Search

International audience; Located in the French Mediterranean region, the French Alternative Energies and Atomic Energy Commission (CEA) Cadarache Centre is subject to significant rainfall events which can lead to groundwater floods with rapid kinetics and high amplitude. This study aims to improve the understanding of the aquifer’s heterogeneity and piezometric responses at the scale of the Cadarache centre (900 ha) in order to better apprehend the groundwater floods. The deployed method consists in: (i) taking an inventory of reliable data; (ii) conducing a preliminary study of the spatial variability of rainfall at the centre scale with the method of double-mass curves and the comparison of the rainfall amounts of the different stations; and (iii) conducting a comprehensive statistical study of piezometric data (graphs of sorted piezometric data for three different flood events, definition of an original clustering method using an automatic feature detection). The results are of two kinds: (i) several clusters were identified; three appeared to be representative of specific areas; (ii) for each specific area, piezometers were linked with different plurikilometric NW–SE and NE–SW lineaments identified by previous bibliographic studies. This extensive study on nearly 80 piezometers has thus allowed to define a method of automatic feature extraction of the sorted groundwater levels, and to establish a clustering of piezometers according to their statistical behaviour. Even if the statistical “black box” method was applied without any prerequisite on the location of the boreholes, it allowed establishing a very relevant relation between sorted water height and geological structure constraining the flows.

International audience; The complexity of karst groundwater flow modelling is reflected by the amount of simulation approaches. The goal of the Karst Modelling Challenge (KMC) is comparing different approaches on one single system using the same data set. Thirteen teams with different computational models for simulating discharge variations at karst springs have applied their respective models on one single data set coming from the Milandre Karst Hydrogeological System (MKHS). The approaches include neural networks, reservoir models, semi-distributed models and fully distributed groundwater models. Four and a half years of hourly or daily meteorological input and hourly discharge data were provided for model calibration. The validation comprised forecasting one year of discharge, without the observed discharge data. The model performance was evaluated using the volume conservation, Nash-Sutcliffe efficiency (NSE) and the Kling-Gupta efficiency (KGE) applied on the total discharge and individual flow components. As a result, the comparison of model performances is a challenging task due to the differences in the model architecture but also required time steps: some of the models require aggregated daily steps while others could be run using hourly data, which provided some interesting differences depending on how the data was transformed. The use of instantaneous data (e.g. value at noon) produces less bias that averaging hourly data over one day. The transformation of hourly into daily data produces a decrease of Nash and KGE of 0.05 to 0.08 (i.e. from 1 to ~0.93). The resulting simulations (forecasted values for year 2016) produced KGEs ranging between 0.83 and 0.37 (0.83 to −0.24 for NSE). Although the simulations matched the monitored flows reasonably well, most models struggled to simulate baseflow conditions accurately. In general, the models that performed the best for this exercise were the global ones (Gardenia and Varkarst), with a limited number of parameters, which can be calibrated using automatic calibration procedures. The neural network models also showed a fair potential, with one providing reasonable results despite the relatively short dataset available for warming-up (4.5 years). Semi-and fully distributed models also suggested that with some more effort they could perform well. The accuracy of model predictions does not seem to increase by using models with more than 9–12 calibration parameters. An evaluation of the relative errors between the forecasted and the observed values revealed that for most models, 50% of the forecasted values contained more than 50% of difference against the observed discharge rate, with 25% having a difference larger than 100%. A significant part of the poorly forecasted values corresponded to base-flow which was surprising given that as base-flow is generally much easier to predict than peak flow. Hence, this shows that modelling approaches and criteria for the calibration are too oriented towards peak-flow sections of the hydrographs, and that improvements could be gained by more focus on the base-flow.

Understanding, simulating and forecasting dynamic and nonlinear natural phenomena are necessary in a climate change context and increased sensitivity of societies to natural hazards. Nevertheless, even though powerful computing tools and algorithms have been widely used to understand and to predict natural disasters, these tasks are still challenging for scientists. Indeed, one of the most dangerous natural phenomena, flash floods keep being a challenge for modelers, despite (i) the existence of some effective hydrological simulating tools, and (ii) the increasing availability of descriptive data, especially rainfall and discharge. In particular, on one hand, environmental data contain an important amount of noise leading to additional uncertainties and on the other hand, physically based models strongly depend on assumptions about the behavior of the basin, that is often more variable in space and time than what is modelled. With the objective of applying data assimilation to improve forecasting properties of the physical model, it is necessary to dispose of a differentiable model. In order to mitigate this issue, a hybrid physical and statistical approach is proposed in this study. It was shown in previous works that deep neural networks are able to identify any differentiable function by using the universal approximation property. Deep neural networks are also good candidates to perform the digital twin of the physical model. Thus, three different neural networks models were designed in this study, and each one is implementing a different type of non-linear filter model, in order to achieve the dynamic character of the catchment area (recurrent, feedforward and static models). The study area is located in the Gardon de Sainte-Croix basin (France), which is known for its sudden and violent floods that caused casualties and a lot of damage. The chosen physical-based model is semi distributed conceptual hydrological SOCONT model, RS Minerve (https://www.crealp.ch/down/rsm/install2/archives.html). Neural networks design was done by using a rigorous complexity selection and regularization methods to promote a good generalization. The three models obtained were thus compared. The feed forward model gave the best results on tests events (Nash score=0.98−0.99), making full use of the inputs with previous observed discharges whereas the recurrent model gave interesting results representing satisfactorily the dynamics of the physical model (Nash score=0.8−0.97). The static model, whose inputs contain only rainfall, is less efficient, showing the importance of dynamics in that kind of system (Nash score=0.62−0.84). Beyond data assimilation, these results open paths of inquiry for building digital twins of physical model, allowing also a great reduction of computing time.

International audience; Peatlands are habitats for a range of fragile flora and fauna species. Their eco-physicochemical characteristics make them as outstanding global carbon and water storage systems. These ecosystems occupy 3% of the worldwide emerged land surface but represent 30% of the global organic soil carbon and 10% of the global fresh water volumes. In such systems, carbon speciation depends to a large extent on specific redox conditions which are mainly governed by the depth of the water table. Hence, understanding their hydrological variability, that conditions both their ecological and biogeochemical functions, is crucial for their management, especially when anticipating their future evolution under climate change. This study illustrates how long-term monitoring of basic hydro-meteorological parameters combined with statistical modeling can be used as a tool to evaluate i) the horizontal (type of peat), ii) vertical (acrotelm/catotelm continuum) and iii) future hydrological variability. Using cross-correlations between meteorological data (precipitation, potential evapotranspiration) and water table depth (WTD), we primarily highlight the spatial heterogeneity of hydrological reactivity across the Sphagnum-dominated Frasne peatland (French Jura Mountain). Then, a multiple linear regression model allows performing hydrological projections until 2100, according to regionalized IPCC RCP4.5 and 8.5 scenarios. Although WTD remains stable during the first half of 21th century, seasonal trends beyond 2050 show lower WTD in winter and markedly greater WTD in summer. In particular, after 2050, more frequent droughts in summer and autumn should occur, increasing WTD. These projections are completed with risk evaluations for peatland droughts until 2100 that appear to be increasing especially for transition seasons, i.e. May–June and September–October. Comparing these trends with previous evaluations of phenol concentrations in water throughout the vegetative period, considered as a proxy of plant functioning intensity, highlights that these hydrological modifications during transitional seasons could be a great ecological perturbation, especially by affecting Sphagnum metabolism.

Issu de : ITISE 2019 - International Conference on Time Series and Forecasting, Granada, Spain, 25-27 September 2019; International audience; Flash floods frequently hit Southern France and cause heavy damages and fatalities. To enhance persons and goods safety, official flood forecasting services in France need accurate information and efficient models to optimize their decisions and policy in crisis management. Their forecasting is a serious challenge as heavy rainfalls that cause such floods are very heterogeneous in time and space. Such phenomena are typically nonlinear and more complex than classical flood events. This analysis had led to consider complementary alternatives to enhance the management of such situations. For decades, artificial neural networks have been proved very efficient to model nonlinear phenomena, particularly rainfall-discharge relations in various types of basins. They are applied in this study with two main goals: first, modeling flash floods on the Gardon de Mialet basin (Southern France); second, extract internal information from the model by using the KnoX: knowledge extraction method to provide new ways to improve models. The first analysis shows that the kind of nonlinear predictor strongly influences the representation of information, e.g., the main influent variable (rainfall) is more important in the recurrent and static models than in the feed-forward one. For understanding “long-term” flash floods genesis, recurrent and static models appear thus as better candidates, despite their lower performance. Besides, the distribution of weights linking the exogenous variables to the first layer of neurons is consistent with the physical considerations about spatial distribution of rainfall and response time of the hydrological system.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]GEUSI [ADD1_IRSTEA]Gestion intégrée de la ressource et des infrastructures; International audience; In order to better manage complex situations of natural resource management, models are built in a participative way, involving the stakeholders of these situations in participatory modelling activities. The impact that this activity of participatory modelling has on the stakeholders is at the heart of the Companion Modelling approach but this impact is hardly possible to evaluate on the field. In this paper we propose a general framework to study in vitro the impact of participatory modelling on natural resources management. We illustrate our framework by proposing an experimental setting that looks at participatory modelling in the context of water management. We realized a pilot experiment and show that this experimental setting can be used to test, in the laboratory, the hypothesis that participatory modelling of a common pool resource situation has an impact on the way the resource is managed and increases the cooperative behaviour of stakeholders.

The risk of drought impacting the drinking water and agricultural production is worrying in the developed countries, especially in a changing climate context. To manage and prevent this phenomenon, real-time monitoring and predictive systems are emerging as the key solutions. In the field of artificial intelligence, neural networks are one of these predictive systems. This family of parameterized models is a composition of neuronal functions, which apply a non-linear transformation from their inputs to their outputs. These networks are able to learn a hydro(geo)logical system behaviour using a database composed of observed inputs (rainfall, evapotranspiration, etc.) and outputs (groundwater level, discharge, etc.), thanks to an algorithm minimizing a cost function between observed and simulated outputs. However, it remains difficult to assess the uncertainty generated by these models, possibly leading to misinterpretations by the end users. These uncertainties are mainly of three types. The first is related to the input data. Indeed, hydrosystems are surface elements whereas meteorological inputs are punctual elements. The interpolation error can, therefore, be significant because of the lack of knowledge between gauging stations. The second is the neural network model architecture itself. It is possible to deal with this source of uncertainty using regularization methods. Finally, the neural networks are submitted to uncertainties related to parameter initialization, before the training step. The initial parameters may have an important impact on the results. In this paper, we address the prediction of the Blavet groundwater level (Bretagne, France). In order to assess uncertainties, we will first focus on the parameters initialization of the model. Neuronal models are optimized using cross-validation and early stopping. Then, an ensemble model is realized, in which each member is the result of a unique set of parameters initialization. The purpose of the study is to define how many initializations are necessary to obtain a reasonable confidence interval for forecasts, with the smallest interval and the higher rate of observed points inside this interval. The best model will be determined using cross-validation scores thereby ensuring optimal robustness. We show that, in this case study, an ensemble model of 20 different initializations is sufficient to estimate uncertainty while preserving quality. In the second part, the resulting ensemble model will be used to estimate the global model uncertainty using probability density functions (pdf) applied to the distribution of groundwater level data and cross-validation scores of forecasts. It reveals that the groundwater level predictions are composed of two mixed distributions. Therefore, we will use the expectation-maximization algorithm (EM) to obtain parameters of mixed models. Mixed normal and mixed Gumbel laws, among five mixed distributions assessed, give the best groundwater distribution and are able to generate an abacus drawing uncertainty of model.

Selon de nombreuses prévisions, les rivières méditerranéennes devraient être soumises dans les années à venir à de fortes pressions environnementales liées aux changements climatiques, ce qui laisse entrevoir des conséquences certaines pour la faune qu'elles hébergent. Pour tenter de percevoir ces conséquences, la faune invertébrée de l'une de ces rivières, la Cèze, a été étudiée en période d'étiage. La biodiversité des assemblages de la Cèze a été estimée en tenant compte de l'hétérogénéité des échanges entre la rivière et sa nappe. Ainsi trois habitats ont été échantillonnés : le fond de la rivière (habitat benthique), la zone d'interface entre la rivière et l'aquifère poreux (zone hyporhéique) et celle d'interface avec l'aquifère karstique (sources). Les résultats montrent que la composition des assemblages d'espèces récoltées est différente dans les trois habitats, avec seulement 11 % de taxons communs. L'habitat benthique se révèle plus riche en Éphéméroptères et en Trichoptères, l'habitat hyporhéique en Oligochètes, Crustacés et Plécoptères, et les sources en Diptères. Les espèces stygobies (spécialisées à la vie en milieu souterrain) sont totalement absentes de l'habitat benthique, alors qu'elles sont bien représentées dans les sources et la zone hyporhéique. Les deux habitats situés à l'interface entre le cours d'eau et les systèmes souterrains produisent un enrichissement considérable de la biodiversité du cours d'eau, qui se traduit par un doublement du nombre de taxons par rapport à l'habitat benthique. Ce travail montre que l'intégration des zones situées à l'interface entre la rivière et les eaux souterraines est cruciale pour l'obtention d'une estimation aussi complète que possible de la biodiversité des hydrosystèmes. In a close future, Mediterranean rivers will support strong disturbances due to climate changes, with inexorable consequences for their fauna. The invertebrate biodiversity of the Cèze River (South of France) was studied in a karstic area. The benthic layer (river bottom), the hyporheic layer (interface with the groundwater inside river sediment) and the lateral springs (interface with the karstic systems) were sampled in early summer. Only 11 % of the taxa were sampled simultaneously in the three habitats. Ephemeroptera and Trichoptera were more species rich in the benthic zone; Oligochaetes, crustaceans and Plecoptera in the hyporheic zone and Diptera in the lateral springs. Stygobite species lack in the benthic layer, they were more diversified in the springs and in the hyporheic zone. When all habitats are considered together, the total taxonomic richness is twice of the benthic one. The habitats located at the interface between river and groundwater thus must be included in the river biodiversity estimations.

IEEE International Conference on Environmental Engineering (EE), Milan, ITALY, MAR 12-14, 2018; International audience; This paper investigates the way to adapt neural network design to non-Gaussian environmental data. The process of model selection is specifically investigated in order to make the model more robust. It appears that the standard method of cross-validation gains to be applied on an ensemble model, rather than a unique model, in order to apply additional regularization. Specific architectures of neural networks based on multilayer perceptron were used simultaneously with various methods of complexity selection. Prediction results on floods and droughts show that the sensitivity to the initial value of parameters could be greatly reduced. A case-study is chosen on a exceptionally complex hydrosystem, subjected to critical water resource conflicts.

Flash floods pose significant hazards in urbanized zones and have important implications that should probably increase due to global changes. Early warning is thus a priority that could be done by using forecast models. When such events occur on karst basins, well known for their intrinsic complexity, anisotropy and heterogeneity, the lack of knowledge regarding the various hydrodynamic behaviours involved in karst systems prevents to use physical models. A generic black box method seems thus to be adequate; specifically, artificial neural network modelling seems to be a relevant method. To model hydrosystem behaviour efficiently, neural networks need to dispose of relevant data sets constituting input and output variables, and rigorous application of regularization methods. In this study, we propose to apply two kinds of models: feedforward and recurrent neural networks to flash flood forecasting. These models are designed using a specific methodology to diminish their complexity. They are applied to the Lez karst aquifer, located in southern France, and their performances are compared. Recurrent model can be used at longer lead time for operational flash flood forecasting. Nevertheless, for short horizon of prevision, performances of feedforward model are higher than those showed by recurrent one. The comparison of both models is then necessary to guide the improvement of operational flash flood forecasting.

International audience; The Liane River is a small costal river, famous for its floods, which can affect the city of Boulogne-sur-Mer. Due to the complexity of land cover and hydrologic processes, a black-box non-linear modelling was chosen using neural networks. The multilayer perceptron model, known for its property of universal approximation is thus chosen. Four models were designed, each one for one forecasting horizon using rainfall forecasts: 24h, 12h, 6h, 3h. The desired output of the model is original: it represents the maximal value of the water level respectively 24h, 12h, 6h, 3h ahead. Working with best forecasts of rain (the observed ones during the event in the past), on the major flood of the database in test set, the model provides excellent forecasts. Nash criteria calculated for the four lead times are 0.98 (3h), 0.97 (6h), 0.91 (12h), 0.89 (24h). Designed models were thus estimated as efficient enough to be implemented in a specific tool devoted to real time operational use. The software tool is described hereafter: designed in Java, it presents a friendly interface allowing applying various scenarios of future rainfalls, and a graphical visualization of the predicted maximum water levels and their associated real time observed values.

Karst hydrosystems constitute important water resource but their recharge and emptying process are poorly known and quantified. Water resource management is thus difficult. Nevertheless, it is a major issue when rainfall is not uniformly distributed during the year, as in Mediterranean climate. This study proposes a method based on neural networks permitting to simulate karst emptying as a function of the pumping volume during the dry period. Applied to the Lez karst system, the model provides excellent simulations of the water level at the main outlet of the system by using mean pumping discharge and zero rainfall hypothesis during dry period. An arbitrary extreme scenario is also provided by introducing a mean pumping volume.

The ability of the multilayer perceptron to model the inverse relation of a fictitious watershed is investigated. Comparison is done between a new formulation of data assimilation and the standard multilayer perceptron applied to three kinds of models: static, feedforward and recurrent. It appears that both techniques are equivalent and allow a very good estimation of the inverse relation. This study aims at proposing methods to supplement or adapt historical databases to modern instrumentation. Datasets will thus be used over a longer time-series to better apprehend the consequences of global warming.

International audience; Karst aquifers provide water resources for a large part of the Mediterranean population and water resources become a strategic problem during summer when the population increases due to tourism. To help managers to optimize the exploitation of karst aquifer waters this paper studies the ability of a neural model to efficiently simulate and forecast water table levels a few months ahead during the dry season. The neural model is a recurrent multilayer perceptron that learns the relations between inputs (mainly rainfall and pumping discharge) and output (water table level). After a training step using several years of data, the model is assessed on a never seen year to be validated. Particular attention is devoted to the dry season (May\textendashSeptember). The model achieved good forecast of the maximal observed drawdown. To investigate the usefulness of the model as an operational tool, various scenarios of future rainfalls and future pumping are proposed and associated forecasts are analysed. Scenarios are the following: zero-rainfall scenario and mean forecast rainfall 1 week ahead; in both cases, the model proposes accurate estimations of the maximal drawdown. In a second step, pumping scenarios are investigated: a mean pumping scenario during the dry season and the maximum pumping scenario taking into account the actual pumps. One more time, it is demonstrated that the maximal drawdown can be efficiently forecast for the dry season. The methodology is generic and can thus be used with profit by managers on other karst aquifers.

Artificial neural networks (ANN) are nonlinear models widely investigated in hydrology due to their properties of universal approximation and parsimony. Their performance during the training phase ...

International audience; Aqueous mineral suspensions and pastes are greatly used in industry and waste treatment processes. But unfortunately due to their inherent complexity (numerous parameters to consider and non-linearity of temporal behaviour), their physicochemical stability, controlled by their dispersion state, is difficult to predict. A way to have a better understanding of these systems is to apprehend stability by studying settling behaviour of suspensions in function of solid concentration and interparticle interactions. To this purpose, previous works on settling optical analysis were used in addition to rheological approach, to determine in some mineral systems, a suspension typology in function of solid mass fraction and predominant particle interaction. In order to generalize this work to various mineral aqueous suspensions, a modelling study is proposed in this paper. The aim of this work is to predict the stability of mineral suspensions based on a specific index: phase separation index (PSI) previously established using 10 discriminating parameters currently measured in industrial and academic areas. Because of their well-known ability to model non-linear processes, a neural networks-based procedure was used to classify the settling behaviour in four classes: diluted, concentrated (cohesive and non cohesive) and solid suspension. The method was proved to be very efficient, delivering 100% of good classification on various sets of test sets up to 60% of the database allowing thus to predict the suspension stability for applications such as inks, paints, cosmetics. In this research, the predominant influence of mass fraction parameter was showed. (C) 2014 Elsevier B.V. All rights reserved.

Summary As it is the case with many karst aquifers, the Lez basin (southern France) is heterogeneous and thus difficult to model. Due to its supply of fresh water and its ability to reduce flooding however, more in-depth knowledge of basin behavior has proven critical. In addressing this challenging issue, an original methodology based on neural networks is presented herein so as to better understand the hydrodynamic behavior of such systems. Dedicated architecture containing several sub-networks, each being associated to a specific “homogeneous” geological zone that corresponds to a sub-basin contributing discharge, is described. A method, previously proposed for variable selection, has been applied to determine both the relative contribution of the considered zone and its response time. Given the difficulty of verifying such non-observable knowledge, a specific validation step has also been provided. This methodology has been successfully applied to the difficult case of the Lez karst basin, yielding improved knowledge on basin behavior and a revised delimitation of its feeding basin. A new approach has been adopted for the basin, leading the way to additional fieldwork and revised methodologies, particularly regarding the protection of water supply. It should be emphasized that the proposed methodology is generic and applicable to all kinds of aquifers with available and sufficient rainfall and discharge data.

Neural networks are increasingly used in the field of hydrology due to their properties of parsimony and universal approximation with regard to nonlinear systems. Nevertheless, as a result of the existence of noise and approximations in hydrological data, which are very significant in some cases, such systems are particularly sensitive to increased model complexity. This dilemma is known in machine learning as bias–variance and can be avoided by suitable regularization methods. Following a presentation of the bias–variance dilemma along with regularization methods such as cross-validation, early stopping and weight decay, an application is provided for simulating and forecasting karst aquifer outflows at the Lez site. The efficiency of this regularization process is thus demonstrated on a nonlinear, partially unknown basin. As a last step, results are presented over the most intense rainfall event found in the database, which allows assessing the capability of neural networks to generalize with rare or extreme events.

In southern France, flash flood episodes frequently cause fatalities and severe damage. In order to inform and warn populations, the French flood forecasting service (SCHAPI, Service Central d'Hydrométéorologie et d'Appui à la Prévision des Inondations) initiated the BVNE (Bassin Versant Numérique Expérimental, or Experimental Digital Basin) project in an effort to enhance flash flood predictability. The target area for this study is the Gardon d'Anduze basin, located in the heart of the Cévennes range. In this Mediterranean mountainous setting, rainfall intensity can be very high, resulting in flash flooding. Discharge and rainfall gauges are often exposed to extreme weather conditions, which undermines measurement accuracy and continuity. Moreover, the processes governing rainfall-discharge relations are not well understood for these steeply-sloped and heterogeneous basins. In this context of inadequate information on both the forcing variables and process knowledge, neural networks are investigated due to their universal approximation and parsimony properties. We demonstrate herein that thanks to a rigorous variable and complexity selection, efficient forecasting of up to two-hour durations, without requiring rainfall forecasting as input, can be derived using the measured discharges available from a feedforward model. In the case of discharge gauge malfunction, in degraded mode, forecasting may result using a recurrent neural network model. We also observe that neural network models exhibit low sensitivity to uncertainty in rainfall measurements since producing ensemble forecasting does not significantly affect forecasting quality. In providing good results, this study suggests close consideration of our main purpose: generating forecasting on ungauged basins.

International audience; “Cévenol flash floods” are famous in the field of hydrology, because they are archetypical of flash floods that occur in populated areas, thereby causing heavy damages and casualties. As a consequence, their prediction has become a stimulating challenge to designers of mathematical models, whether physics based or machine learning based. Because current, state-of-the-art hydrological models have difficulty performing forecasts in the absence of rainfall previsions, new approaches are necessary. In the present paper, we show that an appropriate model selection methodology, applied to neural network models, provides reliable two-hour ahead flood forecasts.

International audience; A neural network model is applied to simulate the rainfall-runoff relation of a karst spring. The input selection for such a model becomes a major issue when deriving a parsimonious and efficient model. The present study is focused on these input selection methods; it begins by proposing two such methods and combines them in a subsequent step. The methods introduced are assessed for both simulation and forecasting purposes. Since rainfall is very difficult to forecast, especially in the study area, we have chosen a forecasting mode that does not require any rainfall forecast assumptions. This application has been implemented on the Lez karst aquifer, a highly complex basin due to its structure and operating conditions. Our models yield very good results, and the forecasted discharge values at the Lez spring are acceptable up to a 1-day forecasting horizon. The combined input selection method ultimately proves to be promising, by reducing input selection time while taking into account: i) the model's ability to accommodate nonlinearity, and ii) the forecasting horizon.

International audience; The present study focuses on the modeling of the Lez karstic system (France) using artificial neural networks. Two methods of variable selection were compared: cross-correlation and cross-validation. In both cases, the artificial neural network forecasts closely matched the measured discharge, giving Nash criteria higher than 0.8, which can thus provide satisfactory 2-day forecasts.

International audience; After a presentation of the nonlinear properties of neural networks, their applications to hydrology are described. A neural predictor is satisfactorily used to estimate a flood peak. The main contribution of the paper concerns an original method for visualising a hidden underground flow Satisfactory experimental results were obtained that fitted well with the knowledge of local hydrogeology, opening up an interesting avenue for modelling using neural networks.