Next Article in Journal
A Comparative Study of Factor Graph Optimization-Based and Extended Kalman Filter-Based PPP-B2b/INS Integrated Navigation
Next Article in Special Issue
Glacial Lake Outburst Flood Monitoring and Modeling through Integrating Multiple Remote Sensing Methods and HEC-RAS
Previous Article in Journal
Design of a Data Acquisition, Correction and Retrieval of Na Doppler Lidar for Diurnal Measurement of Temperature and Wind in the Mesosphere and Lower Thermosphere Region
Previous Article in Special Issue
Characterization of Three Surges of the Kyagar Glacier, Karakoram
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 15
Issue 21
10.3390/rs15215142
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Inventory of Glacial Lake in the Southeastern Qinghai-Tibet Plateau Derived from Sentinel-1 SAR Image and Sentinel-2 MSI Image

by
Yuan Zhang
1,
Jun Zhao
1,*,
Xiaojun Yao
1,
Hongyu Duan
1,
Jianxia Yang
1 and
Wenlong Pang
2
1
College of Geography and Environmental Science, Northwest Normal University, Lanzhou 730070, China
2
Xining Center of Natural Resources Comprehensive Survey, China Geological Survey, Xining 810000, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2023, 15(21), 5142; https://doi.org/10.3390/rs15215142
Submission received: 12 September 2023 / Revised: 23 October 2023 / Accepted: 25 October 2023 / Published: 27 October 2023
(This article belongs to the Special Issue Glacial Lakes and Related Hazards: Mapping, Monitoring, and Risk Assessment)
Browse Figures
Versions Notes

Abstract

:
The glacial lakes in the Southeastern Qinghai–Tibet Plateau (SEQTP) have undergone dramatic expansion in the context of global warming, leading to several glacial lake outburst floods (GLOFs) disasters. However, there is a gap and incompleteness in glacial lake inventories across this area due to the heavy cloud cover. In this study, an updated and comprehensive glacial lake inventory was produced by object-based image analysis (OBIA) and manual vectorization based on the Sentinel-1 SAR and Sentinel-2 MSI images acquired in 2022. Detailed steps regarding the OBIA were provided, and the feature set of Sentinel-1 SAR images suitable for extracting glacial lakes was also determined in this paper. We found that the mean combination of ascending-orbit and descending-orbit images is appropriate for mapping glacial lakes. VV-polarized backscattering coefficients from ascending-orbit achieved a better performance for delineating glacial lakes within the study area. Moreover, the distribution of glacial lakes was characterized in terms of four aspects: size, type, elevation, and space. There were 3731 glacial lakes with a total area of 1664.22 ± 0.06 km2 in the study area; most of them were less than 0.07 km2. Ice-contacted lakes were primarily located in the Palongzangbo basin (13.24 ± 0.08 km2). Nyang Qu basin had the most abundant glacial lake resources (2456 and 93.32 ± 0.18 km2). A comparison with previously published glacial lake datasets demonstrated that our dataset is more complete. This inventory is useful for evaluating water resources, studying glacier–glacial lake interactions, and assessing GLOFs’ susceptibility in the SEQTP.

Graphical Abstract

1. Introduction

Glacial lakes are formed by the advance or retreat of a glacier [1] that either impounds a lake or exposes a basin, which is then recharged by meltwater from modern glaciers [2,3]. Emerging as a product of glacial motion, the development and evolution of glacial lakes provide insights into the regional climate state [4]. In the context of intensifying global warming, the degradation of glaciers in alpine regions has led to an increasing abundance of glacial lakes in these areas [5,6,7]. This phenomenon has dual implications. On the one hand, it slows down the rapid depletion of water resources in mountainous regions. On the other hand, it amplifies the frequency and intensity of glacial lake outburst floods (GLOFs) [8,9]. This is particularly true for maritime-type glaciated areas, where glacial movement is more pronounced. Glacial lakes play a vital role in both the energy and water cycles of glaciated regions [10], while also serving as indicators of environmental change [11,12]. For instance, the position of glacier-eroded lakes and the scour marks on their ledges can signal the remnants of ancient glaciers and the direction of glacial movement. Modern glacial lakes are also key aids in identifying unreported historical GLOF events [13]. Additionally, hillslope erosion and landscape evolution are also relevant to glacial lakes, because GLOFs can erode channel banks and landscapes [14]. A comprehensive and detailed glacial lake inventory is of the utmost importance and necessity for assessing regional lake resources, quantifying the magnitude of GLOF susceptibility, and reducing the risks from glacial hazards.
More than 30 glacial lake inventories were compiled at different spatial and temporal scales in recent decades [8], such as in the Himalayas [15], the Tien Shan [16], the Third Pole region [17] and the Hengduan Mountains [18]. This proliferation can be attributed to the rapid advancement in remote sensing technology and computer science, as large-scale glacial lake inventories can be created more easily with the help of high-quality remote sensing images and good computer performance. Among various methods, the threshold classification approach, combined with band ratios and the Normalized Difference Water Index (NDWI), was the most applied [19,20]. Machine learning methods such as decision trees and Random Forests (RF) were also employed in extracting glacial lake outlines [21,22]. In these studies, manual vectorization was inevitably involved in the production process due to its high precision. Some scholars developed a fully automated method called the Lake Detection Algorithm (LDA) based on Landsat OLI image for mapping glacial lakes [23]. However, it had high requirements for the data sources and lacked general applicability. In addition to optical images, SAR images (mostly Sentinel-1 SAR images) were also utilized to demarcate glacial lake outlines [24,25,26,27], especially for some regions that were heavily affected by clouds. Typically, the backscatter coefficients of SAR imagery were preprocessed first, including intensity ratios and median filtering. Subsequently, threshold classification or machine learning methods were applied to extract lake boundaries. However, these studies are usually limited to individual lakes or a few glacial lakes. Meanwhile, it is challenging to create datasets of glacial lakes over a large area from SAR images alone due to the inherent characteristics of SAR images (layover and foreshortening) and the sporadic distribution of glacial lakes within a region.
The Southeastern Qinghai–Tibet Plateau (SEQTP) is often covered by clouds under the influence of climate and topography. The published glacial lake inventories covering the SEQTP only included glacial lakes within a 1 km or 10 km buffer zone around the modern glacier [8,17,19,20,28], with the latest year of 2019. However, the high cloud cover had led to inaccurate glacier boundaries over the SEQTP. Some outlines of glaciers within this area in the Second Chinese Glacier Inventory (SCCI) were still substituted by boundaries from the First Chinese Glacier Inventory (FCCI) compiled in the 1970s [29]. Therefore, the completeness of the published inventories needs to be enhanced. Also, the boundaries of glacial lakes beyond the 10 km buffer zone of glaciers should be compiled in the inventory from the water resource evaluation perspective.
The small proportion of glacial lakes in the study area, along with the limited availability of positive samples, causes a poor performance when using the deep learning method. Object-based image analysis (OBIA) is also adopted in lake-mapping [30,31]. However, the significant differences among features across a wide area make it difficult to achieve well-segmented image objects, which leads to the requirement of a lot of post-processing [32,33]. The acquisition of images encompassing the glacial lake and the surrounding area can break through these two limitations and cause SAR images and the OBIA method to be more effective in glacial lake extraction. Hence, we initially clipped the Sentinel-1 SAR images before applying the OBIA. The input boundaries were the buffer zone of three or five times the lake area in the glacial lake inventory produced by Wang et al. [8]. This approach yielded Sentinel-1 SAR images from the site of glacial lake development. For the glacial lakes outside the buffer zone, we acquired Sentinel-2 MSI images with cloudiness <5% from August to October 2022 through the Google Earth Engine (GEE) platform and used manual interpretation to obtain their boundaries. It is noteworthy that Sentinel-1 SAR images were still selected for areas that did not meet the conditions of cloud cover. The aims of this study were: (1) to detect the polarization modes and features of Sentinel-1 SAR images that are suitable for extracting glacial lake boundaries; (2) to produce a comprehensive glacial lake inventory for the SEQTP, enabling the accurate assessment of regional water resources and providing data support for subsequent studies on glacial hazards in this area.

2. Study Area

The study area in this paper consists of the Yigongzangbo basin, Nyang Qu basin and Palongzangbo basin (Zangbo and Qu mean river in Tibetan), collectively known as the Southeastern Qinghai–Tibet Plateau (termed SEQTP hereinafter), covering a total area of 43,486.9 km2. The topography of the SEQTP is dominated by high mountains and gorges, with an average elevation of over 4500 m (Figure 1). The major mountains in the study area are the Nyainqentanglha Mountains and the Mishmi Mountains. Against the climatic background of the Indian Ocean monsoon and continental westerly winds, the SEQTP has a hot and rainy summer and a dry and cold winter. Temperature differs significantly across this area, with generally higher temperatures in the north and west than in the south, with annual mean temperatures of 10 °C and 8 °C, respectively [34]. The spatial difference of annual precipitation is similar to that of temperature, with more precipitation in the north and west (900 m) and less precipitation in the south (650 m) [35,36,37]. Rivers are widely distributed in the territory, with the Yigong River, Nyang River and Palong River being the major rivers. They are the tributaries of the Yarlung Zangbo River (Brahmaputra River in India). Yigong River and Palong River intersect at Tongmai and then flow southward to converge into the Yarlung Zangbo River at a place known as the Great Bend of the Yarlung Zangbo River. Nyang River flows into the Yarlung Zangbo River near the township named Qiangna. There were 4186 glaciers identified in the study area in 2005; all of them are temperate glaciers [29]. The degradation of these glaciers provides important recharge to the rivers and leads to the extensive formation of glacial lakes within the SEQTP. Glacial lakes are vital water storage for agricultural irrigation and economic actions (e.g., hydroelectric power generation) in the study area and the lower Yarlung Zangbo River region. Meanwhile, they are also the initiators of many natural disasters in the SEQTP (e.g., GLOFs and debris flow) [34].

3. Data and Methods

3.1. Data

Three types of data were adopted to achieve the glacial lake mapping in this paper, including remote sensing images from Sentinel-1/2 satellites, glacial lake products produced by Wang et al. [8] and SRTM DEM. The SAR sensor carried by the Sentinel-1 satellite has four acquisition modes and various data products; the Sentinel-1A IW GRD images (spatial resolution: 10 m) were selected in this paper. We used the GEE cloud platform (https://code.earthengine.google.com, accessed on 20 August 2022) to pre-process and acquire Sentinel-1A SAR images from August to November in 2022. Some studies [2,17] identified that the images during this period are suitable for mapping lakes. Glacial lakes cover a small proportion of the study area and are scattered. Therefore, the glacial lake dataset produced by Wang et al. [8] was applied to create buffers for downloading the Sentinel-1 SAR images within those buffer zones. This dataset was compiled by manual vectorization combined with NDWI maps and false-color composite images, with absolute and relative area errors of 2.28 km2 and 13.2%, respectively, in 2018. This action reduces memory usage and effectively suppresses errors from different features (mountain shadows and ice crevices, etc.) during the glacial lakes mapping. Sentinel-2 MSI images were adopted to demarcate glacial lakes outside buffer zones. SRTM DEM was used to obtain the slope in the SEQTP to attenuate the effect of mountain shadow on the results.

3.2. Methods

The inventory of glacial lakes across the SEQTP in 2022 was produced based on two procedures (Figure 2) to obtain the complete lake outline in the perennially cloud-covered region. The primary steps included the OBIA, manual vectorization and post-processing of the results. Firstly, a buffer zone was created for the glacial lake outline created by Wang et al. [8] within the study area (area of buffer equal to three times or five times the glacial lake area). Secondly, Sentinel-1 SAR images acquired during August through November in 2022 were clipped using those buffer zones. Subsequently, the clipped images underwent multiresolution segmentation and RF methods to extract glacial lake outlines. Finally, the initial outlines were manually checked and revised by referring to Sentinel-2 MSI images and Google Earth images. Additionally, we manually compiled glacial lakes located outside the buffer zones based on Sentinel-2 MSI images with less than 5% cloudiness and Sentinel-1 SAR images (for the middle of the study area with greater than 30% cloudiness). The combination of both results constitutes the final glacial lake inventory. The key methods involved in OBIA and the post-processing of the results are elaborated below.

3.2.1. Polarization Mode Selection

Sentinel-1 SAR data acquired from the GEE platform include two polarization schemes: VV-polarized image and VH-polarized image. The glacial lake has different representations in these two polarized images (Figure 3). The VH-polarized images are fine-grained, exhibiting fewer differences between glacial lakes and their surroundings compared to the VV-polarized images (Figure 3b1–e1). Lakes with islands in the interior may not be extracted correctly from VH-polarized images since the islands lack discernible differences from the surrounding water (Figure 3a2,c2,e2). However, the islands within the lake can be identified from the VV-polarized image (Figure 3b2,d2). In addition, the lake boundary is more visible in the VV-polarized image than that in the VH-polarized image (Figure 3a3–e3), indicating that the results obtained from the former are more accurate when mapping glacial lakes. Therefore, the VV-polarized images are suitable for mapping the glacial lake in the SEQTP.

3.2.2. Image Segmentation

Object-based image classification greatly improves the classification efficiency compared to pixel-based image analysis, because its smallest unit for analysis is an object composed of many homogeneous pixels [38]. This is the basis for object-based image classification. In this study, the multiresolution segmentation in eCognition developer 9.0 was utilized to segment Sentinel-1 SAR images. This approach is especially effective for processing data with texture and a low contrast, such as SAR images [38]. A total of three parameters needs to be set when executing this algorithm, which are “segmentation scale”, “shape weight” and “compactness weight”, respectively. The range of the above three parameters was determined by the control variables method and visual interpretation. Then, a tool named the estimation of scale parameter (ESP) was adopted to ascertain the exact parameter value.
Segmentation scale controls the maximum heterogeneity allowed within the objects and directly dictates the size of the segments. Firstly, shape weight and compactness weight were set as the default (0.1 and 0.5, respectively). Then, the scale parameters from 1 to 30 were iteratively experimented with an increment of 10. It is obvious that the glacial lakes were severely over-segmented when the scale parameter was 1 (Figure 4). As scale parameters increase, the image object becomes larger. The objects change less when the segmentation scale is 20 and 30, indicating that the segments are saturated. Glacial lakes with different areas correspond to varying optimal segmentation scales. When the scale is 20, the glacial lakes with an area larger than 0.01 km2 were well segmented. The object boundaries closely matched with the glacial lake outlines. However, for glacial lakes with an area less than 0.01 km2, a scale of 10 is optimal. Therefore, a scale parameter below 20 was selected as the optimal range to yield complete classification results. Additionally, the shape weight and compactness weight were assigned to 0.2 and 0.8, respectively, after conducting 18 trials (parameters are shown in Table 1).
The optimal segmentation scales among features were significantly different because of the large heterogeneity between them. Therefore, the ESP tool in eCognition developer 9.0 was adopted to determine the optimal value of segmentation scale. This tool can calculate the local variance (LV) in the homogeneity among the image segments at different segmentation scales. The best scale is indicated by the peak rate of change of LV (ROC-LV) [39]. Table 2 presents the specific parameter settings in the ESP tool.

3.2.3. Object Features’ Construction and Selection

Object features describe relevant information about the objects and are important reference criteria for image classification. Figure 3 shows that the VV-polarized backscattering coefficient is appropriate for mapping glacial lakes. However, both the VV-polarized images obtained from the ascending orbit and those obtained from the descending orbit have their strengths and weaknesses. Some lakes are easier to distinguish on the ascending-orbit image than on the descending-orbit image (Figure 3b1,d1), but the opposite is also true (Figure 3b3,d3). The ascending-orbit image is contrary to reality; for example, the gullies may appear high and the ridges may appear low. The descending-orbit image is consistent with the real ground, but some of the glacial lakes in these images are not in line with the extent of the real lake (Figure 3d2). This phenomenon may be related to the angle of the radar incident pulse to the ground during imaging [30]. Therefore, VV-polarized images from both ascending orbit and descending orbit were scored in this paper.
A feature set was constructed using the backscattering coefficients of multi-temporal Sentinel-1 SAR images and slopes. The former consists of 12 features, specifically VV-polarized backscattering coefficients (ascending orbit and descending orbit) for from August to November in 2022 (median composites of monthly images) and the mean combination of VV-polarized backscattering coefficients from ascending orbit and descending orbit for each month.
The RF method was employed for feature selection, which can determine the importance of different features for accurate classification by calculating Out-Of-Bag-Error (OOB-Error) (Equation (1)) [40]. Out-Of-Bag refers to the original samples that were not selected during the sampling process (approximately 37% of the total samples) [41].
V I M A = 1 N ∑ t = 1 N ( B n t M A − B O t M A )
where MA represents a feature; VI(MA) denotes the importance of the feature; N is the number of decision trees (100 for this paper); B n t M A is the OOB-Error of the tth decision tree when noise is added to MA, and B O t M A is the OOB-Error of the tth decision tree when MA is not affected by noise. If the OOB-error of MA before and after adding noise is larger, this indicates that the feature has a greater impact on the classification results and is of higher importance.

3.2.4. Object Classification and Manual Vectorization

Object classification is the last procedure of the OBIA method, including rule-based classification and machine learning classification. RF, a machine learning algorithm, was selected to detect glacial lake objects in this paper. A total of 1960 training sample objects were input into the algorithm for glacial lake extraction. The number of positive and negative samples was equal (980) due to the slight difference in area between glacial lakes and the background. The extraction results were validated using 680 validation sample objects (comprising 340 positive and 340 negative samples). Overall Accuracy (OA) and Kappa coefficients were chosen as the criteria for selecting the features.
Manual digitization offers the highest accuracy and is widely adopted in research on glacial lake inventory [8,17,19]. To ensure the precision of our dataset, the preliminary results were manually revised by recognizing certain signs, such as the color and texture in the TCC images. This revision was conducted with reference to contemporaneous Sentinel-2 MSI images and high-spatial-resolution images in Google Earth. This approach was also adopted to map glacial lakes outside the buffer zones.

3.2.5. Error Assessment

Factors that contribute to boundary errors regarding the glacial lakes from manual interpretation involve the spatial resolution of images, cloud cover, mountain shadows, and subjectivity of interpreters [8,42,43,44]. Errors from cloud cover and mountain shadows can be maximally eliminated by cloud filter and experts’ experience. There is a criterion that 50% of the boundary pixels were included in the vectorization [10,45]. Hanshaw and Bookhagen [46] reported that area errors of glacial lakes attributed to manual digitization were compatible with a Gaussian distribution. Thus, only the errors caused by the spatial resolution of image and vectorization were assessed in this paper, which can be calculated by Equation (2), proposed by Hanshaw and Bookhagen [46]:
E r r o r 1 σ = P G × G 2 2 × 0.6872
where Error(1σ) is the area error of a single glacial lake within one standard deviation; P is the perimeter of the lake; G is the spatial resolution of image (10 m for Sentinel-1 SAR image and Sentinel-2 MSI image); 0.6872 is the revised coefficient under 1σ.
The cumulative area error for multiple glacial lakes can be obtained by Equation (3) according to the error propagation law:
E m = ∑ i = 1 n ( E r r o r ( 1 σ ) i 2
where Em is the cumulative area error for multiple glacial lakes; n is the number of glacial lakes in a specified range; i is the ith glacial lake in this range.

3.2.6. Classification of Glacial Lakes

Scholars have proposed diverse criteria for the classification of glacial lakes based on different research purposes. In this paper, the glacial lakes in the study area were classified into ice-contacted lake, ice-uncontacted lake, supraglacial lake and non-glacier-fed lake by referring to the criteria of Wang et al. [8]. The class attribute of glacial lakes derived from the OBIA was directly inherited from the glacial lake inventory of HMA. Other glacial lakes, however, were classified with the help of Google Earth 3D images and DEM during the vectorization process. Non-glacier-fed lakes are primarily formed by glacier ploughing and plucking and recharged by precipitation [3]. They typically formed during the Quaternary glacial period, with few modern glaciers in their upper reaches [3,47]. Thus, it is difficult to distinguish them from thermal melt lakes and tectonic lakes. The most practical approach to define the glaciation extent is to specify a minimum elevation threshold or a distance threshold from the modern glaciers. In this paper, the minimum elevation threshold was set. Regional climatic background needs to be considered when using the lowest elevation to indicate the glaciation frontier. Mool et al. [48] specified 3000 as an elevation threshold to record glacial lakes in Nepal. The SEQTP is geographically close to Nepal and shares a similar climatic context (highland climate). Therefore, 3000 was also taken as the lowest elevation of the glacial lake in the study area to filter the results.

4. Results

4.1. Optimal Segmentation Scale

The optimal segmentation scale of the Sentinel-1 SAR images in the three sub-basins (the Nyang Qu basin, the Yigongzangbo basin and the Palongzangbo basin) was determined separately by the ESP tool (Figure 5). Obviously, ROC-LV has two peaks, corresponding to the two segmentation scales of eight and fifteen, respectively. In this paper, eight was selected as the optimal scale to ensure that glacial lakes with a small area (<0.01 km2) were completely detected.

4.2. Optimal Feature Set

The importance of the 13 features was computed using the RF method (Table 3). We incrementally added features in descending order of importance to extract glacial lakes and calculated the OA and Kappa coefficient of the results for each feature combination (Figure 6). With the increase in the number of variables involved in classification, the OA initially undergoes a sharp rise, followed by slight fluctuations, then a gradual increase, and finally a decline. When the features with top six importance scores were incorporated, the OA rose from 87.88% to 96.05%, which is attributed to the low correlation and high importance score among these top-ranked features. After adding the 7th and 9th features, the classification accuracy changes less (from 96.05% to 96.04%). OA reached its peak when the first 12 features were incorporated (96.59%), along with a Kappa coefficient of 0.91. This indicates that the first 12 features are the optimal combination for outlining glacial lakes.
It can be observed that the mean combinations of ascending-orbit and descending-orbit images are more suitable for mapping glacial lakes (Table 3). The mean combinations for the four months are ranked as the 1st, 3rd, 4th, and 10th, respectively. The three features for October 2022 were ranked low (10th, 11th, and 13th, respectively), suggesting that this is not a favorable month for glacial lake extraction. VV-polarized backscattering coefficients from the ascending orbit scored higher than ones from the descending orbit, signifying that the former had a better performance when delineating glacial lakes within the study area. Moreover, the importance score for the slope is 1.29 (ranking 2), which indicates that slope is a crucial feature for glacial lake extraction.

4.3. Distribution of Glacial Lakes

A total of 3731 glacial lakes (1664.22 ± 0.06 km2) were extracted in 2022 within the SEQTP, which encompass supraglacial lakes, ice-contacted lakes, ice-uncontacted lakes and non-glacier-fed lakes (Figure 7a). Most of them were ice-uncontacted lakes (1671), with an area of 77.95 ± 0.16 km2. Non-glacier-fed lakes exhibited the smallest scale (0.03 km2), with a number and area of 1513 and 46.93 ± 0.11 km2. Conversely, ice-contacted lakes had the largest scale, numbering approximately a third of the count of non-glacier-fed lakes (318), yet they covered 30.02 ± 0.11 km2 (scale: 0.09 km2). The largest glacial lake in the study area, Jionglaco, is an ice-contacted lake. It covered an area of 5.43 ± 0.22 km2. Supraglacial lakes presented the smallest number and area of glacial lakes, 229 and 9.33 ± 0.06 km2, respectively. We categorized the glacial lakes within the study area into five sizes based on the natural Jenks approach (Figure 7b). The number of glacial lakes with an area of less than 0.07 km2 was 3279, accounting for 87.86% of the total glacial lakes in the study area. Only five glacial lakes had an area greater than 2.13 km2. As the area size increased, both the number and area of glacial lakes decreased. The glacial lakes with an area greater than 2.13 km2 are ice-uncontacted lakes and ice-contacted lakes, indicating that these are larger than other types.
The glacial lakes over the study area were concentrated above 4200 m (Figure 8). The number and area of glacial lakes in this range were 3561 and 125.51 ± 0.48 km2, respectively, representing 95.42% and 76.38% of the total amount in the study area. Above 3600 m, the number of glacial lakes increases and then decreases with elevation rises, peaking in the interval of 5000–5200 m (1152 and 30.87%). Only 41 glacial lakes were identified as being below 3600 m. Glacial lakes with large areas tend to be found at high elevations (Figure 8a). For example, glacial lakes with an area larger than 0.75 km2 exclusively develop in the range of 3800–5000 m. The area of glacial lakes fluctuates as elevation rises, with the maximum area being found between 4400 m and 4600 m (26.72 ± 0.09 km2 and 16.26%). Additionally, glacial lakes within 3800–4000 m were substantial due to the existence of the five glacial lakes with an area greater than 2.13 km2 in this range (24.10 ± 0.11 km2 and 14.67%). The smallest glacial lake area was found to be between 3000 and 3200 m (0.02 ± 0.002 km2). Figure 8b illustrates the elevation distribution of different glacial lake types. Supraglacial lakes have a lower position compared to the other three types, with an average elevation of 4681.654 m in 2022. This is because they mostly develop on the ice lobes of large glaciers with low relief. Ice-contacted lakes had the highest average elevation (4982.26 m). Ice-uncontacted lakes are mainly developed in the range of 3800–4000 m (12.58 ± 0.07 km2), whereas non-glacier-fed lakes are primarily developed between 4400 and 4600 m (15.53 ± 0.07 km2).
The SEQTP was divided into 14 × 6 nets to spatially characterize the distribution of glacial lakes. Most of the glacial lakes were developed in the northwest, west and southwest parts of the study area, while the northeast, east and southeast parts had fewer glacial lakes (Figure 9a). Among the three sub-basins, Nyang Qu basin had the largest glacial lake area (93.32 ± 0.18 km2). The glacial lakes in the Palongzangbo basin and Yigongzangbo basin covered an area of 46.71 ± 0.13 km2 and 24.19 ± 0.10 km2, respectively. Different types of glacial lakes exhibited distinct spatial distribution patterns. Non-glacier-fed lakes were primarily developed in the Nyang Qu basin (Figure 9a,b), covering an area of 41.03 ± 0.10 km2 and accounting for 87.43% of the total area of non-glacier-fed lakes. In contrast, the area of this type within the Yigongzangbo basin was only 1.17 ± 0.02 km2 (2.49%). The distribution pattern of ice-uncontacted lakes was similar to that of non-glacier-fed lakes, i.e., the largest area was found in the Nyang Qu basin (43.02 ± 0.13 km2 and 55.11%), and the smallest area was found in the Yigongzangbo basin (7.03 ± 0.04 km2 and 9.45%). The biggest area and smallest area for ice-contacted lakes were observed in the Palongzangbo basin and Nyang Qu basin, with values of 13.24 ± 0.08 km2 (44.10%) and 5.23 ± 0.04 km2 (17.42%), respectively. Supraglacial lakes were primarily concentrated in the Yigongzangbo basin, with an area and percentage of 4.08 ± 0.04 km2 and 43.68%. Conversely, the Palongzangbo basin had the least supraglacial lakes (1.21 ± 0.02 km2 and 12.96%).

5. Discussion

5.1. Assessment of Glacial Lake Volume

Bathymetric data from an ice-contacted lake within the SEQTP, named Bienongco, were measured on-site in August 2020 using Unmanned Surface Vehicles (USV, APACHE3). The maximum and average depths and volume of Bienongco were 181 m, 85.4 m, and 102.3 × 106 m3, respectively [49]. Duan et al. [49] validated published area-volume relations for glacial lakes on the Qinghai–Tibet Plateau using this measured bathymetry and found that they all underestimated the Bienongco’s volume. Zhang et al. [50] surveyed the bathymetry of 16 glacial lakes in the Himalayan and Qinghai–Tibet Plateau during 2017–2021. Eight of them were located within the scope of this study (Figure 1). Notably, the measured volume of the Bienongco differed between the two surveyed tasks due to differences in the density of the USV track. The route taken by our research team was denser than that of Zhang et al. [50], so the interpolation results (volume) we obtained were more accurate. Hence, seven lake bathymetries obtained by Zhang et al. [50], along with one measured by our team, were utilized to fit an area–volume formula in this study (V = 74.33 A2.29, where V is the volume in 106 m3 and A is the area km2) (Figure 10). Since the acquired field bathymetric data were from ice-contacted lakes, this formula was applied to evaluate the volume of ice-contacted lakes within the study area.
The total volume of 318 ice-contacted lakes within the SEQTP was estimated to be 5440.95 ± 145.76 × 106 m3 (error was obtained from the law of error propagation) (Figure 11). Approximately 89.27% of the ice-contacted lakes had a volume of less 1 × 106 m3. The lake with centroid coordinates of 96°13′25″E and 29°56′12″N exhibited the smallest volume (1 m3). Among the three sub-basins, Yigongzangbo basin exhibited the largest lake volume (3986.93 ± 73.77 × 106 m3), primarily due to a lake named Jionglaco. developed in this basin. Although the largest number of ice-contacted lakes was observed in Nyang Qu basin, their volume was the smallest (95.04 ± 4.90 × 106 m3). This was attributed to the fact that the size of ice-contacted lakes in this basin was small (0.04 km2). The ice-contacted lake volume in the Palongzangbo basin was 1266.25 ± 29.05 × 106 m3. Three ice-contacted lakes greater than 100 × 106 m3 were developed in this basin.
In this study, the empirical equations proposed by Cook et al. [51] and Konoval et al. [52] were selected to verify the accuracy of the obtained relationship (Supplementary Table S1). Table 4 demonstrates that the ice-contacted lakes volume from our equation differ significantly from the results of Konovalov et al.’s [52] volume calculation. The agreement between the results from this paper and Cook et al.’s [51] equation is high and the differences are less than 20%. Cook et al. [51] categorized glacial lakes with an error of less than 24% from the real volume as predictable types, indicating that the area–volume relation obtained in this study is relatively plausible. The larger the glacial lake area, the greater the difference in lake volume assessed using our equation and Cook et al.’s [51] equation. The volume evaluated in this paper is greater than that obtained from Cook et al.’s [51] calculation for ice-contacted lakes with an area of less than 0.013 km2 (Supplementary Table S1).

5.2. Comparison with Other Glacial Lake Datasets

The glacial lake dataset produced in this paper was compared with those produced by Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28]. Different criteria regarding the minimum area and glacial lake development region were adopted in the above inventories. To ensure fair comparison, we selected our glacial lakes based on their criteria (Table 5). Considering the area expansion of glacial lakes during the timespan between this paper and theirs, only the number was compared. The numbers of glacial lakes created by this study are all higher than that obtained in other studies (under equal conditions), indicating the good completeness of our results (Table 5). It is worth noting that the obtained data are an updated version of the glacial lake dataset created by Zhang et al. [17], which includes a much higher number of glacial lakes than the values given in Table 5. We screened these according to the criteria outlined in their paper, published in 2015. Inconsistencies between the criteria in the paper and the published dataset were also found in the inventories produced by Shugar et al. [28]. We examined the glacial lakes that were included in other inventories but were not identified by our study and concluded four reasons for these discrepancies (Table 6).
Table 5. Number of glacial lakes compiled in this study and other studies.
Table 5. Number of glacial lakes compiled in this study and other studies.
Dataset SourceLatest YearMinimum Area (km2)Development AreaCount (Them)Count (us)Discrepancy
Shugar et al. [28]20150.05Within 1 km of glacier polygon in the RGI6.0781948
Zhang et al. [17]20160.0027Within 10 km of glacier polygon in the RGI3.22072218858
Chen et al. [19]20170.0081Within 10 km of glacier polygon in the SCGI1081184362
Wang et al. [8]20180.0054Within 10 km of glacier polygon in the SCGI1565221230
Note: RGI6.0 represents the Randolph Glacier Inventory (v6.0). RGI3.2 represents the Randolph Glacier Inventory (v3.2). SCGI depicts the Second Chinese Glacier Inventory. Discrepancy refers to the number of glacial lakes that have been included in other glacial lakes inventories but were not identified in this study.
Table 6. Summary of reasons for discrepancy.
Table 6. Summary of reasons for discrepancy.
Sequence NumberNameDescription
1Incorrect extraction in other studiesBare land (Figure 12a), glacial tongue (Figure 12c) or mountain shadow (Figure 12d) were misclassified as glacial lakes in other datasets.
2Area thresholdSome glacial lakes in this study were excluded because they were smaller than the area threshold due to splitting (Figure 12b), being partially covered by ice floes (Figure 12e) or being interpreted from high-spatial-resolution images (Figure 12f).
3Glacial lakes omitted in this studyImage quality resulted in some glacial lakes being missed in this study (Figure 12h).
4Glacial lake specificationOther glacial lake inventories include rivers (Figure 12i) and lakes with dense vegetation growing around them (Figure 12g).
Remotesensing 15 05142 g012
Figure 12. The comparison between the glacial lakes compiled in this study and other studies. Backgrounds of (a,d–g,i) are the Sentinel MSI TCC images from August to October 2022. Background of (b) is the Sentinel-1 SAR image (VV-polarized backscatter coefficient for August, September and August as RGB). Background of (c) is the Sentinel-1 SAR image for August 2022. Background of (h) is the Landsat OLI image acquired on 19 August 2016 (Bands 6, 5, 2). (j) is the frequency of reasons for discrepancies between this paper and the Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28].
Figure 12. The comparison between the glacial lakes compiled in this study and other studies. Backgrounds of (a,d–g,i) are the Sentinel MSI TCC images from August to October 2022. Background of (b) is the Sentinel-1 SAR image (VV-polarized backscatter coefficient for August, September and August as RGB). Background of (c) is the Sentinel-1 SAR image for August 2022. Background of (h) is the Landsat OLI image acquired on 19 August 2016 (Bands 6, 5, 2). (j) is the frequency of reasons for discrepancies between this paper and the Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28].
Remotesensing 15 05142 g012
To illustrate the consistency of the spatial distribution of glacial lakes in the various datasets, the numbers of glacial lakes from different datasets were compared over 15.5 km × 16.7 km grids for the SEQTP. An evident correlation was observed between the dataset produced in this study and that of Wang et al. [8], Chen et al. [19] and Zhang et al. [17] (Figure 13a–c), demonstrating a high consistency in the spatial distribution of these datasets. Also, the glacial lakes in this study are more comprehensive compared to other datasets (most of the data points are distributed below the 1:1 line). The low correlation (R2 = 0.24) between our dataset and the dataset from Shugar et al. [28] was attributed to the limited amount of data (Figure 13d). Although the total number of glacial lakes in this paper was more than that of Zhang et al. [17] (Table 4), some glacial lakes were missed by this study due to the image quality (Figure 12h,i and Figure 13c).

5.3. Limitation and Prespective

Three limitations of our glacial lake dataset should be acknowledged. Firstly, some glacial lakes were omitted due to image quality. Only Sentinel-1A SAR images and Sentinel-2 MSI images from August to November 2022 were adopted to demarcate glacial lakes within the study area. However, it is challenging to produce a complete glacial lake inventory over a large area using images within one calendar year [53], especially for a highly cloud-covered region like SEQTP. Therefore, glacial lake distribution in current dataset mainly represented the characteristics between August and November 2022. Secondly, some glacial lakes may be misclassified. Although we classified the glacial lakes with the help of SRTM DEM and Google Earth images, their accuracy is still not guaranteed due to the lack of measured data. Finally, the accuracy of glacial lake volume is limited. The volume of ice-contacted lakes in the study area was determined by a volume–area formula based on measured glacial lakes. Nevertheless, this relationship had significant uncertainty due to the insufficient number of samples. For example, Jionglaco had an estimated volume of 3584.68 ± 72.96 × 106 m3.
Three plans will be implemented in the future. Firstly, the current glacial lake dataset will be supplemented with high-quality Sentinel-2 MSI images from recent years. Secondly, bathymetric data from large glacial lakes within the SEQTP (e.g., Jionglaco) will be measured to explore a more accurate volume–area relationship. Lastly, the temporal resolution of the glacial lake dataset will be improved. SEQTP is severely subjected to GLOFs, and the high-frequency mapping of glacial lakes can provide effective information for early identification prior to disasters in this region.

6. Conclusions

In this study, we determined a feature set of Sentinel-1 SAR images that is suitable for extracting glacial lakes and produced an updated and comprehensive glacial lake inventory across the SEQTP using the OBIA and manual vectorization. The most appropriate image objects for extracting glacial lakes can be obtained with the scale parameter, shape and compactness of 8, 0.2 and 0.8, respectively. The mean combination of ascending-orbit and descending-orbit images is more suitable for mapping glacial lakes. The feature named “Mean combination of ascending-orbit and descending-orbit image for August 2022” received the highest importance score (2.68).
A total of 3731 glacial lakes with an area of 164.22 ± 0.06 km2 was extracted in 2022 within the SEQTP. Among these, the majority were ice-uncontacted lakes, with a count and total area of 1671 and 77.95 ± 0.16 km2, respectively. Most glacial lakes have an area smaller than 0.07 km2. Glacial lakes are distributed unevenly in three sub-basins. Nyang Qu basin has the largest glacial lake area (93.32 ± 0.18 km2), including the most non-glacier-fed lakes (41.03 ± 0.10 km2) and ice-uncontacted lakes (43.02 ± 0.13 km2). Conversely, the Yigongzangbo basin has the smallest glacial lake area, primarily consisting of ice-contacted lakes (11.55 ± 0.07 km2). The highest number of glacial lakes was observed in the elevation range of 5000–5200 m (1152), while the largest area of glacial lakes was found in the elevation interval of 4400–4600 m (26.72 ± 0.09 km2). Although our dataset is more comprehensive than previous ones, it still misses some lakes, which need to be added before publication.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/rs15215142/s1, Table S1: Comparison of water volume in ice-contacted lakes.

Author Contributions

Conceptualization, X.Y. and J.Z.; methodology, X.Y. and J.Z.; software, Y.Z.; validation, Y.Z.; formal analysis, Y.Z.; writing—original draft preparation, Y.Z.; writing—review and editing, J.Z. and X.Y.; visualization, Y.Z., H.D., J.Y. and W.P.; supervision, J.Z.; project administration, X.Y. and J.Z.; funding acquisition, J.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research was jointly funded by National Natural Science Foundation of China (Grant No. 42071089), the Innovation Star Program of Gansu Province (Grant No. 2023CXZX-253) and the Graduate Research Grant Program of Northwest Normal University (Grant No. 2022KYZZ—B065).

Data Availability Statement

The data presented in this study are available on request from the author.

Acknowledgments

Authors appreciate the ESA and GEE cloud platform for their freely available Sentinel satellite images. Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28] are also acknowledged for providing us with the glacial lake datasets over the study area. Y.Z. thanks J.Z., X.Y., H.D., J.Y. and W.P. for their detailed guidance and advice during the preparation of this manuscript.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Qin, D.; Yao, T.; Ding, Y.; Ren, J. Glossary of Cryospheric Science, 2nd ed.; China Meterological Press: Beijing, China, 2016; ISBN 978-7-5029-6473-3. [Google Scholar]
  2. Wang, X.; Liu, S.; Yao, X.; Guo, W.; Yu, P.; Xu, J. Glacier lake investigation and inventory in the Chinese Himalayas based on the remote sensing data. Acta Geogr. Sin. 2010, 65, 29–36. [Google Scholar]
  3. Yao, X.; Liu, S.; Han, L.; Sun, M.; Zhao, L. Definition and Classification System of Glacial Lake for Inventory and Hazards Study. J. Geogr. Sci. 2018, 28, 193–205. [Google Scholar] [CrossRef]
  4. Xu, D.; Liu, C.; Feng, Q. Dangerous glacial Lake and outburst features in Xizang Himalayas. Acta Geogr. Sin. 1989, 44, 343–351+385-352. [Google Scholar]
  5. López-Moreno, J.I.; Valero-Garcés, B.; Mark, B.; Condom, T.; Revuelto, J.; Azorín-Molina, C.; Bazo, J.; Frugone, M.; Vicente-Serrano, S.M.; Alejo-Cochachin, J. Hydrological and Depositional Processes Associated with Recent Glacier Recession in Yanamarey Catchment, Cordillera Blanca (Peru). Sci. Total Environ. 2017, 579, 272–282. [Google Scholar] [CrossRef]
  6. Yao, T. Tackling on Environmental Changes in Tibetan Plateau with Focus on Water, Ecosystem and Adaptation. Sci. Bull. 2019, 64, 417. [Google Scholar] [CrossRef]
  7. Song, C.; Sheng, Y.; Ke, L.; Nie, Y.; Wang, J. Glacial Lake Evolution in the Southeastern Tibetan Plateau and the Cause of Rapid Expansion of Proglacial Lakes Linked to Glacial-Hydrogeomorphic Processes. J. Hydrol. 2016, 540, 504–514. [Google Scholar] [CrossRef]
  8. Wang, X.; Guo, X.; Yang, C.; Liu, Q.; Wei, J.; Zhang, Y.; Liu, S.; Zhang, Y.; Jiang, Z.; Tang, Z. Glacial Lake Inventory of High-Mountain Asia in 1990 and 2018 Derived from Landsat Images. Earth Syst. Sci. Data 2020, 12, 2169–2182. [Google Scholar] [CrossRef]
  9. Yang, C.; Wang, X.; Wei, J.; Liu, Q.; Lu, A.; Zhang, Y.; Tang, G. Chinese glacial lake inventory based on 3S technology method. Acta Geogr. Sin. 2019, 74, 544–556. [Google Scholar]
  10. Slemmons, K.E.H.; Saros, J.E.; Simon, K. The Influence of Glacial Meltwater on Alpine Aquatic Ecosystems: A Review. Environ. Sci. Process. Impacts 2013, 15, 1794–1806. [Google Scholar] [CrossRef]
  11. Wang, X.; Ding, Y.; Zhang, Y. The Influence of Glacier Meltwater on the Hydrological Effect of Glacial Lakes in Mountain Cryosphere. J. Lake Sci. 2019, 31, 609–620. [Google Scholar] [CrossRef]
  12. Zhang, G.; Bolch, T.; Allen, S.; Linsbauer, A.; Chen, W.; Wang, W. Glacial Lake Evolution and Glacier–Lake Interactions in the Poiqu River Basin, Central Himalaya, 1964–2017. J. Glaciol. 2019, 65, 347–365. [Google Scholar] [CrossRef]
  13. Zheng, G.; Bao, A.; Allen, S.; Ballesteros-Cánovas, J.A.; Yuan, Y.; Jiapaer, G.; Stoffel, M. Numerous Unreported Glacial Lake Outburst Floods in the Third Pole Revealed by High-Resolution Satellite Data and Geomorphological Evidence. Sci. Bull. 2021, 66, 1270–1273. [Google Scholar] [CrossRef] [PubMed]
  14. Cook, K.L.; Andermann, C.; Gimbert, F.; Adhikari, B.R.; Hovius, N. Glacial Lake Outburst Floods as Drivers of Fluvial Erosion in the Himalaya. Science 2018, 362, 53–57. [Google Scholar] [CrossRef] [PubMed]
  15. Li, J.; Sheng, Y.; Luo, J. Automatic Extraction of Himalayan Glacial Lakes with Remote Sensing. J. Remote Sens. 2011, 15, 29–43. [Google Scholar]
  16. Wang, X.; Ding, Y.; Liu, S.; Jiang, L.; Wu, K.; Jiang, Z.; Guo, W. Changes of Glacial Lakes and Implications in Tian Shan, Central Asia, Based on Remote Sensing Data from 1990 to 2010. Environ. Res. Lett. 2013, 8, 044052. [Google Scholar] [CrossRef]
  17. Zhang, G.; Yao, T.; Xie, H.; Wang, W.; Yang, W. An Inventory of Glacial Lakes in the Third Pole Region and Their Changes in Response to Global Warming. Glob. Planet. Chang. 2015, 131, 148–157. [Google Scholar] [CrossRef]
  18. Wang, X.; Chai, K.; Liu, S.; Wei, J.; Jiang, Z.; Liu, Q. Changes of Glaciers and Glacial Lakes Implying Corridor-Barrier Effects and Climate Change in the Hengduan Shan, Southeastern Tibetan Plateau. J. Glaciol. 2017, 63, 535–542. [Google Scholar] [CrossRef]
  19. Chen, F.; Zhang, M.; Guo, H.; Allen, S.; Kargel, J.S.; Haritashya, U.K.; Watson, C.S. Annual 30 m Dataset for Glacial Lakes in High Mountain Asia from 2008 to 2017. Earth Syst. Sci. Data 2021, 13, 741–766. [Google Scholar] [CrossRef]
  20. Dou, X.; Fan, X.; Wang, X.; Yunus, A.P.; Xiong, J.; Tang, R.; Lovati, M.; Van Westen, C.; Xu, Q. Spatio-Temporal Evolution of Glacial Lakes in the Tibetan Plateau over the Past 30 Years. Remote Sens. 2023, 15, 416. [Google Scholar] [CrossRef]
  21. Wangchuk, S.; Bolch, T. Mapping of Glacial Lakes Using Sentinel-1 and Sentinel-2 Data and a Random Forest Classifier: Strengths and Challenges. Sci. Remote Sens. 2020, 2, 100008. [Google Scholar] [CrossRef]
  22. Wang, X.; Wu, K.; Jiang, L.; Liu, S.; Ding, Y.; Jiang, Z.; Guo, W. Wide expansion of glacial lakes in Tianshan Mountains during 1990–2010. Acta Geogr. Sin. 2013, 68, 983–993. [Google Scholar]
  23. Bhardwaj, A.; Singh, M.K.; Joshi, P.K.; Snehmani; Singh, S.; Sam, L.; Gupta, R.D.; Kumar, R. A Lake Detection Algorithm (LDA) Using Landsat 8 Data: A Comparative Approach in Glacial Environment. Int. J. Appl. Earth Obs. Geoinf. 2015, 38, 150–163. [Google Scholar] [CrossRef]
  24. Wang, J.; Chen, F.; Zhang, M.; Yu, B. ACFNet: A Feature Fusion Network for Glacial Lake Extraction Based on Optical and Synthetic Aperture Radar Images. Remote Sens. 2021, 13, 5091. [Google Scholar] [CrossRef]
  25. Zhang, M.; Chen, F.; Tian, B.; Liang, D.; Yang, A. High-Frequency Glacial Lake Mapping Using Time Series of Sentinel-1A/1B SAR Imagery: An Assessment for the Southeastern Tibetan Plateau. Int. J. Environ. Res. Public Health 2020, 17, 1072. [Google Scholar] [CrossRef] [PubMed]
  26. Zhang, B.; Liu, G.; Zhang, R.; Fu, Y.; Liu, Q.; Cai, J.; Wang, X.; Li, Z. Monitoring Dynamic Evolution of the Glacial Lakes by Using Time Series of Sentinel-1A SAR Images. Remote Sens. 2021, 13, 1313. [Google Scholar] [CrossRef]
  27. Zhang, M.; Chen, F.; Tian, B.; Liang, D. Using a Phase-Congruency-Based Detector for Glacial Lake Segmentation in High-Temporal Resolution Sentinel-1A/1B Data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2019, 12, 2771–2780. [Google Scholar] [CrossRef]
  28. Shugar, D.H.; Burr, A.; Haritashya, U.K.; Kargel, J.S.; Watson, C.S.; Kennedy, M.C.; Bevington, A.R.; Betts, R.A.; Harrison, S.; Strattman, K. Rapid Worldwide Growth of Glacial Lakes since 1990. Nat. Clim. Chang. 2020, 10, 939–945. [Google Scholar] [CrossRef]
  29. Liu, S.; Yao, X.; Guo, W.; Xu, J.; ShangGuan, D.; Wei, J.; Bao, W.; Wu, L. The contemporary glaciers in China based on the Second Chinese Glacier Inventory. Acta Geogr. Sin. 2015, 70, 3–16. [Google Scholar]
  30. Korzeniowska, K.; Korup, O. Object-Based Detection of Lakes Prone to Seasonal Ice Cover on the Tibetan Plateau. Remote Sens. 2017, 9, 339. [Google Scholar] [CrossRef]
  31. Nie, Y.; Liu, Q.; Liu, S. Glacial Lake Expansion in the Central Himalayas by Landsat Images, 1990–2010. PLoS ONE 2013, 8, e83973. [Google Scholar] [CrossRef]
  32. Zhao, H.; Chen, F.; Zhang, M. A Systematic Extraction Approach for Mapping Glacial Lakes in High Mountain Regions of Asia. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2018, 11, 2788–2799. [Google Scholar] [CrossRef]
  33. Yang, Y.; Liu, Y.; Zhou, M.; Zhang, S.; Zhan, W.; Sun, C.; Duan, Y. Landsat 8 OLI Image Based Terrestrial Water Extraction from Heterogeneous Backgrounds Using a Reflectance Homogenization Approach. Remote Sens. Environ. 2015, 171, 14–32. [Google Scholar] [CrossRef]
  34. Ke, C.; Kou, C.; Ludwig, R.; Qin, X. Glacier Velocity Measurements in the Eastern Yigong Zangbo Basin, Tibet, China. J. Glaciol. 2013, 59, 1060–1068. [Google Scholar] [CrossRef]
  35. Qi, M.; Liu, S.; Yao, X.; Grünwald, R.; Gao, Y.; Duan, H.; Liu, J. Lake Inventory and Potentially Dangerous Glacial Lakes in the Nyang Qu Basin of China between 1970 and 2016. J. Mt. Sci. 2020, 17, 851–870. [Google Scholar] [CrossRef]
  36. Duan, H.; Yao, X.; Zhang, D.; Qi, M.; Liu, J. Glacial Lake Changes and Identification of Potentially Dangerous Glacial Lakes in the Yi’ong Zangbo River Basin. Water 2020, 12, 538. [Google Scholar] [CrossRef]
  37. Liu, J.; Yao, X.; Gao, Y.; Qi, M.; Duan, H.; Zhang, D. Glacial Lake Variation and Hazard Assessment of Glacial Lakes Outburst in the Parlung Zangbo River Basin. J. Lake Sci. 2019, 31, 1132–1143. [Google Scholar] [CrossRef]
  38. Niu, C.; Jiang, W.; Huang, X.; Xie, J. Analysis and Comparison between Two Object-Oriented Information Extraction Software of Feature Analyst and eCognition. Remote Sens. Inf. 2007, 2, 66–70+105. [Google Scholar]
  39. Drăguţ, L.; Csillik, O.; Eisank, C.; Tiede, D. Automated Parameterisation for Multi-Scale Image Segmentation on Multiple Layers. ISPRS J. Photogramm. Remote Sens. 2014, 88, 119–127. [Google Scholar] [CrossRef]
  40. Genuer, R.; Poggi, J.-M.; Tuleau-Malot, C. Variable Selection Using Random Forests. Pattern Recognit. Lett. 2010, 31, 2225–2236. [Google Scholar] [CrossRef]
  41. Zhang, L.; Gong, Z.; Wang, Q.; Jin, D.; Wang, X. Wetland Mapping of Yellow River Delta Wetlands Based on Multi-Feature Optimization of Sentinel-2 Images. Natl. Remote Sens. Bull. 2019, 23, 313–326. [Google Scholar] [CrossRef]
  42. Gardelle, J.; Arnaud, Y.; Berthier, E. Contrasted Evolution of Glacial Lakes along the Hindu Kush Himalaya Mountain Range between 1990 and 2009. Glob. Planet. Change 2011, 75, 47–55. [Google Scholar] [CrossRef]
  43. Hall, D.K.; Bayr, K.J.; Schöner, W.; Bindschadler, R.A.; Chien, J.Y.L. Consideration of the Errors Inherent in Mapping Historical Glacier Positions in Austria from the Ground and Space (1893–2001). Remote Sens. Environ. 2003, 86, 566–577. [Google Scholar] [CrossRef]
  44. Paul, F.; Huggel, C.; Kääb, A. Combining Satellite Multispectral Image Data and a Digital Elevation Model for Mapping Debris-Covered Glaciers. Remote Sens. Environ. 2004, 89, 510–518. [Google Scholar] [CrossRef]
  45. Fujita, K.; Sakai, A.; Nuimura, T.; Yamaguchi, S.; Sharma, R.R. Recent Changes in Imja Glacial Lake and Its Damming Moraine in the Nepal Himalaya Revealed by in Situ Surveys and Multi-Temporal ASTER Imagery. Environ. Res. Lett. 2009, 4, 045205. [Google Scholar] [CrossRef]
  46. Hanshaw, M.N.; Bookhagen, B. Glacial Areas, Lake Areas, and Snow Lines from 1975 to 2012: Status of the Cordillera Vilcanota, Including the Quelccaya Ice Cap, Northern Central Andes, Peru. Cryosphere 2014, 8, 359–376. [Google Scholar] [CrossRef]
  47. Yi, C.; Cui, Z. Classification and Sedimentary Types of Glacial Lakes in the Halasi River Catchment, the Altay Mountains, Xinjiang. Oceanol. Limnol. Sin. 1994, 25, 477–485. [Google Scholar]
  48. Mool, P.K.; Bajracharya, S.R.; Joshi, S.P. Inventory of Glaciers, Glacial Lakes and Glacial Lake Outburst Floods in Nepal; Hill Side Press (P) Ltd.: Kathmandu, Nepal, 2001; ISBN 978-92-9115-194-3. [Google Scholar]
  49. Duan, H.; Yao, X.; Zhang, Y.; Jin, H.; Wang, Q.; Du, Z.; Hu, J.; Wang, B.; Wang, Q. Lake Volume and Potential Hazards of Moraine-Dammed Glacial Lakes—A Case Study of Bienong Co, Southeastern Tibetan Plateau. Cryosphere 2023, 17, 591–616. [Google Scholar] [CrossRef]
  50. Zhang, G.; Bolch, T.; Yao, T.; Rounce, D.R.; Chen, W.; Veh, G.; King, O.; Allen, S.K.; Wang, M.; Wang, W. Underestimated Mass Loss from Lake-Terminating Glaciers in the Greater Himalaya. Nat. Geosci. 2023, 16, 333–338. [Google Scholar] [CrossRef]
  51. Cook, S.J.; Quincey, D.J. Estimating the Volume of Alpine Glacial Lakes. Earth Surf. Dynam. 2015, 3, 559–575. [Google Scholar] [CrossRef]
  52. Konovalov, V.G.; Rudakov, V.A. Remote Assessment of Reserve Capacity of Outburst Alpine Lakes. Lëd I Sneg 2016, 56, 235–245. [Google Scholar] [CrossRef]
  53. Lesi, M.; Nie, Y.; Shugar, D.H.; Wang, J.; Deng, Q.; Chen, H.; Fan, J. Landsat- and Sentinel-Derived Glacial Lake Dataset in the China–Pakistan Economic Corridor from 1990 to 2020. Earth Syst. Sci. Data 2022, 14, 5489–5512. [Google Scholar] [CrossRef]
Remotesensing 15 05142 g001
Figure 1. The location of the study area in the Qinghai-Tibet Plateau (a) and the geographical setting of the SEQTP (b).
Figure 1. The location of the study area in the Qinghai-Tibet Plateau (a) and the geographical setting of the SEQTP (b).
Remotesensing 15 05142 g001
Remotesensing 15 05142 g002
Figure 2. The glacial lake mapping workflow.
Figure 2. The glacial lake mapping workflow.
Remotesensing 15 05142 g002
Remotesensing 15 05142 g003
Figure 3. The manifestation of some glacial lakes in Sentinel-2 MSI true-color composite (TCC) images and Sentinel-1 SAR images with different polarization modes and flight directions. (a1–a3) are the Sentinel-2 TCC images. (b1–b3,c1–c3) are the VV-polarized and VH-polarized Sentinel-1 SAR image with a flight direction of ascending orbit, respectively. (d1–d3,e1–e3) are the VV-polarized and VH-polarized Sentinel-1SAR image with a flight direction of descending orbit, respectively. The red plates indicate the location of the glacial lakes in the figure.
Figure 3. The manifestation of some glacial lakes in Sentinel-2 MSI true-color composite (TCC) images and Sentinel-1 SAR images with different polarization modes and flight directions. (a1–a3) are the Sentinel-2 TCC images. (b1–b3,c1–c3) are the VV-polarized and VH-polarized Sentinel-1 SAR image with a flight direction of ascending orbit, respectively. (d1–d3,e1–e3) are the VV-polarized and VH-polarized Sentinel-1SAR image with a flight direction of descending orbit, respectively. The red plates indicate the location of the glacial lakes in the figure.
Remotesensing 15 05142 g003
Remotesensing 15 05142 g004
Figure 4. The segmentation results (red boundaries) of different segment scales with a fixed shape weight and compactness weight of 0.1 and 0.5, respectively.
Figure 4. The segmentation results (red boundaries) of different segment scales with a fixed shape weight and compactness weight of 0.1 and 0.5, respectively.
Remotesensing 15 05142 g004
Remotesensing 15 05142 g005
Figure 5. The optimal scale parameter.
Figure 5. The optimal scale parameter.
Remotesensing 15 05142 g005
Remotesensing 15 05142 g006
Figure 6. The relation between the number of variables and the Overall Accuracy (OA) and Kappa coefficient.
Figure 6. The relation between the number of variables and the Overall Accuracy (OA) and Kappa coefficient.
Remotesensing 15 05142 g006
Remotesensing 15 05142 g007
Figure 7. Number and area of glacial lakes of different types (a) and sizes (b) within the study area.
Figure 7. Number and area of glacial lakes of different types (a) and sizes (b) within the study area.
Remotesensing 15 05142 g007
Remotesensing 15 05142 g008
Figure 8. The altitude distribution of glacial lakes in terms of number, area (a) and type (b) within the study area. Bubble colors in subfigure (b) represent the types of glacial lakes shown in the horizontal axis.
Figure 8. The altitude distribution of glacial lakes in terms of number, area (a) and type (b) within the study area. Bubble colors in subfigure (b) represent the types of glacial lakes shown in the horizontal axis.
Remotesensing 15 05142 g008
Remotesensing 15 05142 g009
Figure 9. The spatial distribution of glacial lakes in terms of area and type within the study area (a) and glacial lakes area of different types in the three sub-basin (b).
Figure 9. The spatial distribution of glacial lakes in terms of area and type within the study area (a) and glacial lakes area of different types in the three sub-basin (b).
Remotesensing 15 05142 g009
Remotesensing 15 05142 g010
Figure 10. Empirical relationship between area and volume of ice-contacted lakes over the study area.
Figure 10. Empirical relationship between area and volume of ice-contacted lakes over the study area.
Remotesensing 15 05142 g010
Remotesensing 15 05142 g011
Figure 11. Estimated ice-contacted lake volume across the SEQTP in 2022.
Figure 11. Estimated ice-contacted lake volume across the SEQTP in 2022.
Remotesensing 15 05142 g011
Remotesensing 15 05142 g013
Figure 13. Correlation of number of glacial lakes produced by this study with that of Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28] over a 15.5 km × 16.7 km grid. The blue circles are correlations between our result and their results, the red lines are the fitting curves between our result and their results, and the green line is the 1:1 diagonal line.
Figure 13. Correlation of number of glacial lakes produced by this study with that of Wang et al. [8], Chen et al. [19], Zhang et al. [17] and Shugar et al. [28] over a 15.5 km × 16.7 km grid. The blue circles are correlations between our result and their results, the red lines are the fitting curves between our result and their results, and the green line is the 1:1 diagonal line.
Remotesensing 15 05142 g013
Table 1. The trial parameters for determining shape and compactness.
Table 1. The trial parameters for determining shape and compactness.
Trial NameFixed VariableIteration RangeStep
Shape weight
determination		Scale parameter: 20
Compactness weight: 0.5		0.1–0.90.1
Compactness weight
determination		Scale parameter: 20
Shape weight: 0.2 (obtained from above trial)		0.9–0.10.1
Table 2. Parameter settings in the ESP tool.
Table 2. Parameter settings in the ESP tool.
ParameterValueParameterValue
Increment of scale parameter1Number of loops15
Starting scale parameter5Shape weight0.2
Use of hierarchy1Compactness weight0.8
Table 3. The rank, name and score of variables.
Table 3. The rank, name and score of variables.
Rank of VariablesName of VariablesScore
1Mean combination of ascending-orbit and descending-orbit images for August 20222.68
2Slope1.29
3Mean combination of ascending-orbit and descending-orbit images for September 20221.06
4Mean combination of ascending-orbit and descending-orbit images for November 20220.94
5Ascending-orbit images for August 20220.93
6Descending-orbit images for November 20220.87
7Ascending-orbit images for September 20220.78
8Ascending-orbit images for November 20220.76
9Descending-orbit images for September 20220.75
10Mean combination of ascending-orbit and descending-orbit images for October 20220.74
11Ascending-orbit images for October 20220.73
12Descending-orbit images for August 20220.72
13Descending-orbit images for October 20220.45
Table 4. Comparison of ice-contacted lakes volume derived from this study with those calculated using empirical relationships from Cook et al. [51] and Konovalov et al. [52].
Table 4. Comparison of ice-contacted lakes volume derived from this study with those calculated using empirical relationships from Cook et al. [51] and Konovalov et al. [52].
Area Scale (km2)Volume Derived from This Study (106 m3)Cook et al.’s [51] Volume (106 m3)Konovalov et al.’s [52] Volume (106 m3)
ValueError (%)ValueError (%)
<0.070.120.123.920.2726.45
0.07–0.251.551.686.974.7363.35
0.25–0.757.948.9510.4724.9166.55
0.75–2.1343.8951.6514.72141.5268.92
>2.13229.06281.2818.18758.6769.72
Note: Cook et al.’s [51] equation: V = 0.1217 A1.4129. Konovalov et al.’s [52] equation: V = 0.8 A1.2 for ice-contacted lakes with an area of <0.01 km2, V = 0.6 A1.3 for ice-contacted lakes with an area of 0.01–0.05 km2, V = 0.4 A1.4 for ice-contacted lakes with an area of >0.05 km2. In the above equations, the volume unit is 106 m3; the area unit is m2. The values in the fourth and sixth columns are averages for each lake volume in the corresponding area interval.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhang, Y.; Zhao, J.; Yao, X.; Duan, H.; Yang, J.; Pang, W. Inventory of Glacial Lake in the Southeastern Qinghai-Tibet Plateau Derived from Sentinel-1 SAR Image and Sentinel-2 MSI Image. Remote Sens. 2023, 15, 5142. https://doi.org/10.3390/rs15215142

AMA Style

Zhang Y, Zhao J, Yao X, Duan H, Yang J, Pang W. Inventory of Glacial Lake in the Southeastern Qinghai-Tibet Plateau Derived from Sentinel-1 SAR Image and Sentinel-2 MSI Image. Remote Sensing. 2023; 15(21):5142. https://doi.org/10.3390/rs15215142

Chicago/Turabian Style

Zhang, Yuan, Jun Zhao, Xiaojun Yao, Hongyu Duan, Jianxia Yang, and Wenlong Pang. 2023. "Inventory of Glacial Lake in the Southeastern Qinghai-Tibet Plateau Derived from Sentinel-1 SAR Image and Sentinel-2 MSI Image" Remote Sensing 15, no. 21: 5142. https://doi.org/10.3390/rs15215142

APA Style

Zhang, Y., Zhao, J., Yao, X., Duan, H., Yang, J., & Pang, W. (2023). Inventory of Glacial Lake in the Southeastern Qinghai-Tibet Plateau Derived from Sentinel-1 SAR Image and Sentinel-2 MSI Image. Remote Sensing, 15(21), 5142. https://doi.org/10.3390/rs15215142

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop