Volume 15, numéro 3, 2002

FR :

Dans le contexte de la rareté des ressources en eau, une approche globale de la gestion à long terme d'un système de stockage/transfert/distribution d'eau est proposée. L'objectif principal de la gestion d'un tel type de système est de gérer les réserves et les délestages de manière à minimiser les écarts entre offre et demande, ceci à partir d'une prédiction de la demande et des apports.

Ainsi, on propose une approche à horizon glissant et surtout une procédure d'adaptation des pondérations du critère fondée sur la logique floue. Cette notion d'adaptation du critère parait tout à fait judicieuse quand on connaît la difficulté de définir les pondérations de tels problèmes d'optimisation et son influence sur la pertinence de la solution obtenue. On vérifie ici l'apport essentiel de la logique floue qui permet d'appréhender finement les enjeux en présence dans la gestion de long terme du système stockage/transfert/distribution d'eau. Le problème de gestion à long terme est résolu par une heuristique améliorée utilisant la programmation linéaire et la programmation dynamique pour réduire les effets de la discrétisation spatiale qui est si limitative dans ce contexte.

L'approche de gestion proposée est effectivement appliquée à un cas d'étude qui permet de mettre en évidence sa relative simplicité de mise en œuvre.

EN :

Since the origins of history, irrigation of agricultural lands has been reported to be an activity of great concern for many human societies. At the beginning, natural irrigation systems such as the Nile River flowing through the sands of Egypt have provided to neighbouring populations some amenities in their hard life. However irregular cycles of floods and droughts were a serious impediment to permanent settlements and to a continuous improvement of life conditions. Consequently human ingenuity has been continuously challenged by the development of new ways and means to master water resource system (WRS).

During the last century, improved civil engineering techniques and the development of digital control systems and techniques have dramatically increased the power of human societies over their water resources. However many problems, which received some attention in the past, now require new approaches, given the steady increase in water demand and the introduction of environmental conservation considerations. Today, intelligent systems techniques appear to be able to give some insight in this direction by improving the efficiency of the different decision steps involved in the management and control of such systems. This paper focuses on the problem of the long-term management of a water resource system composed of a network of dams and river reaches. This system is viewed as a hybrid dynamic system, called here a storage/transfer/distribution system. The main long-term management objective of such a system is to manage reserves and releases so as to minimise the deficit between supply and demand by taking into account predictions of demand and contributions.

Thus, in the present context of water resource scarcity, a complete approach for long-term management of a storage/transfer/distribution system is proposed. To take into account major uncertainties related to the operations of this kind of system, a sliding horizon approach (it consists of readjusting each week the release plan over the whole coming year, according to the present reserves, the most recent long-term demand estimation and the programmed release for the next week). In addition, an adaptation procedure of weighting parameters of the minimisation criterion based on fuzzy logic is implemented. The definition of an optimisation objective function is in this case a very intricate question since it involves competition, uncertainty and geographical dispersion. However, it is crucial to guarantee the quality of long-term management. This is why Fuzzy Logic is used as a particularly appropriate means to refine on-line the formulation of the objective function of the recurrent optimisation problem. Fuzzy Logic is also shown to be very useful in defining what is at stake in the long-term management. This criterion adaptation concept seems judicious, in view of the difficulty of defining the weighting parameters of such optimisation problems and their influence on the relevant solution obtained.

The long-term management problem is solved with improved heuristics using linear programming and dynamic programming in order to reduce the effects of spatial discretisation, which is so restrictive in this context. The suggested approach is applied to a case study, which highlights its relative simplicity of implementation.

FR :

Le Lac Saint-Jean est un réservoir hydroélectrique ayant une superficie de 1 053 km² et une profondeur moyenne de 6 m. Ses eaux se renouvellent de quatre à cinq fois par année, dont deux fois durant la saison printanière. Entre juin et octobre, l'épaisseur de la couche photique du lac est environ de 4 m et la thermocline est presque toujours plus profonde que 20 m avec une température moyenne saisonnière de 16°C. Les éléments nutritifs (phosphates et nitrates) demeurent à un niveau très bas durant toute la saison. Le taux moyen de chlorophylle a est de 1,3 mg m^-3 avec des valeurs maximales de 3 à 4 mg m^-3. Le coefficient de corrélation entre la chl. a et la température, pour l'ensemble des données du lac, est de 0,56 (p<0,01).

La communauté phytoplanctonique (>20 µm) compte environ vingt espèces mais la dominance est assurée seulement par les Diatomées Asterionella formosa et Tabellaria flocculosa. Ces deux espèces représentent plus de 90 % de tous les organismes entre juin et octobre. D'autres espèces comme Melosira (Aulacoseira) distans, M. islandica et M. ambigua ne sont importantes qu'au début de juin tandis qu'Anabaena flos-aquae ne l'est qu'à la mi-juillet. La communauté zooplanctonique regroupe 17 espèces réparties en 4 catégories : Copépodes, Cladocères, Rotifères et Protozoaires. À partir de la mi-juillet, le zooplancton est surtout dominé par le Rotifère Polyarthra vulgaris et par le Protozoaire Codonella cratera. Ayant des cycles vitaux relativement courts, ces deux espèces forment à elles seules entre 65 et 82% de tous les individus. Dans les 5 premiers mètres de la colonne d'eau, le coefficient de corrélation entre la densité de ces deux espèces est de 0,55, (p<0,01). La variabilité spatio-temporelle des biomasses et la distribution des organismes planctoniques du lac sont influencées par l'hydrodynamique de l'écosystème, particulièrement par les variations saisonnières du débit qui influent sur les nutriments, la température et également par le brassage intense qui maintient la thermocline à un niveau au-dessous de 20 m.

EN :

The lake Saint-Jean is well known to fishermen and renowed for its freshwater salmon (Salmo salar). Of its 21 tributaries, three provide about 75% of its water; they are the Peribonka, the Mistassini and the Ashuapmushuan rivers. Although it is the fifth largest lake in the province of Quebec, its water is renewed four to five times a year; about a quarter of its 1 053 km² area is only 3 m deep, with a maximum depth of 63 m in the middle. This kind of shallow lake is very sensitive to winds.

The present study looked at the spatial and temporal variability of phytoplankton and zooplankton; ten stations were distributed over the whole lake (Figure 1 and Table 1) and water samples were taken at different depths (1, 2, 5, 10, 15 and 20 m when possible) from June to October 1997. Several physicochemical parameters were measured; phytoplankton and zooplankton samples were collected as well and the main species were identified and counted. The mean surface water temperature was 16° C compared to the average of the 1, 2 and 5 m data of 15°C and the deep layer (10, 15 and 20 m) of 13°C; from mid-August to mid-October, the water temperature is the same at all depths because of very good mixing due to the strong winds. Referring to the same layers, the mean values of chlorophyll a for the study period were 1.6, 1.3 and 0.6 mg m-3. For the early June sampling, values of chlorophyll a varied from 0.6 to 1.9 mg m?3. As for July and August, values between 3 and 4 mg m^-3 were recorded; the overall mean of chlorophyll a was 1.3 mg m^-3. The correlation coefficient between chlorophyll a and the temperature, for all data, was 0.56 (p<0.01). As stated before, a very good mixing of the upper water layer is encountered, bringing the phytoplankton to depths below the photic layer evaluated at about 4 m; such conditions of light limitation do not foster a good photosynthetic activity.

Nitrates and orthophosphates were always low during the study, respectively lower than 0.8 mg L^-1 and between 0.1 and 0.2 mg L^-1. Linear or exponential regression (Figure 5) between the sum of nitrites and nitrates and chlorophyll a gave correlation coefficients of -0.48 or -0.52 (p<0.01), suggesting a decrease in inorganic nitrogen caused by photosynthetic activity. Of the 26 species of phytoplankton identified with dimensions above 20 µm (Table 2), only two dominated during the study period; these were the diatoms Asterionella formosa and Tabellaria flocculosa with an occurrence between 72 and 87% for the former and between 11 and 24% for the latter. The geographical variability is presented (Figure 6A) along with the seasonal and vertical variability. Such distributions likely result from the combination of the inflow from the Peribonka river, the Coriolis effect and the outflow to the Grande-Decharge river.

The main zooplankton pesent in the lake have a short life cycle; two of the 17 species identified were encountered most of the time, the rotifer Polyarthra vulgaris and the protozoan Codonella cratera accounting for 10 to 29% and 36 to 72% of those species identified (July to October). The correlation coefficient between these two species was 0.55 (p<0.01) in the first 5 meters of the water column (Figure 8). The reciprocal diversity index (Simpson) varied between 0.63 and 0.97 for the zooplankton community, reflecting the dominance of only few species; such an index was not calculated for the phytoplankton because the nanoplankton (<20 µm) was not sampled and is often very important in such systems. However, the dominance of the protozoae and the rotifera, organisms with rather short life cycles (few hours to few days), might have some relation to the hydrodynamics of the ecosystem.

With a thermocline always deeper than the photic layer, the water column is mixed intensively and the phytoplankton is often at depth with little or no light for photosynthetic activity. Based on the low nitrate and chlorophyll a levels, we consider the lake Saint-Jean as oligotrophic. This project was the first systematic study of planktonic populations in the lake and was just a part of a larger study about the distribution and feeding of rainbow smelt (Osmerus mordax) larvae. This will be one of the main steps to understand trophic relationships in this ecosystem famous for sport fishing of landlocked Atlantic salmon (Salmo salar).

FR :

Une procédure d'évaluation des risques pour l'écosystème aquatique engendrés par un dépôt de matériaux de dragage dans une gravière type a été élaborée, et testée avec des échantillons de sédiments d'un canal du Nord-Est de la France. La procédure comporte une étape d'évaluation sommaire des risques, à partir de quotients des concentrations mesurées par les critères de danger correspondants, et une étape d'évaluation détaillée où des essais de toxicité et de lixiviation en colonnes sont mis en œuvre. Le scénario testé retient trois hypothèses, qui concernent (a) les effets sur les peuplements d'invertébrés benthiques, représentés notamment par Hyalella azteca et Chironomus riparius, (b) les effets sur les peuplements d'organismes pélagiques, représentés par Chlorella vulgaris, Ceriodaphnia dubia, et Brachionus calyciflorus, et (c) la pollution de la nappe alluviale associée. Différentes modalités d'exposition (essais normalisés, microcosmes) ont été testées. Dans le contexte particulier des trois sédiments étudiés, ces hypothèses se sont avérées plus ou moins discriminantes, la pollution de la nappe étant la plus sensible. Des améliorations de la procédure doivent être envisagées qui concernent à la fois la formulation des hypothèses (risques à court et long terme sur les organismes pélagiques), et les protocoles d'essai, tant pour les organismes du sédiment (rôle de la nourriture notamment) que pour les essais de lixiviation en colonnes.

EN :

When contaminated by metals or synthetic organic compounds, dredged sediments may have negative impacts on receiving ecosystems. Therefore, there is a need for an operational risk assessment approach. Such a framework is proposed for dredged material deposits in open gravel quarries, which is a rather common means of disposal in France. The first step of the assessment relies upon chemical characterisation of the sediments; the resulting concentrations are divided by "probable effect concentrations" and pooled together, in order to calculate a global hazard quotient. According to the value of this quotient, several decisions can be taken: (a) undertake a detailed risk assessment, (b) dispose of the materials without further constraints, or (c) in case of uncertainty, do some biological testing (with Hyalella azteca and Chironomus riparius) in order to allow decisions. The second step is a detailed ecological risk assessment. Three different assessment endpoints have been proposed, which are (1) the deposit should have no effect on the structure and abundance of benthic invertebrates in the quarry, (2) it should have no long term effect on pelagic species, and (3) it should not cause groundwater pollution, as such quarries are in fact cross sections of shallow alluvial groundwater aquifers. A fourth assessment endpoint should be introduced, regarding health risks for recreational uses, including fishing, but this endpoint was not implemented in the current version of the approach. The analysis phase includes aquatic bioassays (bacteria - Metplate TM-, algae, microcrustaceans Ceriodaphnia dubia, rotifers Brachionus calyciflorus), and leaching assays in columns under ascendant flow.

The proposed approach was tested with 3 sediments from a canal located in the north-eastern region of France. Microcosm assays were introduced in parallel to the proposed tests, in order to explore alternatives to standardised bioassays. According to their hazard quotient, the 3 sediments showed a contamination gradient; one of them should not have entered the detailed risk assessment phase, while another would have been further tested with H. azteca and C. riparius. In that case, this latter sediment would not have entered the detailed phase either, as it was not toxic to these species. However, the detailed risk assessment approach was applied to the three sediments, so as to test completely the relevance of the framework.

The three sediments were not significantly toxic to either C. riparius or H. azteca. However, some effects were observed in microcosms, including genotoxicity to molluscs. In this case, no risk characterisation could be made. Pore waters extracted from the three sediments were not toxic or slightly toxic to bacteria, algae, and C. dubia; an EC10 value could be determined only for B. calyciflorus. Therefore, due to exposure calculations, it seems there is a risk to pelagic species. However, as 3 bioassays out of 4 were negative or inconclusive, a refinement step would seem to be necessary. The highest concentrations of cadmium, copper, chromium, nickel and zinc were measured in the first lixiviates of the most contaminated sediment. Yet, even in that case, the total extracted fraction remained less than 10% of the total load. This fraction was below 1% for the other sediments, whatever the metal. Maximum concentrations and predicted concentrations at 1 year were compared to drinking water standards. This comparison showed a real risk of degrading groundwater quality for that most contaminated sediment, and a transient risk due to cadmium and nickel for the following one on the contamination gradient.

Considering these results, the design of the first step of the proposed assessment approach may be discussed, as one sediment which would not have been assessed in depth according to its hazard quotient did show a risk to pelagic species. This discrepancy underlines the fact that some sediment toxicity may exist below the lowest threshold. As it would be unrealistic to enter systematically into detailed risk assessments, the proposed thresholds in the decision diagram must be pragmatic compromises rather than absolutely safe boundaries. Moreover, protocol improvements are needed for sediment toxicity bioassays. Chronic endpoints are preferable, as they are more sensitive and more relevant. Another issue is related to the role of additional food: not adding food may increase the apparent toxicity, but the sediment organic content, which is an alternative food source, may also be a contaminant carrier. Furthermore, the second assessment endpoint (risk to pelagic species) should be reformulated, as it includes in fact two different questions. Short-term risks related to the deposition phase could be assessed with standardised bioassays like C. dubia survival and reproduction and algal growth, while longer term risks related to contaminant diffusion could be assessed with microcosms. Genotoxic effects were observed at rather high levels, as compared to published results. This result stresses the interest of introducing sensitive and early markers in the risk assessment process, although their real meaning for ecosystems is not yet fully elucidated. Finally, the leaching tests in columns are not completely satisfactory, as the column filling implies that one must first dry the sediments, which will alter their structure. Other application trials along with field validation studies should be carried out prior to the introduction of this scenario in operational or regulatory frameworks.

FR :

Cette étude présente les principaux résultats se rapportant d'une part à l'évolution des consommations en chlore par des composés organiques en solutions synthétiques de minéralisation variable. D'autre part, quelques essais sont consacrés aux possibilités de formation de sous- produits de la chloration tels les trihalométhanes.

Les différences de réactivité entre les composés simples (phénol, résorcinol, aniline) et les substances humiques, déjà observées en eau distillée, ont pu également être mises en évidence lors de leur dilution dans des milieux minéralisés (eaux de sources ou de forages). L'influence de la minéralisation, appréhendée par le paramètre "conductivité" ou par des teneurs variables en éléments minéraux spécifiques, s'est concrétisée par des écarts souvent considérables entre les potentiels de consommation en chlore en eau distillée et ceux en eaux minéralisées.

La présence d'ions bromures dans les eaux les plus minéralisées induit en particulier une formation plus importante des THM bromés. L'intervention d'ions chlorures et sulfates, en provoquant une diminution des potentiels de consommation en chlore, peut rendre plus complexe le schéma réactionnel du chlore avec les différentes entités réactives présentes dans une eau naturelle.

L'application du procédé de chloration à des eaux de surface algériennes a permis de distinguer entre les eaux très minéralisées du sud algérien et celles moins minéralisées du nord du pays.

Leur réactivité vis-à-vis du chlore s'est trouvée directement liée à leurs caractéristiques physico-chimiques.

Les différences de réactivité relevées peuvent être attribuées d'une part à la variabilité de la proportion en matériel humique et d'autre part aux teneurs relatives en éléments minéraux promoteurs ou inhibiteurs de la réaction de chloration.

EN :

The organic compounds studied during this work are simple aromatic structures (phenol, resorcinol, aniline), as well as macromolecular structures (humic substances). These compounds are representative of natural organic matter in surface waters. Various studies (De Laat et al., 1982; Norwood and Christman, 1987; Legube et al., 1990) showed that chlorination of these organic substances led to a strong chlorine consumption and the formation of volatile and non-volatile organohalogenated compounds. However, the majority of these experiments were conducted in aqueous media at very low ionic strength or in distilled water. The aim of the present study was to observe the reactivity of organic compounds in mineralised media. Indeed, Algerian waters are often very mineralised, excessively hard with high amounts of chloride and sulphate (Achour, 2001). In addition, humic substances can account for 60 to 90% of the total organic carbon (TOC) in most surface waters used to produce drinking water (Achour and Moussaoui, 1993).

The first part of this study involved the investigation of the impact of the overall parameter "total mineralisation", which can be directly related to conductivity measurements in the dilution waters. In addition to distilled water, dilution media for the organic compounds included four ground waters initially free of organic matter and residual chlorine. Figure 1 shows the geographic location of sampling sites for the waters of Drauh and Oued Biskra (Biskra) as well as those of waters of Ifri (Bedjaïa) and Youkous (Tebessa). Their main physico-chemical characteristics are presented in Table 1. The influence of certain specific inorganic constituents such as ammonia, bromide, chloride and sulphate is also observed on the chlorination of the organic compounds tested. The objective was to determine and to compare chlorine consumption potentials respectively for the simple organic compounds (D) and the humic substances (PCCl2) in media with differing degrees of mineralisation and under controlled conditions (pH, contact time and chlorine dose applied). Chlorination ratio is expressed for the simple organic compounds by r=moles of introduced chlorine/moles of organic compound and for the humic substances by m=mg of introduced chlorine/mg of humic substances.

Tests were performed to evaluate the formation of trihalomethanes (chloroform and brominated trihalomethanes). Analysis was done by gas chromatography using a head space method. The main experimental chlorination conditions for the synthetic solutions of organic compounds are summarised in Table 2. The results obtained (Table 3) show that the organic compounds tested remain very reactive towards chlorine, whatever the mineralisation of the dilution media. Potentials of chlorine consumption in distilled water and in mineralised waters are different (E), showing the obvious effect of the degree of mineralisation (Table 4). The reactivity of humic substances seems intermediary between the structures of resorcinol and phenol or aniline. The highest variations (E) between distilled water and the natural mineralised waters were observed in the case of the most highly mineralised water (water of Oued Biskra); generally the gaps decreased in the order E (Oued Biskra) > E (Ifri) > E (Youkous) > E (Drauh). In studies at different pH values, the oxidation capacity of chlorine was exerted more vigorously at circumneutral pH values (Table 5). For phenol and humic substances, gaps are more meaningful within five minutes of contact time, during the first step of the chlorination reaction (Table 6).

Total mineralisation does not seem to be the only parameter able to affect the reactivity of organic matter. The composition of the inorganic matrix must be considered too. Thus, according to the relative proportions of ammonia and bromides, competitive reactions (chlorine/organic matter and chlorine/inorganics) will condition chlorine consumption and the formation of trihalomethanes (THMs). For waters with a low ammonia/bromide ratio, an increase of the bromide concentration can induce a significant degradation of the chlorine consumption potentials, even for resorcinol, which is habitually more reactive with chlorine than with bromide (Figure 2). In waters with high levels of bromides (waters of Drauh and Oued Biskra), bromoform may be formed (Table 7). The presence of high concentrations of chloride and sulphate in aqueous solutions with humic substances seems to induce a reduction of chlorine consumption potentials (Figures 3 and 4). Several hypotheses can be advanced to explain this phenomenon, especially the decrease of activity of the chlorinating element by the increase of the ionic strength of the aqueous media. At the same time, it can be assumed that chlorinated entities may be formed with less reactivity than hypochlorous acid. The reduction of chlorine consumption is less significant in the presence of sulphates than chlorides. Moreover, the cation associated with sulphate seems to have an effect on the reactivity of the humic substances towards to chlorine.

In the second part of the work, chlorination was applied to four surface water samples. The sampling sites (Figure 1) relate to reservoir waters of Souk El Djemâa (Tizi-Ouzou), Aïn Zada (Sétif), Foum El Gherza (Biskra) and Fontaine des gazelles (Biskra). Amounts of humic material are relatively important and constitute about 60% of the TOC found in the raw waters. The mineralisation in these waters is considerable, particularly for waters in the southern part of the country (Table 8). Chlorine consumption and THMs formed seem to be correlated with the nature and concentration of organic matter, as well as with the inorganic component of these waters (Table 9). Chlorination of water samples taken from the north of Algeria (Aïn Zada and Souk El Djemâa) leads to higher values of chlorine consumption potentials than for water samples taken from the south of the country. Taking into account the results obtained on synthetic solutions, this trend can be explained by the appreciable contents of humic substances as well as by a lower degree of mineralisation of waters in the north. It also interesting to note that proportion of brominated trihalomethanes compared to the total trihalomethanes increases not only with the initial concentration of bromides but also with increasing levels of chlorides and sulphates. Highly mineralised waters, containing humic material, seem therefore to favour numerous competitive reactions during chlorination.

FR :

L'objectif principal du présent travail est la prévision en temps réel des charges de polluants dans un réseau d'assainissement urbain. La méthodologie préconisée dans cette étude se base sur deux outils. En premier lieu, le modèle de la courbe de tarage a été utilisé afin d'exprimer la corrélation entre les charges de polluants et les débits de ruissellement. Ce modèle a été sélectionné en raison de sa simplicité et de la disponibilité des paramètres nécessaires pour sa mise en œuvre. L'hypothèse de synchronisme systématique entre les pointes de l'hydrogramme et du pollutogramme dans ce modèle constitue une des faiblesses que nous proposons de surmonter dans le cadre du présent travail. Ainsi, le modèle de la courbe de tarage a été modifié par l'introduction d'un terme de déphasage qu'on identifie en temps réel. D'autre part, la constance des paramètres mis en jeu dans le modèle classique de la courbe de tarage constitue un autre obstacle pour la reproductibilité des phénomènes au cours du même événement et d'un événement à l'autre. Afin de surmonter cette deuxième faiblesse, le filtre de Kalman a été utilisé pour identifier les paramètres d'un modèle dynamique en fonction des erreurs de prévision constatées à chaque pas de temps. La méthodologie a été testée avec succès sur le secteur I de la ville de Verdun du Québec. Le modèle établi a été validé à l'aide de trois critères de performance, à savoir, le coefficient de Nash, le rapport des pointes mesurées/prévues et leur déphasage. Selon ces critères, les résultats trouvés par le modèle dynamique concordent bien avec les mesures.

EN :

It is normally unrealistic to send the total combined water volume generated during a rainfall event to a wastewater treatment plant and this approach is not retained as a viable solution when physical and economic constraints need to be accounted for. It becomes therefore pertinent to reduce the pollution from a given area by limiting water treatment to the most polluted portion of the runoff volume. For this purpose, various municipalities have expressed an urgent need for an automated system that could dynamically manage all the hydraulic components of their urban drainage basins. However, such a system of management in real time requires short-term forecasting of the water quality in the drainage basins. The main object of this work is the development of tools for the real-time forecasting of pollutant loads in an urban sewer network. The method used in this study is based on two tools: the rating curve model and the Kalman filter.

The rating curve model is used to explain the correlation between pollutant loads and runoff. This model was selected because of its simplicity and the availability of the parameters necessary for its implementation. The rating curve model has several important characteristics. First of all, the formulation of the model is independent of the accumulation phase and the load accumulated over the basin is assumed to be unlimited. A second characteristic consists in the normalized form in which runoff is present in the model as a flow rate, so that the rating curve model can integrate the quantitative and qualitative aspects of urban runoff in a simple formulation, which requires parameters available in real time.

The assumption of systematic overlap between the hydrograph and pollutograph peaks constitutes the main weakness of this model, which we propose to overcome within the framework of this work. Thus, the rating curve model was modified by the introduction of a lag term identified in real time. In order to define the time lag parameter in real time, a mobile window has been programmed to scan the two observation vectors of flow rates and loads. Theoretically speaking, the time lag corresponds to the maximum of the cross correlation function between flow rate and load vectors observed in real time. Three cases are therefore possible. In the first case, an increase of the pollutograph precedes that of the hydrograph and the time lag is positive. In this case and in a context of real-time management, loads are determined using a forecast model for flow rates. Measured flow rates are considered in this work as forecasted flow rates. If the hydrograph precedes the pollutograph, the time lag "d" is negative and the loads are related to the flow rate measured at an instant that precedes forecast time by "d" times the time step. When, finally, the two curves are perfectly synchronous, the "d" parameter is equal to zero and the flow rates are forecasted on the basis of the flow rates measured at the time of forecasting. The model is thus sufficiently flexible and adapted to the various foreseeable conditions.

In addition, the constancy of the parameters concerned in the classic rating curve model constitutes another weakness with respect to the reproducibility of the phenomena during the same event and from one event to another. In order to overcome this second weakness, the Kalman filter was used to identify the parameters of a dynamic model according to the forecast errors noted with each time step. Use of the Kalman filter also allowed us to eliminate the calibration procedure required by the static model. With this filter, the dynamic model continuously readjusts its parameters to satisfy the non-stationary behaviour of hydrological phenomena.

The methodology was tested successfully on the sector I of the town of Verdun (Quebec). The established model was validated using three performance criteria, namely, the Nash coefficient, the peak ratio and the lag between measured and forecasted values. According to these criteria, the results obtained with the dynamic model agree well with measurements.

FR :

Une régionalisation débit-durée-fréquence des débits de crue est réalisée sur les sous bassins du Cris qui draine une superficie d'environ 14 300 km² à l'ouest de la Roumanie. Cette régionalisation concerne 78 sous bassins dont les chroniques de débit quotidien et de pointe observées sont de trente ans en moyenne et pour lesquels nous disposons des pluies maximales de bassin de 1 jour à 10 jours calculées à partir de 92 postes pluviométriques. La régionalisation est menée selon deux approches : une approche régionale classique et une approche à partir de trois modèles adimensionnels de référence établis sur trois sites observés de France. La différence fondamentale entre les deux approches réside en ce que l'une prend en compte l'information spatiale pluie-débit inventoriée du Cris et que l'autre considère essentiellement l'information pluie-débit de chaque site de référence français. L'approche modèles de référence a pour base conceptuelle une typologie des crues qui pour un site cible est prédéfinie par un critère de choix, tandis que l'approche classique nécessite que soient définis des régions hydrologiques homogènes. Cette démarche est menée sur les trois sous bassins hydrographiques du Cris et permet d'étendre la région hydrologique homogène à l'ensemble du bassin du Cris. L'approche régionale comme l'approche modèles de référence privilégie la loi exponentielle adaptée aux valeurs supérieures à un seuil pour ce qui concerne les quantiles de crue de faible période de retour et pour des durées de 1j à 10j selon la dynamique de crue des sous bassins. Pour les quantiles de crue de grande période de retour les deux approches sous tendent le modèle du GRADEX, forme d'extrapolation des distributions observées par le gradex des pluies maximales. Quelle que soit l'approche de régionalisation, en tout site cible doivent être disponibles deux descripteurs de régime : le débit de pointe décennal Q10 et une durée caractéristique de crue D. Afin de comparer essentiellement l'incertitude des modélisations sur les quantiles de crue, D et Q10 sont connus et déduits des observations. Les résultats présentés montrent une bonne validité du modèle régional ajusté sur l'ensemble du Cris. Ceci indique que la zone étudiée est relativement bien homogène. Concernant les modèles de référence, leur critère de choix n'apparaît pas pertinent lorsqu'on s'intéresse aux faibles périodes de retour, mais se révèle significatif pour les fortes périodes de retour. Ce résultat est en grande partie dû à la méthode d'extrapolation appliquée. Celle ci est liée à la méthode du GRADEX et utilise l'information locale sur les gradex de pluie, comme cela est souvent le cas en France. Il est à noter que ces modèles de référence établis sur des chroniques de débit et pluie d'avant 1992 n'ont pas été réactualisés. L'exemple du bassin du Cris montre qu'ils n'en gardent pas moins un caractère opérationnel pour l'estimation des quantiles de crue de durée d (0 < d(j) < 10) et de période moyenne de retour T (5 < T(an) < 1000).

EN :

When local information on streamflows is insufficient for estimating flood quantiles, a Regional Flood Frequency Analysis (RFFA) is usually carried out. Once homogeneous hydrological regions are defined, methods such as the "Index-Flood" method can be applied. In most cases, the variable understudy is the maximum peak flood (instantaneous or mean daily value, depending available data). However, the severity of a flood is not only defined by its peak, but also by its volume and duration.

For this reason, Cemagref has developed for several years an approach which, in addition to the classic flood/frequency relationship, takes into account the notion of duration: the flood-duration-frequency approach (QdF). In a similar manner to the rainfall intensity-duration-frequency analysis, averaged discharges are computed over different fixed durations d. For each duration, a frequency distribution of maximum averaged discharges is studied. Finally, a continuous formulation is fitted, as a function of the return period (T) and the duration (d) over which discharges have been averaged. Rare quantiles are determined using rainfall frequency information, according the so-called "GRADEX" method ("aesthetic" version). The regionalization of the QdF approach leaded to define three sets of dimensionless QdF curves. The originality of the method is that each set was fitted to one unique basin, located in France, and corresponding to a distinct hydrological regime. A choice criterion, involving maximal rainfall distributions, determines which reference basin has to be considered. In this way, the QdF regionalization differs from the classic concepts mentioned in the first paragraph. The classic approach takes into account the whole streamflow-rainfall information available on a homogeneous hydrological region, while the QdF regionalization developed at Cemagref considers the streamflow-rainfall information available on the three French reference basins. The aim of this research is to apply classical regional concepts (homogeneous regions) to the QdF approach and to compare results with thus obtained using the three French reference basins. One interest of the study is that it has been conducted in Romania, ie, outside the area where the three reference basins are located.

The case study is carried out on the Cris river sub-catchments which cover an area of about 14 300 km², in the west of Romania. This regionalization concerns 78 sub-catchments having about thirty years of streamflow measurements (daily flow and instantaneous flood peaks). For validation purpose, two basin sets are constituted: a calibration set (54 basins) and a validation set (24 basins). Furthermore, maximal rainfalls over 1 day to 10 days are available for these basins, from 92 rain-gauge stations.

First, regional QdF curves are deduced according the definition of homogeneous regions. Three regions are defined, corresponding to the three main sub-catchments of the Cris basins. After different tests, it is shown that the whole Cris basin can be considered as a unique homogeneous region. Then, the regionalization involving the three French reference basins is carried out. In both cases, methods are applied with the exponential law adjusted on peak over threshold values, for small return periods. Durations are ranging from 1 day to 10 days, according to the flood dynamic of the studied basins. For long return periods, both approaches use the GRADEX method, which extrapolates discharge distributions according to the rainfall distributions. Whatever the regional approach used, two descriptors have to be estimated for each target site, in order to unscale the dimensionless regional QdF curves. Theses two descriptors are the 10-year-return-period peak flood Q10 and a flood characteristic duration D. In order to compare the uncertainty of the two approaches, D and Q10 are local values, deduced from observations.

The results presented shows a good validity of the regional model fitted to the whole Cris basin. This indicates that this region is quite homogeneous. Concerning the reference models, their choice criterion does not appear to be pertinent for small return periods, but becomes relevant for high return periods. In this case, estimations are comparable to thus provided using the whole regional information available on the Cris basin. This result is mainly due to the extrapolation method used. It is related to the GRADEX method and takes into account the local information about rainfall gradex, as it is often done in France. One point to be noticed is that these reference models have not been updated since they have been established in 1992. Despite that, the Cris basin example shows their operational ability for estimating flood quantiles of duration d (0 < d(day) < 10) and return period T (5 < T(year) < 1000). From an operational point of view, the three reference basins can be a valuable option. Indeed, a regional analysis is sometime difficult to carry out: data not available or too costly, time available for the study to short, etc. On the other hand, the three reference basins method is easier to apply and requires less data. However, uncertainties and hypothesis of the method should be kept in mind. In particular, we should be aware that high return period quantiles are given according the GRADEX method, ie using the frequency rainfall information. Consequently, the hypothesis of this method should be respected.