The field survey at Korumdu lake (Figs. 1, 2) was run during the summers of 2015–2019. We installed water level and water temperature data loggers (Hobo U20) at lake bottom and ground surface on the moraine to collect measurements once per hour since 21 August 2015. Water level logger measurements (water pressure data) at lake bottom were converted to the water level (meter) using atmospheric pressure data at the adjacent ground surface on the moraine. A time-lapse camera (Brinto) was installed as well, which took one oblique image of the area per day.

In addition, we obtained aerial images of the lake basin acquired by Phantom-4 (DJI) and JABO H601G (Medix) uncrewed aerial vehicles (UAVs) with a mounted camera (Ricoh GR) on 21 August 2015, 12 August 2016, 6 August 2017, 20 July 2018, and 4 August 2019. High-resolution orthoimages and digital surface models (DSMs; resolution of 0.2 m) were made using the Pix4D mapper (Pix4D SA) of structure from motion (SfM) software and ground control points (GCPs). We collected the GCPs around the lake using a Trimble GeoExplore 6000 Global Navigation Satellite System (GNSS). The absolute positions of GCPs were corrected during post-processing using data from the Kyrgyz GNSS reference station and had an accuracy of 30–40 cm. Surface elevation changes of the moraine complex surrounding the lake were computed in ArcGIS 10.5 by comparing UAV-derived DSMs from 2015 and 2016.

The daily volume and discharge of the lake during the summers of 2017–2019 were calculated using the daily water level data and the 2017–2019 UAV-derived DSMs combined with the 2016 UAV-derived DSM (without water) in ArcGIS 10.5. For the water volume of the lake's bottom layer, we used the 2016 DSM because the 2017–2019 DSMs had water at the lake bottom. In addition, we investigated whether satellite remote sensing data could replace in situ water level logger data to calculate lake water levels using the combined DSMs. We found that the water level logger measurements agreed with the derived water levels based on UAV-derived DSMs combined with satellite imagery. For example, we confirmed the position of the water level by comparing a UAV orthorectify image or satellite data with 1 m contour lines from the combined UAV-derived DSMs. Finally, we obtained the water level and lake area from satellite data. Using this method, we reconstructed the water level data between 4 and 31 August 2019 based on 9 satellite images from PlanetScope and Sentinel-2 because we do not have water level data after our last field survey on 4 August 2019. We also investigated the changes in lake area during 2017–2019 using PlanetScope images.

Finally, we examined the meteorological and thermal conditions using air and ground temperature data loggers (TR-52i, T&D; accuracy ±0.3 ∘C) to log data at 1 h intervals around the lake (Fig. 2). Mean annual air temperature (MAAT) between 2015 and 2017 and mean annual ground surface temperature (MAGST) during 2015–2019 were calculated.

Short-lived glacial lakes in the northern Teskey Range were identified in ArcGIS 10.5 using satellite images (Landsat 7 Enhanced Thematic Mapper Plus (ETM+, SLC-off), Landsat 8 Operational Land Imager (OLI), Sentinel-2, and PlanetScope) acquired during 2013–2018 (Supplement Table 1). We used the definition by Daiyrov et al. (2018) for short-lived lakes, which is based on seasonal changes in lake area over the summer months of each year. Specifically, a short-lived lake is a temporary lake, lasting just 1 or 2 years, that suddenly appears or increases substantially in area and then disappears or shrinks within the same year. We counted the number of lakes that appeared from June to September each year. In addition, the number of lakes was tracked in each given year to examine how it changed from one year to the next. Polygon shapefiles of lakes were digitized manually from the images using ArcGIS 10.5. We also investigated the area changes of short-lived glacial lakes during summer months in a given year.

ALOS AVNIR-2 data taken on 17 September 2007 indicated that most of the lake basin had been covered by Korumdu glacier. Thus, the lake basin developed in a depression that formed during the retreat of the glacier in recent years. The UAV orthoimage of the basin in 2019 indicated a length of 360 m, a width at mid-length of 110 m, and a total area of 0.062 km2. The lake basin volume increased from 264 000 to 330 000 m3 from 2017 to 2019 (Fig. 2) due to retreat of the glacier terminus. In the field, we observed ice-exposed ridge and debris sliding on the basin's slope, indicating that the ice was melting around the shore, thus increasing the basin's width.

The lake had no discernable surface drainage channel, but we found an outlet point where meltwater from the lake emerges from a subsurface ice tunnel within the ice-cored moraine complex that connects to the lake (Fig. 2). The length of the outlet ice tunnel is 60 m from the entrance of the lake basin. During the fieldwork, we observed meltwater draining at the outlet point on 30 July 2015, 6 August 2017, and 4 August 2019 but not on 12 August 2016 and 20 July 2018.

Concerning lake size changes, in 2015, the lake appeared sometime before 30 July and then shrank significantly by 21 August (Fig. 3). In 2016, according to the water level data and on-site time-lapse camera images, the lake area did not form. For 2017–2019, more images of the area could be acquired and thus a more detailed evolution of changes in lake size is shown with a sequence of PlanetScope satellite images in Fig. 4. The images show that the lake appears suddenly at the end of July to the beginning of August and then shrinks and vanishes by the end of August (Fig. 4). Although the timing of lake expansion differs slightly over the years 2017–2019, the lake always appears in summer. The time-lapse on-site images show the same behavior from a different view (Fig. 5). These images also indicate that the lake began to expand from mid-July and reached its maximum size at some time between late July and early August. Thus, these data demonstrate that the lake is a short-lived glacial lake.

Changes in water level, area, volume, and discharge of Korumdu lake were studied in detail during the three summers of 2017–2019. For 2017, Fig. 6a shows the water level increasing from 6 July, reaching a maximum on 3 August, and then the lake is empty on 19 August. Within 29 d, the water level increases 13 m, the area reaches 0.36 km2 (Fig. 6b), and the volume reaches 234 000 m3 (Fig. 6c). The resulting rate of lake volume increase is 8070 m3 per day. During the emptying of the lake, 234 000 m3 of water drains in 17 d, with half of the volume draining from 3 to 7 August 2017 (Fig. 6c), resulting in a maximum net outflow discharge of 0.66 m3/s (Fig. 6d). Although the water level increases intermittently before 3 August, the net outflow is nearly continuous. The lake water temperature averages about 1 ∘C (Fig. 6a). The water temperature fluctuates more when the lake is shallower because the heating of shallower water by solar irradiance is stronger than cooling from inflowing ice meltwater.

In 2018, the water level peaks three times, though it reaches only about half that of 2017 (Fig. 6a). The first peak, on 25 July, occurs with a lake depth of 3.5 m and a volume of 21 000 m3 (Fig. 6a, c). The second, and maximum peak, occurs on 11 August with a lake depth of 6 m and a volume of 53 000 m3. The third peak occurs on 17 August with a lake depth of 5 m and a volume of 39 000 m3. The maximum net discharge occurs after the second peak, reaching 0.32 m3/s (Fig. 6d). Similar to 2017, the net inflow rate also clearly varies over time in 2018.

In 2019, the lake water level rises and falls before 22 July, when it rises sharply (Fig. 6a). Then, the water level shows a maximum around 30–31 July, reaching a lake depth of 5 m and a volume of 53 000 m3. The 2019 maximum level occurs on 11 August, with a lake depth of 6.5 m and a corresponding volume of 74 000 m3 (Fig. 6a, c). The maximum discharge occurs right after the second peak, reaching 0.48 m3/s (Fig. 6d).

Over all 3 years, the highest water level is in 2017 (Fig. 6a). In general, each year differs in the timing of lake-level increase, number of peaks, and maximum water volume. All 3 years have relatively small net discharge rates (maxima of 0.66, 0.32, and 0.48 m3/s in 2017, 2018, and 2019), which is consistent with the absence of reported flooding.

During each of these years, the lake level rose and fell several times, indicating repeated storage-drainage cycles. In the field, we observed sudden small increases in water level in 2016 and 2017, with the lake level increasing tens of centimeters within 3 h (Fig. 7). These results indicate that water level fluctuations occurred frequently due to closing and opening of the outlet ice tunnel.

During the fieldwork, we observed lake water draining out an outlet point in 2015, 2017, and 2019 but not in 2016 and 2018. In years in which drainage occurred, the lake elevation exceeded that of the outlet ice tunnel entrance. For example, the water levels were at 3810 m a.s.l. on 21 August 2015, 3816 m a.s.l. on 6 August 2017, and 3810 m a.s.l. on 4 August 2019, all above that of the outlet ice tunnel entrance at approximately 3807.5 m a.s.l. In contrast, in 2016 and 2018, lake water levels were lower than 3807.5 m a.s.l., the outlet tunnel elevation (Fig. 8a, c). Therefore, the field-survey results indicate that the key factor determining where drainage occurs is the relative elevations of the lake surface and outlet ice tunnel entrance.

Between 2015 and 2016, debris sliding and horizontal backwasting around the lake exposed an ice ridge of up to 7 m height (Fig. 9). The backwasting indicates that melting of debris-covered ice occurred, which is supported by comparing the UAV-derived DSMs from both years (Fig. 9c). For instance, along the cross-sectional profile a–a' in Fig. 9b, the surface elevation decreased by about 5 m (Fig. 9c). These results are consistent with closure in the outlet ice tunnel during ice-melt period due to debris sliding and ice–debris deposition. During our fieldwork in 2016, we observed water flow at the entrance of an ice tunnel. After 2 or 3 h, the lake level increased (Fig. 7), consistent with the cause being closure of the ice tunnel.

In the northern part of the Teskey Range, the discontinuous mountain permafrost zone lies above 3100–3200 m a.s.l. (Daiyrov et al., 2018). Around Korumdu lake (3806 m a.s.l.), the mean annual air temperature (MAAT) during 2015–2017 was -4.8 ∘C, and the mean annual ground surface temperature (MAGST) during 2015–2019 was -2.9 ∘C. Thus, the buried ice of the ice-cored moraine complex at Korumdu lake is maintained under a permafrost environment. Melting of buried ice causes surface changes including expansion of the lake basin and expansion and deposition (closure) in the outlet ice tunnel.

To determine when other short-lived glacial lakes in the northern Teskey Range formed, we used satellite images of 2013–2018 (see Supplement Table 1). Based on the satellite imagery, a total of 160 short-lived glacial lakes could be identified. A classification of these lakes by month of appearance is shown in Fig. 10. Most lakes appeared in June (43 lakes) during the snow-melt period and in July (90 lakes) during the ice-melt period. The total numbers and the proportions of the numbers for these two periods varied during the 6 years. The number of lakes vary greatly by year and by appearance date, indicating that the formation of these short-lived glacial lakes can not be explained solely by an increase in meltwater during summer. Such variability has been argued to be related to geomorphological conditions such as drainage through ice tunnel inside of ice-cored moraine complex (Daiyrov et al., 2018).

Concerning reappearances, 81 lakes appeared only once during 6 years. Of the remaining, 19 lakes appeared twice, 7 lakes appeared three times, 2 lakes appeared four times, and 2 lakes appeared all 6 years. These results are consistent with tunnel closure being the main cause of formation. Short-lived glacial lakes that reappear during many years likely have an environment that either favors tunnel closure and hence lake formation or an increase in meltwater from glacier during summer (Daiyrov et al., 2020).

We first consider four previously studied short-lived lakes of tunnel type in this area. The Kashkasuu (2006), western Zyndan (2008), Jeruy (2013), and Karateke (2014) lakes appeared in May–June and expanded in area until June–July, and then all had relatively large drainage events leading to serious damages (Narama et al., 2010a, 2018). This timing of lake appearance suggests an ice tunnel closure that is caused by the freezing of stored water during winter or deposition of ice–debris mixture as sketched in Fig. 11a (Popov, 1987; Narama et al., 2010a, 2018). We call this the deposition–freezing type of ice tunnel closure.

In contrast, Korumdu lake appeared during July–August (except in 2016) and produced relatively little drainage during emptying. This different appearance time might reflect a different formation process. As we observed subsidence and downwasting changing the lake basin (Fig. 9), the blockages of the outlet ice tunnel at its entrance or interior were likely caused by deposition of ice–debris mixture from thermal erosion. This type of blockage (deposition–collapse type) is sketched in Fig. 11b. The water level fluctuations support this mechanism. The fluctuations of lake water level and discharge spikes reveal changes in the ice tunnel morphology (Fig. 6d). A sudden blockage of an outlet ice tunnel can cause a rapid increase in water level within a few weeks. Also, the water level increase was sporadic, indicating that the outlet ice tunnel was not completely closed, the blockage was temporary, and the size of the ice tunnel is small. As a result, lake drainages can occur any time in summer, depending on how the outlet ice tunnel responds to changes in water pressure or deposition of ice–debris mixture through melting processes.

In the northern Teskey Range, the Toguz-Bulak glacial lake appeared in June and disappeared in September every year from 2010 through 2019 due to the inflow of glacier meltwater (Daiyrov et al., 2020). This lake has a surface drainage channel from the lake, but its incoming glacial runoff controls its behavior, such as its area. Thus, as for short-lived glacial lakes with surface drainage channels like Toguz-Bulak, the evolution of the glacier mass balance during summer (amount of snow and ice melt flowing into the lake) also plays an important role for the formation and evolution of short-lived glacial lakes having a subsurface outlet ice tunnel.

In 2017, there were two trends in water volume of Korumdu lake (Fig. 6c). The first period (5–25 July) involved sporadic fluctuations superimposed on an increase in water volume, indicating incomplete closure of the ice tunnel. Then, in the second period of 26 July to 3 August, the volume continuously and rapidly increases, indicating complete closure of the ice tunnel. Hence, we argue that the main factor of these rapid lake-area changes is tunnel closure. The lake area reached its maximum in 2017. This indicates that the tunnel closure was longer in 2017 than in 2018 or 2019. Longer periods of tunnel closure are associated with the formation of larger short-lived glacial lakes (Narama et al., 2018). Thus, the period of closure is likely determined by the morphology of the ice tunnel and deposition condition of tunnel-closure point (e.g., when melting can open the blocked region).

Many of the other short-lived glacial lakes in the northern Teskey Range, observed via satellite imagery, are likely to belong to the deposition–collapse type as well. However, some likely have a larger influence from the water balance between drainage and storage, related to increasing glacial meltwater and tunnel size. Consider Jeruy glacial lake between 2014 and 2016 (Fig. 12). Ice melting caused distinct changes and rapid deposition within the outlet ice tunnel, which likely led to tunnel closure. Thus, morphology and surface characteristics of an ice-cored moraine complex within the mountain permafrost zone are prone to frequent changes, and the deposition–collapse type is likely the main type for the short-lived glacial lakes in the northern Teskey Range. If the deposition–collapse processes occur in summer when the melting rate is high, the formation of a short-lived glacial lake is highly likely.

Of the 160 short-lived lakes we identified in 2013–2018, only Jeruy lake in 2013 and Karateke lake in 2014 showed considerable drainage. The estimated maximum discharges from Jeruy (182 000 m3) and Karateke (123 000 m3) lakes were 14.9 and 11.5 m3/s, respectively (Narama et al., 2018). These lakes had relatively large outlet tunnels, with one at Jeruy, as well as one at Karateke, having a cross section of about 8 m2 (Fig. 12b). Earlier, in 2008, the western Zyndan lake (437 000 m3) emptied at a higher discharge rate of 27 m3/s (Narama et al., 2010a). Most of the water in these three cases drained over a period of several hours. In contrast, Korumdu lake did not show such high drainage rates during 2017–2019, draining at a maximum rate of 0.66 m3/s in 2017, taking 17 d to drain 234 000 m3, and its tunnel cross section was much smaller than those at Jeruy lake or Karateke lake.

In addition, Korumdu lake exhibited sudden fluctuations of water level over several hours, which we argue was related to closure of the small outlet ice tunnel caused by deposition of and blockage by debris. The relatively small tunnel size of this lake resulted in slower lake discharge even when lake volume reached its maximum (330 000 m3). During 2017–2019, the lake size was largest in 2017, yet discharge rates were in the same order of magnitude every year. These results show that, at least for Korumdu lake, the dimensions of the outlet ice tunnel were the dominant factor controlling lake discharge rates.

However, tunnel dimensions could increase in the future due to thermal erosion and flowing water, allowing greater discharge rates. Meltwater and increasing temperature can accelerate thermokarst processes enlarging the outlet ice tunnel (Sakai et al., 2000; Kääb et al., 2001; Miles et al., 2018). In addition, although lake basin size changes on ice-cored moraine complexes depend on the details of the thermal erosion, the basin area of Korumdu lake has increased each year due to glacier retreat. If these conditions and evolution also apply to other short-lived glacier lakes in the Teskey Range, large-scale flooding events during their discharge may become more frequent in the future due to increasing temperature.