Contrasting physical and chemical conditions of two rock glacier springs

Rock glaciers are increasingly influencing the hydrology and water chemistry of Alpine catchments. During three consecutive summers (2017–2019), we monitored by recording probes and fortnightly/monthly field campaigns the physical and chemical conditions of two rock glacier springs (ZRG, SRG) in the Zay and Solda/Sulden catchments (Eastern Italian Alps). The springs have contrasting hydrological conditions with ZRG emerging with evident ponding (pond‐like), and SRG being a typical high‐elevation seep (stream‐like). Water temperature was constantly low (mean 1.2°C, standard deviation 0.1°C) at both springs. Concentrations of major ions (dominated by SO42−, HCO3−, Ca2+ and Mg2+) and trace elements (As, Sr, Ba, U, Rb) increased, and water became more enriched in heavy stable isotopes (δ18O, δ2H) towards autumn. This solute and isotopic enrichment had an asymptotic trend at SRG, and a unimodal pattern at ZRG, where peaks occurred 60–80 days after the snowmelt end. Wavelet analysis of electrical conductivity (EC) and water temperature records revealed daily cycles only at SRG, and significant weekly/biweekly fluctuations at both springs attributable to oscillations of meteorological conditions. Several rainfall events triggered a transient (0.5–2 h) EC drop (of 5–240 μS cm−1) and water temperature rise (of 0.2–1.4°C) at SRG (dilution and warming), whereas only intense rainfall events occasionally increased EC (by 15–85 μS cm−1) at ZRG (solute enrichment and thermal buffering), with a long‐lasting effect (6–48 h). Building on previous research, we suggest that rock glacier springs with differing flow conditions, that is, stream‐like and pond‐like, have contrasting fluctuations of water parameters at different timescales. Thus, for pond‐like springs, peaks of EC/solute concentrations might indicate a seasonal window of major permafrost thaw. Our quantitative description of the hydrochemical seasonality in rock glacier outflows and the physical and chemical response to precipitation events provides relevant information for water management in mountain areas under climate change.

, HCO 3 À , Ca 2+ and Mg 2+ ) and trace elements (As, Sr, Ba, U, Rb) increased, and water became more enriched in heavy stable isotopes (δ 18 O, δ 2 H) towards autumn. This solute and isotopic enrichment had an asymptotic trend at SRG, and a unimodal pattern at ZRG, where peaks occurred 60-80 days after the snowmelt end. Wavelet analysis of electrical conductivity (EC) and water temperature records revealed daily cycles only at SRG, and significant weekly/biweekly fluctuations at both springs attributable to oscillations of meteorological conditions. Several rainfall events triggered a transient (0.5-2 h) EC drop (of 5-240 μS cm À1 ) and water temperature rise (of 0.2-1.4 C) at SRG (dilution and warming), whereas only intense rainfall events occasionally increased EC (by 15-85 μS cm À1 ) at ZRG (solute enrichment and thermal buffering), with a long-lasting effect (6-48 h). Building on previous research, we suggest that rock glacier springs with differing flow conditions, that is, stream-like and pond-like, have contrasting fluctuations of water parameters at different timescales. Thus, for pond-like springs, peaks of EC/solute concentrations might indicate a seasonal window of major permafrost thaw. Our quantitative description of the hydrochemical seasonality in rock glacier outflows and the physical and chemical response to precipitation events provides relevant information for water management in mountain areas under climate change. [IPCC], 2019) is paralleled by an increased influence from paraglacial and periglacial landforms (Brighenti, Tolotti, Bruno, Engel, et al., 2019;Haeberli et al., 2016). In particular, rock glaciers (i.e., creeping bodies of rock fragments providing evidence of mountain permafrost) have been addressed as significant water reservoirs at a global scale (Jones et al., 2018) as their subsurface ice melts much more slowly than that of glaciers (Haeberli et al., 2016). Their hydrological importance is also promoted by an increasing storage capacity made available by the loss of internal ice (Jones et al., 2019;Wagner et al., 2016). In fact, the internal structure of rock glaciers influences their hydrological behaviour. Growing evidence suggests the importance of a fine-grained basal layer with low hydraulic conductivity, which constitutes a groundwater storage system in these landforms. This sub-permafrost layer exerts a major control on base flow conditions, with fractures and depressions occurring in the basal bedrock only playing a minor role (Jones et al., 2019;Wagner et al., 2020). Most of the rainwater is quickly exported from rock glaciers across lateral flows in the suprapermafrost layer, which is made of coarse blocky materials with high hydraulic conductivity. The presence of fractures and ice-free areas in the ice-sediment matrix allows some water to cross the intrapermafrost zone, enhancing the recharge of the sub-permafrost aquifer (Wagner et al., 2020). This mixed water contribution can support surficial waters emerging as lakes, ponds, or streams at the rock glacier forefields.
The seasonal snowmelt is a key hydrological driver of rock glacier springs. In fact, their discharge is higher during early summer, and decreases towards autumn as the snow on the rock glacier and its catchment progressively melts away Krainer & Mostler, 2002). During late summer/autumn, only a small fraction of discharge (<5%, Krainer et al., 2011) can be sustained by internal ice melt Jones et al., 2019), which strongly influences the hydrochemistry of rock glacier springs.
Long-term studies on headwaters fed by rock glaciers attributed the increase of electrical conductivity (EC) and concentrations of major ions and trace elements observed in the last decades, to the progressive permafrost thawing Steingruber et al., 2020;Thies et al., 2007). The seasonal timing of solute export from rock glaciers is particularly important. In fact, high concentrations of metals and metalloids have been observed in rock glacier springs, often exceeding drinking water and environmental quality standards , and affecting the water chemistry further downstream along the river network (Brighenti, Tolotti, Bruno, Engel, et al., 2019).
Solute concentrations typically increase during late summer because the contribution of diluting snowmelt decreases, and those of reacted groundwater and permafrost ice melt increase (Caine, 2010;Krainer et al., 2015;Munroe, 2018;Williams et al., 2006). Different studies revealed contrasting patterns of solute increase in rock glacier springs, reporting either an asymptotic behaviour of EC/solute concentrations with the plateau corresponding to autumn and winter (Harrington et al., 2018;Krainer et al., 2007Krainer et al., , 2015, or a unimodal trend with late summer peaks . However, the drivers of these contrasting seasonal trends have yet to be investigated. Even the influence of precipitation on rock glacier hydrochemistry is still being debated.  found intense rainfall events caused solute-enrichment in a rock glacier pond whereas most studies reported a dilution effect from rainfall with transient solute-depletion in rock glacier streams (Berger et al., 2004;Harrington et al., 2018;Krainer et al., 2007;Krainer & Mostler, 2002).
Rock glaciers are efficient thermal buffers, and for this reason they represent potential climate refugia for cold-adapted terrestrial and aquatic organisms under global warming .
However, quantitative analyses aimed at understanding the main parameters driving intensity, duration, and temporal patterns of the response of T water and solute concentrations to precipitation events are still lacking. Rainfall events can also disrupt the diel fluctuations of water parameters in rock glacier springs. These oscillations typically occur during the seasonal snowmelt, and smooth down as summer progresses (Berger et al., 2004;Krainer et al., 2007;Krainer & Mostler, 2002). Periodic fluctuations of EC and T water might also occur over longer timescales (e.g., associated with switching warm and cold periods) but quantitative assessments of such oscillations have been never attempted.
In this study, we investigated the physical and chemical condi-

| STUDY AREA
We studied the permanent springs emerging from two tongue-shaped rock glaciers located in distinct subcatchments (Zay and Solda/Sulden) of the upper Solda/Sulden catchment, Eastern Italian Alps ( Figure 1; Table 1). The geographical, climatic, geological, and hydrological settings of the catchment are described in Engel et al. (2019) and Brighenti, Tolotti, Bruno, Engel, et al. (2019). Geologically, the area belongs to the Austroalpine domain, represented by a crystalline basement and its sedimentary cover (Montrasio et al., 2015; Table 1).
Although the ice abundance and distribution of these rock glaciers is unknown, the two landforms are classified as intact (i.e., containing ice) by Kofler et al. (2020), in agreement with the rock glacier inventory of the Autonomous Province of Bolzano/Bozen [APB] (2020a) which classified both landforms as active (i.e., intact and with motion).
Oversteepened fronts with unstable boulders, localized scouring, and sparse/absent vegetation suggest the occurrence of creeping activity for both rock glaciers, but no detailed studies exist on their kinematic behaviour.
The Zay spring (ZRG) is five meters wide, and it originates with an evident ponding from the south-western side of the Zay rock glacier terminus. In part, this rock glacier is hydrologically connected with the Ausserer Zay glacier, whose front is located~0.5 km away from the rock glacier upstream margin, and the glacier-fed stream flows parallel to the rock glacier along its northern margin (APB, 2020a; see Supporting Information S1).
The Solda spring (SRG) is~30 cm wide, and it originates as a typical mountain seep from the Solda rock glacier, which is composed of two merged bodies and is partially reshaped by an unsealed road at its terminus (APB, 2020a; Supporting Information S1). The spring lies 40 m downstream from the western side of the rock glacier front, and it partially drains moraine deposits (Supporting Information S1). To distinguish the two rock glacier outflows according to their distinct water flow conditions, we hereafter define them as pond-like (ZRG) and stream-like (SRG) springs.
Two automatic weather stations managed by the APB were used to retrieve meteorological data. The Madriccio/Madritsch station (2825 m a.s.l.) is located within the same glacial cirque as SRG, at a distance of 1.4 km and an elevation 232 m higher. A second weather station is located in the Solda/Sulden village (1900 m a.s.l.), and it was used to validate precipitation data for the Zay spring (see Section 3.2).
The Madriccio station is located 5.6 km away from ZRG and at a 133 m higher elevation, and it is separated from the Zay subcatchment by the Rosim subcatchment ( Figure 1a).

| Field activities and laboratory
The two springs were investigated during three consecutive summers (2017)(2018)(2019)

| Data analysis
We used the meteorological data recorded by the two automatic  and corrplot (Wei, 2017) R packages for data visualization. δ 18 O was discarded a-priori from PCA analysis because it was strongly collinear with δ 2 H.
The relationship between precipitation events and the variation of water parameters was estimated by determining new variables from the physico-chemical and meteorological dataset. For each precipitation event (minimum intensity ≥ 5 mm h À1 over the recorded interval), we calculated the following parameters: total precipitation (mm); duration (h); and the average and maximum intensity (mm h À1 ) of precipitation. The minimum inter-event time (to consider two separate precipitation events) was set at 1 h, as suggested by Molina-Sanchis et al. (2016). The type of precipitation was estimated for each event as a function of the liquid-solid state of water, based on the average air temperature during the event (US Army Corps of Engineers, 1956): rainfall (T air > 3 C), rainfall/snow (1 C ≤ T air ≤ 3 C), snow/rainfall (À1 C ≤ T air < 1 C) or snow (T air < À1 C). We could use the weather dataset of the Madriccio station also for the analyses at ZRG (located in an adjacent subcatchment) because all precipitation events associated to responses in EC/T water at this spring were also recorded by the Solda weather station (located at the Zay closing section), and as such could be considered as catchment scale events.
When a transient shift (δ) above the instrumental accuracy of T water (> 0.1 C) and/or EC (> 5 μS cm À1 ) was detected (for T water we used the records of HOBO Pendant UA-001-08, with lower accuracy, only when the U-24 was not recording, i.e., during 2017 and June/ July 2018), we calculated the following weather variables: lag time (h) between the onset of the precipitation event and the onset of the response; variation of EC (δEC = EC minimum/maximum À EC at the precipitation onset); variation of T water (δT = T water minimum/ maximum À T water at the precipitation onset); δEC and δT duration (h; time elapsed to return to pre-event values or to the trend of the corresponding cycles, assessed by visual inspection of EC and T water plots).
To estimate the drivers and temporal trends of water parameters we used a combination of linear models (LM), generalized linear models (GLM) and generalized additive models (GAM) that were performed using the basic R functions (for LM, GLM) or the package mgcv (for GAM; Wood, 2020) in R 4.0.0 (R Development Core Team, 2020). To identify trends in the recorded time series we applied GAMs separately for each station, assuming daily averages of T water and EC as response variables, days after the snowmelt as an explanatory variable, and year as covariate. Moreover, to prevent heterogeneity of residuals, we discarded some outliers in the series before running the GAM analyses, following Zuur et al. (2009). We also used LM, GAM, and GLM to identify the drivers of the physical and chemical response to precipitation events, separately for each station, by setting the absolute variations in the parameters (δEC, δT, δEC duration, δT duration) as response variables and the parameters of precipitation (total precipitation, intensity, maximum intensity, days elapsed after the snowmelt) as predictors, and precipitation type as categorical covariate. We validated the models with residual graphs and following the procedures outlined by Zuur et al. (2009). The distribution family (Gaussian or Gamma) of GAM/-GLM was selected based on the Akaike Information Criterion (Zuur et al., 2009). Thin plate regression spline is the default smoother used for GAM in the mgcv package (Wood, 2020).
The periodicity in T water and EC time series was analysed with wavelet analysis. This spectral analysis investigates periodical phenomena in time series by partitioning the variability in the series into different components according to different frequencies (Morlet et al., 1982). We applied Morlet wavelet transformation with the R 4 | RESULTS

| Climatic conditions during the monitoring period
During the 3 years of monitoring, the Madriccio station recorded typical alpine conditions, with a mean air temperature of À1.2 C and winter snow cover lasting for 206-222 days (Figure 1f). An early onset of snow accumulation occurred in 2016 (10 th October) when compared with 2017, 2018 (late October) and 2019 (mid-November).
In 2017 there was an earlier end to the snowmelt (10 th June) compared to 2018 (16 th June) and 2019 (29 th June). Summer 2017 was warmer (August air temperature = 8.6 ± 3.9 C) and wetter (total precipitation during the snow free period equal to 506 mm) than the summers of 2018 (7.5 ± 3.3 C; 390 mm) and 2019 (7.8 ± 2.9 C; 504 mm). The first two axes of the PCA explained 72.2% of the total variance within the dataset (Figure 2; Table 2). The gradient along PC1

| Trends of water parameters at multiple timescales
The continuous monitoring of EC and T water revealed contrasting patterns for the two springs during the study period (Figure 5a). GAM analysis confirmed the importance of the days after the snowmelt as a significant explanatory variable (Figure 5b, Table 3). Seasonal minima in T water were recorded during the snowmelt period at both stations.
As summer progressed, water temperature showed a positive unimodal behaviour at the stream-like spring (SRG), peaking during mid-summer, and a continuous increase at the pond-like spring (ZRG).
In contrast, EC continuously increased at SRG and had a positive unimodal behaviour at ZRG, where peaks occurred during late summer before the autumn decline (Figure 5a).
Wavelet analysis showed distinct cycles in the T water and EC series at the two springs ( Figure 6). For SRG 1-day cycles and 1-

| Effect of precipitation events
The two springs showed contrasting responses to precipitation events. The analysis of the time series at the Madriccio station over the three summers detected 218 precipitation events.
At SRG, only one snow/rainfall event (from a total of N = 18), 19 rainfall/snow events (N = 38) and 40 rainfall events (N = 145) were associated with a response in EC and/or T water . Almost 94% of rainfall or rainfall/snow events >2 mm occurring when snow cover was absent were associated with a negative response in EC and a positive response of T water (larger than the instrument accuracy). Total precipitation was the best predictor for the response of EC and T water at the two springs in GLM analyses, and also gave the best performance (explained deviance) when periods of snow cover presence were excluded from the analysis. During snow-free periods, we observed a low fall in EC (of 5.5-11.1 μS cm À1 ) to precipitation events down to a threshold of around 5 mm, after which δEC declined as a   Table 3). Models revealed that the response of T water to precipitation events was most likely to occur for the events during relatively warm conditions (T air > 4 C) and when snow cover was absent. The delay between precipitation onset and the associated response in T water and EC ranged between 30 min and 2.5 h.
At ZRG, only seven precipitation events influenced T water . The response was weakly negative and close to the instrument accuracy (À0.2 < δT < À0.1 C), long-lasting (1-2 days), and it was only recorded from late July to early September. We found no relationship between the meteorological parameters and δT. Only nine (total N = 96) precipitation events influenced EC with values above the instrument accuracy.
These events occurred during relatively warm conditions (T air > 4 C) and when snow cover was absent. δEC was positive and exponentially related to total precipitation (R 2 = 0.79, p = 0.001), as also shown by GAMs (Figure 7; Table 3). The δEC lag-time ranged from 6 to 9.5 h (except 2 h on one occasion), and we found no relationship between precipitation duration and δEC duration.

| DISCUSSION
The two springs investigated in this study are typical examples of rock glacier outflows but, despite exhibiting many common features, they display significant differences in hydrochemistry and seasonal trends of water parameters. In these ways, they embody the contrasting hydrochemical patterns that can be found in the literature on rock glacier hydrology.

| Snow and permafrost drive spring similarities
As outflows from intact rock glaciers, the two springs have compara- hydrological behaviour . For example, an anticipated snow accumulation precludes the efficient cooling of the active layer during winter (thermal buffering of snow against cold air), whereas an anticipated snowmelt, and warm summers, accelerate the warming of the active layer and promote ice melting (Schoeneich et al., 2011). In fact, while the snowmelt water typically has very low solute concentrations  and as such promotes water dilution during early summer, permafrost ice is enriched in solutes and its thawing enhances solute concentrations in the water during late summer . Thus, we attribute the higher

| Lithology and geomorphology drive spring differences
Despite these similarities, the two rock glacier springs differed in several hydrochemical parameters and in their seasonal behaviour. The hydrological connection between the Zay rock glacier and the glacier located upstream (Brighenti, Tolotti, Bruno, Engel, et al., 2019) explains the higher turbidity values (peaking in the seasonal period of glacier ablation), the more depleted isotopic signal, and the lower concentrations of major ions recorded at the Zay spring when compared with the Solda spring. The different bedrock composition of the two rock glaciers can explain the distinct concentrations of major ions and trace elements in the two springs. These differences progressively decline towards autumn, when the baseflow contribution becomes the dominant component of rock glacier discharge (Wagner et al., 2020). Although some studies did not find evidence of a lithological origin of high trace element concentrations in rock glacier waters (Krainer et al., 2011;Thies et al., 2007), and some attributed solute export from permafrost thawing to the release of legacy contaminants (Scapozza et al., 2020), rock weathering is considered the major driver of trace element export in rock glacier waters. This is especially under permafrost thawing conditions  when variations in the availability of weathering products results in the preferential export of different combinations of solutes (Steingruber et al., 2020;Tolotti et al., 2020 ium, and strontium were much higher at the Solda spring, where the bedrock is enriched in these elements . Notably, arsenic concentrations were high, exceeding by two to four times the limits for drinking water quality (EU, 2020), but did not show a clear seasonality. This suggests that permafrost thawing does not enhance arsenic concentrations, which instead might originate from the F I G U R E 5 (a) Series of EC and T water (daily average) at the two springs over the logging period (24 June-20 September 2017, 9 June-26 September 2018, 18 June-15 October 2019). As a reference, we provide the daily values of air temperature and total precipitation, and the presence of snow cover at the Madriccio station (Source: APB, 2020b). (b) Model fit of GAMs analyses performed with T water and EC in the two springs, setting the days after the snowmelt as a smoothing term (see Table 3 for model numerical results). Solid line is the smoother, shaded area represents 95% confidence interval weathering of the calcareous bedrock filling the cracks in this area of tectonic contact (Montrasio et al., 2015). In fact, a recent study attributed the origin of arsenic in rock glacier waters to the presence of Asenriched carbonate ores associated with quartz dykes (Eder, 2019). However, wavelet transformations detected fluctuations at weekly to bi-weekly timescales, suggesting that both rock glaciers are hydrochemically and thermally responsive to the medium-term meteorological variability typical of the alpine summer.

| A window of permafrost thaw revealed by water parameters?
The contribution from permafrost ice melt in rock glacier outflows is more likely to occur in the late summer, when the 0 C isotherm reaches the interstitial ice and triggers partial melting Leopold et al., 2011;Williams et al., 2006). At the F I G U R E 6 Wavelet power spectrum of electrical conductivity and water temperature recorded at SRG and ZRG during each summer. As a reference, the daily values of air temperature ( C, black line), snow height (cm, grey area) and total precipitation (mm, grey bars) at the Madriccio station are plotted in the centre of the figure (Source: APB, 2020b). Horizontal axes represent the timeline, shown only in the plot of weather conditions. Vertical axes indicate the fluctuation period (days). The wavelet power spectrum (coloured space, 250 power levels) represents the affinity of each variable to each period over the series. White contours delineate the areas of significant periods (p < 0.01, method 'white noise'), and the black line indicates a ridge in the power spectrum (i.e., strongest affinity of the variable with the corresponding period) Solda rock glacier, the increased permafrost ice melt contribution might have caused the onset of water temperature decline and the solute enrichment at the end of August/early September each year during the study period. The behaviour of water temperature resembles that previously reported for active (Krainer et al., 2015(Krainer et al., , 2007Krainer & Mostler, 2002;Munroe, 2018) and inactive (Harrington et al., 2018) rock glaciers, as well as for periglacial taluses (Millar et al., 2013). In contrast, the steady increase of water temperature at the Zay pond-like spring suggests a slow thermal response of the rock glacier to the degree-days previously accumulated over the season. At this spring, EC and solute concentrations had a unimodal trend, peaking 60-80 days after the snowmelt end. Similar timings of maximum EC and solute concentrations were also detected by studies on ponds influenced by rock glaciers  and permafrost (Colombo et al., 2019), and may reveal the period of permafrost thaw. This 'window of permafrost thaw' ends when the cold air and the declining solar radiation promote the re-cooling of the active layer in autumn and prevent further internal ice melting.
The isotopic enrichment occurring during the same period at the Zay pond-like spring might reveal the release of water from permafrost ice melt that has undergone several freeze/thaw cycles (Williams et al., 2006), and/or a higher contribution from liquid precipitation, which was isotopically more enriched when compared with the spring water.
At the Solda stream-like spring, the window of permafrost thaw might be identified differently. In the late summer, any solute enrichment occurring during this period (at least, during dry days), while isotopic values do not change, cannot be attributed to the concentrationeffect, and might reveal permafrost ice melt. Unfortunately, discharge was not monitored in our study, and this hypothesis thus remains speculative and should be tested in future research.

| Contrasting responses to rainfall events
Several precipitation events which occurred during the monitoring period triggered a transient response of EC at the two springs. The variations were more marked when the snow cover was absent, precipitation was very likely in the form of rainfall, and precipitation amounts were high. However, our third hypothesis (H3) on the precipitation features driving the response characteristics of the springs cannot be accepted. In fact, the intensity and duration of EC response to precipitation events was evident and controlled by the precipitation characteristics only at the Solda stream-like spring, where a rapid (in the order of hours) dilution and warming effect associated to the rainfall events occurred. Although a fast and transient solute dilution from rainfall has already been reported in rock glacier outflows (Harrington et al., 2018;Krainer et al., 2007;Krainer & Mostler, 2002), our study is the first to provide a measure of the intensity and duration of these responses, suggesting a threshold of 5 mm rainfall after which the increase of rainfall strongly correlates with an increased effect on EC and water temperature. The positive thermal response to rainfall (during snow free periods) at Solda, which is unique in the literature Geiger et al., 2014;Harrington et al., 2018;Krainer et al., 2015Krainer et al., , 2007Krainer & Mostler, 2002;Winkler et al., 2016), might be promoted by the increased subsurface water flow coming from the surrounding moraine deposits that occupy part of the spring catchment. During F I G U R E 7 Scatterplots of precipitation parameters versus the associated response of water conditions at the two springs. We fitted 95% confidence intervals of linear (LM) and generalized additive (GAM) models. Red points indicate the values that were discarded from the models because snow cover was present (precipitation-δEC plot) or because T air was low (<4 C; precipitation-δT plot) during the event rainfall events, water percolating the moraine sediments might mix with the rock glacier waters before outflowing at the Solda spring, with the observed effect of dilution and warming. However, we suggest that this physical and chemical influence has a short duration, as the scarce development of these morainal debris promotes a quick water routing and hinders water retention.
At the Zay pond-like spring, a long-lasting effect of rainfall events was observed in terms of solute enrichment, with an intensity that is F I G U R E 8 Schematic representation of the distinct behaviour of rock glacier springs at multiple timescales during the snow free season.
(a) Stream-like springs such as the Solda one exhibit an asymptotic behaviour of solute concentrations as summer progresses (a 1 ). Rainfall events (arrows, thickness, and length indicate increasing precipitation amount) cause a rapid dilution effect (a 2 ) at these springs, where diel fluctuations of solute concentrations are evident soon after the snowmelt period, and progressively smooth towards the end of the summer (a 3 ). (b) Pond-like springs such as the Zay one exhibit a unimodal behaviour of solute concentrations as summer progresses (b 1 ), with peaks corresponding to the window of permafrost thaw (WPT). Rainfall events can cause a delayed and long-lasting effect of solute enrichment (b 2 ) at these springs where diel fluctuations of solute concentrations do not occur (b 3 ). Callouts indicate the number of studies supporting this evidence (see main text for references) correlated to the volume of rainfall. However, only a few rainfall events triggered a detectable response in EC, and we could not identify the event parameters responsible for this response. A similar hydrochemical response to precipitation was described in a rock glacier pond by , who concluded that rainfall events occurring during the snow free season enhance solute concentrations because infiltrating rainfall flushes out the weathering products derived from permafrost thaw.

| Potential drivers of contrasting hydrological patterns in rock glacier outflows
Building on previous studies, we suggest that the distinction between stream-like and pond-like systems might be a good predictor of physical and chemical patterns at multiple timescales in rock glacier waters In contrast, at pond-like springs (such as the one at Zay) solute concentrations peak during late summer, rarely increase after rainfall events, and are not influenced by snowmelt cycles (Figure 8b). A prevalence of slow and distributed water pathways in the rock glacier interior might cause a buffered hydrological behaviour that we observed at the pond-like spring (Harrington et al., 2018;Winkler et al., 2016).
The Zay rock glacier has a gentle slope for most of its longitudinal profile, the water velocity at the spring is very low, and a wet meadow with raised grass tufts located in the rock glacier forefields indicates a diffuse water table emergence (Hayashi, 2020). The presence of a ponding system reveals the emergence of the aquifer close to the rock glacier front. Displacement (translatory flow; see Sprenger et al., 2019) and/or uplifting mechanisms (e.g., transmissivity feedback; Bishop et al., 2004) have been suggested as key processes promoting the outflowing of old groundwater during rainfall events in aquifers and have been also suggested to occur in periglacial taluses (Muir et al., 2011). These hydrological processes would explain the solute enrichment occurring after rainfall events at the Zay ponding spring, particularly during the window of permafrost thaw when the rock glacier aquifer is enriched in solutes that are flushed from theintra-permafrost layer.
Unfortunately, these hypotheses linked to contrasting geomorphological settings are supported by few studies, and the hydrological regime and internal structure of the Zay and Solda rock glaciers are still unknown. Further research involving hydrological, hydrochemical, and geophysical characterization of rock glaciers is necessary to better elucidate the linkages between the internal structure of these landforms and their hydrochemical behaviour at multiple timescales, as well as to verify the existence of these two major systems featuring distinct hydrological dynamics.

| CONCLUSIONS
This study details the seasonal trends in water temperature and solute concentrations of rock glacier springs and, to the best of our knowledge, represents the first attempt to quantitatively describe the fluctuation of these parameters at multiple timescales, including the effects from precipitation events. Our results provide additional insights on the importance of hydromorphological settings in driving the physical and chemical attributes of rock glacier outflows. The seasonality of these springs is dictated by the dilution effect from melting snow paralleled by an increasing influence from internal ice melt, resulting in continuously cold waters and increasing concentrations of major ions and trace elements as summer progresses. Based on the limited literature, we hypothesized that the distinct patterns of water routing determine the hydrochemical behaviour of rock glacier outflows. Stream-like springs with channelized base flow pathways display diel cycles of water parameters, and rainfall events cause solute dilution because the rainwater is efficiently routed across the rock glacier. In contrast, pond-like springs, relying on a shallow and distributed aquifer, have smoothed diel cycles of water parameters, and react slowly to changing atmospheric conditions in particular rainfall. These events can trigger solute enrichment, likely promoted by displacement or uplifting mechanisms occurring in the rock glacier aquifer. A seasonal window of major permafrost thaw can be detected for these rock glacier springs because of the efficient export of weathering products during this period.
Given the increasing hydrological influence of rock glaciers in deglaciating and glacier-free catchments, a better understanding of the drivers for distinct physical and hydrochemical patterns would help inform the management of water resources under ongoing climate change. For example, drinking water monitoring might be intensified during the window of permafrost thaw for pond-like springs, with less frequent water quality testing needed in autumn and winter.
In contrast, stream-like rock glacier springs might be of greater concern given the prolonged seasonal persistence of high solute concentrations in these waters. These differing patterns might also be ecologically relevant since the enrichment of heavy metals might hinder the survival of cold-adapted species that are also sensitive to water contamination.