Société géologique de France

Bull. Soc. géol. France, 2002, t. 173, n° 3, pp. 245-253

Mots clés. - Karst côtier, Almyros, Intrusion saline, Conduit, Modélisation.

Résumé. - Les variations de salinité et de débit à la source aérienne naturellement salée de l'Almyros d'Héraklion en Crète ont été suivies pendant deux cycles hydrologiques. Le fonctionnement du système karstique côtier de l'Almyros est présenté mettant ainsi en évidence l'influence de la dualité de l'écoulement dans le karst (conduits et matrice fissurée) sur la qualité de la ressource en eau en zone côtière. Le mécanisme principal de salinisation de l'eau alimentant la source Almyros est identifié et validé par un modèle mathématique hydraulique qui rend compte de façon remarquable des observations.

Key words. - Coastal karst, Almyros, Saline intrusion, Conduit, Modelling.

Abstract. -Variations in salinity and flow rate in the aerial, naturally salty spring of Almyros of Heraklion on Crete were monitored during two hydrological cycles. We describe the functioning of the coastal karstic system of the Almyros and show the influence of the duality of the flow in the karst (conduits and fractured matrix) on the quality of the water resource in the coastal area. A mechanism of saltwater intrusion into this highly heterogeneous system is proposed and validated with a hydraulic mathematical model, which describes the observations remarkably well.

Introduction. - Fresh groundwater is a precious resource in many coastal regions, for drinking water supply, either to complement surface water resources, or when such resources are polluted or unavailable in the dry season. But coastal groundwater is fragile, and its exploitation must be made with care to prevent saltwater intrusion as a result of withdrawal, for any aquifer type, porous, fractured or karstic. In karstic zones, the problem is very complex because of the heterogeneous nature of the karst, which makes it difficult to use the concept of representative elementary volume developed for porous or densely fractured systems. The karstic conduits focus the major part of the flow in preferential paths, where the water velocity is high. In coastal systems, these conduits have also an effect on the distribution of the saline intrusion. As was shown e.g. by Moore et al. [1992] and Howard and Mullings [1996], both freshwater and saltwater flow along the fractures and conduits to reach the mixing zone, or the zone where these fluids are superposed in a dynamic equilibrium because of their differences in density ; but the dynamics of such a saltwater intrusion are generally unknown and not represented in models. Such coastal karstic systems are intensely studied at this moment in the Mediterranean region [Gilli, 1999], both as above sea-level or underwater springs, for potential use in areas where this resource would be of great value for economic development.
This article discusses the freshwater-saltwater exchange mechanisms in the karstic aquifer of the Almyros of Heraklion aquifer (Crete) and explains the salinity variations observed in the spring. First, the general hydrogeology of the study site is described, then the functioning of the spring : a main conduit drains the freshwater over several kilometres and passes at depth through a zone where seawater is naturally present. The matrix-conduit exchanges are the result of pressure differences between the two media. These processes are represented in a mathematical model that confirms their relevance.

General hydrogeology of the studied site. - The karstic coastal system of the Almyros of Heraklion (Crete) covers 300 km² in the Ida massif whose borders are a main detachment fault, and the Sea of Crete in the north, the Psiloritis massif (highest summit at 2,456 m) in the south and west, and the collapsed basin of Heraklion filled in by mainly neogeneous marl sediments in the east. The watershed basin consists of the two lower units of characteristic overthrust formations of Crete (fig.1 ) : the Cretaceous Plattenkalk and the Cretaceous Tripolitza limestones. The two limestone formations are locally separated by interbedded flysch or phyllade units that form an impervious layer [Bonneau et al., 1977 ; Fassoulas, 1999] and may lead to different flow behaviour within the two karstic formations. Neo-tectonic activity has dissected these formations with large faults and fractures. The present-day climate in Crete is of Mediterranean mountain type, with heavy rain storms and snow on the summits in winter. Rainfall is unevenly distributed over the year, with 80 % of the annual total between October and March and a year-to-year average of 1,370 mm. The flow rate of the spring is high during the whole hydrologic cycle, with a minimum in summer on the order of 3 m³.s^-1 and peak flow in winter reaching up to 40 m³.s^-1. The water is brackish during low flow, up to a chloride content of 6 g.l^-1, i.e. 23 % of seawater, but it is fresh during floods, when the flow rate exceeds 15 m³.s^-1. During the 1999-2000 and 2000-2001 hydrologic cycles, the water was fresh during 14 and 31 days, respectively. The water temperature is high and varies very little during the year (see table l).
In the areas of Kéri and Tilissos (fig. 1), immediately south of the spring, the city of Heraklion extracts water from the karstic system through a series of 15 wells with depth reaching 50 to 100 m below sea level. Initially, when the wells were drilled, the water was fresh, but nowadays the salinity rises progressively, but unequally from well to well (fig. 2). The relatively constant temperatures and salinities of the wells, during the hydrological cycle, contrast with the large salinity variations at the spring (fig. 2 and table I). They show that the karstic system is complex and comprises different compartments, where each aquifer unit reacts to its individual pressures (pumping, rainfall) according to its own hydrodynamic characteristics [Arfib et al., 2000]. The Almyros spring seems disconnected from the surrounding aquifer and behaves differently from that which feeds the wells (upper Tripolitza limestone). It is recharged by fresh water from the mountains, which descends to depths where it probably acquires its salinity. The spring would thus be the largest resource of the area, if it was possible to prevent its pollution by seawater. A general functioning sketch is proposed (fig. 3), which includes the different geological units of interest.

Identification of the functioning of the Almyros spring through monitoring of physical and chemical parameters. - The functioning of the aquifer system of the Almyros spring was analysed by monitoring, over two hydrological cycles, the level of the spring, the discharge, the electric conductivity and the temperature recorded at a 30 min time interval. In the centre of the watershed basin, a meteorological station at an altitude of 800 m measures and records at a 30 min time interval the air temperature, rainfall, relative humidity, wind velocity and direction ; moreover, an automatic rain gauge is installed in the northern part of the basin at an altitude of 500 m.
The winter floods follow the rhythm of the rainfall with strong flow-rate variations. In contrast, the summer and autumn are long periods of drought (fig. 7). The flow rate increases a few hours after each rainfall event ; the water salinity decreases in inverse proportion to the flow rate a few hours to a few days later. Observations showed that the water volume discharged at the Almyros spring between the beginning of the flow rate increase and the beginning of the salinity decrease is quite constant, around 770,000 m³ (fig. 4) for any value of the flow rate, of the salinity and also of the initial or final rainfall rates. To determine this constant volume was of the upmost importance when analyzing the functioning of the Almyros spring. The lag illustrates the differences between the pressure wave that moves almost instantaneously through the karst conduit and causes an immediate flow rate increase after rainfall and the movement of the water molecules (transfer of matter) that arrives with a time lag proportionate to the length of the travel distance. The variation of the salinity with the flow rate acts as a tracer and gives a direct indication of the distance between the outlet and the seawater entrance point into the conduit. In the case of the Almyros, the constant volume of expelled water indicates that sea-water intrusion occurs in a portion of the conduit situated several kilometres away from the spring (table Il), probably inland, with no subsequent sideways exchange in the part of the gallery leading up to the spring.
As the lag between the flow rate and the salinity recorded at the spring is constant, one can correct the salinity value by taking, at each time step, with a given flow rate, the salinity value measured after the expulsion of 770,000 m³ at the spring, which transforms the output of the system so as to put the pressure waves and the matter transfer in phase [Arfib, 2001]. After this correction, the saline flux at the spring, equal to the flow rate multiplied by the corrected salinity, indicates the amount of sea-water in the total flow. This flux varies in inverse proportion to the total flow rate in the high-flow period and the beginning of the low-flow period, thereby demonstrating that the salinity decrease in the spring is not simply a dilution effect (fig. 5). The relationship that exists between flow rate and corrected salinity provides the additional information needed to build the conceptual model of the functioning of the part of the Almyros of Heraklion aquifer that communicates with the spring. Freshwater from the Psiloritis mountains feeds the Almyros spring. It circulates through a main karst conduit that descends deep into the aquifer and crosses a zone naturally invaded by seawater several kilometers from the spring. The seawater enters the conduit and the resulting brackish water is then transported to the spring without any further change in salinity. The conduit-matrix and matrix-conduit exchanges are governed by the head differences in the two media.

Mathematical modelling of seawater intrusion into a karst conduit

Method. - The functioning pattern exposed above shows that such a system cannot be treated as an equivalent porous medium and highlights the influence of heterogeneous structures such as karst conduits on the quantity and quality of water resources. Our model is called SWIKAC (Salt Water Intrusion in Karst Conduits), written in Matlab®. It is a 1 D mixing-cell type model with an explicit finite-difference calculation. This numerical method has already been used to simulate flow and transport in porous [e.g. Bajracharya and Barry, 1994 ; Van Ommen, 1985] and karst media [e.g. Bauer et al., 1999 ; Liedl and Sauter, 1998 ; Tezcan, 1998]. It reduces the aquifer to a single circular conduit surrounded by a matrix equivalent to a homogeneous porous medium where pressure and salinity conditions are in relation with sea-water. The conduit is fed by freshwater at its upstream end and seawater penetrates through its walls over the length L (fig. 6) at a rate given by an equation based on the Dupuit-Forchheimer solution and the method of images. The model calculates, in each mesh of the conduit and at each time step, the head in conditions of turbulent flow with the Darcy-Weisbach equation. The head loss coefficient l is calculated by Louis' formula for turbulent flow of non-parallel liquid streams [Jeannin, 2001 ; Jeannin and Marechal, 1995].
The fitting of the model is intended to simulate the chloride concentration at the spring for a given matrix permeability (K), depth (P) and conduit diameter (D) while varying its length (L) and its relative roughness (kr). The spring flow rates are the measured ones ; at present, the model is not meant to predict the flow rate of the spring but only to explain its salinity variations.

Results and discussion. - The simulations of chloride concentrations were made in the period from September 1999 to May 2001. The depth of the horizontal conduit where matrix-conduit exchanges occur was tested down to 800 m below sea level. The diameter of the conduit varied between 10 and 20 m, which is larger than that observed by divers close to the spring but plausible for the seawater intrusion zone. The average hydraulic conductivity of the equivalent continuous matrix was estimated at l0^-4 m/s. A higher value (l0^-3 m/s) was tested and found to be possible since the fractured limestone in the intrusion zone may locally be more permeable but a smaller value (10^-5 m/s) produces an unrealistic length (L) of the saline intrusion zone (over 15 km).
For each combination of hydraulic conductivity, diameter and depth there is one set of L (length) and kr (relative roughness) calibration parameters. All combinations for a depth of 400 m or more produce practically equivalent results, close to the measured values. When the depth of the conduit is less than 400 m, the simulated salinity is always too high. Figure 7 shows results for a depth of 500 m, a diameter of 15 m and a hydraulic conductivity of 10^-4 m/s. The length of the saltwater intrusion zone is then 1,320 m, 4,350 m away from the spring and the relative roughness coefficient is 1.1. All the simulations (table II) need a very high relative roughness coefficient which may be interpreted as an equivalent coefficient that takes into account the heavy head losses by friction and the variations of the conduit dimensions which, locally, cause great head losses. The model simulates very well the general shape of the salinity curve and the succession of high water levels in the Almyros spring but two periods are poorly described due to the simplicity of the model. They are (1) the period following strong freshwater floods, where the model does not account for the expulsion of freshwater outside the conduit and the return of this freshwater which dilutes the tail of the flood and (2) the end of the low-water period when the measured flux of chlorides falls unexpectedly (fig. 5), which might be explained by density stratification phenomena of freshwater-saltwater in the conduit (as observed in the karst gallery of Port-Miou near Cassis, France [Potié and Ricour, 1974]), an aspect that the model does not take into account.

Conclusions. - The good results produced by the model confirm the proposed functioning pattern of the spring. The regulation of the saline intrusion occurs over a limited area at depth, through the action of the pressure differences between the fractured limestone continuous matrix with its natural saline intrusion and a karst conduit carrying water that is first fresh then brackish up to the Almyros spring. The depth of the horizontal conduit is more than 400 m. An at tempt at raising the water level at the spring, with a concrete dam, made in 1987, which was also modelled, indicates that the real depth is around 500 m but the poor quality of these data requires new tests to be made before any firm conclusions on the exact depth of the conduit can be drawn.
The Almyros spring is a particularly favorable for observing the exchanges in the conduit network for which it is the direct outlet but it is not representative of the surrounding area. To sustainably manage the water in this region, it is essential to change the present working of the wells in order to limit the irreversible saline intrusion into the terrain of the upper aquifers. It seems possible to exploit the spring directly if the level of its outlet is raised. This would reduce the salinity in the spring to almost zero in all seasons by increasing the head in the conduit. In its present state of calibration, the model calculates a height on the order of 15 m for obtaining freshwater at the spring throughout the year, but real tests with the existing dam are needed to quantify any flow-rate losses or functional changes when there is continual overpressure in the system. The cause of the development of this karstic conduit at such a great depth could be the lowering of the sea level during the Messinian [Clauzon et al., 1996], or recent tectonic movements.