Search

International audience; The objective of flood frequency analysis (FFA) is to associate flood intensity with a probability of exceedance. Many methods are currently employed for this, ranging from statistical distribution fitting to simulation approaches. In many cases the site of interest is actually ungauged, and a regionalisation scheme has to be associated with the FFA method, leading to a multiplication of the number of possible methods available. This paper presents the results of a wide-range comparison of FFA methods from statistical and simulation families associated with different regionalisation schemes based on regression, or spatial or physical proximity. The methods are applied to a set of 1535 French catchments, and a k-fold cross-validation procedure is used to consider the ungauged configuration. The results suggest that FFA from the statistical family largely relies on the regionalisation step, whereas the simulation-based method is more stable regarding regionalisation. This conclusion emphasises the difficulty of the regionalisation process. The results are also contrasted depending on the type of climate: the Mediterranean catchments tend to aggravate the differences between the methods.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; International audience; On the 3rd October 2015, heavy localized precipitations have occurred in South Eastern France leading to major flash floods on the Mediterranean coast. The severity of those floods has caused 20 fatalities and important damage in almost 50 municipalities in the French administrative area of Alpes-Maritimes. The local recording rain gauges have shown how fast the event has happened: 156 mm of rain were recorded in Mandelieu-la-Napoule and 145 mm in Cannes within 2 hours. As the affected rivers are not monitored, no anticipation was possible from the authorities in charge of risk management. In this case, forecasting floods is indeed complex because of the small size of the watersheds which implies a reduced catchment response time. In order to cope with the need of issuing flood warnings on un-monitored small catchments, Irstea and Météo-France have developed an alternative warning system for ungauged basins called the AIGA method. AIGA is a flood warning system based on a simple distributed hydrological model run at a 1 km² resolution using real time radar rainfall information (Javelle, Demargne, Defrance, Pansu, & Arnaud, 2014). The flood warnings, produced every 15 minutes, result of the comparison of the real time runoff data produced by the model with statistical runoff values. AIGA is running in real time in the South of France, within the RHYTMME project (https://rhytmme.irstea.fr/). Work is on-going in order to offer a similar service for the whole French territory. More than 200 impacts of the 3rd October floods have been located using media, social networks and fieldwork. The first comparisons between these impacts and the AIGA warning levels computed for this event show several discrepancies. However, these latter discrepancies appear to be explained by the land-use. An indicator of the exposure of territories to flooding has thus been created to weight the levels of the AIGA hydrological warnings with the land-use of the area surrounding the streams for which the warnings are issued. This paper aims to explain how this indicator has been created and to assess its relevance with the example of the 3rd October 2015 flood. By completing this approach, the AIGA warnings may characterize not only the flood hazard but more inclusively the risk of flooding, allowing to forecast this type of event.

/ Le 3 octobre 2015, des précipitations d'une intensité exceptionnelle se sont abattues sur le sud-ouest de la France. Elles se sont traduites par des inondations localisées qui ont entraîné la mort de vingt personnes et d'importants dégâts matériels dans les Alpes-Maritimes. Avec un bilan provisoire estimé par la CCR entre 500 et 650 millions d'euros (Caisse Centrale de Réassurance, 2016), cet évènement se place au moins en quatrième position des catastrophes naturelles survenues en France, en termes de coûts économique et humain. L'ampleur des dégâts liés à cet évènement peut être mise en relation avec deux facteurs principaux. Premièrement, les Alpes-Maritimes sont un département où la densité humaine est particulièrement importante, notamment sur les zones côtières où sont regroupés la grande majorité des dégâts (Direction Territoriale Méditerranée du Cerema, 2015). Mais le manque d'anticipation a également été pointé du doigt lors de cet évènement. Les cumuls de pluie 2h relevés par les pluviographes de Météo-France témoignent de la rapidité du phénomène avec notamment 156mm à Mandelieu-la-Napoule et 175mm à Cannes (Javelle et al.). Deux communes qui concentrent à elles seules 11 des 20 décès survenus ce jour-là. La vigilance météorologique émise par Météo-France était alors de couleur orange sur le département. Pour autant, aucun des cours d'eau entrés en crue ce 3 octobre n'était concerné par une quelconque vigilance car, s'agissant de petits bassins versants côtiers aux temps de réaction très réduits, ils ne bénéficient pas du système de surveillance des cours d'eau Vigicrues (Gerard, 2013). Face à ces enjeux majeurs, le centre Irstea Aix-en-Provence a développé, en partenariat avec Météo-France, une méthode d'anticipation des crues rapides sur les petits bassins versants dite méthode AIGA (Javelle, Demargne, Defrance, Pansu, & Arnaud, 2014). Elle transforme les estimations de pluies provenant des radars météorologiques en débit des cours d'eau, en s'appuyant sur les caractéristiques régionales des bassins versants. Ces données sont confrontées en temps réel avec celles d'une base de référence sur les pluies et les débits, qui permet ainsi d'estimer le risque de crue associé. Pour compléter AIGA, une méthode expérimentale d'évaluation de l'exposition des territoires au risque inondation basée sur l'occupation du sol, a été créée. Ce poster s'intéresse donc à la combinaison de ces deux protocoles, qui est ici testée sur l'évènement du 3 octobre grâce à des données géolocalisées de dégâts qui vont permettre de valider leur pertinence.

International audience; In absence of streamflow data to calibrate a hydrological model, its parameters are to be inferred by a regionalization method. In this technical note, we discuss a specific class of regionalization methods, those based on spatial proximity, which transfers hydrological information (typically calibrated parameter sets) from neighbor gauged stations to the target ungauged station. The efficiency of any spatialproximity-based regionalization method will depend on the density of the available streamgauging network, and the purpose of this note is to discuss how to assess the robustness of the regionalization method (i.e., its resilience to an increasingly sparse hydrometric network). We compare two options: (i) the random hydrometrical reduction (HRand) method, which consists in sub-sampling the existing gauging network around the target ungauged station, and (ii) the hydrometrical desert method (HDes), which consists in ignoring the closest gauged stations. Our tests suggest that the HDes method should be preferred, because it provides a more realistic view on regionalization performance.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; National audience

International audience; Insufficient hydrological knowledge hampers the reanalysis of river quality measurements. For most of the pollution-control sites monitored by the French Water Agency, streamflow data are generally missing or are available only as few point gauging records. This paper presents a comparative study of methods of daily streamflow time series reconstruction based on rainfall-runoff modeling. Their performance is evaluated through data running from 1990 to 2009 over a rather dense streamflow monitoring network. Results show that a small number of point gauging measurements may help to significantly improve the reconstruction of daily streamflow data series for almost ungauged sites. Point gauging records are also useful to define the optimal neighborhood used during the regionalization process. At-site monitoring of chemical and physical status of surface water should additionally include a series of flow measurements well-distributed in time, especially in regions where hydrological data are scarce; L'exploitation rétrospective des données physico-chimiques de qualité des eaux courantes requiert la connaissance du contexte hydrologique (chroniques de débits) dans lequel s'effectue le prélèvement d'eau. Cette information hydrologique n'est pas toujours connue au point de prélèvement ou de façon seulement discrète (débits ponctuels). Cet article présente les résultats d'une analyse comparative de méthodes de calcul des débits des cours d'eau par simulation pluie-débit applicables au droit des stations du réseau de contrôle de surveillance (RCS) gérées par l'Agence de l'Eau Rhin-Meuse. L'étude repose sur un réseau de mesure hydrométrique relativement dense et un patrimoine de données hydrologiques couvrant la période 1990-2009. Les résultats de l'étude montrent que la disponibilité d'une information hydrométrique ponctuelle même parcimonieuse au point de prélèvement, permet d'améliorer significativement la reconstitution des chroniques de débits journaliers et la définition du voisinage hydrologique utilisé lors de leur reconstitution. Tout suivi physico-chimique des eaux courantes devrait donc être doublé d'un suivi hydrométrique minimal, indépendant du suivi physico-chimique de manière à couvrir toute la gamme de débits de la rivière et permettant de reconstituer par modélisation hydrologique une chronique de débits aux points des réseaux RCS dépourvus de cette information à caractère opérationnel ou patrimonial

In Europe, flash floods mainly occur in the Mediterranean area on small catchments with a short concentration time. Anticipating this kind of events is a major issue in order to reduce the resulting damages. But for many of the impacted catchments, no data are available to calibrate and evaluate hydrological models. In this context, the aims of this study is to develop and evaluate a warning method for the Southern French Alps. This area is of particular interest, because it regroups different hydrological regimes, from purely Mediterranean to purely Alpine influences. Two main issues should be addressed: - How to define the hydrological model and its parametrization for an application in an ungauged context? - How to evaluate the final results on 'real' ungauged catchments? The first issue is a classic one. Using a 'observed' data set (154 streamflow stations with catchment areas ranging from 5 to 1000 km2 and distributed rainfall available on the 1997-2006 period), we developed a regional model specifically for the studied area. For this purpose, the AIGA method, initially developed for Mediterranean catchments was adapted, in order to take into account snowmelt and to produce baseflows. Then, different parametrizations were tested, derived from different simple regionalisation techniques: - the same parameters set for the whole area defined as the median of the local calibrated parameters; - the same technique as the previous case, but by considering different sub-areas, defined as 'hydro-climatically' homogeneous by previous studies; - and finally the neighbour's method. The second issue is more original. Indeed, in most studies the final evaluation is done using gauged stations as they were 'ungauged', ie keeping the at-site discharge data only for validation ant not for calibration. The main disadvantage of this approach is that the evaluation is made at the scale of the gauged catchments, which are in general greater than the catchments impacted by flash floods. Furthermore, many events are missed, since flash floods can occur very locally. In this study, we try to evaluate the results on observations collected by witnesses on 'real' ungauged catchments. The proposed method consists to use an historical data-base of flood damages reports. These data have been collected by local authorities (RTM). Finally, 139 ungauged locations were considered, where we simulated discharges for the entire 1997-2006 period. The comparison of these modelled discharges with the occurrence of an observed discharge makes it possible to determine a local 'modelled' discharge threshold above it most of the damages are observed. The pertinence of this threshold (and consequently of the model used for the simulation) is assessed by considering classical contingency statistics: probability of detection (POD), false alarm rate (FAR) and critical success index (CSI). The main advantage of this historical approach is the availability of many events in the database on very small catchments (50% less than 20 km2). The preliminary results show that on gauged basins, the base flow and the snowmelt added modules improve the performance of the AIGA method when locally calibrated. But when results are applied on real ungauged catchments, improvements become less obvious, with a small advantage for neighbour's method. These results shows the difficulty arising with ungauged catchments, specially when target catchments are smaller than the gauged 'parents'. It also illustrates the interest of the damages database used as 'proxy' data to investigate the model performances at smaller scales. This work has been done in the framework of the RHYTMME project, with the financial support of the European Union, the Provence-Alpes-Côte d'Azur Region and the French Ministry in charge of Ecology.

Development of operational flash flood warning systems is one of the challenges in operational hydrology: flash floods are devastating but difficult to monitor and predict due to their nature. To provide flash flood warnings for ungauged basins, Météo-France and Irstea (formally Cemagref) have developed a discharge-threshold flood warning system called AIGA, which combines radar-gauge rainfall grids with a simplified distributed rainfall-runoff model run every 15 minutes at a 1-km2 resolution. Operational since 2005 in the Southern part of France, the AIGA system produces, every 15 minutes, a map of the river network with a color chart indicating the range of the estimated return period of the ongoing flood event. To increase forecast lead time and quantify the forcing input uncertainty, the rainfall-runoff distributed model ingests the 11 precipitation ensemble members from the PEARP ensemble prediction system of Météo-France. Performance of the experimental probabilistic precipitation and flow forecasts is evaluated from a variety of ensemble verification metrics (e.g., Continuous Ranked Probability Skill Score, Relative Operating Characteristic score) for different French basins. We also discuss planned enhancements and challenges to assess other sources of hydrological uncertainty and effectively communicate the uncertainty information to forecasters for better risk-based decision making.

We address in this study the routing of distributed hydrological information to the outlet of watersheds, in the fields of flood forecasting on ungauged watersheds in the French Mediterranean area. Flood forecasting can benefit of rainfall data provided in real-time by radar networks. This data used as an input to rainfall runoff models gives access to flood anticipation on small ungauged watersheds prone to flash floods. Within the framework of the AIGA method, developed by IRSTEA (formerly Cemagref) and Météo-France (the French National Weather Service) to provide floods alerts, a rainfall-runoff model is implemented at the spatial resolution of the radar data, thus providing a map of the 1 km² pixel contributions to the runoff. We have focused on the routing function of the rainfall-runoff model at the 1 km² cell scale, this scale being the first step of the runoff routing from the production area to the outlet of the catchment. We have then produced streamflow hindcasts for selected observed events using: - two different kinds of routing functions within our rainfall-runoff model, the first being a uniform routing function, the second being distributed routing functions, - two different hindcasts configurations, the first being a research configuration with the calibration of the production function of our rainfall-runoff model, the second being a configuration close to the real-time conditions of an operational flood warning service, i.e. with no possibility to calibrate the model parameters. Results indicate that the AIGA method forecasting ability, assessed via a multicriteria analysis evaluating the forecasting lead-time, contingency statistics and the similarity between the calculated and observed hydrographs, is enhanced (both in research and operational configurations) when implementing a distributed routing function.

Appréhender le fonctionnement actuel des bassins versants français est un élément essentiel pour répondre aux exigences de la Directive Cadre européenne sur l’Eau. La connaissance du régime hydrologique – a minima au travers de descripteurs statistiques - en sites non jaugés est de fait nécessaire. Cet article présente les résultats de différentes techniques d’estimation de deux variables descriptives des crues : le débit de pointe maximal annuel de fréquence décennale QIXA10 et une durée caractéristique issue d’une sélection d’hydrogrammes tronqués de l’écoulement de base. Sont présentées la procédure de sélection du jeu de stations hydrométriques fiables, le mode d’extraction des échantillons de données et le calcul des variables aux sites instrumentés à spatialiser. Le point principalement abordé concerne l’établissement de formules empiriques pour l’estimation en bassins non jaugés et les incertitudes associées. Leurs performances sont mises en perspective par rapport à celles d’autres méthodes exploitant la proximité géographique (voisinage glissant et approche géostatistique). / Knowing the river flow regime along the French river network is required to answer basic questions related to the European Water Framework directive. The river flow regime should be at least estimated at ungauged locations through relevant statistics. A comparison of different approaches applied to estimate two flood regime descriptors is presented: the instantaneous peak discharges with 10 year return period and a flood duration derived from a selection of hydrographs removed from baseflow. This paper focuses on the framework developed to establish the reference dataset of gauging stations, the sampling technique applied to extract variables from an extended set of time series and the procedure to compute the two variables. Several empirical formulas at both national and regional levels are presented. An analysis of the predictive performance is performed for comparing the accuracy of the tested approaches including the empirical formulas, an approach using geographical neighbourhood and a geostatistical method.

The study aims at estimating flow duration curves (FDC) at ungauged sites in France and quantifying the associated uncertainties using a large dataset of 1080 FDCs. The interpolation procedure focuses here on 15 percentiles standardised by the mean annual flow, which is assumed to be known at each site. In particular, this paper discusses the impact of different catchment grouping procedures on the estimation of percentiles by regional regression models. In a first step, five parsimonious FDC parametric models are tested to approximate FDCs at gauged sites. The results show that the model based on the expansion of Empirical Orthogonal Functions (EOF) outperforms the other tested models. In the EOF model, each FDC is interpreted as a linear combination of regional amplitude functions with spatially variable weighting factors corresponding to the parameters of the model. In this approach, only one amplitude function is required to obtain a satisfactory fit with most of the observed curves. Thus, the considered model requires only two parameters to be applicable at ungauged locations. Secondly, homogeneous regions are derived according to hydrological response, on the one hand, and geological, climatic and topographic characteristics on the other hand. Hydrological similarity is assessed through two simple indicators: the concavity index (IC) representing the shape of the dimensionless FDC and the seasonality ratio (SR), which is the ratio of summer and winter median flows. These variables are used as homogeneity criteria in three different methods for grouping catchments: (i) according to an a priori classification of French Hydro-EcoRegions (HERs), (ii) by applying regression tree clustering and (iii) by using neighbourhoods obtained by canonical correlation analysis. Finally, considering all the data, and subsequently for each group obtained through the tested grouping techniques, we derive regression models between physiographic and/or climatic variables and the two parameters of the EOF model. Results on percentile estimation in cross validation show that a significant benefit is obtained by defining homogeneous regions before developing regressions, particularly when grouping methods make use of hydrogeological information.

International audience; The real-time vision of a large portion of French rivers is not a coloured one, not even a black and white, but a blind one, du to the large number of basins which are not monitored. The event of June 2010 in Southern France, where 25 people died during the flood of the Nartuby and Argens rivers gives a dramatic example of this blind vision. In such a context a simple insight, aiming at providing flood alerts rather than pretending to represent exactly the physical process at stake in the catchment, seems welcomed. The aim of this study is to assess the ability of a simple conceptual and parsimonious rainfall-runoff model to deliver flood alerts in ungauged catchments. We use a downward approach: the model is calibrated on gauged catchments and its performances are checked on nested catchments which are not used for the calibration. The model is calibrated over 20 catchments of southern France with areas ranging from 150 km² to 2170 km². The rainfall information is provided by a radar network at an hourly time-step for 15 events that have occurred between 2005 and 2008. The "ungauged" catchments used for the model evaluation are 47 nested basins located within the 20 calibration catchments. We derive the model parameters for these ungauged catchments, from the gauged ones. The model used belongs to the GR models family developed at the Cemagref institute. It works on en event basis thank to a production parameter S and two routing parameters: a shape parameter B and a lag parameter C. Our knowledge of the production parameter benefits from a previous regionalisation work based on several characteristics of the catchments such as the description of their drainage network, the ground-water typology and a climatic variable. The model calibration rests upon the production parameter only. We implement two versions of the model: a lumped one and a distributed one. The results of this downward approach show that the model tested here is able to deliver successful flood alerts in ungauged areas. The distributed version performs slightly better than the lumped version. This study also show that fully distributing the routing parameters results in better alerts than using uniform routing parameters in the distributed version. To fully assess the performance of the model, further evaluation should be carried out in an operational mode, i.e. using an initialisation procedure to determine the value of the S parameter at the beginning of each event instead of using a calibrated value. We anticipate this study by checking how well the calibrated S correlate with the parameter of a continuous soil moisture accounting model run at a daily time-step. The correlation being better in the case of the distributed version of the model, this version seems hence quite promising for an operational use.

International audience; Flash floods result from the combination of meteorological and hydrological conditions. Recognition of the coupled meteorological / hydrological nature of flash floods is now evident in interpretative studies and in the development of predictive models. There is a real need for research to improve the understanding of the major atmospheric and hydrologic factors leading to extreme flood events, especially those affecting small to medium ungauged basins. It has been shown that most of flash flood events can be attributed to stationary Mesoscale Convective Systems (MCSs). Due to their very localized nature, the observation of such event using a gauging network is problematic. Weather radars can provide better spatial rainfall estimations, even if radar assessment of rainfall can be significantly influenced by the orography itself. Therefore accurate monitoring of severe storm rainfall intensities remains a major challenge. In terms of the hydrological processes associated to flash flood, it is known that the knowledge of the soil characteristics is an important source of uncertainty in the understanding of the hydrological behavior of the catchments. This contribution presents a modelling approach aiming at quantifying the respective impact of rainfall estimation and soil variability on the simulated discharge for an extreme event in southern France. The method is illustrated for two ungauged medium size basins using different radar and gauge rainfall estimates and an existing soil data base. The hydrological distributed modelling approach was implemented within the numerical modelling LIQUID platform on the Cévennes-Vivarais region, used without any calibration phase, and validated with a regional approach using distributed post-flood estimation (Braud et al., EGU2009). For the same event, previous work using an event-based TOPMODEL approach with a calibration phase, showed that the model efficiency significantly increases when the spatial variability of rainfall is taken into account. Nevertheless, for some of the catchments, the mis-performances remained unexplained. The model was run using different radar image treatments and the sensitivity on the hydrological response was assessed. The results show the important contribution of the volume-scanning protocol for the quantitative radar rainfall estimation at regional and local scale. Nevertheless, this study also highlights the importance of the description of the soil types to properly set up the hydrological model.

The Mediterranean region is subject to flash flood events that lead to many damages and sometimes to fatalities. Because not every little river's tributaries can be monitored, flash floods occur most often on small ungauged catchments with limited available data. In this context, the calibration of the hydrological model used within a flood warning system is a tricky task. The aim of this study is to evaluate the potential benefit of 'proxy data' in areas where no classical continuous flow measurements are available. Administrative services for example collect and archive large amounts of regional and local information over long periods based on observed damages, some of which can be used as proxy indicators for flood occurrences (e.g. infrastructure maintenance, insurance claims,...). To determine if such data could be helpful, a benchmark test is carried out, comparing three cases. In the first case the entire streamflow data series is used. In the second case, data from case 1 is degraded: only the date and value of the flood peak are retained. In the last case, only the date of the flood peak is used. This last case aims to imitate the case, where only proxy data collected by administrative services are available. The first case corresponds to the classical gauged basin case, which is used as reference. These 3 cases are carried out using a simple conceptual hydrological model (from the GR model family) and data coming from around 150 catchments located in the South East of France and covering a 10-year period (1997-2007). The performance of simulated runoff is evaluated using contingency statistics (CSI). A split-sample test is used in order to assess the robustness of the different calibrations. Results indicate that performances decrease from case 1 to case 3, case 2 being intermediate. Since the methodology developed for case 3 can be applied at ungauged locations, it seems that obtained alerts can be very helpful when no classical hydrological data is available to calibrate the hydrological model used in the flood warning system. This work has been made in the framework of the ongoing RHYTMME project. This project aims at developing a warning system in the south of French Alps, using radar rainfall data (http://rhytmme.cemagref.fr/synopsis).