Table 1 shows a summary of collection methods, sampling resolution and locations in the Weierbach catchment. Stream water depth at the catchment outlet was measured using a differential pressure transducer at a 15 min interval (ISCO 4120 Flow Logger) (Fig. 1). Stream electrical conductivity at the outlet was also measured at 15 min intervals using a conductivity meter (WTW). Rainfall was measured with a tipping bucket rain gauge (52203 model, manufactured by Young, Campbell Scientific Ltd.). One rain gauge was installed within a small clearing of the study catchment (see Fig. 1), and another one installed in an open area at the Roodt meteorological station, located ≈ 3.5 km distant from the Weierbach one (49∘48′22.2′′ N, 5∘49′52.7′′ E). Data gaps due to instrument failure were filled with rainfall data from a nearby weather station (49∘47′39.2′′ N, 5∘49′13.2′′ E).

Four groundwater wells were instrumented with real-time TD-Divers data loggers (Schlumberger Water Services) and WTW conductivity meters – each recording at 15 min intervals. GW1 was located in a plateau, and GW2, GW3 and GW4 in the transition zone between riparian and hillslope settings (Fig. 1). Wells were around 2 m deep and were screened at least for the lowest 50 cm up to a metre.

The volumetric water content (VWC) of soils was measured using water content reflectometers (CS616-L model, Campbell Scientific), which use the time-domain reflectometry method. Four probes were installed at 10 cm depth, parallel to the surface and along a 5 m transect perpendicular to the stream (Fig. 1): riparian zone, foot of the hillslope, mid-hillslope and plateau positions.

Fortnightly, cumulative rainfall (R) and throughfall samples under deciduous trees (TH1) and coniferous trees (TH2) were collected using conical, volumetric rain gauges. A ten-bottle sequential rainfall sampler was installed at the rain gauge located within the Weierbach (modified from Kennedy et al., 1979). Three automatic water samplers (ISCO 3700 FS and 6712 FS) were installed immediately upstream of the weir to collect stream water samples (AS) frequently (0.5 to 4 h) during storm events. Sampling was triggered by flow conditions. Events were considered separately if they were separated by a period of at least 24 h without rainfall. Stream water at the catchment outlet (SW) and wells (GW1 to GW4) were sampled fortnightly, as well as prior to, during, and following precipitation events. Soil water was sampled fortnightly using Teflon suction lysimeters, installed at three locations: deciduous hillslope (SS1), coniferous hillslope (SS2), and riparian zone (SSr). Three soil depths for each location: 10 cm for the organic layer (Ah horizon), 20 and 60 cm for the mineral layers (B and C horizons). Overland flow (OF) that occurred on lower hillslope was sampled using 1 and 2 m long gutters sealed to the soil surface, which diverted surface runoff to 1 or 2 L plastic, blackened (to prevent light penetration which causes diatom growth) water bottles. Note that what we refer to as OF might in fact originate within the forest litter layer (Buttle and Turcotte, 1999; Sidle et al., 2007). All gutters were covered to avoid the influence of precipitation. Gutters were regularly cleaned with Milli-Q water to avoid diatom growth on their surfaces.

All water samples were analysed for electrical conductivity (EC), anion and cation concentrations (Cl-, NO3-, SO42-, Na+, K+, Mg2+, Ca2+), silica (SiO2) and UV absorbance at 254 nm (Abs 254 nm). UV absorbance at 254 nm can be considered as a proxy of DOC (Edzwald et al., 1985). Samples were analysed at the Luxembourg Institute of Science and Technology chemistry laboratory after filtration through WHATMAN GF / C glass fibre filters (< 0.45 µm). Prior to analysis, samples were stored at 4 ∘C. Dissolved anions and cations were analysed by ion chromatography (Dionex HPLC), SiO2 by spectrophotometry (ammonium molybdate method), and UV absorbance was measured by a Beckmann Coulter spectrophotometer. Isotopic analyses of 18O / 16O and 2H / H were conducted using a LGR Liquid-Water Isotope Analyser (LWIA) at the Luxembourg Institute of Science and Technology (model DLT-100, version 908-0008) (Penna et al., 2010). The analyser was connected to a LC PAL liquid auto-injector for the automatic and simultaneous measurement of 2H / H and 18O / 16O ratios in water samples. According to the manufacturer's specifications (Los Gatos Research Inc., 2008), the DLT-100 908-0008 LWIA provides isotopic measurements with a precision below 0.6 ‰ for 2H / H and 0.2 ‰ for 18O / 16O. Data were transformed into δ notation relative to Vienna Standard Mean Ocean Water (VSMOW) standards (δ2H and δ18O in ‰).

Diatom analysis was conducted for multiple sample types: stream water, overland flow, epilithon, epipelon, and diatoms attached to different substrates outside the streambed (i.e. litter, bryophytes, vegetation and soils).

A small set of stream water and overland flow samples was set aside for geochemical and isotopic analysis (≈ 70 mL); the rest of the sample was centrifuged (1250 rpm, 8 min) to concentrate the diatoms.

In addition to high-frequency sampling during rainfall events, seasonal sampling campaigns were carried out throughout the Weierbach catchment to assess the geographic and intra-annual variability of diatom communities. The following substrates were sampled in the catchment: (i) litter, bryophytes from the two hillslope classifications (hardwood and coniferous) and surface soil samples; and (ii) litter, bryophytes, and vegetation in the riparian zone. Each sample was comprised of five sub-samples collected on a 5 m transect parallel to the stream (a subsample collected every metre). Only material from the top surface, where there was greatest incident sunlight, was collected into 1 L plastic bottles. Sample bottles containing different substrata were filled with carbonated water (1 L), carefully shaken and left to settle overnight at 0 ∘C. The next day, the diatom-filled, carbonated water was recovered by passing it through a 1 mm screen. Sample substrate was then rinsed with additional carbonated water to remove as many diatoms from the sampled substrate as possible. This procedure was repeated several times until a 2 L sample volume was achieved. The recovered sample, now with substrate removed, was stored at 0 ∘C for a minimum of 8 h to allow diatoms to settle, and the supernatant removed by aspiration.

During the same catchment-wide campaigns, epilithic (in-stream stone substrata) and epipelic (in-stream sediment or soil substrata) samples were also collected, treated and counted following European standards CEN 13946 and CEN 14407 (European Committee for Standardization, 2003, 2004). For epilithic samples a minimum of five stones from the main flow and well-lit stream reaches were brushed to collect the diatom biofilm, while epipelic samples were collected by disturbing small pools with sediment bottoms and then pipetting a superficial layer of 5–10 mm of sediment from reach pools.

All samples were preserved with 4% formaldehyde and treated with hot hydrogen peroxide to obtain clean frustule suspensions. After eliminating the organic matter from the diatom suspensions, diluted HCl was added to remove the calcium carbonate and avoid its precipitation later, which would make diatom frustule observation difficult. Finally, oxidized samples were rinsed with deionized water by decantation of the suspension several times, and permanent slides were mounted with Naphrax^®.

Diatom valves were identified and counted (≈ 400 valves) on microscopic slides with a light microscope (Leica DMRX^®). For the autecological assignment of the diatom species we relied on (1) the Denys (1991) diatom ecological classification system refined by Van Dam et al. (1994), which is, as far as we know, the only formal classification of the occurrence of freshwater diatoms in relation to moisture; and (2) the associated hydrological units assigned by Pfister et al. (2009) to the five diatom occurrence classes defined by Van Dam et al. (1994). We express these results as relative abundance (percentage) of aerial valves, i.e. categories 4 and 5 of Van Dam's et al. (1994) classification.

Two-component hydrograph separation was performed using δ18O isotopic composition and the mass balance approach (Pinder and Jones, 1969; Sklash and Farvolden, 1982; Pearce et al., 1986; Sklash et al., 1986). The incremental mean method proposed by McDonnell et al. (1990) was used to adjust δ18O rainfall isotopic composition, so that the bulk isotopic composition of rainfall from the beginning of the event to the time of stream sampling was calculated (i.e. rain that had not yet fallen was excluded from the estimate).

Spatial end-member contributions to stream water were explored using EMMA (Christophersen and Hooper, 1992), which assumes that (i) the stream water is a mixture of end-member solutions with a fixed composition, (ii) the mixing model is linear and relies on hydrodynamic mixing, (iii) the solutes used as tracers are conservative, and (iv) the end-member solutions are distinguishable from one another. Catchment end-members included shallow groundwater (GW1-4), soil water (SS120, SS160, SS260), soil water from the riparian zone (SSr), rainfall (R), throughfall (TH1-2), snow (SN) and overland flow (OF). We applied the diagnostic tools of Hooper (2003), which have been recently applied in the literature (James and Roulet, 2006; Ali et al., 2010; Barthold et al., 2011; Neill et al., 2011; Inamdar et al., 2013). Our approach followed three main steps.

We identified tracers that exhibit conservative linear mixing assuming that stream water chemistry is controlled by physical mixing of different sources of water and not by equilibrium mixing (Christophersen and Hooper, 1992; Hooper, 2003; Liu et al., 2008). The latest would imply equilibrium reactions among solutes of different charge, which may be approximated by high-order polynomials. Hooper (2003) suggested that conservative and linear mixing of tracers can be evaluated using bivariate scatter plots. In this study, stream water concentrations and isotopic compositions (of all samples collected during storm events and low flows at the catchment outlet) were considered conservative when they exhibited at least one linear trend with one other tracer (i.e. r2 > 0.5, p value < 0.01) (James and Roulet, 2006; Ali et al., 2010; Barthold et al., 2011).

The hydrometric response for water years 2011–2012 is shown in Fig. 2. Diatom sampling commenced in November 2010 when the catchment started to progressively wet up (see groundwater depths and soil volumetric water content in Fig. 2). Annual precipitation for the water year 2011 was 671 mm, a ∼ 20% decrease compared to the average of the preceding 4 years (873 mm, as measured by the nearby meteorological station, Roodt), and 838 mm for the water year 2012. In January 2011, a 10-year return period rain-on-snow event produced a peak flow of 1.5 mm h-1. The high winter discharge levels decreased progressively from February to June 2011 due to reduced precipitation during this period. Afterwards, a dry period extended from July to November 2011. A longer wet period was measured the following year (from December 2011 to July 2012).

During wet antecedent conditions, streamflow response of the basin was double peaked, with a first peak timing coincident with the rainfall input and the second, delayed peak coming a few hours later. On the contrary, when the catchment was dry, the hydrological response was shorter and only a single sharp peak occurred.

We determined hydrological connectivity along a HRS transect via hydrometric observations. Water tables in the saprolite and fractured schist bedrock responded significantly to rainfall events. The magnitude of water level change was well correlated with the precipitation amount. Soil volumetric water content (VWC) decreased with distance upslope (VWC hillslope foot > VWC hillslope middle > VWC hillslope plateau (Fig. 2)). The riparian zone showed unchanging values close to saturation during wet periods (≈ 70 %), which decreased slightly when the catchment was dry (≈ 65 %). For all monitored events, VWC at 10 cm depth responded quickly to incident rainfall at all transect locations (i.e. hillslope foot, middle and plateau), suggesting a vertically infiltrating, wetting front.

During dry antecedent conditions (summer and spring), threshold-like behaviour between soil moisture and discharge was observed at the hillslope foot (Fig. 3a). Only when the VWC was higher than ≈ 27–30 % did discharge increase significantly (threshold 1 in Fig. 3a). A second threshold appeared when the catchment was wet (autumn and winter); stream discharge increased significantly when VWC was above 40 % (threshold 2 in Fig. 3a). This likely indicated connectivity between the hillslope and riparian compartments and the stream channel. A similar relationship was observed between VWC and depth to groundwater levels (i.e. GW1, GW2 and GW3; Fig. 3b).

Two-component hydrograph separation results using δ18O isotopic composition (i.e. pre-event water vs. event water) showed that, in winter, when the catchment was wet and flow response was double-peaked, the first peak had a larger contribution of event water than the delayed peak. For instance, the first peak of the November 2010 event showed a maximum of 50 % event water contribution. This contrasted with the delayed peak that exhibited only a maximum of 16 % event water contribution (Fig. 4a). When the catchment was dry, the response consisted of one sharp peak composed largely of event water. A maximum event-water contribution of 60 % was estimated for a storm event that occurred in June 2011 (Fig. 4b).

Twelve different tracers measured in the different water compartments of the catchment were used to assess end-member contributions to stream water (Fig. 5). Ten out of the twelve tracers presented linear trends in the solute–solute plots of stream water samples with at least one other tracer (EC, Cl-, Na+, K+, Mg2+, Ca2+, SiO2, Abs, δ2H and δ18O; r2 > 0.5, p value < 0.01, Fig. 6). These tracers were retained for the PCA analysis. Weaker linear trends were found between NO3- and the other tracers (r2 < 0.13) and between SO42- and the other tracers (r2 < 0.43). NO3- and SO42- did not reach the pre-defined threshold of collinearity (r2 > 0.5), and were therefore not retained.

A PCA was performed on the correlation matrix of stream concentrations and isotopic compositions for the ten selected tracers. The first three principal components explained 91.3 % of the variance in stream concentrations and isotopic compositions and were selected to generate a three-dimensional mixing space (U space, Table 2). Plots of residuals of each solute plotted against observed concentrations and isotopic compositions suggested that three components were needed to obtain a well-defined mixing subspace. End-member tracer concentrations and isotopic compositions were then projected into the mixing space (Fig. 7). All stream water samples are plotted inside the mixing domain defined by the end-members. Rainfall, throughfall, soil water and soil water from the riparian zone end-members are plotted in the upper right quadrant of the U1–U2 mixing space (Fig. 7a). Shallow groundwater samples were located in the lower left quadrant and snow in the lower right quadrant. Overland flow is plotted in the upper left quadrant and was located furthest away from stream water samples and with the largest interquartile ranges. Most of the stream water samples were clustered in the immediate vicinity of the soil water from the riparian zone samples, half-way between the throughfall and the groundwater samples. Snow seems to contribute to some stream water samples that are placed slightly more toward the lower right quadrant (Fig. 7a). The large distance between stream water and overland flow samples suggests a minor role of the latter in total runoff generation. Event peakflow samples are highlighted in Fig. 7b. In general, results show that when the catchment was wet, there was a higher contribution of groundwater to streamflow (events 1–2 and 10–11) than when the catchment antecedent condition was dry (events 3–9). However, compared to winter (events 1–2), a much higher contribution of throughfall was estimated during summer (events 5–8), when the pre-storm catchment state was dry.

In order to better understand water pathways during each event separately, we plotted stream water samples collected for each event and end-member tracer signatures in the previously determined two-dimensional mixing space (Figs. 8 and 9). We accounted for end-member temporal variability by plotting not only end-member samples collected the same month as the event occurred, but also the preceding and following months. Groundwater and rainfall signals remained relatively constant throughout the year, whereas throughfall, riparian and soil water presented higher temporal variability. Results showed that runoff mixing patterns changed between events. During autumn and winter, when the catchment was wet (events 1–2, and 10–11), stream water signal composition was most similar to riparian, soil water and groundwater. Only samples collected during the rain-on-snow event (event 2) might have a small contribution of not only overland flow but also snow. Mixing patterns changed during spring and summer when the catchment was drier (i.e. events 3 to 9). As previously seen in Fig. 7b, groundwater seems to have a much lower contribution to stream water, since stream water samples are now plotted in an intermediate position between throughfall and soil water from the riparian zone (with the exception of event 3, which still has a significant groundwater contribution). Note that overland flow did not occur and the soils were dry during these spring and summer events.

The qualitative and semi-quantitative analysis of diatom microflora revealed 230 taxa in the Weierbach catchment. Diatom communities from samples collected during the seasonal campaigns in the streambed (i.e. epilithon, epipelon and stream water samples) during low flow were usually composed of species from oligotrophic environments, mainly occurring in water bodies, but also rather regularly on wet and moist surfaces (i.e. the riparian zone hydrological functional unit of Pfister et al., 2009), such as Achnanthes saxonica Krasske, Achnanthidium kranzii (Lange-Bertalot) Round & Bukthiyarova, Fragilariforma virescens (Ralfs) D. M. Williams & Round, Eunotia botuliformis F. Wild, Nörpel & Lange-Bertalot, and Planothidium lanceolatum (Brébisson) Lange-Bertalot. Important seasonal changes in the relative abundance of aerial diatoms amongst the sampled habitats were not observed (Table 3). The null hypothesis of equal distributions was tested with the Mann–Whitney U test for the samples from the riparian zone and the hillslope (too small an amount of stream water at low flow and streambed samples). P values were high (0.21 and 0.73 for the riparian zone and the hillslope samples, respectively) and the null hypothesis was accepted. No diatom valves were found in groundwater or rainfall samples.

The riparian zone was characterized by several species that prefer aerial habitats, mainly living on exposed soils or epiphytically on bryophytes. Such species occur mainly in wet and moist or temporarily dry places or live nearly exclusively outside water bodies (categories 4 and 5 of Pfister et al., 2009), such as Chamaepinnularia evanida (Hustedt) Lange-Bertalot, C. parsura (Hustedt) C. E. Wetzel & Ector, Eunotia minor (Kützing) Grunow, Hantzschia abundans Lange-Bertalot, Nitzschia harderi Hustedt, Orthoseira dendroteres (Ehrenberg) Round, R. M. Crawford & D. G. Mann, Pinnularia borealis Ehrenberg, P. perirrorata Krammer, Stauroneis parathermicola Lange-Bertalot and S. thermicola (J. B. Petersen) J. W. G. Lund.

Diatoms were completely absent in samples from dry litter on the hillslope and only occurred on bryophytes. Almost no diatoms were found in overland flow samples. The relative abundance of aerial valves was higher in hillslopes and riparian samples compared to streambed samples (Table 3). However, we found a higher number of aerial diatoms (in absolute numbers) in the riparian zone. This emphasizes the importance of the riparian zones as the main terrestrial diatom source during rainfall, when diatoms are mobilized from moist or temporarily dry habitats into the stream channel (Table 3).

A series of 11 rainfall events were sampled from November 2010 to December 2011 during both wet and dry catchment conditions (Table 4 and Fig. 2). The main aerial species found in stream water during storm events were as follows: Chamaepinnularia evanida, C. obsoleta (Hustedt) C. E. Wetzel & Ector, C. parsura, Humidophila brekkaensis (J. B. Petersen) R. L. Lowe et al., H. perpusilla (Grunow) R. L. Lowe et al., Eolimna tantula (Hustedt) Lange-Bertalot, Eunotia minor, Pinnularia obscura Krasske, P. perirrorata, Stauroneis parathermicola, and S. thermicola.

Stream water samples taken throughout storm hydrographs showed a systematic increase in aerial diatoms as a response to incident precipitation and increasing discharge (Figs. 8 and 9). During events, the minimum increment of aerial valves' relative abundance was 8.1 % (event 2), whereas the maximum increment was 27 % (event 11). The maximum percentage of aerial valves was 43.5 % (event 10).

No significant relationship was found between the percentage of aerial diatoms and instantaneous discharge (r2=0.13, n=101; discharge on the x axis), most probably due to different diatom abundances on the rising limb of the hydrograph than on the recession limb (i.e. hysteretic effects). Two events showed clockwise hysteretic loops (events 1 and 2); five events showed counter-clockwise hysteretic loops (events 4, 5, 6, 8, and 10) and three showed figure-eight shaped hysteretic loops (events 7, 9 and 11). Although a clear pattern was not observed, results suggest that clockwise hysteretic loops predominated during wet conditions (the greater percentages of aerial diatoms in streamflow were immediately before peakflow), and counter-clockwise hysteretic loops during dry conditions (the greater percentages were immediately after peakflow).

Aerial valves comprised less than 15 % of the total diatoms in low flow samples for all events except 6, 9 and 10 (which had 19.2, 17.1, and 25.6 %, respectively). Due to technical problems, no low-flow sample was collected for event 3. No relationship was observed between antecedent event rainfall and the percentage of aerial valves observed during low flow (n=10, r2=0.08 and 0.09 for 10 and 20 days of antecedent rainfall, respectively).

At event scale, there were significant correlations between maximum percentage of aerial diatoms and event rainfall and maximum event discharge (r2=0.54, p < 0.05, n=10, Fig. 10a; r2=0.76, p  < 0.05, n=10, Fig. 10b, respectively; the multi-peak event sampled in December 2011 was considered as an outlier). High percentages (> 35 %) of aerial diatom relative abundance were measured during dry catchment conditions, compared to when the catchment was wet, where maximum relative abundances were low (< 15%). Alternatively, higher maximum percentages of aerial diatom proportions (> 35%) were measured during dry catchment conditions, when events were shorter and more intense.

A significant correlation between percentage of aerial diatoms with UV absorbance at 254 nm was found (r2=0.55, p < 0.05, n=76, Fig. 10c). During rainfall events in the Weierbach catchment, the relative abundance of aerial diatoms was associated with increased organic matter concentrations in the stream. A similar trend was observed with K+ (r2=0.25, p < 0.05, n=76), which is also associated with organic matter content. The relative abundance of aerial diatoms was not correlated with any other tracers.

Our central hypothesis for this study was that aerial diatoms could indicate connectivity within the HRS system. In order to test this hypothesis, we sampled from potential upland catchment sources (i.e. hillslope and riparian zones), and within the streambed (i.e. epilithon, epipelon and stream water samples).

Before testing our central hypothesis, we tested for the existence of distinguishable diatom species assemblages on the hillslope, the riparian zone and the stream. Only if diatom assemblages are distinguishable between these zones can their presence in the channel during rainfall events serve as a proxy for HRS connectivity. Results showed clear differences in diatom species assemblages between the hillslopes, riparian zone and streams, with higher relative abundance of aerial diatoms on the hillslopes and in the riparian zones compared to the stream (Table 3). Diatoms are usually abundant in moist environments (Van de Vijver and Beyens, 1999; Nováková and Poulíčková, 2004; Chen et al., 2012; Vacht et al., 2014), but in spite of the presence of diatoms in bryophyte-covered areas of the hillslopes, we did not find any diatom valves in hillslopes covered by dry litter. Moreover, the quantities of aerial diatoms found on the hillslopes covered by bryophytes and in the overland flow gutter samples were small and sometimes not sufficient to fully characterize the zone (due to the rarity of some species but also linked to sampling difficulties). This constrained the use of aerial diatoms to infer hillslope-riparian zone connectivity in some parts of the Weierbach catchment because of a limited diatom reservoir on hillslopes.

Despite the highest relative abundance of aerial valves on the hillslope compared to the riparian zone, the riparian zone was still the largest aerial diatom reservoir (in absolute numbers) with the highest probability of connecting to the stream (Table 3). We did not observe significant seasonal differences in diatom species assemblages among the different sampled habitats.

We examined the aerial diatoms transported in the stream water during runoff events. We observed an increase in the relative abundance of aerial diatoms with discharge for all sampled events regardless of antecedent wetness conditions. Hence, during storm events there was an increase in the relative proportion of diatoms in categories 4 and 5 of Van Dam's et al. (1994) classification. Similar results were reported by Pfister et al. (2009). These observations imply hydrological connectivity between the riparian soil surface and the stream for all events. The use of aerial diatoms to infer hydrological connectivity in the Weierbach catchment thus remains limited to the riparian-stream system as no diatoms were found on the hillslopes covered by dry litter.

Even though aerial diatoms do not live in microhabitats with flowing water, they were found in stream water samples during low flow conditions preceding storm events (Table 3). This indicated that the “stock” of aerial diatoms in the catchment before the sampled events was not completely exhausted during previous events. Similar conclusions were drawn by Coles et al. (2015), who examined diatom population depletion effects during rainfall and found that while aerial diatom populations in the riparian zone were depleted in response to rainfall disturbance, rainfall was unlikely to completely exhaust the diatom reservoir.

We hypothesize that the transport of diatoms from the riparian zone to the stream might take place either through (i) a network of macropores in the shallow soils of the riparian zone or (ii) overland flow in the riparian zone. The potential for diatoms to be transported through the subsurface matrix was investigated using fluorescent diatoms and soil columns by Tauro et al. (2015). Results demonstrated that sub-surface transport of diatoms through the sub-surface matrix was unlikely. However, the potential for transport of diatoms through heterogeneous macropore networks remains unexplored. The increased relative abundance of aerial diatoms in the stream event water could also be explained by as yet undocumented surface or near-surface pathways.

Two-component hydrograph separation and EMMA provide valuable information on water sources and flowpaths. Using these methods we learned that in the Weierbach catchment, during spring and summer, the hydrological response was largely composed of event water (see an example of dry antecedent catchment conditions in Fig. 4b). Similar conclusions were drawn by Wrede et al. (2014) using dissolved silica. Accordingly, EMMA results suggest canopy throughfall, rainfall and riparian soil water were the main water sources (Figs. 8 and 9). As observed in other headwater catchments (e.g. Penna et al., 2011), discharge likely increased due to channel interception and riparian runoff leading to clear and singular hydrograph peaks (Fig. 4b). During fall and winter, when the catchment was at its wettest state, double peaked hydrographs characterized the event hydrological response. Hydrograph separation indicated that the first peak was mainly event water and the delayed, second peak was mostly pre-event water (Fig. 4a; Wrede et al., 2014). During these events, soil water, groundwater, and throughfall contributed substantially to total discharge (Figs. 8 and 9). Hillslope overland flow was insignificant during most sampled events. Only for event 2 – the largest storm on record – was overland flow a significant contributor to stream discharge, likely due to rapid snowmelt onto a surface-saturated area (Figs. 8 and 9).

During all sampled events the relative abundance of aerial diatoms increased with discharge indicating hydrological connectivity between the riparian zone and the stream. These findings are consistent with the hydrograph separation results. Aerial diatoms could reach the stream as saturated areas expand during rainfall events. Accordingly, we found a significant correlation between percentage of aerial diatoms with UV absorbance (proxy of DOC). DOC concentrations associated with runoff storm often come mainly from the near-stream riparian zones (Boyer et al., 1997). Controls on surface saturated and subsurface mixing processes are currently being investigated in the Weierbach riparian zone using infrared imagery and groundwater metrics (Pfister et al., 2010).

Hydrological connectivity between hillslopes and the stream has also been previously defined by water table connections between the hillslope and the riparian zone (Vidon and Hill, 2004; Ocampo et al., 2006; Jencso et al., 2010; McGuire and McDonnell, 2010). While our results showed that overland flow did not occur on hillslopes during most sampled events, the VWC measurements and timing of the hydrograph response suggest that subsurface hydrological connectivity along the HRS system occurs during wet catchment conditions (Fig. 3). Hence, if aerial diatoms found on the hillslopes, might reach the stream through sub-surface flowpaths remains unknown. Others have demonstrated that tracer transport can occur on larger timescales that extend beyond individual events (McGuire and McDonnell, 2010). Whether this may also be true for diatoms remains to be explored.

Storm hydrograph separation using stable isotope tracers has resulted in major advances in catchment hydrology. However, despite their usefulness, these methods do not provide unequivocal evidence of hydrological connectivity in the HRS system. In comparison, diatoms can provide evidence of riparian-stream connectivity. Further research is needed to better understand diatom transport processes (and associated water flowpaths) in headwater catchments. Future studies should focus on expanding our understanding of terrestrial diatom taxonomy and ecology, which are scarce or lacking for a large number of taxa (Wetzel et al., 2013, 2014). Even though this new data source will have its own individual measurement uncertainty (McMillan et al., 2012), diatoms offer the possibility to tackle open questions in hydrology and eco-hydrology.

A key issue with the concept of hydrological connectivity is how it can be applied across and between environments. Uncertainties increase when applying two-component hydrograph separation at large scales. For instance, Klaus and McDonnell (2013) note that quantifying the spatial variability in the isotope signal of rainfall and snowmelt can be difficult in large catchments and in catchments with complex topography. Similarly, some studies showed that, for meso-scale catchments, only qualitative results of the contribution of a runoff component can be obtained by the hydrograph separation techniques (Uhlenbrook and Hoeg, 2003). For aerial diatoms to be useful and a way forward to increase our understanding of hydrological pathways at a range of scales, they must be also relevant across environments and scales (Bracken et al., 2013). The current concepts related to HRS connectivity are best suited to humid, temperate settings (Beven, 1997; Bracken and Croke, 2007) and represent only very specific settings (Bracken et al., 2013). Previous investigations in Luxembourg have shown that freshwater diatom assemblages in headwater streams have regional distributions strongly affected by geology, as well as anthropogenic factors (e.g. organic pollution sources and eutrophication) (Rimet et al., 2004). Hence, we speculated that diatoms have potential in headwater systems, and at larger catchment scales to determine connectivity between contrasting geological zones.

The need to account for the temporal variability in end-member chemistry and to collect high-frequency data on both – stream water as well as potential runoff end-members – has been well recognized (Inamdar et al., 2013). As noted by Tetzlaff et al. (2010), seasonality should also be considered when using living organisms to trace water flowpaths. Diatom end-members must be sampled seasonally in order to ensure that populations have not undergone demographic changes. Indeed, this increases the sampling needs and the overall laboratory procedures of an already time-consuming approach (i.e. sampling, pre-treating the samples, mounting permanent slides and diatom identification). A potential alternative to reduce processing time is to develop new techniques such as to dye diatom valves and use them to trace water flowpaths (see Tauro et al., 2015). The use of dyed diatoms under field conditions for experimental hydrology remains unexplored.