Supraglacial pond evolution in the Everest region, central Himalaya, 2015-2018

6 Supraglacial ponds are characteristic of debris-covered glaciers and can greatly enhance local melt rates. They 7 can grow rapidly and coalesce to form proglacial lakes, presenting a major hazard. Here, we use Sentinel-2A 8 satellite imagery (10 m) to quantify the spatiotemporal changes of 6,425 supraglacial ponds for 10 glaciers in 9 Everest region of Nepal between 2015 and 2018. During the study period, ponded area increased on all glaciers, 10 but showed substantial temporal and spatial variation. The rate of pond growth accelerated compared to 200011 2015 (Watson et al., 2016). Both Imja and Spillway Lake expanded and Khumbu Glacier continued to develop a 12 chain of connected ponds. 54% of ponds were associated with an ice-cliff, but the proportion of ponds with cliffs 13 decreased during the study period. Pond location showed limited correspondence to slope, but favoured areas 14 of lower surface velocity. Ideal conditions for pond formations have advanced up-glacier, and are now 15 predominantly found at mid-elevations. Results indicate high-resolution imagery (< 10 m) is essential, as using 16 Landsat data would miss 55–86 % of the total ponds found. Finally, glaciers were classified by stage of 17 development (Komori, 2008; Robertson, 2012), with two transitioning between 2015 and 2018, suggesting lakes 18 in the region are evolving rapidly. Furthermore, some glaciers displayed characteristics of multiple classes, so 19 we propose an adapted classification system. Overall, our results demonstrate a trend of pond expansion in the 20 Everest region and highlight the need for continued monitoring for hazard assessment. 21


23
Mass loss from glaciers in the Himalayas has increased rapidly over the past 30 years, in response to climate 24 change (e.g. Bolch et al., 2012;Kääb et al., 2012;Quincey et al., 2009;Gardelle et al., 2013). Here, 'summer-25 accumulation type' glaciers rely on summer-monsoon snowfall for mass gain (Bolch et al., 2012), which is 26 thought to be reducing as temperatures rise (Fujita, 2008). The shrinkage of these freshwater reservoirs will 27 have significant regional-and local-scale impacts (Immerzeel et al., 2010;Bolch et al., 2012;Dehecq et al., 28 2018). Seasonal monsoonal rainfall is the dominant source of water in the Himalaya, but glacial meltwater 29 provides up to 40% of water supplies during the dry season (Immerzeel et al., 2010). Thus, the consistent 30 negative mass balance of Himalayan glaciers may lead to long-term reductions in perennial flow supplied to 31 major rivers outside of monsoon season (Xu et al., 2009), and could threaten the water and food security of an 32 estimated 70 million people in the densely populated downstream catchments (Immerzeel et al., 2010). 33 Additionally, increased supraglacial meltwater storage will likely increase the frequency of glacier related 34 hazards in the region, particularly from Glacier Lake Outburst Floods (GLOFs) (Thompson et al., 2010). GLOFs are 35 highly destructive and Nepal (along with Bhutan), has been identified as the most economically vulnerable to 36 whereas the number of ponds generally declined, despite the considerable spatial variability observed in 2016 179 (Fig. 2c). 180 Two water bodies in the study area were assessed separately, due to their large size; supraglacial lake 181 'Spillway Lake' on the terminus of Ngozumpa Glacier, and proglacial lake 'Imja Lake' fronting the Imja/Lhotse 182 Shar Glacier complex (Fig. 4). Spillway Lake, located at the terminus of Ngozumpa Glacier, underwent a net gain 183 of 40,565 m² during the study period (December 2015 -April 2018) and reached a size of 286,367 m². This 184 remains the largest surface water store in the study region. The only proglacial lake identified in this study was 185 Imja Lake, which expanded up-glacier by 455.6 m to reach and area of size of 1,493,142.68 m 2 by 2018. Whilst 186 these two water bodies are currently the only very large water bodies in the study area, our data show 187 substantial growth and coalescence of surface ponds on Pangbung and Khumbu glaciers (Fig. 3b, c). 188

Glacier-scale Pond Changes 189
Despite the overarching trends across the region, changes in supraglacial pond number and area varied from 190 glacier to glacier and across individual glaciers ( Fig. 2; Table 2). For the largest three glaciers, most of the ponded 191 area was located near the terminus, and this persisted throughout the study period ( Fig. 3a-c). The number of 192 ponds, although demonstrated an overall decrease, remained relatively similar over the four year period, but 193 ponded area increased (Fig. 2), which was primarily as a result of coalescence on the lower glacier tongues. (Fig.  194 3a-c). For example, on the terminus of Ngozumpa Glacier, two smaller ponds ( On the seven smaller glaciers, there was substantial spatial and temporal variability in the number and 202 area of surface ponds, both between the glaciers and across individual glaciers ( Fig. 3d and Fig. 4). For example, 203 Lhotse Shar and Imja Glacier neighbour each other, in the east of the study region (Fig. 1). However in April 2018, 204 Lhotse Shar Glacier had over three times the ponded area (135,420 m 2 ) of Imja Glacier (42,603 m 2 ) as well as 205 almost three and a half times the number of ponds (96 and 28 respectively). Sumna Glacier in the west of the 206 region showed major variations in ponded area and number over time, as its percentage pond cover ranged 207 from 2.16 % to 4.87 % during the four year study period. For most of the smaller glaciers, ponds occupied similar 208 locations at each time step (e.g. Ama Dablam Glacier SI. Fig. 2a and Lhotse Glacier SI. Fig. 2b), in addition to the 209 growth of new ponds (e.g. Lhotse Glacier SI. Fig. 2b). Sumna Glacier was the exception to this and showed a 210 distinctive change in the spatial pattern of the pond locations during the study period: at the start of the study 211 (2015), it had more ponds at higher elevations, but by the end (2018), ponds were most concentrated near the 212 terminus (Fig. 3d). Overall, our results show an increase in both ponded area and number of ponds on the 213 smaller glaciers, but this showed substantial spatial and temporal variation, even between neighbouring glaciers 214 and on the same glacier. 215

Glacier Elevation Profile 217
The study glaciers have an average slope > 18° at the tongue and < 25° at higher elevations. In general, areas 218 where the mean overall slope is lower (>10°) contain more ponds Overall, slope angle tends to decrease closer 219 to the glacier termini, where slopes are generally between 2° and 4° (Fig. 5). However, with the exception of 220 Sumna Glacier (see section 3.1.2.) there is no apparent relationship between elevation and pond number/ area 221 on the study glaciers (Figs. 5 and 6). We assessed this in further detail by dividing each glacier into 10 equal 222 elevation bands (to account for differences in total length and to facilitate direct comparison between glaciers) 223 and calculating the number and area of ponds in each elevation band (Figs. 5 and 6). Bands are numbered from 224 1 (the glacier head wall) to 10 (terminus). On six of the ten study glaciers, the number of ponds decreased from 225 the head of the glacier to terminus, whilst the pattern was reversed on the remaining 4, so that pond numbers 226 increased with distance up glacier ( Fig. 5 and 6). in slope in band 6 preceded a decrease in pond number from 7 to 1 (Fig. 5). 244

Glacier Velocity 245
Generally, glacier velocity decreased from source to terminus, being highest in bands 1-3 and lowest in the bands 246 8-10 (Fig. 7). Where velocities were higher, the number of surface ponds was generally lower (Fig. 7). For 247 instance, in bands 1 and 2 on Lhotse Shar Glacier there were no ponds recorded (velocity >20 ma -1 ). However, 248 further down glacier from band 3, velocities were lower (<20 m a -1 ) and the number of ponds increased by 14 249 (14.6 %) (Fig. 7). The main exception to this trend is Nuptse Glacier, where velocity is low in bands 1-2 (> 2 ma -250 1 ) and there are no ponds, but the area of higher velocity in bands 4 to 5 (>14 ma -1 ) contains a total of 34 ponds 251 (Fig. 7). In general, where velocities are lowest, for instance nearer the glacier terminus, pond number increases, 252 following a similar trend to that of glacier slope (section 3.2.1.). For example, from band 5 onwards on Lhotse 253 Shar Glacier, there is almost no recordable velocity and pond number reaches its highest (23 ponds) (Fig. 7). The 254 relationship between total pond area and glacier velocity is similar to that for pond number (Fig. 8): higher 255 velocities coincide with lower ponded area, whereas lower velocities have a higher ponded area (Fig. 8). Sumna 256 Glacier displays this relationship clearly: it has 23 % of its ponds in Bands 2-4 and 70 % in bands 6-8. This 257 corresponds to velocities of 3 to 7 ma -1 and < 3 ma -1 respectively. Overall, our results indicate that lower 258 velocities correspond with higher ponded area and pond number, and higher velocities generally relate to fewer 259 ponds. 260

Ice Cliffs 261
Although ice cliffs were identified on all 10 glaciers 2015-2018, the number of cliffs changed markedly during 262 this period, with cliffs seen to increase and decrease from year to year, and between glaciers (Fig. 9). Ngozumpa 263 and Khumbu glaciers had the highest percentage area of the glacier covered by ice cliffs, with 4.3% and 3.92% 264 respectively, and Sumna glacier the least (1.1%; Fig. 9). The greatest temporal variability was observed on Ama 265 Dablam Glacier, where ice-cliff coverage decreased from 2.59 % in 2015 to 1.3 % in 2016, and then rapidly 266 increased to 3.3 % in 2018 (Fig. 9). The number of supraglacial ponds with a corresponding ice-cliff exceeded the 267 number without: on average, across all of the study glaciers, 54 % of ponds had a coincident ice-cliff ( Fig. 9). 268 During the study, the number of ponds without an ice cliff increased on average by 1.6 % of the total glacier 269 surface area and this was most noticeable on the seven smaller glaciers (Fig. 9). For instance, on Sumna Glacier, 270 the area of the ponds with a cliff remained relatively stable (~ 0.8 %), whereas the area of ponds without a cliff 271 increased from 1 % in 2015 to 1.75 % in 2018 (Fig. 9). In comparison, the number of ponds with an ice cliff 272 increased by just 0.9 %, observed most notably on the three larger glaciers (Fig. 9). For example, on Ngozumpa where there was a decrease (0.32 %). 277

Future Lake Development 278
Each of the 10 study glaciers was assigned a number, according to the stage of lake development described in 279 established lake classification schemes (Komori, 2008; Table 3). In 2015 (Fig. 10b), the study 280 region was dominated by glaciers in Stage 2 of lake development, with 60% showing ponds that have coalesced, 281 were ice-dammed and have large areas (>20,000m 2 ). Only Ngozumpa Glacier was defined as Stage 3, due to the 282 presence of the large terminal Spillway Lake (Fig. 10). The remaining three glaciers (Sumna, Lhotse Nup and Ama 283 Dablam Glaciers) were all classified as Stage 1, with supraglacial ponds forming in their lower ablation zones (Fig.  284 10b). 285 Over the four-year study period, (December 2015-April 2018) two glaciers (Ama Dablam and Lhotse Nup) 286 transitioned to a new stage of lake development (Fig. 10). Both progressed from Stage 1, where a few 287 supraglacial ponds were identified, to Stage 2, where ponds had begun to coalesce (Fig. 10). Two glaciers 288 (Pangbung and Lhotse Shar) partially transitioned from Stage 1 to Stage 2, and from Stage 2 to Stage 3, 289 respectively (Fig. 10). Features that fit more than one stage were identified on these two glaciers, meaning that 290 they could not be assigned a stage using the current classification. For example, on Pangbung Glacier, ponds 291 were appearing on the lower ablation zone (characteristic of Stage 1), but some ponds were also beginning to 292 coalesce (Stage 2; Fig. 3c). On Lhotse Shar Glacier, coalescing was observed (Stage 2), but there was also stable 293 expansion of its proglacial lake (Stage 3). As a result, the current development cannot be captured by existing 294 schemes.  Table 2). 302 Similarly, the rate of expansion on Lhotse Glacier in our study (17.0 % a -1 ) is four times greater than that found 303 by Watson et al. (2016;4.1 % a -1 ). This is a major concern in terms of risks to downstream communities, as these 304 very high rates of pond growth will rapidly increase the water volumes available for outburst floods and will also 305 encourage pond coalescence and lake formation. 306 One potential explanation for the observed acceleration in pond growth is climatic controls: warmer 307 temperatures should increase melt rates and hence encourage pond expansion, whilst increased precipitation 308 could add water directly to the ponds. Data on climate trends proximal to our study glaciers are very limited. 309 However, available data (Salerno et al., 2015) suggest that minimum and mean air temperatures have risen in 310 the Everest area between 1994 and 2013, at elevations above 5000 m. However, warming was most marked in 311 spring and winter, and would thus have a more limited impact on ice melt, and it was concurrent with a reduction 312 precipitation, which would decrease direct inputs to the ponds (Salerno et al., 2015). As such, we suggest that 313 the observed increase in ponding may at least partly reflect changes in the dynamics of our study glaciers, which 314 provide the conditions that promote pond formation. Between 2000 and 2017, glaciers in East Nepal decelerated 315 by -1.8 ± 0.1 ma -1 (17.0 ± 1% a -1 ) and thinned, which in turn reduced driving stresses (Dehecq et  The three larger glaciers (Ngozumpa, Khumbu and Pangbung) increased in ponded area between 2015 and 2018, 324 but decreased in pond number (Fig. 2). This suggests that pond area is increasing via coalescence, as seen at the 325 terminus of Ngozumpa Glacier (Fig. 3a) and the eastern margins of Khumbu Glacier (Fig. 3b). This has implications 326 for both glacier lake related hazards and ice loss rates in the future. As ponds continue to join, and the area of 327 supraglacial ponds increases, the likelihood of proglacial lakes formation is increased (Komori, 2008 Generally, the smaller glaciers in the region do not have large, extensive ponding at their termini, but 338 the ponded area on all seven of the smaller glaciers increased during the study period (Fig. 2). For example, the 339 ponded area and pond number found on Ama Dablam Glacier increased by 38,958 m 2 (48 %) and 29 (43 %) 340 respectively 2015 to 2018 (Fig. 2b, c). Given the increase in area of the ponds but variations in pond number, we 341 suggest that this is primarily due to the coalescing of smaller area ponds (e.g. Ama Dablam Glacier SI. Fig. 3)   use Sentinel 2, which is 10 m resolution (1 pixel = 100 m 2 ). Our results demonstrate that ponds < 100 m² (one 352 pixel in Sentinel data) accounted for 3% -8% and those < 400 m² (four pixels in Sentinel data) comprised 28% -353 59% of total ponds found in the region (SI. Fig. 4). Of the total ponds identified, between 55 % and 86 % were 354 below 900 m 2 (one pixel in Landsat data). As such, using Landsat imagery to map pond changes in the region 355 would have missed the majority of the total number of ponds. Whilst these ponds are comparatively small in 356 area, including them in assessments is vital, as they inform us about where ponds are nucleating, and hence 357 controls on their formation. These data also indicate locations that may become ponded in the future, and 358 therefore subject to enhanced melt rates, and/or areas that may eventually coalesce with other ponded 359 sections. Furthermore, recent work has demonstrated that Sentinel-2 imagery has a better spectral contrast 360 between debris-cover ice and supraglacial ponds than Landsat of RapidEye, which affirms its suitability ( (Figs. 5 and 6). However, contrary to theory, 368 supraglacial ponds were also found in areas with much greater slope gradients (> 10°), both at the termini and 369 further up-glacier (Figs. 5 and 6). For instance, the highest number of ponds on Ama Dablam Glacier (76 %) were 370 located in the high accumulation zone (band 1-2). We also observed changes in the number of ponds after beaks 371 in slope (Fig. 5) Our results generally support this, as there were more ponds and a greater ponded area closer to the terminus 393 (bands 8-10), where velocities were lower (Figs. 7 and 8) For example, on Sumna, Imja and Lhoste Shar glaciers 394 no ponds were found in bands 1 or 2 where velocity was > 20 ma -1 , but as velocity decreased ( < 20 ma -1 ) pond 395 numbers began to increase (Fig. 7). However, as with glacier slope, there are exceptions to this relationship. For 396 example, the peak number of ponds corresponds with peak velocities on Lhotse and near-peak velocities on 397 Nupste (Fig. 7). The cause of these velocity patterns is difficult to elucidate, but we suggest that it may relate to 398 the pattern of debris inputs from the surrounding hillslopes (and hence the debris characteristics and 399 distribution), the location of tributary glaciers and/or meltwater inputs from the surrounding slopes. Overall our 400 data indicate that ice velocities influence pond area and number, but that the relationship is far for simple and 401 requires further detailed study. 402

Ice-Cliffs 403
All ten of the glaciers included in this study had surface ice-cliffs, and the majority of ponds were associated with 404 adjacent ice cliffs (54 %). This follows theory, as ice cliffs experience enhanced melt at a rate three to six times 405 greater than that of debris-covered ice (Kirkbride, 1993 (Josberger, 1978;Röhl, 408 2006). Despite this, our results show that the proportion of ponds without an ice cliff increased substantially 409 during the study period (2015-2018; Fig. 9). For example on Sumna Glacier, ponded area without an adjacent 410 cliff almost doubled during the study, whereas ponded area with a cliff actually decreased, meaning that there 411 were fewer ponds with an ice cliff than without in 2018 (Fig. 9). On Ngozumpa, in contrast, the percentage of 412 ponds with cliffs were on average three times higher (4.3 %) than those without (1.25 %). Overall, our results 413 show that ponds are expanding and that they are usually associated with ice cliffs (Figs. 2 and 9). However, as 414 ponded area increases, the proportion of the ponds with ice cliffs reduces (Fig. 9). We speculate that this may 415 be because pond growth is outstripping ice cliff formation, but this requires further investigation. However, Komori (2008) found glaciers in Bhutan took on average 40 years to pass through Stage 1 and 2, and 421 enter into Stage 3, whilst those the in Aoraki/Mt Cook region took 8-30 years for the same transition (Robertson, 422 2012). Given the short time-frame of this study (4 years), our results demonstrate that proglacial lakes in the 423 Everest region are evolving rapidly and quicker than other regions, both in the Himalaya and globally. This has 424 important implications for hazard assessments in the region, as these rapid changes require high temporal 425 resolution monitoring to determine potential changes in GLOF risk and could result in hazardous lakes forming 426 quickly in new locations. Our results therefore highlight the need for frequent monitoring of the hazards posed 427 by glacier lake growth in the Everest region, particularly given that Nepal (along with Bhutan) is at the greatest 428 economic risk from GLOFs (Carrivick and Tweed, 2016). 429 Two of our study glaciers exhibited characteristics of multiple classes of proglacial lake development 430 (Fig. 10). Specifically, on Pangbung Glacier, ponds were appearing on the lower ablation zone, which is indicative 431 of Stage 1 but some ponds were beginning to coalesce (Stage 2) ( Table 3). On Lhotse Shar Glacier, we observed 432 coalescing (Stage 2), but also the stable expansion of the proglacial lake (Stage 3). As such, the proglacial lake 433 development observed in our study region does not fit within the four stage classification system of Komori 434 (2008) and  ; Table 3). Additionally, changes in ponded area, number of ponds and ice-cliffs were 435 observed on all glaciers (Figs. 2 and 9), but these substantial differences were not accounted for in the current 436 model. For example, both Khumbu and Nuptse glaciers remained in Stage 2 throughout the study period, but 437 experienced increases in ice-cliffs (45 % and 76 % respectively; Fig. 9). Ngozumpa Glacier remained in Stage 3,438 but showed a marked increase in both ponded area (33 %) and the number of ponds with ice-cliffs (24 %) (Fig.  439 2c and Fig. 9). As such, the current classification model does not account for many of the observed changes in 440 pond area and characteristics, which could contribute to proglacial lake growth, and does not account for 441 glaciers that bridge different categories. Thus, on the basis of our observations, we propose a six-stage 442 categorisation which accounts for the 'in-stage' changes observed, such as appearance of ice-cliffs and marginal 443 pond expansion (Fig. 11). The inclusion of two additional stages (shown in red), now include 'in-stage' changes 444 observed in this study, such as marginal expansion, ice-cliff formation and pond drainage. The inclusion of these 445 stages would enable us to assign our study glaciers to just one stage, making the model more suitable for 446 evaluating and communicating glacier lake hazard potential. 447

Lake Development Trajectories and Outburst Risk 448
A key observation of this study was the coalescing of smaller area ponds on the eastern margins of Khumbu 449 Glacier and the Ngozumpa terminus (e.g. Fig. 3a  During the 4 years of our study, Imja Lake expanded by 456 m up-glacier and Spillway Lake increased in 461 area by ~ 14% a -1 . At the same time, we saw large ponds form via coalescing on the margins of Khumbu and 462 Ngozumpa glaciers, and coalescence begin on Ama Dablam Glacier ( Fig. 3 and SI. Fig. 2a). We also observed an 463 increase ponded area increased on all of our study glaciers (Fig. 2). As such, our results show that increasing 464 volumes of water are being stored on, and in front of glaciers in the Everest region, and comparison to previous 465 work (Watson et al., 2016) suggests that trend is accelerating. The glaciers in our study region appear to be 466 rapidly moving along the trajectory of proglacial lake formation, and may be doing so quicker than other regions 467 (Komori, 2008;). These developments have major implications for downstream GLOF risk, as 468 lake volume and speed of expansion are key factors in the hazard potential of proglacial lakes (Rounce et al., 469 2017b). As such, there is an urgent need for high resolution monitoring of ice-surface water volumes and 470 proglacial lake development in the Everest Region. 471 472

Conclusions 473
This study represents the most up-to-date assessment of supraglacial ponds in the Everest region of Nepal. All 474 10 glaciers demonstrated an overall ponded area increase over the period 2015 to 2018, ranging from 13.6 % to 475 108 %. Given the short time span of this study, this is a marked acceleration in the rate of pond growth compared 476 to that of 2000 -2015 (Watson et al., 2016). This is a major concern for proglacial lake formation and thus future 477 risk to downstream communities. A shift towards pond coalescing was observed, most notably on the three 478 larger glaciers (Ngozumpa, Khumbu and Pangbung). This suggests a transition towards larger surface ponds and 479 lakes. Despite this, smaller area ponds (< 900 m 2 ) continue to form (accounting for between 55 % and 86 % of 480 the ponds identified in this study) highlighting the need for higher resolution imagery for future remote sensing 481 studies. Our data show the conditions for pond formations are now predominantly found at mid-elevations on 482 glaciers in the Everest region, and the area where ponds can form may be advancing up-glacier. This has 483 important implications for future ice loss in the region. Velocity has been shown to influence both ponded area 484 and pond number, however this relationship requires further study. Ice cliffs were found on all 10 study glaciers, 485 with the highest proportion found on the three larger glaciers, however the rate of formation appears lower 486 than the rate of pond formation in this region. Two glaciers (Ama Dablam and Lhotse Nup) progressed to a new 487 stage of proglacial lake development during the four year study period. This transition is markedly faster than 488 observations elsewhere (e.g. Komori, 2008;, and suggests proglacial lakes are evolving much 489 quicker here than other regions, and as such the situation requires continued high temporal resolution 490 monitoring. Given this, our results show existing classification schemes are too simplistic to suitably evaluate 491 and communicate glacier lake hazards, leading to the proposal of a new, six-stage model more suited to 492 evaluating and communicating GLOF hazards. 493

Conflict of Interest 494
The authors declare that the research was conducted in the absence of any commercial or financial relationships 495 that could be construed as a potential conflict of interest. 496

Acknowledgements 497
We acknowledge a number of freely available datasets used in this study. We are also grateful to Dehecq et al. 498 (2015) for velocity data.

3
Stable expansion of ponds up-glacier and shift to moraine dammed proglacial lake.

4
Glacier retreats out of the proglacial lake.