The Kongsfjord basin, located in northwestern Svalbard, covers a total land area of approximately 1430 km2, of which about 75 % is currently glacier-covered. The basin surrounds Kongsfjorden, which stretches over ∼ 20 km towards the Arctic Ocean to the west. This study centres on three adjacent glacier systems that terminate in the inner part of Kongsfjorden (Fig. 1a): Isachsenfonna–Kongsbreen, Holtedahlfonna–Infantfonna–Kronebreen, and Kongsvegen.

At the northeastern fjord head, the ice field Isachsenfonna drains through the outlet glacier Kongsbreen (together 378 km2), which splits due to a bedrock ridge into two branches, Kongsbreen North and Kongsbreen South. To the east, the terminal glacier Kronebreen is fed by two contributory catchments: the larger ice field Holtedahlfonna and the smaller Infantfonna (together 373 km2). Roughly 5 km from its front, Kronebreen is confluent with Kongsvegen (108 km2), and the two glacier tongues share a common terminus which is currently dominated by Kronebreen (Sund et al., 2011).

Surface elevation (Fig. 1a), basal topography (Fig. 1b), and ice thickness of the Kongsfjord basin were recently mapped by airborne and ground-based ice-penetrating radar (Lindbäck et al., 2018). Together with their accumulation areas, the Kongsfjord glaciers span an altitudinal range from sea level to 1400 m a.s.l. (above sea level) in the northern parts of their catchments (Fig. 1a). Below the equilibrium line altitude, which for Holtedahlfonna–Kronebreen was estimated around 610 m a.s.l. for the period 1961–2012 (Van Pelt and Kohler, 2015), Kronebreen is heavily crevassed, and its slope is disrupted by several ice falls. In the ablation area, the subglacial topography is characterized by extended troughs as deep as -180 m a.s.l. under Kongsbreen North, -130 m a.s.l. under Kronebreen, and -70 m a.s.l. under Kongsvegen (Fig. 1b). Both Kronebreen and Kongsvegen are fully grounded below sea level over their lowermost 10 and 9 km, respectively, whereas Kongsbreen South is mostly resting on bedrock above the water line. A few nunataks are present in the lower parts of the glaciers at approximately 500–600 m a.s.l. Further inland, the subglacial topography smoothens into wide, flat valleys around 200 m a.s.l., and ice thickens to a maximum of 740 m in the middle of Isachsenfonna.

Like most tidewater glaciers in Svalbard, the Kongsfjord glaciers have a negative mass balance dominated by calving, a finding that both in situ measurements and modelling experiments have corroborated over the last decades (e.g. Lefauconnier et al., 1994b; Nuth et al., 2012). More recently, Van Pelt and Kohler (2015) used a coupled surface-energy-balance–snow model to include superimposed ice and internal accumulation in their surface mass balance calculations. For Kronebreen and Kongsvegen, they found that surface mass balance is controlled by mass gain through precipitation (0.87 m w.e. a-1) and mass loss through run-off (0.79 m w.e. a-1) and that the refreezing of water in the firn layer (0.30 m w.e. a-1) provides a strong buffer for run-off. Their results also reveal a 31 % increase in run-off since 2000, coinciding with higher surface melt rates and lower refreezing rates in response to recent warming and firn line retreat.

Kronebreen is one of the fastest-flowing glaciers in Svalbard (Liestøl, 1988; Sund et al., 2011). Near the terminus, mean annual surface velocities are persistently high (1.6–2.15 to m d-1) (Lefauconnier, 1987; Lefauconnier et al., 1994a; Kääb et al., 2005), and summer velocities peak up to 3.2 m d-1 (Schellenberger et al., 2015). Such high velocities can only be achieved through basal sliding, which is consistent with the finding of widespread temperate basal conditions at several glaciers in the region (Björnsson et al., 1996; Sevestre et al., 2015). Kongsbreen has a more complex velocity pattern due to its split into two separate terminal branches: the northern, marine-terminating branch of the glacier is fast-flowing, with a maximal speed of 2.7 m d-1 (measured in 2012, Schellenberger et al., 2015), whereas the southern, partially land-terminating branch is slow-moving. Both Kronebreen and Kongsbreen North exhibit clear seasonal flow variability, characterized by relatively stable and low velocities in autumn and winter and an acceleration synchronous with the melt season. Schellenberger et al. (2015) observed that, at both glaciers, a large part of the ice flow variability followed fluctuations in surface meltwater production and rainfall, a correlation that they attributed to variations in basal lubrication. Evidence from later studies also strongly suggest that multi-annual changes in ice flow at Kronebreen are controlled by changes in the subglacial hydrology (How et al., 2017; Vallot et al., 2017). Inter-annual variability in ice velocity and upglacier propagation of seasonal speed-up were both found to be higher at Kronebreen than at Kongsbreen North (Schellenberger et al., 2015), indicating different basal conditions at the two glaciers. Kongsvegen, a surge-type glacier that is currently in its quiescent phase, contrasts with the other Kongsfjord glaciers in its dynamics. Since its last surge in 1948 (Melvold and Hagen, 1998), the glacier has experienced very little motion (2–8 m a-1), negligible calving, and significant surface steepening (Hagen et al., 1999, 2005; Nuth et al., 2012).

Glaciers in Kongsfjord are polythermal, usually with basal temperatures at or close to the pressure melting point, meaning that water drains at their base all year through (Björnsson et al., 1996; Sevestre et al., 2015; Nuth et al., 2019). Glacial run-off is periodically released into the fjord, where it produces sediment-rich freshwater plumes that are of great importance for the glacio-marine dynamics and the ecosystem of Kongsfjorden (Svendsen et al., 2002; Everett et al., 2018).

In recent years, several locations of subglacial discharge have been identified at the Kronebreen–Kongsvegen terminal complex: one main outlet in the northern half of the glacier tongue and several secondary outlets near the Kronebreen–Kongsvegen junction (Trusel et al., 2010; Kehrl et al., 2011; Everett et al., 2018). Contrasts in plume timing and duration have been suggested to reflect local differences in subglacial drainage efficiency across glacier termini (Schild et al., 2016; Slater et al., 2017). How et al. (2017) monitored plumes at Kronebreen and combined the data with observations of borehole water pressure, supraglacial lake drainage, and surface velocities, as well as modelling of meltwater routing, to infer the morphology of the subglacial hydrological system during summer 2014. Their results indicated that meltwater was largely drained to the northern part of the terminus through a channelized system connected to the upper catchment area (Holtedahlfonna), hence the broader and more persistent plume on the northern side. By contrast, the intermittent activity of the southern plume during the 2014 melt season was attributed to distributed drainage of a smaller catchment area.

Spatio-temporal variations in basal properties at Kronebreen were also investigated by Vallot et al. (2017) through inverting surface velocities for 2013–2015 to determine a basal friction coefficient. Their modelling results supported conclusions drawn by the earlier studies, i.e. that subglacial hydrology organization is as much a key factor to basal sliding as are local characteristics such as bed topography and summer melt.

While the relationship between meltwater supply, subglacial hydrology, and ice velocity is well documented at Kronebreen (Schellenberger et al., 2015; How et al., 2017), similar knowledge is lacking at the neighbouring outlet glaciers (Kongsvegen, Kongsbreen North, and Kongsbreen South), and the subglacial hydrology at the upper ice fields (Isachsenfonna, Holtedahlfonna, and Infantfonna) remains to be examined.

Mapping of Kongsfjord's supraglacial drainage system was achieved using TopoToolbox, a collection of scripts to analyse flow pattern in digital elevation models (DEMs) (Schwanghart and Scherler 2014). TopoToolbox first seeks to determine surficial water movement within the DEM. Flow direction calculation is based on a single flow direction algorithm that examines elevation gradients between grid cells. Next, information on flow direction is used to derive (1) flow accumulation, i.e. the number of cells draining in each grid cell, (2) the drainage network, i.e. the estimated flow paths in the DEM, and (3) drainage basins, i.e. the upslope areas contributing to each grid cell in the drainage network. We generated a map of the supraglacial stream network, shown in Fig. 2c, and the watersheds associated with each moulin (Fig. 2c, d). The selection of the moulin locations is described in Sect. 3.3.2.

To simulate the subglacial hydrology of the Kongsfjord basin, we apply the Glacier Drainage System model (GlaDS) (Werder et al., 2013), a two-dimensional model developed to describe subglacial water drainage and which participated in the Subglacial Hydrology Model Intercomparison Project (SHMIP) (de Fleurian et al., 2018). GlaDS has been implemented in Elmer/Ice, an open source, finite-element model (Gagliardini et al., 2013), which in this study serves only as platform for the subglacial hydrology model. The implementation of GlaDS in Elmer/Ice is detailed in Gagliardini and Werder (2018).

The subglacial hydrology model GlaDS accounts for both inefficient distributed drainage and efficient channelized drainage to compute the evolution of the hydraulic potential at the glacier base. The distributed network of linked cavities is featured in the form of a continuous water sheet across the model domain which is fed by a source term at each node on the mesh. We define the source term from the sum of a space and time varying surface water input, as described in Sect. 3.4.2, and a uniform, steady basal melting of 10-3 m a-1 (∼ 3.17 × 10-11 m s-1). Discrete channels that form on the edges of the mesh elements constitute the efficient drainage system. Between each pair of nodes, channel segments develop from water exchange with the adjacent sheet, as well as from water input at moulins, which act as local sources. Moulins are shaped as vertical cylinders at some of the mesh nodes (see Sect. 3.3.2), and deliver water directly from the glacier surface to the basal channel network. Depending on their cross-sectional area and on the subglacial water pressure, moulins can store a certain volume of water, as revealed by tracer-experiments (Werder and Schuler, 2010). For a complete description of the equations solved by the GlaDS model, we direct readers to Werder et al. (2013).

The problem is solved on a mesh constructed by discretizing the domain using irregular triangulation with a mean edge length of 250 m, which yields 11 824 nodes in total. The mesh resolution is a compromise between numerical computation time and spatial precision of the model outputs. To ensure computational stability, the simulation time step does not exceed 1 d, and adaptive time stepping down to ∼ 1.5 min is applied. An additional gain of computation time is achieved through partitioning and parallel simulation of each model run.

As boundary conditions, channels are not allowed to develop along the outer limits of the model domain, and the hydraulic potential is set to zero at the glacier fronts because we consider that all three termini (Kongsbreen North, Kongsbreen South, and Kronebreen–Kongsvegen) are freely connected to the fjord. Indeed, by imposing the water pressure to be equal to sea pressure at the depth of the outlets, we have 1φ=φm+pw=ρwgzb+ρwgzsl-zb=0, where φ is the hydraulic potential, φm the elevation potential, pw the sea water pressure, zb the bed elevation, and zsl the sea level elevation (= 0 m a.s.l.).

The values for the model parameters are set out in Table 1. Parameters in bold are those to which we assigned values different from Werder et al. (2013). We adapted values for conductivity parameters, as well as for sheet width below the channel, to facilitate channelization and to maximize the influence of meltwater input on the subglacial drainage efficiency. More specifically, our value for the sheet conductivity is lower than the one originally proposed by Werder et al. (2013), but inspection of current literature revealed that others find better agreement to observations by using lower values (e.g. Dow et al., 2018, 2020) or even suggest a seasonally variable conductivity (Downs et al., 2018). We performed sensitivity tests and found that lower sheet conductivities lead to higher water pressures all year round, whereas the original value from Werder et al. (2013) yields unrealistically low water pressures in winter. We also found our model results to be largely insensitive to changes in channel conductivity. In addition, we set a uniform basal sliding speed of ∼ 315 m a-1 (10-5 m s-1) for the entire domain, which corresponds roughly to the glacier-averaged surface speed measured by satellite imagery in the Kongsfjord basin (Schellenberger et al., 2015).

In order to investigate the sensitivity of our subglacial hydrology model to water recharge distribution, we design four model experiments with various degrees of complexity in the way that water is delivered to the glacier beds. The control experiment is the simplest scenario, in which basal water supply is spatially continuous and identical to surface meltwater production. The next two experiments assume discrete water recharge to the basal drainage system through moulins, with the latter including only water that is drained in moulin catchments. Finally, the fourth and most realistic scenario builds upon local knowledge of glacial hydrology at Kongsfjord. This experiment accounts for immediate surface-to-base water transfer via crevasses in the ablation zone, as well as for catchment-wide water collection into moulins in the accumulation zone. Water input is taken from a raster of modelled recharge values, as described in Sect. 3.4.2 below.

Here we use the results of Experiment 4, the most realistic water supply configuration, to describe the subglacial hydrology of Kongsfjord. At the beginning of each melt season (early June), the subglacial hydrological system starts to evolve from its channel-free winter configuration. As rising amounts of surface-generated meltwater penetrate to the glacier beds, the inefficient subglacial drainage system rapidly becomes highly pressurized. Pressure first increases in the ablation zone where surface melting is highest and crevasses allow immediate transfer of meltwater to the bed (Fig. 8a). Water flow at the ice-bed interface dissipates energy through the melting of ice, thereby opening small channels where melt-opening exceeds creep closure. The first channels (0.2–1.0 m3 s-1) appear in late June, at the front of Kongsbreen North and Kronebreen–Kongsvegen, before rapidly extending to the upper ablation zone.

By mid-July, channels have grown to accommodate larger water fluxes (> 5.0 m3 s-1). In the lower half of the model domain, moulins have generated channels that connect to the main channel trunks, thereby forming an arborescent network (Fig. 8b). At the peak of the melt season (July–August), subglacial water pressure varies considerably (0 %–100 % of the ice overburden pressure) across the Kongsfjord basin, with a domain-wide mean peak value of 81.5 % of the ice overburden pressure. There is a strong inverse correlation between basal water pressure and bedrock topography, the most pressurized regions being the low-lying valleys below the accumulation basins and the subglacial overdeepenings in the ablation area.

By late August, large channels (> 5.0 m3 s-1) have formed as far as 30–35 km from the glacier fronts (Fig. 8c). There are no channels upstream of the highest moulins, which indicates that, in the absence of moulins, the efficient drainage system would extend no further than the lower part of the accumulation zone. Although the broad-scale structure of the channelized system remains stable throughout the melt season, channels constantly rearrange to accommodate the short-lived variations in meltwater supply. Most noticeably, the subglacial system remains highly pressurized despite channelization even during the second half of the melting period. In most of the upper part of the glacier basin, basal water pressures in August (Fig. 8c) are up to 20 % higher than in June (Fig. 8a) for comparable water inputs.

Finally, towards the end of the melt season, the decrease in water supply leads to the gradual closure of the subglacial channels, and the residual water drains through the distributed system. Channels in the upper reaches are the first to recede, whereas at the glacier termini channels persist until late October.

In this study, we made various assumptions about how meltwater is delivered to the glacier beds. Here we discuss each of these approximations and how they impact the development of the subglacial drainage system in Kongsfjord.

Our reference experiment (Experiment 1) assumes direct and immediate surface-to-bed water transfer over the entire model domain by considering each node of the mesh as a sink for the local meltwater. Despite still being used in models of subglacial hydrology (e.g. Cook et al., 2020), the assumption of perfect vertical permeability may only be appropriate in heavily crevassed glacier zones, such as the terminal part of Kronebreen. This approximation overestimates the amount of water supply to large parts of the glacier bed, in particular at the highest elevations, but underestimates input in areas where recharge is concentrated at moulins. Accordingly, Experiment 1 drives higher water pressures and less channelization in the upper accumulation zone (Fig. 4) and hence the highest mean annual pressures overall compared to the other scenarios (Fig. 3b).

In Experiment 2, the domain-wide total recharge is evenly distributed among 13 moulins instead of being directly fed to the subglacial hydrological system. The assumption of equal recharge at moulins is unrealistic as it implies that meltwater produced at lower elevations is injected into moulins at higher elevations. However, results from this experiment can be easily compared with those of Experiment 1 since both scenarios apply the same water volume to the subglacial system (Fig. 3a). Experiment 2 drives the highest water pressures at the onset of the melt season (Fig. 3b), as well as earliest and largest channel discharge in the accumulation zone (Fig. 6a, d), but the lowest pressures in middle and late summer (Fig. 3b). This indicates that local recharge favours faster channelization in areas where moulins receive large volumes of water (Fig. 4f) and that more efficient drainage at higher regions of the glacier in early summer leads to lower pressurization of the subglacial system during the rest of the melt season.

In Experiment 3, the input to each moulin is collected from their upstream catchments (some of them extending beyond the model domain), generating a slightly larger total volume of water supply than in experiments 1–2 (Fig. 3a). Yet paradoxically, Experiment 3 produces overall lower water pressures than Experiment 1 (Figs. 3b, 4g) and less channelization in the ablation zone (Fig. 6b, c, e, f) compared to the other scenarios. We attribute this counterintuitive result to the water supply configuration that does not account for recharge in the ablation zone where meltwater production is highest.

Experiment 4 aims at the most realistic representation of the actual surface hydrology of the Kongsfjord basin by considering supraglacial streams, supraglacial lakes, moulins, and crevasses. The total water volume delivered to the bed is higher than in the other experiments (Fig. 3a) due to the different drainage catchment definition (Fig. 2d). Similar to Experiment 1, this scenario leads to both earlier and higher subglacial discharge in the ablation zone compared to experiments 2 and 3 (Fig. 6b, c, e, f), thus opening larger channels in the lower half of the model domain (Fig. 7d). However, unlike Experiment 1, Experiment 4 produces a more efficient drainage in the upper half of the glacier basin as well and therefore yields lower water pressures both in summer and in winter despite feeding larger volumes of meltwater to the subglacial system than the reference scenario. This is in line with findings by Koziol and Arnold (2018), which suggested that a more efficient drainage system in summer leads to lower water pressures in winter. This result also shows how a more upward-extended channelized drainage system, maintained by higher recharge at higher elevations, can lower glacier-wide basal pressure.

Notwithstanding these differences between the water input configurations, water pressure remains high (roughly around 70 %) during the entire melt season in all experiments (although less in Experiment 2), suggesting that channels are not very efficient at depressurizing the subglacial hydrological system. As depicted in Fig. 9, the subglacial water pressure displays annual hysteresis under variations in water input. Experiments 1, 3, and 4 exhibit counter-clockwise hysteresis (Fig. 9a, c, d), meaning that for comparable water inputs, water pressure is significantly higher during falling recharge at the end of the melt season than during rising recharge at the start of the melt season. This indicates that water pressure continuously builds up during the melt season and only slowly decreases afterwards. This supports our interpretation that, in experiments 1, 3, and 4, channels contribute only very little to the modulation of subglacial water pressure in our study area. For Experiment 2, however, the relationship between water input and water pressure is more complex (Fig. 9b). For higher water inputs (above 60 m3 s-1), the hysteresis loop turns clockwise, indicating that the subglacial drainage system becomes more efficient with increasing recharge. This change in direction in hysteresis between mid-June and early September indicates that subglacial channels are able to exert some control, although limited, on basal water pressure, which explains why Experiment 2 yields the lowest pressures in midsummer and late summer (Fig. 3b).

In contrast to some previous studies (e.g. Schoof, 2010; Chandler et al., 2013), we find no support that channels are able to substantially increase the efficiency of the subglacial drainage system as water pressures remain high during the entire melting period in all of our experiments (Fig. 3b). Moreover, in our most realistic water supply scenario (Experiment 4), high water pressures (> 90 % of the ice overburden pressure) until late summer over large parts of the model domain (Fig. 8c) suggest that channelization coincides with high pressurization of the subglacial drainage system. We identify two factors behind the low efficiency of the channels in our study area.

First, the small surface gradients that characterize our model domain drive high pressurization of the channels. Subglacial channels open through the melting of the overlying ice driven by turbulent heat dissipation, which requires steeper downglacier bed and surface slopes than those in our study area. Hence, higher water pressures are required to counteract creep closure and keep the channels open even in areas that receive high water input from moulins. The finding that subglacial channels do not play an important role in regulating subglacial water pressure at such low sloping glaciers has been reported in other studies applied to the Greenland Ice Sheet (Meierbachtol et al., 2013; Moon et al., 2014; Dow et al., 2015).

Second, Arctic glaciers such as those of the Kongsfjord basin experience shorter melt seasons than mid- and low-latitude glaciers, and they thus receive overall less water supply from the surface, which also inhibits channelization. In particular, minimal wintertime water input due to generally low temperatures in the Arctic and limited recharge from episodic warm spells is unable to sustain year-round subglacial channels. Compared to overwintered channels that more readily grow to accommodate meltwater input, first season channels lack the capacity to evacuate large volumes of water injected to glacier beds. Poinar et al. (2019) suggested that high-elevation firn aquifers can provide steady water input to the subglacial system outside of the melt season, thereby facilitating subglacial channel formation and winter-persistence even in low sloping areas. Evidence of a perennial firn aquifer was revealed by ice-penetrating radar and GPS observations in the upper reaches of Holtedahlfonna (Christianson et al., 2015). This aquifer is recharged by downward percolation of summertime meltwater, and discharge occurs through downhill flow in the winter. Deep firn water storage is accounted for in the model that provided the input time series to our subglacial hydrology model (Van Pelt et al., 2019) but does not permit considerable wintertime contribution to recharge in the absence of supply from the surface.

Studies of glacier motion suggest that ice acceleration typically occurs after the onset of the melt season when water input to the ice-bed interface is rising faster than the subglacial drainage system can accommodate, driving an increase in basal water pressure, ice-bed separation, and sliding. The sensitivity of ice motion to water input usually decreases over the melt season as the subglacial drainage system evolves from an inefficient distributed to efficient channelized configuration, thereby gradually lowering basal water pressures despite increasing surface melt and thus causing ice deceleration in late summer (Davison et al., 2019). We found that varying the water input configuration introduces differences in the way the subglacial drainage system copes with the early summer increase in water supply, which may be crucial for glacier sliding. In an ice flow model forced by a subglacial hydrology model such as that used in this study, Experiment 1 would cause unrealistically high sliding in the upper accumulation zone as these regions receive too much water. Conversely, the absence of water recharge in the ablation zone (Experiment 3), in driving only moderate pressurization of the subglacial system, would underestimate the rate of sliding at lower elevations. Our results also suggest that limited efficiency of the channelized system, which we attribute to low hydraulic gradients, further enhances sensitivity of sliding to basal recharge variability by maintaining high water pressures and faster sliding speed during the entire summer. We thence deduce that substantial sliding likely occurs at most of the Kongsfjord glaciers over the entire melt season and that sliding rates remain high even in the late summer, which is consistent with observations by Schellenberger et al. (2015) for Kronebreen. For Kongsbreen North, however, Schellenberger et al. (2015) show that surface velocities tend to decrease over the melt season, which is not in accordance with our findings. Their results suggest that subglacial drainage efficiency increases during summer and is able to lower water pressure in that part of the glacier system. Such behaviour is not captured by our model and may be due to the existence of permanent canals carved in the subglacial sediment (Walder and Fowler, 1994) which are not accounted for in the GlaDS model.

One limitation of this study is the lack of coupling between subglacial hydrology and ice dynamics. Given the regional scale of our experiments, running the model in an uncoupled version allowed us to keep the computational cost of our study at a reasonable level. Therefore, we have primarily focused our discussion on the behaviour of the subglacial hydrological system under different melt input configurations and have only cautiously discussed its likely impacts on sliding. Implementing a two-way coupled model would be relevant for future studies seeking to investigate in more detail the feedbacks between subglacial hydrology and ice dynamics.

Another limitation is that our modelling approach ignores some aspects of the glacial hydrological system. We assume that the location of surface-to-bed connections such as moulins and crevasses remains unchanged during the 14-year period of our simulations. In reality, supraglacial hydrology is a dynamic system that is governed by variations in ice velocity and in surface mass balance. Glacier acceleration enhances crevassing, thus increasing the area of direct meltwater transfer to the ice-bed interface. Intense melt events not only promote opening of new moulins through melt-enlargement of existing crevasses but also drive hydrofracturing, a mechanism through which water-filled crevasses or supraglacial meltwater ponds abruptly propagate downwards (Alley et al., 2005; van der Veen, 2007; Das et al., 2008). Supraglacial lake drainage is a widespread mechanism for moulin activation in the upper ablation zone and hence may contribute significantly to subglacial reorganization (Bingham et al., 2003; Andrews et al., 2018; Hoffman et al., 2018). Firn aquifers lying upstream of crevasse fields can also drive hydrofracturing (Poinar et al., 2017). Consequently, neglecting the spatio-temporal evolution of crevasses, moulins, and lake drainage events in our model is definitely a limiting factor to the accurate simulation of water recharge to glacier beds. This limitation could be overcome by using a spatially distributed model of supraglacial hydrology that would predict both the location and the timing of new surface-to-bed hydrological connections for given surface melt scenarios (e.g. Clason et al., 2012, 2015; Koziol et al., 2017).

Finally, the GlaDS model is built on hard-bed physics that may not be relevant everywhere in our study area. The presence of abundant sediment alters water drainage at the glacier base as water can also flow through porous sediment or canals carved in the sediment (Walder and Fowler, 1994), for instance. While GlaDS includes the presence of porous sediment through the parameter of water sheet conductivity, the model does not account for formation of canals within the sediment. Canals could locally increase drainage efficiency compared to the prediction obtained from the Röthlisberger (1972) theory alone as implemented in GlaDS.