Volume 11, Number 1, 1998

FR:

Le concept de risque, trop souvent imprécis, est analysé et un essai de définition rationnelle est tenté pour tenir compte à la fois des aspects de probabilité et de conséquences dommageables. On montre que toute analyse de risque en sciences de l'eau doit résulter d'un dialogue entre hydrologue et gestionnaire qui doit respecter des critères de clarté , d'efficacité et de cohérence. On insiste particulièrement sur la nécessaire prise en compte par l'hydrologue d'une expression, même simplifiée, des conséquences dommageables réelles pour le gestionnaire ou les membres de la cible du risque.

EN:

The often vaguely defined concept of risk is analyzed and an attempt is made to provide a rational definition which embeds both probabilistic aspects and consequences associated with decisions. Any risk analysis in the field of water resources involves a dialogue between the hydrologist and the decision-maker, a dialogue which should be clear, eflicient, and coherent. We insist inparticular on the need for the hydrologist to use an expression, even simplified, of the actual damage potential, by the decision- maker or the people at risk .

FR:

Un modèle de génération stochastique de pluies est couplé ˆ un modèle de calcul de l'index de leurs érosivités, dérivé de l'Equation Universelle de l'Erosion des Sols (USLE). Le premier fonctionne au pas de temps de 30mn, il est calé sur une série pluviographique de 15 ans de la Tunisie centrale. Le second modèle fonctionne par calcul automatique des cumuls et moyennes de l'érosivité des pluies générées.

En mode opérationnel, ces deux modèles sont exploités pour simuler les aléas de l'envasement annuel des réservoirs collinaires de la zone aride et semi-aride de la Tunisie : le bassin versant est considéré comme une "boite noire" où l'agressivité climatique est la principale variable (quelques pluies extrêmes font l'essentiel de l'érosion), les autres facteurs sont considérés constants durant la durée de service du réservoir.

Nous observons sur trois bassins versants répartis du nord au sud de la frange comprise entre 500mm et 250mm (de pluie moyenne annuelle), que la distribution annuelle des index d'érosivité des pluies peut être assimilée ˆ la distribution des transports solides.

Sur l'un de ces bassins versants (OUED EL HISSIANE : 15,9 ) nous observons également que les valeurs extrêmes de l'érosion sont proportionnelles aux valeurs extrêmes de l'index d'érosivité des pluies. Seulement l'automne et le printemps sont des saisons érosives.

Dans le cas de petits bassins versants non-jaugés, comme ceux pour l'aménagement de réservoirs collinaires, le générateur nous permet de constituer des chroniques d'érosivité de pluie. Si on considère que les autres paramètres sont constants, ce modèle nous aide à déterminer les intervalles de confiance de durées de service probables.

Une analyse de sensibilité par la modification des paramètres du générateur (nombre d'épisodes, hauteur de pluie, maximum et durée d'averse etc ...) valide la méthodologie. De même une analyse régionale montre les faibles fluctuations des résultats sur l'étendue aride et semi-aride de la Tunisie.

Ces deux résultats nous ont conduit à proposer un abaque régional de prédétermination des fluctuations probables des durées de service des réservoirs collinaires, compte tenu de la connaissance préalable de la durée de service moyenne probable.

Cette méthode directement opérationnelle peut être utilisée pour l'aménagement, la planification, et la gestion des réservoirs collinaires. Elle améliore les études de faisabilité, notamment lorsqu'on la couple aux calculs économiques.

EN:

A stochastic rainfall generator and a model for calculating the erosivity index of the Universal Soil Loss Equation (USLE) are coupled. The first one operates on a 30 min time step and it is calibrated on a 15 year historical pluviographic database for central Tunisia. The second model works like an automatic calculator, where the rain erosivity index can be averaged or summed.

These two models are applied to simulate the annual siltation of hill reservoirs in arid and semi-arid areas of Tunisia. The catchment is considered as a "black-box" where mostly climatic aggressivity is varying (some extreme rainfall events cause most of the erosion) and the other erosion parameters are considered to remain constant over the lifetime of the reservoir.

We know that the WISCHMEIER rain erosion index has no upper limit. When all the other factors are maintained constant, erosion is proportional to it. This climatic index is calculated for each rain event, but we can also calculate the annual average or sum it over many years. Soil losses (or solid transport) during a period are proportional to the value of this index summed over the same period.

We demonstrate on three representative catchments lying from north to south (from 500 mm to 250 mm annual average rainfall) that the annual rain erosion index distribution can be considered comparable to the solid transport distribution (Galton distributions with equal statistical parameters).

For the OUED EL HISSIANE catchment (15.9 km²), the annual extreme value of erosion is proportional to the annual extreme value of the rain erosion index. Only spring and autumn storms need be considered to calculate the annual sediment loss.

On gauged small catchments like those of hill reservoirs, the rainfall generator can provide an erosivity rainfall record. Considering that the other erosion parameters are constant, this model allows one to calculate the confidence intervals of a reservoir's predicted durations. We equate the annual erosivity index distribution to the annual solid supplies distribution, and we generate and sum over several consecutive years many series of "solid supplies" to simulate the progressive silting up of the reservoirs (represented by the chosen period : sum over T=5 years, 10 years, 15 years ...). In the context of setting up hill reservoirs, this study concerns the potential duration and possible services of hydraulic structures (including the risk of rapid silting up of the reservoirs). This generator can be computed over many years and avoids the limitation due to the short observed time series.

The methodological benefit of such an approach is therefore to avoid the difficulty of limited observed samples. It enlarges the data analysis of the series dispersal and residuals. This approach, based on observations and a statistical method, supports experimental analysis. It gives a clear idea of the rate of siltation, according to climatic hazards.

A sensitivity analysis, where we modify the rainfall generator parameters (number of episodes and quantity of rain, peak and duration of storm...), is used to validate the methodology. A regional analysis shows a very low fluctuation considering the whole arid and semi-arid area of Tunisia. These two results led us to set up a regional abacus. This abacus reports the forecasts of the fluctuations of the time length service for hilly and small dams, including the probable average of their service duration. It allow a predetermination of the small dams' life time together with a confidence interval, and it can be used for small dam construction, planning and management. It improves the first feasibility study and can be included in economic calculations.

FR:

La caractérisation précise de l'aléa climatique nécessite l'exploitation de mesures reposant sur la période d'observation la plus longue possible. Souvent cette information est constituée de mesures au sol à partir de postes pluviométriques. L'évolution dans l'espace et dans le temps des réseaux de pluviomètres introduit un biais dans toute étude stochastique spatiale ou ponctuelle reposant sur des séries de valeurs échantillonnées à partir d'un tel réseau. On se propose dans cet article de quantifier la potentialité d'un réseau de pluviomètres à intercepter des surfaces de pluie, en fonction de leur aire et des caractéristiques de ce réseau à une date donnée. On procède par simulation à partir du réseau de pluviomètres géré par Météo-France sur la région Languedoc-Roussillon, étudié sur une période de 123 ans. On définit la notion de pourcentage d'observation, qui représente la proportion de surface pluvieuse affectant la région et qui ont été interceptées par le réseau de mesure. Toutes études statistique reposant sur des séries de mesure échantillonnées à partir du réseau seront biaisées, étant donné qu'entre 1958 et 1993, on observe qu'une proportion des surfaces pluvieuses de moins de 2000 km² qui ont touché la région étudiée. Ce pourcentage d'observation est ensuite utilisé pour débiaiser les estimations de l'aléa pluvieux régional reposant sur le réseau de pluviomètres.

EN:

The most usual rainfall risk assessment, based on a stochastic approach, or an accurate quantiles estimation, requires long series of observations. Most of the time when long periods of observation are considered, the available information consists of data from daily rain gauge networks which are evolving in space and time during these periods. As the rainy surfaces which generate the highest intensities are localised in space, the intergauge distances may be too large to "observe" all the rainfall events occurring over a given network. Thus it could bias the stochastic results based on values sampled from such a network, especially when extreme rainfall events are considered. The aim of this paper is to estimate the capacity of a daily rain gauge network to intercept rainy surfaces according to their area and the network density. The results have been used to estimate the bias introduced in rainfall risk assessment using the regional frequencies of isohyets areas observed in the studied region.

The network studied is the Languedoc-Roussillon daily rain gauge network, in a French region along the Mediterranean sea. The network has been developed by Météo-France since 1870. The number of gauges put into service has varied during the 1870-1993 period of observation: from 3 gauges in 1870, the maximum reached was 353 gauges in 1969 and 1972, which represented a spatial mean density of 12.6 gauges/ 1000 km². Since 1972 the number of gauges has decreased; in 1993 the gauge density was the same as in 1963, with 10.6 gauges/ 1000 km². Nevertheless the clustered gauges have been reduced, as have the maximum intergauge distances, and the network has become more homogeneous over the region.

Using simulation, the percentage of rainy surfaces which have affected the region, and which have been observed by the rain gauge network, has been estimated, as a function of the rainy surfaces area and the rain gauge density. It could be interpreted as the empirical expression of the probability to observe a given rainy surface with a given network configuration. Two periods have been considered, 1870-1957 and 1958-1993.

Two simulation methods have been used: in the first the rainy surfaces have been considered to be static and in the second their motion has been taken into account. It has appeared that considering the motion of rainy surfaces yields the same results as the static method but with a different rainy surface geometry. The small differences between the percentage of rainy surfaces observed by the network in both cases can be explained by the simulation methods.

It has been shown that the average probability over the period from 1870 to 1957 of observing a given rainy surface is 2 to 4 times less than the average probability over the 1958-1993 observation period, during which the gauge density has increased and the network has become more homogeneous over the region: over the 1870-1957 period the rain gauge network intercepted 50% at least of the rainy surfaces equal to or larger than 700 km2 but in the 1958-1993 period 50% at least of the rainy surfaces were observed if their area exceeded 80 km². If the rainfall event which affected the N"mes hydrological system on 2-3 October 1988 is considered, these results have shown that the average probability over the 1870-1957 observation period to observe such an event is 2 times less than over the 1958-1993 observation period.

In a recent study, a rainfall risk assessment has been made over the Languedoc-Roussillon region, using the frequencies of the isohyets areas defined for different rain thresholds, for 24-hour and 48-hour durations. These isohyets areas have been estimated on the basis of a sample of 93 rainfall events selected over the Languedoc-Roussillon region from 1958 to 1993 (Neppel et al., 1998). A method to estimate the bias introduced by the network in the estimation of the isohyets area return periods has been carried out, using this empirical probability estimated with the static simulation method. It has been shown that the bias only affects the more frequent isohyets area quantiles, corresponding to return period of 1 year for 48-hour duration and 1 to 3 years for 24-hour duration. Moreover, for this sample and this network, it has been shown that the bias would be negligible compared to the quantiles 5% confidence limits, whatever the return period and the time step. It must be noted that with this sample the 5% confidence limits of the quantiles sometimes reach 100% of the quantiles. The results are related to the sample and the network configuration, and they should not be extended to other areas or other samples: a larger sample over the same region could lead to narrower confidence limits, in which case the bias might no longer be negligible. In particular, the use of historical data needs to consider the longest observation period. Usually the rain gauge density decreases over such observation periods, which leads to a lower empirical probability of observing rainy surfaces according to their area. Thus the bias influence may increase, especially compared with the quantiles 5% confidence limits which are reduced when the sample is enlarged. Nevertheless the method described here is general and may be transposed to other geographical zones, provided that the isohyets area frequencies and the empirical probability of observing a rainy surface according to its area, corresponding to the network under consideration, are known.

The current tendency in France is to reduce the number of daily rain gauges, managed by volunteers, and to replace them by automatic rain gauges. However in such a case the density would decrease and reach that observed in 1900. When rainfall risk assessment is considered, this study has shown the drawbacks of such a policy.

FR:

Les régions tropicales subissent une déforestation importante. En Amérique du Sud,la forêt est généralement remplacée par une prairie, C'est pourquoi nous avons étudié le comportement hydrologique de 2 petits (1,5 ha) bassins versants. Un bassin (bassin B) est recouvert par une forêt primaire, tandis que le second (bassin A) a été défriché et transformé en prairie (Digitaria swazilandensis, programme ÉCÉREX, ORSTOM/CTFT). Ces bassins, situés en Guyane Française, sont proches (500 m), escarpés et principalement constitués par des sols à drainage vertical ralenti. Le climat est de type tmpical humide avec une température moyenne (26 °C) et des précipitations moyennes annuelles (3500 à 3900 mm/an) élevées. L'évapotranspiration réelle et potentielle de la forêt primaire sont respectivement égales à 1470 mm/an et 1565 mm/an, En période d'étiage, nous avons observé un écoulement permanent à l'exutoire du bassin A, alors que le bassin B en est dépourvu. Deux crues (24 mai 1992 et 15 mai 1993) ont été étudiées, simultanément sur les 2 bassins. Pendant les crues, nous avons prélevé des échantillons d'eau des précipitations (pluie et pluviolessivat), des ruisseaux et du sol. Sur ces sites, l'eau circulant dans les couches peu profondes du sol présente une concentration élevée en K+ et faible en Cl-. Une signature opposée caractérise l'eau des couches pmfondes du sol. L'analyse des relations existant entre les traceurs chimiques (K+, Cl-) et isotopique l80) ainsi l'étude des propriétés hydrodynamiques du sol permet de décomposer qualitativement l'hydrogramme de crue en 3 réservoirs: sol superficiel (écoulement hypodermique), sol intermédiaire (de 0 à - 0,4 m), sol profond (bassin B) ou nappe (bassin A). Une décomposition quantitative a été effectuée en utilisant des traceurs chimique (Cl-) et isotopique l80). Nous avons ainsi montré que les crues sur les 2 bassins sont dominées par l'écoulement issu des couches intermédiaires du sol qui représente environ la moitié de l'écoulement total de crue. Cependant,les mécanismes de génération des crues diffèrent sur les 2 bassins. Sur le bassin A, les couches profondes du sol sont saturées avant la crue et participent donc à la totalité de la crue. Au contraire, sur le bassin B, les couches profondes de sol atteignent la saturation peu de temps avant le pic de crue et participent donc essentiellement aux écoulement pendant la décrue. Ces résultats confirment les études hydrologiques réalisées précédemment (FRITSCH, 199Ù) et permettent d'identifier les mécanismes de genèse des crues et ainsi de mettre en évidence l'effet de la déforestation.

EN:

The tropical regions are subjected to fast deforestation. In South America, the tropical rain forest is being replaced by grassland. Thus, we have studied the hydrological behaviour of two small (1.5 ha) watersheds. One basin (hereafter named "B" basin) is still covered by primary forest while the second one (hereafter named "A" basin) was cleared and transformed to grassland (Digitaria swazilandensis, ÉCÉREX program, supported by ORSTOM/CTFT). These basins, located in French Guyana, are close to one another (500 m), steep, and are principally constituted of soils showing lateral drainage. The tropical humid climate is characterized by a high mean interannual temperature (26_¡C), which varies slightly from month to month, and by a high mean annual precipitation (3500 to 3900 mm yr-1). Precipitation mainly occurs during the main wet season from May to June and during a secondary wet season from December to January. Real evapotranspiration of the natural forest is 1470 mm yr-1 and potential evapotranspiration is 1565 mm yr-1. During the low-water level period, we have observed perennial runoff at the outlet at the "A" basin while the "B" basin is without permanent flow. We have studied two runoff events (24 May 1992 and 15 May 1993) in both basins. On 24 May 1992, the runoff event was caused by a rainfall lasting for about 10 hours. Total precipitation was 53.8 mm. The main event amounted to 32 mm. The main peak of the hydrograph corresponded to the heaviest rainfalls. On 15 May 1993, the runoff event was caused by a rain lasting for about 13 hours. Total precipitation was 64.0 mm. The main peak of the hydrograph (86.2 L s-1) corresponded again to the heaviest rainfalls. Spatial variability of the precipitation amount was high, especially for the most intense events that have the largest standard deviations. Interception by the canopy amounted to 5.3% of the rainfall in 1992 and 4.3% in 1993. High rapid runoff coefficients were observed, i.e., 0.28 for 24 May 1992 and 0.43 for 15 May 1993. No overland flow was observed in the watershed.

Samples of rainwater, throughfall, stream water, and soil water were regularly collected in both watersheds during the runoff events. Temporal variations in the isotopic composition of the stream water at the outlet of the watershed paralleled variations in rainwater but with a distinct shift. The difference between the two signatures could be due to a mixture between:

- Rainwater and water present in the watershed before the event and whose isotopic composition is different and variable over space.

- Rainwater and water originating from various reservoirs whose contribution to the stream varies with time.

The analysis of runoff events using the isotope tracer method revealed the existence in the stream of a mixture of water originating from rain and from one or several other reservoirs in the watershed. Isotope tracers alone were not sufficient to estimate the depth of the soil water contributing to the runoff event. On one hand, temporal variability in the isotopic composition of rainwater was very similar to the vertical spatial variability in the isotopic composition of soil water. On the other hand, surface evaporation in the watershed was negligible: the isotopic signature of water originating from soil during runoff events was the consequence of successive infiltrated rain events. Oxygen-18 content in rain water strongly varied with time but only slightly with space because of the small area of the watershed. Because of this temporal variability, an average isotope content of rainwater could not be used when calculating the contribution of "new water" at the outlet of the watershed.

Using chemical and isotope tracers is a way to identify and quantify the contribution of the various water reservoirs to runoff. We were thus able to separate runoff hydrographs into simple components (water from superficial layer, intermediate layer and deep layer). In these watersheds, shallow water was characterized by relatively high concentration in potassium and very low concentration in chloride. An opposite signature characterized deep water

A "deep water" chemical tracer (chloride) - isotope tracer (18O) diagram shows the evidence of a hysteresis relationship:

1. The decreasing limb of this relationship (rising segment of the hydrograph) is due to a decrease in heavy isotope content resulting from the decrease of oxygen-18 content in the precipitation and from the arrival of water from upper soil layers with low concentrations of chloride.

2. The increasing limb (falling segment of the hydrograph and recession) is associated with the arrival at the outlet of deep waters containing relatively high concentrations of chloride and heavy isotopes.

Using chemical (Cl-) and isotope (18O) tracers, quantitative hydrograph separation was achieved with a simple 2- or 3- component conservative-mixing model. This information allowed qualitative hydrograph separation into 3 reservoirs: superficial soil layers, intermediate soil layers (0 to -0.4 m), deep soil layers ("B" watershed) or ground water ("A" watershed).

Thus, the runoff event of both basins was dominated by the intermediate soil layers reservoir, which represents half of the total flow for both basins. However, the processes of runoff generation differ: in the "A" watershed, the deep soil layers were saturated before the rain: the contribution is significant throughout the runoff . In the "B" watershed, the deep soil layers become saturated a few times before the peak flow: their contribution dominates during the recession. These results confirm previous hydrological studies (Fritsch, 1990), which showed the high reactivity of the watershed, and give a better insight into the mechanisms involved.

Some of these observations can also be used at a larger scale:

1. Identification of the reservoirs contributing to the runoff event by analyzing the relationships between oxygen-18 content and the flow rate, and between isotope and chemical tracers.

2. Simultaneous samplings along the stream in order to detect a possible zonation of the watershed. These samples must be taken during a runoff event as well as during a low-water level period to check whether the tracer concentrations in the continuous or discontinuous water table supplying the stream are heterogeneous. If the signature of the water table is heterogeneous or if the stream is supplied by several water tables with different chemical concentrations, the watershed must be divided into several homogeneous sub-watersheds.

FR:

Les nappes d'eau souterraine constituent des ressources vulnérables que l'activité industrielle croissante contribue à polluer trop fréquemment. Les produits les plus manipulés se voient donc directement concernés dans les cas de pollution d'aquifères. A ce titre, les produits pétroliers, notamment ceux que l'on considère comme étant domestiques (gazole, fuel, essence, etc.) se situent au premier rang.

Dans le but de mieux aborder ce type de contamination, des études sont menées sur un site expérimental de grandes dimensions et parfaitement contrôlé, dans lequel 476 l de gazole routier (GOR86) ont été déversés. Une première tentative de récupération du polluant, par infiltration de tensioactifs depuis la surface, démontre la capacité du mélange utilisé à mobiliser le polluant mais conduit à la réduction de la perméabilité du milieu poreux faisant ainsi apparaître une limitation des performances pour ce type de procédés de décontamination.

Des expériences au laboratoire, conduites sur des colonnes de milieu poreux, ont été entreprises pour rechercher les causes de cette perte de conductivité hydraulique. Avec le sable considéré ici, les argiles ne jouent pas un rôle prépondérant dans le colmatage. Il est démontré également que le gradient de charge, quand il est augmenté, permet d'injecter une quantité plus importante de tensioactifs sans pour autant pallier ce problème de colmatage. Enfin, il apparaît que les tensioactifs, en présence des ions calcium, peuvent s'agglomérer pour former des associations de micelles lamellaires, cylindriques et/ou mixtes, assez volumineux pour être filtrés à la surface du sol à traiter. Au bout d'un certain temps, le milieu poreux est totalement colmaté et par conséquent le débit d'infiltration de la solution devient nul. Sur le plan de la récupération du polluant, ce paramètre s'avère être particulièrement important. Sa variation influence donc fortement les quantités de polluant mobilisé.

Cette étude démontre la faisabilité du procédé. Toutefois elle souligne aussi que, sous peine de perdre de leur efficacité, les solutions de tensioactifs doivent conserver leurs caractéristiques initiales (stabilité de la taille des particules colloïdales notamment) pendant toute la durée du traitement. Dans la mesure où l'agglomération des micelles est en cause, il faudra rechercher les moyens d'obtenir une dispersion plus stable, par exemple par addition d'agents solvants.

EN:

In case of groundwater contamination by hydrocarbon spills, one of the main problems is how to recover residual hydrocarbons trapped in the porous medium, forming a long term pollution source. In order to develop a better approach to such problems, experiments were conducted in a large experimental controlled site, called SCERES (Site Contrôlé Expérimental de Recherche pour la Réhabilitation des Eaux et des Sols) made of an impervious concrete basin (25 m x 12 m x 3 m) packed with two layers of quartz sand and fitted with specific instrumentation.

This research site was experimentally polluted with diesel oil (476 litres), then the shape of the impregnation body was identified and oil saturation values were quantified thanks to a specific coring programme before and after each step of the experiments (water table fluctuations, hydraulic pumping, surfactant infiltration, etc.). The principle of the remediation technique is based on a surfactant flushing from the soil surface. This process was carried out when all of the removable diesel oil had been recovered by hydraulic pumping in the central recovery well. The surfactant infiltration allowed a displacement of the main part of the residual pollutant in the vertical zone of the impregnation body, in spite of a progressive reduction of hydraulic conductivity leading to a plugging of the porous matrix. This phenomenon has been observed by many scientists (Allred et Brown, 1994 ; Celik et al., 1982) and explained as being caused by a precipitation of the anionic surfactant in presence of calcium ions.

This decrease of hydraulic conductivity, due to surfactant infiltration and its effects, was studied in the laboratory by the implementation of experiments using columns filled with the same sand as in the SCERES basin. The results showed that the reduction of the hydraulic conductivity of the porous medium cannot be due to the precipitation of the anionic surfactants in presence of calcium ions of the sand. If this were situation, all the porous medium in the column would have exhibited this loss of hydraulic conductivity. In our case, only the upstream part of the physical model was influenced by this phenomenon.

Furthermore, the sand devoid of its clay minerals has nearly the same behaviour as a natural sand, with respect to the infiltration flowrate of the surfactants. This proves that the clay minerals, present in small proportion in this matrix, are not mainly responsible for the plugging problem. The influence of the hydraulic gradient, another studied parameter, was observed and permitted to apprehend the behaviour of the surfactant solution during its infiltration through the porous medium. . The increase of this parameter, even if it allows the injection of more surfactant, could not resolve the problem of the reduction of hydraulic conductivity.

Finally, all the experiments indicated that the plugging effect (cancellation of the flowrate) happened after nearly 20 hours of surfactant infiltration through the porous matrix. The time parameter seems to be important since its variation can influence the stability of the surfactant solution. The results indicated that we can infiltrate a greater volume of the solution if it has been recently prepared. In other words, this is certainly due to the time necessary for the formation of liquid crystals obtained in presence of the calcium ions in the water used for the preparation of the solutions (tap water). After nearly 20 hours these crystals should have attained a size sufficient to be retained by a filtration effect, near the soil surface. We suggest that this filtration is the major factor responsible for the reduction of hydraulic conductivity and the plugging.

The present study shows the feasibility of this process and it highlights the need to conserve the initial characteristics of the surfactants (especially the size stability of the colloidal particles) during all the treatment. Because of the agglomeration of the micelles, we have to search for means to ensure a good dispersion in the aqueous medium, for example by adding a solvent agent.

FR:

Le présent travail porte sur l'évaluation des différentes formes de phosphore, dans les sédiments d'un réservoir eutrophe en climat méditerranéen (El Kansera, Maroc) et la détermination des relations entre ces formes. L'application d'un schéma de fractionnement chimique du phosphore a permis de déterminer les proportions de quatre formes différentes : le phosphore lié au fer Fe(OOH)-P, le phosphore lié au calcium CaCO3-P, le phosphore organique soluble dans l'acide POSA et le phosphore organique résiduel POR. La forme biodisponible a été déterminée par ailleurs, par des bioessais en utilisant une Chlorophycée (Scenedesmus crassus) isolée à partir des eaux de la retenue du barrage étudié.

Une étude par analyse factorielle des correspondances a mis en évidence l'influence de la nature des sédiments étudiés et de leurs propriétés physico-chimiques sur la répartition des formes du phosphore dans les sédiments. En période de basses eaux, les sédiments sont caractérisés par la prédominance des formes inorganiques ; alors qu'en période de hautes eaux, la forme POSA est la plus abondante. La confrontation des résultats obtenus par la méthode des bioessais avec ceux relatifs aux extractions chimiques séquentielles de phosphore a montré par ailleurs, que la forme biodisponible du phosphore est fortement corrélée aux formes inorganiques {Fe(OOH)-P et CaCO3-P} et indépendante vis à vis des formes organiques {POSA et POR}.

EN:

We have studied in this work the different forms of phosphorus in the sediments of a eutrophic reservoir in a Mediterranean climate (El Kansera – Morocco). The studied sediments have been collected at eight sampling stations located at different points in the middle of the lake during two sampling sessions in contrasting weather conditions: the first sampling occurred in November 1992, a period of autumn mixing and low water levels, whereas the second took place in January 1994 during a high water level period, just after the first winter floods of the Beht river whose waters are retained in the studied reservoir.

The evaluation of the different forms of phosphorus was carried out using both a chemical extraction method and a bioassay method. The application of the chosen chemical extraction scheme allowed the determination of the proportions of four different phosphorus forms: iron-bound {Fe(OOH)-P}, calcium-bound {CaCO₃-P}, acid-soluble organic {POSA} and residual organic {POR}. The bioavailable form {Pbio}, representing less than 32 % of the total sediment phosphorus content, has been evaluated using a Scenedesmus crassus algal strain isolated from El Kansera reservoir waters.

According to the results of our analyses, the concentrations of iron-bound phosphorus are higher overall than those of calcium-bound phosphorus. The sediments collected during the low water period contain spatially stable values of POSA (0.03 – 0.09 mg.g^-1). These values are lower than those determined in the sediments collected during the high water period (0.24 – 0.42 mg.g^-1). This form represents less than 11% of the total phosphorus in the low water period whereas it represents more than 25% and up to 56% of the total phosphorus in the filling period. This result can be explained by the freshness of the organic matter supplied by the continuous water flow in the high water period and the degradation of this organic matter during the low water period. Moreover, the anaerobic conditions of the hypolimnetic layers and the low external supply during the low water period favour organic phosphorus mineralisation.

A Factorial Correspondence analysis study has shown the influence of the nature and the physico-chemical properties of the studied sediments on the distribution of the different phosphorus forms. In the low level period, the predominance of the inorganic forms is obvious while POSA is the most abundant form in the period of rising water levels. This differentiation can be explained by the different transport dynamics of suspended organic or inorganic particles according to the water level and also to the external or internal supply of the organic matter.

We have also determined the statistical relations among the estimated different forms of phosphorus in El Kansera reservoir sediments. This statistical analysis was carried out to establish the correlation between the bioavailable phosphorus determined by the bioassay method and the phosphorus forms determined by chemical speciation. The goal of the study was to determine, according to statistically significant correlations, the chemical forms of the bioavailable phosphorus contained in the studied sediments.

The comparison between the results we have obtained by the bioassay method and those obtained by sequential chemical extraction has shown that the bioavailable form of phosphorus is strongly correlated with the inorganic forms {Fe(OOH)-P} and {CaCO₃-P}. The high correlation found between the iron-bound form of phosphorus and the bioavailable one is in good agreement with previous work described in the literature that has shown that an important part of the iron-bound and aluminium- bound phosphorus is bioavailable, whatever the extraction method may be.

It has also been demonstrated that the forms {Pbio}, {Fe(OOH)-P}, {CaCO₃-P} and {Pinorg}are notably independent from the {POSA} and {Porg} forms, the correlation coefficients being relatively low. The residual organic form of phosphorus {POR} does not seem to be linked to any variable. Therefore, we can say that this form is not bioavailable for the Scenedesmus crassus species.

FR:

Cette étude a été réalisée dans le cadre d'un projet multidisciplinaire sur la gestion et la protection de l'habitat des salmonidés et sur l'évaluation des perturbations que subissent les habitats de ces poissons dans les eaux courantes suite aux coupes forestières et à la construction de routes. Afin d'identifier les voies d'écoulement responsables de la qualité des eaux de surface d'un petit bassin versant forestier, une étude approfondie a été entreprise sur l'évolution de la qualité de l'eau de pluie lors de son passage à travers la phytocénose et la couverture pédologique jusqu'au ruisseau. La signature chimique des compartiments du bassin versant servira d'intrant quant à l'application et l'analyse du modèle EMMA (end-members mixing analysis).

La signature chimique de l'eau du ruisseau s'explique par un graphe x-y (graphe de mélange) sur lequel la composition chimique des compartiments et celle du ruisseau sont reportées. Si trois compartiments circonscrivent la signature chimique du cours d'eau, alors on peut émettre l'hypothèse que ces compartiments se mélangent de façon conservatrice pour donner la qualité des eaux de surface du bassin versant. Plusieurs traceurs (conductivité électrique, SO₄^2-, Cl^-, NO₃^-, K⁺, Al_t et Fe_t) naturels n'ont pas servi à l'identification des compartiments parce que le modèle ne tient pas compte de certaines conditions, tels l'activité biologique, l'état hydrique des profils, etc. Seuls le pH, Na⁺, Ca²⁺, Mg²⁺ et SiO₂ se sont avérés des traceurs utiles.

La nappe phréatique a été incluse par défaut dans le modèle puisqu'il était connu qu'elle assurait la base de l'écoulement du cours d'eau en tout temps de l'année. Les sols de la plaine d'inondation semblent également prendre part à la qualité de l'eau du ruisseau, particulièrement les horizons B podzoliques, lesquels sont saturés d'eau pendant toute la période sans gel. C'est donc dire que l'écoulement de l'eau souterraine et l'écoulement hypodermique au niveau des horizons B de la plaine d'inondation sont les voies d'écoulement qui expliquent le mieux la qualité des eaux de surface du bassin versant.

Toutefois, la séparation de l'hydrogramme par l'équation du bilan massique a montré qu'un modèle à trois réservoirs (nappe phréatique, horizons B des versants sud et nord) ne peut pas donner des résultats satisfaisants quant à la simulation de la charge chimique des eaux de surface. Le modèle élimine systématiquement trop de compartiments pouvant s'avérer explicatifs de la qualité de l'eau du ruisseau. Un modèle mécaniste développé à partir des variations du niveau de la nappe phréatique, de la conductivité hydraulique et de la composition chimique des solutions de sol permettrait de reproduire plus rigoureusement l'hydrogramme du ruisseau. Le modèle EMMA demeure tout de même un bon outil pour réfuter ou confirmer une hypothèse de recherche car il met clairement en relation la composition chimique des compartiments à celle du ruisseau et enlève parfois tout doute quant à l'action d'un processus susceptible d'alimenter le cours d'eau.

EN:

Intense forest harvesting is suspected as a cause of soil acidification. Inputs of acidity into the soil system may lead to high concentrations of metal ion species in the soil solution and surface waters. Some of these metal ions, e.g., Al³⁺, can cause toxic responses to fish and aquatic invertebrates. Timber-induced soil and surface water acidification is considered to be a short-lived phenomenon during the growing season following the cut. Vegetation loss could mean increased frequency of high Al³⁺ and H⁺concentrations in stream water rather than increased mean levels.

Many types of tracers are useful for hydrograph separation. Isotopic (e.g., deuterium and tritium) and natural chemical tracers (e.g., pH, Cl^-, SiO₂) have been used extensively to interpret chemical data gathered in catchment studies. The ability of computer simulation models to reproduce the hydrograph and chemical species in streamwater varies. In some cases, the use of too many hydrological parameters (i.e., over-parameterization) can make the validation of reactions responsible for streamwater chemistry almost impossible. Recently, advances in hydrological modelling have been made by considering that streamwater chemistry is a mixture of groundwater and soil solutions at different depths. One model that originates from this hypothesis is EMMA (end-member mixing analysis). Chemical species that are variable with depth within a same soil profile were shown to be highly correlated with streamwater discharge. Generally, chemical species that show high concentrations in surface horizons increase in streamwater during high flow, whereas chemical species found in high concentrations in lower horizons are higher during low flow.

In order to identify end-members that can potentially contribute to streamwater chemistry of a small catchment in central New Brunswick, we investigated the chemistry changes of rainwater entering the catchment, passing through vegetation and soils and reaching the stream channel. The chemical composition of the catchment's end-members will serve as input in order to run and analyse the EMMA model. Furthermore, a better knowledge of water flowpaths that dominate in the catchment could be valuable information for the Department of Fisheries and Oceans, who, in 1990, initiated a multidisciplinary project on 1) the protection and management of the salmonid habitat, and 2) the effects of forest harvesting and road construction on the freshwater habitats of these fish. Harvest operations are planned from 1996 until 1999.

Streamwater chemistry is explained by a x-y graph (mixing diagram) on which the end-members and streamwater chemical composition are plotted. Because end-member chemistry is stable over time and space, mean values of tracers are plotted on the mixing diagram. Streamwater chemical compositions have all been plotted on the graph since they vary significantly with flow. If the chemical composition of three end-members enclose the streamwater chemical composition, then it can be assumed that these end-members mix conservatively to produce streamwater chemistry. If two chemical species mix non-conservatively, then the model will not accurately indicate the relative contribution of each end-member. Generally, the mixing diagram does not validate conservative mixing, but it can be used to test a mixing hypothesis. For example, if streamwater chemistry falls largely outside the end-members chemical composition, then at least one end-member is incorrectly characterized (or missing), or the end-members do not mix conservatively.

The relative contribution of selected end-members are obtain from the mass balance equation: C_tQ_t=C₁Q₁ + C₂Q₂ + C₃Q₃, where 1, 2 and 3 refer to the three end-members, C1,2 and 3 are the soil water concentrations of conservative elements for each end-member, and Q1,2 and 3 are the amounts of soil water for a given end-member. With this equation, the concentrations of a number of elements for each end-member (C₁, C₂, C₃) are used simultaneously to estimate a single value for each Q₁, Q₂ and Q₃. Since we want to quantify the contribution of each end-member to the total stream discharge, i.e., the mix of the three end- members, Q_t is set to 1. Once values for Q₁, Q₂ and Q₃ are calculated, the results are interpreted in terms of a hydrograph separation to show the contribution from each end-member to the overall stream discharge.

Two soil toposequences that correspond to typical soil profiles along the northern and southern hillslopes were selected. From June to November 1995, wet deposition, throughfall, soil solutions at four depths and streamwater were collected. Samples were analyzed for pH, electrical conductivity, Na⁺, Ca²⁺, Mg²⁺, K⁺, SO₄^2-, Cl^-, NO₃^-, SiO₂ , Al_t and Fe_t.

Many natural tracers (electrical conductivity, SO₄^2-, Cl^-, NO₃^-, K⁺, Al_t and Fe_t) have not been used to identify end-members because the model does not always consider adequately some conditions or processes that go on in the catchment, e.g., biological activity and Eh. Because they vary considerably with depth, solution pH, Na⁺, Ca²⁺, Mg²⁺ and SiO₂ have been shown to be useful tracers.

Groundwater has been included in every diagram as one of the three end-members mixing conservatively to produce streamwater since it is certain that it contributes a large portion to the total discharge under any hydrological condition. Soils along the stream seem to contribute the rest of the streamwater chemistry, particularly B horizons which are submerged all summer by groundwater. Thus, groundwater and subsurface flow at the base of the soil profiles along the stream seem to be the principal flow mechanisms that control streamwater chemistry in the catchment.

However, hydrograph separation shows that a three end-member model (i.e., groundwater, B horizons from the northern and southern hillslopes) is not enough to simulate streamwater chemistry. Saturated subsurface flow in the B horizons from both sides of the stream should contribute approximately the same amount to the total discharge since groundwater affects both end- members throughout the growing season. In that respect, groundwater level fluctuations at this depth of the soil profiles should not be considered as a cause of this discrepancy. What can be said at this point is that one end-member that is incorrectly defined in space, and that has a similar chemical composition to saturated subsurface flow coming from the southern hillslope, is the primary source (with groundwater) of stream discharge during events. It is thus better to interpret this information in terms of solution type rather than in terms of physical origin (northern or southern hillslope). In this manner, the stream water is provided by both hillslopes.

In conclusion, the model eliminates systematically too many end-members that could partially explain streamwater chemistry. Results show that a more complex mixture is necessary to reproduce streamwater chemistry. A mechanistic model based on groundwater level fluctuations, hydraulic conductivity and soil solution chemistry would possibly have better success in reproducing the stream hydrograph. However, EMMA remains a useful tool to refute or confirm the possible action of a flow mechanism by correlating the chemical composition of end-members with streamwater chemistry.