3.1 Observation of the groundwater table

We installed 36 wells and seven piezometers (Figure 1(a)). We placed the piezometers directly into stream channel boreholes drilled with a percussion hammer (Cobra TT, Eijkelkamp, Netherlands). Wells were drilled with a portable drilling system (Gabrielli & McDonnell, 2012) down to fresh bedrock and cased the wells with a 4 cm diameter PVC pipe screened at the bottom (Table A2.1). We filled the space between the borehole and the pipe with gravel and sealed the borehole with bentonite at the top. We observed the water level every 15 min at 22 of the 36 wells with a water level sensor (Orpheus Mini, OTT, Kempten, Germany, resolution of 1 mm and accuracy of ±0.05% FS) and approximately biweekly in all wells and piezometers via manual measurements. Measurements started in July 2018 and continued until February 2020. Inflow into the study section was measured using a steam gauge (Figure 1(a)) with a pressure transducer (ISCO 4120 Flow Logger). Discharge was derived from a water level – discharge rating curve that was built with 73 discharge measurements (2011–2021) with graded buckets and tracer injections. Water level was continuously monitored with a pressure transducer (ISCO 4120 Flow Logger) at a V notch weir (Figure 1).

3.2 ERT measurements

Electrical resistivity tomography (ERT) measurements were carried out for identifying the depth of the interfaces between solum-subsolum and subsolum-fractured bedrock. The ERT measurements were conducted along eight transects in the stream corridor and on the adjacent footslopes (Figure A1.1). We used a resistivity metre (IRIS Instruments, Syscal Pro 120, ten-channel) with multicore cables equipped with 120 stainless steel rod electrodes with 50 cm spacing increments. Data was processed following the approach outline by Gourdol et al. (2021) that surveyed the Weierbach catchment. The recorded resistivity data were pre-processed, removing values with a potential lower than 10 mV. The cleaned resistivity data was inverted using the Boundless Electrical Resistivity Tomography inversion code (Günther & Rücker, 2016). ERT images were processed to extract subsurface regolith layers through the derivative method. It was assumed that the subsurface regolith interfaces are located where the resistivity changes in space were at a maximum (Chambers et al., 2014). The changes in resistivity can be obtained by using the first or the second derivatives, targeting maximum gradients or zero values (Gourdol et al., 2021; Sponton & Cardelino, 2015). Zero values of the second derivative of the logarithm of resistivity were identified using Paraview (version 5.5.2, Kitware, Inc. Ahrens et al., 2005) to obtain the location of the interface between two different subsurface layers (Hsu et al., 2010). The two interfaces so obtained mirror the solum-subsolum-fractured bedrock morphology of the Weierbach regolith of previous studies (Glaser et al., 2016; Gourdol et al., 2021; Moragues-Quiroga et al., 2017). Starting from the location of the interfaces obtained from the ERT profiles, we derived the elevation of the solum-subsolum interface and the subsolum-fractured bedrock interface for every well and piezometer at the study site through linear interpolation (Figure A1.1 and Table A2.1). We were not able to clearly identify subsurface interfaces in the riparian wetland due to the relatively low resistivity of the Leptosols in this area. This is because the distinction between solum and subsolum is not reliable when resistivity contrast is low (Gourdol et al., 2021). To overcome this problem, a hand-drilling campaign was carried out to determine solum-subsolum-fractured bedrock interfaces in the riparian wetland (details in Data S1).

3.3 Analysis of precipitation events and groundwater flow direction

Precipitation was measured using a tipping bucket rain gauge located 3.5 km from the site. Individual events were defined with at least 1 mm of precipitation and separated by at least 2 h without precipitation. For every event, we calculated: groundwater response time (defined as the lag time between the beginning of the precipitation event and a vertical 1 cm rise of the water table), streamflow response time (defined as the lag time between the beginning of the precipitation event and an increase of 0.1 L/s of streamflow), total precipitation depth, event intensity, and number of days without precipitation prior to the event. For every continuously monitored well, we evaluated the correlation between level increase and the response time with the initial groundwater level and precipitation characteristics (i.e., depth, intensity and number of antecedent dry days). We correlated the spatial differences in the average groundwater response time and increase per well with the regolith depth above the fractured bedrock and with the distance from and elevation above the streambed using the Spearman rank correlation coefficient (Rs). Significance was evaluated with the Mann–Whitney test (significance level: p-value < 0.05).

We computed groundwater flow directions by assuming that they were equal to the slope of a planar groundwater table determined by three adjacent wells. We quantified the direction by angle α (degrees) on the xy plane (cf. Rodhe & Seibert, 2011). The stream is oriented with −72° on the xy plane (cf. Figure 2(b)). α-values pointing towards the stream channel indicate that the stream is gaining conditions, whereas α pointing away from the channel suggests losing conditions. The subsurface groundwater flow direction was calculated every 15 min. For the wells equipped with continuous water level loggers (Figure 2(a), 29 triangles), and for all wells every 2 weeks (Figure 2(b), 65 triangles). For every triangle we also derived the direction of the fractured bedrock fall line (defined as the direction on the xy plane [degrees] of the slope of fractured bedrock surface) and the direction of the surface topography fall line (defined as the direction on the xy plane [degrees] of the slope of the topographic surface) using the same approach used for obtaining the direction of the groundwater table. Data analysis has been conducted with Matlab R2020a (The Mathworks, Natick, MA).

4.1 Groundwater and streamflow dynamics

Streamflow varied widely (Figure 3, arithmetic mean of 6.5 L/s, median of 1.7 L/s, interquartile range of 9.1 L/s, St.Dev. of 11.52 L/s), with extended no-flow periods during summer (no discharge for a total of 194 days during the study period), persistent streamflow during winter and spring, and variable event responses. We defined three different hydrologic conditions (dry, intermediate, wet) based on streamflow and groundwater levels (shown as grey shades, Figure 3; details in Table 2) to classify groundwater behaviour.

We defined wet conditions when discharge Q exceeded 0 L/s for 14 consecutive days regardless the groundwater elevation in the wells.

Intermediate conditions were only defined when they immediately occurred before or after the wet conditions. Intermediate conditions were specified when Q = 0 or Q > 0 L/s lasted less of 14 consecutive days, and when the monitored groundwater in at least 17 (75%) of the wells was above the subsolum-fractured bedrock.

Dry conditions were specified when Q = 0 or Q > 0 L/s lasted less of 14 consecutive days and when the monitored groundwater in less than 17 (75%) of the wells was above the subsolum-fractured bedrock.

Dry conditions (dark grey shading, Figure 3, 28.7% of the observation period) (Table 2) had no-flow for 95.3% of the time. The average streamflow in this period was 0.01 L/s. Groundwater (Figure 3(b)) displayed flashy and short-lived increases after precipitation. The groundwater table in the east and west footslopes (wells 7W1 and 9W6, Figure 3(b)) showed larger increases and faster recessions than tables in the riparian wetland (well 9W6, Figure 3(b)). During intermediate conditions (light grey shading, Figure 3, 11.0% of the observation period, Table 2), precipitation occurred more frequently and with higher amounts; no-flow persisted for 67.4% of the time (Table 2) and average streamflow was 0.12 L/s. During wet conditions (white shading, Figure 3), which covered 60.3% of the observation period (Table 2), streamflow was persistent (average discharge was 10.8 L/s with a maximum of 118 L/s).

To quantify the drivers of groundwater dynamics, we analysed the average response time and increase after precipitation events for groundwater and streamflow (Figure 4). During dry conditions, a groundwater response was triggered in all wells by small precipitation depths (<3 mm for 21 of the 22 wells; median = 2.2 mm, min = 1 mm, max = 3.5 mm, Figure 4(a)). This groundwater rise occurred a few hours after the onset of precipitation (<3 h, for 17 of the 22 wells; median = 2.48 h, min = 1.21 h, max = 5.15 h, Figure 4(d)). If streamflow was generated following precipitation, it appeared several hours after the rise of the groundwater table (between 0.25 and 2.95 h; median = 1.75 h, Figure 4(g)). During dry conditions and before precipitation events (Figure 5(a)), the groundwater table in the footslopes was below the streambed elevation (Figure 5(d)) in the fractured bedrock (Figure 5(g)). The groundwater table in the riparian wetland was above the subsolum-bedrock interface and – for some wells – above streambed elevation (Figure 5(d),(g)). In response to precipitation events, groundwater in the footslopes rose rapidly from the fractured bedrock into the subsolum – and in a few instances – above streambed elevation. After events, the groundwater level decreased rapidly towards the pre-event level.

During intermediate conditions, the minimum precipitation depth necessary for a groundwater response was <3 mm for 21 of the 22 wells (median = 1.5 mm, min = 1.3 mm, max = 3.37 mm, Figure 4(b)). Groundwater rise usually occurred less than 5 h after the beginning of a precipitation event for most wells (median = 3.56 h, min = 2.17 h, max = 6.53 h; Figure 4(e)). In response to precipitation, streamflow increased (or re-appeared) almost synchronously with the groundwater table (Figure 4(h)). During intermediate conditions, the groundwater table was above the fractured bedrock and rose within the subsolum layer after the events and decreased to the pre-event level within 2–3 days (Figure 5(h)). The groundwater in the riparian wetland was always above the streambed elevation before and after precipitation events, while the groundwater in the footslopes was mostly at the level of the streambed (Figure 5(e)).

During wet conditions, the groundwater table was always above streambed elevation (Figure 5(f)) and in the upper subsolum (Figure 5(i)). The groundwater table variation in the footslopes mirrored streamflow variations (Figure 2(a),(b)), while the groundwater table in the riparian wetland was more stable with less frequent and smaller peaks (Figure 2(b)). The precipitation depth necessary to trigger an increase in the groundwater level was highest in wet conditions (median = 3.24 mm, min = 1.81 mm, max = 4.1 mm, Figure 4(c)) and the groundwater response to precipitation events was more delayed compared to dry and intermediate conditions (median = 5.58 h, min = 2.35 h, max = 12.68 h Figure 4(f)). In contrast to dry and intermediate conditions, streamflow response occurred before groundwater response (median = 4.54 h, min = 1.26 h, max = 11.41 h, Figure 4(i)). In the riparian wetland, the groundwater table reached the surface in several wells (8.5%, 6.8%, 14.6%, and 50.3% of the time for wells 7W2, 7W3, 9W3, and 10W2, respectively). Some wells displayed artesian behaviour for extended periods of time (46.5% and 14.9% of the time for wells 7W3 and 10W2, respectively).

We conducted a correlation analysis for every well to determine whether precipitation characteristics and initial groundwater table elevation drove temporal variability of groundwater dynamics in different hydrologic conditions (Figure 6 and Data S3). The increase in the groundwater level was always positively correlated with precipitation depth (Figure 6(a)), and it was positively correlated with precipitation intensity for the whole study period and for intermediate and wet conditions (Figure 6(b)). The increase in the groundwater level was also negatively correlated with the initial groundwater level for most wells during the whole study period (Figure 6(d)). We found no significant correlation between groundwater increase and number of antecedent dry days (Figure 6(c)). The observed groundwater response time was always negatively correlated with precipitation intensity (Figure 6(f)) and the correlation was significant for the whole study period and for intermediate and wet conditions. For the whole study period, the groundwater response time was also significantly positively correlated with initial groundwater levels for most wells (Figure 6(h)). We found no significant correlation between groundwater response time and precipitation depth or number of antecedent dry days (Figure 6(e,g)).

We conducted a correlation analysis to determine whether the regolith thickness and distance from and elevation above the stream channel drove spatial variability in groundwater dynamics for different hydrologic conditions. During wet conditions, the average groundwater increase and response time were significantly positively correlated with regolith thickness above the fractured bedrock and elevation above and distance from the streambed. During dry conditions, the average groundwater increase and response time were respectively significantly correlated with elevation above the streambed and regolith thickness above the fractured bedrock (Data S3).

4.2 Spatiotemporal dynamics of groundwater flow direction

α showed spatial patterns with clear differences between the east footslope, west footslope and the riparian wetland. The patterns are illustrated for two sets of 2 days using manual level measurements before and after precipitation events (Figure 7). The first example of 2 days relates to dry conditions (Figure 7(a),(c),(e)). Before the event (falling limb, 26 August 2019), α pointed towards the stream channel on the east footslope and towards the hillslope on the west footslope (Figure 7(e), red arrows). Observed groundwater levels were mostly below the streambed (median_{GW_wells} = 7.2 cm and median_{GW_piezo} = 11 cm below the streambed) with groundwater in the stream channel generally at the same level or below groundwater in the footslopes (e.g., the level differences between the piezometers 4P1 and the adjacent well 4W1 was ΔGW_4P1-4W1 = −4.7 cm). In the riparian wetland, α pointed towards the stream channel. The event on 7 September led to an increase in the groundwater level throughout the study reach (median_{GW_wells} = 1.1 cm, median_{GW_piezo} = 1.8 cm below the streambed on 9 September 2019; Figure 7(a),(c),(e), blue arrows). As a result, groundwater in the streambed rose above the level of the adjacent groundwater in several sections (e.g., ΔGW_4P1-4W1 = 15.3 cm) and α pointed towards the footslopes in some sections of the stream corridor (Figure 7(e)).

The second example relates to wet conditions (Figure 7(b),(d),(f)). Before the event (falling limb, 25 February 2019), the groundwater table was above the streambed elevation (median_{GW_wells} = 7.8 cm, median_{GW_piezo} = 3 cm above the streambed in the wells). Yet, groundwater in some piezometers was above that of the adjacent wells (e.g., ΔGW_6P2-6W2 = 3.5 cm), and at the same level or below the streamwater level (e.g., ΔGW_6P2-SW = −1 cm). α pointed towards the hillslope on the west footslope and in a few locations close to the stream channel on the east footslope (Figure 7(f), red arrows). However, α pointed towards the stream on the east footslope upstream of the stream riffle and in the riparian wetland. This occurrence of both gaining and losing conditions in different sections of the stream was not persistent during wet conditions. After a series of precipitation events (total precipitation = 49 mm between 1 and 8 March 2019) the groundwater level increased and α pointed towards the stream channel throughout the study reach (8 March 2019, median_{GW_wells} = 16.6 cm, median_{GW_piezo} = 5.2 cm above the streambed; Figure 7(f), blue arrows). Groundwater in the piezometers was generally below the adjacent groundwater in the footslopes (e.g., ΔGW_6P2-6W2 = −6 cm) and at the same level or above the streamwater level (e.g., ΔGW_6P2-SW = 0 cm). An exception was the east footslope close to the riffle, where local gradients pointed away from the stream channel.

The analysis of groundwater flow directions with continuously-monitored wells is generally consistent with the biweekly data and provided clear short-term dynamics of α and the duration of the gaining and losing conditions for different sections of the stream corridor (Figure 8). On the east footslope, α pointed towards the stream channel in dry, intermediate, and wet conditions in most triangles for most of the time (Figure 8(a),(d),(g)). However, during intermediate and dry conditions, α occasionally pointed away from the stream channel (Figure 8(a),(d)). On the west footslope, the groundwater flow direction towards the stream channel became relatively more important compared to the groundwater flow direction pointing away the stream channel with increasing wetness conditions (Figure 8(b),(e),(h); Tr2, Tr19). Here, spatial differences were observed in triangles closer to the stream, where α pointed mostly parallel (≃ − 72°) to the stream channel (Figure 8(b),(e),(h); Tr17), and in triangles closer to the riparian wetland, where α pointed constantly towards the stream channel (Figure 8(b),(e),(h); Tr4). In the riparian wetland, α nearly continuously pointed towards the stream channel for different hydrologic conditions (Figure 8(c),(f),(i)) with some exceptions close to the stream (cf. Triangle 22).

The continuously observed groundwater wells allowed us to capture the behaviour of α during the transition from dry to wet conditions (Figure 9(a)–(d)) and during events (Figure 9(e)–(l)). The transition was accompanied by a gradual change in α towards the stream on the east (Figure 9(b)) and west (Figure 9(c)) footslopes and partly in the riparian wetland (Figure 9(d)). α also varied between the directions of the fall line of the surface topography and the direction of the fall line of the factured bedrock surface for the majority of the continuously monitored wells.

The continuous measurements offered insights into event-based changes of α. During dry conditions, precipitation events were followed by sporadic re-appearance of the streamflow in the channel (from Q = 0 to Q > 0 L/s for 3 h Figure 9(a),(e)) and a shift of groundwater flow direction on the east footslope, with α pointing towards the stream between events and towards the hillslope after events (Figure 9(b),(g)). On the west footslope, α always pointed towards the hillslope (Figure 9(c)) and was almost perpendicular to the stream channel (α ~ −160°) after events (Figure 9(i)). During wet conditions, precipitation events were followed by an increase of streamflow (Figure 9(f)) and a more pronounced groundwater flow direction towards the stream. While this variation did not considerably change α on the east footslope (Figure 9(b),(h)), it caused a change in the groundwater flow direction on the west footslope. Here, α shifted from pointing towards the footslope during recessions to pointing towards the stream during and few days (up to 3 days) after events (Figure 9(c),(j)). In the riparian wetland, α always pointed towards the stream channel (Figure 9(d)) and approached the direction of the surface fall line after sporadic precipitation events in dry conditions (Figure 9(k)) and persistently during wet conditions (Figure 9(l)).

5.1 Drivers of near-stream groundwater dynamics

5.2 Drivers of near-stream groundwater flow directions depending on hydrologic conditions

5.3 Implications for runoff generation and hydrological connectivity

Without additional data on groundwater flow direction, one may have interpreted the observed response of groundwater before streamflow response in dry conditions at the study site as groundwater contributing to streamflow generation. Such an interpretation would be in agreement with hillslope-stream connectivity studies, which concluded that streamflow generation is driven by groundwater inflow when groundwater responds to events before streamwater (Beiter et al., 2020; Haught & Van Meerveld, 2011; Rinderer et al., 2016). However, α and the detailed pattern of the groundwater level, that was consistently below the streamwater level, clearly showed that groundwater flow direction pointed towards the hillslope during and after precipitation events in dry conditions. This is evidence for the lack of groundwater contributions to streamflow generation in the study reach.

α and groundwater level above the streamwater level jointly revealed that groundwater contributes to streamflow generation both before and after precipitation events during wet conditions in most of the study reach. This is despite the fact that a response of groundwater after streamwater was previously linked to a lack of groundwater contribution to the streamflow generation (Beiter et al., 2020). The nearly instantaneous streamflow response in the Weierbach following precipitation largely consists of event water in dry and wet conditions (Wrede et al., 2015). This has commonly been interpreted as surface runoff generated in saturated areas or by direct precipitation in the stream channel (Glaser et al., 2016; Rodriguez & Klaus, 2019; Schwab et al., 2018) and is consistent with the spatial extent of saturated areas (Glaser et al., 2020) and the observations in the current study.

The high spatial density and high-frequency observations across hydrologic conditions in this study showed the potential pitfalls of using only one or a few wells for characterising runoff generation or hydrological connectivity. Using groundwater flow direction proved useful to avoid misinterpretations of hillslope-stream connectivity derived from time lags, as the lag time between the streamwater and groundwater response to events was interpreted contrary to our results in the same hydrogeological setting of the Luxembourg Ardennes (e.g., Beiter et al., 2020).