To document the supraglacial lake expansion, we analysed 25 Level 3B tiles collected by the PlanetScope Dove satellite constellation between 27 March and 26 October 2017 (Table S1 in the Supplement). These 4-band data have a ground sampling distance of 3.7 m but are resampled to 3 m during orthorectification, and digital numbers (DNs) contain scaled at-sensor radiance values for the blue (B: 455–515 nm), green (G: 500–590 nm), red (R: 590–670 nm), and near-infrared (NIR: 780–860 nm) spectral ranges. We also used several RapidEye level 3B tiles for pond coverages and geomorphic interpretations. These are 5-band data (B: 440–510 nm; G: 520–590 nm; R: 630–685 nm; red edge: 690–730 nm; NIR: 760–850 nm) with a ground sampling distance of 6.5 m, resampled to 5 m during orthorectification . Due to the high density of clouds during the monsoon, few scenes are cloud-free over the whole study area. We therefore masked clouds and cloud shadows manually in the region of the supraglacial lake before mapping ponded water (Fig. b–i). For each scene, we calculated the Normalised Difference Water Index (NDWI) based on DNs for the G and NIR bands NDWI=G-NIRG+NIR; e.g. and used an Otsu adaptive histogram-based approach to select an optimised NDWI threshold , identifying ponded water as those pixels exceeding this threshold. Finally, the pond cover products were again inspected manually to remove terrain and cloud shadows before determination of multi-temporal lake area, and we used a ±1 pixel buffer for lake area uncertainty e.g..

We analysed two along-track Pléiades triplets with acquisition dates of 23 March and 14 December 2017, bounding the lake's filling and drainage. The two scenes had maximum base-to-height ratios of 0.55 and 0.32, respectively. Their panchromatic bands (480–830 nm, ground sampling distance of 0.7 m) were processed using the Ames Stereo Pipeline to generate DEMs and orthoimages at 2 and 0.5 m resolution, respectively. The two Pléiades DEMs were 3-D-coregistered using off-glacier terrain , then differenced to produce a map of surface elevation change (dH) spanning the 2017 monsoon period.

This geodetic difference encompassed the majority of the ablation season, so for the glaciers we focused on zones of enhanced surface lowering not solely attributable to ice cliffs and supraglacial ponds, which are known hotspots of melt for Himalayan debris covered glaciers e.g.. Ice cliffs tend to have curvilinear forms, with their planimetric length much greater than their width (e.g. ; ). For our study area, we are able to neglect advection and emergence of these features due to glacier dynamics e.g., as the lowest 5 km of Khumbu Glacier is stagnant . Over a short interval, melt along the inclined cliff surface was thus expressed as a thin arc of surface lowering e.g., clearly identifiable in Fig. . We ignored these cliff areas and areas of elevation change within ponds. We thus identified 11 zones of prominent elevation change that were clearly associated with the lake drainage according to the PlanetScope and RapidEye imagery (Table , Fig. ). Field visits in May 2017, October 2017, and May 2018 enabled direct observation of many of the most prominent zones of change.

To assess the error on the elevation difference obtained by differencing of two Pléiades DEMs, we follow the tile method of and split the stable terrain dH maps into n×n tiles, with n varying from 2 to 200. The corresponding individual tile area thus varies from 91.2 km2 (n=2) to 0.01 km2 (n=200). For each tile, we compute the absolute value of the median dH. We then calculate our dH error (σdH) as the average of these n2 absolute values, and σdH ranges from 0.12 m (n=2) to 0.64 m (n=200). In Fig. b, dH is plotted as a function of the individual tile area. The relationship is well represented by a logarithmic fit which we use as our error model. Consequently, for all our zones of change we estimate an error based on the zone area and only analyse elevation changes of magnitude greater than this error.

Using the pre-lake March 2017 Pléiades DEM, we identified 142 closed surface depressions and determined depth–area–volume relationships for each by progressively filling them with an increment of 0.1 m depth following e.g.. We then calculated stored water volumes in the supraglacial lake area for each PlanetScope scene by estimating the volume of each individual pond in the area of the supraglacial lake, then summed these to estimate the total ponded volume in the study area (Fig. ). On 16 July the lake was partially obscured by clouds, so for this scene we instead estimated the water level and volume from a partial shoreline data set (Figs. S1 and S2 in the Supplement). This approach assumes negligible topographic changes in the proximity of the supraglacial lake during the study period, but many studies have noted the local ablation associated with supraglacial ponds . Thus, the resulting volume estimates carry considerable uncertainty (in this case calculated using the ±1 pixel areal uncertainties), but are nonetheless useful for providing minimum values of supraglacial water storage during this period.

We also measured areal changes associated with active channel migration and bank erosion along the Khumbu proglacial stream as far as Pheriche using RapidEye level 3B imagery from November 2012, 2015, 2016, and 2017. The images were coregistered in ENVI (RMSE <1 m), then we calculated changes in the NDWI and Normalised Difference Vegetation Index (NDVI=NIR-RNIR+R) for 2012–2015 and 2016–2017, enabling us to resolve periods preceding and spanning the 2017 Changri Shar outburst. Outburst floods from Imja Khola during 2015 and 2016 may have affected bank erosion and channel migration change over the period of analysis, but this tributary joins the Khumbu proglacial stream below Pheriche.

We considered the major NDVI changes (all decreases) to indicate bank erosion and reactivation, while marked spatial changes in NDWI indicated stream migration. We calculated a 3×3 focal mean to reduce noise, then eliminated low-magnitude changes in the indices based on a visual inspection of the histogram (thresholds in Table ). We manually trimmed the results to zones within the channel, also eliminating areas severely affected by shadows. Finally, we aggregated areas of bank erosion and stream migration in 1 km bins along the main Khumbu River to compare rates of change preceding and bounding the event (Table ).

The study period coincided with automated water level measurements collected every 30 min in the proglacial stream near Pheriche village (Fig. ). A rating curve has been developed for this position based on 34 field-calibrated fluorescein discharge measurements collected since November 2010 and was used to calculate discharge for the period of analysis. Based on the analyses of and , we estimated a discharge uncertainty of 15 % for stage values within the calibrated range and 20 % for stage values above the maximum stage-discharge measurement. From this record, we estimate background discharge (hereafter, base flow) from 17:00 on 15 July to 09:00 on 17 July (all times given in Nepal Time, NPT; UTC + 05:45) using a half-hourly cubic spline interpolant fitted to measurements for 10–15 and 17–20 July (i.e. interpolating between preceding and subsequent 09:30 measurements to estimate discharge at 09:30 on 16 July) and determine the flood discharge as the difference between observed discharge and estimated base flow.

The dynamics of the lake system formation are relatively straightforward to interpret. A significant obstruction to the coupled supraglacial and englacial drainage system must have formed during winter 2016–2017, as occurs seasonally for other debris-covered glaciers . This may have been the consequence of a significant conduit collapse or freeze-on of accumulated englacial debris, as has been observed through glaciospeleology e.g.. However, the impediment to drainage appears to have been unusually effective in early 2017, preventing the development of preferential flow paths which would lead to increasingly efficient drainage. Thus, as winter snow in the ablation area melted due to the onset of pre-monsoon conditions, this water accumulated in a large surface depression that opened over recent years by heightened ablation along supraglacial ponds and ice cliffs. The accumulated water would have had a positive surface energy balance through the pre-monsoon, leading to peripheral ablation and further increasing the depression capacity and lake volume .

The supraglacial ponds initially grew in isolation, then coalesced between 18 May and 17 June as the water levels rose (Fig. ). By 19 June, new peripheral ponds began to fill, suggesting the flooding of englacial conduits up to 300 m away from the main water body. These secondary ponds mostly coalesced with the main surface water body before its eventual drainage. Based on the pond shorelines and Pléiades DEM, we estimate a steady water supply rate of 0.14 m3 s-1 for 17 June to 13 July.

The process of pond drainage is slightly less clear due to the lack of observations during 14–15 July. The available PlanetScope imagery indicates that the lake began to drain between 13 and 15 July, and was still underway on 16 July. Given the total duration of the flood at Pheriche (∼36 h) and the landslide deposit identified on 16 July, we expect that drainage began around midday on 15 July. Based on the lack of down-glacier surface change on Changri Shar, we infer that the lake drained englacially or subglacially, rather than along the surface. Hydrofracture is an unlikely scenario as the ice is nearly stagnant in this area; rather, this could have been accomplished by penetrating the internal blockage or establishing a new connection to relict conduits. In either case the water re-emerged at the surface ∼700 m away, just prior to the Changri Shar terminus.

The textureless appearance of the flooded entrance to Khumbu Glacier imaged on 16 July (Fig. c) suggests that the water had only recently reached this position. This assessment is supported by the rapid subsequent drainage of the flooded water and incision of the debris deposit, which had occurred by 17 July. As this subsurface conduit would have closed at least partially since the prior monsoon, the sudden input of water and debris likely overwhelmed the conduit's capacity. Using an empirical relation for peak tunnel discharge Qp=46Vp0.66, with Vp the lake volume in 106 m3;, we estimate a peak discharge of 59 m3 s-1. Some water may have been retained in both glaciers' drainage network, and the flood at Pheriche is likely to have incorporated additional meltwater and debris along its glacial and proglacial flowpath, but this discharge estimate is very close to the maximum discharge of 56±11 m3 s-1 measured by the Pheriche gauge.

As with Changri Shar, the lack of surface change on Khumbu Glacier suggests a subsurface flow path for much of the glacier's length. However, the floodwaters appear to have reached the glacier surface 2.3 km from the terminus, where several segments of conduit collapse are evident. We interpret this as due, at least in part, to the elevated hydrological base level of Khumbu Glacier , the terminus area of which has experienced extensive ponding in recent years .

There is a notable contrast in the magnitudes of proglacial stream migration and bank erosion between the 2012–2015 and 2016–2017 periods (Table ). During the 2012–2015 period, channel migration was a continuous background process that predominantly remained within the stream banks. This period encompassed the Gorkha earthquake , which would have increased debris supply and stream migration. The 2016–2017 NDWI and NDVI changes show a greater magnitude of channel migration than 2012–2015, despite the shorter time interval. The area of bank erosion is also greatly enhanced during 2016–2017. The magnitude of geomorphic change associated with the flood appears to be uncommon, since we were unable to find similar areas of bank erosion in any of the historic satellite image archives.

The double peak of discharge observed at Pheriche (Fig. ) is unusual for outburst floods. A possible cause is the blockage of the Khumbu stream inlet by the landslide in the Changri Shar proglacial gorge (Fig. ). This is likely to have initiated around peak flow through the gorge and could have led to a substantial decline in discharge, followed by a later, sudden increase as preferential flow paths developed through the debris . A second explanation is the possibility of multiple flow paths for the flood through the lower part of Khumbu Glacier. As Khumbu Glacier exhibits a low terminus slope and high hydraulic base level, the flood may have temporarily overwhelmed the subsurface drainage network and, exploiting fractures and secondary pathways common for these glaciers, partially emerged at the glacier surface. This would result in two or more flow paths of differing efficiency, possibly leading to distinct discharge peaks on the Pheriche hydrograph. This possibility is supported by the appearance of highly turbid water in the ponds between zones E and G (Fig. ) during drainage. A third possibility is that the increased discharge late on 16 July corresponds to a delivery of water stored elsewhere within the glacier system. This stored water might connect to the drainage system more efficiently by the opening of conduits and channels during the flood. Regardless, it is clear that the increase in discharge at Pheriche only lasts until 10:00 on 17 July, so the flood's direct contribution to discharge was short-lived.

The use of novel satellite platforms for observing and interpreting this event enabled the development of a detailed chronology of surface changes rarely available for such events. Our observations of the drainage and outburst of the Changri Shar supraglacial lake have several implications for cryospheric hazards and debris-covered glacier hydrology. First, this is a short-lived event, with a lake system of 1.36×106 m3 filling and draining within one ablation season. This is important because, despite the lake's short duration and relatively small volume, the event led to considerable glacial, fluvial, and geomorphic change. These changes disrupted major trails, which are the primary corridor for local trade and tourism, in at least three locations Figs. , and ;. As suggested by and , the hazard posed by such features is significant, yet traditional glacial lake monitoring approaches, which rely on repeat optical imagery such as Landsat and Sentinel-2, would have had difficulty observing the lake's formation due to the timing of repeat passes and cloud cover. Considering all Landsat 8 or Sentinel-2 scenes over the period of our analyses, we find only two that are mostly cloud-free over the supraglacial lake in the 2 months leading up to lake drainage. Pond observations during the monsoon are intermittent at best and thus we recommend the adoption of high-frequency repeat optical imagery (as in this study) and synthetic aperture radar data products e.g. for improved monsoon monitoring of glacier hydrology.

Furthermore, the scarcity of seasonal observations of glacier hydrology (limited to a few closely monitored glaciers) suggests that short-lived or seasonal outburst floods may be a regular feature of debris-covered glaciers in the region. This is important because both and this study indicate that outburst floods from sources other than large proglacial lakes have had downstream effects on transportation networks and the livelihoods of local communities. The few observations of outburst floods from high-elevation debris-covered glaciers suggest a distinct seasonal cycle of hydrological development that contrasts with clean ice glaciers e.g.. Rather than a gradual up-glacier progression of an efficient, connected drainage network e.g., debris-covered glaciers may impound significant volumes of water internally and at the surface before establishing efficient drainage through the lowest portion of the glacier . This key difference is likely related to the melt-inhibiting thick debris near the terminus on such glaciers, which reduces the terminus area's sensitivity to seasonal warming. Instead, the zone of maximum melt (and seasonal sensitivity) is usually in the middle of the ablation area, leading to significant meltwater generation before efficient drainage pathways have been established for the lower glacier .

The geomorphic evidence from this study suggests that supraglacial lake outburst floods of this magnitude are not particularly common in the Khumbu catchment (indeed, no large supraglacial lake formed on Changri Shar in 2018, and past years show no evidence of such a lake). Still, supraglacial water storage is increasing for many Himalayan glaciers e.g.. This is expected as climate warms and debris-covered glaciers stagnate, which are precursors to proglacial lake formation . In the case of Changri Shar, a very large closed surface depression had been opened by ice cliffs and supraglacial ponds prior to this event, creating the capacity to store 1.36×106 m3 of water. Consequently, as the excavation and pitting of near-stagnant debris-covered glacier termini by ice cliffs and supraglacial ponds becomes more prevalent with a warming climate, other glaciers in the region are likely to develop large supraglacial water bodies. While the coalescence of ponds forming a large supraglacial lake represents an early stage of base-level lake development , these supraglacial lakes also represent an outburst risk (as evidenced here). Thus, the expected increase in moraine-dammed glacial lake outburst floods due to a lagged response to climate warming may also apply to the outburst of supraglacial water bodies, and events similar to the Changri Shar outburst are likely to increase in frequency.

Finally, the rapid transit time we observe for the flood's passage of the lower Khumbu Glacier suggests that the glacier's subsurface drainage system has, or can develop, an efficient configuration in response to sufficient water supply. We base this assessment on the sudden interruption of peak discharge observed at Pheriche, which most likely corresponds to the blockage of the Changri Shar stream portal as observed in the PlanetScope image on 16 July (Fig. c). This image was captured at 09:50, implying a transport time of 3–5.5 h for water to travel a (straight-line) distance of 4.9 km through Khumbu Glacier. Consequently, we estimate a mean travel velocity of at least 0.25–0.45 m s-1; the water also passed 4 km from the glacier to Pheriche during this time but we cannot determine its transit time. Prior dye-tracing studies have considered flow velocities >0.2 m s-1 to indicate hydraulically efficient drainage through a system of major channelised conduits e.g., which we thus interpret for drainage through Khumbu Glacier during this event. The subsurface drainage initially exploited a pre-existing marginal flow path maintained by normal discharge from Changri Shar and Changri Nup glaciers, similar to that inferred for Ngozumpa Glacier by , thus enabling the system's rapid accommodation of the surplus water. It appears that subglacial or deep englacial flow paths were utilised by the flood for Changri Shar Glacier (∼700 m subsurface transit) and much of Khumbu (2.8 km subsurface transit) Glacier. The outburst seems to have bypassed the coupled supraglacial and shallow englacial drainage networks inferred by and until the lowermost portion of Khumbu Glacier, where at least some water emerged at the surface and routed through the terminal chain of ponds. Our interpretations of hydraulically efficient subsurface drainage and subsurface-to-surface routing reflect the response of the drainage system to flood conditions, so additional observation is needed to understand the structure of the drainage system with normal meltwater inputs.