The study area covers the main springs of Western Crete which are used for domestic and irrigation purposes (Figure 1, Table 1). The total water demand in the study area (Prefecture of Chania) is 130 × 10⁶ m³ on an annual basis and approximately 85% of this amount is allocated to agriculture and 13% to domestic uses [22] . The most important springs in terms of discharge and water use are Agyia (domestic supply in the city of Chania), Stylos (large scale bottling factory), and Meskla (Table 1). No regular monitoring of the springs’ discharge occurs in the particular area but some indicative measurements from the Water Resources Department of the Regional Authority of Crete illustrate Agya, Almyros, Armenoi and Stylos as the most important springs in terms of discharge (from 63 to 80 × 10⁶ m³, Table 1). The geological substrate of the study area (Figure 1) comprises calcareous rocks (the White Mountains massif, approx. 42% of the study area), Flysch formations (western part, approx.: 27% of the area) and Neogene and Quaternary sediments (lowland, south part, approx.: 31% of the study area).All the springs under study are located within calcareous areas (limestones) with the exception of the

lowland springs (Koleni, Stylos and Armenoi) that are found within alluvial deposits, overlaying usually calcareous formations (Figure 2).

A series of water samples for isotopic analysis has been collected over the period 2005-2007, from 15 different

sites (11 springs and 4 rain samples, Figure 2) at a seasonal basis (10 sampling campaigns). The samples were collected in plastic bottles of 100 ml, following the zero-head air principle and transferred to Demokritus Research Center for isotopic analysis in δ¹⁸O and Tritium (TU). The equipment used for the analyses of δ¹⁸O samples is a SIRA Series III sotope Ratio Mass Spectrometer while for the Tritium a Packard-Tricarb LSA Liquid scintillator has been used. The Electrical Conductivity (in μS/cm) of the water samples were measured in-situ with a portable Hanna Inc. Instrument. Moreover, a water quality campaign in July 2006, took place in the same sites to identify the qualitative status of the springs and propose specific management measures.

The Oxygen isotopic composition of the water samples were the main focus of this effort, which is defined as delta (δ) values, expressed in the per mille deviation from the standard isotopic composition known as SMOW (Standard Mean Ocean Water, [15] [23] . The isotopic ratio of ¹⁸O/¹⁶O is defined in the following equation:

‰

If the value of δ is negative then the sample is “depleted” in the light O isotope (low amount of evaporation and high recharge altitude) while if it is positive the sample is “enriched” in comparison to the standard which means lower recharge altitude and more exposed to the evaporation effect.

All the isotopic values together with the additionally collected data (Tritium, EC, Water quality parameters) underwent a thorough statistical analysis including descriptive statistical estimations, hierarchical clustering and regression analysis. Particularly, the δO₁₈ values have been used to identify the groups of springs that have similar behavior and may be affected from the same aquifer. For this process the hierarchical clustering algorithm as well as a box plot analysis in SPSS software has been used. A pair-wise regression analysis followed between the stable isotope concentrations and the physicochemical parameters in order to identify common relationships and trends. Moreover, a correlation matrix has been also estimated to identify links between the major ions concentrations and the stable isotopes levels in all the springs. These results have been combined to interpret the interrelationships between sampling points and identify the common recharge areas. The rainfall samples collected from 4 different locations in Western Crete on January 2006 (single rainfall event), covering the local altitude natural variations, were used to estimate the relationship between altitude and δ¹⁸O values by using the least squared equation approach. The resulted equation has been then used to estimate the recharge altitude of the sampling sites. The average recharge zone altitude for each sampling site was estimated by using the following the following equation:

where y = the estimated average recharge zone altitude (m), a = the slope of the least squared regression line connecting rainfall δ¹⁸O values and altitude, x = the sampling points average δ¹⁸O value and b = the offset in y axis of the least squared regression line mentioned above. The particular recharge zone altitude estimations is an approximation of the reality since the available data to justify the accuracy of these estimations are limited but combined with other sources of information such as the geological and topographic substrates can provide the basis for the recharge zones identification.

The clustering exercise used the hierarchical clustering algorithm in R software to identify group of sites with similar behaviour as far as δ¹⁸O values are concerned. Regression analysis between δ¹⁸O and TU, Electrical Conductivity and Ca were used to validate the aforementioned groups. Then, the particularly outcome was combined with the recharge zone altitude estimations and the main geological properties of the area to identify common recharge areas and the potential limits of the respective aquifers.

A water quality survey was attempted in all the sites at July 2006 and the acquired water samples were analysed in the hydrochemistry laboratory of the Hellenic Centre for Marine Research. The examined parameters included: SiO₂, Ca²⁺, Mg²⁺, Na⁺, , and. The correlation matrix of these chemical analyses, including the same dates’ δ¹⁸O and Tritium values has been estimated to depict the potential relationships between them while an assessment of the water quality status followed to reassure the classification of springs in groups based on the δ¹⁸O values.

The potential interactions of the examined springs became apparent through the aforementioned approach and a qualitative assessment followed in all of the studied aquifers to present specific management and protection measures.

The results indicated that the monitored springs illustrate a large variation in their δ¹⁸O values both in seasonal and interannual basis (Table 2). Particularly, the highest average δ¹⁸O value was detected in the Agya spring with the value of −8.15‰ followed by the Armenoi spring (−8.10‰). The lowest δ¹⁸O value was observed in the Koleni spring (−4.81‰) followed by the spring named Argyroupoli with the value of −4.83‰ (Table 2). Furthermore, Argyroupoli and Almyros had a low standard deviation (0.46 and 0.47 respectively) indicating that the intrannual fluctuations are limited while Ag. Eirini and Fournes 2 had a high standard deviation (0.77 and 0.67 respectively) value showing that the δ¹⁸O values throughout the year were fluctuating significantly. Also, the skewness was calculated (Table 2) which describes the degree of asymmetry of δ¹⁸O values’ distribution around its mean [24] . The majority of the springs had a positive value of skewness and therefore most of their δ¹⁸O values were lower than their means. Fournes 2 was the only spring which its skewness was negative showing that most of its values were higher than the mean. Moreover, Ag. Eirini and Meskla (followed by Stylos and Armenoi) that present high StD and Skewness are probably affected more from snowmelt than the other sites. This is because snowmelt in high altitude has very low δ¹⁸O values while rainfall in moderate and low altitudes present relatively high δ¹⁸O values, especially in Crete. Thus, the aforementioned sampling sites undergo the impact of extremely low δ¹⁸O values more frequently than the other sites.

The seasonal fluctuation of δ¹⁸O values (Figure 2 and Figure 3) indicates that the highest values were observed in October 2006 and July 2006, while the lowest ones in January 2006 and November 2005. The springs such as Koleni, Almyros, Argyroupoli, Meskla, Agiya, followed more closely this particular seasonal pattern than the rest of the sampling sites. Armenoi and Stylos illustrated their highest values in April and November 2006 while their lowest values were the same of the other sites. This behavior could be related to the main water supply mechanism that changes throughout the year. Particularly, for the sampling sites that have recharge zones affected by snowmelt mostly the spring and summer periods are the ones with the highest δ¹⁸O values (Armenoi

and Stylos) due to the characteristic δ¹⁸O values of snowmelt and the extended travel time that can fluctuate from 2 - 5 months. The sites that are both affected by snowmelt and rainfall have higher seasonal fluctuations but always in the lower levels of δ¹⁸O values (Meskla, Agya, Ag. Eirini) while the ones that are located in lower altitudes and are mostly affected by rainfall have significant seasonal fluctuations at a lower δ¹⁸O level (Almyros, Koleni, Argyroupoli).

The similarity index of the cluster analysis indicated that Stylos and Armenoi are very close not only in absolute values but also in their seasonal and interannual fluctuations. The same stands for Ag. Eirini and Sempronas 2 while Meskla and Fournes 2 illustrate the most dissimilar behavior in relation to the all other sites.

Thus, according to the seasonal fluctuations and the cluster analysis two main group of springs can be observed from the data; the first one includes Agiya, Amenoi, Stylos and Meskla while the second one Koleni, Argyroupoli, Almyros and Potistiri (Figure 4).

The Box and Whisker plot of all the data (Figure 5) shows the substantially different behavior of the various sampling sites as far as δ¹⁸O is concerned. The majority of springs did not have a normal distribution apart from the springs named Ag. Eirini and Fournes 2 which presented a distribution close to normal. The highest median of δΟ₁₈ values appeared in Stulos spring and the lowest one in Koleni (Figure 5). The maximum range of the δ¹⁸Ο values was found in Fournes 2 while the minimum one in Almyros. The clustering groups of sampling sites described above are also accredited by the Box and Whisker plot since Stulos and Armenoi have very similar distribution of values while the same stands for Almyros, Argyroupoli and Potistiri, as well as for Agyia and Meskla and to a lesser extend for Ag. Eirini and Sempronas (Figure 5).

The δ¹⁸O values of the sampling sites have been correlated with other measured parameters such as Electrical Conductivity (EC), Tritium and Ca to identify potential relationships between these parameters and distinguish group of sites with similarities in their behavior.

Regarding EC, Almyros spring had the greatest values, ranging from 2909 μS/cm to 2180 μS/cm which was expected due to its proximity to the sea and the dominant geological substrate that includes limestones with fluctuating amounts of gypsum (Figure 6). The average conductivity in the rest sampling sites fluctuated between 633.40 μS/cm and 252.20 μS/cm which are typical values for karstic groundwater with relatively limited residence time.

The least squared regression between EC and δ¹⁸O indicates a quite low correlation coefficient (0.364) between the two parameters which is mainly due to Almyros site that appears as an outlier in the analysis. According to their scatterplot (Figure 6), there is a positive linear relationship between conductivity and δ¹⁸O which groups the sites in a similar way as the previously mentioned analyses.

The Tritium values show a slightly positive linear relationship with the δ¹⁸O values but again the correlation coefficient is low (0.1531) showing that this relationship is not statistically significant (Figure 7). The greatest range of Tritium values was observed in Fournes 2 with the maximum and minimum ones to be 16.7 TU and 4.2 TU respectively (Figure 8). The average Tritium values of the rest of the springs fluctuated between 5.66 TU and 2.76 TU while Sempronas 2 had the lowest average Tritium value (2.79 TU). In terms of groundwater age, based on the Tritium concentrations, Ag. Eirini and Sempronas have relatively low values which indicates a mixture of recent and old (prior to 1952) water [15] while the rest of the sampling sites have Tritium values from 4 to 10.5 TU which indicates modern water. The regression analysis between Tritium and δ¹⁸O values (Figure 7) indicates a clear grouping of the samping sites (group 1: Stylos, Armenoi, Agyia, Meskla, group 2: Almyros, Potistiri, Koleni and Argyroupoli, group 3: Ag. Eirini and Sempronas).

The relationship between Ca and δ¹⁸O values in the particular sampling sites were weak but reassured some of the aforementioned groups (Figure 9). Almyrosand Koleni springs had the highest Ca values (107 mg/l and 61mg/l respectively) while Sempronas 2 and Ag. Eirini had the lowest ones (20 mg/l and 35 mg/l respectively). Stulos, Armenoi and Agyia presented very similar Ca values while Argyroupoli and Koleni had almost identical

concentrations (Figure 9). The different concentrations of Ca in the studied springs express the high variation of the geological formations and the groundwater residence times in the particular area. Springs with relatively low Ca concentrations such as Sempronas and Ag. Eirini indicate probably aquifers with limited calcareous content and/or short travel times. Proximity in the sea also affects the Ca concentrations significantly, such as the case of Almyros spring.

The rainfall sampling sites and their δ¹⁸O values have been used to estimate the relationship between δ¹⁸O and recharge altitude in the study area (Figure 10). The coefficient of determination for the identified relationship was very high (0.84) and therefore the associated regression equation could be used to estimate the potential, average recharge zone altitude for all the sampling sites (Table 3).

Stulos spring had the greatest average recharge altitude (1040 m) followed by Armenoi and Agiya springs

(1020 m). Koleni and Argyroupoli had the lowest average recharge altitude (460 and 480 m respectively), followed by Almyros (560 m) and Potistiri (570 m, Table 3). The potential seasonal fluctuations in the recharge zone altitudes of each site indicate that Ag. Eirini and Meskla have the highest values (range of 610 and 620 m respectively), while Fournes and Potistiri have the lower variations (350 and 380 m respectively, Table 3).

The scatterplot of recharge altitude against δ¹⁸O values (Figure 11), illustrates again the strong negative linear relationship between these parameters while re-validates the grouping effects mentioned before in the particular study.

By combining the sampling sites clustering and topographic information together with their recharge zones altitude and the dominant geological conditions, an effort has been made to delineate the recharge zone boundaries for each site as well as the common aquifers shared by each group of sites (Figure 12). These recharge zones present only the potential areas of water supply for each group of springs, based on the fluctuations of the average recharge altitudes for each sample and the dominant geological substrate. Nevertheless, more work needs to be done to validate and possibly amend these areas by using more isotopic data and reference values to optimize and/or accredit the relationship between recharge altitude and isotopic values.