Saturated areas through the lens: 2. Spatio‐temporal variability of streamflow generation and its relationship with surface saturation

Investigating the spatio‐temporal variability of streamflow generation is fundamental to interpret the hydrological and biochemical functioning of catchments. In humid temperate environments, streamflow generation is often linked to the occurrence of near stream surface saturated areas, which mediate hydrological connectivity between hillslopes and streams. In this second contribution of a series of two papers, we used salt dilution gauging to investigate the spatio‐temporal variability of streamflow in different subcatchments and for different reaches in the Weierbach catchment (0.42 km2) and explored the topographical controls on streamflow variability. Moreover, we mapped stream network expansion and contraction dynamics. Finally, we combined the information on the spatio‐temporal variability of streamflow with the characterization of riparian surface saturation dynamics of seven different areas within the catchment (mapped with thermal infrared imagery, as presented in our first manuscript). We found heterogeneities in the streamflow contribution from different portions of the catchment. Although the size of the contributing area could explain differences in subcatchments' and reaches' net discharge, no clear topographic controls could be found when considering the area‐normalized discharge. This suggests that some local conditions exert control on the variability of specific discharge (e.g., local bedrock characteristics and occurrence of perennial springs). Stream network dynamics were found not to be very responsive to changes in catchment's discharge (i.e., total active stream length vs. stream outlet discharge relationship could be described through a power law function with exponent = 0.0195). On the contrary, surface saturation dynamics were found to be in agreement with the level of streamflow contribution from the correspondent reach in some of the investigated riparian areas. This study represents an example of how the combination of different techniques can be used to characterize the internal heterogeneity of the catchment and thus improve our understanding of how hydrological connectivity is established and streamflow is generated.

ity. Moreover, we mapped stream network expansion and contraction dynamics.
Finally, we combined the information on the spatio-temporal variability of streamflow with the characterization of riparian surface saturation dynamics of seven different areas within the catchment (mapped with thermal infrared imagery, as presented in our first manuscript). We found heterogeneities in the streamflow contribution from different portions of the catchment. Although the size of the contributing area could explain differences in subcatchments' and reaches' net discharge, no clear topographic controls could be found when considering the area-normalized discharge. This suggests that some local conditions exert control on the variability of specific discharge (e.g., local bedrock characteristics and occurrence of perennial springs). Stream network dynamics were found not to be very responsive to changes in catchment's discharge (i.e., total active stream length vs. stream outlet discharge relationship could be described through a power law function with exponent = 0.0195). On the contrary, surface saturation dynamics were found to be in agreement with the level of streamflow contribution from the correspondent reach in some of the investigated riparian areas. This study represents an example of how the combination of different techniques can be used to characterize the internal heterogeneity of the catchment and thus improve our understanding of how hydrological connectivity is established and streamflow is generated.

| INTRODUCTION
The spatio-temporal variability of surface saturated areas and its impact on the hydrological behaviour of catchments have been on top of research agendas for several decades. Surface saturated areas are recognized as key areas in generating run-off in humid temperate regions (Ambroise, 2004;Hewlett, 1961)-mediating the onset and offset of hydrological connectivity between hillslopes and streams (Birkel, Tetzlaff, Dunn, & Soulsby, 2010;Bracken & Croke, 2007;Tetzlaff et al., 2007). In these environments, the development of surface saturated areas is primarily due to the occurrence of saturation excess (Dunne & Black, 1970a) in near stream areas with low relief and shallow water table (i.e., riparian zone) and up to the previously dry low-order channels (Bracken & Croke, 2007;Dunne, Moore, & Taylor, 1975;Montgomery & Dietrich, 1989). Both riparian surface saturated areas and stream networks are known to be highly dynamic (Dunne et al., 1975;Godsey & Kirchner, 2014;Shaw, 2016;Whiting & Godsey, 2016), quickly extending in response to precipitation and fostering the establishment of hydrological connectivity between the riparian zone and the stream-eventually triggering run-off generation (Bracken & Croke, 2007). Moreover, riparian surface saturated areas and stream network expansion and contraction dynamics reflect local groundwater (GW) dynamics. The spatial extent of riparian surface saturated areas can be considered as a valuable indicator of the hydrological state of the catchment and, in particular, of GW storage during baseflow conditions (Ambroise, 2016;Gburek & Sharpley, 1998;Myrabø, 1997). Similarly, stream network dynamics have been defined as a visible expression of subsurface processes otherwise hidden (Godsey & Kirchner, 2014). For these reasons, an accurate characterization of surface saturation and stream network dynamics is required to fully interpret the hydrological behaviour of catchments in humid temperate environments and to accurately predict run-off dynamics and associated flowpaths.
Understanding how run-off is generated within a catchment and which features, namely, catchment topography, geology, vegetation, and climate, control its variability is crucial to interpret catchment responses and stream water biogeochemical signatures and fluxes (Bergstrom, Jencso, & McGlynn, 2016;Pinay, 2005). Some experimental studies on this subject have adopted catchment discretization into defined landscape units such as hillslopes, riparian areas, and streams (Jencso et al., 2009;McGlynn, 2003;McGlynn & McDonnell, 2003;McGlynn, McDonnell, Seibert, & Kendall, 2004), providing fundamental information on run-off source area dynamics in terms of hillsloperiparian-stream (HRS) connectivity (defined as water table continuity across hillslope, riparian zone, and stream). These studies helped to clarify the relative role of different landscape units as spatial sources of run-off and the importance of riparian zones in regulating the portion of "new" and "old" water in stormflow (McGlynn et al., 2004;McGlynn & McDonnell, 2003). Other studies have focused on characterizing spatial and temporal variability of run-off by measuring discharge along continuous stream reaches (Anderson & Burt, 1978;Bergstrom, Jencso, et al., 2016;Floriancic et al., 2018;Genereux, Hemond, & Mulholland, 1993;Huff, O'Neill, Emanuel, Elwood, & Newbold, 1982;Kura s, Weiler, & Alila, 2008;Payn, Gooseff, McGlynn, Bencala, & Wondzell, 2012;Shaw, Bonville, & Chandler, 2017). Unlike the studies employing catchment discretization, these studies take into account the dynamics of surface water (cf. Blume & van Meerveld, 2015), specifically the increase or decrease of streamflow between two measurement points. When applied over a whole stream network, this approach has the advantage of providing a general indication of heterogeneities in streamflow generation within the catchment. This heterogeneity can be directly linked to hydrologic dynamics, structure, and vegetation to understand how different processes are integrated along the stream to produce the total discharge volume at the outlet.
In humid temperate regions, spatio-temporal variability of streamflow is very often linked to the location and temporal variability of surface saturated areas (Bracken & Croke, 2007). However, studies combining variability in streamflow generation with surface saturation dynamics are extremely rare (Shaw et al., 2017;Ward, Schmadel, & Wondzell, 2018). Moreover, these studies tend to focus only on stream network dynamics. To the best of our knowledge, riparian surface saturation dynamics have only been investigated in relation to measurements of discharge at the catchment outlet or in relation to GW level fluctuations (Birkel et al., 2010;Dunne & Black, 1970b;Lana-Renault, Regüés, Serrano, & Latron, 2014;Latron & Gallart, 2007;Martínez Fernández, Ceballos Barbancho, Hernández Santana, Casado Ledesma, & Morán Tejeda, 2015;Tanaka, Yasuhara, Sakai, & Marui, 1988). The study of Kirnbauer and Haas (1998) in an Alpine catchment is a unique exception in this respect. They used stream gauges downstream of surface saturated areas to quantify their contribution to run-off. Combining a detailed description of surface saturation dynamics with the investigation of streamflow variability along the stream network could provide new insights on the spatial and temporal variability of HRS connectivity in humid temperate environments and, in particular, on the role of valley bottoms in regulating this connectivity. Additional experimental investigation along this line of work is needed across catchments with a range of geological and climate conditions to advance our understanding of surface saturation dynamics and its link to run-off generation.
The main obstacle to comparing surface saturation dynamics with streamflow dynamics along the stream network stems from the need to map surface saturation at the same temporal resolution to which streamflow is measured and for different locations within a restricted timeframe. Even though time consuming, mapping the active portion of the stream network can be achieved by walking along the stream and recording the active/inactive portions using a GPS receiver (Godsey & Kirchner, 2014;Shaw, 2016). More challenging is the mapping of riparian surface saturation, where classic approaches such as field surveys based on the "squishy boot" method may not provide an adequate spatio-temporal resolution (Pfister, McDonnell, Hissler, & Hoffmann, 2010). In this regard, recent technological development is represented by ground-based remote sensing techniques (i.e., thermal infrared [TIR] or digital imagery) with which surface saturation can be mapped at a higher temporal (i.e., minutes to weeks) and spatial (i.e., centimetres to metres) resolution (Glaser et al., 2016;Glaser, Antonelli, Chini, Pfister, & Klaus, 2018;Pfister et al., 2010;Silasari, Parajka, Ressl, Strauss, & Blöschl, 2017).
Here, we investigate the link between surface saturation dynamics (read as both riparian surface saturation and dynamics of expansion and contraction of the active portion of the stream network) and streamflow generation in the Weierbach catchment in Luxembourg.
The Weierbach catchment is a long-term studied catchment, nowadays considered as a reference catchment for rainfall-dominated mountainous catchments (Zuecco, Penna, & Borga, 2018). The catchment's hydrological response is influenced by a storage threshold (Martínez-Carreras et al., 2016), and it is characterized by a single spiky peak in case of dry antecedent conditions and by a first spiky peak followed by a broader peak of longer duration in case of wet antecedent conditions (Martínez-Carreras et al., 2016;Wrede et al., 2014). The riparian zone in this catchment presents seasonally dynamic surface saturated areas whose possible influence on the connectivity and hydrological response of the system has never been clarified.
This contribution is the second in a series of two papers. Here, we leverage (a) information obtained by monitoring the spatio-temporal Our findings will be discussed in light of the current perceptual model of the Weierbach catchment (Martínez-Carreras et al., 2016;Scaini et al., 2018;Wrede et al., 2014) and compared with previous research in a broader context.

| Hydro-meteorological measurements and catchment storage calculation
Hydro-meteorological measurements of stream discharge at the out-

| Monitoring of saturated areas in the riparian zone and stream network dynamics
Riparian surface saturation and stream network dynamics have been surveyed simultaneously, weekly, or fortnightly from November 2015 to December 2017. Riparian surface saturation has been monitored in seven different locations via ground-based TIR imagery, and its dynamic has been characterized through postprocessing of the TIR camera outputs (i.e., sequential images or videos) following the methodology outlined in Glaser et al. (2018). The riparian surface saturated areas were seasonally variable and were found to be particularly responsive to GW fluctuations. The development of surface saturation in the seven different areas is influenced by local riparian morphology that leads to small differences in the relationship between surface saturation and outlet discharge observed for the different areas. For a thorough characterization of the dynamics of riparian surface saturation, the reader is referred to the accompanying manuscript

| Salt dilution gauging
Stream discharge was measured via salt dilution gauging at 12 locations along the stream network ( Figure 1). The measurements were carried out within a few hours on 11 dates with no rain and contrasting hydrological states ( Figure 3). Salt dilution gauging (Day, 1976) is a common method for measuring discharge in small streams with irregular streambed morphology (Moore, 2004). Measurement locations were selected based on two criteria: (a) include a surface saturated area between the upstream and downstream measurement locations and (b) maximize the possibility for complete mixing of the injected salt solution and stream water between the injection point and the measurement location (i.e., by injecting just upstream of a riffle or a narrowing of the stream section), which is an important requisite for salt dilution gauging (Day, 1977;Moore, 2004). We injected a solution of a known amount of NaCl. Electrical conductivity was recorded at the discharge measurement locations using a WTW Multi 3420 device, equipped with a TetraCon 925 probe (Xylem Analytics, Weilheim, Germany).
Replicates of the salt dilution gauging were carried out for 15 of the measurements, covering different flow states and measurement locations. In these cases, a second injection was carried out after the stream electrical conductivity returned to its background value. We found an average error between the two replicates of~3%, with a minimum of 0% and a maximum of 10%, which was sufficient for our application.

| Data analysis
The discharge values (L/s) retrieved via salt dilution gauging at the different locations along the stream were used to characterize the spatial and temporal variability of streamflow generation in the catchment.
We compared the streamflow at the outlets of the different subcatchments (i.e., catchment area above each discharge measurement location- Figure 1 upper right panel) in terms of area-normalized discharge (specific discharge-mm/day). We calculated the net discharge between two measurement locations (i.e., for a reach- Figure 1 upper right panel) as their difference in discharge (L/s). In order to compare reaches of different lengths, we expressed the net discharge as specific discharge (mm/day) and normalized it by the reach stream length (m). Similarities between specific discharge produced by each subcatchment and similarities between normalized specific discharge contributions from the different reaches were tested with the Mann-Whitney-Wilcoxon test (α = 0.05). The test was applied by taking into account the subcatchments against each other for the dates where a F I G U R E 3 Streamflow dynamics at the catchment outlet and days of salt dilution gauging (dashed red lines). The same days are reported along the flow duration curve (FDC; 2-year study period) for the catchment outlet measure of discharge (or net discharge) was available for the considered pair. The same procedure was followed for the reaches. Spearman's rank correlation test rho (ρ; α = 0.01) was applied in order to test monotonic relationships between the subcatchments' specific discharge and the other hydrometric measurements (i.e., stream discharge at the outlet, estimated catchment storage, GW levels, and soil VWC-daily-averaged values) and between the normalized specific discharge contribution of the different reaches and the hydrometric measurements (i.e., daily-averaged values).
We used multiple linear regression analyses to investigate which topographic characteristics influenced discharge (L/s) and specific discharge (mm/day) contributions of the different subcatchments and reaches (non-normalized values). We extracted several topographic features from a DEM (5-m resolution) and from a high-resolution LIDAR DEM (~5-cm resolution). We extracted topographic features for the different subcatchments and for the portion of the catchment draining between the two measurement locations defining a reach (i.e., upslope catchment area). Specifically, we considered catchment area, riparian area, percentage of riparian area, riparian buffer (riparian area/hillslope area), median slope (only for the reaches because too homogeneous between the subcatchments), percentage of steep slope (i.e., >15 ), median elevation, median flow length (only for the subcatchments), and reach length (only for the reaches). The models were ran for each discharge measurement date and averaged through time. Variance inflation factor analysis and backward selection were carried out to select the significant variables to retain in the models.
To investigate the relationship between stream network dynamics and stream outlet discharge, we related outlet discharge and stream length dynamics of the three headwater areas and tested the occurrence of monotonic relationships with Spearman's rank correlation (α = 0.01). We also related outlet discharge and the total active stream length (i.e., total stream length considering the entire catchment) and fitted a power law equation (stream length = a * Q β ) to this relationship following the approach of Godsey and Kirchner (2014). The total active stream length versus discharge relationship can provide an estimation of how much the total active stream length changes (in percentage) with changes in discharge. This estimation is represented by the β power law scaling exponent. Following a similar approach, we related the total active stream length and estimated catchment storage and GW level (measured in well GW5 on the plateau-the closest well to two out of three monitored headwater areas). We fitted equations that approximated the trend of these relationships. The goodness of fit of all the fittings was tested with Kolmogorov-Smirnov test (p value > 0.1). All the hydrometric variables are daily averages.  (11). The map also reports the names for reaches. For the discharge measurement point names, we refer to Figure 1 that the number of observations that could be used to explore these relationships was consistently lower than for the other investigated relationships. This was due to the low quality of the TIR images collected during some of the stream gauging dates. Therefore, and due to its low statistical significance, we avoided any quantification of the strength of this relationship. However, we believe that a description of the trends that could be observed from the scatterplots (cf. Figures 8 and 9) would provide us with valuable information.

| Spatio-temporal variability of streamflow
Streamflow within the catchment was found to be highly variable in to contribute considerably more than others (Figure 4b). During wet conditions, the contribution of these reaches increased further, and R-M1 and R-M3, R-R1, R-R2, and R-R3, and R-S2 and R-S1 became more active as well (Figure 4c).
Considering all the measurement dates, some reaches were found to contribute positively to streamflow more frequently than others (i.e., R-L1 and R-S2 in Figure 4d). Similarly, some reaches showed overall higher variability in contribution (i.e., R-S2, R-L1, R-M2, and R-R1) compared with others ( Figure 5a). Between the reaches with most variable contribution, R-L1 and R-M2 were found to be particularly similar (Mann-Whitney-Wilcoxon test p value = .7). Specific discharge contributions of the different subcatchments appeared to be quite homogeneous within the same stream gauging dates (Figure 4).
Only the subcatchment with outlet in point SW2 produced systematically higher specific discharge than the other subcatchments, even though this difference was not statistically significant (Mann-Whitney-Wilcoxon test p value always higher than.05; Figure 5b For the reaches, a colour is assigned if the reach includes the correspondent riparian surface saturated area; for the subcatchments, a colour is assigned if the subcatchment outlet is right downstream the correspondent riparian surface saturated area different reaches, we observed that reaches R-S0, R-S1, R-L1, R-M1, R-R2, and R-R3 were not correlated to other reaches and, except R-L1 and R-R3, never correlated with catchment outlet discharge, estimated catchment storage, or GW levels (ρ ≤ 0.75 or nonsignificant correlations). VWC never correlated with normalized specific discharge contributions of any reach, except for R-S2 (ρ not lower than 0.81). Also in this case, VWC measured in riparian location did not correlate with any reach (nonsignificant correlations).

| Stream network dynamics and relationship between riparian surface saturation and reaches' streamflow contribution
During the study period, the stream network never dried out completely. We observed only occasionally lack of flow at the outlet (e.g., in January 2017, when the stream was partially frozen) or moments in which appreciably downstream sections of the stream became ephemeral (e.g., in September 2016, the stream stopped flowing in proximity of the discharge measurement point "QS3" and started flowing again close to the surface saturated area S2). In general, we could always detect flow starting points at the three headwater locations, even though sometimes the water reinfiltrated after few metres.
Stream network expansion and contraction dynamics in the three headwater stream reaches (R-L1, R-M3, and R-R3) were all positively monotonically related to catchment's outlet discharge (Spearman's rank test ρ not lower than 0.81- Figure 6). R-L1 was the reach that expanded the least, with a difference between its maximum and minimum observed starting points of about 8 m. In contrast, R-R3 extended about up to 60 m above its minimum observed starting point. R-M3 extended upward along both the main reach direction (i.e., "R-M3 centre" in Figure 6) and on the right (i.e., "R-M3 right" in Figure 6). In both cases, the reach extended about 30 m above its minimum observed starting point.
We could observe one or more particularly stable flow starting points for all three headwater reaches ( Figure 6). In R-L1, this

| Relationships between topography and streamflow contribution
A set of topographic characteristics have been initially considered in the multiple linear regression analyses for discharge and specific discharge of subcatchments (Table 1)   explained a good proportion of the discharge variance (i.e., 98-99% for subcatchments' models and 44-94% for reaches' models).
The predictive power and significance of the models considering specific discharge (mm/day) as response variable were considerably lower (Table 3-third column). For the subcatchments, only seven out of the 11 models were significant (p value ≤ .05) and explained between 35% and 78% of specific discharge variance. In these cases, median subcatchment elevation significantly predicted subcatchments' specific discharge most of the time (positive regression coefficient). All the models considering reaches' specific discharge were found to be not significant.

| DISCUSSION
Until now, previous research carried out in the Weierbach catchment has led to considerable advancement in our level of understanding of how its hydrological response is generated. Through modelling and tracer-based studies Glaser et al., 2016 Figure 4a). We observed quite homogeneous specific discharge contributions from the different subcatchments (cf. Figure 4). Subcatchments' specific discharge was well correlated with the hydrometric measurements of outlet discharge, GW levels, and thus, estimated catchment storage. Consistent with observations by Seibert, Bishop, Rodhe, and McDonnell (2003), we found that the correlation between subcatchments' specific discharge and GW levels decreased with increasing distance of the wells from the stream (i.e., GW2 and GW3 generally better correlated with specific discharge than GW5 and GW6). In our case, this trend existed in particular for the most upstream subcatchments, which were the catchments exhibiting the power on subcatchments' specific discharge was median elevation, once again mainly during dry conditions. These results suggest the presence of one/several important controlling factor/factors on specific discharge-especially when considering the different reacheswhich has/have not been taken into account. As the high spatial and temporal variability observed in the reaches' normalized specific discharge (cf. Figures 4 and 5a) and their variable relationship with the different hydrometric measurements suggest, the specific discharge produced by each reach could be the result of very location-specific factors. For example, the presence of perennial springs in some of the reaches often resulted in generally higher normalized specific discharge (cf. Figure 5a). These reaches were the ones corresponding to PSA and PSpA locations observed via TIR imagery (described in our first contribution). An exception was the reach "QR1-SW4," which was identified as a losing reach during low flow and appeared to be quite active in terms of normalized specific discharge during higher flow. This is probably due to the activation of temporary springs in the streambed during wetter conditions. The activation of a temporary spring additionally to the perennial ones could be observed via TIR imagery in area S2, which also became very active during wetter conditions (cf. Figure 4c). Spring location and the delivery of water from the hillslopes to the streams are likely substantially controlled by bedrock characteristics as schists/slate weathering degree, fractures' size and orientation within the catchment (Gourdol et al., 2018;Scaini et al., 2018), and/or presence of faults (Shaw, 2016;Whiting & Godsey, 2016). The aforementioned bedrock characteristics have been shown to be variable within the Weierbach catchment (Gourdol et al., 2018), representing a substantial source of variability that can be hardly disentangled.
The stream network was observed to be dynamic, but it was not very responsive to changes in catchment outlet discharge. Stream network dynamics in the three headwater locations showed to be well monotonically related to catchment outlet discharge ( Figure 6). However, the relationship between the total stream length and catchment outlet discharge suggested relatively small responsiveness of the stream network to changes in discharge (i.e., the low β value in the power law relationship). This is typical of catchments with stream heads "anchored" by perennial springs (cf. Figure 6), as reported by Whiting and Godsey (2016) and Shaw et al. (2017). In accordance with the observations by Withing and Godsey (2016), we detected a higher stability in reaches L1 and M3 compared with R3 (cf. Figure 6), which was the headwater location with the smaller accumulation area during higher flow (likely to be supported by longer, deeper, and slower flowpaths) and a flatter topography. Even though stream network dynamics were found to not to be very responsive to changes in the outlet discharge, they were found to be very well correlated to estimated catchment storage and GW depth. This suggest the total active stream length to reflect subsurface processes variability rather than surface water dynamics at the outlet.
Riparian areas have their surface saturation positively correlated to normalized specific discharge from the correspondent reach. PSpA. This could be related to the fact that both surface saturation and streamflow contributions from the hillslopes are influenced by GW fluctuations (Antonelli et al., in review;Glaser et al., 2019Glaser et al., , 2016Martínez-Carreras et al., 2016;Wrede et al., 2014). surface saturation developing in PSA and PSpA is related to their high streamflow contribution. Although this could be the case during rainfall events, in moments when the system is not affected by the occurrence of precipitations, the level of surface saturation in one area and the streamflow generated by the correspondent reach seem not to really influence each other but rather be influenced by common factors as GW dynamics and springs locations. Note that because surface saturation has been quantified as percentage of saturated pixels and not as area, we could not quantify if areas with an absolute larger surface saturation provided more streamflow contribution than others.
The described positive relationship between surface saturation and streamflow contribution disappears when we consider the percentage contribution of a specific reach to the total catchment outlet discharge. We noticed that some areas contributed for very stable percentage of total discharge regardless of the general catchment wetness conditions and the level of surface saturation in the area (cf. Figure 9). A more stable percentage of contribution seemed to be associated to the reaches located in the middle and west part of the stream (i.e., reaches M3, M2, M1, and L1), in contrast to the reaches located in the east (R2 and R2) and lower part (S2) of the stream. As previously mentioned, the investigations of Gourdol et al. (2018) employing soil drilling and electrical resistivity tomography revealed some heterogeneities in the subsurface structure of the Weierbach catchment. In particular, they have shown that the northern and western part of the catchment is characterized by overall thinner solum (i.e., "true soil," where pedogenic processes are dominant; cf. Gourdol et al., 2018) and shallower hard bedrock compared with the eastern portion of the catchment. This may determine differences in the way different sides of the catchment deliver water to the stream. However, the mechanism behind the consistency of the relative contribution of some specific reaches to the total catchment outlet discharge remains of difficult interpretation.
The key role of near stream surface saturation in mediating hydrological connectivity between hillslopes and streams has been acknowledged across a range of landscapes and climate conditions, such as-just to mention a few-catchments in boreal and temperate environments (Birkel et al., 2010;Devito, Creed, & Fraser, 2005;Tetzlaff et al., 2007), Mediterranean (Lana-Renault et al., 2014;Latron & Gallart, 2007;Niedda & Pirastru, 2014), and alpine environments (Kirnbauer & Haas, 1998;von Freyberg, Radny, Gall, & Schirmer, 2014). Similarly to what we observed in the Weierbach catchment, GW dynamics and local topography-and in some cases, the presence of perennial GW springs-have been recognized as the main controls on surface saturation dynamics in the majority of the aforementioned studies. Thus, we believe our results to provide a good representation of the spatio-temporal dynamics of surface saturation and streamflow generation occurring in most headwater catchments.
Recent studies have reaffirmed the need for catchments' interfaces to be characterized for their own processes and fluxes in order to have a better perception of where and when connectivity may take place in a catchment (Blöschl et al., 2019;Wohl et al., 2019). Failure in assessing possible heterogeneities may lead to erroneous processes conceptualization and discrepancies between processes observed at smaller scales and responses that may occur at larger scales (Krause et al., 2017;Ward & Packman, 2019). In this study-together with its accompanying manuscript-we go beyond the sole characterization of the surface saturation versus outlet baseflow discharge relationship of a catchment (Ambroise, 2016;Latron & Gallart, 2007). Our results suggest that a deeper understanding of the role played by riparian surface saturation in mediating hydrological connectivity along the HRS continuum (and how it translates into the total discharge volume observed at the outlet) is possible-probably only-if considering the riparian zone (and the multitude of its hydrological processes) as a complex feature of the system, rather than as a single homogeneous entity (as suggested by Ledesma et al., 2018). Interfaces in hydrology have been traditionally considered as a boundary condition (Blöschl et al., 2019) where complexity is commonly reduced for the sake of simplicity in experimental and conceptual model designs (Krause et al., 2017). However, Blöschl et al. (2019) also recognize the need to start looking for more typical cases where this simplification can be applied or not. In our catchment, we observed that, although the seasonal dynamics of surface saturation in the different investigated areas seem to be synchronous (Antonelli et al. in review), this does not necessarily translate into similar hydrological behaviour in terms of streamflow contribution for all areas. This kind of variability is at the base of the difference between variable active and variable contributing areas (or periods) described by Ambroise (2004) and has important implications for investigating and modelling catchments' responses.
This is fundamental in studies that focus on biogeochemical transformations occurring in the riparian zone (Blume & van Meerveld, 2015;Laudon et al., 2016;Ledesma et al., 2018). Indeed, variable dynamics of surface saturation could provide indications on potentially different buffer capacities of distinct riparian sections, both in terms of water quantity and quality.

| CONCLUSION
In this contribution, we have explored the spatio-temporal variability of streamflow generation in the Weierbach catchment. We investigated possible links to the occurrence and dynamics of surface saturation and active stream length. We carried out our investigations at a finer scale compared with previous studies and showed that a considerable level of heterogeneity can be found within a small, homogeneous (e.g., vegetation coverage and pedological and geological characteristics) headwater catchment.
We found that the net discharge contribution variability between different subcatchments and between different reaches could be explained by the contributing area. However, this was not the case when considering the area-specific discharge contribution of different subcatchments and reaches. In this case, no clear topographic control was able to explain the variability in contribution, suggesting that very local factors may influence streamflow generation, such as bedrock characteristics or the presence of perennial springs. We related the surface saturation dynamics observed within the catchment to the streamflow dynamics. The stream network expansion and contraction dynamics reflected the general wetness state of the catchment (i.e., they were related to GW fluctuations and changes in the estimated catchment storage), but they were not very responsive to changes in outlet discharge (i.e., perennial springs would "anchor" the channel head in specific locations for most of the time). Finally, we showed that the surface saturation versus streamflow contribution relationship in different riparian areas could mirror the degree of connectivity of the areas to the subsurface system.
Besides providing new information on subcatchment scale processes in the Weierbach catchment, we have shown that a combination of a thorough investigation of surface saturated area dynamics within the catchment through TIR imagery with sequential measurements of stream discharge can be used to improve our perception and understanding of the internal heterogeneity of catchments. Our approach is in line with the "Roadmap for Eco-hydrological Interface Research" proposed by Krause et al. (2017), because we applied a combination of approaches from different disciplines to investigate the complexity of the riparian-stream interface and identify hotspots of hydrological connectivity/streamflow generation. This information is also fundamental in studies that have their focus on nutrients and tracers transport and eco-hydrological processes in the riparian zone.
Future research should focus on analysing and linking the observed catchment's internal heterogeneities with reference to stream water isotopic and chemical signature or through simulation approaches.

ACKNOWLEDGMENTS
We would like to thank Jean-Francois Iffly, the Observatory for Cli-