Glacial Area Changes in the Ili River Catchment ( Northeastern Tian Shan ) in Xinjiang , China , from the 1960 s to 2009

The Ili River originates in the Tian ShanMountains of Northwest China before flowing into Kazakhstan and Lake Balkash. Melting snow and ice are its major contributors. We analyzed glacial changes in the upper Ili River basin between the 1960s and 2007/2009 using topographic maps and satellite imagery from a Landsat TM. The relationships between glacial changes and glacial size, topographic factors, and debris cover were examined. Our results found that total glacial area decreased by 485 ± 177.3 km (24.2% ± 8.8%) during the study period, and there were no advancing glaciers. Additionally, 331 glaciers disappeared and 18 disintegrated into two or three smaller glaciers.This study demonstrated a linear relationship between glacial area change and elevation. Changes in glaciers smaller than 1 km were affected by both glacial size and topographic factors, while larger ones were affected by size only. Area losses in debris-covered glaciers were smaller by 2.5% to 7.5% compared to clean ice of the same size in this basin. As in other glaciated regions, glacial retreat in the Ili River basin is attributed to global warming. The slightly increasing precipitation over the study period could not offset the ice melting.


Introduction
The Intergovernmental Panel on Climate Change (IPCC) reported that alpine glaciers respond to regional climate over a period of decades and thus serve as indicators of regional climate change [1].Most alpine glaciers in the world have shown shrinkage in area since the later part of the 19th century, as did the glaciers in the Tian Shan Mountains [1][2][3].However, advancing or surging glaciers were found in the Karakoram Himalaya, northwestern Himalaya, and Qilian mountains during the same period [4][5][6].
Global-warming-induced snow and ice losses from the cryosphere impact socioeconomic development because rising sea levels cause coastal flooding, melting glaciers cause flooding from rapid melting and glacial lake outburst, and glacial shrinkage reduces a dependable irrigation supply [7][8][9][10].Cryosphere-related disasters such as glacier/snow melt floods, glacial lake outburst floods, ice jam floods, and avalanches have frequently been observed in the Tian Shan region [11].In these mountains, hazards related to drifting snow, avalanches, ice jam floods, and glacial melt floods have occurred [11,12].There were no reports of glacial lake outburst floods in the study area, and many glacial lakes continue to expand in the Ili River basin as a result of glacial melting [13].Thus, it is important to know how glaciers have changed in this area.However, there has been little research to quantify the actual changes to glaciers [14,15].
Changes in length, area, thickness, and volume in response to temperature and precipitation changes are not uniform across the glaciers over periods of years to decades [6,[16][17][18][19].This variable pattern of responses to climate changes is expected to be affected by glacial size and topographic factors.Glaciers that are smaller in area experience greater shrinkage [16][17][18].However, there is no information about the relationship between glacial changes and topographic factors.In addition, there is seemingly conflicting research regarding the influence that debris cover plays on changes to glacial surface elevation, area, and length [20,21].There are debris-covered glaciers with no visible or only little area changes but significant mass changes.Glaciers in the Ili River watershed have a wide range of surface areas, elevations, and debris covers.This examination of changes in the Ili River basin's glaciers, from the 1960s to 2009, used topographic maps and satellite imagery from the Landsat Thematic Mapper (TM) to document glacial size, topographic factors affecting glacial area changes, and the amount of debris cover.

Study Area
The Ili River, originating in the central northwest Tian Shan Mountains and flowing from China into Kazakhstan (Lake Balkhash), is fed by melting snow and ice [22].Its total drainage area is 4.2 × 10 5 km 2 , including both the Chinese and Kazakhstan sections.Because this region lies on the west side of the mountains, it receives more precipitation than other parts of the Tian Shan Mountains; its westward opening allows westerly uplift [22].Precipitation at the yearround snow line is around 800-1000 mm per year.Snowfall, during the winter and spring, accounts for 40%-50% of the annual precipitation [22].This is different from other glaciated areas in the Tian Shan Mountains where most of the precipitation occurs during summer.Although the Ili River is mostly in Kazakhstan, 77.5% of the glaciers that feed it are in China [23].Our study focused on the glaciers within China (Figure 1).The World Glacier Inventory (http://nsidc.org/data/docs/noaa/g01130 glacier inventory/) has coded the Ili River basin as 5X04 with five tributary basins originating in China (5X042, 5X043, 5X044, 5X045, and 5X046).The first Ili River basin glacier inventory, performed in the 1960s, found 2373 glaciers totaling 2022.66 km 2 and occupying an estimated 127.96 km 3 in China [24].Snow line elevations ranged from 3630 to 3840 m above sea level (m asl).

Dataset and Methodology
3.1.Glaciers in the 1960s.The first Chinese inventory of glaciers (FCI) was compiled using primarily topographic maps scaled at 1 : 50,000 or 1 : 100,000.There were few glaciers from maps scaled at 1 : 1,000,000 because of the data quality [22].Although glacier boundaries were manually corrected using 1 : 43,000 and 1 : 60,000 aerial photographs and analytic photogrammetric methods [22], notable errors remained in the mapped delineation of glaciers.No vectorized glacier boundaries were included in this inventory compilation.Instead, glacier boundaries were determined from attribute tables or by manually measuring the areas on maps.We reexamined the early maps depicting the glaciers at scales of 1 : 50,000 and 1 : 100,000 (Table 1 [25].Detailed information from the seven-parameter transformation was shown by Wang et al. [26].The seven parameters of each map were calculated utilizing national trigonometric reference points.Errors introduced by this method were shown to be no greater than 0.5 meters [6,27,28].The Landsat MSS images were orthorectified by USGS using the Universal Transverse Mercator coordinate system referenced to the WGS84 ellipsoidal elevation.Finally, glacier boundaries on maps and MSS images were digitized manually into a shape file format using ArcMap software.Glaciers were white and easy to distinguish from other surface features in bands 4, 3, and 2 of the composite MSS images.Glaciers that measured at least 0.01 km 2 were included in this study.Detailed information was found in 1 : 50,000 topographic maps for 30% of the glaciers, in 1 : 100,000 maps for 35% of the glaciers, and in MSS images for the remaining 35% of the glaciers, which were larger than 2 km 2 .These glaciers formed the basic data for the changes in glaciers in the source region of the Ili River.

Glaciers in 2007 and 2009.
To develop new glacier inventories and up-to-date satellite images of the Ili River watershed, this study acquired eight Landsat TM scenes from the USGS (https://lpdaac.usgs.gov/getdata/) that had spatial resolutions of 30 meters (Table 1).All Landsat scenes were orthorectified in level 1T by the USGS using global ground control points (GCPs) from Global Land Survey 2005 datasets.The GCP report files for TM images were available, but those for MSS images were not.Table 2 shows the GCP residuals, which have a maximum of 4.6 m and a mean of 4.1 m.Where there were no GCP files for MSS data, we compared tie points of MSS images with TM images to assess the accuracy.The mean error was 13.2 m.Methods for glacier mapping using Landsat data include (1) manual delineation of glacier outlines on false color image composites [16,29], (2) segmentation of ratio images with reflectance thresholds [30], (3) the normalized-difference snow index (NDSI) [31,32], and (4) supervised classification techniques [33].In this study, ratio images of the raw digital numbers from TM band 3 and TM band 5 with a threshold of 2.1 were used to obtain a surface mask for the glaciers in the Ili River basin.This thematic method has been commonly used and proven to be effective for extracting glacial information [30][31][32]34].The drawbacks of this method are its inability to identify glacier debris cover and its inability to determine portions of the glaciers covered by debris, which must be done by manual digitizing.Current automatic methods for delineating glacier debris cover continue to need manual modifications [35][36][37][38].For the 80 glaciers in the Ili watershed that had debris covering their terminus areas, manual delineation with Landsat TM was applied to extract debris boundaries.Manual digitization of debriscovered glaciers was based mainly on the recognition of distinctive surface features such as supraglacial lakes and differences in colors and textures of the glacial surface in various red-green-blue composites of Landsat images.These special features made it easier to differentiate debris from surrounding periglacial landforms [39].This manual digitization was performed by experienced glaciologists.The debris terminal ended at the outlets of subglacial streams near glacial termini.Finally, the resultant contiguous ice coverage was further divided into individual glacier polygons, based on topographical ridgelines, using ASTER GDEM data (http://gdem.ersdac.jspacesystems.or.jp/) [40].
The debris cover percentage used hereinafter was calculated based on current glaciers.Our purpose was not to study the changes of debris cover; thus the debris cover area in the 1960s was not extracted.

Error Assessment
Two methods are used to assess the error in measurements of glacial changes: one based primarily on image resolutions and coregistration accuracies of multi-images [41] and the other on calculating the uncertainty of glacial data using given buffers [42,43].The latter was applied in this study because the former lacks assessment of errors in the debris-covered areas.
The 1960s glaciers were extracted from topographic maps and Landsat MSS images.Thus, the uncertainty which arises during glacial boundary delineation needed to be assessed in terms of the accuracy of both map data and MSS images.In topographic maps, the accuracy of glacial boundary delineation is dependent on scanning resolution, map quality, and map scales.Previous research demonstrated that the accuracy of these maps can be further confirmed using data from a global positioning satellite (GPS) [25,44].However, instead of using GPS data, we used the root mean square of geometrical corrections and the digitizing accuracy to confirm glacier boundaries.The root mean square for each map was calculated from ground control points (GCPs), and mean errors of 4.2 m and 9.8 m were found for the 1 : 50,000 and 1 : 100,000 maps, respectively.When digitizing maps, the error is around one pixel, or 5 m and 10 m for the 1 : 50,000 and 1 : 100,000 maps, respectively [45].We applied the law of error (1) and found that the total errors were 6.5 m and 14.2 m for the 1 : 50,000 and 1 : 100,000 maps, respectively In satellite images, the main factors that cause uncertainty in glacier delineation include cloud and snow cover, the sensor spatial resolution, and mountain shadows [6,43,46].Moreover, extracting glacier boundaries also brings about errors [34,47].No images with clouds and seasonal snow were used in our study except for one MSS image that showed a 1% cover of seasonal snow.We modified the shadow areas using images from Google Earth by the Export to KML tool.In addition, the method of segmentation using TM band 3 and TM band 5 could also partially discern shadows [32].After we performed these corrections, hindrances from clouds, snow, and shadows became negligible.Accuracy of less than onehalf pixel was usually achievable for glaciers without debris, cloud, and seasonal snow cover [42].Thus, we determined the uncertainty to be 28.5 m for Landsat MSS images and 14.3 m for TM images in clean ice areas.However, the error caused by debris cover was larger [39].Our partner compared the debris boundary with GPS points and found that the error in debris-covered ice was 31.3 m in Landsat TM images [48], which is nearly one pixel.Thus, a one-pixel buffer was used for debris-covered glaciers in TM images and 57 m for glaciers in MSS data.This analysis determined an uncertainty of 7.6% for glaciers mapped using data from the 1960s and 6% for glaciers mapped using TM data from 2007 and 2009.

Glaciers in the 1960s.
The first inventory of glaciers, which took place in the 1960s, found 2373 glaciers with a total glacial area of 2022.7 km 2 .Our more accurate study of the 1960s maps resulted in identification of 2119 glaciers with a total area of 2002.9 ± 152.2 km 2 .Details of the differences are shown in Table 3.The large difference in both glacier numbers and areas as a result of this update is probably a consequence of error in the 1 : 1,000,000 maps that were used by the FCI.Some glacial areas were likely missed while some seasonal snow was mistaken for glacier ice.There are no other data available to validate the accuracy of the FCI, and we used both the topographic maps and MSS to cross-validate the glaciers found in the 1960s.The biggest difference was found in the area of glacier 5X046G0048, which had originally been inaccurately measured to be 36.5 km 2 using the 1 : 1,000,000 scale maps.Using the MSS image, its area was found to be 88.9 km 2 (Figure 2), making it the largest glacier in the study area.Another problem identified in the 1960s inventory was that 33.5 km 2 was originally described as glacier ice, but it was actually seasonal snow.In addition, 9.3 km 2 was not recognized as glacier ice.
The corrected 1960s maps showed that median glacial elevation ranged from 3158 to 4567 m asl (Figure 3), and the lowest elevation of a glacier's terminus was 2635 m asl; 97% of the total glacial area was located between 3500 and 4200 m asl.
In the 1960s, there were 23 glaciers in the Ili River basin larger than 10 km 2 .Although 96.7% of the glaciers in the basin would be considered small (≤5 km 2 ), these made up 57.6% of the total glacial area in the basin (Table 3).There were only 80 debris-covered glaciers in the basin, but these glaciers accounted for 34% of the total glacial area (686.7 km 2 ) and included the two largest glaciers in the basin.The surfaces of individual glaciers were 1.3% to 26.4% covered in debris.Debris covered 16.2% of the surfaces of the two largest glaciers.Most of the debris cover was found between 3500 m and 4300 m asl (Figure 3), and the peak debris area was found between 3900 and 4000 m asl.

Glacial Changes.
When comparing glacial area from the 1960s to 2007 and 2009, we found that almost all those in the Ili River basin were receding; three small glaciers remained stable.During the same time period, the total glacial area decreased by 485 ± 177.3 km 2 , or 24.2% ± 8.8% (∼0.6% ⋅ a −1 ).
The number of glaciers decreased from 2119 in the 1960s to 1813 in the late 2000s.During the same period, the mean glacial size decreased from 0.94 km 2 to 0.72 km 2 .The speed of glacial retreat appeared to be dependent on the subbasin in which it was located (Table 4).The most drastic shrinkages were 47.2% ± 5.7% in the Kunes river subbasin and 42.6% ± 9.4% in the Huoerguosi River.Glaciers in the Kax River and Kuksu River subbasins experienced moderate recession rates of 27.5% ± 5.0% and 29.2% ± 9.0%, respectively.The least amount of shrinkage, found in the Tekes River subbasin, was 18.3% ± 10.8%.
Between the 1960s and the late 2000s, 331 glaciers disappeared completely, and 18 glaciers disintegrated into two or three smaller glaciers.All those that disappeared were debrisfree.The area of these glaciers was 34.8 km 2 , accounting for 1.7% of the total glacial area in the Ili River basin.Of those that disappeared, 59.4% were smaller than 0.1 km 2 , 30.3% were between 0.1 km 2 and 0.2 km 2 , and the remaining 10% were larger than 0.2 km 2 .A scatter plot of the median altitude versus glacial area of the disappearing glaciers is shown in Figure 4. Most were at around 3100 to 4200 m asl, and the median altitude decreased linearly with glacial area.Furthermore, nearly 60% of them were smaller than 0.1 km 2 , independent of their locations or altitudes.However, the larger ones (>0.3 km 2 ) were generally at very low altitudes  (lower than 3600 m asl), indicating that these glaciers disappeared because of warming temperatures.Disintegrated glaciers (Table 5), those that broke up into smaller pieces, existed in all subbasins except for the Kunes River basin.Seventeen broke up into two glaciers, and one broke up into three glaciers.Disintegration occurred at elevations between 3015 m and 3970 m, at the glacier terminal or in the upper branch joints.Most (14) of them disintegrated between 3400 and 3800 m elevation.

Changes in Glacial Area Related to Topographic Variables.
To analyze the effect of topography on changes in surface area, the median elevation and mean surface slope of each glacier in the basin were calculated using SRTM DEM.It is necessary to point out that ASTER GDEM is not suited for topographic factor extraction on a changing glacial surface because its obtained date is uncertain.Figure 5(a) shows the results of comparing glacial shrinkage by dividing all glaciers into one of eight groups according to mean surface slope.The groups were divided by five-degree increments (e.g., one division might represent glaciers with 5 ∘ to 10 ∘ mean surface slope).were divided into 15 100-meter elevation groups.For example, all glaciers with a median surface elevation between 2900 and 3000 m were grouped together in the 3000 m division.Figure 5(c) groups glaciers of similar size to compare glacial size to shrinkage.
Figure 5(a) shows that shrinkage was greater for glaciers smaller than 0.08 km 2 .It also shows that the slopes of large glaciers (>0.08 km 2 ) had nearly no effect on shrinkage.Glaciers smaller than 0.08 km 2 showed no relationship between glacial area and shrinkage.
Figure 5(b) shows a logarithmic relationship between shrinkage and area for glaciers at similar median elevations.For example, at 3700 m median elevation, large (>1 km 2 ) glaciers experienced less shrinkage than smaller ones.Figure 5(c) shows that the relationship between glacial shrinkage and area is more pronounced for those larger than 1 km 2 .It also indicates that shrinkage in small glaciers (<1 km 2 ) was greater at lower elevations whereas shrinkage in larger glaciers was less affected by elevation.Our results concur with previous research which found that lower median elevation resulted in relatively more loss in area [42].Bolch's study did not remove the impact of glacial size, and the exact relationship was not as distinct as was found in our study [42].
In summary, in glaciers smaller than 0.08 km 2 , glacial recession appeared to be primarily controlled by median elevation (Figures 5(a Table 6 shows that the majority of overall glacial area is concentrated in glaciers larger than 1 km 2 .Glaciers larger than 5 km 2 accounted for over half the total glacial area in the Tekes River subbasin.Glaciers larger than 1 km 2 accounted for 74.3% of the total glacial area in the Kax River subbasin and 70% of the total glacial area in the Kuksu River subbasin.Of the five subbasins, the Kunes River basin held the smallest glaciers.The distribution of glacial area in the five subbasins was consistent with their glacial changes.The larger the proportion of large glaciers (>10 km 2 ), the smaller the glacial area loss.For example, glaciers in the Tekes River basin, with the largest proportion of glaciers larger than 10 km 2 , had the smallest area shrinkage.Consequently, the distribution of glacial area is probably one of the main factors of change for the basin scale.
Most glaciers at lower elevations disappeared during the study period.For example, 80% of those at a median elevation of less than 3400 m disappeared (Figure 6).Through regression analysis, Figure 6 demonstrates an exponential relationship between the percentage of disappearing glaciers and median glacial elevation.All glaciers below 3300 m asl disappeared, and 38% of those between 3300 m and 3400 m disappeared.Only 1% of the glaciers at a median elevation of 3800 m disappeared.As expected, smaller glaciers at lower elevations were the most likely to lose area or disappear.with debris.Compared with debris-free glaciers, these shrunk less (Figure 7(a)).The area lost from debris-covered glaciers was between 7.5% and 20%, significantly less than the area lost from debris-free glaciers.Area loss by debris-covered glaciers is smaller by 2.5% to 7.5% than clean ice of the same size.This difference becomes smaller in large glaciers.The general result is that glaciers with debris cover greater than 25% lost less surface area than the others (Figure 7(b)).

Discussion
There were no advancing or surging glaciers in the Ili River basin during the measurement periods.It is different from the Aksu catchment in the central Tian Shan Mountains where 10 advancing glaciers were found between 1990 and 2010 [49].Our results are similar to those found for glaciers in the western Nyainqentanglha Range and the Nam Co Basin, Tibet, where also no advancing/surging was found from 1976 to 2009 [42].Sorg et al. reported that, of the entire Tian Shan range, the outer ranges had the strongest annual area shrinkage rates since the middle of the twentieth century; they were between 0.38% and 0.76% a −1 [50].Thus, glaciers in the Ili River basin should show the greatest area shrinkage (0.6% a −1 ) of the range, the same as in the glaciers of Big Naryn basin (central Tian Shan), where the area shrinkage was 23.4% (0.6% a −1 ) from the mid-20th century to 2007.In many mountainous regions, various glaciers grow and shrink at different rates, and some glaciers will advance or surge while others shrink [6,20,31].Some studies pointed out that glaciers of different sizes respond to climate change differently because of the time lag [17,42,51].Previous studies found that larger glaciers tended to shrink less, as a percentage of total area, than smaller glaciers [42,51,52]; our results agree.These studies also implied that small glaciers (<0.4 km 2 ) were easily affected by topography, and glaciers that shrank in area or disappeared had lower mean elevations [53].However, this research showed only the relationship between topographic settings and glaciers smaller than 0.4 km 2 .Our results not only showed that changes in small glaciers (<1 km 2 ) are affected by topography and size together, but also showed that larger glaciers were less affected by topographic factors.These diverse factors result in a real world situation that is difficult to simulate with hydrological or land surface models.
Debris cover affects the response of glaciers to climate change by altering surface ablation rates [54,55].Previous research pointed out that heavily debris-covered glaciers, with stagnant low-gradient terminus regions, typically had stable fronts in the Himalayas [20].Our results agree that debris-covered glaciers have less shrinkage than debris-free glaciers.Thus, we suggest that the vast debris cover of glaciers in the Tekes River basin is one factor that was responsible for the lower shrinkage seen there rather than in other basins.However, previous studies of debris-covered glaciers on Tomur Peak showed that thinning of debris-covered ice was not statistically different from that of debris-free ice [19,56].This should not conflict with our results.Studies have shown that ice cliffs and supraglacial lakes always develop in debris-covered areas, and these cliffs result in 69% of the melting of debris-covered areas in the Himalayas [57].These quick ablation ice cliffs could cause extensive thinning of debris-covered ice, but the percentage of ice cliffs is small, and they can not affect the overall area shrinkage seen in debriscovered glaciers.This calls for studies of ice thickness changes in debris-covered and debris-free glaciers while observing ice cliff distribution in debris areas.
Research suggests that glacial recession is mainly influenced by global warming [42,44,49,52].Meteorological data obtained from the Zhaosu weather station (43.15 ∘ N, 81.133 ∘ E; 1851 m asl) were used to analyze the drivers of change to the glaciers in the Ili River basin (Figure 8).This station recorded a significant temperature increase between 1956 and 2012.Temperature has risen by 1.4 ∘ C in the past 56 years and the warming tendency has accelerated, as evidenced by the decadal average temperatures.Precipitation has slightly increased during the same period but decreased between 1978 and 1998.Thus, we suggest that glacial recession between the 1960s and 2007/2009 can be attributed primarily to air-temperature increases, even though precipitation slightly increased at the same time.This greater precipitation could not offset ice losses resulting from surface melting driven by higher temperatures.

Summary and Conclusions
This study used multitemporal remote sensing and historical topographic maps to analyze glacial changes in the area of the Ili River.Our results updated the first glacier inventory from the 1960s by employing additional sources of data and enhanced analytical methods.The use of satellite data revealed that glaciers in the study area shrunk by 24.2% ± 8.8% between the 1960s and 2007/2009.No expanding or advancing glaciers were identified in the study area during the same time period.The greatest glacial shrinkage occurred in the Kunes River basin, which contained most of the smaller glaciers.During the 40-year study period, 331 glaciers disappeared, and 18 disintegrated into two or three glaciers.

Figure 1 :
Figure 1: Glaciated areas in the Ili River basin.

Figure 3 :
Figure 3: Proportion of glacial area covered by debris at varying elevations per 2007/2009 mapping data.

Figure 4 :
Figure 4: Scatter plot showing the median altitude versus area of disappearing glaciers.

Figure 5 (Figure 5 :
Figure 5: Glacial change in area (%) versus mean slope, median elevation, and glacial area.(a) Glacial mean slope plotted as a function of glacial shrinkage and area, (b) glacial median elevation plotted as a function of glacial shrinkage and area, and (c) glacial area plotted as a function of glacial shrinkage and median elevation.
Figure 5(c)  shows that the relationship between glacial shrinkage and area is more pronounced for those larger than 1 km 2 .It also indicates that shrinkage in small glaciers (<1 km2 ) was greater at lower elevations whereas shrinkage in larger glaciers was less affected by elevation.Our results concur with previous research which found that lower median elevation resulted in relatively more loss in area[42].Bolch's study did not remove the impact of glacial size, and the exact relationship was not as distinct as was found in our study[42].In summary, in glaciers smaller than 0.08 km 2 , glacial recession appeared to be primarily controlled by median elevation (Figures5(a) and 5(c)).Those between 0.08 km 2 and 1 km 2 showed a linear relationship between size and median elevation (Figures 5(b) and 5(c)).Shrinkage was not affected by median elevation for glaciers larger than 1 km 2 (Figures 5(b) and 5(c)).Table6shows that the majority of overall glacial area is concentrated in glaciers larger than 1 km 2 .Glaciers larger than 5 km 2 accounted for over half the total glacial area in the Tekes River subbasin.Glaciers larger than 1 km 2 accounted for 74.3% of the total glacial area in the Kax River subbasin and 70% of the total glacial area in the Kuksu River subbasin.Of the five subbasins, the Kunes River basin held the smallest glaciers.The distribution of glacial area in the five subbasins was consistent with their glacial changes.The larger the proportion of large glaciers (>10 km 2 ), the smaller the glacial area loss.For example, glaciers in the Tekes River basin, with the largest proportion of glaciers larger than 10 km 2 , had the smallest area shrinkage.Consequently, the distribution of glacial area is probably one of the main factors of change for the basin scale.Most glaciers at lower elevations disappeared during the study period.For example, 80% of those at a median elevation of less than 3400 m disappeared (Figure6).Through regression analysis, Figure6demonstrates an exponential relationship between the percentage of disappearing glaciers and median glacial elevation.All glaciers below 3300 m asl disappeared, and 38% of those between 3300 m and 3400 m disappeared.Only 1% of the glaciers at a median elevation of 3800 m disappeared.As expected, smaller glaciers at lower elevations were the most likely to lose area or disappear.

Figure 6 :
Figure 6: Percentage of glacial area lost by glaciers that disappeared as a function of median glacial elevation.

Figure 7 :
Figure 7: Area change (%) of debris-covered glaciers.(a) Percent area shrinkage of glaciers of various sizes, debris-covered and debrisfree, and (b) percent decrease in glacial area as a function of the percentage of the glacier covered in debris.

Figure 8 :
Figure 8: Changes in annual air temperature and precipitation between 1956 and 2012.

Table 1 :
Topographic maps list for the study area.

Table 2 :
Landsat images list for the study area.

Table 3 :
The differences in glaciers before and after update.
E Figure 2: Glaciers in the 1960s before and after update.Glacier 5X046G0048 was in the black boundary after update.

Table 4 :
Glacial area changes in five tributary subbasins of the Ili River.

Table 5 :
Details regarding the 18 glaciers that disintegrated.

Table 6 :
Area percentage of each area group in all subbasins in the 1960s.