Search

International audience; Headwater streams are generally naturally prone to flow intermittence, mainly due to their upstream position in the network and to their high reactivity to natural or human disturbances. These intermittent rivers and ephemeral streams (IRES) have recently seen a marked increase in interest, especially to assess the impact of drying dynamics on aquatic ecosystems. However, the understanding of IRES dynamics and their geographic extent remains limited because gauging stations are preferentially located along perennial rivers and, consequently, the proportion of IRES is highly underestimated. A new source of observational data named ONDE have recently become available in France to complement discharge data from the conventional French River Flow Monitoring network. The ONDE network provides frequent discrete field observations (five inspections per year or more) of the flow states on 3 300 sites throughout France and located mostly in headwater areas. At each ONDE site, one flow states is assigned each day of observation (including "dry" condition). This study focuses on a region homogeneous in terms of geology, topography and climate located in western France (20 000 km2) and containing 240 ONDE sites more or less impacted by flow intermittence. The objective of this work is to reconstruct the daily drying dynamic at ONDE sites by combining available flow states observations over the period 2012-2016 with time series of discharge and groundwater level, climate indices (rainfall, evapotranspiration and air temperature) and geomorphological characteristics. Four statistical classification models are tested: (1) k-nearest neighbors model (k-NN); (2) Artificial Neural Networks model (ANN); (3) Random Forest model (RF) and (4) Least Absolute Shrinkage and Selection Operator model (LASSO). A performance analysis is carried out over the 5-year period 2012-2016 to assess their ability to simulate the daily drying dynamic at ONDE sites by cross validation. In a second step, the influence of each explanatory variable in predictions is examined and main drivers of flow intermittency are identified. ANN, RF and LASSO models perform equally well over the period 2012-2016 with both F.score and the probability of detection (POD) > 0.7. The worst results are obtained with the k-NN model (F.score around 0.5, POD < 0.4). ANN, RF and LASSO models tend to slightly less under-estimate zero-flow conditions in comparison to ONDE observations. The main drivers controlling the drying development are in order of importance: the proportion of days with dry states observed at ONDE site between 2012 and 2016, the non-exceedance frequency of discharge and groundwater level and the antecedent rainfall. The proportion of days with dry states observed at ONDE site informs the temporariness of the streams and the additional variables controls the inter-annual variability. Finally, ANN, RF and LASSO models succeeded to reconstruct the drying dynamics locally by differentiating each stream and could be potentially used to analyze the past and future temporal variability of flow intermittence.

International audience; This paper presents a coupled observation and modelling strategy aiming at improving the understanding of processes triggering flash floods. This strategy is illustrated for the Mediterranean area using two French catchments (Gard and Ardèche) larger than 2000 km2. The approach is based on the monitoring of nested spatial scales: (1) the hillslope scale, where processes influencing the runoff generation and its concentration can be tackled; (2) the small to medium catchment scale (1–100 km2), where the impact of the network structure and of the spatial variability of rainfall, landscape and initial soil moisture can be quantified; (3) the larger scale (100–1000 km2), where the river routing and flooding processes become important. These observations are part of the HyMeX (HYdrological cycle in the Mediterranean EXperiment) enhanced observation period (EOP), which will last 4 years (2012–2015). In terms of hydrological modelling, the objective is to set up regional-scale models, while addressing small and generally ungauged catchments, which represent the scale of interest for flood risk assessment. Topdown and bottom-up approaches are combined and the models are used as “hypothesis testing” tools by coupling model development with data analyses in order to incrementally evaluate the validity of model hypotheses. The paper first presents the rationale behind the experimental set-up and the instrumentation itself. Second, we discuss the associated modelling strategy. Results illustrate the potential of the approach in advancing our understanding of flash flood processes on various scales.

National audience; Présentation de la méthodologie de comparaison utilisée dans ExtraFlo

International audience; We compare three approaches to estimate the distribution of extreme rainfall at ungauged sites. Two approaches rely on the univariate generalized extreme value distribution (GEV). SIGEV interpolates linearly the GEV parameters estimated locally. RFA is a regional method which builds circular homogeneous neighbourhood around each site in order to increase the sample size. The observations in the neighbourhood, properly normalized, are assumed to follow the same GEV distribution. Then the normalizing factor (called the index value) has to be interpolated to ungauged sites. The third method is the stochastic hourly rainfall generator called SHYPRE. By characterizing precisely rainfall events, SHYPRE is able to simulate long rainfall series with statistics similar to the observed series. The distribution of extreme rainfall is estimated empirically from the simulated series. The three approaches are evaluated and compared on datasets from over 1000 rain gauges in the South of France. The evaluation framework that we follow is based on the computation of high-level quantiles and aims at assessing the goodness-of-fit of the three approaches and their sensitivity to the training data. Our conclusions are threefold : SIGEV, as implemented, should be avoided because of its lack of robustness, RFA and SHYPRE despite the fact that they are based on very different hypotheses on rainfall provide comparable performance and finally, the main challenge regarding the estimation at ungauged sites concerns the spatial interpolation of the parameters, whatever the approach taken.; Nous comparons trois approches pour l’estimation de la distribution des pluies extrêmes en des sites non jaugés. Deux de ces approches reposent sur la loi des valeurs extrêmes généralisée (GEV). La méthode SIGEV interpole linéairement les paramètres de la GEV estimés localement aux sites jaugés. RFA est une méthode régionale qui définit des voisinages homogènes circulaires autour de chaque site ce qui permet d’augmenter la taille de l’échantillon. En effet, RFA fait l’hypothèse que les observations aux stations du voisinage suivent la même loi GEV à un facteur de normalisation près. Ce facteur, appelé index value doit être interpolé aux sites non-jaugés. La troisième approche se base sur un générateur de pluie horaire appelé SHYPRE. À l’aide d’un caractérisation précise des événements pluvieux, SHYPRE est en mesure de simuler de longues séries de pluie ayant des statistiques semblables aux séries d’observations. Ces trois approches sont évaluées et comparées sur un jeu de données comprenant plus de 1000 stations dans le sud de la France. La comparaison des méthodes repose sur le calcul de quantiles de haut niveau et a pour but d’évaluer la justesse et la sensibilité des méthodes. Nos conclusions sont les suivantes : SIGEV tel que mis en ½uvre ne devrait pas être retenu en raison de son manque de robustesse, RFA et SHYPRE ont des performances comparables bien que ces méthodes soient très différentes et nous concluons que le défi le plus important à relever pour ces trois approches réside dans l’interpolation spatiale des paramètres.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU [Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; An abundance of methods have been developed over the years to perform the frequency analysis (FA) of extreme environmental variables. Although numerous comparisons between these methods have been implemented, no general comparison framework has been agreed upon so far. The objective of this paper is to build the foundation of a data-based comparison framework, which aims at complementing more standard comparison schemes based on Monte Carlo simulations or statistical testing. This framework is based on the following general principles: (i) emphasis is put on the predictive ability of competing FA implementations, rather than their sole descriptive ability measured by some goodness-of-fit criterion; (ii) predictive ability is quantified by means of reliability indices, describing the consistency between validation data (not used for calibration) and FA predictions; (iii) stability is also quantified, i.e., the ability of a FA implementation to yield similar estimates when calibration data change; and (iv) the necessity to subject uncertainty estimates to the same scrutiny as point estimates is recognized, and a practical approach based on the use of the predictive distribution is proposed for this purpose. This framework is then applied to a case study involving 364 gauging stations in France, where 10 FA implementations are compared. These implementations correspond to the local, regional, and local-regional estimation of Gumbel and generalized extreme value distributions. Results show that reliability and stability indices are able to reveal marked differences between FA implementations. Moreover, the case study also confirms that using the predictive distribution to indirectly scrutinize uncertainty estimates is a viable approach, with distinct FA implementations showing marked differences in the reliability of their uncertainty estimates. The proposed comparison framework therefore constitutes a valuable tool to compare the predictive reliability of competing FA implementations, along with the reliability of their uncertainty estimates.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; We present a conditional density model of river runoff given covariate information which includes precipation at four surrounding stations. The proposed model is non parametric in the central part of the distribution and relies on Extreme-Value Theory parametric assumptions for the upper tail of the distribution. From the trained conditional density model, we can compute quantiles of various levels. The median can serve to simulate river runoff, quantiles of level 5% and 95% can be used to form a 90% confidence interval, finally, extreme quantiles can estimate the probability of large runoff. The conditional density model is based on a mixture of hybrid Paretos. The hybrid Pareto is built by stitching a truncated Gaussian with a Generalized Pareto distribution. The mixture is made conditional by considering its parameters as functions of covariates. A neural network is used to implement those functions. A penalty term on the tail indexes is added to the conditional log-likelihood to guide the maximum likelihood estimator towards solutions that are preferred. This alleviates the difficulties encounter with the maximum likelihood estimator of the tail index on small training sets. We evaluate the proposed model on rainfall-runoff data from the Orgeval basin in France. The effect of the tail penalty is further illustrated on synthetic data.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; The objectives of the FloodScale project were to progress in the knowledge, understanding and simulation of hydrological processes leading to flash floods in the Mediterranean area. These flash floods and the rainfall events that trigger them lead, each year, to fatalities and a large amount of damage in this region, but processes leading to their generation remained poorly known and hierarchized. The FloodScale project contributed significantly to the 'hydrology' part of the international HyMeX (Hydrological Cycle in the Mediterranean Experiment) program. Several aspects were tackled in the project: observation, understanding and simulation of flash floods, with in the background, two fundamental questions in hydrology: the change of scale problem and the prediction in ungauged catchments. These two questions are particularly relevant for flash floods that often occur in small ungauged catchments but that may interest a large territory, requiring understanding and modelling at small scale and over a large area. For the observation of flash floods and data collection allowing tackling the two scientific questions mentioned above, a multi-scale observation strategy was set up in two meso-scales catchments of the Cévennes region: the Gard and Ardèche catchments. Three scales were considered in the sampling strategy: 1/ the hillslope scale for documenting hydrological processes; 2/ the small catchment scale (1-100 km2) where the spatial variability of landscapes and hydrological processes was monitored and described and 3/ the meso-scale catchment scales (100-2000 km2) relevant for management and warning. These observations combined continuous measurements during four years (rainfall, water level in streams, discharge, soil moisture) and opportunistic measurements during the four autumns of the project: gauging of flooding rivers, sampling of soil moisture and geochemistry sampling of rainfall, streams and soil water. The observation strategy was found relevant and allowed documenting both the 'normal' catchment behavior and a selection of flash floods across scales. The project also allowed evaluating several new measurements techniques such as the saturometer for estimating soil infiltration capacity but also the interest of non-contact measurements for 1/ estimating soil depth or monitoring water fluxes thanks to electrical resistivity, 2/ estimating flooding discharges (use of portable surface velocity radars, fixed cameras and analysis of citizens videos). Methods for merging radar rainfall and pluviometers data were also proposed and improved and proved relevant to describe properly rainfall time and space variability. We should also mention methodological progress for quantifying rainfall and discharge uncertainty. Analysis of the collected data showed a large variability of soil hydraulic properties on the hillslope scale, whereas those properties were quite stable across hillslopes. At the scale of the whole Cévennes-Vivarais region, geology and land use were found discriminant for explaining soil hydraulic properties spatial variability. A high degree of soil depth variability is also a characteristic of the soils in the region and it is necessary to take it into account into models. Hillslope experiments, but also the analysis of opportunistic or continuous geochemistry sampling show the large contribution of quick lateral sub-surface flow in granite and schist geology with a natural vegetation, but also their role in sedimentary and agricultural areas. Soil water storage inferred from existing data bases was also found to be greatly underestimated because they do not account for water storage in the deep weathered and fractured bedrock layers. The soil-bedrock interface was found to be pervious, with high infiltration capacities. Analysis of soil moisture data combined with the hydrological response show threshold effect at the transition between dry and wet conditions with a low hydrological response below this humidity threshold and a large variability in the response when soil water content is above this threshold. The simulation and modelling of flash floods was performed using a hypothesis testing framework, where various functioning hypotheses are assessed by comparing modeled and observed values. The project results proved the validity and interest of this iterative approach. At the small catchment scales, various models, based on different hypotheses about dominant processes were used. The results show that one process is generally not sufficient to reproduce the observed response and that sub-surface fluxes must be included in the modeling as well as the spatial variability of soil water storage. More specifically, a distributed modelling approach, designed from field observations was shown to be relevant and robust to the change of scale. At the regional scale, top-down and bottom-up approaches were set up and assessed. The studies confirm the need to account for water storage in deeper weathered horizons for a correct simulation of soil water storage capacity. They also show that geology must be also taken into account to correctly simulate the spatial variability of the hydrological response. Both approaches allow a satisfactory simulation of this variability, in particular during flash floods. Those models provide simulations in ungauged catchments. During the project a large effort was dedicated to data collection, validation and documentation for inclusion in several data bases. The collected data set document various aspects of flash floods at various scales and is quite unique. All the collected data were only partially analyzed and exploited during the project, with a focus on sites analyses to verify the consistency of the data and have a first insight into active processes. The most promising perspective of the work is certainly a global analysis of the data sets by pooling all the data from the various sites together to obtain a picture of hydrological responses on the scale of the whole region. The effort of documenting rainfall and discharge uncertainty on rainfall and discharge also offers rich perspectives to better take uncertainty into account in models, in their evaluation and calibration.; Le projet FloodScale avait pour objectif de progresser sur la connaissance, la compréhension et la simulation des processus hydrologiques conduisant aux crues rapides en zone Méditerranéenne. Ces crues rapides et les épisodes pluvieux importants qui les génèrent conduisent chaque année à des pertes humaines et matérielles très importantes dans cette région mais les processus conduisant à leur genèse restaient encore mal connus ou hiérarchisés. Par ailleurs, FloodScale représente une contribution importante au volet « hydrologie » du programme international HyMeX (Hydrological Cycle in the Mediterranean Experiment). Le projet a abordé plusieurs volets : l’observation, la compréhension et la simulation des crues rapides, avec en toile de fond, deux questions fondamentales en hydrologie : le changement d’échelle et la prédiction en bassins non jaugés. Ces deux problématiques sont particulièrement pertinentes pour les crues rapides qui affectent souvent de petits bassins sans mesure des débits et peuvent concerner de larges étendues, nécessitant une compréhension et une modélisation à échelle fine et sur de vastes territoires. Pour l’observation des crues rapides et le recueil de données permettant d’aborder les questions ci-dessus, une stratégie multi-échelle a été mise en ½uvre sur deux bassins de méso-échelle des Cévennes : le Gard et l’Ardèche. Trois échelles spatiales ont été considérées : 1/ le versant pour la documentation des processus ; 2/ l’échelle du petit bassin (1-100 km2) pour la description de la variabilité spatiale des paysages et des réponses hydrologiques et 3/ l’échelle du bassin de méso-échelle (100-2000 km2) pertinente pour la gestion et la prévision. Ces observations ont combiné des mesures en continu sur quatre ans (pluie, hauteur d’eau, débits, humidité des sols) et des mesures sur alerte durant les quatre automnes du projet : jaugeages en crue, échantillonnage de l’humidité des sols, échantillonnages géochimiques. Cette stratégie d’observations s’est révélée pertinente et a permis de documenter non seulement plusieurs épisodes de crues rapides à différentes échelles mais aussi le fonctionnement « normal » des bassins versants. Le projet a aussi permis de tester et valider des méthodes de mesure innovantes telles que le saturomètre pour les capacités d’infiltration des sols, ainsi que l’intérêt de mesures sans contact pour 1/ l’estimation des profondeurs de sols ou le suivi des flux d’eau grâce à la résistivité électrique, 2/ les mesures de débit de crue (utilisation de radars de vitesse, de caméras fixes, analyses de vidéo amateurs). Des méthodes de fusion de données radar et pluviomètres ont aussi été proposées et développées et se sont aussi révélées pertinentes pour améliorer la connaissance de la pluviométrie et sa variabilité spatiale et temporelle. Mentionnons aussi les avancées méthodologiques permettant de quantifier les incertitudes sur les pluies et les débits. L’analyse des données collectées a permis de mettre en évidence une forte variabilité spatiale des propriétés hydrodynamiques des sols intra-versant, mais une relative stabilité de ces propriétés inter-versants. A l’échelle des Cévennes, la géologie et l’occupation des sols apparaissent comme des éléments discriminants sur la variabilité spatiale de ces grandeurs. La variabilité des profondeurs des sols est aussi une caractéristique des milieux étudiés, qu’il est nécessaire de prendre en compte dans les modèles. Les expérimentations de versants, mais aussi l’analyse des données géochimiques des campagnes sur alerte ou mesurées en continu confirment la prépondérance d’écoulements latéraux de sub-surface rapides en terrain granitiques ou sur schiste et végétation naturelle, et leur importance, même sur terrains sédimentaires et agricoles. Il a aussi été montré que les capacités de stockage des sols issues de bases de données sols étaient largement sous-estimées car elles ne prennent pas en compte le stockage dans les horizons de socle altéré et fracturé. L’interface sol-bedrock s’est par ailleurs révélée perméable, avec des capacités d’infiltration élevées. Les mesures d’humidité des sols combinées à l’analyse de la réponse hydrologique mettent en évidence un effet de seuil au moment du passage de conditions de sol sèches à humides avec une réponse hydrologique très faible en conditions sèches et une forte variabilité de la réponse hydrologique en conditions humides. La modélisation et la simulation des crues rapides ont été abordées dans un esprit d’évaluation/comparaison de différentes hypothèses de fonctionnement en s’appuyant sur les données collectées et l’analyse des différences entre simulation et observation. Les résultats du projet ont aussi permis de valider la pertinence de cette approche itérative. A l’échelle des petits bassins, différents modèles reposant sur différentes hypothèses de processus dominants ont été mis en ½uvre et testés. Ces essais montrent qu’un seul processus ne peut suffire à expliquer les réponses observées et que les flux de sub-surface doivent impérativement être considérés pour une simulation correcte des observations, de même que la variabilité spatiale des capacités de stockage des sols. En particulier, une modélisation distribuée, construite à partir des observations de terrain s’est révélée particulièrement pertinente et robuste au changement d’échelle. A l’échelle régionale, deux approches de modélisation « top-down » et « bottom-up » ont été mise en ½uvre et évaluée. Les études confirment la nécessité de prendre en compte les horizons altérés pour la bonne représentation des capacités de stockage dans les sols. Elles mettent en évidence la nécessité de bien prendre en compte la géologie pour simuler correctement la variabilité des réponses hydrologiques et que les deux types d’approches permettent de représenter de manière satisfaisante cette variabilité, notamment pour les crues rapides. Ces modèles permettent d’obtenir des simulations en bassins non jaugés. Durant le projet, un effort important a été consacré à l’acquisition, la validation et la mise à disposition de riches jeux de données documentant différents aspects pertinents pour la compréhension des crues rapides. Ces données ont été exploitées de manière fragmentaire durant le projet, avec une focalisation sur des analyses par site d’études visant à confirmer la cohérence des données acquises. Une des perspectives les plus prometteuses est maintenant la mise en perspective des données des différents sites et une analyse globale des comportements à l’échelle de toute la région cévenole. Les efforts menés pour caractériser les incertitudes sur la pluie et les débits offrent aussi d’intéressantes perspectives pour une meilleure prise en compte dans les modèles, leur évaluation et leur calibration.