1. Introduction
Glacial lakes serve as an indicator of climate change (Moser and others, Reference Moser2019). The formation of glacial lakes can initiate positive feedback mechanisms (Tweed and Carrivick, Reference Tweed and Carrivick2015), where the presence of lakes accelerates ice loss through processes such as calving and subaqueous melting (Carrivick and Tweed, Reference Carrivick and Tweed2013; Sutherland and others, Reference Sutherland, Carrivick, Gandy, Shulmeister, Quincey and Cornford2020). This, in turn, leads to increased melting and glacier retreat (King and others, Reference King, Bhattacharya, Bhambri and Bolch2019; Watson and others, Reference Watson, Kargel, Shugar, Haritashya, Schiassi and Furfaro2020), further contributing to lake expansion (Carrivick and others, Reference Carrivick, Tweed, Sutherland and Mallalieu2020). In the Himalayas, numerous glacial lakes, especially moraine-dammed ones, have recently formed and expanded around alpine glaciers (Zhang and others, Reference Zhang, Yao, Xie, Wang and Yang2015, Reference Zhang, Wang, An and Wei2023a; Shugar and others, Reference Shugar2020). These growing lakes are often bordered by steep slopes and held by fragile moraines, which are susceptible to failure (Westoby and others, Reference Westoby, Glasser, Brasington, Hambrey, Quincey and Reynolds2014). Catastrophic events such as rockfalls, glacier calving, and avalanches into the lake can generate large waves and trigger flooding due to outbursts (Byers and others, Reference Byers, Rounce, Shugar, Lala, Byers and Regmi2019; Fischer and others, Reference Fischer, Korup, Veh and Walz2021).
Over the past decades, numerous large Glacial Lake Outburst Floods (GLOFs) have occurred in the Himalayan region (Nie and others, Reference Nie, Liu, Wang, Zhang, Sheng and Liu2018; Shrestha and others, Reference Shrestha2023), which have caused significant damage to downstream communities across the globe (Richardson and Reynolds, Reference Richardson and Reynolds2000; Zhang and others, Reference Zhang, Wang, An and Wei2023a; Sattar and others, Reference Sattar2025). Additionally, small-scale GLOFs occur frequently but remain unnoticed due to limited damage or remote locations (ICIMOD, 2011). GLOFs and related events threaten millions of people worldwide (Veh and others, Reference Veh, Korup, von Specht, Roessner and Walz2019; Taylor and others, Reference Taylor, Robinson, Dunning, Rachel Carr and Westoby2023), and the frequency of such events is expected to increase in the future (Zheng and others, Reference Zheng2021). The continued ice loss due to climate warming (Kraaijenbrink and others, Reference Kraaijenbrink, Bierkens, Lutz and Immerzeel2017) and expansion of glacial lakes due to climate change therefore represent a globally important natural hazard that requires urgent attention if future loss of life from GLOFs is to be minimised. Therefore, improved monitoring of glacial lakes and a better understanding of GLOF mechanisms are essential.
Lake volume is an important parameter in GLOF hazard assessments (Fujita and others, Reference Fujita2013; Cook and Quincey, Reference Cook and Quincey2015). Precisely estimating the storage capacity of glacial lakes is vital for forecasting flood peak discharge, modelling flood progression, and creating early warning systems. Glacial lakes are most often located in remote places, making it difficult to carry out detailed bathymetric surveys. Consequently, lake depths and volumes for unmeasured lakes are often estimated using Volume–Area (VA) empirical relationships developed mainly from small bathymetric datasets (Huggel and others, Reference Huggel, Kääb, Haeberli, Teysseire and Paul2002; Wang and others, Reference Wang2012; Loriaux and Casassa, Reference Loriaux and Casassa2013; Muñoz and others, Reference Muñoz, Huggel, Frey, Cochachin and Haeberli2020; Zhang and others, Reference Zhang, Wang and An2023b). In the Himalayan region, the harsh climate and remoteness of moraine-dammed lakes often result in a shortage of ground-based monitoring data, particularly regarding lake water depth. To date, in-situ measurements of lake water depth in the Himalayas have been conducted at only a few glacial lakes, mainly concentrated in the central and eastern Himalayas (Table S1). Most in-situ bathymetry-based studies are concentrated in the Everest region of the Nepal Himalaya (Table S1), with only a few lakes monitored in the Lunana region of the Bhutan Himalaya, and in Tibet (Zhang and others, Reference Zhang2023c).
To the best of our knowledge, only a few lakes in the Indian Himalayas have been surveyed using in-situ bathymetry, including South Lhonak (Sharma and others, Reference Sharma, Pradhan, Sharma and Shrestha2018), Drang Drung (Ramsankaran and others, Reference Ramsankaran, Verma, Majeed and Rashid2023) and Neelkanth Lake (Deswal and others, Reference Deswal, Sharma, Saini, Dalal and Kumar2020). However, many of these studies relied on manual depth measurements using a rope and canoe, resulting in scattered and spatially limited depth points rather than full-coverage bathymetric surveys. This approach introduces significant uncertainties in volume estimation and lake basin characterisation, affecting hazard assessments. The lack of systematic, high-resolution depth data highlights the need for better in-situ survey strategies. To address these limitations, various approaches have been employed across the Himalayas, including portable echo sounders mounted on kayaks (Ramsankaran and others, Reference Ramsankaran, Verma, Majeed and Rashid2023), and echo sounders integrated with Uncrewed Surface Vehicle (USV) (e.g. Haritashya and others, Reference Haritashya2018; Li and others, Reference Li2021; Qi and others, Reference Qi2022; Duan and others, Reference Duan2023; Zhang and others, Reference Zhang2023c). Among these, the USV-based survey of glacial lakes is more convenient, efficient and time-effective and provides robust datasets, compared to the conventional manual depth measurements. In this study, we address these gaps by employing a hybrid approach using both USV- and kayak-mounted echo sounders, which enables denser, spatially continuous depth measurements, improving the accuracy and reliability of glacial lake volume estimations in the region.
In recent decades, the western Himalayan region has experienced significant growth of glacial lakes (Allen and others, Reference Allen, Linsbauer, Randhawa, Huggel, Rana and Kumari2016; Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Sharma, Murari, Nüsser and Schmidt2023, Reference Das, Das, Mandal, Sharma and Ramsankaran2024), and several lakes have been categorised as potentially dangerous glacial lakes for future GLOF (Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). The small number of published lake-bathymetry measurements in the western Himalayas has resulted in a limited understanding of the morphology and storage capacity of glacial lakes in this region. In this study, we report the results of the first in-situ surveys conducted on four glacial lakes—Kya Tso Lake (KTL), Panchi Nala Lake (PNL), Gepang Gath Lake (GGL) and Samudra Tapu Lake (STL)—located in the Chandrabhaga basin, western Himalayas, India. First, we used bathymetric data collected with high-resolution (i.e. precision and density of depth measurements) echo sounders mounted on a USV and a kayak to model lake basin morphology using Triangulated Irregular Networks (TIN) and calculate their volumes. We then compared our in-situ volume estimates with those derived from empirical VA relationships, highlighting the observed discrepancies. Lastly, we identified key challenges encountered during USV surveys in high-altitude glacial lakes and provided recommendations for improving future surveys.
2. Study areas
This study investigates four glacial lakes (i.e. KTL, PNL, GGL and STL) in the western Himalayas (Table 1; Fig. 1), selected for their contrasting topographical and morphological settings. All four lakes are fed by temperate glaciers within the greater Himalayan range (Table 1; Fig. 1) and exhibit distinct geomorphic characteristics (Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). KTL is a bedrock-dammed lake that is no longer connected to its parent glacier, while PNL, GGL and STL are moraine-dammed lakes that remain in contact with active glacier termini. These lakes were selected for bathymetric surveys based on accessibility and their alignment with established research frameworks for analysing glacial dynamics. Specifically, GGL and STL were prioritised because they have been classified as potentially dangerous glacial lakes for future lake outburst floods in earlier studies (Allen and others, Reference Allen, Linsbauer, Randhawa, Huggel, Rana and Kumari2016; Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024), and thus warranted immediate attention for detailed bathymetric assessment. In addition, we included PNL, which is also categorised as moderately susceptible to future outburst floods (Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). KTL, although not classified as currently dangerous, was included to increase the diversity of surveyed lake types—particularly a bedrock-dammed, ice-disconnected lake—thus allowing comparative analysis of contrasting morphological settings. Increasing the number and diversity of lakes with in-situ bathymetric data also supports the development of improved empirical VA relationships and contributes to a more nuanced understanding of lake evolution and outburst flood potential in different geomorphic settings. The characteristics of each lake are discussed below.
Location of studied glacial lakes in upper Chenab (Panchi Nala, Gepang Gath and Samudra Tapu) and Zanskar (Kya Tso) basins, western Himalaya (upper panel). PlanetScope false colour composite images of glacial lakes as of August 2024 (lower panel). Gl = Glacier.
Characteristics of the studied lakes
a Glacier IDs are obtained from the Global Land Ice Measurements from Space (GLIMS; https://www.glims.org/) inventory.
b Area and length were measured for August 2024 (see supplementary information for details).
2.1. Kya Tso Lake
The KTL, situated at the headwaters of the Tsarap Chhu River (a tributary of the Zanskar River), lies at an elevation of 5144 meters above sea level (m asl; Table 1). The lake spans 445 m east-west and 330 m north-south, forming a bedrock-dammed cirque lake following glacial recession (Fig. 2). As of 2024, KTL is detached from its parent glacier, and the distance between the lake and the parent glacier is ∼446 m. Glacier meltwater serves as the lake’s primary recharge source. The right flank of the lake features steep slopes with unconsolidated loose sediments, whereas the left flank comprises gently sloping terrain dominated by exposed bedrock (Fig. 2).
Field photographs of Kya Tso Lake (KTL). KTL is detached from the parent glacier. The outlet is characterised by narrow and shallow channels, mostly occupied by unsorted boulders. The right-side mountain slope of the lake is steeper than the left side, characterised by talus/scree deposits. KTL is a bedrock-dammed cirque lake, formed on palaeo-cirque depressions. A water discharge measurement instrument was installed in August 2024 by the National Centre for Polar and Ocean Research (NCPOR), Govt. of India.
From 1971 to 1993, KTL exhibited significant expansion: its length increased from 282 m to 448 m, and its surface area grew from 0.04 km2 to 0.09 km2 (Text S1; Tables S3–S5; Fig. S1). However, post-1993, both parameters have stagnated (Tables S3–S5; Fig. S5). This stagnation contrasts with the growth patterns observed in other proglacial lakes across the western Himalayas. While KTL’s small size, lack of recent growth, and remote location (far from human settlements/infrastructure) reduce its immediate relevance for hazard assessment, bathymetric data could enhance future VA relationship models, contributing to broader regional datasets.
2.2. Panchi Nala Lake
The PNL is located at the headwaters of the Panchi Nala (nala = stream), a tributary of the Bhaga River in the Chandrabhaga basin (Table 1). It is situated at an altitude of 4560 m asl in 2024, with a length of 420 m from east to west and a width of 285 m from north to south. The lake is dammed by an end moraine. The recharge of PNL occurs mainly from the connected glacier’s meltwater. The terminus of Panchi Nala Glacier, with abundant ice cliffs covered by thick debris, stretches into the lake (Fig. 3a). PNL is turbid in nature and characterised by several sediment mounds (Fig. 3b). The lake cuts through the end moraine, having a width of ∼3 m as of June 2024 (Fig. 3c).
Field photographs of Panchi Nala Lake (PNL). PNL is a pro-glacial lake formed due to the blocking of water by frontal moraine and associated glacial recession. (a) The terminus of PNL as of June 2024 was characterised by exposed ice cliffs. The lake is characterised by several sediment deposits (boulders, sediment mounds), which depict the shallow depth of the lake. (b) View looking towards the up valley. (c) View looking down the valley.
PNL has evolved significantly in size and length during the last 53 years (Tables S3–S5; Fig. S2). Between 1971 and 2024, the lake’s length increased from 101 to 376 m (Text S1; Tables S3–S5; Fig. S5), corresponding to an average annual growth rate of 5.18 m a−1, while its area expanded from 0.01 to 0.11 km2, with an average growth rate of 0.002 km2 a−1 (Tables S3–S5; Fig. S5). Previous studies (Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024) have categorised PNL as moderately susceptible for future lake outburst floods due to several factors, such as the potential for mass movements into the lake from the surrounding steep valley walls and the downstream presence of a meteorological station. A meteorological station, situated ∼8 km downstream (∼3800 m asl) from the PNL along the Leh–Manali highway, is maintained by the Defence Geoinformatics Research Establishment (Government of India), and houses resident scientists as well as an adjacent Indo-Tibetan Border Police transit camp—underscoring its significance as critical infrastructure potentially exposed to GLOF hazards.
2.3. Gepang Gath Lake
The GGL is one of the top five prioritised glacial lakes in the western Himalayas (Allen and others, Reference Allen, Linsbauer, Randhawa, Huggel, Rana and Kumari2016; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). It is situated at an elevation of 4070 m asl, with a length of 2.49 km and a maximum width of 567 m as of 2024 (Table 1). GGL is fed by a ∼12.55 km2 compound valley glacier, which originates from four cirques on the northern slopes of Mount Gepang Goh (∼5870 m asl). GGL is bounded by the calving front of the Gepang Gath Glacier to the east, lateral moraines to the north and south and a recessional end moraine to the west (Fig. 4a and b). Ice calving is an active process at GGL (Fig. 4b and c). The lateral moraine trough acts as a ‘gutter’, trapping debris from rockfalls, snow avalanches and fluvial transport. However, large and rapid mass movements could potentially breach the lateral moraines and enter GGL, posing a significant hazard. The lake outlet is ∼6 m wide, and a bridge has been constructed to regulate the outflow, having a freeboard of 5 m as of August 2024 (Fig. 4d and e).
Field photographs of Gepang Gath Lake (GGL). (a) GGL as of August 2024. View looking up the valley. Floating ice chunks are very common in GGL. The water of GGL is sediment-free. Both sides of the lake are characterised by a lateral moraine ridge, which acts as a ‘gutter’ and prevents landslides on the lake from steep valley sides. (b) Close-up view of the terminus from the right lateral moraine. (c) The concave terminus shape depicts the fast retreat of the glacier. Calving is a very frequent event in GGL. (d) Lake outlet as of 7th August 2024, approx. 6 m wide. (e) A man-made structure is located on the left side of the outlet. The height difference between the lake water level and the structure is approx. 5 m. (f) GNSS base station and hydroboat survey at GGL. (g) Kayaking on GGL for manual depth measurements.
Since its formation from a small moraine-dammed lake in the 1970s, GGL has experienced substantial growth (Patel and others, Reference Patel, Sharma, Laluraj, Thamban, Singh and Ravindra2017; Kumar and others, Reference Kumar, Mehta and Shukla2021; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). The retreat of the Gepang Gath Glacier has directly facilitated the lake’s expansion (Fig. S3), with its surface area increasing from 0.15 km2 in 1971 to 1.04 km2 in 2024 (Text S1; Tables S3–S5; Fig. S5), corresponding to an average annual growth rate of 0.02 km2 a−1. The lake’s length has expanded from 580 m in 1971 to 2482 m in 2024, reflecting an average growth rate of ∼36 m a−1 (Tables S3–S5). Several studies (Allen and others, Reference Allen, Linsbauer, Randhawa, Huggel, Rana and Kumari2016; Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024) have categorised GGL as vulnerable to future GLOFs. The lake drains via Sissu Nalla, which joins the Chandra River ∼11 km downstream. Given its upstream location from Sissu village—a settlement that has experienced significant expansion in recent years—GGL presents notable hazard implications.
2.4. Samudra Tapu Lake
The STL is located at an elevation of 4156 m asl and is fed by the Samudra Tapu Glacier, the second-largest glacier in the Chandra Basin (Table 1). The lake is surrounded by steep mountain slopes composed of unconsolidated sediments (Fig. 5a). As of August 2024, STL measured 2.35 km in length and had a maximum width of 870 m (Tables S3–S5). Geomorphologically, STL is situated on an outwash plain and is impounded by low-gradient moraine deposits (Fig. 5b). The lake’s moraine is primarily composed of stratified sand and gravel, indicative of glaciofluvial deposition (Fig. 5b). STL is characterized by high turbidity, with substantial amounts of suspended sediment in the water (Fig. 5c).
Field photographs of Samudra Tapu Lake (STL). (a) Valley morphology of Samudra Tapu Glacier. STL is situated on the outwash plain depressions. Lake water is blocked by the small sediment ridges. The outlet is approx. 10 m wide. (b) Close-up view of the lake outlet showing the glaciofluvial deposit types. (c) View looking towards the up valley. The water of STL is highly turbid as compared to GGL. (d) Water ice interaction of STL. The frontal ice face has approx. free board of 23 m as of 17 August 2024. (d) Water-ice interaction near the valley edges of the right side of the terminus. Blue, white and red circles indicate the location of the zoomed panel.
Over the past five decades (1971–2024), STL has expanded from 0.24 km2 to 1.63 km2 in surface area (Text S1; Tables S3–S5; Figs S4–S5). The expansion of STL has been more pronounced (∼0.03 km2 a−1) than other studied lakes (Fig. S5). Previous studies (Allen and others, Reference Allen, Linsbauer, Randhawa, Huggel, Rana and Kumari2016; Rao and others, Reference Rao, Joshi, Roy, Sweta and Raju2023; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024) categorised STL as one of the potentially dangerous glacial lakes in the western Himalayas.
3. Data and methods
3.1. In-situ bathymetry survey
USV-based bathymetric surveys were conducted for three glacial lakes (KTL, GGL, STL) in August 2024, while a kayak-based survey was carried out for PNL in August 2022. All surveys were conducted during the ablation season (i.e. August) to minimise the presence of floating ice chunks on the water surface. Figure 6 provides a schematic overview of the methodology used for bathymetric data acquisition and processing.
Flowchart of methodology used in this study.
3.1.1. USV-based survey
A USV system (Satlab Hydroboat 1100 model) was used for this study. It comprises of four core modules: power, telemetry, navigation and data acquisition (Fig. 7a). The vessel is equipped with a lithium-ion battery powering dual propellers, offering up to ∼ 3 hours of endurance. Remote communication between the USV, controller and laptop is maintained via a telemetry system with a range of ∼ 1.5 km. Navigation is supported by a Real-Time Kinematic (RTK)-enabled Global Navigation Satellite System (GNSS) receiver (SL900), providing centimeter-level positional accuracy and is integrated with Hydral GCS software for real-time route tracking (Fig. 7b). Depth data were recorded using a 200 kHz single-beam echo sounder with a beam angle of 5°, operating in auto mode to dynamically adjust power, gain and threshold in response to depth conditions. This allowed improved echo returns and reduced noise across variable bathymetric profiles. A fixed sound velocity of 1480 m s−1, calibrated for 0°C water, and a boat draft of 0.1 m were used for accurate depth estimation. Technical specifications and system parameters of USV are detailed in Table 2.
A USV module, data collection procedure and post-processing of depth datasets. (a) Component of the Satlab Hydroboat 1100 USV module. (b) Steps associated with the bathymetry survey and data collection. (c) An example of post-processing of echograms in SLHydro software for Gepang Gath Lake. The red line denotes the simulative echo depths, and the blue dotted line (overlaid on the red line) represents the digital echo signal received by the transducer.
USV-based bathymetry survey specifications adopted in this study
Each lake survey was customised based on site-specific conditions such as lake size, depth and the presence of floating ice chunks to ensure optimal coverage and data quality. Bathymetric data were collected along a predefined grid of longitudinal and transverse transects and monitored in real time during the survey. Minor deviations from planned transects were made as needed to safely navigate around large icebergs while maintaining continuous data collection. Surveys were conducted during early morning hours to minimise wind-induced disturbances. Multiple GNSS base stations were strategically (i.e. maximising uninterrupted visibility and line-of-sight communication between the USV and the GNSS base station) placed along the lake edges, depending on lake length and terrain accessibility, to support accurate positioning. In floating ice-dominated sections of GGL (see Fig. 4b), we employed a hybrid method where the USV was tethered to a kayak and manually manoeuvred through dense ice patches. This ensured safe navigation without compromising sonar data collection (see Section 5.3 for operational challenges). This method was tested beforehand in controlled environments—at Powai Lake (19.12769°N, 72.90451°E), located adjacent to the Indian Institute of Technology Bombay (IITB) campus, and in the IITB campus swimming pool—to ensure operational feasibility. Supplementary equipment (i.e. inflatable kayak, rope, rugged laptop) and field strategies were employed as needed. The USV recorded depth data at one-second intervals throughout each of the surveys. The total survey distances recorded were 4.44 km for KTL, 28.51 km for GGL and 18.24 km for STL.
3.1.2. Kayak-based survey
We conducted the depth survey of PNL using a portable echo sounder mounted on an inflatable kayak. The hydrographic setup included a Catable Sea Eagle 380x Explorer kayak, an Airmar P66 single-beam transducer and a Simrad GO5 XSE fishfinder console (Table 3). The system recorded depth, geographic coordinates and timestamps through the Simrad’s built-in GPS interface. The complete survey kit also included safety and utility equipment such as paddles, life jackets, a 12 V battery, a rope with lead weight and a measuring tape. Ramsankaran and others (Reference Ramsankaran, Verma, Majeed and Rashid2023) demonstrated the capabilities of kayak-based survey procedure at high-altitude lakes in the western Himalaya. Technical specifications of the kayak-based system adopted in this study are summarised in Table 3.
Kayak-based bathymetry survey specifications adopted in this study
Based on the lake characteristics and morphology (see Fig. 3), the depth of the PNL was expected to be shallow, and therefore, the transducer was operated at a frequency of 200 kHz. The transducer was positioned at the side of the kayak with its face towards the forward direction of the movement. It was ensured that the transducer remained at a depth of 10 cm from the water surface while surveying to avoid errors due to air traps. Locations near large chunks of mud and rocks, as well as shallow depths of <1 m, were avoided during the survey to avert any possible collision and aground, which may cause damage to the kayak and the echo sounder. The Airmar P66 echo sounder collected depth points at approximately 1-m spatial intervals.
3.1.3. Manual depth measurements for validation
The bathymetry data recorded by the echo sounder were collected in continuous mode (Fig. 7c), while validation was carried out in waypoint mode. Manual depth measurements were performed at selected locations in GGL and STL to validate the echo sounder data. These validation points were primarily chosen in the shallower zones of the lakes to increase confidence in measurement accuracy. Depth was measured using a dead weight attached to a rope and a measuring tape. During the process, extra care was taken to ensure that the rope remained taut and vertical underwater, ensuring reliable depth readings. A total of six and 10 points were measured at GGL and STL, respectively, for cross-comparison with echo sounder depth data.
3.2. Bathymetry data processing and volume estimation
During post-processing, several noise artefacts were identified in the echogram, particularly in areas where the simulated depth (red line) and the digital echo signal (blue line) in the SLHydro Sounder interface did not align (Fig. 7c). The digital depth represents the raw signal derived from the echo sounder, based on the time taken for the sonar pulse to return from the lakebed. In contrast, the simulative depth is a modelled curve generated by the software, which applies bottom-detection algorithms to create a best-fit line from the raw data. Accurate depth measurements are typically indicated by close agreement between the two. However, in regions with rapid depth variations or weak bottom returns, noticeable mismatches occurred (Fig. 7c), indicating signal distortion or measurement uncertainty. The primary sources of noise could be attributed to: (i) intermittent GPS signal loss, (ii) motion-induced errors—particularly in deeper areas where return echoes were delayed and (iii) variable acoustic penetration due to turbidity and salinity differences. We closely inspected the echogram profiles and made corrections where applicable to ensure dataset reliability. Final post-processed depth points were exported at 1-m intervals for interpolation. The Airmar P66 transducer provides a bathymetry file with a binary format (.sl2 extension), which was processed in the ReefMaster version 2 software package (https://reefmaster.com.au/) for depth estimation.
Final post-processed depth points are heterogeneously distributed across all studied lakes (Fig. 8). Bathymetric models were created within ReefMaster software using the TIN interpolation technique. The data points were triangulated, and interpolation between them was performed based on the slope of the connecting triangles. The interpolated values were then gridded and further refined using Gaussian smoothing. Bathymetric modelling requires a shoreline with zero depth. Since no ‘zero depth’ data were collected from the outline extent of the lakes, we generated lake extents using multispectral PlanetScope imagery acquired closest to the survey dates (see supplementary Text S1 for details). Points were generated at 1-m intervals along the derived lake boundary polylines and assigned depths of zero. These points were appended onto the validated lake bathymetric points to complete the sounding databases for bathymetric modelling. Zero depth points were removed at the glacier calving front to avoid forcing the interpolated bathymetry to zero (Yao and others, Reference Yao, Liu, Sun, Wei and Guo2012; Haritashya and others, Reference Haritashya2018; Li and others, Reference Li2021). This was required since the TIN method does not interpolate outside the convex hull and therefore would not otherwise interpolate bathymetry to the calving front.
Sampling points of the USV for investigated lakes in the western Himalayas (background image: PlanetScope RGB): (a) Kya Tso Lake, (b) Panchi Nala Lake, (c) Gepang Gath Lake, and (d) Samudra Tapu Lake.
Depth maps were generated on a 1-m resolution grid and georeferenced to the WGS 84 ellipsoid. Lake volume was then estimated by integrating depth values across the mapped area, using the following equation:
\begin{equation}V = {\text{ }}\mathop \sum \limits_{i = 1}^n {D_i}*P\end{equation} where V is the volume (m3); 
${D_i}$ is the depth (m) at the i-th pixel; n is the number of pixels in the lake area; 
$P$ is the pixel size (m2) of the bathymetric map.
3.3. Uncertainty analysis
We estimated volume uncertainty by aggregating depth uncertainty from different sources. Following Haritashya and others (Reference Haritashya2018), we quantified depth uncertainty using water depth-normalised 2-σ (standard deviation). To assess this uncertainty, we randomly selected > 10 000 points distributed along and across all surveyed lakes. We validated the interpolated bathymetry depth by comparing it with depths obtained from the echo sounder. The overall (i.e. mean) depth uncertainty (2-σ) was found to be ∼1.5%. Additionally, the echo sounder data accuracy is suggested to be 0.01 m (± 0.1%) of depth. The sound velocity variations due to water temperature change have relatively high uncertainty, as temperature likely varies between 0°C and 4°C in the glacial lake, as reported for several proglacial lakes in the central Himalayan region (Chikita and others, Reference Chikita, Joshi, Jha and Hasegawa2000; Haritashya and others, Reference Haritashya2018). Moreover, for ± 2°C temperature uncertainty, the corresponding depth error is about ± 0.7% (e.g. about ± 0.2 m for a 30-m water depth) was considered, as proposed by Haritashya and others (Reference Haritashya2018). We excluded very small refraction-induced positional errors from error analysis for near-nadir projecting sound transmission through a single beam sonar system (i.e. Satlab 1100 hydroboat USV model). Our overall uncertainty for volume analysis was ± 2.3%. However, we acknowledge that the overall volume uncertainty may be underestimated, particularly in zones located farther from the actual survey tracks, where interpolated values contribute more significantly. This limitation is inherent in spatial interpolation and is more pronounced in areas with sparser data density.
4. Results
4.1. Comparison of manual and interpolated depths
The manual measurement points did not directly align with the Hydroboat survey path (Fig. 8c and d), so the comparison was made between the manually recorded depths and the interpolated depths derived from the echo sounder data (Table 4). The depth values recorded by the echo sounder showed good agreement with the manual measurements (Table 4). Manual depths ranged from 2.75 m to 15.11 m, with depth differences generally <0.5 m (Table 4). These differences may be attributed to wind conditions during manual measurements, which could have caused the lead-weighted rope to drift slightly, measuring a slant height rather than the true vertical depth. Although we selected areas with a high density of echo sounder data for manual measurements (see Fig. 8c and d), the interpolation process may still introduce minor deviations from the actual depth.
Manual depth measurements at various locations of the Gepang Gath and Samudra Tapu lakes
a Error = (Sonar − Manual survey)/Manual survey × 100%.
4.2. Bathymetry and lake volume
4.2.1. Kya Tso Lake
Based on the TIN data, the lake bathymetry of KTL was modelled as shown in Figure 9a. The KTL profile exhibits an oval shape (Fig. 9a). The basin bottom of the lake is relatively flat, while its flanks are steep, with slopes ranging from 60° to 70°. The right side is steeper than the left (Fig. 9a). From the water depth profile, it is observed that the depth gradually increases from the moraine dam to the centre of the lake and then gradually decreases towards the lake inlet (Fig. 10a). The depth changes from the lowest (0.24 m) to the highest (15.76 m) over a distance of 270 m from the lake outlet (Fig. 10a). Since KTL is detached from the parent glacier, changes in depth are represented as a linear decrease towards the lake inlet, assuming zero water depth at the lake boundary. On August 2024, KTL had a surface area of 0.09 km2, with a maximum depth of 15.76 m and an average depth of ∼9 m (Table 1). The volume of KTL was estimated to be 0.89 ± 0.02 × 10⁶ m3.
Basin morphology of the studied glacial lakes: (a) Kya Tso Lake, (b) Panchi Nala Lake, (c) Gepang Gath Lake, and (d) Samudra Tapu Lake.
Bathymetric map (right panel) and water depth profile (left panel) of the studied lakes: (a) Kya Tso Lake, (b) Panchi Nala Lake, (c) Gepang Gath Lake, and (d) Samudra Tapu Lake. The depth profiles were generated along the central axis of each lake, as indicated by the black lines on the maps.
4.2.2. Panchi Nala Lake
The bathymetric profile of PNL shows a gradual deepening from the moraine-dammed outlet towards the glacier terminus (Fig. 9b). The lake bottom is relatively undulated, characterized by multiple small depressions and sediment mounds, which were also observed during field surveys (see Fig. 3). Along the longitudinal profile, the depth increases from 0.12 m to 2.23 m within the first 100 m from the lake outlet. Between 220 m and 350 m, the depth exhibits a steep increase from 4.23 m to 10.43 m (Fig. 10b). Near the glacier snout (i.e. 350–450 m along the longitudinal profile), the average depth remains ∼10 m (Fig. 10b). This pattern reflects the characteristic up-valley deepening of a glacially eroded basin. On August 2024, PNL had a surface area of 0.11 km2, with a maximum depth of 10.52 m and an average depth of 4.25 m (Table 1). The lake’s volume was estimated at 0.44 ± 0.02 × 10⁶ m3.
4.2.3. Gepang Gath Lake
GGL bed morphology is characterized by a gently sloping (deepening) bed from the lake outlet to the mid-point, followed by a slightly steeper decline towards the lake-glacier boundary (Fig. 9c). The deepest part of the trough is located closer to the calving front. Such up-valley deepening is common in glacial valley systems, as similar has been reported for central Himalayan lakes (Haritashya and others, Reference Haritashya2018). Most of the central section of the lake (800–1500 m) along the longitudinal transect shows depths between 20 m and 40 m, with a steep gradient of >40° (Fig. 10c). Between 1500 m and 2500 m along the longitudinal profile, lake bed is flat, characterised by three small circular troughs with depth >45 m (Fig. 10c). The maximum and average depth of GGL are 46.34 m and 22.46 m, respectively, and it contains 24.26 ± 0.4 × 106 m3 of water as of August 2024 (Table 1).
4.2.4 Samudra Tapu Lake
STL is the biggest and deepest of the four lakes studied here (Fig. 9d). STL’s bottom profile ranges from 0.01 m around the lake outlet to as much as ∼59 m deep near glacier terminus (Figs. 9d and 10d), characterised by heterogeneous mounds and depressions along long profile (Fig. 10d). Going up valley, the lake bottom gradually increases, maximum depth inclined mostly towards left flank (Fig. 9d). Most of the central section of the lake along the longitudinal transect shows depths between 20 m and 30 m, and locally reduces up to 15 m, which reflects a different lake bottom morphology compared to the GGL (Figs. 10c and d). The deepest section of the lake appears as a narrow 180-m long longitudinal trough located ∼100 m down valley from the glacier terminus (Figs. 9d and 10d). The central section of STL does not show a gradual deepening towards the glacier terminus as at GGL. Rather, in our interpolated bathymetric map, the bed appears undulating, characterized by several major trough areas (Fig. 10d), but the deepest points are located near the glacier terminus, like GGL and PNL. STL has a maximum and average depth of 58.65 m and 15 m, respectively, and it contains 24.69 ± 0.4 × 106 m3 of water in 2024, which makes STL both ∼27% deeper and ∼2% volumetrically larger than GGL (Table 1). Interestingly, the average depth of STL is ∼33% less than GGL.
5. Discussion
5.1. Comparison of in-situ volumes with other Himalayan lakes
No prior bathymetry survey of the studied lakes exists. Based on the previous studies (Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024), we classified KTL and PNL as small lakes (size < 0.1 km2), while GGL and STL were categorised as fast-growing large lakes (size > 1 km2) in the western Himalayan region. Considering the size, growing pattern, and associated GLOF potentiality, we compared the depth and volume of GGL and STL with other in-situ bathymetry-derived depths and volumes to understand the regional trend of lake evolution across the Himalayas (Table S1). For comparisons with the present findings, we compiled the depth and volume information of other glacial lakes for which the in-situ bathymetry data are available in the Himalayan region (Fig. 11; Table S1). Bathymetry data are currently available for a small number of glacial lakes in the Himalayas, which are mainly concentrated in central Nepal, Bhutan and several basins of Tibet (Table S1). Based on the bathymetry data from the USV, we obtained the maximum and average depths of GGL of ∼46 and ∼22 m, respectively, in August 2024, with a lake volume of ∼24.26 ± 0.41 × 106 m3. STL showed maximum and average depths of ∼59 and ∼15 m, respectively, in August 2024, and it contains ∼24.69 ± 0.42 × 106 m3. In August 2024, GGL and STL had surface areas of 1.07 km2 and 1.63 km2, respectively, with corresponding lengths of 2.48 km and 2.43 km.
Distribution of (a) area-volume and (b) area-maximum depth of glacial lakes based on in-situ bathymetry across the Himalayas. Selected lakes mentioned in the discussion section are labelled for reference. The lakes investigated in this study are marked with black circles. Detailed information on 61 glacial lakes is provided in Supplementary Table S1. Abbreviations: KTL – Kya Tso Lake; PNL – Panchi Nala Lake; GGL – Gepang Gath Lake; STL – Samudra Tapu Lake.
Based on in-situ bathymetric data, Duan and others (Reference Duan2023) reported maximum and average depths as 181 m and 85.41 m, respectively, with a total lake volume of ∼102 × 10⁶ m3 for Bienong Co (1.15 km2 in size), a moraine-dammed glacial lake in southeastern Tibet in August 2020 (Fig. 11). Similarly, Longbasaba Lake, another moraine-dammed glacial lake located in the headwaters of the Pumqu River on the northern slope of the central-eastern Himalayas, had a surface area of 1.22 km2 in 2009, a maximum depth of 102 m, and a volume of 64 × 10⁶ m3 (Yao and others, Reference Yao, Liu, Sun, Wei and Guo2012). Despite having a comparable surface area, GGL’s volume is significantly lower, ∼76% less than Bienong Co and ∼62% less than Longbasaba Lake. In contrast, STL has a larger surface area, ∼41% and ∼34% greater than Bienong Co and Longbasaba Lake, respectively, yet holds only ∼25% and ∼16% of their total volumes. Lugge Glacial Lake in Bhutan had an area of ∼1.17 km2 in 2004, with a maximum depth of 126 m, and a volume of 58 × 106 m3 (Yamada, Reference Yamada2004). The area of Lugge Glacial Lake is lower than that of STL, but the maximum depths and volume are much higher than the corresponding values of STL. At the time of the bathymetry survey, South Lhonak Lake in India and Imja Lake in Nepal had a similar surface area of ∼1.3 km2 (Haritashya and others, Reference Haritashya2018; Sharma and others, Reference Sharma, Pradhan, Sharma and Shrestha2018), which is ∼20% larger than GGL and ∼25% smaller than STL. In 2016, South Lhonak Lake had a maximum depth of 131 m and a total volume of 66 × 10⁶ m3 (Fig. 11). In 2014, Imja Lake had a maximum depth of 150 m and a volume of 78 × 10⁶ m3 (Fig. 11). Our analysis revealed that the volumes of both GGL and STL are ∼63% and ∼68% smaller than those of South Lhonak and Imja Lakes, respectively. The areas of Raphsthren Glacial Lake in Bhutan and Tsho Rolpa Glacial Lake in Nepal were 1.4 km2 and 1.5 km2, respectively (Fig. 11; Table S1), at the time their bathymetries were conducted (Geological Survey of India, 1995; ICIMOD, 2011). These areas are 14% and 8% smaller than that of STL. However, their volumes are significantly larger, being 171% and 252% higher than that of STL, respectively. In 2015, Lower Barun Lake had a surface area of 1.8 km2, a maximum depth of 205 m and a volume of 112 × 10⁶ m3 (Fig. 11). Despite having only 9% more surface area than STL, Lower Barun Lake is 247% deeper and 353% larger in volume than STL. Several other lakes across the Himalayas show heterogeneous depth and volumetric storage, despite having comparable surface area (Fig. 11; Table S1).
Heterogeneity in volumetric storage could be attributed to variable depth, lake basin morphology, overdeepening of glacial beds, moraine height and sediment deposition patterns. Glaciation style is a key factor driving heterogeneous lake evolution in alpine environments. For instance, large lakes associated with long valley glaciers often exhibit greater depths compared to smaller lakes linked to steeper, shorter glaciers (Richardson and Reynolds, Reference Richardson and Reynolds2000). Glacial lakes typically form in irregular depressions (i.e. overdeepenings) carved by glacial activity (Haeberli and others, Reference Haeberli, Linsbauer, Cochachin, Salazar and Fischer2016). Variations in basin shape, slope and depth result in volume heterogeneity, even among lakes with comparable surface areas (Carrivick and Tweed, Reference Carrivick and Tweed2013; Linsbauer and others, Reference Linsbauer, Frey, Haeberli, Machguth, Azam and Allen2016). Haritashya and others (Reference Haritashya2018) reported similar observations of four glacial lakes in the Nepal Himalayas. Zhang and others (Reference Zhang2023c) reported variability in the volumetric assessment of 16 pro-glacial lakes in the eastern Himalayas. Subglacial drainage channels, buried ice, ice-ramps near the glacier-lake boundary (Haeberli and others, Reference Haeberli, Linsbauer, Cochachin, Salazar and Fischer2016) or sediment layers can further alter a lake’s effective water-holding capacity (Cook and Quincey, Reference Cook and Quincey2015; Steffen and others, Reference Steffen, Huss, Estermann, Hodel and Farinotti2022). Additionally, lakes receiving high sediment loads (from erosion, avalanches or glacial flour) may experience reduced volumes due to infilling (Hardmeier and others, Reference Hardmeier, Schmidheiny, Suremann, Lüthi and Vieli2024), despite stable surface areas. For example, the highly turbid waters of STL indicate significant sediment production, likely contributing to bed deposition and infilling, in contrast to the relatively sediment-free GGL.
5.2. Comparison of volumes between in-situ bathymetry and empirical volume-area estimates
Due to the scarcity of glacial lake bathymetry data and its critical importance for glacial lake hazard assessments, researchers have developed VA empirical relationships to estimate lake volume using parameters such as surface area, width and length (Fig. 12 and references therein). In this study, we applied 20 published VA empirical equations—derived from datasets spanning global and Himalayan regions—to estimate the volumes of the studied glacial lakes (Fig. 12). These equations were previously compiled based on lake morphological types, with varying sample sizes used in their derivation (Table S2).
The percentage difference between in-situ and modelled volume estimates for the studied glacial lakes, based on 20 commonly used empirical equations. Details of each equation are provided in Supplementary Table S2. References for equations are Eq. 1: Huggel and others (Reference Huggel, Kääb, Haeberli, Teysseire and Paul2002); Eq. 2: Loriaux and Casassa (Reference Loriaux and Casassa2013); Eq. 3: Cook and Quincey (Reference Cook and Quincey2015); Eq. 4: Watson and others (Reference Watson, Quincey, Carrivick, Smith, Rowan and Richardson2018); Eq. 5: Evans and others (Reference Evans1986); Eq. 6: O’Connor and others (Reference O’Connor, Hardison and Costa2001); Eq. 7: Emmer and Vilímek (Reference Emmer and Vilímek2014); Eq. 8, 9, and 10: Wood and others (Reference Wood2021); Eq. 11: Kapitsa and others (Reference Kapitsa, Shahgedanova, Machguth, Severskiy and Medeu2017); Eq. 12: Sakai and others (Reference Sakai2012); Eq. 13: Wang and others (Reference Wang2012); Eq. 14: Fujita and others (Reference Fujita2013); Eq. 15: Khanal and others (Reference Khanal, Hu and Mool2015); Eq. 16: Sharma and others (Reference Sharma, Pradhan, Sharma and Shrestha2018); Eq. 17: Patel and others (Reference Patel, Sharma, Laluraj, Thamban, Singh and Ravindra2017); Eq. 18: Miles and others (Reference Miles2018); Eq. 19: Watson and others (Reference Watson, Quincey, Carrivick, Smith, Rowan and Richardson2018); Eq. 20: Zhang and others (Reference Zhang2023c). Note: Data for Samudra Tapu Lake using Eq. 6 are excluded due to an unusually high deviation of 1736%.
The results indicate that none of the empirical equations agreed with in situ bathymetry-derived volumes (Fig. 12; Table S2). Substantial discrepancies (±32–1736%) were observed between volumes estimated using existing equations and those measured via in-situ bathymetry across all four lakes. For example, applying the formula proposed by Huggel and others (Huggel and others, Reference Huggel, Kääb, Haeberli, Teysseire and Paul2002), deviations of 47%, 240%, 57% and 180% were observed for KTL, PNL, GGL and STL, respectively (Eq. 1 in Fig. 12). Similarly, the equation by Cook and Quincey (Reference Cook and Quincey2015) yielded volumetric deviations ranging from 54% for GGL to 276% for STL (Eq. 3 in Fig. 12). Notably, all equations except that of Wang and others (Reference Wang2012) produced positive deviations (overestimates) in volumetric measurements (Eq. 13 in Fig. 12). Patel and others (Reference Patel, Sharma, Laluraj, Thamban, Singh and Ravindra2017), using their empirical equation (Eq. 17 in Fig. 12), reported volumes of ∼27 × 10⁶ m3 and ∼68 × 10⁶ m3 for GGL and STL, respectively, as of 2014. However, our in-situ measurements revealed deviations of ∼112% for GGL and 336% for STL (Table S2).
Several studies have developed VA empirical relationships by adding data from specific regions (Wang and others, Reference Wang2012; Loriaux and Casassa, Reference Loriaux and Casassa2013; Cook and Quincey, Reference Cook and Quincey2015; Zhang and others, Reference Zhang2023c), suggesting that regional factors may influence the predictability of lake volume. However, based on the observations from the Himalayan regions, Qi and others (Reference Qi2022) reported that regional differences do not consistently improve volume predictability. Cook and Quincey (Reference Cook and Quincey2015) critically evaluated the performance of existing empirical relationships across different regions and highlighted that even lakes within the same geographic area exhibit significant variability in volume predictability. Their findings suggest that region-specific equations are not inherently more reliable than those derived from broader datasets. Instead, lake origin and morphology play a dominant role in determining volume predictability and the suitability of empirical equations for accurate forecasting (Cook and Quincey, Reference Cook and Quincey2015). For instance, Zhang and others (Reference Zhang2023c) developed a Himalayan-specific equation using bathymetric data from 59 moraine-dammed lakes. Yet, our in-situ volume measurements for moraine-dammed GGL and STL reveal substantial deviations (96–247%) from their estimates (Fig. 12; Table S2).
For water volume estimation, incorrect values could provide an overestimate of potential water storage, while for risk assessment purposes, underestimates of lake volume and consequently, the possible peak discharge could be potentially dangerous (Muñoz and others, Reference Muñoz, Huggel, Frey, Cochachin and Haeberli2020; Qi and others, Reference Qi2022). Since the different empirical formulas have various estimate errors for individual lakes, several studies (Cook and Quincey, Reference Cook and Quincey2015; Haeberli and others, Reference Haeberli, Linsbauer, Cochachin, Salazar and Fischer2016; Muñoz and others, Reference Muñoz, Huggel, Frey, Cochachin and Haeberli2020) have recommended the use of different approaches for lake volume estimation rather than relying on a single method.
Our study area, the upper Chenab basin, contains several potentially dangerous glacial lakes and is a GLOF-prone basin in the western Himalayas (Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Das and others, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). Previous studies have focused on glacial lake area change, volume estimation using empirical equations, and risk assessment of potentially dangerous glacial lakes in the basin (Patel and others, Reference Patel, Sharma, Laluraj, Thamban, Singh and Ravindra2017; Prakash and Nagarajan, Reference Prakash and Nagarajan2017; Kumar and others, Reference Kumar, Mehta and Shukla2021; Das and others, Reference Das, Sharma, Murari, Nüsser and Schmidt2023, Reference Das, Das, Mandal, Sharma and Ramsankaran2024). Due to numerous gaps in current knowledge regarding the volumes of glacial lakes in the Chenab basin, the simulation of GLOF development pathways is difficult to estimate precisely. Hence, the present study provides new datasets on in-situ bathymetry, which could be further used in future to develop a more robust VA equation for the region. Future efforts are necessary to continue and extend glacial lake bathymetric measurement programs for further volume assessment and model calibration.
5.3. Challenges and recommendations
We faced several challenges during the USV surveys. Previous studies (Haritashya and others, Reference Haritashya2018; Muñoz and others, Reference Muñoz, Huggel, Frey, Cochachin and Haeberli2020; Li and others, Reference Li2021; Qi and others, Reference Qi2022; Duan and others, Reference Duan2023; Zhang and others, Reference Zhang2023c) have not detailed the field difficulties associated with bathymetric surveys in high-altitude regions. However, Ramsankaran and others (Reference Ramsankaran, Verma, Majeed and Rashid2023) highlighted the procedures, challenges and recommendations for conducting kayak-based surveys in Himalayan glacial lakes. Building on standard procedures, our USV-based survey methodology offers a scalable and replicable approach for bathymetric mapping in comparable glacial lake environments. Here, we outline key challenges and offer practical recommendations to enhance the efficiency and reliability of USV surveys in high-altitude glacial lake settings.
I. For this study, we deployed a Satlab Hydroboat 1100 USV equipped with a 200 kHz single-beam echo sounder with a maximum penetration capacity of 200 m. While this system is effective for lower depths (<20 m), acoustic noise emerged in higher depths (>20 m) at constant speeds due to delayed echo returns (Fig. 7c). Such challenges highlight the need for prior knowledge of tentative maximum lake depths to select appropriate transducer frequencies. Lower frequencies (i.e., 50 kHz) improve penetration in deeper areas (>10 m) but reduce resolution in shallower zones (<10 m), suggesting for use of a dual-frequency transducer (e.g., low frequency for deep areas and high frequency for shallow areas) to balance depth penetration and resolution.
II. The limited telemetry range (∼1.5 km) of Satlab 1100 USV restricts real-time communication and data transmission between the boat and control system during the survey of long lakes (i.e. GGL, STL; ∼2.7 km length). Several base stations are required at different locations along the lake edges, which is difficult due to steep terrain and inaccessibility. To overcome this limitation, a USV with a higher telemetry range (>5 km) is recommended for surveying large lakes (such as Imja: ∼3.1 km, Lower Barun: ∼2.7 km, Thulagi: ∼3 km, Tso Rolpa: ∼3.8 km, South Lhonak: ∼2.7 km) in the Himalayas.
III. High-altitude conditions significantly reduce the USV’s battery performance, limiting operational time. In our surveys, a single battery charge provided ∼3 hours of runtime—insufficient to fully cover large lakes like GGL and STL, given their size and required survey line density. To address this, we recommend carrying additional battery sets to ensure uninterrupted operations. Moreover, transporting the USV and associated equipment to remote high-altitude lake sites presents major logistical challenges. The USV’s large dimensions and heavy protective cases demand considerable manpower and extended time for transport. Ensuring continuous power supply further complicates logistics, as it requires carrying a portable generator—often bulky and difficult to handle in rugged terrain. These constraints necessitate more porters and additional days in the field. To improve feasibility and efficiency in such environments, we strongly recommend using more compact, lightweight USV systems with enhanced battery capacity, which would ease transportation and reduce logistical demands.
IV. The strong winds in the glacial valley during afternoons made surveying much more difficult since these winds helped build strong water currents that caused USV to bounce during the survey. Studies reported that waves generated bouncing of USV lead to air pockets between the boat and water that scatter acoustic signals and introduce noise in echograms (Shabangu and others, Reference Shabangu, Ona and Yemane2014). It is suggested to perform the survey during the first half of the day, especially during morning hours when calm weather persists. It is because, in glaciated regions, the wind speed will be relatively higher during the afternoons, resulting in more bounce of USV, even at slow speeds.
V. The relationship between USV speed and depth of water is one of the important factors that need to be kept in mind during the survey. Using a single-beam high-frequency sounder, one needs to adjust the boat speed with varying depths. Constant boat speed during auto mode can lead to motion-induced errors and noise in echo signals, especially in deeper areas where return echoes take longer to reach the sensor. We observed similar noise in echograms during the survey at GGL, depth >30 m. It is suggested to adjust boat speed dynamically based on depth: slower speeds (<1 m s−1) for deeper areas and faster speeds for shallower areas. In addition, a USV equipped with adaptive speed control algorithms is recommended to optimize the survey.
VI. Among the four lakes surveyed, STL exhibited the highest turbidity (see Fig. 5). Surveying in turbid lakes can produce additional acoustic noise due to suspended sediments scattering the echo-sounder signals (Santos and others, Reference Santos, Oliveira, Carinhas, Pinto, Zacarias and Freitas2020). While we could not isolate the specific contribution of turbidity to signal scattering, any noise artefacts observed in the raw data were successfully removed during post-processing of the bathymetry datasets.
VII. During the survey, we encountered operational challenges due to large floating ice chunks in the lakes, particularly in GGL, where they were more prevalent than in the other three lakes. These ice chunks hindered the auto-mode survey plan by obstructing predefined paths, destabilizing the USV and posing risks of equipment damage. Although the USV was equipped with a 360-degree omnidirectional camera for obstacle monitoring, this proved insufficient for navigating ice-dense zones. To address this, we implemented a hybrid approach: the USV was tethered to a kayak, which was manually manoeuvred through ice-dominated areas. This strategy enabled access to otherwise inaccessible regions and conserved battery life by deactivating the USV’s propellers during kayak-led navigation. For future surveys in high-altitude glacial lakes, where ice calving is a common phenomenon, we recommend USVs equipped with advanced obstacle detection and avoidance systems to enhance safety and efficiency.
High-altitude glacial lakes present unique challenges for USV-based bathymetric surveys, including environmental factors (wind, ice, turbidity) and operational limitations (battery life, telemetry range, transducer performance). Addressing these challenges requires a combination of advanced technology, adaptive survey strategies and careful planning. By implementing the recommendations above, one can improve the efficiency, accuracy and safety of USV surveys in these challenging environments.
6. Conclusions
In this study, we have modelled the basin morphology of four lakes in the western Himalayas based on an in-situ bathymetry survey. The bathymetric survey revealed significant variations in lake depths and storage capacities, with maximum depths of 16 m, 10 m, 46 m and 59 m for Kya Tso Lake, Panchi Nala Lake, Gepang Gath Lake and Samudra Tapu Lake, respectively. Their corresponding volumes were estimated at 0.89 × 10⁶ m3, 0.44 × 10⁶ m3, 24.12 × 10⁶ m3 and 24.69 × 10⁶ m3. A comparison with volume estimates from commonly used empirical equations revealed substantial discrepancies (±36–1736%), highlighting the limitations of these models in accurately predicting glacial lake volumes. The findings highlight the importance of conducting targeted in-situ bathymetric surveys, particularly for potentially dangerous glacial lakes in the region, to support more accurate hazard assessments based on lake-specific geomorphological characteristics. Future studies should focus on integrating high-resolution USV-based bathymetric data with advanced hydrodynamic modelling to improve risk assessment and mitigation strategies for GLOFs in the region.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/jog.2025.10082.
Data availability statement
The in-situ bathymetry and multi-temporal lake boundary datasets can be downloaded from Zenodo at https://doi.org/10.5281/zenodo.16677319.
Acknowledgements
We would like to thank the porters for their invaluable assistance during the fieldwork. We sincerely appreciate the technical support for the USV provided by Satlab and Hi-Target. Thanks to Dr Parmanand Sharma from the National Centre for Polar and Ocean Research (NCPOR) for supporting us during the USV survey at Kya Tso and Samudra Tapu Lakes. The first author extends gratitude to Luvkesh Attri, PhD scholar at the Hydro-Remote Sensing Applications (H-RSA) group, IIT Bombay, for support during the USV survey. The authors thank Sayantan Mandal, Ajay Godara, and Navin Kumar, PhD scholars at H-RSA group, IIT Bombay, for their support during the Kayak-based survey at Panchi Nala Lake. This work was supported by the institute postdoctoral program of the Indian Institute of Technology Bombay, Mumbai and by the Science and Engineering Research Board (SERB), Department of Science and Technology (DST), Government of India for the project titled ‘Modelling the evolution of large glacier-fed lakes in Western Himalayas’ under Core Research Grant (CRG). We thank the Chief Editor, Dr Hester Jiskoot, the Scientific Editor, Dr David Rounce, the Associate Chief Editor, Dr Nicolas Eckert, and the two anonymous reviewers for their constructive and insightful comments, which have greatly improved the quality of the manuscript.
Author contributions
Suresh Das conducted USV surveys at KTL, GGL and STL, analysed depth data, created figures and drafted the original manuscript. RAAJ Ramsankaran performed a kayak-based survey at PNL, supervised the study, acquired funding and contributed to writing and reviewing the manuscript.
Competing interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Open access
