ALBERT

Description: This paper describes and illustrates a methodology to conduct postflood investigations based on interdisciplinary collaboration between social and physical scientists. The method, designed to explore the link between crisis behavioral response and hydrometeorological dynamics, aims at understanding the spatial and temporal capacities and constraints on human behaviors in fast-evolving hydrometeorological conditions. It builds on methods coming from both geosciences and transportations studies to complement existing postflood field investigation methodology used by hydrometeorologists. The authors propose an interview framework, structured around a chronological guideline to allow people who experienced the flood firsthand to tell the stories of the circumstances in which their activities were affected during the flash flood. This paper applies the data collection method to the case of the 15 June 2010 flash flood event that killed 26 people in the Draguignan area (Var, France). As a first step, based on the collected narratives, an abductive approach allowed the identification of the possible factors influencing individual responses to flash floods. As a second step, behavioral responses were classified into categories of activities based on the respondents' narratives. Then, aspatial and temporal analysis of the sequences made of the categories of action to contextualize the set of coping responses with respect to local hydrometeorological conditions is proposed. During this event, the respondents mostly follow the pace of change in their local environmental conditions as the flash flood occurs, official flood anticipation being rather limited and based on a large-scale weather watch. Therefore, contextual factors appear as strongly influencing the individual's ability to cope with the event in such a situation.

Description: This paper presents an initial prototype of a distributed hydrological model used to map possible road inundations in a region frequently exposed to severe flash floods: the Gard region (South of France). The prototype has been tested in a pseudo real-time mode on five recent flash flood events for which actual road inundations have been inventoried. The results are promising: close to 100% probability of detection of actual inundations, inundations detected before they were reported by the road management field teams with a false alarm ratios not exceeding 30%. This specific case study differs from the standard applications of rainfall-runoff models to produce flood forecasts, focussed on a single or a limited number of gauged river cross sections. It illustrates that, despite their lack of accuracy, hydro-meteorological forecasts based on rainfall-runoff models, especially distributed models, contain valuable information for flood event management. The possible consequences of landslides, debris flows and local erosion processes, sometimes associated with flash floods, were not considered at this stage of development of the prototype. They are limited in the Gard region but should be taken into account in future developments of the approach to implement it efficiently in other areas more exposed to these phenomena such as the Alpine area.

Description: In flash flood prone areas, roads are often the first assets affected by inundations which make rescue operations difficult and represent a major threat to lives: almost half of the victims are car passengers trapped by floods. In the past years, the Gard region (France) road management services have realized an extensive inventory of the known road submersions that occurred during the last 40 years. This inventory provided an unique opportunity to analyse the causes of road flooding in an area frequently affected by severe flash floods. It will be used to develop a road submersion susceptibility rating method, representing the first element of a road warning system. This paper presents the results of the analysis of this data set. A companion paper will show how the proposed road susceptibility rating method can be combined with distributed rainfall-runoff simulations to provide accurate road submersion risk maps. The very low correlation between the various possible explanatory factors and the susceptibility to flooding measured by the number of past observed submersions implied the use of particular statistical analysis methods based on the general principals of the discriminant analysis. The analysis led to the definition of four susceptibility classes for river crossing road sections. Validation tests confirmed that this classification is robust, at least in the considered area. One major outcome of the analysis is that the susceptibility to flooding is rather linked to the location of the road sections than to the size of the river crossing structure (bridge or culvert).

Description: This paper presents some analytical results and numerical illustrations on the asymptotic properties of flood peak distributions obtained through derived flood frequency approaches. It confirms and extends the results of previous works: i.e. the shape of the flood peak distributions are asymptotically controlled by the rainfall statistical properties, given limited and reasonable assumptions concerning the rainfall-runoff process. This result is partial so far: the impact of the rainfall spatial heterogeneity has not been studied for instance. From a practical point of view, it provides a general framework for analysis of the outcomes of previous works based on derived flood frequency approaches and leads to some proposals for the estimation of very large return-period flood quantiles. This paper, focussed on asymptotic distribution properties, does not propose any new approach for the extrapolation of flood frequency distribution to estimate intermediate return period flood quantiles. Nevertheless, the large distance between frequent flood peak values and the asymptotic values as well as the simulations conducted in this paper help quantifying the ill condition of the problem of flood frequency distribution extrapolation: it illustrates how large the range of possibilities for the shapes of flood peak distributions is.

Description: This paper investigates the influence of mean areal rainfall estimation errors on a specific case study: the use of lumped conceptual rainfall-runoff models to simulate the flood hydrographs of three small to medium-sized catchments of the upper Loire river. This area (3200 km2) is densely covered by an operational network of stream and rain gauges. It is frequently exposed to flash floods and the improvement of flood forecasting models is then a crucial concern. Particular attention has been drawn to the development of an error model for rainfall estimation consistent with data in order to produce realistic streamflow simulation uncertainty ranges. The proposed error model combines geostatistical tools based on kriging and an autoregressive model to account for temporal dependence of errors. It has been calibrated and partly validated for hourly mean areal precipitation rates. Simulated error scenarios were propagated into two calibrated rainfall-runoff models using Monte Carlo simulations. Three catchments with areas ranging from 60 to 3200 km2 were tested to reveal any possible links between the sensitivity of the model outputs to rainfall estimation errors and the size of the catchment. The results show that a large part of the rainfall-runoff (RR) modelling errors can be explained by the uncertainties on rainfall estimates, especially in the case of smaller catchments. These errors are a major factor limiting accuracy and sharpness of rainfall-runoff simulations, and thus their operational use for flood forecasting.

Description: The 2 June 2008 flood-producing storm on the Starzel river basin in South-West Germany is examined as a prototype for organized convective systems that dominate the upper tail of the precipitation frequency distribution and are likely responsible for the flash flood peaks in Central Europe. The availability of high-resolution rainfall estimates from radar observations and a rain gauge network, together with indirect peak discharge estimates from a detailed post-event survey, provided the opportunity to study in detail the hydrometeorological and hydrological mechanisms associated with this extreme storm and the ensuing flood. Radar-derived rainfall, streamgauge data and indirect estimates of peak discharges are used along with a distributed hydrologic model to reconstruct hydrographs at multiple locations. Observations and model results are combined to examine two main questions, (i) assessment of the distribution of the runoff ratio for the 2008 flash flood and how it compares with other less severe floods; and (ii) analysis of how the spatial and temporal distribution of the extreme rainfall, and more specifically storm motion, controls the flood response. It is shown that small runoff ratios (less than 20%) characterized the runoff response and that these values are in the range of other, less extreme, flood events. The influence of storm structure, evolution and motion on the modeled flood hydrograph is examined by using the "spatial moments of catchment rainfall". It is shown that downbasin storm motion (in the range of 0.7–0.9 m s−1) had a noticeable impact on flood response by increasing the modeled flood peak by 13%.

Description: This paper presents a modeling study aiming at quantifying the possible impact of soil characteristics on the hydrological response of small ungauged catchments in a context of extreme events. The study focuses on the September 2002 event in the Gard region (South-Eastern France), which led to catastrophic flash-floods. The proposed modeling approach is able to take into account rainfall variability and soil profiles variability. Its spatial discretization is determined using Digital Elevation Model (DEM) and a soil map. The model computes infiltration, ponding and vertical soil water distribution, as well as river discharge. In order to be applicable to ungauged catchments, the model is set up without any calibration and the soil parameter specification is based on an existing soil database. The model verification is based on a regional evaluation using 17 estimated discharges obtained from an extensive post-flood investigation. Thus, this approach provides a spatial view of the hydrological response across a large range of scales. To perform the simulations, radar rainfall estimations are used at a 1 km2 and 5 min resolution. To specify the soil hydraulic properties, two types of pedotransfer function (PTF) are compared. It is shown that the PTF including information about soil structure reflects better the spatial variability that can be encountered in the field. The study is focused on four small ungauged catchments of less than 10 km2, which experienced casualties. Simulated specific peak discharges are found to be in agreement with estimations from a post-event in situ investigation. Examining the dynamics of simulated infiltration and saturation degrees, two different behaviors are shown which correspond to different runoff production mechanisms that could be encountered within catchments of less than 10 km2. They produce simulated runoff coefficients that evolve in time and highlight the variability of the infiltration capacity of the various soil types. Therefore, we propose a cartography distinguishing between areas prone to saturation excess and areas prone only to infiltration excess mechanisms. The questions raised by this modeling study will be useful to improve field observations, aiming at better understanding runoff generation for these extreme events and examine the possibility for early warning, even in very small ungauged catchments.

Description: Recently Feed-Forward Artificial Neural Networks (FNN) have been gaining popularity for stream flow forecasting. However, despite the promising results presented in recent papers, their use is questionable. In theory, their “universal approximator‿ property guarantees that, if a sufficient number of neurons is selected, good performance of the models for interpolation purposes can be achieved. But the choice of a more complex model does not ensure a better prediction. Models with many parameters have a high capacity to fit the noise and the particularities of the calibration dataset, at the cost of diminishing their generalisation capacity. In support of the principle of model parsimony, a model selection method based on the validation performance of the models, "traditionally" used in the context of conceptual rainfall-runoff modelling, was adapted to the choice of a FFN structure. This method was applied to two different case studies: river flow prediction based on knowledge of upstream flows, and rainfall-runoff modelling. The predictive powers of the neural networks selected are compared to the results obtained with a linear model and a conceptual model (GR4j). In both case studies, the method leads to the selection of neural network structures with a limited number of neurons in the hidden layer (two or three). Moreover, the validation results of the selected FNN and of the linear model are very close. The conceptual model, specifically dedicated to rainfall-runoff modelling, appears to outperform the other two approaches. These conclusions, drawn on specific case studies using a particular evaluation method, add to the debate on the usefulness of Artificial Neural Networks in hydrology. Keywords: forecasting; stream-flow; rainfall-runoff; Artificial Neural Networks

Description: This paper investigates the influence of mean areal rainfall estimation errors on a specific case study: the use of lumped conceptual rainfall-runoff models to simulate the flood hydrographs of three small to medium-sized catchments of the upper Loire river. This area (3200 km2) is densely covered by an operational network of stream and rain gauges. It is frequently exposed to flash floods and the improvement of flood forecasting models is then a crucial concern. Particular attention has been drawn to the development of an error model for rainfall estimation consistent with data in order to produce realistic streamflow simulation uncertainty ranges. The proposed error model combines geostatistical tools based on kriging and an autoregressive model to account for temporal dependence of errors. It has been calibrated and partly validated for hourly mean areal precipitation rates. Simulated error scenarios were propagated into two calibrated rainfall-runoff models using Monte Carlo simulations. Three catchments with areas ranging from 60 to 3200 km2 were tested to reveal any possible links between the sensitivity of the model outputs to rainfall estimation errors and the size of the catchment. The results show that a large part of the rainfall-runoff (RR) modelling errors can be explained by the uncertainties on rainfall estimates, especially in the case of smaller catchments. These errors are a major factor limiting accuracy and sharpness of rainfall-runoff simulations, and thus their operational use for flood forecasting.

Description: This paper presents a modeling study aiming at quantifying the possible impact of soil characteristics on the hydrological response of small ungauged catchments in a context of extreme events. The study focuses on the September 2002 event in the Gard region (South-Eastern France), which led to catastrophic flash-floods. The proposed modeling approach is able to take into account rainfall variability and soil profiles variability. Its spatial discretization is determined using Digital Elevation Model (DEM) and a soil map. The model computes infiltration, ponding and vertical soil water distribution, as well as river discharge. The model is set up without any calibration and the soil parameter specification is based on an existing soil database. To perform the simulations, radar rainfall estimations are used at a 1 km2 and 5 min resolution. To specify the soil hydraulic properties, two types of pedotransfer function (PTF) are compared. It is shown that the PTF including information about soil structure reflects better the spatial variability that can be encountered in the field. The study is focused on four small ungauged catchments of less than 10 km2, which experienced casualties. Simulated specific peak discharges are found to be in agreement with estimations from a post-event in situ investigation. Examining the dynamics of simulated infiltration and saturation degrees, two different behaviors are shown which correspond to different runoff production mechanisms that could be encountered within catchments of less than 10 km2. They produce simulated runoff coefficients that evolve in time and highlight the variability of the infiltration capacity of the various soil types. Therefore, we propose a cartography distinguishing between areas prone to saturation excess and areas prone only to infiltration excess mechanisms. The questions raised by this modeling study will be useful to improve field observations, aiming at better understanding runoff generation for these extreme events and examine the possibility for early warning, even in very small ungauged catchments.