Background & Summary

The cryosphere, covering approximately 52–55% of the Earth’s land surface, is a unique and vital Earth system component comprising all forms of frozen water, including glaciers, snow, permafrost, and sea ice1. Interactions between the cryosphere and other Earth system spheres significantly influence global energy, water, and carbon-nitrogen balances, thereby exerting profound impacts on weather and climate, the hydrological cycle and water resources, ecosystems, and even human sustainable development2,3,4,5. Among the various components of the cryosphere, mountain glaciers and polar ice sheets are the most significant. They are dynamic bodies of ice formed through the accumulation and transformation of snowfall and other forms of solid precipitation6,7,8,9. In the arid region of northwestern China, mountain glaciers serve as an irreplaceable freshwater resource, playing a critical role in sustaining the ecological environment and supporting human activities10,11.

Over the past century, the global climate has undergone a pronounced warming trend that is both persistent and intensifying, resulting in widespread and irreversible glacier retreat worldwide, with some regions experiencing complete glacier disappearance12,13,14,15,16,17. This ongoing glacier retreat not only serves as a direct indicator of global climate change but also represents a critical aspect of its impacts. The reduction and loss of glaciers inevitably affect water resource availability, ecosystem stability, and contribute to global sea-level rise. For regions dependent on glacier runoff, glacier disappearance signifies a diminished freshwater supply, potentially leading to significant challenges in agriculture, drinking water security, and energy production18,19,20,21. As a major mountain range in northwestern China, the Qilian Mountains not only constitute a critical part of the northeastern margin of the Tibetan Plateau but also serve as an important water source region for inland river basins in Asia22,23. Glaciers in the Qilian Mountains, as the primary freshwater resource of the region, play a vital role in regulating the regional water cycle, maintaining ecological balance, and supporting local socio-economic activities24,25,26. However, consistent with global changes, glaciers in the Qilian Mountains have also exhibited an accelerated retreat in response to the intensifying global warming24,25. Therefore, obtaining up-to-date and high-precision glacier data for the Qilian Mountains is of great importance for in-depth studies of regional glacier changes, as well as for formulating climate adaptation strategies, ensuring water resource security, and promoting sustainable development. However, the previously conducted First and Second National Glacier Inventories of China27,28 are relatively outdated (with data corresponding to approximately 1970 and 2007, respectively) and no longer accurately reflect the current status of glacier changes. Consequently, updating and acquiring new glacier data will provide a more accurate basis for research on regional glacier dynamics.

This study utilized high-resolution satellite imagery from 2018 to 2019, including the Gaofen series (Gaofen-1: GF-1; Gaofen-2: GF-2, and Gaofen-6: GF-6), Ziyuan-3 (ZY-3), and Landsat 8, as primary data sources. Glacier boundaries were delineated through visual interpretation, with relevant attributes refined according to the GLIMS (Global Land Ice Measurements from Space) coding scheme, resulting in the Glacier Inventory Dataset of the Qilian Mountains for 2018–2019. This dataset not only provides intuitive observations of the Qilian Mountains’ glaciers through remote sensing techniques but also offers researchers high-precision details on glacier boundaries and area changes. It systematically reveals the spatiotemporal distribution and change trends of glaciers in the Qilian Mountains, thereby facilitating enhanced glacier monitoring and conservation efforts in the region.

Methods

Data and framework

In this Qilian Mountain glacier inventory primarily utilized high-resolution satellite remote sensing images acquired during 2018 and 2019. These high-resolution data are provided by the China Centre for Resources Satellite Data and Application (https://www.cresda.cn/) and the Geospatial Data Cloud (https://www.gscloud.cn/). The 2018 dataset comprised 18 scenes from the GF-1 satellite, 18 scenes from GF-2 satellite, and 6 scenes from the ZY-3 satellite, mainly covering glaciers within Qinghai Province. The 2019 dataset included 6 scenes from GF-1 satellite, 2 scenes from GF-2 satellite, 8 scenes from GF-6 satellite, and ZY-3 satellite, primarily covering glaciers within Gansu Province. Satellite images with minimal cloud and snow cover—or those in which such cover did not affect glacier boundary identification—were selected. Additionally, Landsat 8 imagery acquired during the period of 2018–2020 was used as reference data. Cloud cover and seasonal snow accumulation directly compromise the accuracy of visual interpretation. To mitigate these interferences, we rigorously screened high-resolution imagery during the data selection phase, retaining only those with cloud coverage below 10% and acquired on clear days during the ablation period, thereby substantially reducing cloud and snow influences at the source. Nevertheless, minor residual interferences remain inevitable. To address this, we adopted the methodology of Muhammad & Thapa29,30, integrating Landsat 8 imagery by applying a cloud probability threshold to mask residual cloudy pixels; concurrently, we computed the Normalized Difference Snow Index (NDSI) and excluded snow-covered pixels; finally, we filled the resulting gaps with contemporaneous cloud- and snow-free Landsat 8 pixels to ensure the reliability of the outcomes. The detailed information of the primary images are presented in supplementary information.

The GDEM V3 digital elevation model (DEM) with a 30 m resolution was used as the data source for glacier attribute extraction. The GDEM V3 dataset Developed jointly by Ministry of Economy, Trade and Industry (METI) and National Aeronautics and Space Administration (NASA) is freely available to the public. On August 5, 2019, NASA and METI jointly released ASTER GDEM V3, which builds upon V2 by adding 360000 new optical stereo pairs, primarily to reduce elevation voids and anomalies over water bodies. GDEM V3 can be downloaded from the Geospatial Data Cloud (https://www.gscloud.cn/).

All raster layers were processed using QGIS software. The Qilian Mountains extend approximately 800 km in an east-west direction; therefore, the choice of the projection coordinate system for the vector layer must avoid distortion caused by this longitudinal elongation. Ignoring the characteristics of the projection system could affect the accuracy of glacier area calculations. Therefore, both raster datasets and vector layers were projected using the Asia North Albers Equal Area Conic projection with the WGS-84 ellipsoid. The satellite imagery was first processed through band combination to generate false-color composites that enhance the spectral contrast between snow/ice and other surface features. Subsequently, a high-resolution panchromatic band was fused with the false-color imagery to produce enhanced-resolution false-color images.

In this study, glacier boundaries in the Qilian Mountains were delineated using visual interpretation. Compared to automated or semi-automated methods, visual interpretation is more time-consuming but yields highly reliable glacier boundary data. To mitigate the limitations of visual interpretation—namely, its strong dependence on the interpreter’s expertise, experience, and attention, as well as potential subjective discrepancies among interpreters in assessing the same image features, which may reduce classification consistency—the delineation was independently performed by multiple experts with over a decade of glacier inventory experience. Each glacier in the Qilian Mountains was manually outlined, followed by three rounds of cross-validation and independent review, thereby minimizing methodological uncertainty and potential bias to the greatest extent possible. With reference to Landsat 8 imagery and DEM data, the primary image was geometrically corrected, followed by visual interpretation to delineate glacier boundaries. Upon completion of glacier boundary vectorization, cross-verification through visual inspection was conducted. After finalizing revisions, glacier attributes were extracted by integrating with DEM data, the workflow diagram was illustrated in Fig. 1.

Fig. 1
Fig. 1
Full size image

Workflow for the Production of the 2018–2019 Glacier Inventory Dataset of the Qilian Mountains.

Glacier extraction and coding

In this study, glacier boundary information was extracted by vectorizing remote sensing imagery. Common methods for extracting glacier area information from satellite images include visual interpretation, brightness thresholding, the Normalized Difference Snow Index (NDSI), and supervised classification31,32. Compared to other methods, visual interpretation enables rapid extraction of key features from large volumes of imagery, minimizing reliance on complex data processing. It provides an intuitive representation of information, facilitating quick understanding and decision-making. Therefore, this study employed visual interpretation to process remote sensing images and delineate glacier boundaries.

The First Chinese Glacier Inventory adopted the WGI coding system, which was established in 1977 by the World Glacier Inventory Temporary Technical Service (WGI/TTS) as specified in the “World Glacier Inventory Guidelines”33. The coding structure consists of the country code (CN for China), the continent code (5 for Asia), the watershed code and the glacier identification number. The watershed code is composed of five digits and characters representing the hierarchy from first- to fifth-order basins. The first-order basin code is denoted by uppercase letters; in the Qilian Mountains, there are two first-order basins with glaciers: the Inland River Basin (Y) and the Yellow River Basin (J). The second- to fourth-order watershed codes are assigned by starting from the outlet of the higher-level watershed and sequentially coding each tributary in a clockwise direction using Arabic numerals. The fifth-order watershed code, representing individual rivers, is denoted by English letters. Glacier identification numbers correspond to the glaciers within the fifth-order watershed, numbered clockwise using Arabic numerals and padded with leading zeros to ensure four digits. In the First Chinese Glacier Inventory, the country code was omitted from the coding scheme.

In the Second Chinese Glacier Inventory, The GLIMS (the Global Land Ice Measurements from Space) identification system was incorporated while retaining the WGI coding scheme. GLIMS identification system assigns a unique identifier to each glacier. The GLIMS code consists of 14 alphanumeric characters include latitude and longitude of glacier reference point. However, GLIMS lack intuitive representation of glacial geographic location and does not incorporate watershed information. Moreover, for adjacent small glaciers, the codes are highly prone to duplication, thereby compromising the uniqueness of identifiers. Therefore, in this study continues to use the WGI coding method. In cases of omissions or glacier separated, unencoded glaciers are updated only up to the fifth-level watershed code without assigning new glacier identification numbers.

Glacier attribute extraction

Glacier attribute information was extracted utilizing the analytical tools available in the QGIS. The Field Calculator was used to compute geometric parameter of glaciers, such as area and perimeter, retaining three significant digits.

Using DEMs as input, slope gradient and aspect were derived by performing raster terrain analysis tools. Subsequently, the Zonal Statistics tool was used to extract elevation and slop values within each glacier boundary.

Aspect is a circular variable, its mean cannot be directly calculated using zonal statistics. Instead, the extraction of glacier mean aspect followed the method proposed by Davis et al.34.

Data Records

The 2018-2019 Qilian Mountains Glacier Inventory Dataset (designated as the Third Glacier Inventory of the Qilian Mountains to distinguish it from the First and Second Inventories) which is stored in Shapefile format is freely available from the National Tibetan Plateau Data Center35.

Data Overview

The attribute table of the Third Glacier Inventory of the Qilian Mountains Dataset comprises 18 fields, including glacier identification codes, remote sensing image acquisition dates, glacier area, perimeter, elevation, slope, and aspect (Table 1). During the cataloging process, glaciers in the eastern section of the Altun Mountains within Gansu Province were also vectorized and their information collected. These glaciers belong to the Danghe and Haerteng Rivers, and to ensure the integrity of the watershed, they were included in the statistical scope of glaciers in the Qilian Mountains. Unless otherwise specified, subsequent analyses will not separately address glaciers in the eastern Altun Mountains section. Within the Qilian Mountains-eastern Altun Mountains region, there are 11 third-level watersheds, which belong to the Hexi inland drainage system, the Qinghai inland drainage system, and the Yellow River drainage system (Table 2).

Table 1 Description of the Attribute Fields in the 2018-2019 Qilian Mountains Glacier Inventory Dataset.
Full size table
Table 2 Tertiary Watersheds in the Qilian Mountains and the Eastern Altyn-Tagh Region.
Full size table

Technical Validation

Uncertainty assessment

In glacier inventory mapping utilizing remote sensing imagery, the delineation of glacier boundaries constitutes a critical data processing procedure. The precision of glacier boundary interpretation, which benefits from clearly discernible demarcations between glacial and non-glacial areas in the imagery, depends not only on spatial resolution but also on acquisition timing and surface conditions. To quantify the uncertainty \({U}_{A}\) in area estimation during feature identification from raster imagery, it is assumed that feature boundaries pass through the center of each pixel. Following co-registration, the area uncertainty \({U}_{A}\) of the raster image is calculated using the following equation36,37:

$${U}_{A}=\frac{1}{2}\times{l}\times{\lambda}$$

Where l represents the feature perimeter and \(\lambda\) denotes the spatial resolution of the imagery.

Using the perimeter attribute from the Qilian Mountains glacier inventory data, we calculated the corresponding area extraction errors for individual glaciers at multiple spatial resolutions commonly found in remote sensing imagery: 0.5 m, 1 m, 2.5 m, 5 m, 10 m, 15 m, and 30 m (Fig. 2). Higher raster image resolution leads to significantly improved accuracy in area identification, while also reducing the minimum detectable target area. Glacier boundaries exhibit irregular shapes, and the uncertainty in glacier area extraction is generally inversely proportional to glacier size, though exceptions exist. For instance, glaciers with similar areas but more circular boundary geometries—characterized by smaller perimeters—demonstrate lower uncertainty. Therefore, to comprehensively evaluate the relationship between glacier dimensions and area extraction uncertainty, the area-to-perimeter ratio (A/P) is adopted as an evaluation metric.

Fig. 2
Fig. 2
Full size image

Glacial area extraction error by raster images with different resolutions.

The A/P ratio exhibits an inverse relationship with uncertainty. With an uncertainty threshold set at 0.2 (values below this threshold representing acceptable error margins), the maximum uncertainties in glacier area delineation at raster resolutions of 0.5 m and 1 m remain below this threshold, measuring 0.07 and 0.14 respectively. At coarser resolutions ranging from 2.5 to 30 m, area delineation uncertainties exceed 0.2 when the A/P ratio falls below 5.74, 12.49, 24.96, 37.50, and 49.98 for the respective resolutions. When the uncertainty threshold is set at 0.5, the error constitutes half of the identified feature’s area. With raster resolutions of 5 m and above, cases where uncertainty exceeds 0.5 occur, with corresponding A/P ratios falling below 4.82, 9.45, 14.95, and 19.91 respectively. At raster resolutions of 10 m and coarser, the maximum uncertainty in feature area identification exceeds 1.0, indicating that the margin of error is comparable to the area of the feature itself, thus merely meeting the minimum requirement for identification. The corresponding A/P ratios for these resolutions fall below 4.05, 7.06, and 9.45, respectively. In glacier inventory studies, the minimum glacier area has typically been set at 0.01 km²38. Using 30 m resolution Landsat imagery as an example, this corresponds to approximately 11 – 12 pixels. However, the associated area extraction uncertainty in such cases already exceeds 0.5.

The aforementioned uncertainty quantification is based on theoretical calculations. In practical area extraction processes, accuracy is also influenced by surface conditions of the features. For instance, the terminus of Laohugou Glacier No. 12 is partially covered by sparse supraglacial debris distributed along the marginal zone, with widths ranging from 14 to 50 m. These widths correspond to 4 and 2 pixels in 15 m and 30 m ground resolution raster images, respectively. In automated glacier boundary identification, factors such as water bodies and cloud cover may cause misinterpretations, while visual interpretation or manual digitization strongly depends on the field experience and subjective judgment of the operator.

Distribution of glaciers in the Qilian Mountains

In 2018/2019, the Qilian Mountains host a total of 2797 modern glaciers, covering a combined area of 1489.53 km². Among these, the Laohugou Glacier No.12 (code: 5Y448D0012) in the Shule River basin is the only one exceeding 20 km² in size, its area to be 20.09 km². The distribution of glacier numbers and areas in the Qilian Mountains is uneven across different size classes (Fig. 3b,c). As glacier size classes increase, the number of glaciers decreases progressively. The two size categories with the greatest number of glaciers are those smaller than 0.01 km² and between 0.01 to 0.50 km². Glaciers smaller than 0.10 km² dominate in quantity, totaling 1147 glaciers, which accounts for 41.01% of all glaciers in the Qilian Mountains. However, their total area is only 49.03 km², representing just 3.29% of the total glacier area in the region. This area is only twice that of the Laohugou No.12 Glacier (20.09 km²), despite the much larger number of small glaciers. The 0.10 to 0.50 km² size category also contains a considerable number of glaciers, with 1,003 glaciers totaling 247.19 km², accounting for 35.86% of the glacier count and 16.59% of the total glacier area in the Qilian Mountains. Glacier area is more evenly distributed among the size classes of 0.10–0.50 km², 0.50–1.00 km², 1.00–2.00 km², and 2.00–5.00 km², which contribute 16.59%, 13.94%, 18.29%, and 24.22% of the total glacier area, respectively. Among these, the 2.00–5.00 km² category has the largest total glacier area but represents only 4.33% of the total glacier count in the Qilian Mountains.

Fig. 3
Fig. 3
Full size image

Spatial distribution of glaciers in the Qilian Mountains (a). Spatial distribution of glacier area and number across tertiary watersheds; (b). Glacier area statistics; (c). Glacier aspect statistics; (d). Distribution of maximum, median, and minimum glacier elevations; (e). Mean glacier slope distribution; f. Distribution range of mean glacier elevations).

The largest number of glaciers in the Qilian Mountains face north, accounting for 49.34% of the total glacier count. Glaciers oriented northwest rank second, comprising 17.41%, while those facing west are fewer, representing 4.93%. In terms of area, north-facing glaciers have the greatest cumulative extent, covering 50.83% of the total glacier area in the Qilian Mountains. Glaciers oriented primarily southward are relatively few in both number and area. Southeast-facing glaciers account for the smallest area proportion at 2.63% and represent 2.82% of the total glacier count. The least numerous are southwest-facing glaciers, constituting 3.32% of the glacier count and covering 4.35% of the total glacier area.

There is an overall trend of decreasing glacier size and number from east to west across the Qilian Mountains. Northern slope watersheds exhibit greater glacier area and higher glacier counts than those on the southern slopes (Fig. 3a). The Shule River watershed contains the most extensively developed glaciation in the eastern Qilian–Altun Mountains, accounting for 24.92% of the total number and 31.39% of the total area of glaciers in the region. In contrast, glaciers are sparsely distributed in the Bayinguole, Buha, and Datong River watersheds. Among them, the Bayinguole River watershed has the fewest glaciers, with only 11 glaciers (0.39% of the total), and the smallest total glacier area (0.14%). Although the Buha River is the largest tributary in the Qinghai Lake watershed, only 27 glaciers are present in its upper reaches, accounting for 0.95% of the total glacier number and 0.58% of the total glacier area. The Datong River, an outflow river of the Yellow River watershed and the only one in the Qilian Mountains originating from a glacierized region, hosts 66 glaciers, contributing 2.32% to the total number and 0.88% to the total area of glaciers in the eastern Qilian–Altun Mountains.

The average glacier elevation in the Qilian-Altun Mountains region is 4855.31 m a.s.l., with a general west-to-east decreasing trend. Glaciers on the southern slopes are slightly higher in elevation than those on the northern slopes (Fig. 3f). The Beida, Shule, and Dang Rivers exhibit a wide range of glacier elevations, corresponding to the large number and total area of glaciers within these watersheds. In contrast, the Shiyang River watershed has the lowest and most concentrated average glacier elevations. The Datong River and Hei River watersheds show similar elevation distribution patterns, while the Yuka-Tataleng River watershed features the highest average glacier elevations, characterized by a relatively symmetric unimodal distribution.

The distribution patterns of maximum, median, and minimum glacier elevations are generally consistent across the watersheds (Fig. 3d). The Beida, Shule, and Dang River watersheds show the widest elevation ranges. The Shiyang River watershed has the lowest maximum, median, and minimum glacier elevations among all watersheds. In terms of maximum elevation, the Shiyang, Hei, and Datong River watersheds exhibit similar values. The Yuka–Tataleng River watershed features the highest glacier elevations across all three metrics (minimum, median, and maximum), with particularly distinct differences in minimum and median elevations compared to other watersheds. The Ha’erteng River watershed shows a similar maximum elevation distribution to that of the Yuka–Tataleng River watershed.

The distribution of average glacier slopes across watersheds is shown in Fig. 3e. The Beida River watershed has the steepest average glacier slope, reaching 30.44°, while the Shiyang River watershed exhibits the widest range of average slope values. The Bayinguole River watershed has the gentlest average slopes, with a mean of 23.95°. Overall, glacier slopes in the eastern Qilian-Altun Mountains tend to show a west-to-east variation, with slopes generally increasing and then decreasing, which is closely related to glacier morphology and types.

Usage Notes

This dataset has been publicly released and is freely available for download from the National Tibetan Plateau Data Center. When using the data, geographic information software such as ArcGIS, QGIS, or SuperMap can be employed for data reading, editing, and other operations.