**3.3.1 Differences between the modelling of riverine and flash floods**

Conventional large riverine floods and flash floods do not only differ in their general characteristics but also in their time of concentration (Tc or lag time) and duration of flood peaks. As it has been mentioned, for flash floods Tc is not more than 6 hours (NOAA definition). This extremely short lead time makes warning and prevention very difficult. In unexplored and ungauged catchments the single source of information is field surveys. Flood reconstruction is often only possible from the falling limb of the hydrograph or by assessing the aftermaths of the event (including the study of deposits – Costa, 1983).

As far as the triggering process is concerned, flash floods are generally associated with intense and convective rainfalls – often further enhanced by orographic effects (Horváth, 2005). Large riverine floods, on the other hand, are often preceded by days of incessant rainfall over hundred or thousand square kilometres affecting several drainage basins. It is to be noted here that in humid continental environments flash floods do not only occur in the summer, but rain-on-snow events may also generate winter flash floods (as shown for Southwest-Hungary by Pirkhoffer et al., 2009a,b; Czigány et al., 2010).

For flash flood modelling and forecasting usually an area of 10 to 200 km2 is selected, i.e. about one or two orders of magnitude less than for large riverine floods. During flash floods peak flow may exceed baseflow several hundred times - although peak discharge only lasts for a few hours. Moreover, as flash floods are usually triggered on the upper reaches of the stream, where the channel is narrow, stage increase is even more pronounced than flow changes. Figure 4 shows an idealized, rapidly rising and slowly attenuating hydrograph, typical of many hill regions, where Tc is extremely short. The second flow peak, triggered by a moderate rainfall, is due to higher soil moisture. The time of residence in the reservoirs (e.g. canopy or surface storage) of the hydrological cycle is usually much shorter for flash floods. Rainfall intensity largely exceeds infiltration rate, and, thus, excess runoff into intermittent streams reach the beds of permanent rivers, where a rapid rise of water will be observed (Fig. 4.).

Numerical modelling is further complicated by a plethora of environmental factors to be considered during the simulation process. Judging from the available data on documented flash floods, it can be claimed that built-up areas of the highest risk are located along the boundaries of areas of higher elevation and adjacent lowlands as well as at the abrupt narrowing of river valleys (bottlenecks). As periodically water-filled gullies often function as preferential flow paths during convective rainstorms, adequate knowledge on topography is essential for highly accurate flash flood prediction.

#### **3.3.2 Runoff modelling**

In case studies and pilot catchments we used 50-m and 10-m Digital Elevation Models (DEMs) based on topographic maps. In the field TOPCON HiPER Pro RTK GNSS high-

maps (topography, drainage network, CORINE land cover, soil depth, soil class). The meteorological variables required are rainfall, potential evaporation (for bare soil, closed

Conventional large riverine floods and flash floods do not only differ in their general characteristics but also in their time of concentration (Tc or lag time) and duration of flood peaks. As it has been mentioned, for flash floods Tc is not more than 6 hours (NOAA definition). This extremely short lead time makes warning and prevention very difficult. In unexplored and ungauged catchments the single source of information is field surveys. Flood reconstruction is often only possible from the falling limb of the hydrograph or by

As far as the triggering process is concerned, flash floods are generally associated with intense and convective rainfalls – often further enhanced by orographic effects (Horváth, 2005). Large riverine floods, on the other hand, are often preceded by days of incessant rainfall over hundred or thousand square kilometres affecting several drainage basins. It is to be noted here that in humid continental environments flash floods do not only occur in the summer, but rain-on-snow events may also generate winter flash floods (as shown for

For flash flood modelling and forecasting usually an area of 10 to 200 km2 is selected, i.e. about one or two orders of magnitude less than for large riverine floods. During flash floods peak flow may exceed baseflow several hundred times - although peak discharge only lasts for a few hours. Moreover, as flash floods are usually triggered on the upper reaches of the stream, where the channel is narrow, stage increase is even more pronounced than flow changes. Figure 4 shows an idealized, rapidly rising and slowly attenuating hydrograph, typical of many hill regions, where Tc is extremely short. The second flow peak, triggered by a moderate rainfall, is due to higher soil moisture. The time of residence in the reservoirs (e.g. canopy or surface storage) of the hydrological cycle is usually much shorter for flash floods. Rainfall intensity largely exceeds infiltration rate, and, thus, excess runoff into intermittent streams reach the beds of permanent rivers, where a rapid rise of water will be

Numerical modelling is further complicated by a plethora of environmental factors to be considered during the simulation process. Judging from the available data on documented flash floods, it can be claimed that built-up areas of the highest risk are located along the boundaries of areas of higher elevation and adjacent lowlands as well as at the abrupt narrowing of river valleys (bottlenecks). As periodically water-filled gullies often function as preferential flow paths during convective rainstorms, adequate knowledge on

In case studies and pilot catchments we used 50-m and 10-m Digital Elevation Models (DEMs) based on topographic maps. In the field TOPCON HiPER Pro RTK GNSS high-

assessing the aftermaths of the event (including the study of deposits – Costa, 1983).

canopy and water surfaces) and daily mean air temperature (De Roo et al., 2000).

**3.3.1 Differences between the modelling of riverine and flash floods** 

Southwest-Hungary by Pirkhoffer et al., 2009a,b; Czigány et al., 2010).

topography is essential for highly accurate flash flood prediction.

**3.3 Methods** 

observed (Fig. 4.).

**3.3.2 Runoff modelling** 

precision GPS and SOKKIA surveying instruments were employed to improve the spatial resolution of the generated DEM to 1 m. This, however, was achieved only locally – usually in the immediate vicinity of watercourses.

Surface runoff was simulated using HEC-HMS, which has the advantage of working with distributed precipitation data available from weather radar and a continuous soil-moistureaccounting model (Pirkhoffer et al., 2009b). Radar images and various meteorological data were obtained from the Hungarian Meteorological Service (OMSz), while hydrological data (e.g. water stage and discharge) were received from the Research Institute for Environmental Protection and Water Management (VITUKI Rt.), the South-Transdanubian Environmental Protection and Water Management Directorate (DDKÖVIZIG) and the MECSEKÉRC Rt., a successor enterprise to the former uranium mining company. To obtain field data we monitored soil moisture (using Time Domain Reflectrometry technique), canopy cover and precipitation at 14 monitoring stations in a 1.7 km2 pilot catchment (Pirkhoffer et al., 2009b; Czigány et al., 2010). Runoff output data were then compared with observed flow.

#### **3.3.3 Rapid screening and GIS-based risk assessment**

The first comprehensive, but least detailed type of approach to flash flood modelling is rapid screening that usually employs ARC GIS and SGA GIS softwares (Pirkhoffer et al., 2009b; Czigány et al., 2011a). The input data for this analysis comprise various topographical, geological, soil and land use parameters. Rapid screening models serve to delineate the area with a natural hazard or rate vulnerability and risk in that area (Cobby et al., 2009; Czigány et al., 2008) in order provide a general overview of its level for experts, decision-makers and the public.

Fig. 4. Hydrograph of a typical flash flood (a) and that of a flood event with saturated soils (b). Tc = time of concentration; LNQ = maximum discharge; LKQ = minimum discharge

Flash Flood Hazards 39

represents the stream (valley inundation model). In this case, however, we have to define the valley floor through visual interpretation, wherever possible, including bottlenecks and broader floodplains (Fig. 5c). As it has been mentioned flood levels are approximated by

Flood risk was determined by the complex, superimposed impact of the 50-m resolution input grid databases of passive factors through appropriate weighting. Figure 5 summarizes the major elements of a GIS based runoff model and its mapping possibilities. Obviously, the number of included input parameters will determine the accuracy of the output

Fig. 5. Parameters for the construction of GIS inundation maps: a. land cover; b. time of

runoff concentration; c. valley width; d. height above channel

height above the channel (Fig. 5d).

vulnerability map.

First, the catchments potentially affected by flash floods are identified. In the next step, risk assessment has to be carried out individually for each catchment as catchment properties influence flood level and stream behaviour. The impacts of floods are most pronounced along the watercourses and at the outflow point of the catchment. Therefore, all catchments are assigned with a unique ID number and the outflow points (usually in built-up areas) receive the same ID.

There are two approaches available for flood risk assessment: the first is based on passive factors, i.e. those which do not change significantly with time, while the second method focuses on the active factors, i.e. those which show significant variations with time (precipitation, canopy cover and soil moisture content). Passive environmental factors are determined with relatively high accuracy; spatially and temporally correct data on active factors, however, are difficult to obtain (Bálint & Szlávik, 2001).

The environmental factors incorporated in the 1:100,000 risk map are classified into five categories: topographical parameters (derived from the DEM), drainage netwok (from the river network database of Hungary), land cover (from CORINE Land Cover 2000), soils (from the AGROTOPO Hungarian soil database) and hydrological conditions. The three topographical properties were average slope, slope range and valley density for the catchment. Four soil parameters, which influence surface runoff, infiltration and interception were considered: soil depth, physical soil type, the ratio of barren/vegetationcovered surfaces (in limestone areas). Data on the hydrological factors contributing to flash flood generation were borrowed from the river network database of Hungary, created in accordance with the Water Framework Directive of the European Union. As confluences (number of tributary rivers) are prone to enhance the magnitude of flash floods (also proven during the Mátrakeresztes flash flood event – Horváth, 2005), first the number of stream confluences per unit area (1 km2) were determined. Then drainage density was incorporated in the model.

GIS-based risk mapping represents a transitional type of models. It is closely related to rapid screening, but already points towards numerical analyses. The basic difference from rapid screening is that flooding in this case is not directly associated with a given rainfall event. Rainfalls are incorporated in a rather hypothetical manner: the extent of flooding is determined from a threshold height above the valley floor or the mean stream stage.

GIS models are primarily based on topography: all parameters, including runoff, Tc and drainage network, are derived from a topographic map or a DEM (Digital Elevation Model). The spatial resolution is at least 5 or 10 m. (Errors tend to be significant: between the calculated and the actual watercourses may reach 100 m.) The models which ignore infiltration and the canopy effect and define runoff direction and volume exclusively from topographic models are called impervious surface (IS) models. However, to obtain a true picture of runoff behaviour, the impact of soils and land use (Fig. 5a) cannot be neglected. Figure 5b clearly show the differences in runoff according to IS models with light colours, while the black zones indicate runoff also influenced by soils and land use.

Channel widths vary greatly in areas of high relief, ranging from 0.5 m to dozens of metres. When the spatial resolution of the topographic map exceeds channel width, a channel as a physical entity will not be shown on the final output map but a theoretical centerline

First, the catchments potentially affected by flash floods are identified. In the next step, risk assessment has to be carried out individually for each catchment as catchment properties influence flood level and stream behaviour. The impacts of floods are most pronounced along the watercourses and at the outflow point of the catchment. Therefore, all catchments are assigned with a unique ID number and the outflow points (usually in built-up areas)

There are two approaches available for flood risk assessment: the first is based on passive factors, i.e. those which do not change significantly with time, while the second method focuses on the active factors, i.e. those which show significant variations with time (precipitation, canopy cover and soil moisture content). Passive environmental factors are determined with relatively high accuracy; spatially and temporally correct data on active

The environmental factors incorporated in the 1:100,000 risk map are classified into five categories: topographical parameters (derived from the DEM), drainage netwok (from the river network database of Hungary), land cover (from CORINE Land Cover 2000), soils (from the AGROTOPO Hungarian soil database) and hydrological conditions. The three topographical properties were average slope, slope range and valley density for the catchment. Four soil parameters, which influence surface runoff, infiltration and interception were considered: soil depth, physical soil type, the ratio of barren/vegetationcovered surfaces (in limestone areas). Data on the hydrological factors contributing to flash flood generation were borrowed from the river network database of Hungary, created in accordance with the Water Framework Directive of the European Union. As confluences (number of tributary rivers) are prone to enhance the magnitude of flash floods (also proven during the Mátrakeresztes flash flood event – Horváth, 2005), first the number of stream confluences per unit area (1 km2) were determined. Then drainage density was incorporated

GIS-based risk mapping represents a transitional type of models. It is closely related to rapid screening, but already points towards numerical analyses. The basic difference from rapid screening is that flooding in this case is not directly associated with a given rainfall event. Rainfalls are incorporated in a rather hypothetical manner: the extent of flooding is

GIS models are primarily based on topography: all parameters, including runoff, Tc and drainage network, are derived from a topographic map or a DEM (Digital Elevation Model). The spatial resolution is at least 5 or 10 m. (Errors tend to be significant: between the calculated and the actual watercourses may reach 100 m.) The models which ignore infiltration and the canopy effect and define runoff direction and volume exclusively from topographic models are called impervious surface (IS) models. However, to obtain a true picture of runoff behaviour, the impact of soils and land use (Fig. 5a) cannot be neglected. Figure 5b clearly show the differences in runoff according to IS models with light colours,

Channel widths vary greatly in areas of high relief, ranging from 0.5 m to dozens of metres. When the spatial resolution of the topographic map exceeds channel width, a channel as a physical entity will not be shown on the final output map but a theoretical centerline

determined from a threshold height above the valley floor or the mean stream stage.

while the black zones indicate runoff also influenced by soils and land use.

factors, however, are difficult to obtain (Bálint & Szlávik, 2001).

receive the same ID.

in the model.

represents the stream (valley inundation model). In this case, however, we have to define the valley floor through visual interpretation, wherever possible, including bottlenecks and broader floodplains (Fig. 5c). As it has been mentioned flood levels are approximated by height above the channel (Fig. 5d).

Flood risk was determined by the complex, superimposed impact of the 50-m resolution input grid databases of passive factors through appropriate weighting. Figure 5 summarizes the major elements of a GIS based runoff model and its mapping possibilities. Obviously, the number of included input parameters will determine the accuracy of the output vulnerability map.

Fig. 5. Parameters for the construction of GIS inundation maps: a. land cover; b. time of runoff concentration; c. valley width; d. height above channel

Flash Flood Hazards 41

Therefore, the soils acted as an impervious surface triggering extreme surface runoff. Soil moisture content only slightly decreased in the following two-week period, thus the second storm with less cumulative rainfall induced flash floods again on 31 May and 1 June. Over the period of 1 May to 16 June the cumulative number of rainy days reached at least 21 at all rain gauges operated by the Hungarian Meteorological Service in Southwest-Hungary (Table 1 and Fig. 6). Groundwater tables in the observation wells of the area indicated a

Fig. 6. Total cumulative rainfall (a) and number of rainy days (b) in Southern Transdanubia between 1 May and 16 June 2010 (data provided by the Hungarian Meteorological Service)

Table 2 clearly illustrates the extreme precipitation characteristics of the mentioned 47-day period. At many rain gauges in the study area precipitation reached or even exceeded 50% of the mean annual rainfall. The long-term average May precipitation in Pécs is 84 mm, while the cumulative precipitation in May 2010 was nearly threefold higher. The return time

The extremity of rainfall is also clearly reflected in the actual intensity values. For short periods, intensity values reached 30 mm h-1, while 10-minute intensity was 51.6 mm h-1 at the Keszthely main meteorological station. For small mountainous catchments it is essential to know the areal extent of the rainfall zone. Due to the scarcity of rain gauges, we have to rely on radar images. Convective cells are around 5 to 10 km across, thus radar images of adequate (at present 2 by 2 km) resolution are extremely helpful in the estimation of the areal extent of precipitation for modelling purposes. Heavy rainfall characterized the settlements of Sásd and Csikóstőttős on 15 May 2010 (Fig. 7) and maximum rainfall and intensity were observed basically in the same area on the following day (16 May 2010).

On 15 May 86 mm of rain fell on the upper catchments of the Baranya Canal, where Tc is shortest within the catchment, with similar flood stages. As a consequence, rapidly rising flood stages were just slightly off from the previous records (Fig. 8). South of the divide, in the mountainous Bükkösd Stream catchment, the rainfall was much more prolonged and high water stages persisted longer at the Szentlőrinc stream gauge than at the gauges

mean rise of 1 to 1.2 m over the entire region (DDKÖVIZIG, 2010).

of such precipitation is estimated at 400 years.

upstream (Fig. 9).
