Sensitivity of the hydrological response to the variability of rainfall fields and soils for the Gard 2002 flash-flood event

Journal of Hydrology 394 (2010) 134–147

Contents lists available at ScienceDirect

Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol

Sensitivity of the hydrological response to the variability of rainfall fields and soils for the Gard 2002 flash-flood event Sandrine Anquetin a,*, Isabelle Braud b, Olivier Vannier a,b, Pierre Viallet c, Brice Boudevillain a, Jean-Dominique Creutin a, Claire Manus a a b c

Laboratoire d’étude des Transferts en Hydrologie et Environnement, Université de Grenoble (CNRS, UJF, IRD, INPG), France Cemagref, UR HHLY, F-69336 Lyon, France HYDROWIDE, BP 53 – Domaine Universitaire, 38041 Grenoble Cedex 9, France

a r t i c l e

i n f o

Keywords: Flash-floods Distributed hydrological model Radar rainfall Soil variability

s u m m a r y In the general context of field experiment design, this paper presents a modeling study that quantifies the respective impact of rainfall estimation and soil variability on the simulated discharge for an extreme event in southern France. The CVN distributed hydrological model, built within the LIQUIDÒ modeling platform is used. The method is illustrated for two medium sized catchments, Saumane (99 km2) and Uzès (88 km2) using raingauges and two radar estimates. The soil properties are extracted from an existing soil database provided for the whole region. The model parameter specification uses available observation and a priori hydrological knowledge. No parameter adjustment is performed. For model evaluation on the regional scale, simulated maximum peak discharges are compared with post-flood estimations for 32 catchments. The area of these catchments ranges from 2.5 to 99 km2 and model results are satisfactory. Then, the study focuses on the Saumane and Uzès catchments. A sensitivity analysis highlights the role of the Manning roughness coefficient on the simulated hydrographs dynamics. The impact of the bottom boundary condition of the infiltration and water redistribution module is also shown for the gauged Saumane catchment. Then the impact of rainfall input and soil spatial variability is presented. The results show that (i) the use of radar data is necessary to properly simulate the flood dynamics; (ii) although radar volume-scanning strategy has been shown to give more accurate results on a pixel/gauge comparison of the rainfall estimations, it is not necessarily the case when catchment averaged amounts are considered, especially for catchments in mountainous areas; (iii) the impact of the variability in soil properties on the simulated discharges is of the same order of magnitude as the impact of differences in rainfall estimation; (iv) the flood dynamics presents two phases: the first one, mainly controlled by the soil properties and the second one, since the soils are saturated, controlled by the rainfall variability. Therefore, uncertainties on both observations need to be mitigated in order to improve flash-flood understanding. Ó 2010 Elsevier B.V. All rights reserved.

1. Introduction There is no doubt that flash-floods represent one of the most destructive natural hazards in the Mediterranean region (Gaume et al., 2009) and are still poorly understood. During the last two decades, several extreme flood events occurred in Southern France (i.e. Nîmes, 1988; Vaison-la-Romaine, 1992; Aude, 1999; Gard, 2002). These events are still poorly understood, mostly due to the lack of experimental sites and long-term hydro-meteorological data with adequate space–time resolution (Foody et al., 2004; Anquetin et al., 2004; Borga et al., 2008). * Corresponding author. Address: LTHE, BP 53 X, 38041 Grenoble Cedex, France. Tel.: +33 (0) 476 82 51 08; fax: +33 (0) 476 82 50 14. E-mail address: [email protected] (S. Anquetin). 0022-1694/$ - see front matter Ó 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2010.07.002

Flash-floods result from the combination of meteorological and hydrological conditions. Recognition of the coupled meteorological/hydrological nature of flash-floods is now obvious in interpretative studies and in the development of predictive models (Creutin and Borga, 2003; Anquetin et al., 2004; Collier, 2007). It has been shown that most flash-flood events are attributed to precipitation generated in stationary Mesoscale Convective Systems (MCSs) (Hernandez et al., 1998; Homar et al., 1999). Due to their very localized nature and to their wide variability in space and time, the observation of such events using raingauge networks is problematic. Weather radars provide better spatial rainfall resolution, even if radar assessment of rainfall is significantly influenced by orography (Joss and Waldvogel, 1990; Pellarin et al., 2002; Germann et al., 2006). Moreover, it has been demonstrated that the more intense the rainfall, the less reliable the radar rainfall

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

estimates become (BASC, 2005). Thus, accurate monitoring of severe storm rainfall intensities remains a major challenge. Flash-floods are rare events that usually affect small to medium basins. The mitigation of the resulting distributed risk imposes the study of specific questions dealing with ungauged river basins. Indeed, people exposure to the dangerousness of the event is the highest in small-scale catchments because traditional defenses are usually weak (Montz and Gruntfest, 2002; Ruin et al., 2008; Creutin et al., 2009). Several methods for predicting flash-floods in ungauged river basins are now accepted. The flash-flood guidance (Georgakakos, 2006; Norbiato et al., 2008) and the discharge threshold exceedance approach (Reed et al., 2007; Younis et al., 2008) are built to give an early warning suitable for the organization of civil defense. These methods rely either on conceptual or physically based hydrological models. To improve the forecasting chain, there is a real need for research to improve (i) the understanding of the major atmospheric and hydrologic factors leading to extreme flood event and (ii) their representations within the prediction models. The importance of considering the spatial distribution of rainfall for process-oriented hydrological modeling is now accepted. Sensitivity studies of the runoff response to the spatial variability in precipitation highlight that detailed rainfall information is necessary for small catchments in complex terrain, and for runoff processes that respond directly to precipitation (Yates et al., 2000; Nicotina et al., 2008; Sangati et al., 2009). Morin et al. (2006) note that one of the key issues is the spatial resolution at which the rainfall data are represented in the hydrological model. They performed sensitivity analysis of maximum radar cell intensity and its extensions to simulated peak discharges. Their results show that peak discharge could be twice as high if the convective cell was initiated just a few kilometres away from the catchment. Delrieu et al. (2009b), in the introduction of the special issue dedicated to ‘‘Weather radar and Hydrology”, suggest that ‘‘weather radar technology offers a unique means for characterizing the rainfall variability over the range of scales and with the space–time resolutions required for a large variety of hydrological problems”. However, radar-based precipitation estimation may lack consistent, quantitative accuracy. Moreover, the formulation of hydrological models in distributed form may be problematic due to process complexity and scaling issues. The literature addressing this problem includes numerous approaches recently reviewed in the Advanced in Water Resources special issue (32 (7), 2009). The qualification of the different radar treatment is either done using a regional radar pixel – raingauge comparison (Joss and Waldvogel, 1990; Westrick et al., 1999; Pellarin et al., 2002; Germann et al., 2006; Delrieu et al., 2009a) or by examining the resulting simulated discharge at the event scale (Carpenter and Georgakakos, 2004; Tetzlaff and Uhlenbrook, 2005; Chancibault et al., 2006; Cole and Moore, 2008). The results of a regional evaluation are obviously linked to the observation ground network that may introduce bias in mountainous regions where the density of raingauges is the smallest. Furthermore, the hydrological evaluation is shown to be strongly dependent on the catchment size (Tetzlaff and Uhlenbrook, 2005; Morin et al., 2006). Therefore, the evaluation of the radar-based quantitative precipitation is not straightforward. However, radar rainfall estimation remains a natural approach to area-wide flood forecasting and warning at any location, whether gauged or ungauged. The first objective of this paper is to propose an assessment of radar-based precipitation estimation complementary to the regional radar pixel – raingauge comparison proposed by Delrieu et al. (2009a). We analyse the response of the non-calibrated distributed hydrological model CVN (Manus et al., 2009) to different radar data sets issued from various data processing. We investigate the influence of the spatial and temporal rainfall variability in terms of peak

135

discharge time and amplitude as proposed by Sangati et al. (2009). The September 2002 Gard flash-flood event (Delrieu et al., 2005), shortly described in Section 2, is the case study and the simulations are focused on small catchments ranging from a few km2 to about 100 km2. For the same event, Le Lay and Saulnier (2007) used the eventbased n-TOPMODELs calibrated model. They showed that the model efficiency significantly increases when the spatial variability of rainfall is taken into account. Nevertheless, for some of the catchments, mis-performance remained unexplained and further insight is required in order to better understand the missing factors that are influential on the hydrological response for such extreme events. Our hypothesis is that soil spatial variability can be a significant factor, which is usually neglected, in existing distributed models (for some reviews see Todini, 2007; Kampf and Burges, 2007; Furman, 2008). The second objective of this paper is to quantify the respective impact of rainfall estimation and soil variability on the simulated discharges. The model CVN, built within the LIQUIDÒ hydrological modeling platform and presented in Section 3, takes into account both the rainfall variability as described by raingauges or radar and the soil variability as described using an existing soil data base. Sensitivity studies on (i) radar rainfall processing, (ii) space and time resolution of the rainfall and (iii) soil properties are presented and discussed in Section 4. Conclusions and perspectives for future work are finally drawn in Section 5.

2. The region of interest and the case study 2.1. Studied area and main characteristics of the catchments The Cévennes–Vivarais region (Fig. 1), located in the Southeastern part of the Massif Central, is especially prone to flashfloods during the fall season. The topography starts from the Mediterranean shore, and ranges up to 1700 m (Mount Lozère; Fig. 1b) over less than 100 km. The main Cévennes rivers (Vidourle, Gard, Cèze, Ardèche; Fig. 1b) have a typical intermittent hydrological regime: low water levels during the summer, floods occurring mainly in the fall. The above-mentioned catchments are medium size catchments (i.e. 2300 km2 for the largest) with travel times of less than 12 h. In this study, soil characteristics are extracted from the Languedoc–Roussillon soil database (later referred as BDsol-LR) provided by the INRA1 from the IGCS2 program. The soil depth in the studied region and the variability of soil classes are given in Fig. 2b and c, respectively. These graphs illustrate the spatial resolution of the database. The average depth does not exceed 55 cm and more than 50% of the soils are shallow (depth below 50 cm). Manus et al. (2009) show that the average texture over the whole region consists of around 50% sand, 30% silt and 20% clay. Even if the modeling approach is built for the whole Cévennes–Vivarais region, this study is mainly focused on two meso-scale catchments (Saumane, 99 km2 and Uzès, 86 km2; Fig. 1b) and on one extreme event that caused severe flooding in the entire Gard region. These two catchments are representative of the soil diversity of the region. The Saumane catchment is located in a hilly area (i.e. mean slope of the whole catchment 0.38 m m1) while the Uzès catchment is located in a flat area (i.e. mean slope of the whole catchment 0.06 m m1) with an average river slope four times lower than that of the Saumane basin (Fig. 2a). The average soil depth for Saumane is of less than 20 cm whereas the Uzès soil is much deeper (80 cm) leading to a maximum water storage capacity four times larger than in Saumane (300 mm 1 2

The French National Institute of Agronomical Research. http://gissol.orleans.fr/

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

Rhône

136

(a)

(b)

Fig. 1. (a) Location of the region in France. (b) Topography and indication of the two studied catchments, in red, Uzès and Saumane, in black, the 32 catchments used for the regional model evaluation. The 50 km and 100 km Bollène radar circles are indicated. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

against less than 70 mm for Saumane). Fig. 2d presents the soil cartographic units and points out the soil heterogeneity at the catchment scale. The Saumane catchment has an approximately uniform soil type while the Uzès catchment presents a more complex soil organization. 2.2. Radar rainfall estimations of the 2002 Gard flash-flood event The Gard flash-flood event in September 2002 was an exceptional situation due to: (i) the extreme intensity of the event with a maximum accumulated precipitation of about 610 mm recorded in 24 hours (from 1200 UTC on September 8th until 1200 UTC on September 9th); (ii) a spatial extension of at least 3000 km2 for the region affected by precipitation greater than 200 mm in less than 30 hours. This storm triggered catastrophic flash-floods on many upstream tributaries as well as the most important flood ever reported in the major rivers (Gard, Cèze and Vidourle) (Fig. 1b). It took 23 human lives and generated 1.2 billion Euros of damages in less than 24 hours over an area of 20,000 km2 located in the Cévennes–Vivarais region (Delrieu et al., 2005). Within the framework of the OHM-CV3 hydro-meteorological observatory, the radar data was derived from the so-called Bollène 2002 experiment protocol (Delrieu et al., 2009a) designed to evaluate the benefit of a radar volume-scanning strategy in this region. The three operational Plan Position Indicators (PPIs) (elevation angle 3 Observatoire Hydrométéorologique Méditerranéen Cévennes – Vivarais, http:// www.ohmcv.fr/.

of 0.8°, 1.2° and 1.8°) of the Bollène radar (see its location in Fig. 1b) were complemented by 10 PPIs (elevation angle from 0.4° to 18°) allowing a good sampling of the atmosphere (see Bouilloud et al., 2010 for more information on the used of several PPIs to scan the atmosphere for the studied region). Based on this new protocol, innovative algorithms, two time-adaptive strategies and four space–time strategies (Delrieu et al., 2009a), were developed in order to identify and correct various error sources (radar calibration, ground clutter). Improved relationships that link the radar reflectivity (Z) to the rainfall rate (R), usually called the Z–R relationship, were proposed to take into account the rain type and the vertical profile of reflectivity. No radar – raingauge adjustment and/or external meteorological information is applied to constrain the radar processing algorithms. In this study, the rainfall fields estimated with two different algorithms are used. The operational algorithm (rainfall field later referred as OPER) uses the three operational PPIs (i.e. 0.8°, 1.2° and 1.8°). Delrieu et al. (2009a) show that ‘‘in comparison with time-adaptive strategies (T-AD), the space–time adaptive strategy (ST-AD) yields a very significant reduction in the radar-raingauge bias”. Therefore, this study uses one of the four ST-AD precipitation estimator, i.e. ST-AD3, which obtains the best score for point radar – raingauge comparisons at the event time step (Nash (Nash and Sutcliffe, 1970) N = 0.9) for the Gard 2002 event. Radar rainfall fields are established over 1-km2 Cartesian meshes with a time resolution of 5 min. The kriged rainfall data is provided at an hourly time step and over the same grid. An anisotropic variogram is used for the kriging (Creutin and Obled, 1982; Lebel et al., 1987) as suggested by the rain climatology of the region (Molinié et al., submitted for

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

137

Fig. 2. (a) Zoom into the Gard region where the studied catchments and the 32 catchments used for validation are located. (b) Average soil depth (in cm) extracted from the BDSol-LR database. (c) Soil classes: each color corresponds to a different soil cartographic unit. (d) Geological simplified maps (colors) and hydro-landscape units (light black line) of the two studied catchments. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

publication). The identified range values (mean decorrelation distances of the rain gauge amounts) are 65 km in the direction North-30° (i.e., toward the NNE), roughly parallel to the mountain range, and 40 km in the orthogonal direction (North-120°). The OPER and ST-AD3 48-h cumulated rainfalls are compared with the kriged rainfall on the regional scale in Fig. 3, and at the catchment scale in Fig. 4. The two radar rainfall estimators do not equally estimate the rainfall gradients. In term of mean water depth, the largest discrepancy is observed for the Saumane catchment due to its farthest location from the radar. The mean cumulated water depths obtained with the OPER and the ST-AD3 estimations reach 145 mm and 208 mm, respectively, while the raingauge network gives 204 mm. Concerning the Uzès basin, both OPER (400 mm) and ST-AD3 (390 mm) radar protocols produce larger accumulated rainfall than the raingauge network (190 mm). Some hourly raingauges did not work properly during this event, leading to this large underestimation of the kriged rainfall (KRIG). If we use the daily raingauge network, the accumulated rainfall of the Uzès basin reaches 408 mm, which is consistent with radar observations. Nevertheless, the daily observations are inappropriate for flash-flood modeling and we used the hourly data in our simulations. This study uses the 5 min time step OPER and ST-AD3 radar data and the KRIG hourly rain data to simulate discharges.

3. The hydrological modeling strategy 3.1. Model setup The distributed hydrological model CVN is developed within the numerical LIQUIDÒ platform (Viallet et al., 20064; Branger et al., 2010). This flexible modeling environment allows the customization of integrated models for a specific case study. It includes modules that describe the hydrological processes with time and space scales consistent with their own dynamics. A discrete action simulator called ‘‘Scheduler” manages the temporal coupling. Each solver computes its own time step that is estimated at each solver execution according to criteria defined by the module developer. The representation of landscape heterogeneity leads to irregular modeling units. Their irregular shapes makes the spatial coupling more complicated than when regular grids are used. As detailed in Branger et al. (2010), the LIQUIDÒ platform is able to handle these irregular geometries, to synchronize different time steps, and to simulate complex connexions between components, in particular involving feedback. This platform is still being developed. The present study uses the same

4

www.forge.hydrowide.com/.

138

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

Cumulated rainfall (mm) 700

Extended Lambert II coordinates

600

500

400

300

200

100 0

Rain gauges Radar location

Extended Lambert II coordinates

(3) Extended Lambert II coordinates

(1)

Extended Lambert II coordinates

Extended Lambert II coordinates

(2)

Fig. 3. Forty-eight-hour cumulated rainfall estimations at the regional scale given by (1) the kriging (KRIG) and the two radar protocols: (2) OPER; (3) ST-AD3. (+) Raingauges used for the kriging. The black rectangle stands for the zoomed region.

model as the one used by Manus et al. (2009) that can be summarized as follows. The infiltration is computed using the 1D Richards’ equation, solved through the simplified Ross’ algorithm (Ross, 2003), as validated by Varado et al. (2006). When the topsoil is saturated, ponding is generated. In the studied catchments, there is no permanent groundwater and it is assumed that perched water tables can locally be formed, due to the vertical heterogeneity of the soils (type-A saturation excess of Latron and Gallart, 2007). Saturation can also occur due to full saturation of the soil (referred to as type-B saturation excess by Latron and Gallart, 2007). Both overland flow and the flow through the drainage network are modeled with a 1D kinematic wave approximation of the de St-Venant equations with the Manning friction law. The channel is described with a simplified trapezoidal geometry. As proposed by Manus et al. (2009), the soil parameters are estimated using the Rawls and Brakensiek (1985) pedotransfer functions. Soil structure is taken into account in the porosity estimation, which is derived from the BDSol-LR with the Brakensiek et al. (1981) statistical approach. Thus, the specification of the parameters directly uses (i) the available data and (ii) the a priori hydrological knowledge. The model is then implemented for the whole region and is run without any parameter adjustment. Therefore, the model set up relies ‘‘only” on what we know and what we can observe. It is in this sense that we speak about a non-calibrated model.

The two-level spatial discretization is the one proposed by Dehotin and Braud (2008). For the first level, the catchments are sub-divided into Representative Elementary Watersheds (REWs) (Reggiani et al., 1998), or sub-catchments. They are extracted from a Digital Elevation Model (DEM) analysis where the threshold drained area for initiating the network is set at 0.1 km2. The REWs discretization corresponds to the first Strahler order. Due to the 75 m DEM truncation, the river slope is sometimes found to be very small. To improve the Manus et al. (2009) river network, a minimum slope is fixed in this study to half of the DEM resolution divided by the length of the reach. The second level of discretization renders the soil heterogeneity. The sub-catchments are subdivided into hydro-landscapes (Dehotin and Braud, 2008) derived from Geographic Information System (GIS) layers analysis. In the present case, only soil variability as described in the soil map is considered. The soil variability (Fig. 2d) is described using 394 and 411 hydro-landscapes inside the Saumane and Uzès catchments, respectively. The 1D vertical soil water transfer module is solved at this scale, using a constant vertical soil discretization of 1 cm in thickness, and a description of the soil horizons extracted from the soil database. A null flux is considered at the bottom of the horizons. Therefore, the bedrock is assumed impervious. When the top soil is saturated, ponding is generated. The event occurred after three days of light rain on the 2nd, 4th and 5th of September. The initial soil moisture is specified based on

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

139

Extended Lambert II coordinates (km)

Extended Lambert II coordinates (km)

Extended Lambert II coordinates (km)

KRIG

KRIG

Fig. 4. 48-h cumulated rainfall estimations at the hydro-landscape scale given by the kriging (KRIG) and the two radar protocols: OPER, ST-AD3. (Red triangle) Raingauges used for the kriging.

the SAFRAN–ISBA–MODCOU (SIM) model with a 8  8 km2 resolution as provided by Météo-France (Habets et al., 2008). SIM provides a wetness index that is converted into a uniform soil water pressure profile corresponding to a 60% saturation of the first layer. For the case study, sensitivity studies to the initial rate of saturation (not shown here) and on the initial storage deficits (Braud et al., 2010) suggest that, for catchments presenting shallow soil, saturation is quickly reached and the impact on peak discharge is small. For catchments with deeper soil, the initial soil moisture affects the simulated discharges. The 60% saturation initial condition remains nevertheless plausible given the estimated peak discharges. Without any more detailed information, the initialization is based on the initial soil moisture provided by the SIM model. Due to the soil properties and their representations in the model, the available initial water storage for the Uzès catchment is 8 times larger and more heterogeneous than for the Saumane catchment. Transfer within the river network is performed using the 1D kinematic wave approximation of the de St-Venant equation. The network is discretized into river reaches with one reach per subcatchment. The river section is assumed trapezoidal with geometrical parameters assigned according to the Strahler order. In this first version of the model, the routing scheme transferring the ponding generated on the hydro-landscapes to the river reaches is very simple as the ponding is quasi-instantaneously transferred to the closest river reach. Due to the high-resolution modeling approach (Fig. 2c), more than 50% of the hydro-landscape units are less than 500 m away from the closest river reach. The maximum distance does not exceed 2500 m. The impact of neglecting the hillslope travel time is therefore minimized. For instance, if a 0.5 m s1 hillslope velocity is assumed, the neglected time delay is 15 min

delay for more than 50% of the hydro-landscape and reaches a maximum of one hour for the farthest units. 3.2. Model evaluation In this study, the model evaluation is twofold: (i) quantify the regional hydrological responses to one storm using a post-flood field survey over several ungauged basins; (ii) assess the accuracy of the simulated outflow of one gauged catchment for several storms. 3.2.1. Regional evaluation The use of post-flood field survey provides a spatial view of the hydrological response across a large range of catchment scales (Borga et al., 2008). The information retrieved from the survey is the peak discharge and, sometimes, the time of the peak according to witness interviews. The regional model evaluation is based on the simulation of the discharges of 32 watersheds surveyed during the extensive post-flood field experiment of the Gard 2002 event (Gaume and Bouvier, 2004). The size of the catchments and the estimated discharges are given in Table 1. Their locations are indicated in Fig. 2a. The ST-AD3 rainfall fields are used as input to the hydrological model, and the initial soil moisture conditions remain the same as defined in Section 3.1. Fig. 5 compares the estimated and the simulated peak discharges at the 32 validation outlets. Most of the simulated peak discharges are within the range of uncertainty given by the postflood survey, whatever the catchment size. In general, the CVN model tends to overestimate the peak discharge; the explained variance (R2) is equal to 0.60. Apart for the catchments located at

140

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

Table 1 Peak discharges of the 32 catchments used for the model evaluation (Gaume and Bouvier, 2004). The most probable values are in bold whereas the values in brackets give the range of uncertainty. #

Name of the catchment

Area km2

Estimated peak discharge Qe m3 s1

22 26 5 14 2 1 9 13 19 8 4 6 31 11 30 23 18 25 29 15 21 27 32

Aigalade_sec02 Aigalade_sec03 Alzon_sec01 Alzon_sec03 Alzon_sec04 Alzon_sec05 Bourdic_sec02 Braune_sec01 Braune_sec04 Braune_sec05 Braune_sec06 Brestalou_sec01 Brestalou_sec03 Brestalou_sec04 Brestalou_sec05 Courme_sec01 Courme_sec02 Courme_sec03 Courme_sec04 Crieulon_sec01 Crieulon_sec05 Crieulon_sec07 Crieulon at Barrage de la Rouvière Droude_sect04 Galeizon_sec03 Galeizon_sec05 Grabieux_sec02 Ourne_sec02 Ourne_sec03 Quinquillan_sec01 Rieumassel at Barrage Ceyrac Brié at Fontanès

32.2 39.5 8.2 16 3.4 2.5 12 14.6 23.3 11.6 7.3 9.7 82.7 12.6 67.2 32.2 21.3 38.5 50.2 19 25 45.6 94

166 230 [200–300] 330 [270–370] 430 [300–550] 100 [70–120] 100 [80–125] 111 [100–111] 60 [40 - ] 300 [200–400] 230 [170–290] 160 [120–200] 165 [105–210] 500 [350–800] 220 400 [ - 600] 550 [400–650] 350 [260–470] 550 [ - 600] 635 [590 – 730] 320 [285–380] 500 [350–600] 600 [400–850] 1576 (observation)

4.04 38.1 21 24.1 12 10.2 20.5 46.2 14.6

40 [30–50] 400 [320–490] 390 [310–470] 400 [350–500] 300 [250–350] 270 [220–350] 85 503 (observation) 120 [65–155]

3 24 17 20 10 7 16 28 12

the eastern edge (#3, #9, #13, #22, #26) or at the southern edge (#16), the relative error between the simulated and the estimated peak discharges ranges from less than 1% (#1 and #14) to approx-

imately 60% (#12). The mean error value is about 29% whereas the estimation uncertainty associated with the field experiment reaches 21%. This result is fairly good since no model parameter adjustment is performed. As mentioned in Gaume et al. (2003) and Bonnifait et al. (2009), catchments (#3, #9, #13, #16, #22, #26) are located in the region where the rainfall gradient is the strongest. In this case, the peak discharge overestimation is probably associated with rainfall errors. Note that the estimated specific discharges range from 5 m3 s1 km2 to 40 m3 s1 km2 in the 600 mm cumulative rainfall area. These values are much higher than those of the 10 years return period discharge, which is about 2 m3 s1 km2 for such catchment size in this region. 3.2.2. The case of the gauged Saumane catchment To check our physically based distributed hydrological model consistently with the validation made on regional scale, we must use high-resolution rain data in space and in time. Moreover, discrepancies between the rainfall fields observed by radar and by raingauges may be large (Fig. 4), especially in regions not properly covered by the raingauge network. The radar data are the most appropriate, but given the current status of radar data in the study region, it is quite difficult to have a long reliable series of such data at a given location. Within this general context, the hydrological model is evaluated on the gauged Saumane catchment (99 km2) and for three events (September 2000, September and November 2002). Since radar protocol has changed in September 2002, OPER rainfall estimations are used for the three storms. The 48-h accumulated rainfall over the catchment is 187 mm, 145 mm and 70 mm respectively for the three events. For the three simulations, in the absence of more accurate information, the initial conditions are kept identical. Given the uncertainty on the bottom boundary conditions, three alternatives to the zero flux condition (Case 1) used in Section 3.2.1 are tested. Case 2 assumes a gravitational flow at the bottom of the soils. However given the shallow soils encountered in the catchment (average soil depth of 20 cm), this hypothesis does not seem very realistic. Two other cases mimic the existence of a permeable fractured bedrock. They are simulated

Fig. 5. Comparison between the estimated (empty squares) or the observed (empty diamond) peak discharges with the simulated peak discharges (black triangle) at the 32 validation outlets. The range of uncertainties given by Gaume and Bouvier (2004) is indicated. The x-axis (Surface, km2) is given as an indication.

141

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147 Table 2 Simulated peak discharges, runoffs and runoff coefficients for three events on the Saumane catchment. The results of several bottom boundary conditions are presented: Case 1: zero flux; Case 2: gravitational flow; Case 3: gravitational flow at the bottom of an additional 2 m thick layer with porosity 0.1 and hydraulic conductivity Ks = 0.4 mm h1; Case 4: same as Case 3 with Ks = 4 mm h1. Maximum discharge (m3 s1)

a

Runoff (mm)

Runoff coefficient (–)

Nash coefficient

September 2000 (Catchment’s rainfall 187 mm) Observation 273 71 Case 1 739 162 Case 2 264 21 Case 3 711 139 Case 4 392 47

0.38 0.87 0.11 0.74 0.25

– 3.70 0.38 2.55 0.62

September 2002 (Catchment’s rainfall 145 mm) Observation 770 85 Case 1 1062 112 Case 2 660 40 Case 3 1055 108 Case 4 788 63

0.59 0.77 0.28 0.75 0.43

0.55 0.65 0.50 0.73

November 2002 Observation Case 1 Case 2 Case 3 Case 4

0.54 0.78 0.31 0.54 0.78

3.61 1.18 1.22 0.70

(Catchment’s rainfall 70 mm) 147 38a 350 55 13 22 301 38 349 55

After removing an average base flow of 15 m3 s1.

by adding a 2 m depth soil horizon to the existing soils, with a porosity of 0.10 and a gravitational flow condition at the bottom. Two values for the saturated hydraulic conductivity, namely 0.4 mm h1 (Case 3) and 4 mm h1 (Case 4) are tested. The results of the simulation are summarized in Table 2 and the simulated discharges are compared to the observations retrieved from the OHMCV database in Fig. 6. Fig. 6 and Table 2 show that, Cases 1 and 3 lead to an overestimation of both the peak discharge and the runoff

volume. Case 2 leads to an underestimation of both, whereas Case 4 provides satisfactory agreement between the simulated and observed discharges, and with the observed runoff coefficient (Table 2). On September 2000, the Nash efficiency is negative for Cases 1 and 3, and positive for Cases 2 and 4, with a fair agreement with Case 4. All the Nash values are positive for the September 2002 event, with the highest values for Cases 2 and 4. All the Nash coefficients are negative for the November 2002 event, with a simulated hydrograph very different from the observed one. Contrary to the two previous events, a significant base flow is present and the recession is lasting for almost one day. The model does not reproduce the recession as expected because lateral subsurface flow is not yet included in the model. For the September 2000 and 2002 events, the simulated recession is also too fast, but the differences in volume are small. For the September 2002 event, Braud et al. (2010) show that the weight of lateral subsurface flow is negligible on simulated peak discharge and runoff coefficient. This sensitivity study to the soil boundary condition confirms the open question of the imperviousness of the bedrock. It highlights the need to better document the soil properties and thickness over the whole soil profile and not only over the top 2 m. On the gauged catchment, the zero bottom flux condition leads to an overestimation of the peak discharge, which could be a general feature of this model configuration, as shown by the general overestimation of maximum peak discharge in the regional evaluation based on post event field data (Section 3.2.1). Nevertheless, without more detailed information of the bedrock, the simple zero bottom flux hypothesis is used. 3.2.3. Influence of the Manning roughness coefficient of the river reach on the simulated discharges A sensitivity study of the Manning roughness coefficient to the simulated discharge is performed for the two medium sized catchments, Saumane and Uzès (Fig. 7 and Table 3). This parameter is

(b)

(a)

Sep-28 12:00

Sep-29 12:00

Sep-08 12:00

Sep-09 00:00

Sep-09 12:00

(c)

Nov-24 12:00

Nov-25 12:00

Fig. 6. Comparison between the simulated (lines) and the observed (symbol) discharges at the Saumane outlet (99 km2) for the (a) September 2000; (b) September 2002 and (c) November 2002 events. The results of several bottom boundary conditions are presented: Case 1: zero flux; Case 2: gravitational flow; Case 3: gravitational flow at the bottom of an additional 2 m thick layer with porosity 0.1 and hydraulic conductivity Ks = 0.4 mm h1; Case 4: same as Case 3 with Ks = 4 mm h1. The horizontal lines (September 2002 panel) refer to the estimation uncertainties given by Gaume and Bouvier (2004).

142

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

(a)

Sep-08 12:00

(b)

Sep-09 00:00

Sep-09 12:00

Sep-08 12:00

Sep-09 12:00

Fig. 7. Sensitivity of the simulated discharge to the Manning roughness coefficient for (a) the Saumane and (b) the Uzès catchments. In panel (a), the simulated discharges are compared with the observed (symbol) and estimated (horizontal lines) discharges. The horizontal lines (panel a) and the vertical arrows (panel b) refer to the peak discharge estimates and their uncertainties (Gaume and Bouvier, 2004).

Table 3 Sensitivity of the simulated peak discharge and the time of the peak of the to the specification of the Manning roughness coefficient for the Saumane and Uzès catchments. The simulation are performed using the ST-AD3 rainfall estimation.

Saumane (total rainfall 208 mm) Peak discharge (m3 s1) Time of peak 09/09 at (UTC) Uzès (total rainfall 390 mm) First peak discharge (m3 s1) Time of first peak 09/08 at (UTC) Second peak discharge (m3 s1) Time of second peak 09/09 at (UTC)

Observation

n = 0.05 m1/3 s

n = 0.07 m1/3 s

n = 0.125 m1/3 s

770 [650– 1050] 06h00

1673

1559

1392

04h30

04h40

05h00

[130–290] 22h00

486 20h00

448 20h10

359 21h00

[190–400]

845

764

651

10h00

08h40

09h10

10h10

only influential on the flow routing, once it has reached the river. Its variation does not modify the runoff coefficient. On the Saumane catchment, its variation, from n = 0.05 s1 m1/ 3 to n = 0.125 s1 m1/3, leads to a 30 min delay of the peak and a 17% decrease of the peak. The shape of the mean peaks of the hydrographs is only slightly modified (Fig. 7a). The impact is larger for the Uzès catchment with a decrease of 24% of the peak discharge when increasing n from 0.05 to 0.125 s1 m1/3 and the peak is delayed by 1 h and 30 min. The shape of the hydrographs is more significantly changed at the beginning with the simulation of several secondary peaks with the lowest values of the Manning roughness coefficients. When n = 0.125 s1 m1/3, these spurious secondary peaks are smoothed and the main peak is more in agreement with the observations from the post-flood investigation. For

the Uzès catchment, these data shows that the use of n = 0.125 s1 m1/3 reduces the discrepancy between the estimated and the simulated discharges. This value is kept in the following analysis for the Uzès catchment, whereas the default n = 0.05 s1 m1/3 is used for the Saumane catchment, as we have shown in the previous section that it provides reasonable agreement with observations. A global sensitivity analysis was performed for the September 2002 event using the latin hypercube method with 20 intervals for each parameter. The results are discussed in Braud et al. (2010). Four parameters were considered: (1) a multiplicative factor for saturated hydraulic conductivity, (2) a multiplicative factor for soil depth, (3) the Manning coefficient and (4) the initial saturation. The conclusions suggest that a better knowledge of soil depths, hydraulic properties and initial conditions are required at regional scale to strengthen our physically-based modeling approach and to improve our understanding of flash-flood genesis.

4. Modeling Saumane and Uzès watershed responses to the September 2002 storm 4.1. Influence of the rainfall field estimations on the simulated discharges 4.1.1. Impact of the radar protocol The two radar rainfall assessments (OPER and ST-AD3) and the kriged rainfall fields (KRIG) are used as input of the hydrological model to study the sensitivity of the simulated catchment responses to the rainfall estimations. In Fig. 8, the ratio between the simulated (Qs_s) and the estimated (Qs_e) maximum specific peak discharges is plotted as a function of the size of the catchment. The simulations are performed for the catchments used for the evaluation in Section 3.2 and the Uzès and Saumane basins.

143

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

Fig. 8. Ratio between the simulated and the estimated peak discharges as a function of catchment size. The estimations and the associated uncertainty (dotted horizontal lines) come from the post-flood investigation (Gaume and Bouvier, 2004). The CVN simulated discharges are compared with the results obtained with n-TOPMODELs using the numerical configuration proposed by Le Lay and Saulnier (2007).

As reported in Gaume and Bouvier (2004), the uncertainties associated with the estimated peak discharge range from 5% to 60% with a mean value of 30% represented on the graph. In Braud et al. (2010), peak discharge times, derived from the witness interviews, are given whenever available. They showed that the simulated peak discharge times remained within the range of uncertainty of the estimated time. Fig. 8 suggests the following comments. The compared values are of the same order of magnitude and, for the two radar rainfall inputs, more than 55% of the simulated peak discharges are within the range of uncertainty identified by Gaume and Bouvier (2004). Moreover, for the 13 simulated catchments (#1, #3, #4, #5, #8, #9, #13, #14, #15, #17, #19, #20, #29), the correlation coefficients between simulated and observed discharges are fairly good (i.e. 0.88 when KRIG rainfall estimation is used; 0.82 for OPER and 0.79 for ST-AD3). For the same catchments, the correlation coefficient reaches 0.78 with n-TOPMODELs using the configuration as detailed in Le Lay and Saulnier (2007) and forced with the ST-AD3 rainfall estimation. The results of the discharge simulation at the Saumane and Uzès outlets are slightly out of the uncertainty range, except for Saumane when the OPER rainfall estimation is used. But, CVN results remain better than the ones obtained with n-TOPMODELs. Obviously, the rainfall input has an important effect on the simulated discharges. In general, the radar rainfall inputs (OPER and

ST-AD3) lead to an overestimation of the simulated discharges regardless of the size of the catchment. The kriged rainfall inputs lead to different trends depending on the size of the catchment. The KRIG simulated discharges are overestimated for small basins (S < 10 km2) while they are underestimated for larger catchments. The hourly time step mainly explains this result. For example, the KRIG and ST-AD3 estimations of the water depth at Saumane catchment are approximately the same (205 mm vs. 208 mm) but the simulated discharges differ by more than 20%. Therefore, the CVN model is very sensitive to the rainfall time step. The largest overestimation of the simulated discharge is observed for sub-catchments #13 (S = 14.6 km2); #20 (S = 20.5 km2); #22 (S = 32.2 km2) and #26 (S = 39.5 km2). As mentioned above and shown by Bonnifait et al. (2009), the ST-AD3 rainfall estimator overestimates the rainfall fields in the region where the gradients of the rainfall is the strongest. This overestimation may be linked to the rain separation algorithm, which does not properly identify the rain type in this particular region. Therefore, the Z–R relationship is probably not appropriate. The ST-AD3 rainfall estimation leads, in general, to larger discrepancies between the estimated and the simulated specific discharges than the OPER estimator. This overestimation also occurs in the mountainous area (Saumane) where the space–time adaptive strategy is expected to give better rainfall estimations than the operational non-adaptive processing strategy.

Table 4 Characteristics of the study basins, with catchment area, lag time and mean cumulated precipitation at four different spatial and temporal rainfall resolutions. Area (km2)

Saumane Sub-catchment Sub-catchment Sub-catchment Sub-catchment Sub-catchment Sub-catchment Sub-catchment Uzès Sub-catchment Sub-catchment Sub-catchment Sub-catchment

Tc (h)

Mean areal precipitation (mm) at 5-mn resolution and

Mean areal precipitation (mm) at 1-km resolution and

1-km

2-km

4-km

8-km

5-mn

10-mn

30-mn

60-mn

198 163 157 149 117 268 112 280

208 169 160 156 120 288 106 318

208 169 160 156 120 288 105 317

208 169 160 156 120 288 106 318

208 169 160 156 120 288 106 318

397 394 402 378 386

390 385 400 352 380

391 385 400 352 380

397 394 401 378 388

1 2 3 4 5 6 7

99 66.3 59.6 51.2 34.6 30.9 20 12.9

1.25 1.0 0.94 0.87 0.7 0.66 0.52 0.41

208 169 160 156 120 288 106 318

207 169 161 156 121 285 106 313

204 172 165 160 122 269 110 295

1 2 3 4

68.4 43.2 22.4 19.8

1.02 0.79 0.55 0.52

397 394 402 378 386

395 393 400 378 389

392 392 396 384 391

144

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

These results demonstrate that the evaluation of the radar rainfall products varies whether they are based on radar pixel/gauge comparisons on the regional scale (Delrieu et al., 2009a,b) or on the hydrological responses, especially for small catchments in hilly areas.

4.1.2. Impact of the rain space and time resolution The influence of rainfall representation on the modeling of the hydrologic response is expected to depend on the spatial scale of the problem and on the complex interactions between the rainfall variability and the soil variability (Sangati and Borga, 2009; Sangati

Fig. 9. Relative difference of the simulated peak discharge at the Saumane and Uzès outlets as a function of (a) the space and (b) time scales of the input rainfall data. The STAD3 rainfall estimation at 1-km2 and 5-mm is considered as the reference. The space aggregation (Lr) is fixed to 2, 4 and 8-km. The time aggregation (Tr) is fixed to 10, 30 and 60-min. The Lc catchment length scale is fixed to the square root of the size of the catchment. The expression of the Tc catchment lag time is given by Creutin et al. (2009). The results of Sangati et al. (2009) are marked with a star.

145

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

et al., 2009). In general, when addressing the problem of the sampling scale effect, it is found that the measured variability of spatially continuous fields depends on two terms: extent and support (Blöschl and Sivapalan, 1995; Sangati and Borga, 2009). The extent term refers to the overall coverage of the data (i.e. the catchment scale, given by Lc , root square of the catchment area) whereas the support term refers to the resolution of the rainfall (the aggregation length, Lr). In this paper, this concept is extended to the time scales. Therefore, the impact of the lag time of the catchment (Tc) and the time resolution of the rainfall (Tr) on the simulated discharge is also examined in the following. The effect of rainfall aggregation in space and time on flood response modeling is examined here using the ratio of rainfall resolution to the characteristic basin scales (Lr/Lc and Tr/Tc). To point out the impact of rainfall aggregation in space and in time on model simulation, the CVN model is run over Uzès and Saumane. Four rainfall horizontal resolutions (1, 2, 4 and 8 km; i.e. 1, 2, 4 and 8 radar pixels) and four rainfall resolutions in time (5, 10, 30 and 60 min; i.e. 1, 2, 6 and 12 radar time step) are used. Since we have access to the hydrographs of all the REWs, five sub-catchments of Uzès and seven sub-catchments of Saumane (ranging from 10 km2 to 90 km2) have been chosen. Among the different possibilities for the definition of the hydrological characteristic time (Tc) of a basin (Morin et al., 2001; Berne et al., 2004), we chose the lag-time limit identified by Creutin et al. (2009) for small-scale catchments prone to flash-flood. They proposed to link the catchment response time with the watershed size, S, using a power function expressed as follow:

T c ¼ 0:1 S0:55

ð1Þ

The characteristics of the study basins are reported in Table 4 as well as the impact of the space and time aggregation on the catchment and sub-catchment average precipitation.

The runoff analysis is carried out for peak discharges considering the results obtained from 1-km grid size and 5-min time step as the reference (Qref). The rainfall time step is fixed to 5-min for the analysis of the spatial resolution of the rainfall whereas for the analysis of the temporal resolution of the rainfall, the grid size remains fixed at 1-km. The normalized peak discharge difference is computed as follow:

e¼

jQ ref  Q r j Q ref

where Qr represents the peak discharge resulting from aggregation over either Lr or Tr. Fig. 9a and b illustrate the impact of reduced space and time rainfall variability, respectively, on the simulated peak discharge. The results of a similar study (Sangati et al., 2009) performed on Italian catchments are added for comparison. As expected, the figures show that the normalized peak difference increases when smoothing both variabilities. However, smoothing in time (Tr = 60 min) produces larger errors than smoothing in space (Lr = 8 km). The Saumane catchment is more sensitive than the catchment of Uzès; this is due to its shallower depth and its structure that is more homogeneous. Moreover, the small size of the two catchments and their short response time probably explain why the peak time appears more sensitive to the rainfall time steps rather than to the spatial resolution. These results suggest the following comments. When rainfall is averaged in space or in time, rainfall intensity is dampened. Therefore, infiltration excess runoff may be reduced, according to the topsoil saturated hydraulic conductivity. More rainfall can thus infiltrate, leading to the reduction of the peak discharge. When the soil is (fully) saturated, discharge is directly related with the

(a)

Sep-08 12:00

ð2Þ

(b)

Sep-09 12:00

Sep-08 12:00

Sep-09 12:00

Fig. 10. Simulated discharges at (a) the Saumane and (b) Uzès outlets as compared with post-flood estimations and n-TOPMODELs simulations. Simulations with CVN are run for three rainfall inputs, OPER, ST-AD3, KRIG with distributed soils and the ST-AD3 radar data with a uniform soil, corresponding to the dominant soil, all over the catchment. The simulated discharges are compared with the observed (symbol, Saumane catchment) and estimated (horizontal lines, Saumane catchment; arrows, Uzès catchment) maximum peak discharges.

146

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147

rainfall intensity and discharge peak will also be dampened when rainfall is averaged in both space and time. 4.2. Influence of the soil description on the simulated discharge To assess the influence of soil description, a simplified simulation was performed for the Saumane and Uzès catchments. The whole catchments were described using only their major soil type. The ‘‘ST-AD3” estimation of rainfall was used as input, at 5-min and 1-km resolution. The use of the ‘‘OPER” estimation leads to the same results (not shown). In Fig. 10, the comparison between the simulated discharges shows a sensitive impact of soil properties on the discharge. This impact is restricted to the initial response of the basins. This impact is more visible on the Uzès catchment than at the Saumane outlet. It is therefore higher where the soil variability is the highest. When the catchment of Uzès is described with one single type of soil, the value of the first peak is 60% larger than when the soil variability is fully prescribed. As the same initial conditions were applied in both cases, the difference is essentially explained by a difference in the initial soil water deficit. Note that with the distributed soil, the maximum storage capacity of the catchment is 307 mm, whereas it is only 71 mm when the uniform dominant soil is used. A 60% initial saturation leads to a soil water deficit of 111 mm with the distributed soil, which is larger than the maximum capacity of the uniform soil. Thus, in this case, it is not possible to perform a simulation with identical initial soil water deficits. Later in the day of the 9th September, the event is so intense that the soil remains fully saturated whatever the soil properties. This leads to almost no difference between the two simulated discharges. It is important to note that the impact of the soil properties on the simulated discharges is of the same order of magnitude as the impact of the rainfall estimation, even larger during the initial phase of the storm. 5. Conclusion This paper presents a first step towards the set up of a modeling approach for small to medium size catchments. Most of the time, such catchments are ungauged and the hydrological forecasts usually lack observations. In this study, we propose a distributed hydrological model set up with the available data and a priori hydrological knowledge. Rainfall remains the first influential factor on the hydrological responses and high-resolution radar estimates, both in time and space, are needed to properly simulate flashflood. It is suggested to evaluate radar product not only in terms of radar pixel/raingauge comparison but also to quantify the impact of various radar treatment on the catchment outflow for a large range of basin scales. This study also highlights the importance of the soil description to properly set up the hydrological model. It is known that soil characteristics are an important source of uncertainty in the understanding of the hydrological behavior of the catchments. The results suggest that the flood dynamics presents two phases. The first one is mainly controlled by the soil properties. If they are simplified in the hydrological modeling, the results show that the resulting mis-evaluation of the maximum storage capacity of the catchment leads to large error on the simulated discharge. Once the soil is saturated, the rainfall variability controls the flood dynamics. This is the second phase. Therefore, the description of soil spatial variability should thus be enhanced to understand better the first phase, while the influence of the rainfall variability impacts both phases.

So, to better describe ungauged basins, long-term field experiments at regional scale are necessary (i) to document the soil characteristics to understand the hydrological response for a large range of basin sizes, (ii) to systematically organize post-flood investigation to enhance the database of the discharges in small ungauged catchments and, (iii) to improve the rainfall estimation. This is the challenge of the forthcoming HyMeX5 experiment. One of the objectives is to better understand flash-flood events over the Mediterranean region and to improve the predictive capability of hydrometeorological models in simulating and anticipating them. The observation strategy is based on several hydrometorological observatories (Crete, Greece; Liguria and Piemonte, Italy; OHM-CV, France; Catalonia, Spain). The hydrometeorological observation strategy will be approximately the same in all these observatories. It is therefore important to determine where and when observations are needed. Acknowledgements The work presented here is funded by the HYDRATE European Commission FP6 project under the n°GOCE 37024. Radar data and part of the rainfall data were provided by Météo-France. We acknowledge Electricité de France and the flood forecasting service SPC-GD for the availability of the discharge data and the rest of rainfall data. We deeply thank the two anonymous reviewers and the editor that help to improve the manuscript. References Anquetin, S., Creutin, J.-D., Delrieu, G., Ducrocq, V., Gaume, E., Ruin, I., 2004. Increasing the forecasting lead-time of weather driven flash floods. . Berne, A., Delrieu, G., Creutin, J.-D., Obled, C., 2004. Temporal and spatial resolution of rainfall measurements required for urban hydrology. J. Hydrol. 299, 166–179. Blöschl, G., Sivapalan, M., 1995. Scale issues in hydrological modelling: a review. Hydrol. Process. 9, 251–290. Board on Atmospheric Sciences and Climate (BASC), 2005. Flash Flood Forecasting Over Complex Terrain. The National Academies Press, . Bonnifait, L., Delrieu, G., Le Lay, M., Boudevillain, B., Masson, A., Belleudy, P., Gaume, E., Saulnier, G.M., 2009. Distributed hydrologic and hydraulic modelling with radar rainfall input: reconstruction of the 8–9 September 2002 catastrophic flood event in the Gard region, France. Adv. Water Resour. 37 (7), 1077–1089. Borga, M., Gaume, E., Creutin, J.D., Marchi, L., 2008. Surveying flash floods: gauging the ungauged extremes. Hydrol. Process. 22, 3883–3885. Bouilloud, L., Delrieu, G., Boudevillain, B., Kirstetter, P.E., 2010. Radar rainfall estimation in the context of post-event analysis of flash-flood events. J. Hydrol. 394 (1–2), 17–27. doi:10.1016/j.jhydrol.2010.02.035. Brakensiek, D.L., Engleman, W.L., Rawls, W.L., 1981. Variation within texture classes of soil water parameters. Trans. ASAE 24 (2), 335–339. Branger, F., Braud, I., Debionne, S., Viallet, P., Dehotin, J., Henine, H., Nedelec, Y., Anquetin, S., 2010, Towards multi-scale integrated hydrological models using the LIQUIDÒ framework. Overview of the concepts and first application examples. Environ. Modell. Softw. doi:10.1016/j.envsoft.2010.06.005. Braud, I., Roux, H., Anquetin, S., Maubourguet, M.M., Manus, C., Viallet, P., Dartus, D., 2010. The use of distributed hydrological models for the Gard 2002 flash-flood event understanding. Part II: Analysis of the associated hydrological processes. J. Hydrol. 394 (1–2), 162–181. doi:10.1016/j.jhydrol.2010.03.033. Carpenter, T.M., Georgakakos, K.P., 2004. Impacts of parametric and radar rainfall uncertainty on the ensemble streamflow simulations of a distributed hydrologic model. J. Hydrol. 2998, 202–221. Chancibault, K., Anquetin, S., Ducrocq, V., Saulnier, G.M., 2006. Hydrological evaluation of high-resolution precipitation forecasts of the Gard flash-flood event (8–9 September 2002). Quart. J. Roy. Meteorol. Soc. 132, 1091–1117. Cole, S.J., Moore, R.J., 2008. Hydrological modelling using raingauge- and radarbased estimators of areal rainfall. J. Hydrol. 358, 159–181. Collier, C., 2007. Flash flood forecasting: what are the limits of predictabilities? Quart. J. Roy. Meteorol. Soc. 133 (622A), 3–23. Creutin, J.D., Borga, M., 2003. Radar hydrology modifies the monitoring of flash flood hazard. Hydrol. Process. 17 (7), 1453–1456. Creutin, J.D., Obled, C., 1982. Objective analysis and mapping techniques for rainfall fields: an objective comparison. Water Resour. Res. 18 (2), 413–431. 5

HYdrological cycle in Mediterranean EXperiment, http://www.hymex.org.

S. Anquetin et al. / Journal of Hydrology 394 (2010) 134–147 Creutin, J.-D., Borga, M., Lutoff, C., Scolobig, A., Ruin, I., Créton-Cazanave, L., 2009. Catchment dynamics and social response during flash floods: the potential of radar rainfall monitoring for warning procedures. Meteorol. Appl. 16, 115–125. Dehotin, J., Braud, I., 2008. Which spatial discretization for distributed hydrological models? Proposition of a methodology and illustration for medium to largescale catchments. Hydrol. Earth Syst. Sci. 12, 769–796. Delrieu, G., Ducrocq, V., Gaume, E., Nicol, J., Payrastre, O., Yates, E., Kirstetter, P.-E., Andrieu, H., Ayral, P.-A., Bouvier, C., Creutin, J.-D., Livet, M., Anquetin, S., Lang, M., Neppel, L., Obled, C., Parent-du-Châtelet, J., Saulnier, G.-M., Walpersdorf, A., Wobrock, W., 2005. The catastrophic flash-flood event of 8–9 September 2002 in the Gard region, France: a first case study for the Cévennes–Vivarais mediterranean hydro-meteorological observatory. J. Hydrometeorol. 6, 34–52. Delrieu, G., Boudevillain, B., Nicol, J., Chapon, Kirstetter, P.-E., Andrieu, H., Faure, D., 2009a. Bollène-2002 experiment: radar quantitative precipitation estimation in the Cévennes–Vivarais region. J. Appl. Meteorol. Climatol. doi:10.1175/ 2008JAMC1987.1. Delrieu, G., Braud, I., Berne, A., Borga, M., Boudevillain, B., Fabry, F., Freer, J., Gaume, E., Nakakita, E., Seed, A., Tabary, P., Uijlenhoet, R., 2009b. Weather radar and hydrology. Adv. Water Resour. 37 (7), 969–974. Foody, G.M., Ghoniem, E.M., Arnell, N.W., 2004. Predicting locations sensitive to flash flood in an arid environment. J. Hydrol. 292 (1–4), 48–58. Furman, A., 2008. Modeling coupled surface–subsurface flow processes: a review. Vadose Zone J. 7, 741–756. doi:10.2136/vzj2007.0065. Gaume, E., Bouvier, C., 2004. Analyse hydro-pluviométrique des crues du Gard et du Vidourle des 8 et 9 septembre 2002. La Houille Blanche 6, 99–106. doi:10.1051/ lhb:200406014. Gaume, E., Payrastre, O., Rosa da Silvia, B., 2003. Analyse hydrologique des crues des 8 et 9 Septembre 2002 dans le Gard, Report Ministère de l’écologie et du développement durable, Convention 02000246. Gaume, E., Bain, V., Bernardara, P., Newinger, O., Barbuc, M., Bateman, A., Blaskovicová, L., Blöschl, G., Borga, M., Dumitrescu, A., Daliakopoulos, I., Garcia, J., Irimescu, A., Kohnova, S., Koutroulis, A., Marchi, L., Matreata, S., Medina, V., Preciso, E., Sempere-Torres, D., Stancalie, G., Szolgay, J., Tsanis, I., Velasco, D., Viglione, A., 2009. A compilation of data on European flash floods. J. Hydrol. 367 (1–2), 70–78. Georgakakos, K.P., 2006. Analytical results for operational flash flood guidance. J. Hydrol. 317 (1–2), 81–103. Germann, U., Galli, G., Boscacci, M., Bolliger, M., 2006. Radar precipitation measurement in a mountainous region. Quart. J. Roy. Meteorol. Soc. 132, 1669–1692. Habets, F., Boone, A., Champeaux, J.L., Etchevers, P., Franchistéguy, L., Leblois, E., Ledoux, E., Le Moigne, P., Martin, E., Morel, S., Noilhan, J., Quintana-Segui, P., Rousset-Regimbeau, F., Viennot, P., 2008. The SAFRAN–ISBA–MODCOU hydrometerological model applied over France. J. Geophys. Res. 113, D06113. doi:10.1029/2007JD008548. Hernandez, E., Cana, L., Diaz, J., Garcia, R., Gimeno, L., 1998. Mesoscale convective complexes over western Mediterranean area during 1990–1994. Meteorol. Atmos. Phys. 68, 1–12. Homar, V., Romero, R., Ramis, C., Alonso, S., 1999. A case of convection development over the western Mediterranean Sea: a study through numerical simulations. Meteorol. Atmos. Phys. 71, 169–188. Joss, J., Waldvogel, A., 1990. Precipitation measurement and hydrology. Radar Meteorol. D. Atlas, Ed. Am. Meteorol. Soc. 577–606. Kampf, S.K., Burges, S.J., 2007. A framework for classifying and comparing distributed hillslope and catchment hydrologic models. Water Resour. Res. 43, W05423. doi:10.1029/2006WR005370. Latron, J., Gallart, F., 2007. Seasonal dynamics of runoff-contributing areas in a small mediterranean research catchment (Vallcebre, Eastern Pyrenees). J. Hydrol. 335 (1–2), 194–206. Le Lay, M., Saulnier, G.-M., 2007. Exploring the signature of climate and landscape spatial variabilities in flash flood events: case of the 8–9 September 2002 Cévennes–Vivarais catastrophic event. Geophys. Res. Lett. 34 (13), L13401. Lebel, T., Bastin, G., Obled, C., Creutin, J.D., 1987. On the accuracy of areal rainfall estimation: a case study. Water Resour. Res. 23 (11), 2123–2134. Manus, C., Anquetin, S., Braud, I., Vandervaere, J.-P., Creutin, J.-D., Viallet, P., Gaume, E., 2009. A modeling approach to quantify the hydrological response of small Mediterranean catchments to the variability of soil characteristics in a context of extreme events. Hydrol. Earth Syst. Sci. 13, 79–97.

147

Molinié, G., Ceresetti, D., Anquetin, S., Creutin, J.D., Boudevillain, B., submitted for publication. Rainfall regimes in a mountainous mediterranean region: statistical analysis at short time steps. J. Appl. Meteorol. Climatol. Montz, B.E., Gruntfest, E., 2002. Flash flood mitigation: recommendations for research and applications. Environ. Hazards 4, 15–22. Morin, E., Enzel, Y., Shamir, U., Garti, R., 2001. The characteristic time scale for basin hydrological response using radar data. J. Hydrol. 252 (1–4), 85–99. Morin, E., Goodrich, D.C., Maddox, R.A., Gao, X., Gupta, H.V., Sorooshian, S., 2006. Spatial patterns in thunderstorm rainfall events and their coupling with watershed hydrological response. Adv. Water Resour. 29, 843–860. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models. Part 1: A discussion of principles. J. Hydrol. 10 (3), 282–290. Nicotina, L., Alessi-Celegon, E., Rinaldo, A., Marani, M., 2008. On the impact of rainfall patterns on the hydrologic response. Water Resour. Res. 44, W1240. doi:10.1029/2007WR006654. Norbiato, D., Borga, M., Esposti, S.D., Gaume, E., Anquetin, S., 2008. Flash flood warning based on rainfall thresholds and soil moisture conditions: an assessment for gauged and ungauged basins. J. Hydrol. 362, 274–290. Pellarin, T., Delrieu, G., Saulnier, G.M., Andrieu, H., Vignal, B., Creutin, J.D., 2002. Hydrologic visibility of weather radars operating in mountainous regions: case study for the Ardèche catchment (France). J. Hydrol. 3, 539–555. Rawls, W.J., Brakensiek, D.L., 1985. Prediction of soil water properties for hydrologic modelling. In: Jones, E.B., Ward, T.J. (Eds.), Watershed Management in the Eighties: Proceedings of the American Society of Civil Engineers symposium, Denver, April 30–May 1. ASCE, New York, pp. 293–299. Reed, S., Schaake, J., Zhang, Z., 2007. A distributed hydrologic model and threshold frequency-based method for flash flood forecasting at ungauged locations. J. Hydrol. 337 (3–4), 402–420. Reggiani, P., Sivapalan, M., Hassanizadeh, S.M., 1998. A unifying framework for watershed thermodynamics: balance equations for mass, momentum, energy and entropy, and the second law of thermodynamics. Adv. Water Resour. 22, 367–598. Ross, P.J., 2003. Modeling soil water and solute transport – fast, simplified numerical solutions. Agron. J. 95, 1352–1361. Ruin, I., Creutin, J.D., Anquetin, S., Lutoff, C., 2008. Human exposure to flash floods – relation between flood parameters and human vulnerability during a storm of September 2002 in Southern France. J. Hydrol. 361 (1–2), 199–213. Sangati, M., Borga, M., 2009. Influence of rainfall spatial resolution on flash flood modelling. Nat. Hazards Earth Syst. Sci. 9, 575–584. Sangati, M., Borga, M., Rabuffeti, D., Bechini, R., 2009. Influence of rainfall and soil properties spatial aggregation on extreme flash flood response modelling: an evaluation based on the Sesia river basin, North Western Italy. Adv. Water Resour. 32, 1090–1106. Tetzlaff, D., Uhlenbrook, S., 2005. Significance of spatial variability in precipitation for process-oriented modelling: results from two nested catchments using radar and ground station data. Hydrol. Earth Syst. Sci. 9, 29–41. Todini, E., 2007. Hydrological catchment modelling: past, present and future. Hydrol. Earth. Syst. Sci. 11 (1), 468–482. Varado, N., Ross, P.J., Braud, I., Haverkamp, R., 2006. Assessment of an efficient numerical solution of the Richards’ equation for bare soil. J. Hydrol. 323 (1–4), 244–257. Viallet, P., Debionne, S., Braud, I., Dehotin, J., Haverkamp, R., Saâdi, Z., Anquetin, S., Branger, F., Varado, N., 2006. Towards multi-scale integrated hydrological models using the LIQUID framework. In: 7th International Conference on Hydroinformatics 2006, 4–8 September, Nice, France, vol. 1. pp. 542–549. . Westrick, K.J., Mass, C.F., Colle, B.A., 1999. The limitations of the WSR-88D radar network for quantitative precipitation measurement over the coastal Western United States. Bull. Am. Meteorol. Soc. 80 (1), 2289–2298. Yates, D.N., Warner, T.T., Leavesley, G.H., 2000. Prediction of flash flood in complex terrain. Part II: a comparison of flood discharge simulations using input from radar, a dynamic model and an automated algorithm system. J. Appl. Meteorol. 39, 815–825. Younis, J., Anquetin, S., Thielen, J., 2008. The benefit of high-resolution operational weather forecasts for flash-flood warning. Hydrol. Earth Syst. Sci. 12, 1039– 1051.