Water source dynamics in a glacierized alpine river basin (Taillon-Gabiétous, French Pyrénées)



[1] Currently, there is minimal information relating to temporal variability of water source contributions in alpine glacierized basins or the influence of glacier meltwater in a basin-wide context. This study adopts an end-member mixing approach to understand basin-scale water source dynamics in a French Pyrenean, alpine glacierized river system (Taillon-Gabiétous). Major ion and Si data were collected for snow, groundwater tributaries, and four mainstream sites during the 2002/2003 melt seasons. Three conceptual water sources were identified: “quick flow” (dilute, rapidly routed meltwater), “distributed” (SO42− enriched, slow routed subglacial waters), and “groundwater” (Si-enriched groundwater). Water source contributions at nested spatial and temporal scales were determined using end-member mixing and uncertainty analysis. Changes in stream hydrochemistry indicated marked meltwater-groundwater mixing. Quick flow contributions typically decreased over the melt season; groundwater contributions were highest at the beginning of the melt seasons following recharge by snowmelt but also later in the 2002 melt season following prolonged precipitation. Overall, the results suggest an alternative alpine basin melt season hydrological progression compared with previous models (i.e., simple snowmelt to glacier melt to groundwater domination) and emphasize the need to understand water source dynamics to inform related water resource availability, water quality, and stream ecology studies within alpine basins.

1. Introduction

[2] Understanding natural variability in the magnitude, timing and duration of peak snowmelt, glacial ice melt, and groundwater contributions to alpine streamflow has important implications for basin hydrology and related disciplines [Malard et al., 1999]. Alpine basin water source dynamics are first-order controls on physicochemical habitat and ecological community dynamics [Brown et al., 2003]. They also govern runoff availability for hydroelectric power generation and irrigation downstream [Viviroli and Weingartner, 2004]. Therefore it is important that processes influencing present-day alpine basin water sourcing are fully understood. However, currently there is minimal understanding of temporal variability of water source contributions in alpine glacierized basins [Malard et al., 1999], and few studies have examined the influence of glacier meltwater in a basin-wide context [Cooper et al., 2002; Hodson et al., 2002].

[3] Alpine runoff hydrochemistry reflects solute acquisition from reactive mineral phases in subglacial and recently deglacierized environments, producing waters typically enriched in Ca2+, HCO3 and SO42− [Tranter et al., 2002]. High crustal K+: dissolved Si ratios are also typical because K-feldspars are sensitive to high disaggregation rates under glacier-driven physical weathering, and short residence times in higher rock-water contact environments preclude stoichiometric ratios of weathering products being produced in solution [e.g., Hodson et al., 2002]. In older moraines, silicate dissolution may become more stoichiometric (hence lower K:Si ratios) as Si mineral weathering proceeds further for longer residence time waters [e.g., Cooper et al., 2002]. Thus, in a qualitative sense, variability in bulk meltwater solute concentrations can reflect changes in water sourcing to proglacial streams [e.g., Tranter et al., 1996]. Solute acquisition processes offer a potential means of identifying basin-scale water sources for use in more quantitative chemical hydrograph separations. However, this potential has yet to be realized for glacierized environments, despite the wide use of hydrochemical separation methods to elucidate temperate river basin hydrological functioning [Hooper et al., 1990; Wade et al., 1999; Soulsby et al., 2003] and recent successful application in alpine basins influenced by snowmelt but with no glacier [Liu et al., 2004]. This unfulfilled potential reflects a failure to identify conservative properties [Sharp et al., 1995a] with which to fingerprint water routing through glacial river systems.

[4] The French Pyrénées, located at the southern limit of contemporary European glaciation, provide an excellent location to investigate water source dynamics within a glacierized basin and so address the research gaps identified above. The Taillon-Gabiétous basin contains small cirque glaciers and seasonal snowpacks, which are highly dynamic during the summer melt season [Hannah et al., 1999a, 2000a]. The basin also contains a hillslope groundwater system in the wider proglacial environment. Therefore the aims of this study were (1) to determine the principal controls upon bulk meltwater hydrochemistry in four nested subbasins (glacierized cover 2–41%); (2) to characterize conceptual water sources based on conservative hydrochemical signatures; and (3) to quantify spatial and temporal variability in the relative contributions of conceptual water sources to streamflow using end-member mixing.

2. Study Area

[5] Field observations were made within the Taillon-Gabiétous basin, Cirque de Gavarnie, French Pyrénées (43°6′N, 0°10′W; Figure 1) between 26 June (day 177) and 3 September (day 246), 2002 and 2003. Hereafter, dates are referred to using calendar day and time in hours Greenwich Mean Time (GMT). The study area is described by Hannah et al. [1999b, 2000b] and Smith et al. [2001]. Briefly, the basin covers 6.4 km2 and ranges from 3144 m to 1880 m at the most downstream site. The basin is characterized by steep slopes (30°–70°) and glacially overdeepened subbasins containing the Taillon (0.17 km2) and Gabiétous (0.09 km2) Glaciers. The upper basin (Pic du Taillon/Pic des Gabiétous) drains Marboré sandstone interspersed with Cenomanian/Turonian limestone outcrops, and sandy limestone of the Santonien and Conacien series (Figure 1). The lower basin (Vallée des Pouey Aspé) is a wide, gentle gradient valley composed mainly of carboniferous shale and Cenomanian/Turonian limestone. A karst system under Les Gabiétous Massif feeds several small springs in the lower basin [Parc National des Pyrénées, 1991].

Figure 1.

(top) Map of the Taillon-Gabiétous basin showing locations of the four main sampling locations and groundwater streams. (bottom) Geological map of the Taillon-Gabiétous basin and the Cirque de Gavarnie [after Flachére, 1977].

3. Methods

3.1. Field Sampling

[6] Water samples were collected weekly on days with no concurrent precipitation for four nested subbasins. An upper site was located ∼50 m downstream of the Taillon Glacier (Figure 1; glacierized area 41%). Sites A and C were on the Taillon Glacier stream, upstream and downstream, respectively, of the confluence with the Tourettes stream (site B). Glacierized area for sites A, B and C was 6, 2 and 4%, respectively. Samples were collected at low (0600–0900 h) and high flow (1400–1600 h). Because of logistical constraints, water samples were only collected from site A after day 217, 2002. Monthly basin wide sampling of seven to ten hillslope streams and karstic springs (Figure 1) was undertaken throughout the Vallée des Pouey Aspé.

[7] Samples were collected manually and filtered immediately to avoid postsampling changes in water chemistry [e.g., Brown et al., 1994]. For each sample, 30 mL of stream water was passed through a 0.45 μm Whatman cellulose-nitrate filter and discarded. Bottles (60 mL HDPE) were rinsed with 10 mL of filtrate three times before 55 mL was collected, leaving space for expansion during freeze storage. Electrical conductivity (EC) was measured at sites A–C using Campbell Scientific 247 temperature-EC probes. All continuous monitoring sensors were scanned every 10 s, and 15 min averages stored on Campbell CR10X/21X data loggers. Spot measures of EC and pH were made with a WPA CM25 meter and Jenway 3150 meter, respectively. Stream stage and discharge were continuously gauged at sites A, B and C (Figure 1; see Brown et al. [2005] for full details). Discharge data from site A were also used for the upper site because it was logistically impractical to regularly maintain a stream gauge and develop a rating curve here given the human effort required in relation to the altitudinal distance between the Vallée and the Taillon Glacier snout. However, previous studies [Hannah et al., 2000b; Smith et al., 2001] have shown flow regimes at the upper site and site A to be similar because no major tributaries enter between the two sites.

[8] At the beginning of each field season, snow samples were collected from snow pits dug along an altitudinal transect from the Taillon Glacier snout (2500 m) to the lower accumulation zone (2800 m). Snow pits were excavated using aluminum spades to 4 m, or until ice was encountered. Samples were taken from each snow layer by pressing 125 mL HDPE bottles into the snow. Samples were allowed to melt, then filtered as above and refrozen. In 2002, snow collection was repeated after one week to characterise any continuing snowpack ion elution. In both years the study only covered the latter period of snowmelt (i.e., initial snowmelt phase was not captured) so much of the solute load had been eluted prior to sampling.

3.2. Analytical Methods and Data Analysis

[9] Major ion concentrations were determined using a Dionex 4000i ion chromatograph fitted with an IonPac CS12 Cation or IonPac Fast Anion exchange column. Dissolved Si concentrations were determined using the molybdosilic acid method and bicarbonate concentrations estimated from charge-balance deficits [Hodson et al., 2000]. Precision errors for replicate midrange standards were <10% for all determinants and <5% for Ca2+, Mg2+, SO42−, Cl and Si similar to those of Hodson et al. [2002]. Nonmarine snowpack solute concentrations (nmssolute) were calculated thus:

equation image

where x is the ionic solute:Cl ratio for standard seawater [Drever, 1988] and snowsolute was estimated for streams using the mean solute:Cl ratio for each monitoring season. Crustal solute concentrations (prefix asterisk) were calculated following Sharp et al. [1995b]. Ratios of *K and dissolved Si concentrations were examined to qualitatively characterise silicate weathering by K-feldspars in the basin [Hodson et al., 2000].

[10] Estimates of total subglacial SO42− and Si concentrations were made from upper site samples following equations by Tranter and Raiswell [1991]. This method was considered appropriate because independent hydrological modeling studies show upper site meltwaters to be a mixture of rapidly routed snowmelt and ice melt, and meltwater routed more slowly through a subglacial distributed drainage system [Hannah and Gurnell, 2001]. Resource constraints, and national park and UNESCO World Heritage site restrictions, prevented any direct invasive techniques (e.g., drilling) to sample subglacial waters. Where estimated distributed system SO42− and Si concentrations were not significantly different in successive weeks, data were grouped to produce estimates for 2–4 week periods. SO42− and Si were used on account of their quasi-conservative behavior and because they often differ considerably between snow, ice and groundwater sources [Hodson et al., 2002]. SO42− acquisition from suspended sediment is minimal [Brown et al., 1994], while sulphide oxidation downstream of glaciers is believed to be impeded by a lack of reactive renewable mineral surfaces beyond high rock-water contact zones [Fairchild et al., 1994]. Silica was considered conservative due to slow reaction rates, and low diatom abundances in these streams [Brown et al., 2004b].

[11] This paper uses an approach which identifies conceptual water sources, aggregating hydrochemical variability of smaller-scale subsources and flow pathways [Wade et al., 1999] as a basis for end-member mixing analysis. To quantify conceptual water source contributions to streamflow at the four nested subbasin sites, chemical hydrograph separation was undertaken using a three-component end-member mixing Bayesian statistical approach [Brewer et al., 2005], which has been successfully applied by Soulsby et al. [2004] to temperate UK river basins but remains to be tested in glacierized basins. Water source proportions for site A prior to day 217, 2002 were estimated by multiple regressions with the other sites. The end-member mixing approach uses Markov chain Monte Carlo techniques in which end-members vary randomly as bivariate normal distributions. This provides uncertainty estimates, to account for measured spatial variability in conceptual water source end-member chemistries. Separations were conducted for individual years and subseasonal periods to account for temporal water source solute variability: a similar concept to the “event-specific” end-members of Soulsby et al. [2003].

4. Results

4.1. Meteorology, Stream Discharge, EC, and Snow Line

[12] Although discharge magnitude varied between sites, streams displayed broadly similar flows within years (Figures 2 and 3) . Mean discharge at site A was slightly greater in 2003 (0.09 m3 s−1 compared with 0.07 m3 s−1 in 2002); site B discharge was marginally higher in 2002 (0.07 m3 s−1 compared with 0.06 m3 s−1 in 2002) while site C showed the same mean discharge in both years (0.20 m3 s−1). Electrical conductivity was almost invariably greatest at site B and lowest at site A, with diurnal fluctuations at each site. In both field seasons, snow cover below 2400 m was limited to isolated patches that completely melted by approximately day 190. The transient snow line on north facing slopes retreated rapidly so that by day 197 (2002) and day 200 (2003), ice was exposed on the Taillon Glacier at ∼2600 m. The snow line retreated to a maximum altitude of ∼2710 m in 2002 and ∼2750 m in 2003. South facing slopes were devoid of snow cover throughout both monitoring periods [Brown et al., 2004a].

Figure 2.

Time series of SO42−, Si, stream discharge (Q), and electrical conductivity (EC) at (a) upper site, (b) site A, (c) site B, and (d) site C during the 2002 melt season.

Figure 3.

Time series of SO42−, Si, stream discharge (Q), and electrical conductivity (EC) at (a) upper site, (b) site A, (c) site B, and (d) site C during the 2003 melt season.

4.2. Hydrochemistry

[13] Ca2+ and Mg2+ were dominant cations, especially in samples collected from groundwater-fed streams (i.e., hillslope and site B; Table 1). Although snow samples had very low solute concentrations they were enriched in Na+ and K+ compared with stream water. HCO3 was the dominant anion in all samples (except old snow) and Cl concentrations were highest in snow samples. SO42− concentrations were generally low in supraglacial runoff and snow, particularly after one week of sampling in 2002 (i.e., old snow). Average SO42− concentrations were similar at all four main sites with little downstream change but concentrations fluctuated over time (Figures 2 and 3). Si concentrations were negligible in snow and supraglacial samples and lowest in stream at the upper site (Figures 2 and 3). High Si concentrations were evident for hillslope groundwater streams. Mean pH was generally consistent at all sites (7.7–7.9) but highest at the upper site (8.3).

Table 1. Descriptive Statistics for Major Ions, Silica, and pH for 2002 and 2003a
SiteCa2+, μeqL−1Mg2+, μeqL−1Na+, μeqL−1K+, μeqL−1NH4+, μeqL−1Cl, μeqL−1NO3, μeqL−1SO42−, μeqL−1HCO3, μeqL−1Si, mg L−1pH
  • a

    Mean values, standard deviations (in parentheses), and the number of samples (in italics) are provided. Snow (old) samples were collected in 2002, one week after initial snow collections to examine solute concentrations following elution.

Taillon580 (103), 41380 (127), 4124.5 (25.6), 416.5 (6.1), 413.33 (3.6), 4136.1 (53.8), 4117.0 (8.8), 41185.0 (70.9), 41757 (178), 410.24 (0.07), 418.3 (0.3), 39
A925 (123), 37430 (87), 3715.1 (7.48), 375.5 (4.2), 371.7 (2.3), 3717.2 (10.1), 3715.2 (5.6), 37187.4 (39.3), 371157 (167), 370.35 (0.16), 377.9 (0.3), 35
B1120 (262), 49370 (75), 4925.6 (6.7), 496.0 (4.2), 492.81 (11.9), 4922.6 (49.1), 4916.5 (5.7), 49186.4 (33.9), 491298 (296), 490.92 (0.17), 497.7 (0.7), 46
C922 (196), 51390 (85), 5118.2 (7.3), 514.8 (3.7), 516.3 (34.6), 5119.7 (19.4), 5117.0 (6.4), 51182.7 (35.6), 511122 (249), 510.52 (0.11), 517.9 (0.3), 49
Snow190 (180), 2723.6 (27.1), 27132.7 (115.3), 2726.4 (20.8), 27368 (342), 2778.1 (51.4), 2715.4 (13.3), 2739.4 (25.6), 27627 (485), 27(0.0), 27
Snow (old),42.2 (31.3), 122.6 (0.9), 1233.7 (7.1), 123.8 (1.6), 1238.1 (38.3), 12112.8 (58.2), 122.7 (1.8), 121.4 (1.2), 124.9 (81.7), 120.0 (0.0), 12
Supraglacial159 (132), 1142.2 (21.0), 118.2 (7.2), 112.6 (3.2), 111.0 (1.1), 1113.5 (13.0), 112.0 (2.4), 117.0 (7.0), 11190 (150), 110.1 (0.0), 117.5 (0.4), 8
Karstic1036 (214), 24495 (161), 2416.3 (9.6), 245.4 (3.3), 2414.3 (61.9), 2415.0 (8.0), 2419.8 (5.6), 24165.8 (52.8), 241368 (330), 240.67 (0.16), 247.7 (0.5), 19
Hillslope1087 (373), 31338 (115), 3148.0 (11.5), 318.1 (6.5), 313.9 (12.5), 3151.3 (136.4), 316.0 (6.8), 31176.5 (51.6), 311251 (437), 311.5 (0.4), 317.9 (0.4), 29

4.3. Crustal Mineral Weathering and Water Source Chemical Signatures

[14] Stream water had a dominant Ca2+/Mg2+/HCO3 chemistry with moderate SO42− concentrations (Table 1). Mean total SO42− concentration increased between snowpacks and the upper site suggesting sulphide weathering occurs beneath the Taillon Glacier (gypsum is not present in the local rock) demonstrating the potential for SO42− to track distributed water contributions to streamflow in the basin. Ratios of [*Na+ + *K+]:[*Ca2+ + *Mg2+] were all typically low (0.01–0.02 indicating carbonate dissolution dominance over silicate weathering). However, Si concentrations were greatest at site B and increased downstream from the glacier (Table 1). Furthermore, mean molar *K+/Si ratios were very high for the upper site (1.0) but decreased in magnitude and variability downstream from the Taillon Glacier particularly at site B (Figure 4). Because high Si concentrations were found only in hillslope groundwaters, they clearly indicate the potential for tracking groundwater contributions to bulk streamflow [e.g., Anderson et al., 2000].

Figure 4.

Box plots of *K:Si concentrations (molar concentrations) by site for both years' data (n = 37 for each site).

4.4. End-Member Mixing

4.4.1. End-Member Composition

[15] SO42− and Si concentrations produced triplots for both years' data in which “clouds” (mean ±1 SD) of potential end-member chemistries bounded stream samples (Figure 5). Exploratory analysis revealed multitracer end-member mixing applications [e.g., Hooper, 2003] to be less useful because most solutes were nonconservative [Brown et al., 1994]. Three end-members (conceptual water sources) were identified (Table 2) as (1) “Quick flow” waters derived from rapidly routed snowmelt and ice melt passing over the glacier surface and through a short residence time channelized glacier drainage system (low [SO42−] and [Si]); (2) “Distributed” system waters which had passed through a longer residence time subglacial drainage system acquiring a different chemical signature to quick flow (high [SO42−] and low [Si]), and (3) longer residence time hillslope “Groundwater” emerging from older moraines (intermediate [SO42−] and high [Si]). Although distributed and groundwater sources are flow pathways, in this study their distinct chemical signatures enable them to be viewed as conceptual water sources, so allowing reasonable use in end-member mixing. Maximum and minimum distributed SO42− and Si estimates proved consistent for both years, increasing confidence in estimates and the utility of the estimation method. No end-member analysis was undertaken for karstic streams as they were only sampled monthly but they appeared to be predominantly meltwater-fed (Figure 5) as earlier studies have suggested [Parc National des Pyrénées, 1991].

Figure 5.

End-member mixing plots of conceptual water sources for (a) 2002 and (b) 2003 based on SO42− and Si concentrations (error bars for end-members are ±1 SD).

Table 2. Mean Concentrations and Standard Deviations of SO42− and Si Concentrations in Quick Flow, Distributed, and Groundwater for Subseasonal Periods in 2002 and 2003
Time PeriodCalendar DaysSO42−, μeq L−1Si, mg L−1
Quick Flow
   1178–1875.0 ± 3.90.0 ± 0.0
   2197–2445.1 ± 5.70.0 ± 0.0
   1177–18521.2 ± 11.40.0 ± 0.0
   2192–24611.7 ± 3.80.0 ± 0.0
   1178–187228.9 ± 63.40.38 ± 0.06
   2197–202398.6 ± 40.80.45 ± 0.05
   3209–228277.5 ± 63.90.36 ± 0.05
   4237644.6 ± 82.60.47 ± 0.03
   5244438.7 ± 36.80.36 ± 0.02
   1177–185623.2 ± 101.00.46 ± 0.06
   2192327.4 ± 36.40.28 ± 0.02
   3199–241238.5 ± 30.40.22 ± 0.03
   4246413.6 ± 18.10.40 ± 0.03
   1178–83171.2 ± 38.01.38 ± 0.41
   2183–212167.7 ± 52.51.41 ± 0.44
   3212–244156.5 ± 53.51.32 ± 0.42
   1177–187187.1 ± 53.91.51 ± 0.23
   2187–217209.0 ± 52.21.87 ± 0.43
   3217–246192.7 ± 61.91.91 ± 0.32

4.4.2. Spatial and Temporal Water Source Dynamics

[16] Proportions of streamflow contributed by each water source varied considerably between the four main monitoring sites and between years (Table 3). At the upper site in 2002, quick flow (0.42 ± 0.21) and distributed (0.49 ± 0.24) sources were on average the most dominant. In 2003, mean and maximum distributed proportions at the upper site were much less than the previous year and quick flow was dominant (0.72 ± 0.26). Mean quick flow and distributed proportions were lower at site A than the upper site with an increase in groundwater contributions (Table 3). Site B was dominated by groundwater in both years but quick flow contributed a much greater mean proportion of streamflow in 2003 (0.32 ± 0.12) compared with 2002 (0.19 ± 0.13). At site C, the greatest mean water source proportions were from distributed sources in 2002 (0.43 ± 0.20) but quick flow was more dominant in 2003 (0.49 ± 0.20). Mean proportions from each source at site C were typically intermediate of those at sites A and B, reflecting the mix of Taillon and Tourettes stream waters. Mean error (uncertainty) associated with hydrograph separations for individual sites was similar in both years of the study (Table 3). Mean error was typically greatest for distributed and lowest for groundwater contributions.

Table 3. Descriptive Statistics for Quick Flow, Distributed, and Groundwater Proportions at the Four Main Monitoring Sites in 2002 and 2003a
UpperSite ASite BSite CUpperSite ASite BSite C
  • a

    Mean error calculated from 95 percentile range of water source contribution estimates for individual stream water samples.

Quick flow        
   Standard deviation0.
   Mean error (±)
   Standard deviation0.
   Mean error (±)
   Standard deviation0.
   Mean error (±)

[17] Clear temporal variations in mean proportions of water contributing to streamflow were evident at all sites within both study periods (Figures 6 and 7) . In 2002, quick flow was initially dominant at the three Taillon Glacier stream sites (upper, A and C) but showed a decrease over time and low contributions toward the end of monitoring (Figure 6). Distributed contributions were relatively high except at site B where groundwater was always dominant (Figure 6c). All sites showed marked increases in stream discharge after day 228 in 2002, following prolonged precipitation events. These discharge increases were greatest at sites B and C and were predominantly sourced from groundwater (Figures 6c and 6d).

Figure 6.

Time series of end-member mixing hydrograph separation based on mean proportions for (a) upper site, (b) site A, (c) site B, and (d) site C during the 2002 melt season. Two samples (low and high flow) are plotted for each date against measured stream discharge at the time of sampling.

Figure 7.

Time series of end-member mixing hydrograph separation based on mean proportions for (a) upper site, (b) site A, (c) site B, and (d) site C during the 2003 melt season. Two samples (low and high flow) are plotted for each date against measured stream discharge at the time of sampling.

[18] In 2003, water source contributions were clearly less variable than during 2002 (Figure 7). Although quick flow volumes were greatest at the beginning of monitoring and decreased over time (as in 2002), this source typically remained the greatest mean proportional contributor to flow throughout the 2003 monitoring period at the Taillon stream sites (Figures 7a, 7b, and 7d). Proportions of distributed water were relatively constant throughout the monitoring period at all sites. At site B, groundwater was always dominant (Figure 7c) but decreased consistently throughout the melt season similar to the other three sites, with no discernable recharge events (compare 2002).

5. Discussion

[19] The major controls on stream hydrochemistry in the Taillon-Gabiétous basin during the summer monitoring period were (1) diurnal fluctuations in the mixing ratios of dilute and concentrated waters in response to snow and ice ablation (see EC series in Figures 2 and 3), (2) rapid solute acquisition from the weathering of carbonates and pyrite, exemplified by high Ca2+, HCO3 and SO42− concentrations in meltwaters, and (3) enhanced dissolution of silicate minerals within older moraines and hillslopes. These hydrochemical characteristics are often reported in glacierized basins (see Introduction) but interestingly, within the Taillon-Gabiétous basin, they have a clear spatial structure which reflects water sources in different basin areas. Furthermore, hydrochemical mixing models demonstrated that marked temporal (intra-annual to interannual scale) variability accompanied the observed spatial patterns in water source contributions to streamflow.

[20] In both years, the relatively high north face valley snow line (>2400 m [cf. Hannah and McGregor, 1997]) and low snowpack solute concentrations suggested melt onset (and preferential elution) before the initiation of sampling. The dilute nature of rapidly routed quick flow during late June–September resulted in distinct diurnal fluctuations in stream hydrochemistry (SO42−, Si and EC) and conceptual water source contributions even below the predominantly groundwater-fed Tourettes stream (site C; 4% glacierized area). These patterns have only previously been shown for rivers fed by considerably larger glaciers [e.g., Fenn, 1987] and emphasize the importance of quick flow even in basins with reduced snowpack and glacier extent.

[21] Weathering was dominated by carbonate dissolution and sulphide oxidation in the Taillon glacier subbasin, reflecting an abundance of recently exposed mineral surfaces similar to other temperate alpine glaciers [e.g., Tranter et al., 2002]. A negative relationship between *SO42− and discharge at the upper site (r = −0.48, p < 0.01) showed that conceptual water sources which dominated at low flow (i.e., subglacial distributed) had acquired more sulphate. The dominance of distributed waters at low flow is common for larger temperate glaciers but these hydrochemical results confirm this for the much smaller Taillon Glacier and support runoff modeling experiments by Hannah and Gurnell [2001]. In older sediments with no active glacial/periglacial grinding (i.e., sites A–C), sulphides are more likely to have developed oxidation skins that greatly limit dissolution especially along rapid flow paths. This most likely accounts for the relative reduction in sulphide weathering with distance from the Taillon Glacier as observed in other glacierized basins [e.g., Anderson et al., 2000]. Therefore SO42− was ideal for tracing subglacial distributed water contributions in this alpine river system.

[22] At the basin-scale, low [*Na+ + *K+]/[*Ca2+ + *Mg2+] ratios confirmed silicate dissolution was much slower than carbonate weathering. High molar *K+:Si ratios for the Taillon Glacier subbasin indicated highly nonstoichiometric silicate mineral weathering compared with global means [e.g., Anderson et al., 1997; Hodson et al., 2000] indicative of rapid surface exchange reactions whereby H+ is substituted for *K+ in low residence time flow paths. Nevertheless, silicate weathering increased in importance as glacierized area decreased suggesting increased rock-water contact times in older moraines coupled with higher temperatures at lower altitude [Brown et al., 2005, 2006a]. Downstream decreases of *K+:Si ratios along the Taillon stream suggest mixing of meltwaters with longer residence time soil and groundwaters [Hodson et al., 2002]. Overall, low *K+:Si ratios and high Si concentrations show silicate weathering is greatest in the Vallée where hillslope aquifers are the main water source. Therefore Si concentrations were a useful tracer of groundwater in this alpine basin.

[23] Water flow through subglacial and groundwater systems resulted in clear changes in water chemistry (i.e., increased SO42− in distributed, increased Si in groundwater). This enabled basin-wide water source dynamics to be assessed because these conceptual sources imparted clear chemical signatures, facilitating their inclusion as end-members in chemical mixing models [Beck et al., 1990]. Because different volumes of water moved through these systems at various times during the summer melt season, SO42− and Si concentrations fluctuated, suggesting periodic flushing (i.e., distributed became more similar to quick flow; Table 2) followed by higher concentrations when throughflow was lower. In the subglacial system, variable solute concentrations could also be influenced by glacier movement and subglacial channel migration [Fountain and Walder, 1998] leading to changes in water contact time with rock/glacial debris. However, this is unlikely given the Taillon Glacier's small size, thin ice mass, and low meltwater volumes and thus minimal reorganization of ice-walled conduits [Hannah and Gurnell, 2001]. Overall, the approach employed herein considered solute variability by conducting separate end-member mixing for subseasonal periods, enabling more successful hydrograph separations compared with earlier studies [e.g., Sharp et al., 1995a].

[24] End-member mixing demonstrated clear changes in water source contributions to streamflow downstream from the Taillon Glacier. These were driven by groundwater mixing with snowmelt and ice melt runoff, similar to Hodson et al. [2002] and Cooper et al. [2002]. Interestingly, groundwater contributed on average ∼25% of streamflow after only 1.5 km from the Taillon Glacier. Mountain basins have been regarded traditionally as predominantly surface water-fed systems but the results herein are akin to recent studies identifying the importance of groundwaters in upland temperate basins [Soulsby et al., 1998]. These downstream changes in conceptual water source contributions provide first quantitative estimates of groundwater inputs for alpine glacierized basins, which (as other studies indicate) play an important role influencing water quality as well as water quantity [Liu et al., 2004; Brown et al., 2005, 2006b].

[25] Intra-annual changes in streamflow were identified and linked to conceptual water source (quick flow, distributed, groundwater) dynamics. At the beginning of the monitoring period in both years, quick flow and groundwater dominated, particularly at sites A–C. Quick flow dominated runoff during the early melt season due to the wide spatial extent of snowpacks and their relatively quick melt rate at this low-latitude basin [Hannah et al., 2000a]. In addition, the early quick flow dominance may be due to the potentially rapid resumption of flow through the lower Taillon Glacier channelized drainage system during the early melt season [Hannah and Gurnell, 2001], given minimal over-winter closure of ice-walled conduits. The high proportion of groundwater and quick flow together suggested that aquifers had recently been recharged by spring/early summer valley snowpack melting [Flerchinger et al., 1992]. Although snow line retreat allows more rapid routing of quick flow to the proglacial stream, quick flow proportions declined over the melt season as snowpack size decreased. However, groundwater contributions also decreased noticeably over time further emphasizing the strong link between meltwater aquifer recharge and groundwater contributions to alpine streams [Liu et al., 2004]. The decline in groundwater contributions was only arrested once, following a prolonged precipitation episode toward the end of the 2002 melt season [Brown et al., 2005, 2006a].

[26] Intra-annual conceptual water source dynamics in the Taillon Glacier stream in 2002/03 did not conform to the predictable seasonal progression (i.e., early season snowmelt through midseason glacier contributions to late melt season groundwater domination) invoked for larger alpine river systems by Malard et al. [1999]. Although it is difficult to generalize the patterns and processes of water source contributions using data collected over only two melt seasons, results from this study suggested that: (1) only peak quick flow and distributed contributions followed the expected water source progression, (2) high quick flow contributions to streamflow were extended into early July, and (3) groundwater contributions were highest at the beginning of the melt season, following snowmelt recharge of aquifers as described above. Progressive recharge of groundwater is found throughout the summer melt season within the Val Roseg basin, Switzerland where the model was developed [Malard et al., 1999]. However, the Val Roseg basin has a large alluvial floodplain that is seasonally recharged by meltwaters flowing over valley bottom alluvium. Alluvial floodplain aquifers are absent from the Taillon basin; instead, snowpacks on south facing slopes recharge small hillslope aquifers. Earlier and rapid snowpack melting is also influenced by the Taillon basin's southerly latitude; thus hillslope groundwater reserves are recharged earlier than in more northerly alpine basins where snowpacks begin to recede later and more slowly. Consequently, in higher latitude basins, the period of groundwater recharge may be extended leading to peak groundwater contributions later in the melt season.

[27] The mixing analysis and hydrograph separation used herein assumed a simplified basin hydrological system in which water source contributions fit neatly within three conceptual water sources. This was a necessary method to produce a first model of the hydrology of this alpine basin and to address the aim of quantifying broad spatial and temporal variability in the relative contributions of basin-scale water sources. In reality, the three end-members consisted of a range of subsources and flow pathways. For example, the hydrochemistry of the hillslope groundwater system varied between streams, reflecting small-scale variability in geological units, water flow pathways and water residence times. However, although not explicitly defined as individual sources, this spatial variability was reflected in the clouds (mean±SD) of potential end-member chemistries and contributes to the uncertainty estimates associated with the model output. Nevertheless, there exists clear potential for future studies to examine spatial and temporal variability in the contribution of these “sub-end-members” in alpine and other basins using multicomponent end-member mixing.

6. Conclusions

[28] This paper has provided new perspectives upon the hydrological functioning of alpine glacierized basins, by determining the principal controls upon bulk meltwater hydrochemistry as a basis for characterizing conceptual water sources and to quantify their spatial and temporal variability. In addition, the study also presented further independent tracer-based information on the hydrological functioning of the Taillon-Gabiétous basin compared with larger alpine and Arctic systems. Although chemical weathering processes and solute provenance in the Taillon-Gabiétous basin appear to operate in a similar manner to most temperate glacierized basins, conceptual water source dynamics appear less predictable. Intra-annual variations in quick flow, distributed and groundwater proportions in streams demonstrate that simple models of water source variations across the melt season (i.e., early snowmelt, midseason glacier melt, and late groundwater domination) should not be considered as a generic framework, particularly for basins such as the Taillon-Gabiétous where a significant and responsive groundwater contribution is found close to the glacier. Downstream changes in stream hydrochemistry were strongly influenced by hillslope tributary inputs, supporting hypotheses that groundwaters sources have an important influence on mountain basin hydrology. These findings highlight the need for a better understanding of water source interactions in basins traditionally regarded as being predominantly surface water fed. Further knowledge of the patterns and processes controlling water source mixing and associated changes in water quality has applications for understanding contemporary alpine stream ecosystem structure and function and for predicting water source dynamics under scenarios of future climate change [Brown et al., 2003]. More widely, accurately understanding water source dynamics in alpine basins has implications for managing water availability for hydropower and irrigation schemes. The approach detailed in this paper offers a straightforward but effective means of modeling water source contributions within alpine river systems because a conceptual water source approach avoids potential complications resulting from small-scale variability in processes affecting solute concentrations.


[29] This research was funded by a Natural Environment Research Council (NERC) studentship (NER/S/A/2001/05984) to Lee Brown while at the University of Birmingham. Mark Brewer's contribution was funded by the Scottish Executive Environmental and Rural Affairs Department (SEERAD). We thank the Parc National des Pyrénées for permission to undertake fieldwork. Rossa Donovan, Mark Ledger, Debbie Snook, Johannes Steiger, and several University of Birmingham undergraduates provided much appreciated field support. Matt Tefler and Maureen Lamb analyzed samples for Si. Three anonymous reviewers provided helpful comments on an earlier version of the manuscript.