Atmospheric warming is intensifying glacier melting and
glacial-lake development in High Mountain Asia (HMA), and this could
increase glacial-lake outburst flood (GLOF) hazards and impact water
resources and hydroelectric-power management. There is therefore a pressing
need to obtain comprehensive knowledge of the distribution and area of
glacial lakes and also to quantify the variability in their sizes and types
at high resolution in HMA. In this work, we developed an HMA glacial-lake
inventory (Hi-MAG) database to characterize the annual coverage of glacial
lakes from 2008 to 2017 at 30 m resolution using Landsat satellite imagery.
Our data show that glacial lakes exhibited a total area increase of
90.14 km
High Mountain Asia (HMA), consisting of the whole Tibetan Plateau and adjacent mountain ranges such as the Himalayas, Karakoram, and Pamirs, contains the largest area of mountainous glaciers in the world. Atmospheric warming has resulted in widespread glacier retreat and downwasting in many mountain ranges of HMA (Bolch et al., 2012; Brun et al., 2017), which favors the formation and development of a large number of glacial lakes. However, glacial lakes have been incompletely documented over small time intervals. Glacial-lake development varies according to climatic, cryospheric, and lake-specific conditions, including whether the basin geometry is connected to glaciers and the length of lake and glacier contact (Zhao et al., 2018).
There have been many previously published studies devoted to mapping glacial
lakes using remote-sensing data over different regions of HMA. Some works
have focused on investigating the development of relatively large glacial
lakes. Rounce et al. (2017) identified 131 glacial
lakes in Nepal in 2015 that had an area greater than 0.1 km
Because small glacial lakes are highly variable in their shape, location,
and occurrence and are clearly sensitive to the warming climate and glacier
wastage, a growing number of scholars have been paying attention to their
abundance. Salerno et al. (2012) provided a
complete mapping of glacial lakes (including lake size less than
0.001 km
All of these studies significantly help to fill the data gap relating to
information about glacial lakes in the HMA region. At the global scale,
Pekel et al. (2016) used millions of Landsat satellite
images to record global surface water over the past 32 years at 30 m
resolution, and many large and visible glacial lakes were also included.
More recently, Shugar et al. (2020) mapped
glacial lakes with areas
In summary, a homogeneous, annually resolved inventory and analysis of the
spatial and temporal extent of different types of glacial lakes over the
entire HMA region are still lacking. In this study, we developed an HMA
glacial-lake inventory (Hi-MAG) database to characterize the annual coverage
of glacial lakes from 2008 to 2017 at 30 m resolution. A total of 40 481
Landsat scenes were processed using the Google Earth Engine (GEE)
cloud-computing platform to delineate glacial lakes (located within 10 km of
the nearest glacier terminus) larger than 9 (e.g.,
Lakes were manually classified into four categories according to their position relative to the parent glacier or their formation mechanisms (Fig. A1). Category (i) constitutes proglacial lakes, which are usually connected to the glacier tongue and dammed by glacier ice or unconsolidated or ice-cemented moraines (a mixture of ice, snow, rock, debris, clay, etc.). Proglacial lakes are located next to the glacier terminus and receive meltwater directly from their mother glaciers. Category (ii) constitutes supraglacial lakes, which are ponds that form in depressions on low-sloping parts of the surface of a melting glacier and are dammed by ice or the end moraine or stagnating glacier snout. Category (iii) constitutes unconnected glacial lakes, which are not currently directly connected to their parent glaciers, but they may to some extent be fed by at least one of the glaciers located in the basin. They may (but not necessarily) have recently detached from ice contact due to glacial recession. Although not directly connected with the parent glaciers, these glacial lakes are also an outcome of glacier melting in response to atmospheric warming. They can supply fresh water to major river systems of the HMA region, and their changes have significant scientific and socioeconomic implications (Nie et al., 2017; Song et al., 2016). Finally, category (iv) constitutes ice-marginal lakes, which are generally distributed on one side of the glacier tongue, meaning that the lake is dammed by the glacier ice on this side, while on the other side, it is bounded by a lateral moraine. With the increase in atmospheric warming and accelerated melting of glaciers, some glacier tributaries gradually detach from a main trunk glacier. These detachment locations, where glacier melting has been particularly intense, are in some cases also likely to form ice-marginal lakes. We note that such ice-marginal lakes are very common in some parts of the world (e.g., Alaska) but are not common in HMA (Armstrong and Anderson, 2020; Capps et al., 2011). Additionally, purely glacier-dammed lakes are formed by the advance of glaciers and dammed by almost pure glacier ice. Although the dam composition and structure are slightly different between proglacial lakes and glacier-dammed lakes because they are all located in the front of the glacier tongue and driven by the mother glacier, in the process of appending attributed information to each glacial lake, glacier-dammed lakes were merged into the proglacial-lake category.
Every lake was cross-checked manually for its boundary and attribution. We defined an uncertainty of 1 pixel for the detected glacial-lake boundaries and calculated the error in the lake area for the whole HMA region. We also assessed the inventory for climatic and geomorphological influences on lake distribution across HMA.
The HMA region refers to a broad high-altitude region in South and Central
Asia that covers the whole Tibetan Plateau and adjacent mountain ranges,
including the eastern Hindu Kush, western Himalaya, eastern Himalaya,
central Himalaya, Karakoram, western Pamir, Pamir-Alay, northern and western Tien Shan, Dzhungarsky Alatau, western Kunlun Shan, Nyainqêntanglha,
Gangdise Mountains, Hengduan Shan, Tibetan interior mountains, Tanggula
Shan, eastern Tibetan mountains, Qilian Shan, eastern Kunlun Shan, Altun
Shan, eastern Tien Shan, central Tien Shan, and eastern Pamir (Figs. 1 and
6a). It extends from 26 to 45
The HMA region is the source of several of Asia's major rivers, including the Yellow, Yangtze, Indus, Ganges, Brahmaputra, Irrawaddy, Salween, and Mekong. They play a crucial role in downstream hydrology and water availability in Asia (Immerzeel and Bierkens, 2010). Most glaciers in the Tibetan Plateau are retreating, except for the western Kunlun (Neckel et al., 2014; Kääb et al., 2015) and the Karakoram, where a slight mass gain is occurring (Bolch et al., 2012; Gardner et al., 2013). Moreover, glaciers in different mountain ranges show contrasting patterns. Local factors (e.g., exposure, topography, and debris coverage) may partly account for these differences, but the spatial and temporal heterogeneity of both the climate and degree of climate change may be the main reason. Glacial lakes are formed and develop temporally with the retreat or thinning of glaciers and are directly or indirectly fed by glacier meltwater. They are located within 10 km of the nearest glacier terminus (Wang et al., 2013; Zhang et al., 2015).
The HMA climate is under the combined and competing influences of the East
Asian and South Asian monsoons and of the westerlies (Schiemann et
al., 2009). This unique geographical position produces an azonal plateau
climate characterized by strong solar radiation, low air temperatures, large
daily temperature variations, and small differences between annual mean
temperatures (Yao et al., 2012). The annual mean temperature is
1.6
Location of the HMA region. Glacier outlines from the Randolph Glacier Inventory (RGI v5.0), the Second Chinese Glacier Inventory (CGI2), and the Glacier Area Mapping for Discharge from the Asian Mountains (GAMDAM) glacier inventory are drawn in blue. Publisher's remark: Please note that the above figure contains disputed territories.
A total of 40 481 satellite images, including Landsat 5 TM imagery during
2008–2011, Landsat 7 ETM
The SLC-off condition of Landsat ETM
One effective solution to reduce the influence of seasonal lake fluctuations on the mapping is to map glacial lakes and measure their long-term changes during stable seasons, when the lake extents are minimally affected by meteorological conditions and glacier runoff. Here, based on analyses of the mapping times of glacial lakes in different regions, the selected time series of Landsat data were generally from July to November. During this period of each year, the Landsat imagery featured lower perennial snow coverage. Following glacier runoff and precipitation, the area of a glacial lake is large, and changes in this area will be small (Nie et al., 2017; Chen et al., 2017; Zhang et al., 2015). These lakes may also reach their maximum extent around the end of the glacier ablation season (June to August) (Gardelle et al., 2013; Liu et al., 2014), except in the central and eastern Himalaya, where peak ablation extends into post-monsoon September and October. In monsoon-affected areas such as Nepal and Bhutan, monsoon cloud cover from July to mid-September means that clear-sky images can mostly only be obtained from late September to November. The southeastern Tibet region is problematic not only because the observation season is short but also as a result of abundant cloud cover, which is formed by the warm and humid airflow raised by the topography (Zhang et al., 2020; Umesh et al., 2018; Qiao et al., 2016).
As the most highly variable glacial lakes in the study area, supraglacial lakes change preferentially in the year, showing an increase in area during the pre-monsoon and rising to their peak area in the early monsoon (June to July) (Miles et al., 2017a, b). Although the selected image seasons are slightly different due to the meteorological conditions in different regions, they all comply with the same criterion that the lakes were in clear-sky images having little snow coverage. This ensured the initial reliability of the mapping of glacial lakes through the GEE cloud-computing platform. If no valid observations could be obtained, then the optimal mapping time needed to be broadened during the whole year.
To further increase data availability and also as the basis for data selection in the periods beyond the optimum mapping time, we set two criteria for the selection of imagery with valid observations over the potential glacial-lake area by using the cloud-score functions in GEE, including (i) cloud cover being less than 20 % in the 10 km buffer around each glacier outline of a Landsat scene or (ii) less than 20 % cloud cover for the entire scene. The cloud-score functions in GEE may have significant difficulty in detecting clouds in mountain headwaters with high snow and ice cover, where large amounts of snow and ice are likely to be identified as clouds. However, in this study, it was considered better to use much stricter criteria to filter out a larger number of images with lots of clouds or objects that look like clouds (snow or ice) to finally select only images with good observations.
For the development of the Hi-MAG database, we applied a systematic glacial-lake detection method that comprised two steps: initial glacial-lake
extraction and subsequent manual refinement of these lake-mapping results.
The main procedures for glacial-lake mapping using Landsat data, as shown in
Fig. 3, are as follows. (i) The Landsat top-of-atmosphere data were clipped
according to the extent of the glacier buffers and assembled into a
time series dataset. (ii) Poor-quality observations were identified. These
included areas affected by clouds, cloud shadow, topographic shadow, and
SLC-off gaps. Here, we used the Fmask routine (Zhu and Woodcock,
2012) to detect the clouds and cloud shadows in the imagery. Fmask has the
advantage of being able to process a large number of images in a
computationally efficient way. Topographic shadows are located in the areas
where the sunlight is blocked. Generally, on the dark side of high
mountains, the surface gradients are great, and the terrain reliefs are
small. Therefore, topographic shadows were masked using the slopes (larger
than 10
Based on the automated processing, nearly 60 % of glacial lakes in each
year can be correctly classified. Of the other lakes that were not properly
classified, 30 % were missed, and 10 % were misclassified. For such a
large-scale area that is characterized by various and complex climatic,
geological, and terrain conditions, this classification method is simple but
effective. The results are also reasonable since they provide very low
commission errors. To ensure the quality of the inventory, strict quality
control was conducted to visually inspect and correct the mapping errors
after the automated processing using GEE. False lake features, mainly
identified as mountain shadows and river segments, were manually removed by
overlapping mapped lake shorelines in the source Landsat imagery and
higher-resolution imagery in Google Earth. Some glacial lakes may be covered
by ice and clouds for years, grow at steep glacier tongues, or show
heterogeneous reflectance with the surrounding backgrounds. For these
missing glacial lakes, their boundaries were edited further using ArcGIS.
Furthermore, a cross-check and modification were conducted for each glacial
lake based on the lake-mapping results in conjunction with multi-temporal
Landsat imagery. Here, all the Landsat imagery that was used for the
inspection was downloaded manually from the United States Geological Survey
(USGS) Earth Explorer website (
Diagram of the glacial-lake mapping workflow.
Based on the final generated lake inventory data, we used the slope of a
linear regression of the lake area (over the grid cells of 1
Although all the lakes were manually checked and edited, due to the limitation of available images and other factors, the conditions for glacial-lake mapping were not perfectly consistent for each year. For example, the image dates were not consistent across the whole HMA region because of atmospheric disturbances, and there were also influences from varying lake characteristics, image quality (Bhardwaj et al., 2015; Thompson et al., 2012), ice, and shadows that obscured the lakes, which all contributed to detection errors in the lake extent and their annual variation. Generally, these errors were objective and acceptable as a result of the nature of the limited remote-sensing data. For this study, because we used time series data covering a period of 10 years for the estimation of annual changes in lake area and also because the errors only account for a small proportion of the total glacial-lake area for each year, the errors in the observed lake area caused by these different effects do not appear to affect the trends in the statistical results. In addition to the Theil–Sen estimator, a Mann–Kendall trend test was used to detect and further confirm the statistical confidence of the linear regression results, and all the estimated trends were found to fall within the 90 % confidence intervals. The upper and lower change estimates satisfying the 90 % confidence interval for the slope were also derived over the whole HMA region (Fig. A2).
Accuracy assessment of the mapping results is difficult due to the lack of field measurements of glacial lakes in continental-scale areas such as HMA. To obtain quality-controlled data, the glacial-lake vectors over the entire HMA for the years from 2008 to 2017 were rechecked and re-edited individually through dynamic cross-validation by 10 trained experts. This was a time-consuming process but was essential for maximizing the quality of the data.
A key factor influencing the estimation of the uncertainty in the glacial-lake area measurements is the spatial resolution of the satellite data. In
this study, the uncertainty in the glacial-lake area was estimated as an
error of
Numbers and total areas of glacial lakes for different
years.
Assuming an uncertainty of 1 pixel for the detected glacial-lake boundaries,
we calculated the systematic errors for the whole HMA region, and the
results are shown in Fig. 4. For the years between 2008 and 2017, the area
uncertainty in each glacial lake generally ranged from 0.30 % to 50 %,
with the mean value falling around 17 % and the standard deviation around
11 % (Fig. 4a). The maximum and mean values of area uncertainty for the
glacial lakes in 2010 were the lowest, while for 2016, the corresponding
statistics were the highest. This can be attributed to a number of different
factors. The maximum in the area uncertainty in glacial lakes is related to
the shape and size of a certain lake, as can be seen from Eq. (1).
However, its mean value is equal to the sum of the area uncertainties in
each glacial lake divided by the total number, which depends on the total
number of glacial lakes in a given year as well as the shape and area of
each lake. Furthermore, a close relationship can be found between the area
uncertainties and sizes of the glacial lakes (Fig. 4b). Most of the large
glacial lakes (area
The area coverage of glacial lakes increased by 90.14 km
Annual changes in glacial lakes were further analyzed spatially using a
1
We found that glacial lakes exhibited different expansion trends for
different lake types, and supraglacial and ice-marginal lakes have relatively few coverage areas comparing with proglacial and unconnected lakes (Fig. 6b, c). In the Nyainqêntanglha and central Himalaya, around half
of the glacial-lake area consisted of proglacial lakes, where most growth
occurred. In the negative-lake-growth (shrinkage) regions of the eastern
Tibetan mountains and Hengduan Shan, unconnected glacial lakes were
dominantly occupied. As the interaction with a glacier gradually weakens,
part of the water source supplied by that glacier is reduced, and when
combined with the effects from atmospheric warming and a decrease in
precipitation, this means that regions mainly consisting of unconnected
glacial lakes show a trend of decreasing area. Proglacial lakes contributed
approximately 62.87 % (56.67 km
We also noted that the large area growth of lakes occurred in areas with a
relatively large proportion of small glacial lakes, and this was mainly due
to the rapid growth of existing lakes and the formation of new lakes
(Fig. 6d). For example, in some areas of the central and eastern Himalaya and
Nyainqêntanglha that have large annual increases in lake area (greater
than 0.23 km
Glacial-lake area changes and area distribution.
To explore factors that have potentially influenced the glacial-lake
distribution across HMA, we focus on proglacial and supraglacial lakes, for
which the changes are closely related to glaciers, and expansion is most
rapid. Proglacial lakes frequently develop from the enlargement and
coalescence of one or more supraglacial lakes (Thakuri et al.,
2016; Umesh et al., 2018). Proglacial- and supraglacial-lake development from
2008 to 2017 is significantly correlated with initial lake area in 2008
(
For the years before 2008, the year-round Landsat 5 TM data in many years do not fully cover the HMA region. In this study, we constructed the inventory over a 10-year time period. This is shorter than typical glacier response times, which start from a minimum of 10 years for short, steep glaciers to over 150 years for long, debris-covered glaciers (Scherler et al., 2011). Hence, lake expansion is not expected to be coupled with short-term climate trends, particularly for debris-covered glaciers (Umesh et al., 2018). In the inclusion of mass balance forcing of glacial-lake changes, the same questions about response times also occur. Hence, rather than focus on the short-term evolution of lake expansion, we investigated whether the climate and other factors have influenced the overall distribution of lake area, as observed in 2017.
To investigate the factors influencing the predominance of proglacial and
supraglacial lakes, geomorphic, topographic, and climate parameters were
correlated with lake area over a 1
Some regions have comparable numbers of large debris-covered glaciers but substantial differences in total lake area and area growth rates (for example, the central Himalaya compared to central Tian Shan or western Pamir; Table A5). Regional differences in multi-decadal climate trends could play a role in this observation, with Nyainqêntanglha and the central and eastern Himalayan regions all being characterized by rapid warming and decreased precipitation since 1979 (Fig. 7c and d), favoring negative glacial mass balances (Brun et al., 2017). This plausibly explains why the lake area is typically larger in these regions relative to adjacent regions further to the west and north (e.g., the western Himalaya) despite often similar glacier characteristics (in terms of debris cover and glacier length) (Fig. 7e and f). Furthermore, there is very little debris-covered area but rapid warming in the eastern Himalaya, where proglacial lakes are abundant (Fig. 7f). These results emphasize that the distribution of supraglacial and proglacial lakes across HMA is primarily associated with the presence of large debris-covered glaciers, but regional variability in warming and precipitation trends over the past few decades has likely also had some influence (Shugar and Clague, 2011; Zhao et al., 2019; Umesh et al., 2018; Scherler et al., 2018). These results are consistent with previous findings at regional scales, which have demonstrated a rapid expansion of proglacial lakes on debris-covered glaciers, with expansion in the upstream direction demonstrated to occur primarily through a process of subsidence at the lake-contact debris-covered glacier tongue (Harrison et al., 2018; Song et al., 2016, 2017a).
Geomorphic and climatic influences on lake distribution.
We compared our dataset with that of X. Wang et al. (2020) for the closest
period (2017 from the Hi-MAG database and 2018 from X. Wang et al., 2020) over
the spatial extent of our HMA region. The differences in the total number
and area of lakes between these two datasets are 6206 and 223.97 km
To test the spatial correlation of the distributions of the glacial lakes in
the two datasets, we compared the numbers of glacial lakes and their areas
aggregated on a 0.1
To quantitatively and systematically evaluate the accuracy of our data, we implemented stratified random sampling (Song et al., 2017b; Stehman, 2012), in which the glacial lakes were divided into four strata. The sample sizes were the spatial resolution (30 m) of the data, and the strata were designed such that C0W0 indicates that both the results are non-glacial lakes, C0W1 indicates a non-glacial lake in the present data and a glacial lake in X. Wang et al.'s (2020) data, C1W0 indicates a glacial lake in the present data and a non-glacial lake in X. Wang et al.'s (2020) data, and C1W1 indicates that both results are glacial lakes.
A total of 4000 points were randomly selected, as shown in Fig. A5. The
number of samples for C1W1 and C1W0 were 1300 and 700, respectively, and
these numbers have almost the same ratio as that between the total areas for
the two strata (1450.50 vs. 732.77 km
For the 1300 samples that were considered by both datasets to be non-glacial lakes, after the pixel-by-pixel verification, 1215 were found to indeed be non-glacial lakes, while 37 were missed glacial lakes. In contrast, 1260 out of the 1300 samples belonged to the class of glacial lakes, and 25 were misclassified as glacial lakes by both inventories. A total of 307 error pixels were found in the results of X. Wang et al. (2020), constituting about half of the total validation number. For the glacial lakes identified only by our inventory, 678 out of 700 were correctly classified. Our results yielded high overall classification accuracy (88 %), user's accuracy (97 %), and producer's accuracy (82 %) for glacial-lake classification using Landsat data.
The Hi-MAG dataset was also compared with other Landsat-based lake
inventories (Nie et al., 2017; Pekel et al., 2016; Zhang et al., 2015).
The number of lakes in Hi-MAG was found to be 7268 higher, and the area was
644.26 km
The glacial-lake area observed in our lake dataset in the eastern Pamir and western Kunlun Mountains does not conform to the mapped surface water in the GSW for these sub-regions. While there are numerous glacial lakes from an open-water perspective, actually part of them are river segments. Additionally, the Himalaya, eastern Hindu Kush, and some other Tien Shan areas host thousands of glacial lakes that are not readily observable in the GSW dataset. Large discrepancies in mountainous glacial-lake estimates preclude a significant consistency between the GSW and our Hi-MAG lake data over the HMA region. The region with the highest consistency between GSW and Hi-MAG product is interior Tibet. There is little agreement for Tien Shan, where the weather is rainy and snowy in the region above 3000 m, and large quantities of ancient glacial deposits have accumulated. Here, glacial lakes are characterized by small sizes, and due to the influence of their source glaciers and lake beds as well as the water depth and sediment inflow, glacial lakes appear to have heterogeneous reflectance in the images. Errors could exist in datasets produced by automated classification, but, as noted, we also conducted detailed manual editing, so we were not relying exclusively on automatic classification. The Karakoram region seems to have fewer glacial lakes in our estimate, owing to the overestimation of surface water on debris-covered glaciers in the GSW dataset.
The low agreement between our Hi-MAG glacial-lake data and the GSW data is mainly due to their lack of systematic glacial-lake inventory and mapping capabilities. The lake dynamics and differing climate contexts within HMA may also lead to inconsistencies between the sub-regions. Hi-MAG might have made better use of the optimum satellite imaging season to map glacial lakes, potentially resulting in more complete mapping by avoiding conditions such as periods of lake ice that may confound mapping.
There are several important issues and limitations to the datasets produced
and the methods used within this study that are important to highlight to
potential users. (i) Bodies of water smaller than 9 connected pixels
(e.g.,
The Hi-MAG database is distributed under the Creative Commons Attribution 4.0 License. The data can be downloaded from the data repository Zenodo at
In conclusion, the Hi-MAG dataset and others have used Earth observation
satellite data, especially Landsat imagery, to provide a more consistent
delineation of large-scale glacial-lake changes. Some glacial-lake mapping methods have enabled local-scale area estimation or spatial representation of lake extent and change. Such methods result in relatively
good performance for lake areas that remain clear and show homogeneous
reflectance in the image but do not allow for continental-scale mapping of
glacial lakes that have spectral interference from other objects such as
glaciers, snow, clouds, turbidity, and the sedimentation characteristics of
the glacial lake itself or the atmospheric interference and terrain
effects. Automated methods for the extraction of glacial lakes over
large-scale areas have been further developed in this work. However, visual
interpretation and manual editing are still effective ways to ensure high
accuracy of lake inventories and append attributed information for further
analysis. Based on an error of
Mapping of glacial lakes across the Tibetan Plateau and adjoining ranges
reveals a complex pattern of lake occurrence and growth or shrinkage. During
the past 10 years, 2755 glacial lakes with a total area of 90.14 km
The freely downloadable, detailed Hi-MAG dataset can also be used in future studies to provide a sound and consistent basis on which to quantify critical relationships and processes in HMA, including glacier–climate–lake interactions, glacio-hydrologic models, GLOFs and potential downstream risks, and water resources.
Examples of the various types of glacial lakes found in the
HMA region:
Annual changes in lake area between 2008 and 2017 on a
1
Density (number per 100 km
Comparison of the results of
Distribution of validation samples selected using stratified random sampling. Blue polygons are glacier outlines taken from the Randolph Glacier Inventory (RGI v5.0), the Second Chinese Glacier Inventory (CGI2), and the GAMDAM inventory. Yellow polygons refer to buffer areas within 10 km of glacier terminals.
Comparison of the glacial lakes measured in the global
maps as in the
Mountain-wide glacial-lake number and area per year and total loss or gain from 2008 to 2017. The unit of area is square kilometers.
Mountain-wide annual glacial-lake area from 2008 to 2017 for proglacial lakes and unconnected lakes. Supraglacial and ice-marginal lakes have relatively few coverage areas and are not listed in the table. The unit of area is square kilometers. Abbreviations are used to represent the names of mountain ranges to save space (eastern Hindu Kush (EHK), western Himalaya (WH), eastern Himalaya (EH), central Himalaya (CH), Karakoram (K), western Pamir (WP), eastern Pamir (EP), Pamir-Alay (PA), northern/western Tien Shan (N/WT), central Tien Shan (CT), eastern Tien Shan (ET), western Kunlun Shan (WK), eastern Kunlun Shan (EK), Gangdise Mountains (G), Hengduan Shan (H), Tibetan interior mountains (TIM), eastern Tibetan mountains (ETM), Tanggula Shan (T), Qilian Shan (Q), Dzhungarsky Alatau (DA), Nyainqêntanglha (N)).
Areas of different sizes of glacial lakes in 2017 for some regions with large area growth rates. The unit of area is square kilometers.
Summary of correlation coefficients (
Regional summary of key topographic, geomorphic, and
climatological parameters compared to proglacial- and supraglacial-lake area
in 2017. Correlation coefficients are bold where
Statistical results of stratified random sampling.
FC designed the study, performed analysis, and wrote the manuscript. MZ developed the methodology, conducted the lake evolution analysis, and wrote the manuscript. HG provided the funding support and supervision. SA performed the climate and debris-cover analysis. JSK, UKH, and CSW discussed and drew conclusions. All authors contributed to the final form of the paper.
The authors declare that they have no conflict of interest.
We thank Tobias Bolch and Dan H. Shugar for their contributions to this project in its stages of development and Li Wang, Shiguang Xu, Zhengyang Lin, Hang Zhao, Yuanhuizi He, Tianchan Shan, Ning Wang, Zhenzhen Yin, and Jinxiao Wang for the cross-validations of the data that were so integral to this project.
This work was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (grant no. XDA19030101), the International Partnership Program of the Chinese Academy of Sciences (grant nos. 131211KYSB20170046 and 131C11KYSB20160061), and the National Natural Science Foundation of China (grant no. 41871345). Simon Allen was supported by the EVOGLAC project under the Swiss National Science Foundation (grant no. IZLCZ2_169979/1).
This paper was edited by Birgit Heim and reviewed by two anonymous referees.