Volume 16, numéro 4, 2003

FR :

Pour lutter contre les pollutions diffuses en milieu rural, de nombreux programmes d'action se mettent en place. Le développement de recherches sur les connexions parcelle - cours d'eau devrait permettre de mieux comprendre le transfert et la dissipation des polluants dans ce milieu. En particulier, les fossés, structures relativement fréquentes dans les territoires cultivés, peuvent, a priori, avoir une fonction de court-circuit et donc faciliter le transfert des produits phytosanitaires, ou au contraire constituer des éléments de pondération de la pollution. Afin d'éclaircir ce point, une première série d'expérimentations a été menée par le Cemagref (Institut français de recherche pour l'ingénierie de l'agriculture et de l'environnement) dans des fossés de drainage agricole. Une solution aqueuse contenant trois herbicides aux caractéristiques physico-chimiques différentes (isoproturon, diuron et diflufénicanil), et un traceur (chlorures) a été injectée pendant quelques minutes dans quatre fossés. Des échantillons d'eau ont été prélevés à pas de temps fins à deux emplacements en aval du point d'injection. Après dosage par chromatographie au laboratoire, les résultats indiquent une diminution du flux et de la concentration maximale du pic de polluants comparativement à un traceur. En outre, la variation observée est corrélée aux propriétés physico-chimiques des produits, en particulier au coefficient de partage Koc. L'étude présentée montre que la surface de contact (liée à la nature du substrat) et le temps de contact (dépendant essentiellement des conditions d'écoulement) entre les polluants et le substrat sont les paramètres qui influent majoritairement sur la dissipation des produits phytosanitaires.

EN :

The use of pesticide may lead to the contamination of surface and groundwaters. Agricultural nonpoint source pollution originates from land areas which intermittently contribute to the compound transfer to water. Several studies report on the occurrence of pesticides in surface water resources, with concentrations over the limit set by the 80/778 EEC directive for drinking water (0.1 µg/L for each substance and 0.5 µg/L for all pesticides). Numerous herbicides of different chemical families are detected in surface waters, especially triazines and ureas. Their concentrations vary with time and space partly in relation with application patterns and pluviometry. Maximum concentrations are linked to runoff, originating from agricultural fields and primarily occur right after the application periods.

Many methods and levels of actions can be used to reduce water pollution. First, better agricultural practices can be set up, such as choosing the best dose and application period, controlling toxic substance impacts, combining with non-chemical practices. However, pesticide losses from fields can't be totally cancelled because of the complexity of the involved parameters (agricultural practices, climatic conditions, soil physical, chemical and biological properties …). In fact drift during application, runoff or drainage systems may still occur and have an effect on water quality. It may be then pertinent to evaluate to which extent the non treated areas between the fields and the surface water bodies can dissipate pesticide concentrations before they reach them. Pesticides leaving a plot in surface runoff may pass through various landscape components before reaching rivers ; including another field, a ditch, a small brook, a vegetative buffer zone. Besides, pesticide leaving the plot through drainage straightly moves agricultural ditches or streams. However, the contribution of each of these elements in pesticide dissipation is not well known, except for buffer zones (grassed or forest strips) (PATTY (1997). Cemagref (a French research institute), CEH Wallingford and ITCF (Institut Technique des Cereales et des Fourrages) attempt to extend their study to the other elements as agricultural ditches.

This paper deals with the role of farm ditches and small streams in the transport and retention of pesticides from fields to the main river network. Their presence seems to play a significant role in the transfer of nonpoint source pollution (especially in the West of France). Indeed they can either accelerate pesticide transport or reduce it, according to their characteristics (length, flow, bottom sediment or soil characteristics, plants and organic matter contents, etc.). Since 1998, Cemagref has been investigating the retention of pesticides by several natural ditches with varied flows and substratum. A water solution containing three herbicides with different physico chemical properties (diflufenican, diuron and isoproturon) and potassium chloride, a tracer, is introduced with a pump in each ditch for about five minutes with a constant concentration. Water samples are collected in the ditches every two or five minutes at two distances from the injection point. The samples are stored in amber polyethylene terephtalate bottles and frozen. Laboratory analysis is performed by liquid-liquid extraction with dichloromethane and then liquid or gas chromatography depending on the compounds.

The analysis of the water samples highlights a reduction of the maximum concentration and of the accumulative mass of each pesticide with distance compared to the tracer. Indeed, even if all the chloride ions used as tracers are not recovered at each sampling point (due to infiltration or lateral losses), we notice more significant losses for all the studied herbicides. The reduction can reach 70 % of the applied mass for diflufenican compared to the tracer. The retention of pesticides is also linked to their own physical and chemical properties. Thus, diflufenican, which has the highest sorption coefficient value, Koc, is also the most retained pesticide, whereas the total injected mass of isoproturon is recovered in most cases. Diuron has an intermediate behaviour.

In brief, this field experiment proves that the surface and time of contact between pollutants and substratum are likely to play a major role in pesticide retention. An estimated adsorption capacity of each ditch has been assessed, which is based on laboratory sorption experiments on different natural substratum. Despite the few data, a relationship between diflufenican retention in ditches and the estimated adsorption capacity of each ditch has been underscored.

This study also highlighted major limits of field experiments. For example, accurate flow measurements are really difficult to carry out with simple methods for low values. The conventional techniques can't be used with small water height or in ditch where the bottom is filled with plants or grass. Chloride ion was chosen in this study because it is easy to analyze, but the results showed an initial presence of chloride ion in the natural ditch water which incites to replace it by another tracer such as bromide with is not found in the environment in future field experiments.

For all these reasons, some pilot experiments with a physical model (an artificial ditch of 8 m long and 0.4 m wide) are now designed. This equipment allows to adjust and control hydrodynamic parameters such as water flow, water height, and the nature and structure of the substratum. Then, it is possible to quantify both the role played by the substratum, mainly the organic matter content, and the role of the contact time. These parameters could be then taken into account in order to optimize further experiments on adsorption. The primary tests without substratum already give references for hydrodynamic measurements, as the stability of the water flow and the homogeneity of the initial solution concentration.

FR :

La modélisation pluie-débit au pas de temps mensuel, a été étudiée par le biais de quatre modèles qui appartiennent à deux catégories, les modèles conceptuels (modèles à réservoirs), et les modèles basés sur les réseaux de neurones, et la logique floue

Les modèles conceptuels mensuels utilisés sont les modèles de Thornthwaite et Arnell et le modèle GR2M, ainsi que deux modèles représentés par les réseaux de neurones à apprentissage supervisé et le modèle neuro-flou qui combine une méthode d'optimisation neuronale et une logique floue.

Une application de ces modèles a été effectuée sur le bassin de la Cheffia (Nord-Est Algérien), et a confirmé les performances du modèle basé sur la logique floue. Par sa robustesse et son pouvoir d'extrapolation non-linéaire, ce modèle a donné d'excellents résultats, et représente donc une nouvelle approche de la modélisation pluie-débit au pas de temps mensuel.

EN :

Rainfall-runoff modelling is very important for environmental issues, as well as for water management. Due to this importance, several models have been developed to describe the transformation of rainfall to runoff. From these models, we can distinguish three categories: conceptual models; physically-based models and black box models. Conceptual models are designed to approximate within their structures the general sub-processes that govern the hydrological cycle, and they are often used because of their simplicity. The physically-based models are generally distributed models, involve complex descriptions using partial derivative equations, and need some parameter calibration to be adjusted or estimated in situ. These models can not be applied on a monthly scale. In contrast, the black box models rely on linear (or nonlinear) relationships between inputs (rainfall) and outputs (runoff), and they have been widely accepted as a practical tool on different time scales.

In this paper, rainfall-runoff modelling on a monthly scale was studied using four models, from two different categories; conceptual models (reservoir models), and models based on artificial neural network and fuzzy logic. The monthly conceptual models used were the Thornthwaite-Arnell model and the GR2M model with two reservoirs. These models are regarded as mathematical models, and are of simple conception with a reduced number of parameters. In addition, these models are considered the most valid. The two other models were based on artificial neural networks and fuzzy logic, which combine neural optimization methods and fuzzy logic. These models incorporate a flexible mathematical structure that is capable of identifying complex nonlinear relationships between input and output data sets. In contrast to conceptual deterministic models, these models proceed using data learning through input-output systems. Artificial neural network models have been often shown to provide a better representation of the rainfall-runoff relationships. However, it is necessary to investigate different learning methods used with these models.

There are two different learning modes (training). One is data learning (incremental training), which consists of training for each data set, where the weights and biases on the network model are updated each time an input is presented to the network, thus the error between simulated and target (observed) data is minimised for each input. The alternative to data learning is block learning (batch training). In block mode the weights and biases on the network model are updated only after the entire training set has been applied to the network. We have tried a block learning data method, which consisted of learning from the simulation of all data sets. Thus, it evaluates the influence of this model in the streamflow forecasting in real time.

In Algeria, the droughts recorded during the previous years resulted in a reduction of surface water and in unbalanced resources that affected the phreatic underground water due to intensive exploitation. The results from evaluation studies emphasised the instability and vulnerability of surface water resources. The government has decided to carry out an emergency plan, by constructing several reservoirs and dams over the next few years in different regions of the country. However, several hydrometric gauges are disabled, so the series of hydrometric data are short or have gaps, and thus water resource evaluation has become impossible.

One of the objectives of the monthly rainfall-runoff modelling was estimating the stream flow at the mouth of the watershed, so the rainfall-runoff relationship on a monthly scale represents a solution and a reliable method for water management projects. We have selected and applied four models on data from the Cheffia watershed situated in north-eastern Algeria. The catchment of the Cheffia river includes various sub-basins, and has an area of about 575 km². The study was carried out on a twelve-year data set, split into a six-year calibration period, and a six-year validation period. Our research compared the models based on model characteristics, like simplicity and parameterisation, and also conceptual models were compared to parsimonious models. In addition, our research compared modelling results, based on the assessment of quantitative indices and statistics, such as the Nash criterion, the root mean squared error and a comparison of means during the calibration and validation periods.

Model results have confirmed the strong performance of the fuzzy logic based model, for two periods, and this model best stimulated streamflows. Whereas the neural network model based on block learning is unable to reproduce the high runoff values, this model can to be used for simulation of the runoff only. Because of its robustness and non-linear extrapolation power, the neuro-fuzzy logic model gave better results, so it represents a new method of rainfall-runoff modelling in monthly time steps.

FR :

Dans le cas de la gestion en temps réel des réseaux d'assainissement, la première étape peut, par exemple, consister à vérifier qu'une manipulation des organes de contrôle tels que les vannes et pompes est capable de minimiser les déversements vers le milieu naturel. Cette gestion, que l'on appellera " gestion de référence ", permet de déterminer les stratégies de commande sur toute la durée de l'événement pluvieux connu à l'avance. Ce calcul se fait donc à la fin de l'événement pluvieux et permet de dire ce qui aurait pu être fait avec les organes de régulation en terme de minimisation des volumes déversés. La programmation linéaire par les graphes et la programmation linéaire mixte permettent de déterminer une solution optimale. Cet article s'intéresse à la vérification de l'optimalité et à l'applicabilité de la programmation linéaire par les graphes comparée à la programmation linéaire mixte dans le cas de la " gestion de référence " sur le réseau d'assainissement de Saverne (France). En comparant les volumes déversés par ces deux techniques d'optimisation sur 34 événements pluvieux, nous pouvons confirmer que l'approche par les graphes ne donne pas toujours le minimum global. Les résultats ont montré que la programmation linéaire mixte fournit des temps de calcul qui peuvent atteindre plus de 24 heures. Par contre, l'approche par les graphes permet un temps de calcul de l'ordre de 5 minutes en moyenne avec un minimum global en terme de volume déversé atteint qui n'excède pas 5% par rapport à la solution fournie par la programmation linéaire mixte.

EN :

The first stage of real-time management of wastewater systems could, for example, consist of making sure that the use of controls such as valves and pumps can indeed minimise the discharge into the natural environment. This management step, referred to as reference management, is used to determine the control strategies over the entire duration of a rainfall event known in advance. The calculation is therefore performed at the end of the rainfall event and is used to determine what could have been done with the regulation components (e.g. in terms of minimising the volumes discharged). The calculation can also show whether or not it is necessary to control the valves and pumps during the rainfall occurrence (dynamic management) rather than fixing the flow rates in advance (static management) if the receiving body of water is to be protected from discharges.

In the area of operational research, management controls can be determined with the help of linear programming. Here the aim is to minimise the linear function (f), generally called the cost function or objective function, under different linear constraints (g). There are several variants of linear programming. The first one is mixed linear programming, where some variables are required to be integers or even binary. Conventional calculation techniques such as branch-and-bound method provide a global minimum solution, but the calculation is very lengthy.

Another variant of linear programming is graph programming. This optimisation technique consists of modelling the hydraulic behaviour of most of the constructions that can be found in wastewater systems (main drains, storm overflows, storm water basins, etc.). It has been applied to the wastewater system of Saverne in order to minimise the volumes discharged into the natural environment. In order to ensure that the constructions are modelled correctly and to optimise the functioning of the controls, saturation constraints had to be added to the choice of the arcs of the graph. The primal-dual algorithm no longer provides a global minimum solution when these constraints are added. In contrast, the calculation time is much shorter than that for mixed linear programming.

This article is aimed at comparing the results in terms of the minimisation of the discharged volumes by means of linear programming with graphs and mixed linear programming, as part of the reference management applied to the wastewater system of Saverne. The final goal was to be able to select a compromise between the relevance or the accuracy of the results and the means to achieve them.

We have shown that wastewater constructions such as main drains, storm basins and overflows can be modelled simply with the two techniques above. However, it is necessary to add binary variables in the case of mixed linear programming and a degree of arc saturation if the graph approach is used. The branch-and-bound algorithm used for mixed linear programming can be used to obtain a global minimum solution, with a very long calculation time. In contrast, even though the convergence time is very short for linear programming with graphs, the global minimum cannot be ensured because the algorithm used imposes independence with respect to the choice of the arcs to be saturated.

Given the benefits and drawbacks of each approach, we have attempted to use the example of the wastewater system of Saverne to quantify the calculation time and the differences in terms of the discharged volume. The results have shown that mixed linear programming requires calculation times that can last over 24 h. In contrast, with the graph approach, the calculation takes approximately five minutes on average, with a global minimum in terms of volume that does not exceed 5% as compared to the solution obtained by mixed linear programming. We have shown that a solution requiring a much shorter calculation time is available and offers a compromise between exact determination and an optimised associated calculation time.

FR :

Dans cette étude, on propose un modèle hydraulique capable de contribuer à la gestion des eaux de la rivière Sebou au niveau de la retenue d'un barrage de garde situé à l'intérieur de la plaine agricole du Gharb. Le modèle hydraulique élaboré (MHS.1) est du type filaire et utilise un schéma de différences finies. L'écoulement est influencé par la présence du barrage à l'aval et de nombreuses grandes stations de pompage utilisées pour l'irrigation le long du tronçon étudié. Cependant, les données relatives à la quantité d'eau pompée au niveau de ces stations ainsi que par les particuliers sont rarement disponibles. Ainsi, une attention particulière a été attribuée à l'estimation du pompage vue son importance quantitative. Les résultats du calibrage et de la validation du modèle pour des périodes de basses eaux de l'année 1997 sont très satisfaisants. Le modèle donne les valeurs du niveau d'eau aux stations de pompage et permet de suivre l'évolution de la réserve de la retenue du barrage. Ce code regroupe dans un seul outil des données provenant de différentes sources et utilisées pour la première fois dans un modèle hydraulique. Il représente un atout considérable pour les organismes publics gestionnaires des ressources hydriques.

EN :

The studied reach

The Sebou River (600 km) is an important river in Morocco and its waters are solicited for several different uses. The Sebou has an average bottom slope of 10^-4, variable geometry and many meanders. The flow is characterised by considerable annual and seasonal variations (Figure 2). The studied reach is situated between the town of Belksiri and the Lalla Aïcha dam. Flow is influenced by the presence of two dams, the Al Wahda upstream and the Lalla Aïcha in the downstream reach. The first dam was constructed on the Ouergha River, which has a torrential regime. The second dam comprises five principal and two secondary radial floodgates and these gates are opened from the bottom. This dam is completely opened during the period of high flows. The maximum flow during this season is 1800 m³/ s. The dam has a catchment area of 2700 km². The maximum volume of the dam reservoir is 37 Mm³. Its length of influence is about 120 km.

During the dry season, the floodgates are partially closed in order to increase the water level upstream. The maximum level upstream of the dam is 6.5 m NGM (the bottom is at -1 m). This situation facilitates the pumping of water for agriculture, allowing the irrigation of 15,600 hectares of rice. A volume of 200 Mm³ of water is mobilised annually, which, before the construction of the dam, was lost to the Atlantic Ocean.

The hydraulic model MHS.1

The hydraulic model MHS.1 is based on a modification (essentially the representation of the topography and the outputs) of the DYNHYD5 model. It solves the one dimensional Saint-Venant equations of continuity and momentum (equations 1 and 2). The Manning coefficient used in the momentum equation is evaluated initially by the empirical formula (Formula No. 3) proposed by Chow. The factor n0 is evaluated from granulometric measurements that were carried out from upstream to downstream in the studied reach. The others coefficients were evaluated from observations of the river in aerial photos, from the cross sectional areas and available photos, and from field visits. MHS.1 uses a network called ''Link-Node''. The equations of continuity and momentum, expressed in a finite difference manner, give respectively equations 4 and 5. These equations are solved using a Runge-Kutta procedure.

Discretisation of the studied reach

The discretisation of the studied reach was performed using aerial photos achieved by the ORMVAG (L'Office Régionale de la Mise en Valeur Agricole du Gharb) in 1983. These photos were taking in a dry period where the river was nearly dry. This situation permitted a good stereoscopic visualisation of the river morphology. The river reach was divided into 529 grids with a length varying between 50 and 900 m. Data on cross sectional areas from the ORMVAG and other sources were used. Near the town of Souk Tlat (Figure 1), we exploited a new technique called ''Numeral photogrammetry'', which allowed us to reconstitute many cross sectional areas. This technique uses principally stereoscopic pairs of aerial photographs and photogrammetry software. The remaining cross sectional areas were evaluated from observations on aerial photos and from field visits.

Evaluation of the pumped water

One of the important factors that affect flow in the studied reach is the intensive pumping of waters along the river. The pumped water was divided into two types. The first type corresponded to the ten central stations managed by the ORMVAG (Fig. 1). The data of this first type were neither centralised nor easily available. Only the data at the important S2 station were readily available. The second type corresponded to water pumped by individuals and is less quantified than the first type.

Two major hypotheses were adopted. First, the pumped flow at the S2 station was assumed to be equal to 25% of the total flow pumped by all the ORMVAG stations. The stations were classified into three classes according to their theoretical capacity (Table 1). This hypothesis allowed the estimation of the unknown pumped flow at the nine other stations. We further assumed that in the neighbourhood of each station, the flow pumped by individuals was equal to the flow pumped by the station. This latter hypothesis was adopted on the basis of a field investigation in a 7 km characteristic reach. Figure 3 shows the evolution of the overall pumped flow evaluated for the months of June and July 1997. These two months were used respectively for the validation and calibration of the model.

Calibration and validation of MHS.1

The Manning coefficient, estimated initially by the Chow formula (3), varied along the studied reach. It ranged from 0.02 to 0.04 s/m¹ /³, with a mean value of 0.037 s/m¹ /³. In the calibration procedure, the Manning coefficient was modified to the same degree along the studied reach because we assumed that the sources of errors involved in its evaluation are identical for all the grids.

Along the studied reach, the only available measured data are the water levels at the S2 station and upstream of the dam. The period chosen for the calibration was from 07/01/1997 to 07/30/1997. The upstream boundary (at the Belksiri hydrological station) was given as values of the water level as a function of time (Figure 4). The downstream boundary was given as values of the discharge (flow through the dam gates) as a function of time (Figure 5). Figures 6 and 7 give the results of the calibration (month of July). The Manning coefficient decreased for all the reaches by 0.008 s/m¹ /³. These figures show good agreement between the calculated and the observed water level at the S2 station and near the dam. In order to confirm the results of the calibration test, we proceeded with a validation test of the model for the period from 06/04/1997 to 06/30/1997. The results are also satisfactory (Figure 6 and 7, month of June).

Figure 8 shows the evolution of the water level on 12/06/1997. The water level profile remains parallel to the bed profile for the zones situated very far from the dam (the downstream end). From the 45^th kilometre (between stations S7 and S8, see Figure 1), we begin to detect the effect of the dam, characterised by an increase in the water level (and therefore an increase in depth). Figure 9 show the evolution of stream velocity from the upstream to downstream regions on 12/06/1977. Great variations in velocity can be seen due to the changes in river geometry. Also, these variations tend to decrease downstream, reflecting the effect of the dam.

Figure 10 represents the evolution of the water reserve available for the whole reach during the months of June and July. It shows a series of decreases in this variable due to the pumping of water. The reserve reaches very low levels (15 Mm³) compared to its maximal capacity, which is 37 Mm³. Also, there is an interrelationship between the evolution of the reserve, and pumped water and the flow differences between upstream and downstream. The reserve increases when the upstream-downstream flow difference is greater than the pumped flow. Inversely, when the pumped water is greater, the water reserve decreases.

Finally, in this study we proposed a mathematical model that can provide the stages at all locations of the studied reach, specifically at the pumping stations. Water reserve availability can also be provided at any moment, allowing rapid interventions when this variable begins to decrease dramatically. However, more measured water levels at different stations could improve the present results. Also, other considerations must be included such as hydroelectric energy production in dams upstream and river characteristics. Thus, a multipurpose model of the river must be used. More hydraulic data can improve the accuracy of the present model.

FR :

Une grande quantité d'eau est perdue dans les zones de pâturage et prairies naturelles du fait de la présence dans ces régions de plantes à forte transpiration. La gestion du couvert végétal et des bassins versants a été proposée comme moyen pour augmenter la disponibilité des ressources en eau. Des efforts accrûs ont été consacrés au développement de pratiques de gestion et d'outils pour évaluer le potentiel d'augmentation de la ressource en eau. La modélisation hydrologique joue un rôle clé dans ces efforts. Un des outils les plus complets pour la modélisation dans les zones de pâturage et prairies naturelles est le modèle SPUR. Il s'agit d'un modèle de bassin versant, spatialement semi-distribué. Le modèle est constitué de cinq modules principaux qui incluent les aspects suivants: climat, hydrologie, plantes, animaux et économie. La composante hydrologique du modèle prend en compte à un pas de temps journalier les phénomènes de ruissellement, évapotranspiration, percolation et écoulement latéral. Le ruissellement est calculé à partir du numéro de courbe qui dépend du couvert végétal, des pratiques culturales, ainsi que des conditions hydrologiques. Cependant l'utilisation par le modèle de la méthode du numéro de courbe pour déterminer le ruissellement pose de sérieux problèmes quant à l'efficacité du modèle. Dans notre recherche, nous avons substitué la méthode des numéros de courbe par l'équation de Green et Ampt. Un avantage majeur de cette approche est l'utilisation de l'intensité de la pluie comme variable de forçage au lieu de la pluie journalière. De plus, cette équation d'infiltration utilise des paramètres physiques comme la conductivité hydraulique à saturation. L'objectif de cet article est de présenter les performances du modèle SPUR original et modifié sur trois types de couvert végétal : sol nu, sol enherbé et buissons.

Trois années de mesures collectées sur le bassin versant de Throckmorton (Texas, Etats Unis d'Amérique) ont été utilisées pour la calibration et la validation des modèles. Les performances des modèles ont été évaluées en utilisant le coefficient d'efficience de Nash et Sutcliff. Le calage a porté sur la première année de mesure. Pour le modèle original, le calage a consisté à ajuster les numéros de courbe de manière à optimiser l'efficience. Pour le modèle modifié, il n'a été procédé à aucun calage. Les valeurs de conductivité hydraulique à saturation ont été estimées en utilisant des équations de pédotransfert en se basant sur les propriétés texturales et structurales des sols. L'introduction dans le modèle SPUR de l'équation de Green et Ampt a considérablement amélioré la performance du modèle pour la prévision du ruissellement sur tous les sites. L'efficience moyenne pour les prévisions du ruissellement mensuel sur sol nu est de 0.16, alors que celle ci est négative pour le modèle original (-0.11). Pour les sites enherbés l'efficience du modèle modifié est de 0.48 alors qu'elle est négative pour le modèle original. L'utilisation du numéro de courbe a résulté en une surestimation systématique du ruissellement sur tous les sites. De manière générale, le modèle original et le modèle modifié présentent de meilleures performances sur les sites non nus. Ceci est dû au fait que les deux modèles sous-estiment de manière significative l'évaporation sur les sols nus. Un des désavantages des deux modèles est en effet de limiter l'évaporation aux premiers 15 cm du sol. L'introduction de l'équation de Green et Ampt a amélioré les performances du modèle pour la prévision du ruissellement aussi bien à l'échelle mensuelle qu'annuelle. De plus, le modèle modifié est sensible au type d'occupation du sol et est donc adapté comme outil pour l'analyse de scénario en vue de préserver les ressources en eau.

Une analyse de sensibilité a été conduite afin d'évaluer l'impact des paramètres d'entrée sur les sorties des deux modèles. L'analyse de sensibilité a consisté à modifier systématiquement les paramètres d'entrée de plus ou moins 10%. Pour le modèle original, l'analyse a porté sur l'influence du numéro de courbe, et pour le modèle modifié celle ci a porté sur l'étude de l'impact lié aux paramètres utilisés pour calculer la conductivité hydraulique à saturation. Concernant le modèle original, une augmentation du numéro de courbe de 10% entraîne une augmentation du ruissellement de 120% pour le sol nu, et aux alentours de 100% pour les autres sites. L'impact de ces variations sur l'évaporanspiration est minimal, avec une variation maximale de 16% pour le sol nu. Concernant le modèle modifié, la teneur du sol en sable est le paramètre ayant la plus grande influence sur la quantité d'eau ruisselée pour le sol nu. Par contre, pour les lysimètres ayant un couvert végétal, le pourcentage de sol couvert par la canopée est le factor majeur contrôlant la quantité d'eau ruisselée. Les paramètres liés au couvert végétal ont un plus grand impact sur le ruissellement que les paramètres liés aux propriétés intrinsèques du sol.

Globalement l'introduction de l'équation de Green et Ampt a amélioré les capacités prédictives du modèle. Outre le fait que le modèle modifié ne nécessite pas un calage particulier pour la détermination des paramètres de transfert de l'eau dans le sol, il se base sur l'intensité de la pluie pour la détermination du ruissellement. Il a été montré que le modèle modifié est sensible aux changements de type d'occupation du sol. Il peut donc être donc utilisé comme outil pour évaluer l'impact de différents scénarios d'occupation du sol sur les ressources en eau dans les zones de pâturage et prairies naturelles. Toutefois, des améliorations, telles que l'introduction de l'impact du développement de fissures sur l'infiltration (écoulement préférentiel) ainsi que sur le phénomène d'évaporation devraient être prises en compte afin d'améliorer les prévisions du bilan hydrologiques, notamment sur sol nu.

EN :

A large amount of water is lost on rangelands due to the conditions of the watersheds, specifically due to the presence of plant species with a high transpiration rate. Vegetation manipulation and watershed management have been proposed as means to increase water yield. Increasing efforts have focused on developing management practices and tools to evaluate the potential for increased water yield. Hydrologic modeling plays a key role in these efforts, and one the most comprehensive tools for simulating rangelands is the SPUR model. However, some limitations seem arise from the use of the SCS curve number method for simulating the runoff/infiltration process. In this project, the SCS curve number method was replaced by the Green and Ampt infiltration equation. One of the major advantages of the approach is that it relies on rainfall intensity rather than on daily rainfall, and it uses physical parameters such as the saturated hydraulic conductivity. The original and the modified SPUR models were tested for three different covers: bare, grass and mesquite. The use of the Green and Ampt infiltration equations improved model prediction of surface runoff. Furthermore, the model was shown to be sensitive to vegetation manipulation, and could be used as a water resources management tool.

FR :

En septembre 2002, les régions méditerranéennes françaises et notamment le département du Gard ont été affectées par des précipitations d'une extrême intensité. On estime que 80% de ce département a été inondé, on dénombre 23 victimes et les dégâts ont été évalués à 1.2 milliards d'euros. Cette catastrophe hydrologique soulève à nouveau les problèmes de la fréquence de ces événements et de l'augmentation des forts cumuls de pluie ces dernières années. L'objet de cet article est d'apporter quelques éléments de réponse, notamment à travers l'analyse régionale des pluies extrêmes journalières ayant affecté la région Languedoc-Roussillon de 1958 à 2002.

La fréquence régionale des pluies extrêmes est estimée en prenant en compte la superficie couverte par ces événements en fonction des hauteurs pluviométriques. A l'échelle régionale la période de retour de l'événement varie entre 80 ans pour la superficie touchée par au moins 200 mm à 140 ans pour celle couverte par 300 mm.

La stationnarité des fréquences des pluies extrêmes est analysée à partir des chroniques du nombre annuel d'événements pluvieux dépassant 200 mm, 250 mm et 300 mm en 24h maximum, entre 1958 et 2002 sur la région. Les tests de stationnarité ne révèlent pas de tendance significative à l'augmentation de ces fréquences. Les données historiques aboutissent aux mêmes conclusions. L'augmentation réelle des inondations est en fait principalement liée à l'augmentation de la vulnérabilité des bassins.

EN :

In September 2002, the Gard department in the South of France was affected by heavy precipitation that covered a broad geographical area. It was estimated that 80% of the department was flooded; there were 23 victims and the damage was evaluated to be 1.2 billion euros. This hydrological catastrophe raised questions about a possible increase in the frequency of these events during recent years, since several other severe flooding events have been observed in the region over the last 15 years. The aim of this article is to explore these questions through a regional analysis of the extreme daily rainfall that affected the Languedoc-Roussillon region between 1958 and 2002. The daily rain data were used because they are the most available type of information over the observation period. Usually, the rainfall hazard description is based on statistical analysis of the maximum rainfall depth observed at a given rain gauge. However, because the spatial variability of rainfall in the Mediterranean region, such results are only representative of local rainfall conditions. Moreover, this type of analysis does not take into account the spatial coverage of the precipitation, which is another factor influencing the resulting floods. Thus, the regional frequency of extreme rainfall was estimated by taking into account the area covered according to a given rainfall depth. For each rainfall event, a rain field was built using a kriging interpolation (NEPPEL et al., 1997). The isohyet area defined a rainfall threshold from 10 to 300 mm with a step of 10 mm calculated for each rainfall event. For each rainfall depth from 10 to 300 mm with a step of 10 mm, the probability distribution of the isohyet area was estimated. The regional rainfall hazards were described with the Depth-Area-Frequency curves (DAF) for 24-h periods. It was shown that at on regional scale, the return period of the last event varied between 80 years for the surface affected by at least 200 mm and 140 years for the surface covered by 300 mm. Compared with other major events that have occurred in the region, it appears that the September 2002 event one was characterized by :

1. the spatial extension of the heavy rainfall, for example more than 1800 km² were affected by at least 400 mm in less than 24 h;

2. the spatial localisation of the heaviest rainfall depths, which were measured over the highest relief (1000 m to 1500 m) as usual in the 'cévenols' meteorological situation, but rather in the plain where the altitude lies between 200 m and 300 m.

The stationnarity analysis of the extreme rainfall frequency was based on the annual number of events exceeding 200 mm, 250 mm and 300 mm over a 24 h maximum duration, between 1958 and 2002. The hypothesis of random events against the hypothesis of a trend or a sudden break in the mean was examined through several statistical tests. The procedures used were the rank correlation test, PETTITT's test, BUISHAND's test, HUBERT's segmentation procedure, a linear regression procedure, and the turning points procedure. Detailed descriptions of these tests can be found in KENDALL and STUART (1977), LUBES-NIEL et al. (1998) and WMO (2000). Except for the rank correlation test, all the procedures led to the conclusion that the three series are randomly distributed at the level of significance 1%, 5% and 10% respectively. Thus no significant increase in extreme rainfall frequency seems to appear. Although the study period was short, 45 years, compared with climatological variability, LUBES-NIEL et al. (1998) show that the procedures used were adapted in detecting trends in 50-yr time series. In considering historical rainfall data before 1958 in the same region, at least two extreme rainfall events could be compared with the event on 8-9 September 2002: in October 1940, 840 mm of rainfall were measured during 24 h in the Pyrénées-Orientales district and in September 1900, 940 mm were observed over 24 h in Valleraugue, upstream in the Herault catchment. Furthermore, if the evolution of the rain gauge network density is taken into account, one can argue that such an event could have occurred more frequently. Indeed, the number of rain gauges has varied from 162 gauges in 1900 to 330 today. It has been shown that the number of observed rainfall events varied according to the area of the events and the network density (NEPPEL et al., 1998b). For example, an event of 150 km² (corresponding to the area covered by more than 600 mm in September 2002) had a probability of 70% to be observed by the network between 1958 and 1993. If one considers the period 1920-1939, this probability decreases to 30%.

In addition, the basin vulnerability has increased. The regional population has grown from 1,460,000 inhabitants in 1949 to 2,300,000 in 2000. At the same time, urbanization has expanded widely. Moreover, this new population came from other districts, and they are not familiar with the Mediterranean rainfall regime and the resulting flash floods. Buildings have often been constructed near rivers, which are attractive building sites, and sometimes even in the river's main channel, increasing the flooding risk and the flood damages. Thus, rather than climate change, for which the effect on extreme rainfalls cannot be proved, the development of basin urbanisation and vulnerability could explain the apparent increase in floods. As the regional population is expected to reach more than 3,000,000 by 2030, it is necessary to take into account the flood risk in future urban planning.