and the mean field bias at the single event scale.Radar and raingauge rainfall estimates were merged using the same procedure: a mean field bias calculated at the event accumulation scale using rain gauges located in or around the study catchment.Additional details on the procedures can be found in Bouilloud et al. (2010), Delrieu et al. (2014), Marra et al. (2014), Marra and Morin (2015), Boudevillain et al. (2016), and in the references therein.
Earth Syst.Sci.Data, 10, 1783-1794, 2018 www.earth-syst-sci-data.net/10/1783/2018/ For French events 7, 26 and 30 in Table 1, only rainfall data from one local rain gauge are available.These floods have been kept in the database because of the interest in including flood response data for very small basins (< 1 km 2 ) and because the small catchment size of the Valescure basin (4 km 2 ) causes the absence of radar rainfall data to be less detrimental than for floods that hit larger catchments.Note that the available rain gauge is located within the considered 4 km 2 basin.In addition, as the radar closest was quite far from the catchment, located in a zone with complex topography, radar data accuracy was not guaranteed.

Discharge estimation methods
Discharge data in the EuroMedeFF database derive from both streamflow monitoring stations and post-flood indirect estimates of flow peak through IPEC.Streamflow data, permitting recording of flood hydrographs, thus enabling assessment of not only discharge, but also time response and flood runoff volume estimation, were checked for the uncertainties affecting rating curves at high-flood stages by using hydraulic models and topographic data.Discharge data from reservoir operations, water levels and use of the continuity equation, when available, were also included in the database after accurate quality control.
Different methods have been used for the indirect reconstruction of flow velocity and peak discharge from flood marks, such as slope area, slope conveyance, flow-throughculvert, and lateral super-elevation in bends.Amongst these methods, the most commonly used for the implementation of the dataset presented in this paper is the slope conveyance, which consists of the application of the Manning-Strickler equation, under assumption of uniform flow, and requires the topographic survey of cross-section geometry and flow energy gradient, computed from the elevation difference be-tween the high water marks along the channel reach surveyed (Gaume and Borga, 2008;Lumbroso and Gaume, 2012).
Although the identification of river cross sections suitable for indirect peak discharge assessment has sometimes proved not easy (flood marks can be hardly visible or obliterated by post-flood restoration works), and discharge reconstruction in cross sections that underwent major topographic changes is affected by major uncertainties (Amponsah et al., 2016), an appropriate choice of the cross sections permitted us to achieve a spatially distributed representation of flood response for most studied events.Specific details on the IPEC procedures can be found in the references provided in Table 1.

Discussion
Overall, 680 peak discharge data are included in the archive: 32 % (219) were recorded by river gauging stations or based on data from reservoir operations, and 68 % (461) from IPEC surveys.We followed the geomorphic impact-based linear error analysis of the slope conveyance discharge determination presented in Amponsah et al. (2016) for the uncertainty assessment of the IPEC peak flood estimates.Table 2 reports the number of river sections for each of the climatic regions and the corresponding summary statistics of the upstream drainage area.Almost 90 % of the included discharge data are from the Mediterranean region, which is consistent with increasing collation and analysis of flash flood data in this region compared to other climatic regions in Europe (e.g.Gaume et al., 2009;Marchi et al., 2010).The area of the basins included in the archive ranges from 0.27 to 2586 km 2 .Table 2 shows that flash flooding may impact larger basins in the Mediterranean, Alpine and Arid regions than those considered in the Inland Continental region.This supports earlier findings from Gaume et al. (2009).Table 3 reports summary statistics of the upstream drainage area for the two discharge assessment methods (stream gauges and indirect methods).As expected, stream gauges correspond to larger areas, whereas post-flood surveys play major roles in documenting peak discharges for smaller drainage areas (Borga et al., 2008;Marchi et al., 2010;Amponsah et al., 2016).Nevertheless, the database also includes discharge data from a few measuring stations deployed in small research catchments.This allows reduction of the uncertainty related to the estimation of peak discharge in very small catchments (Braud et al., 2014).
The relationship between the unit peak discharge (i.e.peak discharge normalised by the upstream drainage area) and the upstream area was investigated for the EuroMedeFF database to identify the control exerted by catchment size on flood peaks (Fig. 3) and to analyse its variation among the four main climatic regions (Fig. 4a-d).Not surprisingly, the unit peak discharges exhibit a marked dependence on watershed area.The envelope curve, representing the observed upper limit of the relationship, was empirically derived as a powerlaw function for all the floods as well as for the four different main climate regions.The envelope curve representative of all the floods is similar in shape to that reported by Gaume et al. (2009) and Marchi et al. (2010) in previous analyses in the same hydro-climatic context.However, the multiplier reported here is larger than that reported in earlier analyses, due to the inclusion of recent more intense cases documented in large catchments.Inspection of the multiplier and exponent coefficients of the envelope curves reveals that the same exponent provides a good fit for the different climatic regions, whereas the highest multiplier is reported for the Mediterranean region, with an intermediate value for the Alpine-Mediterranean and Alpine basins, and the same lowest value for Inland Continental, Arid-Mediterranean and Arid basins.For small basin areas (1 to 5 km 2 ), Mediterranean and Alpine catchments are shown to experience similar extreme peaks.
Figure 5a-b show the relationship between unit peak discharge based on the two discharge assessment methods and watershed area in a log-log diagram, together with the envelope curves.Indirect estimates of peak discharges show similar dependence of unit peak discharge on catchment size to that reported in Fig. 3, showing that the information content of the overall envelope curve is dominated by the flood obtained based on post-flood campaigns.Indeed, peak data from streamgauging stations show a clearly different exponent of the envelope curve (−0.12) when compared to postflood indirect peak flow estimates (and to the ones previously shown in Fig. 3).The highest values of the peak discharge are often missed by the gauging stations because of insufficient density of streamgauge networks and/or damage to the stations during floods.This sampling problem is more severe in small basins: as a consequence, both the value of the multiplier and the exponent of the envelope equation are lower in Fig. 5a than in the plots that include post-flood peak discharge estimation in ungauged streams (Figs. 3 and 5b).

Conclusions
We presented an observational dataset that provides integrated fine-resolution data for high-intensity flash floods that occurred in Europe and in the Mediterranean region from 1991 to 2015.The dataset is based on a unique collection of rainfall and discharge data (including data from post-flood surveys) for basins ranging in size from 0.27 to 2586 km 2 .The archive provides high-resolution data enabling a number of flash flood analyses.It allows the analysis of the spacetime distribution of causative rainfall, which may be used to investigate methodologies for rainfall downscaling.The data may foster the investigation of the rainfall-runoff relationship at multiple sites within the flash flood environment.This may lead to the identification of possible thresholds in runoff generation which may be related to initial conditions, rainfall rates and accumulations, and catchment properties.Moreover, it allows investigations to clarify the dependence existing between spatial rainfall organisation, basin morphology and runoff response.The archive may be used as a bench-Earth Syst.Sci.Data, 10, 1783Data, 10, -1794Data, 10, , 2018 www.earth-syst-sci-data.net/10/1783/2018/  mark for the assessment of hydrological models and flash flood forecasting procedures in various hydro-climatic settings.The availability of fine-resolution rainfall data may be used to better understand how rainfall spatial and temporal variability must be considered in hydrological models for accurate prediction of flash flood response.Furthermore, the availability of multiple flash flood response data along the river network may be exploited to better understand how calibration of hydrological models may be transferred across events and sites characterised by different severity.
Finally, inspection of the data included in the archive shows the relevance that indirect peak flow estimates have in flash flood analysis, particularly for small basins.This shows the urgency of developing standardised methods for postflood surveys in order to gather flood response data, including flow types, flood peak magnitude and time, damages, and social response.This is key to further advancing understanding of the causative processes and improving assessment of both flash flood hazard and vulnerability aspects (Calianno et al., 2013;Ruin et al., 2014).

Figure 1 .
Figure 1.Location of the flash floods in the central and western Mediterranean, the Alps, and Inland Continental Europe; inset is the eastern Mediterranean (Israel).The length of the arrow represents the area of the largest basin.Colour indicates the magnitude of the largest unit peak discharge.Direction represents the timing of the flash flood occurrence.

Figure 2 .
Figure 2. Budyko plot for the study basins (P : mean annual precipitation, AET: mean annual actual evapotranspiration, PET: mean annual potential evapotranspiration).In case of multiple nested catchments, only data for the largest one are reported.

Figure 3 .
Figure 3. Unit peak discharges versus drainage areas for the studied flash floods.The envelope curve for the upper limit of the relationship is reported.

Figure 4 .
Figure 4. Unit peak discharges versus drainage areas based on climatic regions: (a) Mediterranean catchments, (b) Alpine-Mediterranean and Alpine catchments, (c) Inland Continental, and (d) Arid and Arid-Mediterranean catchments.The envelope curve for each climatic region is reported.

Figure 5 .
Figure 5. Unit peak discharges versus drainage areas based on discharge assessment methods: (a) stream gauges and (b) indirect methods.The envelope curves for the upper limits for each method are reported.

Table 1 .
Summary information on the flash flood database.

Table 2 .
Summary statistics for drainage areas for the EuroMedeFF database under different climatic regions.

Table 3 .
Summary statistics for drainage areas for the EuroMedeFF database based on the two classes of discharge assessment (stream gauges vs. indirect methods).