Conduit NG-04 (27∘57′24′′ N, 86∘41′55′′ E; 4805 m a.s.l.) was surveyed in November 2009 and consisted of a main passage (A) and two shorter side passages (B and C) leading off to the west (Fig. 4). The main passage extended from a large hollow on the glacier surface (basin C-63 in Figs. 3e and 10a) for a distance of at least 473 m, where the survey was discontinued due to deep standing water on the cave floor. Side passage B also connected with a basin on the glacier surface (basin W-6, Figs. 3e and 10b). Side passage C was at least 25 m long but was not surveyed beyond this distance due to the evident instability of the highly fractured walls.

The main passage consisted of an upper level with a flat or gently inclined floor and a lower narrow incised canyon. The passage was highly sinuous, with a sinuosity in the surveyed reach of 5.52. Near A4 (Fig. 4), there was a tight cut-off meander loop off the main passage (Fig. 5a). The base of the abandoned loop had a flat floor and lacked the incised lower level that was present elsewhere in the system. The upper floor level could also be traced along the walls of side passages B and C, which we interpret as twin remnants of a second meander cut-off. The floor of the upper level sloped gently downward from A1 to A14, rose from there to between A18 and A19, and then descended once more. Sandy bedforms on the floor and scallops on the ice walls of this upper level indicate that water flow was from A1 towards A21.

Passage morphology on the upper level was very variable, including tubular, box-shaped, triangular, and irregular sections (Figs. 4 and 5b–d). Throughout most of the system, planar structures were visible in the ceiling or walls of the upper level, running parallel to the passage axis with variable inclination. The structures took the form of (1) “sutures” at the line of contact between opposing walls (S; Fig. 4; Fig. 5b, c), (2) intermittent narrow voids (V; Fig. 4, Fig. 5c), and (3) bands of sorted sand or gravel a few centimetres thick (SB; Fig. 4, Fig. 5d). Some of the voids increased in width inward, in some cases opening out into gaps tens of centimetres across. In some places, bands of sorted sediment could be traced laterally into open voids or sutures. At several points along the main passage, a pair of planar structures occurred on opposite walls of the passage. Side passage B had a narrow, meandering seam of dirty ice running along its ceiling, and in passage C the walls tapered upward to meet at a ceiling suture.

The floor of the incised lower level in both parts of the main passage sloped down towards side passages B and C (arrows, Fig. 4). A pair of incised channels was confluent at C1, whereas a single incised channel was present in passage B, where its lower (western) end was blocked by an accumulation of ice and debris.

The partially debris-filled structures in the walls and ceiling of the upper level are closely similar to many examples of canyon sutures we have observed in cut-and-closure conduits in the Himalaya and Svalbard, marking the planes of closure where former passage walls have been brought together by ice creep and/or blocked by ice and debris (cf. Gulley et al., 2009a, b). Cut-and-closure conduits are typically highly sinuous and have variable cross-sectional morphologies, ranging from simple plugged canyons (incised channels with roofs of névé), to sutured canyons (partially or completely closed by ice creep), horizontal slots (formed by lateral channel migration followed by roof closure), and tubular passages (where passage re-enlargement has occurred under pipe-full (phreatic) conditions; Gulley et al., 2009b). The tubular morphology of the upper passage in NG-04 – combined with the sutures, voids, and sediment bands in the walls and ceiling – indicates that the passage has been re-enlarged under pipe-full conditions following an episode of almost complete closure. For example, the sub-horizontal bands of sorted sand on both conduit walls between A15 and A18 (Fig. 5d) suggest complete suturing of a low, wide reach (horizontal slot) prior to formation of the surveyed passage.

The tubular and box-shaped cross profiles and undulating long profile of the upper passage are consistent with fluvial erosion under pipe-full or phreatic conditions (cf. Gulley et al., 2009b). This contrasts with the canyon-like form and consistent down-flow slope of channels incised under atmospheric (vadose) conditions, typical of simple cut-and-closure conduits. The dimensions of the upper passage (typically 2 m high and 3 m wide) are consistent with high discharges. We conclude that the upper passage formed when water draining from a supraglacial pond in basin C-63 exploited the remnants of an abandoned cut-and-closure conduit (Fig. 3e).

Following formation of the upper passage, the lower level was incised under vadose (non-pipe-full) conditions when the system accessed a new local base level via side passages B and C. We infer that this occurred when a cut-off meander loop between B1 and C2 was exposed by ice-cliff retreat in basin W-6. Water flow between A1 and B2 continued in the same direction as before, but between A14 and A21 flow was reversed and discharge much reduced.

Evolution of conduit NG-04 can be summarized as follows: (1) a cut-and-closure conduit was formed by incision of a supraglacial stream; (2) this conduit was abandoned and almost completely closed, presumably after it lost all or most of its source of recharge following downwasting of the overlying glacier surface; (3) the conduit remnants were exploited and enlarged by water draining from a supraglacial pond in basin C-63; and (4) surface ablation in basin W-6 broke into the conduit, creating a new base level and initiating floor incision. This remarkable cave illustrates how relict drainage systems can be reactivated when connected to new sources of recharge and demonstrates how patterns of drainage can change dramatically within a single system in response to changing surface topography.

In December 2009 a conduit portal was exposed in an ice cliff at the margin of Spillway Lake (27∘56′36′′ N, 86∘42′46′′ E; 4670 m a.s.l.; Figs. 3a and 6). This portal was one of the two main efflux points that discharged water into the lake from upglacier (Thompson et al., 2016). Access to the conduit could be gained via the frozen lake surface (Fig. 7a), although the lake ice was broken up each morning by debris falling from the melting glacier surface above, severely limiting the time available for survey. Consequently, only a short section could be mapped (Fig. 6). The conduit had two main levels, separated by a narrow, partially ice-filled canyon. The floor of the lower part was at lake level, and that of the upper level was 4.8 m higher, close to the summer monsoon level of the lake, as indicated by shorelines exposed around the lake margins. Several notches on the passage walls recorded intermediate water levels. The ice cliff above the upper level was obscured by a mass of icicles, but observations inside the cave showed that the roof tapered up into a narrow debris band or suture.

Although short, this passage is important for understanding the drainage system of Ngozumpa Glacier. The debris band and suture in the roof indicate that, like NG-04, the passage formed by a process of channel incision and roof closure. Additionally, the passage is graded to the seasonally fluctuating surface of Spillway Lake. We therefore conclude that the main drainage on the eastern side of the glacier consists of a cut-and-closure conduit graded to the hydrologic base level of the glacier. For several kilometres upglacier of the portal, the debris-covered ice surface is highly irregular and broken into numerous closed basins, implying that the conduit evolved from a surface stream that predates significant downwasting of the glacier. The significance of these conclusions will be discussed later in the paper.

NG-01, NG-02, and NG-03 (Fig. 3d) were mapped in December 2005 and described by Gulley and Benn (2007). NG-01 had carried water southward into a large basin on the glacier surface (basin C-25, Fig. 10a), whereas NG-02 drained water southward out of the basin. NG-01 (27∘57′58′′ N, 86∘41′50′′ E) was a sinuous canyon passage with three main levels. Debris bands cropped out in the walls of the uppermost level throughout its length, either at the lateral margins of the passage or in the roof (Fig. 7b). The mid-level had a sub-horizontal floor, into which the canyon linking to the lower level had been incised (Fig. 7c). NG-02 (27∘57′55′′ N, 86∘41′51′′ E) was a sinuous canyon passage on two levels, extending in a southwesterly direction from the basin. The upper level had a circular cross profile, and an incised canyon beneath formed the lower level. A suture and debris band were exposed along the entire length of the ceiling of the upper passage, mirroring the planform of the passage (Fig. 7d). The lower level was an asymmetric flat-floored passage with a series of sills along the margins. NG-03 (27∘57′52′′ N, 86∘42′02′′ E) consisted of a single passage graded to a supraglacial pond in basin E-19. Passage morphology varied between a low, wide semi-elliptical cross-section and a more complex form with an elliptical upper section separated by a narrow neck from a lower A-shaped section. At the top of the canyon, the ceiling narrowed to a narrow slot, terminating in a band of coarse, unfrozen sandy debris.

For much of their length, all three conduits follow the trend of debris bands in the walls or roof, leading Gulley and Benn (2007) to conclude that all were structurally controlled. The debris bands were originally interpreted as debris-filled crevasse traces that had been deformed during advection downglacier. When the original work was conducted, the cut-and-closure model had not been developed, and we had yet to learn how to recognize the diverse forms such conduits can take, especially in the later stages of their development. It is now apparent that these conduits have all the hallmarks of cut-and-closure conduits. The continuity and sinuous planform of the debris bands are consistent with formation by the closure of incised canyons, rather than crevasse fills that had been deformed by ice flow. Crevasses in the upper part of the glacier ablation area tend to be short, discontinuous, and oriented transverse to flow, unlike the observed debris bands in the conduit roofs, and ice deformation is unlikely to be capable of generating the highly sinuous patterns observed within the conduit debris bands.

We therefore reinterpret NG-01–NG-03 as cut-and-closure conduits that have undergone cycles of incision, abandonment, partial closure, and later reactivation in response to fluctuating patterns of recharge on the glacier surface. The circular and elliptical cross profiles observed in NG-02 and NG-03 are consistent with phases of phreatic passage enlargement, analogous to that in NG-04. Abandoned, incompletely closed conduits create hydraulically efficient flow paths, which can be readily exploited and enlarged when surface ablation brings them into contact with new sources of recharge.

Direct observation of the subglacial drainage system was not possible. Instead, we use seasonal fluctuations in glacier surface velocity to infer areas subject to variable subglacial water storage. Mean daily ice velocities of the glacier between 29 January 2015 and 5 January 2016 are shown in Fig. 8a. There is no detectable motion (i.e. greater than ∼ 0.01 mday-1) on the main trunk within ∼ 6.5 km of the terminus or on the lowermost 6 km of the E branch. The W branch is the most active, with velocities of ∼ 0.16 mday-1 (∼ 60 myr-1) at 5300 m a.s.l., declining to near zero at 4900 m. The NE branch is slower, although velocities in its upper part could not be determined due to image “lay-over” in steep terrain. The active part of the NE branch does not extend as far down as the confluence with the W branch, and a strip of stagnant ice ∼ 100–200 m wide extends ∼ 3 km down the eastern side of the main trunk from the confluence zone. Thus, neither the E nor the NE branch is dynamically connected to the main trunk.

Evidence for seasonal velocity fluctuations is shown in Fig. 8b, which shows mean daily velocities between 29 January 2015 and 5 January 2016 (341 days) minus mean daily velocities from 19 September 2014 to 18 January 2015 (111 days). Meteorological data from the Pyramid Weather Station, at 5050 m a.s.l. and ca. 12 km east of Ngozumpa Glacier (available through the Ev-K2-CNR Stations at High Altitude for Research on the Environment (SHARE) project), indicate that air temperatures were consistently below freezing between 25 September 2014 and 28 May 2015, defining a minimum winter period for the upper ablation zone. The 111-day interval lies almost entirely within the winter period but is less than half of its total duration, so Fig. 8b yields minimum values for a summer speed-up on the glacier. Most of the active parts of the glacier exhibit some speed-up, although it is much more pronounced in some areas than others. On the W branch, the greatest speed-up (by ∼ 0.015 mday-1 or ∼ 10 %) occurs above the confluence with the NE branch. Areas of lesser speed-up also occur on the main trunk below this point, although these are discontinuous and less than the margin of error and so are likely to be artefacts. Only the northern side of the NE branch is affected by a seasonal speed-up. This area coincides with the tongue of clean ice that descends through the icefall below Gyachung Kang (Fig. 1). Patchy areas of apparent speed-up and slow-down occur elsewhere on the NE branch but may be artefacts. A small speed-up also affects the active part of the E branch, above 5350 m a.s.l.

The seasonal variations in ice velocities in the upper ablation zone are too large to be explained by changes in ice creep rates, which would require fluctuations in driving stress that are inconsistent with the observed surface elevation changes on the glacier (Thompson et al., 2016). We interpret the velocity data as evidence for variations in basal motion (sliding and/or subglacial till deformation) in response to changing subglacial water storage. This interpretation is supported by the spatial distribution of areas affected by the seasonal speed-up, which coincide with, or occur downglacier of, heavily crevassed ice. Much of the upper ablation area of Ngozumpa Glacier consists of icefalls with surface gradients up to 30∘, and fields of transverse crevasses occur across the entire width of the W branch down to an elevation of 5270 m (Fig. 9a). Below this zone, crevasses are largely absent, reflecting decreasing ice velocities and compressive flow (Fig. 9b, c, and d; cf. Fig. 8). Fields of transverse crevasses occur in the upper basin of the E branch, above ∼ 5400 m. Crevasses allow meltwater to be routed rapidly to the bed, and the existence of multiple recharge points will encourage development of a distributed drainage system following the onset of the monsoon ablation season. The lack of a clear seasonal velocity response on the lowermost 10 km of the glacier suggests that subglacial water is transported along the main trunk in efficient conduits, possibly along the glacier margins (see Sect. 4.4).

Supraglacial stream networks are visible below the crevassed zones on all three branches of the glacier. The most extensive network occurs on the tongue of clean ice on the NE branch, where a set of sub-parallel channels descends from ∼ 5180 m to the junction with the W branch at ∼ 4990 m (Fig. 3b, c; Fig. 14a). There are several discontinuous supraglacial channels on the W branch between 5220 m and 5120 m a.s.l., including one along the eastern margin of the glacier. Supraglacial channels occur on both flanks of the E branch below ∼ 5100 m a.s.l. The channels converge at the junction with the main trunk, and after flowing over the glacier surface for several hundred metres the combined stream sinks in a large hollow in basin E-11. Patterns of water storage and release in this hollow are discussed in Sect. 4.4.

Perennial supraglacial channels can only persist if the annual amount of channel incision exceeds the amount of surface lowering of the adjacent ice (Gulley et al., 2009b). The rate at which ice-floored channels incise is controlled by viscous heat dissipation associated with turbulent flow and increases with discharge and surface slope (Fountain and Walder, 1998; Jarosch and Gudmundsson, 2012). Because supraglacial stream discharge is a function of surface melt rate and melt area, significant channel incision requires large catchment areas. Therefore, incised surface channels tend to occur only where potential catchments are not fragmented by crevasses or hummocky surface topography (Fig. 3). At present, these conditions are met in relatively limited areas of Ngozumpa Glacier, below crevassed areas and above hummocky debris-covered areas.

Most of the lower ablation zone of the glacier (below ∼ 5000 m) consists of hummocky debris-covered topography, where the glacier surface is broken up into distinct closed depressions, each of which forms a separate surface drainage basin (Fig. 3d, e). Not including the Spillway Lake basin that drains externally, we defined 111 surface basins in this zone in 2010 (Fig. 10). Some surface basins also occur between 5000 and 5100 m on the W Branch, but these are typically small, shallow, and ill-defined (Fig. 3). This part of the glacier is steeper (3.4∘) and has lower relative relief (∼ 10 m) than the lower glacier, and it appears to be a transitional zone between the channelized upper ablation area and the hummocky debris-covered zone (Fig. 2). Surface basins along the east and west margins of the glacier form a series of depressions within almost continuous lateral troughs, and are considered in Sect. 4.4. Here, we focus on the basins in the central part of the glacier (C1–C69, Fig. 10a) and the terminal zone (T1–T6, Fig. 10b).

Of the 70 basins in the central part of the glacier, 56 (80 %) contained ponds in at least 1 of the 3 years covered by the GeoEye and WorldView imagery. Fifteen of the 42 ponds present in 2010 (36 %) had disappeared by 2012 or 2015, whereas 14 basins that were empty in 2010 contained ponds in 1 or more of the later years. Almost all of the remainder underwent partial drainage and/or refilling. In contrast, the five ponds in the terminal zone of the glacier (below Spillway Lake) clearly exhibited stability. Four showed no significant change in area between 2010 and 2015, while the other showed an increase in area.

Observations on and below the glacier surface show that drainage of perched ponds occurs when part of the floor is brought into contact with permeable structures in the ice (Benn et al., 2001; Gulley and Benn, 2007). The characteristics of NG-01–NG-05 (which all occur within the hummocky debris-covered zone) show that relict cut-and-closure conduits are the dominant cause of secondary permeability in the glacier, providing pre-existing lines of weakness along which perched ponds can drain.

The spatial extent and high temporal frequency of perched pond drainage events on the glacier (Fig. 10a) imply a high density of active or relict conduits within the ice. A rough estimate can be obtained by dividing the number of complete and partial drainage events (35) by the total area of basins in the central part of the glacier (4.62 km2), yielding ∼ 7.6 relict conduit reaches per square kilometre. This is a minimum estimate, because additional conduit remnants could occur below and beyond the margins of observed ponds. Conversely, the number of pond filling events (23 over the 5 ablation seasons spanned by the imagery) shows that drainage routes commonly become blocked. Conduit blockage processes have been described by Gulley et al. (2009b) and include accumulation of icicles or floor-ice at the end of the melt season and creep closure of opposing conduit walls. The interplay between drainage events and conduit blockage maintains a dynamic population of supraglacial ponds, which contribute significantly to ablation of the glacier, through absorption of solar radiation and ice melt, and calving (Thompson et al., 2016).

The stability of ponds in the terminal zone probably reflects a combination of factors. These ponds are flanked by stable slopes of thick debris, which inhibit pond growth by melt or calving. Furthermore, the ponds are located at or close to the hydrologic base level of the glacier, determined by the terminal moraine that encircles the glacier terminus, inhibiting drainage via relict conduits.

Elevation differences between successive DEMs indicate linear zones of enhanced surface lowering along both margins of Ngozumpa Glacier, forming troughs along the base of the bounding lateral moraines (Thompson et al., 2016; Fig. 11). The inner moraine slopes consist of unvegetated, unconsolidated till and undergo active erosion by a range of processes, including rockfall, debris flow, and rotational landslipping (Benn et al., 2012; Thompson et al., 2016). Although debris eroded from the moraine slopes is transferred downslope into the troughs, the troughs underwent surface lowering of 6–9 m from 2010 to 2015, with a total annual volume loss in the moraine–trough systems of ∼ 0.4×106 m3yr-1 (Thompson et al., 2016). This implies that a large volume of ice, debris, or both is evacuated annually from below the lateral margins of the glacier.

The lateral troughs form a series of closed basins, 12 on the west side and 22 on the east (Fig. 10b). Eight of the basins in the west trough and 17 of those in the east contained a pond in 2010, 4 (W) and 7 (E) of which had completely drained by 2012 or 2015. Four new ponds appeared in the eastern trough in 2012 or 2015, and 1 (W) and 7 (E) underwent partial drainage and/or refilling. Three basins on the western side and one on the eastern side showed no fluctuations in pond area.

Benn et al. (2001) provided detailed descriptions of pond filling and drainage cycles in basins W-7 and W-5 (lakes 7092 and 7093, respectively, in their terminology). In October 1998, basin W-7 contained three shallow ponds, but by October 1999 the basin was occupied by a single large pond, and water level had risen by ∼ 9 m. Pond area had increased from 17 890 to 52 550 m2, with 36 % of the increase attributable to backwasting and calving of the surrounding ice cliffs. By September 2000, the pond had almost completely drained and only shallow ponds remained. Pond drainage occurred via an englacial conduit, which had been exposed by retreat of the pond margin. A pond in basin W-5 also underwent fluctuations in area and depth between 1998 and 2000 but did not completely drain during that time. Horodyskyj (2015) used time-lapse photography and a pressure transducer to document rapid pond-level fluctuations in basin W-5, including rises and falls of several metres within hours.

Short-term cycles of pond drainage and filling can also be demonstrated in other basins within the lateral trough systems using optical satellite imagery. Figure 12 shows a series of images of the east side of the glacier close to the junction with the E branch, where a supraglacial stream (Sect. 4.2) flows into a closed depression in basin E-11 (Fig. 10b). A pond occupying the basin expanded in area between March and May 2009 but drained between June and August. In 2015 there is little evidence of the pond in January, but a large pond is present in June.

Widespread, rapid subsidence along both margins of the glacier can be explained by enlargement and episodic collapse of sub-marginal conduits (Thompson et al., 2016). Potential internal ablation rates were calculated from energy losses associated with runoff and supraglacial pond drainage, and the resulting value of 0.12 to 0.13×106 m3yr-1 is around 30 % of the measured volume losses in the moraine–trough systems on the stagnant part of the glacier, the difference being at least partly attributable to sediment evacuation by meltwater.

The sub-marginal conduits are perennial features of the glacier drainage system and discharge water into Spillway Lake during the winter months. Winter discharge may partly reflect slow release of water from supraglacial and englacial storage, but it may also partly consist of subglacial water from the upper ablation zone (see Sect. 4.1). This hints at the possibility that the sub-marginal channels function as the downglacier continuations of the subglacial drainage system, in addition to carrying water transferred more directly from the glacier surface.

Much of the lower ablation zone appears to be bypassed by the sub-marginal conduits, as evidenced by widespread water storage in supraglacial ponds (Sect. 4.3). As noted above, water is intermittently discharged from ponds in the central part of the glacier into the lateral troughs via englacial conduits. Cycles of pond drainage and filling in lateral basins indicate intermittent connections between surface catchments and the sub-marginal meltwater channels (Fig. 10b). In some cases, drainage events can be directly attributed to exploitation of englacial conduits (Benn et al., 2001). The hourly changes in pond level recorded by Horodyskyj (2015) cannot be explained by conduit opening and blockage, and they more likely reflect short-term fluctuations in recharge from surface melt and water release from storage.

In 2010, the area of the Spillway Lake surface catchment was 0.8 km2, of which 0.27 km2 was occupied by the lake system. All of the water leaving the glacier passes through Spillway Lake, entering via portals or upwellings at or close to lake level and leaving via a gap in the western lateral moraine ∼ 1 km from the glacier terminus (1; Fig. 13). (It is possible that water also exits the glacier via groundwater flow, although no springs have been observed in the frontal moraine ramp.) In 2009, conduit NG-05 (Fig. 6; Sect. 3.2) entered the NE corner of the Spillway Lake and is interpreted as the distal part of the eastern sub-marginal conduit. A second conduit portal visible at the NW lake margin in the same year is interpreted as the efflux point of the western sub-marginal stream. The evolution of the Spillway Lake system and its implications for drainage system structure in this part of the glacier are examined in Sect. 5.4 below.