UAV-based Survey of Glaciers in Himalayas: Opportunities and Challenges

Challenges RAAJ Ramsankaran, P.J. Navinkumar, Ajay Dashora, A.V. Kulkarni 1 Hydro-Remote Sensing Applications (H-RSA) Group, Department of Civil Engineering, Indian Institute of Technology, Bombay, Powai, Mumbai-400076, India 2 Earth System Science and Engineering, Department of Civil Engineering, Indian Institute of Technology Guwahati, Guwahati, Assam-781039, India 3 Divecha Centre for Climate Change, Indian Institute of Science, Bengaluru, Karnataka560012, India Abstract


Introduction
Glaciers are perennial features that temporarily store freshwater and they comprise only a fraction of the Cryosphere region. Globally, glaciers are losing ice mass in the early twentyfirst century twice that of the end of previous century [1]. In such scenario, continuous monitoring of glaciers at local and regional levels is essential. On a global scale, numerous efforts are being made to observe and monitor glacier health using ground-based, aerial and spaceborne remote sensing platforms, e.g., [2,3,4,5,6,7].
Despite such efforts, only sparse and limited data from ground-based observations are available for reference or validation purposes. Particularly, the Himalayan region in Southwest Asia has significant gaps in ground-based observations [8]. Remoteness and rugged terrains restrict ground-based observations, which are time-consuming and logistically challenging [9].
Though aerial surveys can cover glaciers at a regional scale at required place and time, harsh weather conditions on high altitudes raise safety concerns. Moreover, repetitive aerial surveys are not feasible for glaciological studies due to high costs. Likewise, existing space borne remote sensing platforms have many limitations for collecting individual glacier level accurate data at high resolutions despite having numerous advantages for global scale glacier monitoring [6]. At this juncture, Unmanned Aerial Vehicles (UAV) as remote sensing platform, which can fly below the clouds has enormous potential to augment the sparse and discontinuous field observations by providing ultra-high-resolution images at relatively low costs.
The technological advances in sensors with navigation modules have increased the potential of UAV as a platform for remote sensing [10]. This new potential has led to new facets in acquiring UAV-based remote sensing data such as the ability to acquire ultra-high-resolution, availability of comprehensive spectral as well as geometric data, and fusion of multi-sensor data [11]. Especially, the remarkable development of UAVs with onboard Global Navigation Satellite System/Global Positioning System (GNSS/GPS) receiver; Inertial Measurement Unit (IMU); high endurance capabilities, and 3D model generation by Structure-from-Motion (SfM) process has paved excellent opportunities for deploying UAV-based RS platform in glacier studies [12].
Although the UAV based RS technology is gaining popularity among researchers, certain hurdles such as limited accessibility of alpine glaciers and the challenges related to flying in high altitudes are hindering the progress of UAVs in glacier research, especially in high mountain alpine regions such as the Himalayas. This is evident from the geographical distribution of available studies, where around 90% of the UAV-based glaciology applications are focussed on polar and sub-polar regions, while few studies, i.e. 10% are only focussed on the Alpine regions [12].
Despite such challenges, in recent years the application of UAVs in Alpine regions (i.e. the Andes, the Alps, the Himalayas) are gaining momentum although still comparatively lower than the Polar and Sub-polar regions [13,14,15,16,17,18,19,20]. Till date, studies demonstrating the UAV applications in Himalayas remain scarce and are focussed only on three glaciers namely, Lirung; Changri Nup and Thulagi glacier areas in Nepal [21,22,23,24,25,26]. It is because the terrain of the glaciers in Himalayas are hostile, highly debris-covered, located at high altittudes (the average elevation of the Himalayas is significantly higher than the Alps and the Andes mountain ranges) making logistics and movements much more difficult. Morerover, applying UAV photogrammetry to the Himalayan glaciers poses additional challenges such as low air pressure, poor GNSS receptions for UAVs with automatic navigations.
As stated in Groos et.al. [27], except few e.g., [28,29], most of the glaciological studies use Commercial off-the-shelf (COTS) UAVs to acquire aerial images in a high spatial resolution e.g., [20,30,31]. The benefit of commercial UAVs is obvious: they are reliable and ready-touse. It is also observed that most of the UAV-based glaciological studies e.g., [21,28,32,33,34] use fixed-wing UAVs rather than rotary-wing UAVs. This is mainly due to the advantages of fixed-wing UAVs in providing better aerodynamics, longer flight duration, cover comparatively larger area per flight and fly higher than rotary-wing [12]. However, great care is required in identifying proper take-off and landing sites in rugged mountainous terrain to prevent physical damages to the fixed-wing UAVs.
On the other hand, to obtain accurate glacier topography, UAV based SfM methods rely on Ground Control Points (GCPs). Recent study by Gindraux et al. [32] reported that on all seasons 17(7) GCPs per km 2 are required to achieve Digital Surface Model (DSM) with vertical (horizontal) accuracy of range 0.1-0.25m (0.03-0.09m). Follwing this, other glacier studies have also used GCPs (varying between seven and seventeen) to generate DEMs between centimetre and decimetre level accuracies e.g., [15,27,31]. However, the topography of most of the glaciers in Himalayan region is complex where collecting minimum GCPs are impractical due to safety reasons.
Despite the increasingly common use of UAVs in glaciology, there are several technical and practical challenges that have yet to be completely overcome when operating UAVs in extreme, high mountain regions. Therefore, studies addressing the practical challenges of UAV surveys in inaccessible terrain and harsh meteorological conditions like Himalayas are very much essential. Towards filling this gap, this article collates the authors' experiences gained from UAV-based surveys carried out at glaciers located in different river basins in Indian Himalayas.
This study used COTS fixed-wing UAV (Sensefly eBee plus) mounted with digital RGB camera to acquire images on the ablation regions of the study glaciers. The study aims to, • Examine the practical challenges faced during UAV surveys of glaciers and their impacts on the UAV data products.
• Generate Digital Elevation Models (DEMs) from UAV collected images and assess the elevation accuracy and • Provide possible strategies and suggestions to identify take-off/landing locations for conducting efficient UAV surveys using COTS fixed-wing UAVs.

Study Sites
UAV surveys were conducted on three glaciers, namely East Rathong, Hamtah and Panchinala-A in Indian Himalayan region. Details of the location, areal extent, elevation, and debris cover of these glaciers are given in Table 1. Figure 1 shows the locations of the study glaciers and the area covered by the UAV surveys.  East Rathong is a South-East facing summer nourished valley glacier [35] and it is the only benchmark glacier in the Eastern Indian Himalayan region, which is being monitored regularly from 2013 [36]. This glacier is covered by debris, i.e. 15% of total glacier area and surrounded by steep lateral walls at the lower ablation regions. It originates at ~6800 m above sea level (a.s.l.) and terminates at ~4720 m a.s.l.
Hamtah glacier is a North facing valley glacier fed by the mid-latitude westerlies. This glacier is one of the benchmark glaciers in Western Indian Himalayas being monitored since the early 2000s [37]. This glacier is mostly covered by debris, i.e. 73% of total glacier area and also surrounded by steep lateral walls. The glacier is ~6 km long with an average width of 0.5 km and lies within the elevation range of 4000 -5000 m a.s.l.
Following [38], the glacier in Figure 1b is named as Panchinala-A. The Panchinala-A glacier is North facing valley glacier which also lies in the Western Indian Himalayan region nourished by mid-latitude westerlies. The glacier is covered by debris i.e. 29% of total glacier area and has less steep walls than that of the other study glaciers. The glacier is ~5 km long with an average width of ~0.55 km and lies within the elevation range of 4300 -5990 m a.s.l.

Methods
The study consists of three stages: data acquisition, data processing, and preparation of data products. Figure 2 illustrates the methods adopted in these three stages. First, data acquisition activities by UAV and the Differential Global Positioning System (DGPS) are discussed for all the three study glaciers. Following this, the processing of collected UAV photos and GCPs through DGPS observations are discussed. Finally, the observations inferred from the UAV extracted data products such as Ortho-mosaicked images, DEMs and their slope characteristics are reported and discussed.

Figure 2:
Block diagram of the processes adopted for UAV in different stages, data acquisition, data processing and data products generation.

UAV Setup and Specifications
To conduct the UAV survey, a commercial-grade off-the-shelf fixed-wing UAV (eBee plus, and their specifications along with the software used for flight planning and data processing are given in Table 2. Table 2: Specifications of the UAV setup and software used in the study

DGPS Survey
Prior to UAV survey, DGPS surveys were conducted on all three study glaciers to georeference and validate the ortho-mosaicked images and DEMs ( Table 3). The DGPS setup included one rover and base station (Trimble Zephyr antennas) with two handheld controllers (GeoExplorer GeoXH handheld). The field photos taken on surveyed sites are shown in Figure 3. Post-Processing UAV collected photos, Orthomosaicked image and DEM generation (dimension 1m × 1m) were made of synthetic clothes with the yellow-black and yellow-red combinations. GCPs were collected on the central point of the artificial targets.  Rover setup (g) for GCP collection at Panchinala-A glacier respectively.

UAV Flight Plan Design and Survey
The UAV survey flight missions were planned and executed with the eMotion version 3 software package. The communication is established between the flight management software and the UAV via a radio frequency (RF) modem. The setup, i.e. laptop with flight management software (eMotion v3) and RF modem, tends to act as ground control station (GCS). With the eMotion software, flight plans are generated as mission blocks with the flight parameters (ceiling height, the radius of UAV coverage area, UAV take-off and landing, and mission block generation) and SRTM DEM of a given terrain. Table 4 shows the data acquisition characteristics of the UAV mission plan adopted for the selected glacier sites. The fixed-wing UAV takes-off with a motion sensor technology, i.e. by moving UAV back and forth three to four times to launch the UAV. Minimum 2ms -1 wind speed is required to fly the fixed-wing UAV. However, on all the surveyed sites, an average wind speed was observed to vary between 4ms -1 and 6ms -1 at the time of UAV flights. All the flight missions were pre-set and automated within eMotion software. At East Rathong, UAV survey was conducted in two locations, viz,  Mode of landing Linear Linear Linear

UAV Ortho-mosaicked Images and DEM Extraction
From the acquired UAV geotagged images, data products such as ortho-mosaicked images and In the initial processing stage, the image scale size was set to the UAV acquired image scale; key point extraction was set to an automatic mode where the images are matched to compute key points, and then automatic tie points are built and analysed. Following this, standard calibration mode was set to optimise the camera parameters.
To improve accuracy, the georeferencing process was carried out by accurately identifying the central points of the collected GCPs targets on three surveyed sites. Following Rossini et al. [31], among the GCPs available within the UAV surveyed site, two-third of them were used for georeferencing and one-third for validation. In East Rathong site, out of 6 GCPs, 5 GCPs were collected at ER2 and one GCP at ER1. Due to insufficient GCPs at ER1, georeferencing and validation was not possible. However, at ER2 the georeferencing was done with all 5 collected GCPs sparing none for validation. It is because the GCPs were confined in a small area and hence, validating in such scenarios is not appropriate. At Hamtah site, out of 5 GCPs, 3(2) GCPs were used for georeferencing (validation). Likewise, in Panchinala-A, out of 9 GCPs, 6(3) GCPs were used for georeferencing (validation).
In the next stage to generate the dense point cloud and mesh, image scale size was set to half of the image size and default optimal level option was set to generate dense point clouds. From the generated dense point cloud and mesh, ortho-mosaicked images and DEMs were reconstructed. Apart from the georeferencing and validation with GCPs, the same procedure was followed to generated ortho-mosaicked images and DEMs for all the three glacier sites.

Results
At East Rathong site, with two flights the fixed-wing UAV (eBee plus) was able to cover 0.78 The UAV coverage area and their flight-related aspects for the selected three glacier sites are shown in Table 5. and ER2, fifteen GCPs on the Hamtah and nine GCPs on the Panchinala-A sites (see Table 2).
The collected GCPs were then post-processed using Trimble Path Finder Office software package. The positional accuracy of the base corrected GCPs in the East Rathong region has 96% of the processed sample points ranged within 5-15 cm, for Hamtah region 82% of processed sample points ranged within 5-50 cm, and 95% of sample points ranged within 5-15 cm for Panchinala-A region.

Accuracy of UAV-derived DEMs
To ascertain the effects of GCPs on UAV data products accuracy, UAV-derived DEMs were generated with and without GCPs for all the three surveyed sites. Care was taken that DGPS  (Table 6).  Figure 4a) and artificial GCP targets (Figures 4b and 4c) on the three study glacier sites are shown in Figure 4.

Topography Maps
For the UAV surveyed area at three study sites, DEMs and their respective slope maps were

Glacier Surface Feature Maps
Some prominent glacier surface features that are visible in UAV-derived ortho-mosaicked images are visually interpreted to generate glacier surface features maps for each surveyed glaciers.
East Rathong: Figure 6 shows the East Rathong glacier area (ER1) of UAV-derived orthomosaicked image. Here glacier ice is found to be exposed at the middle portion of the surveyed glacier area. Glacier landforms such as crevasse (Figure 6a) and ridges are observed along the slightly curved glacier trunk, where the glacier ice is exposed. Meltwater streams are found to be moving from higher elevations towards the lower elevations (Figure 6b). Supraglacial ponds/lakes partially surrounded by ice-cliffs are observed at the debris-area located lower elevations glacier surveyed area (Figures 6c and 6d). Non-linear fractures were dominant on the exposed glacier ice and at some parts of debris-area at higher elevations (Figure 6e). These meltwater streams end up at glacier infills in between and continue flowing in the subsurface towards the glacier snout direction. The glacier ice is less exposed and mostly covered by supraglacial debris at lateral sides and bottom of the surveyed area. Most of the ice-cliffs are found to be covered by debris (Figure 6e). By visually identifying these features in the orthomosaicked images, glacier surface feature map is then generated (Figure 7).  Hamtah: Figure 8 shows the Hamtah glacier area of UAV-derived ortho-mosaicked image.
Two glacier features such as supraglacial debris and ice-cliffs (Figures 8a-8d) are identified.
The glacier surface is found to have more undulations due to heavy debris (Figure 8c). Figure   8b shows the steep walls adjoining the lateral part of the glacier, where the rockfall event has occurred previously. Ice-cliffs are observed near the snout region and at the higher elevations of the surveyed glacier site (Figures 8a and 8d). No supraglacial ponds are found within the surveyed area which indicates that the velocity of the glacier may be active in these areas. By visually identifying these features in the ortho-mosaicked images, glacier surface feature map is then generated (Figure 9).  Panchinala-A: Figure 10 shows the Panchinala-A glacier area of UAV-derived orthomosaicked image. Glacier features like ice-cliffs (Figures 10a-10d), supraglacial lakes/ponds, snow-cover and supraglacial debris are identified. The supraglacial lakes/ponds partially covered by snow/ice are found to be surrounded by ice-cliffs (Figures 10b and 10c). The presence of supraglacial lakes/ponds within the surveyed area of the glacier surface shows that the glacier is stagnant or moving very slow. Supraglacial debris is covered by snow in most of the glacier surface. Compared to other surveyed glaciers, few numbers of supraglacial lakes/ponds are identified in UAV surveyed area of the Panchinala-A glacier. By visually identifying these features in the ortho-mosaicked images, glacier surface feature map is generated ( Figure 11).

Challenges Faced during the UAV Data Acquisition
The challenges experienced during UAV data acquisition over the three glacier sites are studied and reported here. Rathong. UAV take-off location was chosen outside the glacier region, as there was no appropriate place found for UAV take-off and landing around the glacier terminus region.
During the flight, the link between UAV and GCS was lost for about twenty to thirty seconds because of steep wall glacier valleys and hence, UAV's mission was called off. After several attempts, UAV took off again but still received only three to four GNSS satellites throughout the survey. Unexpectedly, the sudden change in weather followed by heavy snowfall in subsequent days led us to abort the planned missions. During the DGPS survey, 15 GCPs were collected (see Figure 4b), but only five GCPs were lying inside the UAV surveyed glacier site.
Unlike natural targets used in the East Rathong glacier, here artificial targets (see Figure 4b) were used for collecting GCPs. The steep-walled glacier valleys obstructed the UAV survey and led to cover a smaller surface area and also affected the accuracy of the UAV data products see Table 4.

Panchinala-A: Here the glacier valley's terrain topographic condition is better than East
Rathong and Hamtah glaciers. Therefore, UAV's GNSS module was able to acquire 9-12 satellite signals. The remaining snow-free area was covered by debris and boulders.
Considering the safety of the UAV, snow covered location was chosen for landing. However, the ground sensor module of UAV identified snow as an obstruction and took additional time to hover around and then finally landed on the heavy debris surface away from the fixed location. As a result, UAV faced major damages (i.e. servo connection mechanism which connects aileron of wings with the central body of UAV). Therefore, to avoid such damages it is recommended to turn-off the UAV's ground sensor at the time of landing.
Based on these observations at three glacier sites, it can be said that the challenges faced during UAV data acquisition on glacier sites vary due to i) nature of terrain and ii) choice of UAV take-off/landing locations. Therefore, the study strongly suggests to select favourable take-off and landing locations while conducting fixed-wing UAV surveys on high mountain glaciers. Table 7 summarizes the challenges faced by fixed-wing UAV (eBee plus) during their data acquisition on three study glacier sites. Snow cover (at landing location) was identified as obstruction by UAV's ground sensor.
Delayed UAV landing and major damages to servo connection mechanism (connects aileron with the central body of UAV). At the same time, it is observed that a single flight above 4000 m.a.s.l. may take ~ 20-40 minutes to cover 1.00 -1.50 km 2 area with an average flying altitude ~175 m above the ground level (provided clear sky conditions, wind speed between 2 ms -1 and 10 ms -1 , good GNSS satellite signal strength with low Dilution of Precision (DOP)). Accordingly, the study assumes that in glaciers located at such high altitudes having low temperatures, thin air density and relatively high wind speeds, 5-8 km 2 of glacier area can be covered using fixed-wing UAVs (such as eBee series) in a single day during ablation season.

Recommended UAV Take-off/Landing Locations
Based on these observations, the study has identified three potential sites favourable for UAV take-off and landing on Himalayan valley-type glaciers.

Top of the Glacier Valley:
In valley glaciers, the study recommends to conduct the UAV survey from the valley top adjacent to the glacier rather than the glacier surface region.
Surveying from the glacier surface, UAV may use around 15% -30% of the battery power only to reach the desired altitude and landing after the end of the survey. When the UAV survey is conducted from the flatter/low relief regions (if available) of the valley top adjacent to the glacier, battery power used by the UAV and its flight time to reach the maximum altitude and landing can be significantly reduced. As a result, more time can be allocated to cover additional glacier area with the same flight conditions. Moreover, UAVs tend to have better visibility of the glacier, and it can acquire good satellite coverage with low DOP. In such cases, one can reduce the number of flight attempts and minimise the risk of damages. However, to identify such locations, the guidance of field experts is essential.
Near the Equilibrium Line Altitude (ELA): When no such locations as recommended above are identified on the valley top adjacent to the glacier, the area near the ELA region in the glacier should be considered. Usually, the ELA region in Himalayan glaciers has wide crosssections and exposed with hard ice without any debris cover. Hence, the landing of a UAV (fixed wing) on hard ice would have minimum impact on UAV than the debris area. Moreover, the chances for maximum GNSS satellite signal receptions and the probability of covering a larger area are also better, i.e. UAV flights can cover both the accumulation and ablation regions from the same take-off and landing location.
Near the Terminus and Ablation Region: When the above two regions are not possible to occupy, then adjacent regions of ablation zones, glacier terminus, and nearby downstream areas should be considered. Generally, the terminus of the valley glaciers has narrow cross-sectional width and more terrain undulations than the ELA regions, which limits the coverage area of UAV survey to the ablation region and has lesser probability of covering an entire glacier in a single flight if the glacier is >1km 2 . However, fixed-wing UAVs will be able to cover an entire glacier in a single flight under ideal flying conditions, if the glacier's, i) area less than 1 km 2 , ii) elevation range is within 1km and iii) terrain has low relief valley similar to Panchinala-A glacier.
By following the above recommendations on mountain glaciers, it is expected that an UAV can cover a large area with less number of flights, minimise UAV's physical damages during landing, acquire good GNSS satellite signals and also possible to derive accurate orthomosaicked images and DEMs.

Summary and Conclusions
This article reports the authors experiences obtained from UAV surveys conducted using fixed- From the UAV surveying experiences on three glacier sites, the study showcases that choosing appropriate locations for UAV take-off and landing is one of the crucial aspects for a successful UAV survey. Furthermore, the study recommends strategies for choosing appropriate takeoff/landing locations for UAVs, especially for fixed-wing UAVs. By following these recommendations, one can optimise the flight endurance, i.e. good GNSS satellite signal availability, minimise number of flights and increase coverage area on high mountain glaciers.
The knowledge developed from this study can be valuable information to the glaciologists and hydrologists, who are interested in using UAVs for mapping and monitoring of glaciers in the Himalayan region and possibly beyond. However, the major limitation of the fixed-wing UAV is its inability to take-off or land vertically. Care should be taken to avoid take-off and landing fixed-wing UAVs on the glacier surface where steep slopes and heavy debris exists.