Volume 13, numéro 4, 2000

FR :

L'une des difficultés rencontrées couramment dans la conception des réseaux de mesure - au moins en ce qui concerne les micropolluants - porte sur la sélection des paramètres à mesurer. C'est notamment le cas pour les pesticides, dont plusieurs centaines sont utilisées en agriculture, mais qu'il est impossible de surveiller dans les eaux en totalité pour des raisons à la fois techniques et économiques. C'est la raison pour laquelle les autorités françaises ont fait procéder à la mise au point d'une méthode de sélection des matières actives utilisées en agriculture basée sur l'évaluation du risque. Dans cette méthode, l'exposition est figurée par un rang combinant les données relatives aux usages des matières actives (superficie, dose par ha) et leurs caractéristiques physico-chimiques. Le danger est représenté par la toxicité, soit pour l'homme, soit pour les espèces aquatiques. Cette approche a été appliquée à l'échelle nationale et dans un certain nombre de régions françaises, dont l'Alsace et la Lorraine. Les résultats des mesures de pesticides réalisées ensuite pendant un an ont été confrontés aux indices d'exposition obtenus. Les substances détectées le plus fréquemment correspondent effectivement à celles dotées des rangs d'exposition les plus élevés (ajustement exponentiel, r²≈0.82) ; cependant, le diuron apparaît à une fréquence plus élevée que celle attendue, en raison de ses usages non agricoles. La corrélation est moins bonne pour les substances dont les rangs d'exposition sont proches de la valeur considérée comme significative pour les eaux superficielles, ce qui peut provenir soit de l'utilisation de données erronées lors de la sélection, soit d'un poids insuffisant attribué à certains facteurs dans la méthode de sélection, soit enfin d'aléas météorologiques.

EN :

Monitoring of micropollutants is a rather recent activity (10-15 years), at least in surface waters; because of the need for sophisticated analytical methods and of the potential number of analytes, this type of activity is confronted with important economic constraints, which require that one make a selection among the range of substances to monitor. Among organic micropollutants, pesticides constitute a well-identified category, since they are used mainly in agriculture; this use on broad surfaces may have important impacts on the quality of surface water. Various methods have been used to select those pesticides likely to have the greatest impacts on water quality; some of these methods might be considered to be "hazard assessment", whereas others correspond to simplified "risk assessment" methods (this appears particularly true for pesticides, of which several hundreds are used in agriculture). Recently, a French panel of experts mandated by different Ministries designed a selection method called SIRIS, which allows one to define three different lists of pesticides according to the media to be monitored (surface or ground-water) and to the monitoring objectives (ecosystem protection, drinking water production). This paper deals with the application of the SIRIS method at a regional level, in the context of a permanent survey of river quality.

As a simplified risk assessment method, SIRIS combines data on hazard and exposure; hazard is estimated by a single parameter, either toxicity for aquatic species or acceptable daily intake (ADI). Exposure represents the probability that a transfer to water bodies may occur; for surface water, this probability is influenced by the crop acreage, the applied dose (kg/ha), the solubility, the pesticide half-life, the hydrolysis and the distribution coefficient between water and organic matter (Koc). These factors are considered in this hierarchical order, and for each substance a score is assigned to each of these factors among three possible values ("o"=slight, "m"=medium, "d"=high, according to the relative influence on transfer); finally exposure is estimated by a relative rank obtained by a combination of these values following a "penalisation" principle. Two tables are available for applying this approach at a regional level: the first contains the values (o,m,d) assigned to more than 300 substances by the expert panel for solubility, half-life, etc., and should be completed with crop acreage and dose. The second table provides the ranks corresponding to the different combinations of o,m,d values. A final rank of 35 was considered by consensus to be a pragmatic threshold for the transfer to surface water. This method was applied in 1996 in two regions in France (Alsace and Lorraine) separately; most of the selected chemicals (but unfortunately not all, due to technical constraints) were then analysed monthly in surface waters (24 sampling points, yielding 144 samples in Alsace and 169 in Lorraine). Occurrences fell between 0% and 60% in Alsace, and between 0% and 90% in Lorraine; in both regions, the most frequently detected chemicals were atrazine and diuron.

The relevance of the selection method may be discussed under several aspects: the choice of the factors, their order, the position of thresholds corresponding to o,m,d values, the value of the overall threshold, and the availability of the data. Some pesticides are not ranked only because no data were available concerning their solubility, hydrolysis rate or Koc, but the relative importance of such gaps cannot be appreciated with the current set of data. Other items may be assessed through the comparison of the exposure rank versus the occurrence. This relationship takes an exponential shape, with some anomalies: for example, the occurrence of diuron in Alsace is higher than expected, based on its exposure rank. This situation can be explained by the fact that there are non-agricultural uses of this substance, such that the exposure rank appears to be underestimated. For other substances, like aldicarb and chlorpyrifos-ethyl, discrepancies are observed between the exposure rank and occurrences, when comparing with substances with higher exposure ranks. This anomaly may be due to poor data quality. For carbendazime, the occurrence in Lorraine appears underestimated, probably because of a dry period deficit after the application. Finally, chlortoluron received the same rank in the 2 regions, but is more frequently detected in Lorraine; crop acreage may have been overestimated in Alsace. However, the dataset is still limited to one year of sampling; some discrepancies may appear less important when more data are available. For chemicals with ranks > 50, there is a good exponential fit between ranks and occurrences (y=0.0235*e^0.0739x ; r²=0.82). This observation means that pesticides with ranks >50 are systematically encountered in surface waters; however, the current threshold (35) should be maintained, because some substances with ranks <50 are also detected. Thus, the SIRIS method appears to be a good tool for selecting agricultural pesticides for monitoring purposes at a regional level.

FR :

Dans la zone semi-aride de Tunisie centrale, un grand barrage a été construit en 1982 sur l'oued Zeroud, dont l'infiltration des crues naturelles constituait jadis la principale source d'alimentation des nappes de la plaine de Kairouan. Ce barrage est destiné à l'écrêtement des crues, l'irrigation et à la recharge des nappes à l'aval de la retenue.

Entre 1988 et 1996, plus de 70 millions de m3 d'eau ont été mobilisés sous forme d'ondes de lâchers à partir du barrage pour la recharge artificielle par infiltration dans le lit de l'oued. Le suivi des ondes de lâchers est effectué sur 3 stations de contrôle disposées sur une distance de 40 km le long du lit. L'analyse de l'évolution des débits a révélé que la capacité d'infiltration du lit augmente progressivement avant de se stabiliser, mettant en évidence l'effet de l'air sur le processus de recharge.

Un modèle conceptuel à réservoirs conjuguant fonctions de production et de transfert avec discrétisation spatiale a permis de modéliser la propagation et l'infiltration des ondes de lâchers le long de l'oued. Son ajustement sur un échantillon de six événements de lâchers a fourni des résultats satisfaisants. Même si la validation reste insuffisante en raison de la rareté des données, ce modèle peut constituer un premier outil d'évaluation de l'efficacité de la recharge et de prédiction de son impact sur la nappe souterraine de la plaine de Kairouan.

EN :

In semi-arid central Tunisia, flood infiltration through the streambed of the ephemeral Zeroud Wadi has long been the principal mechanism of recharging aquifers beneath the plains of the city of Kairouan. In addition to controlling infiltration, a dam built on the Zeroud in 1982 to supply water to Kairouan also protects the city from floods.

From 1988 to 1996 over 70 million cubic meters of dam-held water were mobilized in the form of flood waves over the stream channel to recharge the Kairouan aquifers. Analysis of flow discharges measured at three control stations spanning 40 km of the stream reveals that infiltration is reduced during initial flood-wave spreading and increases gradually before stabilizing. The differential rate of infiltration is thought to be caused by interstitial air initially present.

A conceptual reservoir model based on production and transfer functions with spatial discretization was used to simulate flood-wave propagation and infiltration along the Zeroud. After calibration with six flood events the model produced satisfactory results. Though current data scarcity prevents its validation, the model constitutes a prototype for the evaluation of recharge efficiency and for the prediction of recharge impact on Kairouan aquifers.

FR :

L'exclusion engendrée par la présence de membranes de nanofiltration est due à la superposition de plusieurs mécanismes. Aussi, il est important de pouvoir distinguer ces différents modes de transports afin de les comprendre et de proposer des modèles adéquats. Cette étude s'attachera uniquement à la compréhension du transport de solutés neutres à travers des membranes de ce type. Trois sucres, le glucose, le saccharose et le raffinose, ont été utilisés pour caractériser deux membranes organiques fournies par la société Osmonics. Ce travail montre qu'un modèle simple, basé sur la diffusion, tenant compte des conditions hydrodynamiques du module, est en accord avec les résultats obtenus au laboratoire mais également trouvés dans la littérature. Une bonne adéquation entre le modèle et l'expérience est ainsi obtenue, à la fois sur des modules plans et tubulaires, pour des écoulements laminaires et turbulents.

EN :

Although nanofiltration appeared at the end of the 1970s under various names, it was only really recognized as a useful separation process in the 1980s. Nanofiltration membranes are porous media with a mean pore diameter around 1 nanometer. These membranes do not obey the traditional solution-diffusion model given for reverse osmosis or the convection-diffusion model used to describe ultrafiltration. Although the technique has benefited from a fast technological development, the transport mechanisms are still misunderstood and for a particular separation the choice of a nanofiltration membrane remains empirical.

The main objective of this work was to understand and to model the transport of neutral solutes through nanofiltration membranes. Neutral solutes were chosen to emphasise geometrical exclusions, to avoid any electrical interactions and to identify the preponderant transport mechanisms through these materials. The experiments were carried out with a laboratory filtration apparatus. The membranes were laid out in a parallel plane osmotic cell, which makes tangential filtration possible. The geometry of the filtration cell involved the choice of two organic membranes supplied as flat sheets: a BQ01 and a MX07 membrane. The filtration area was 86 cm². The pressure varied from 7 to 30 bars. The temperature was maintained at 20°C whereas tangential velocity in the cell was fixed at 0.45 m×s^-1 (the Reynolds number was 3350). As the solutions used were slightly concentrated, the pH remained close to neutral pH. Three sugars were chosen as solutes: glucose, saccharose and raffinose. These molecules have two advantages: they are electrically neutral and they have molecular weights close to the membranes' MWCO, as provided by the manufacturer.

First, saccharose was studied on the two membranes with two different concentrations. These experiments showed that the separation of neutral solutes by nanofiltration membranes is due only to a sieving effect. In subsequent experiments a single concentration was used to characterize the retentions of both glucose and raffinose. The results of the filtrations carried out on the three sugars validated the molecular weight cut-off specified by the manufacturer: the MWCO of the BQ01 membrane was estimated to be 1000 Da, and that for the MX07 membrane was 200 estimated as 200 Da.

Schematically, the solute transport can be divided into three stages: in the feed, at the feed/membrane interface, and within the membrane material. In the feed, one notes an increase in solute concentration if one approaches the membrane from upstream. This phenomenon, which is general to any selective transport, is called concentration polarization and is described by film theory. This theory stipulates the creation of an antagonistic diffusive flow, from the membrane towards the feed, seeking to restore the concentration balance within the feed solution. The modification of the concentration at the feed/membrane interface leads to the definition of two retention coefficients: a measured value, the observed retention (R_obs), and a calculated value, the intrinsic retention (R_m). Steric exclusion based on the size difference between the pore and the solute is set up at the interface. Uncharged solutes can be visualised as rigid spheres and the membranes can be regarded as a bundle of cylindrical, parallel, rigid and right capillaries. Since the elements are rigid and the solutes are subjected to the same geometrical constraints at the entry and at the exit of a pore, the partition coefficients are identical at those two ends. Finally, lying between reverse osmosis and ultrafiltration, transport through nanofiltration membranes is often expressed as the sum of convective and diffusive phenomena. However, the experimental results show that the observed retentions are stable or increase when pressure increases. These observations also highlight the fact that the values of infinite retention are always compatible with values close to 1. These observations corroborate the idea that diffusion is the predominant transport mechanism of neutral species through the studied materials (BQ01, MX07). The transport equation of neutral solutes can then be simplified to its diffusive component. The expression of the intrinsic retention is obtained by using Fick's law, the definition of the retention coefficients and the definition of the partition coefficients:

R_m= 1-(1 / 1+J_vα)

The geometrical and physicochemical characteristics of the solutes and of the membranes merge into the α parameter.

The results found with the theoretical relation were confronted with experimental data derived from film theory (in order to take into account concentration polarization). The simple one-parameter model was successful in correlating the results obtained in this work. The model was also tested with data coming from Combe et al. (1997), who studied filtrations of glucose, saccharose and raffinose in a laminar flow system by ceramic nanofiltration membranes laid out in the shape of tubular module. The results obtained show that the simple model also successfully correlates with the performances of these membranes.

With the data obtained in our laboratory as well with the data found in literature, this study shows that a simple one-parameter model, based on the diffusional transport of the solutes within the membrane material, predicts the rejection of neutral solutes by nanofiltration membranes. The simple one-parameter model is able to simulate any filtration carried out by these membranes for different circulation conditions, for diverse geometrical shapes and for various materials.

FR :

La méthodologie débit-durée-fréquence (QdF), appliquée ces dernières années aux étiages, a permis de définir quatre modèles types recouvrant l'ensemble des rivières étudiées. L'identification de la typologie du site étudié et l'estimation de deux descripteurs hydrologiques locaux suffisent au modèle, dit de référence, pour en déduire les courbes QdF (1j ≤ d ≤ 30j) en débit moyen minimum sur l'année (VCNd) ou débit seuil minimum annuel (QCNd) non dépassé sur ces mêmes durées. S'il est relativement aisé de définir les descripteurs hydrologiques, il est plus difficile d'identifier, sans observation de débit, le modèle à prendre en compte. En reconsidérant avec plus de rationalité la démarche d'identification des typologies, et en particulier les distributions multidurées relatives à chaque bassin, il est apparu possible d'évoluer vers un modèle unique pouvant être calé en chaque site observé. Ce nouveau concept de modélisation repose sur la propriété d'affinité des distributions, relatives aux échantillons de valeurs de durées d. Par souci de continuité avec l'approche QdF à référence typologique, la loi statistique log-normale à deux paramètres a été choisie. Le modèle, dont la conceptualisation est indépendante de la loi statistique choisie, aura dans le cas de la loi log-normale trois paramètres à ajuster sur les VCNd échantillonnés. Trente six sous bassins du bassin hydrographique de la Moselle ont été étudiés. Le modèle développé pour les débit moyens VCNd peut être appliqué aux débits seuils QCNd en conservant le même jeu de paramètres, grâce à une relation observée entre débits moyens et débits seuils. Cette nouvelle modélisation rationalise l'approche antérieure basée sur la typologie d'écoulement de basses eaux des bassins versants.

EN :

The flow-duration-frequency (QdF) concept, as applied in recent years to low flows, has made it possible to establish four reference models (GALEA et al., 1999a), corresponding to four typologies. The hydrological variables concerned are the minimum mean discharge of the year defined for various continuous durations d (1day ≤ d ≤ 30day), called VCNd, and the annual minimum threshold discharge not exceeded over these same durations, called QCNd, according to OBERLIN (1992). These QdF models allow a description of the temporal variability of low flows observed for a river basin, from a statistical point of view. The typology of the basin and two local hydrological descriptors have to be known. For ungauged basins, these two descriptors (GALEA et al., 1999b) are well estimated by various methods, such as multivariate analysis relating to the physiographic characteristics of the basin. Nevertheless, the choice of the reference model still remains contentious.

By reconsidering in a more rational manner the step of identification of typologies, and in particular the discharge distributions (for durations d) relating to each basin, it appeared interesting to establish a local model. This new model has a simpler formulation, thanks to a scale invariance assumption. This research (CHAPUT, 1999) was undertaken on 36 sub-basins of the Mosel basin. In order to ensure continuity with the earlier QdF models described above, the two-parameter log-normal law was chosen and adjusted on the distribution of mean discharges. The scale invariance assumption is deduced from the observed parallelism of distributions related to different durations, when discharges are represented in a logarithm scale. This observation means that all of the distributions can be translated to a common point, in order to obtain one "consolidated" distribution, independent on the considered duration. This parallelism has been observed on many basins, and seems to be a realistic assumption. Furthermore, these observations have been made on samples, and do not depend on the choice of statistical law. The methodology described in this paper makes it possible to adjust the local QdF model on sampled discharges. Only three parameters have to be determined: s^c, the "consolidated" standard deviation, ∆e the low flow characteristic duration and VCN(2,1), which represents the quantile of the one-day distribution, with the two-year return period (F=0.5).

This model is also useful for the determination of threshold discharges (QCNd). An observed property gives a relation between the VCN and QCN quantiles, for a fixed return period, considering different durations d: VCN quantiles can been deduced from QCN quantiles by integrating them, according to d. Consequently, the analytical formulation of the VCN model can be derived according to d, in order to obtain a QCN model. This model has the same three parameters s^c, ∆e and VCN(2,1) described above. The comparison between QCN quantiles adjusted on samples and QCN quantiles deduced from the VCN model by derivation shows good results.

As a conclusion, this new modelling approach unifies the typological approach for both mean discharges and threshold discharges. It is based on a local adjustment and avoids having to choose between one of the four former reference models. This local model opens up perspectives for a regional model, as it has been done for floods, for example by the Group of Research in Statistical Hydrology (1996). This will make it possible to estimate the low flow regime on an ungauged basin.

FR :

Pour étudier les distributions de fréquences des variables hydrologiques (pluies et débits) au pas de temps horaire, une méthodologie associant un générateur de chroniques de pluies horaires et un modèle conceptuel global de transformation de la pluie en débit a été développée. Sur une période de simulation donnée, la méthode génère une collection de scénarios de crues vraisemblables utilisée en prédétermination des risques hydrologiques. Les distributions de fréquences des variables hydrologiques sont construites empiriquement à partir des événements de pluies et de crues générés. L'extrapolation des distributions de fréquences des variables hydrologiques vers les fréquences rares se fait de façon empirique en augmentant la période de simulation, et non plus sur l'ajustement direct des distributions observées. Le principe de cette méthode (appelée SHYPRE : Simulation d'HYdrogrammes pour la PREdétermination) est donc d'utiliser les observations pour décrire le phénomène, afin de le reproduire statistiquement et de s'affranchir ainsi du manque d'observation. Son utilisation permet une estimation originale des quantiles de crues de fréquences courantes à rares et présente l'intérêt d'obtenir une information temporelle complète sur ces crues. De plus, on montre que l'approche fournit une estimation de quantiles de crues bien plus robuste que les ajustements statistiques des distributions observées, même pour les événements de fréquences courantes. Cette robustesse provient d'une meilleure prise en compte de l'information pluviométrique et de la stabilité de la paramétrisation du modèle pluie-débit.

EN :

A statistical approach encompassing a stochastic model to generate hourly rainfall and rainfall runoff was used to study frequency distributions of hydrologic variables. The method generates numerous different flood events over a given simulation period to evaluate hydrologic risks. Entitled Simulated HYdrographs for flood PRobability Estimation (SHYPRE), it makes use of observed values to describe hydrological phenomena and successfully reproduces observed-value statistics. Frequency distributions of hydrologic variables are built empirically from model-generated rainfall and flood events. Extrapolation of these frequency distributions to rare frequencies is performed by simulation over longer periods, rather than by direct fit of theoretical probability distributions over observed values. This approach yields different estimations of flood quantiles for common to rare frequencies as well as complete temporal flood data. Moreover, SHYPRE estimates of flood quantiles are more stable than statistical distributions fitted onto observed values, even for frequent events. The improvement stems from better use of rainfall data and from the parametric stability of the rainfall model and rainfall-runoff model.

FR :

Sur le fleuve Niger, les relations entre concentrations en Matières En Suspension (MES) et débits liquides montrent, à l'échelle d'une crue annuelle, des cycles d'hystérésis orthogrades. Le modèle présenté dans cet article reproduit les variations saisonnières de ces MES à partir du seul débit liquide. Il suppose que les MES proviennent de deux sources distinctes : le système " versants + réseau hydrographique secondaire ", siège d'une érosion saisonnière temporaire et le réseau hydrographique principal, siège d'une érosion permanente. Le modèle représente schématiquement la production de MES provenant de ces deux sources par le biais de deux réservoirs de MES. Le premier contient un stock en MES temporaire et limité. Ce stock, maximum au début de la crue annuelle (stock initial), est mobilisé et entraîné au cours de la saison pluvieuse en produisant un flux journalier, supposé être, à un instant donné, proportionnel au stock restant et à une fonction de puissance du débit. Le second, contient un stock de MES illimité et disponible en permanence. La mobilisation de ce stock produit un flux journalier, supposé être aussi une fonction de puissance du débit, et dont l'importance sera limité par la capacité du cours d'eau. Les cinq paramètres du modèle sont calibrés à l'aide des données acquises durant huit années hydrologiques (1991/92 à 1998/99) sur deux stations du Niger amont (Banankoro et Douna). Malgré les limites d'utilisation actuelles liées à la détermination du stock initial, le modèle présenté reconstitue de façon satisfaisante les variations annuelles des concentrations en MES et offre des perspectives intéressantes pour modéliser l'évolution temporelle des MES observées tant pour les fleuves tropicaux unimodaux que pour les petits bassins versants africains. En termes de flux annuels, le modèle n'apporte pas d'amélioration sensible par rapport à un ajustement statistique simple entre les volumes écoulés et les flux de MES. Cependant, il permet aussi de déterminer les variations de flux au cours de l'année, information qui ne peut être obtenue avec un modèle de régression statistique.

EN :

Estimating temporal variability of suspended sediment concentrations in a watershed is important for a number of reasons (e.g., sediment yield estimation, provision of input data for reservoir sediment-deposition models and water quality models). Three different approaches have been adopted for modelling erosion and sediment transport: physical erosion models (Wicks and Bathurst, 1996); conceptual models (Negev, 1967; Pinheiro and Caussade, 1996); and empirical models (Walling, 1977; Asselman, 1977). Physical and conceptual models usually require rainfall intensity data. In African tropical river catchments, however, temporal and spatial variability of rainfall are not well known; in case of the Niger, water discharge is the only reliable hydrological parameter. This study proposes a model of temporal changes in suspended sediment concentrations using only water discharge data, thereby eliminating the need for rainfall parameters.

Daily discharge and weekly suspended sediment concentration data (from 1991/92 to 1998/99) gathered at two monitoring stations of the Upper Niger (Banankoro and Douna) (Figure 1) were used to study relationships between suspended sediment concentration and river discharge.

At each gauging station on the Niger, the relationship between water discharge and suspended sediment concentration during the annual flood is characterized by clockwise hysteresis (Figure 3). Moreover, several other African single-annual-flood rivers-unimodal rivers-also exhibit this type of relationship (Kattan et al., 1987; Olivry et al., 1988; Orange, 1992). This cyclicity suggests a three-stage description of sediment transport dynamics: (1) At the beginning of the rainfall season, sediments are imported by hill-slope surface runoff, re-entrainment of deposits in the channel network, and riverbed erosion. The first two sources consist of easily mobilizable material available throughout the catchment at the beginning of the hydrological year. (2) Sediment availability decreases with time as the soil becomes stabilised by vegetation during the rainy season. Erosion is consequently reduced, despite increased discharge. (3) During the period represented by the falling limb of the flood hydrograph, mobilizable material has been depleted or cannot be entrained; what suspended sediment there is originates upstream and from bank and bed erosion-permanently available sources. Non-seasonal sediment sources are grouped under the label of "continuous erosion."

We propose a lumped conceptual model of suspended-sediment concentration variations over the hydrological year. The model divides the erosion, transport and deposition processes into those acting on hill-slopes and those acting in the channel network, and assumes that both are explainable by water discharge Q(t) alone. The hill-slope/channel distinction is based on the fact that suspended sediment transport in a river depends not only on transport, bank and bed-erosion capacity, but also on the amount of available material in the drainage catchment.

Sediment transport (in tons per day) for the hydrological year is thus computed as the sum of two independent daily contributions of sediment discharge. The first, F _mob (t), originates from a limited reservoir which is full at the beginning of the flood and available only temporarily, during the rainy season. At a given time t, the sediment input F _mob (t) is proportional to the amount remaining in the limited reservoir and to a power function of water discharge (Eq. (1), Eq. (2)). The second reservoir, temporally and quantitatively unlimited, injects a daily sediment discharge F _ec (t) which is a power function of water discharge (Eq. (3)); F _ec (t) is limited only by river capacity. The final concentration is obtained from Equation (4).

The five model parameters were calibrated with concentration and discharge data from hydrological years 1991/92 to 1995/96. The unlimited reservoir parameters were estimated using Equation (6) with data taken during the decreasing stage. Initial sediment in the limited reservoir was estimated with Equation (7), using observed concentrations and concentrations derived from continuous erosion (Eq. (3)). Two parameters related to the decrease in initial sediment loads were obtained by optimization of the mean Nash criterion between observed and calculated concentrations (Figure 6). Some physical interpretations were ascribed to coefficients related to the limited reservoir.

Despite the limitations of assuming a single initial mobilizable reservoir, predictions of temporal sediment-concentration patterns during the annual flood were satisfactory (Figure 7, Table 2). The model also simulated some observed sediment concentration peaks associated with sudden water discharge variations during the rising limb of the annual flood. Best results were obtained by varying initial sediment reservoir estimates for different hydrological years (Table 3)-the model could therefore be improved by highlighting parameters that determine sediment loads at the beginning of the hydrological year.

The model does not give better estimates of annual sediment yield than simple regression of annual water volume (Table 4). However, it is able to reproduce the temporal variability of the sediment flux during the annual flood. The small size of the data set makes evaluation of the performance of this model difficult; for better assessment, it should applied to other data gathered on the same catchments or on other large tropical rivers. Model parameter values could also be explained by drainage basin characteristics.

FR :

L'objectif du présent travail est l'élaboration d'une méthodologie de validation des données hydrométriques mesurées dans un réseau d'assainissement. L'information validée est utilisée aussi bien en temps réel, pour optimiser les consignes de gestion, qu'en temps différé, pour poser le véritable diagnostic et évaluer, sur une base quotidienne, l'efficacité des systèmes d'assainissement.

Le principe de base de la méthodologie proposée repose sur la redondance analytique de l'information provenant d'une part de la mesure directe du débit sur le terrain et d'autre part du débit simulé à partir des variables météorologiques. On compare ainsi d'une part, l'écart entre la valeur prévue par un modèle autorégressif (AR) et la valeur mesurée et d'autre part, l'écart entre la valeur prévue par ce même modèle AR et la valeur simulée par un modèle hydrologique. Parmi les valeurs, mesurée et simulée, celle qui se rapproche le plus de la valeur prévue est retenue. Afin de considérer des modèles non stationnaires et d'éviter le biais d'estimation des paramètres de régression par la méthode standard des moindres carrés, le filtre de Kalman est utilisé pour identifier les paramètres du modèle AR.

La méthodologie proposée a été testée avec succès sur un bassin urbain de la municipalité de Verdun. L'hydrogramme mesuré a été bruité artificiellement à la fois par un bruit blanc et par un certain nombre de perturbations de grandes amplitudes et de différentes formes. Le processus de validation a permis de retrouver pratiquement les mesures initiales, non bruitées. Les critères de performance introduits sont largement concluants.

EN :

We developed an automated methodology for real-time validation of hydrometric data in a sewer network. Our methodology uses real-time validated data to optimise system management and non-real-time data to evaluate day-to-day performance.

Two approaches can be used to validate and correct hydrometric data; the choice depends on the number of level gauges present in a system. In single gauge systems, univariate filtering is used to smooth data. For example, frequency filtering systematically eliminates values corresponding to frequencies higher than a predetermined threshold frequency. In systems with several gauging stations-duplex, triplex, or multiplex systems-the multivariate filtering method proposed here can be used to validate data series from each gauge. Material redundancy in duplex or higher order systems makes it possible to detect a deficient gauge, using a decision rule to set aside erroneous readings before averaging accepted values. Part of the underlying principle of this methodology is heavier reliance on gauges that give readings consistent with previous and subsequent validated values in a given series. Thus isolated positive or negative variations within a series are eliminated if corresponding variation values at other gauges are more consistent. To evaluate persistence, a reading is compared to a value predicted by an autoregressive (AR) model calibrated by the previous validated reading.

This filtering technique constitutes an intelligent alternative to the frequency filtering method mentioned above. In more practical terms, it compares the deviation of an AR model prediction from a measured value with the deviation of the same AR model prediction from a value estimated by a regressive model at other stations in the network. Among the values measured and estimated by the regressive model, the one nearest the AR model prediction is retained.

Our methodology also relies on analytical redundancy generated by direct measurement of flow and hydrological simulation. More precisely, the deviation of the AR model prediction from the measured value is compared with the deviation of the same AR model prediction from a value obtained from a hydrological simulation model. Among measured and simulated values, the one nearest the AR model prediction is retained. To allow consideration of nonstationary models and to avoid the well-known bias of the least squares method, the Kalman filter is used to identify the parameters of the AR model.

The methodology we propose employs three models. The first generates analytical redundancy using hydrological modelling. An autoregressive model is then used to predict future runoff rate values. Finally, a voting process model is used to compare measured and simulated values.

The proposed methodology was tested on the Verdun sewer system in Quebec with successful results. Two types of artificial disturbance of the measured hydrograph were created: white noise was added to measured values and disturbances of large amplitude and various forms were introduced. The methodology produced the initial values and performance criteria were conclusive. Thus on-site testing confirms that this approach allows completely automated detection and correction of most anomalies. Flood peaks were neither underestimated nor overestimated, and total runoff volumes were retained.