REMOTELY SENSED DATA AND IMAGE PROCESSING

Two SPOT-5 scenes used to map thermokarst lakes in the study region were acquired in October 2010. These images have a spatial resolution of 10 m. The SPOT-5 images were orthorectified using ENVI 4.6 image processing software. Ground control points were selected from the 1:100,000 topographic map or by manual selection in the field using the high-precision Global Positioning System (GPS). The selection of ground control points ensured that (1) the points were easily recognizable (e.g., road intersections and building boundaries), (2) there were an adequate number of evenly distributed points across the study region, and (3) the points did not change over time.

The image was mosaicked using the image Mosaic module of ENVI 4.6, based on the geographic coordinate mosaic method. We chose the edge feathering method that uses 50% of the hybrid to fusion by input specified distance value on the edge of the top image and bottom image. The overlap of roads and rivers in the mosaicked images matched very well, and no gaps existed between different images, which demonstrated the high precision of the image mosaic.

Since the ground cover on QTP is relatively uniform, the Kmeans unsupervised classification in ENVI 4.6 was used to delineate water bodies in the study area. Water bodies can be easily distinguished from other land cover types, such as vegetation and bare ground, because they are strong absorbers of infrared wavelengths. Therefore, we applied band 4 (SWIR: 1.58–1.75 μm) of the SPOT-5 multispectral images for the water body classification. We then compared classification results with the actual distribution of water bodies. We found the most accurate classification was produced when six classes were used with a change threshold of 13. Rivers and small streams were manually removed from the classification image, leaving only the thermokarst lakes shown in Figure 1. To ensure the accuracy of the classification results, we compared them with field investigation results from September 2011. From a randomly selected group of 80 lakes, 78 were interpreted correctly as thermokarst lakes (97.5%). The other two lakes were very small and dried up in September 2011.

MORPHOMETRIC ANALYSES

As the extracted water body was raster data, it was converted into vector polygons, and the subsequent data processing and analyses were performed in ArcGIS 9.3. We calculated morphometric variables including area, perimeter, circularity index, elongation index, and the orientation of the major axis for lakes with a circularity index below 0.95 using the methodology described in Morgenstern et al. (2011) (Table 2). In addition, we calculated the curvature of lakeshores for 30 randomly selected lakes in each subregion to characterize the lake shoreline sinuosity. For each of these lakes, we calculated the curvature of 2000 discrete points and used the average value to represent the curvature of the shoreline. To analyze the distributions of the morphometric variables, we used the Kolmogorov-Smirnov test in SPSS 17.0. The results indicated that none of the morphometric variables were normally distributed. Therefore, the rank-based Kruskal-Wallis test was used to illustrate the lake morphometric differences among the three subregions, and the Mann-Whitney U test was used to further examine differences between pairs of subregions.

FIGURE 1.

Location of the study area on the Qinghai-Tibet Plateau. The interpreted thermokarst lakes and the sites of meteorological stations are also shown in this figure. (Revised from two SPOT-5 multispectral images with spatial resolution of 10 m, National Centre for Space Studies, France).

FIGURE 2.

(a) Frequency distribution of summer wind direction in the study area. Frequency distributions of major axis orientation for (b) lakes <5000 m² in the study area, (c) lakes >5000 m² in the study area, (d) lakes >5000 m² in Chumaerhe High Plain (CHP), (e) lakes >5000 m² in Hoh Xil Hill region (HXR), and (f) lakes >5000 m² in Beluhe Basin region (BBR). 0° = North, 90° = East, 180° = South, 270° = West. (The meteorological data in CHP was provided by the Key Laboratory of Plateau Atmospheric Physics Research in Cold and Arid Regions, and the meteorological data in BBR was provided by the Beiluhe Research Station of Cold and Arid Regions Environmental and Engineering Research Institute.)

LAKE DEPTH PROFILES

Geophysical methods have been widely used to identify physical structures of the subsurface because of the close relationships between physical properties of geologic materials and their electromagnetic signatures (Davis and Annan, 1989; Dallimore and Davis, 1987). The difference between the electromagnetic properties of ice, water, and the sediments under lake bottoms makes GPR an effective method for determining lake depth profiles in the field.

The GPR unit used in this study was a pulseEKKO PRO, manufactured by SSI Inc., Canada. Based on considerations of the desired resolution and the depths of lakes in the study area, a 100 MHz antenna was employed. The velocity of pulse propagation was determined by common midpoint (CMP) velocity surveys and verified with a borehole drilled through the lake ice. The EKKO View Deluxe software was used for editing and plotting the data. The bathymetric profiles of lake bottoms were investigated in detail using the GPR. We first measured the depths of 50 randomly selected lakes in CHP, 35 in HXR, and 45 in BBR by boring a hole through the lake ice. Based on the measured lake depths and lake area data acquired from the SPOT-5 images, we selected 15 lakes in January 2013 and an additional 3 lakes in April 2013 with different lake depth and lake size range for the GPR investigations. Of the 18 lakes, lakes 1–7 are in CHP, lakes 8–12 are in HXR, and lakes 13–18 are in BBR.

TABLE 1

Characteristics of the three subregions. (MAGT refer to mean annual ground temperature at the depth of -15 m). CHP = Chumarhe High Plateau; HXR = Hoh Xil hill region; BBR = Beiluhe basin region.

TABLE 2

Overview of morphometric variables calculated for the thermokarst lakes (after Morgenstern et al., 2011).

ALL LAKES IN THE STUDY AREA

We classified 2163 lakes with a total area of 1.09 × 10⁷ m² within the study area. The largest lake was 4.49 × 10⁵ m² and the smallest was 100 m², which corresponds to the area of one pixel of SPOT-5 imagery. The frequency distribution of the areas of all lakes (Fig. 3) is strongly skewed toward lower values, indicating that small lakes are much more abundant than larger ones.

In our field investigation, we found that many small lakes were very shallow and commonly disappeared during the dry season. Over 90% of these lakes have an area less than 5000 m² and did not possess a retrogressive lakeshore caused by thermal erosion. Inclusion of these small lakes would have dramatically skewed the results of morphometric analyses in the study area. Therefore, the lakes were classified into two size groups according to the mean area (5039 m²), either >5000 m² (large) or <5000 m² (small). The descriptive statistics for the shape metrics of these two groups revealed significant differences in lake morphometry (Table 3). The mean circularity index and elongation index of large lakes were 0.57 and 1.39, respectively, and 0.69 and 1.18 for small lakes. This indicates that the large lakes were more elongated than the small lakes. Fig. 2, parts b and c, show the frequency distributions of major axis orientations for the two lake size groups. The major axis orientation of lakes larger than 5000 m² shows a significant ENE-WSW peak, while the orientation of lakes smaller than 5000 m² is nearly uniformly distributed in every direction. This may be because the small lakes do not enlarge appreciably through preferential erosion of the lakeshore.

REGIONAL ANALYSIS

Descriptive statistics for thermokarst lakes in the three subregions (CHP, HXR, and BBR) are presented in Table 4. These regions occupied different proportions of the study area: 41% is in CHP, 28% is in HXR, and 31% is in BBR. The summary statistics show that thermokarst lakes cover about 1.3% of the total area of CHP, 0.6% of HXR, and 1.6% of BBR (Table 4). Lake density (number of lakes per km²) in HXR and BBR was 2.1 and 2.5, respectively, while in CHP it was 3.5. These data indicated that BBR had the highest lake area coverage, while CHP had the highest lake density.

The morphometric variables of thermokarst lakes varied largely between the three subregions (Table 5), particularly for the large lakes. The Kruskal-Wallis test indicated that the lakes larger than 5000 m² had significantly different areas and perimeters (p < 0.005) in the three subregions (Table 6). There were no significant differences in area or perimeter for the lakes smaller than 5000 m² (Table 6). The Mann-Whitney U test further illustrated that the differences primarily existed between BBR and the other two regions (CHP and HXR), while there was no significant difference between CHP and HXR (Table 7). The detailed metrics listed in Table 5 show that the lake areas and perimeters in BBR were much greater than in CHP and HXR.

FIGURE 3.

Frequency diagram of all lakes in the study area.

TABLE 3

Descriptive statistics of thermokarst lakes in the study area.

TABLE 4

Descriptive statistics of the thermokarst lakes in the three subregions.

The Kruskall-Wallis test also revealed significant differences (p < 0.01) in the circularity and elongation indices of large lakes between the three subregions (Table 6). These differences mainly existed between BBR and the other two regions (Table 7). The results of these analyses, and the metrics provided in Table 5 indicated that the thermokarst lakes are more elongated in BBR than the other two regions.

Significant differences (p < 0.05) in major axis orientation also existed for large lakes between the three subregions (Table 6). Figure 2 (parts d-f) shows the frequency distribution of major axis orientation for lakes with areas larger than 5000 m² in the three subregions. These results indicate that the large lakes in the study area shared the same peak in major axis orientation (ENE-WSW), albeit with different frequencies. Lakes in CHP were only strongly oriented ENE-WSW. In HXR, there were two small peaks in the NW-SE and N-S directions, as well as a major ENE-WSW peak. The major axis orientation of lakes in BBR usually had a weak NW-SE peak and the major ENE-WSW peak.

The depths of lakes based on the GPR data (Fig. 4) and manual measurements are shown in Table 5. The results suggest that the lakes in HXR were deeper than the lakes in BBR and CHP. The maximum recorded depth for lakes in the CHP was 1.97 m, while the maximum depth was 2.95 m in HXR. The depths of lakes in BBR (maximum: 2.16 m; minimum: 0.91 m) falls between those observed in CHR and HXR. The ice thickness on the lakes in winter was also obtained from the GPR profiles. The maximum ice thickness was 0.63 m in CHP, 0.53 m in HXR, and 0.52 m in BBR (Table 5).

TABLE 5

Descriptive statistics of morphometric variables of thermokarst lakes >5000 m² in the three subregions (morphometric variables of thermokarst lakes <5000 m² are also shown in brackets). “Normal” refers to the Shapiro-Wilk test for distribution normality at the 0.10 significance level; n = non-normal; N refers to lake number.

The differences of bathymetric profiles in the three subregions are shown in Figure 5. All of the seven selected lakes in CHP had a smooth and regular lake bottom, while the profiles are more complicated in HXR and BBR, with the exception of lake 11 in HXR and lake 18 in BBR. In addition, the curvature of shorelines in HXR was significantly greater than those in CHP and BBR (Table 5), which indicates that lakes in HXR had more complex shorelines than lakes in CHP and BBR.

In summary, significant spatial differences existed in shape metrics of thermokarst lakes in the study area. In CHP, the lakes typically had more regular shoreline shapes and smoother lake bottoms, but they were also smaller and shallower than in the other two regions. In HXR, the lakes were commonly deeper, and had more complex shorelines and lake bottom profiles. In BBR, the lakes were bigger and were usually more elongated than in the other two regions.

TABLE 6

Results of rank-based Kruskal-Wallis test for morphometric variables of lakes >5000 m² and <5000 m² among the three subregions.

TABLE 7

Results of rank-based Mann-Whitney-U test for morphometric variables of lakes >5000 m² between pairs of subregions.

REGIONAL DIFFERENCES IN LAKE MORPHOMETRY

Significant differences existed in the morphometric characteristics of large thermokarst lakes in the three subregions (Table 6). Some studies (Sellmann et al., 1975; Hinkel et al., 2005) on the Arctic coastal plain of Alaska have noted that lake size and shape are closely related with ground-ice content, local relief, and surface age. However, the factors contributing to the differences in morphometric characteristics of thermokarst lakes on QTP are not clear, so here we discuss the relations observed in our investigation.

The formation of large thermokarst lakes in permafrost regions requires sediments with high ice contents, and low surface relief. The thaw of ice-rich ground causes subsidence and encourages lake development in resulting depressions. Flat ground limits drainage and allows water bodies to persist. Compared to HXR, the flat terrain of CHP and BBR provides more favorable conditions for the development of large lakes in these two regions. However, the permafrost in BBR has higher ice contents than in CHP, which contributes to the formation of larger lakes in BBR.

FIGURE 4.

Ground penetrating radar profile of lake-13 in BBR.

FIGURE 5.

The bathymetric profiles of lake bottom in the study area. Lakes No.1–7 are in CHP, lakes No.8–12 are in HXR, and No.13–18 are in BBR.

The shorelines in HXR are much more sinuous than in the other two regions (Table 5). The majority of the thermokarst lakes in HXR are located in lowland areas of the piedmont region, since the steeper terrain does not allow large water bodies to form. The lowland areas in HXR have irregular, rolling topography, which may lead to the complex lakeshores observed in the region. In contrast, the terrain in CHP and BBR is flat, and shoreline morphology is mainly controlled by shoreline retrogression along the oriented axes rather than the local topography. Therefore, the lakeshores in CHP and BBR are more rounded than in HXR.

According to Sellmann et al. (1975), elongated lakes are more likely to be found in low-relief terrain with high ground-ice content. The thaw of ice-rich terrain allows continual lakeshore enlargement, while low-relief terrain ensures that the elongation of the lakeshore is not affected by surrounding topography. HXR commonly has high ground-ice contents and rolling topography, while CHP usually has low-relief terrain and relatively low ground-ice content, so the most elongated lakes were most commonly found in BBR, which had both high ice content and low-relief terrain. The circularity index of elongated lakes is lower than that of nearcircular lakes, so the values of this index in CHP and HXR were much higher than in BBR.

The deepest lakes were located in HXR, which may be due to the high ground-ice contents and rolling topography in this region. The GPR profiles indicated that lakes in CHP usually have smoother bottoms than in HXR and BBR. This is likely due to the sandy substrate and sparse vegetation cover in the region. Strong winds commonly deposit large volumes of sand into the lakes, contributing to the smooth lake bottoms in the region. The lakes in HXR and BBR were mainly distributed in alpine meadow areas where the vegetation cover is greater than 30%, so the winds cannot carry large amount of sediment into the lakes. As a result, the lake bottoms maintain their original morphology.

THE DEVELOPMENT OF ORIENTED THERMOKARST LAKES

Although the large (>5000 m²) thermokarst lakes in this study are not typical of oriented lakes that have been described in the Arctic Coastal Plain of northern Alaska (Black and Barksdale, 1949; Sellmann et al., 1975; Hinkel et al., 2005), they still show distinct oriented patterns. There are many postulated reasons for the formation of the oriented lakes in permafrost regions. The following discussion presents the apparent controls on lake orientation in the study region.

The prevailing summer winds in the study area are ENEWSW (Fig. 2, part a), parallel to the major axis of oriented lakes. Therefore, the lake orientation in the study area may be the result of the long-term prevailing summer winds. According to observations from a field investigation of one oriented lake, enlargement occurs as the result of lakeshore collapse. However, when the collapsed materials covered the exposed ground ice, the rate of shoreline retrogression decreased. In summer, waves carry away the slumped materials and expose ground ice in the lakeshore that is perpendicular to the summer winds. This results in the lakeshore eroding more rapidly in the downwind direction, thereby forming oriented lakes with major axes parallel to the prevailing summer winds. This result is similar to those found in a study of oriented lakes in Northeastern European Russia (Sjoberg et al., 2012), but differs from results from several studies in the North American Arctic Coastal Plain, which suggested that lakes were oriented perpendicular to the prevailing summer wind directions because wind-driven circulation resulted in the thermo-mechanical erosion of ice-rich permafrost (Carson and Hussey, 1962; Cote and Burn, 2002; Hinkel et al., 2005).

On the QTP, abundant convective clouds usually occur between 12:00 and 18:00 (Ye and Gao, 1979; Zhang, 1988). Therefore, the highest incident insolation also occurs before noon, causing the NW (SE-facing) slopes to receive the most solar radiation (Sun et al., 2008). The lakes in HXR and BBR show a minor NW-SE peak in major axis orientation, which may be the result of the solar insolation distribution. Field investigations showed that more than 70% of such lakes are developed in regions with SE-facing slopes. In addition to the major ENEWSW peak in lake orientation and the minor NW-SE peak, lakes in HXR also had a weak N-S peak. This may be the result of other controlling factors such as preexisting microtopographic orientations (e.g., Payette et al., 2004) observed in sub-Arctic regions. Many rectangular borrow pits were excavated in HXR parallel to the Qinhai-Tibet Highway (QTH) during the highway construction. Since much of the highway runs N-S in HXR (Fig. 1), lakes that developed in these rectangular borrow pits usually have major axes in the same direction. Such lakes are commonly found along the QTH, but are quite small in CHP and BBR, with only a few larger examples in HXR having areas larger than 5000 m².

In summary, the orientation of large thermokarst lakes in most of the study region is associated with the prevailing summer winds, but solar insolation is also important in regions with SE-facing slopes. Preexisting microtopographic orientation, such as rectangular pits along the roadway, also influenced lake orientation in HXR, but was seldom a factor in the other regions.

THERMOKARST EXTENT AND POTENTIAL FUTURE LAKE DEVELOPMENT

Large thermokarst lakes (>5000 m²) covered only 1.0% of the study area. This coverage is much lower than the 20% lake coverage of the western Arctic coastal plain in Alaska (Hinkel et al., 2005), the 30% lake coverage of the Tuktoyaktuk Peninsula in Arctic Canada (Cote and Burn, 2002), or the 5.2% lake coverage on the Yedoma landscapes in the Lena Delta in Russia (Morgenstern et al., 2011). If lakes smaller than 5000 m² are included, the coverage on the QTP is still less than 1.5%. According to borehole data from the study region, ice-rich permafrost exists only from 2 to 4 m below the active layer, underlying the well-weathered sandstone and mudstone, which has an ice content less than 10%. In contrast, high ground-ice contents (>20% in volume) occur in the Arctic to depths of 10–20 m below the active layer (Brown et al., 1997). In addition, active layers in the Arctic are much shallower than on the QTP, so the ground ice there may be more easily exposed to initiate thermokarst lakes. These two factors offer an explanation for the low areal coverage of thermokarst lakes on the QTP.

Of the 700 large lakes in the study area, the majority were in undisturbed tundra surfaces, while a small proportion were located in old thermokarst lake basins. For the lakes developed in tundra surfaces outside older thermokarst basins, the surrounding ice-rich permafrost facilitated continuous slumping and retrogression of lakeshores, resulting in large lakes. Monitoring of one thermokarst lake in BBR that developed in undisturbed tundra indicated that the lakeshore retreated 5.2 m between 2007 and 2012. In contrast, lakes located in old thermokarst basins may not increase in size significantly because the basin sediments are ice poor. The high rate of evapotranspiration on the QTP may even cause some lakes to shrink. However, the gradual increase in precipitation (Wu et al., 2005; Ge et al., 2008), and the low permeability of the mudstone substrate may prevent the disappearance of lakes found in old thermokarst lake basins. The majority of the small lakes (<5000 m²) are situated on tundra surfaces outside older thermokarst basins. With climate warming and the associated thaw of permafrost (Cheng and Wu, 2007; Jin et al., 2006), these small lakes may enlarge over time. Therefore, if temperature and precipitation continue to increase, the number of thermokarst lakes and the area they cover may increase in the future.