Thermokarst Lake Changes in the Southern Fringe of Siberian Permafrost Region in Mongolia Using Corona, Landsat, and ALOS Satellite Imagery from 1962 to 2007

This study presents thermokarst lake changes at seven different sites in the continuous and isolated permafrost zones in Mongolia. Lakes larger than 0.1 ha were analyzed using Corona KH-4, KH-4A and KH-4B (1962-1968), Landsat ETM + (19992001), and ALOS/AVNIR-2 (2006-2007) satellite imagery. Between 1962 and 2007, the total number and area of lakes increased by +21% (347 to 420), and +7% (3680 ha to 3936 ha) in the continuous permafrost zone, respectively. These changes correspond to the appearance of 85 new lakes (166 ha) during the last 45 years. In contrast, lakes in the isolated permafrost zone have decreased by −42% (118 to 68) in number and −12% (422 ha to 371 ha) in area from 1962 to 2007. The changes in lake area and number are likely attributed to shifts in climate regimes and local permafrost conditions. Since 1962, the mean annual air temperature and potential evapotranspiration have increased significantly in the northern continuous permafrost zone compared to the southern isolated permafrost zone. Due to ongoing atmospheric warming without any significant trend in annual precipitation, patches of ice-rich subsurface have thawed, and the number and area of lakes have accordingly developed in the continuous permafrost zone. Shrinking of thermokarst lakes in the isolated permafrost zone may be due to disappearing permafrost, deepening of the active layer, and increased water loss through surface evaporation and subsurface drainage.


Introduction
Thermokarst lakes are common features in Arctic and sub-Arctic permafrost regions [1]- [3] and important indicators of permafrost degradation [4] [5]. These lakes are defined as lakes that generally occupy closed basins formed by the settlement of ground following thawing of ice-rich permafrost or melting of massive ice [6]. After a thermokarst lake has formed, the lake size may change due to continued permafrost thaw, variations in air temperature, potential evapotranspiration and precipitation [2] [7]- [10].
The factors controlling their dynamics on decadal scales have been controversial and still are an important topic in permafrost regions.
Most previous studies have investigated the dynamics of thermokarst lakes in the permafrost regions of Siberia, Canada and Alaska using remote sensing imagery with high and medium resolution [2] [3] [8]- [12]. For example, Jones et al. [3] documented that the number of water bodies (>0.1 ha) increased by 10.7% but that the total surface area decreased by 14.9% as result of partial drainage of a few large lakes in the northern portion of the Seward Peninsula of Alaska between 1950 and 2007. Riordan et al. [9] found that the total area and number of lakes decreased by 11.6% and 19% since 1950 respectively, although the amount of precipitation has been stable. They concluded that lake shrinkages are mostly due to a rapid increase in air temperature (i.e., potential evapotranspiration). Plug et al. [2], on the other hand, found that temporal variation of the lake area, which increased for 1978-1992, and decreased for 1992-2001, agreed with those of annual precipitation in the continuous permafrost zone in northwest Canada.
Labrecque et al. [11] also found two phases of significant changes in the lake area in Canadian Yukon Territory, an increasing trend with 1.6% between 1951 and 1972 when precipitation increased slightly, while a decreasing trend with 5% between 1973 and 2001. These studies clearly show the relationship between hydro-climatic changes and variations in lake area.
The spatial and temporal extent of permafrost, as an impermeable ground layer, can have a first order effect on thermokarst lake dynamics. Smith et al. [8] found that the area and number of lakes increased in the continuous permafrost zone while they decreased in the discontinuous, sporadic, and isolated permafrost zones between 1973 and 1998 in western Siberia. This phenomenon is attributed to thawing of near-surface ground ice in the continuous permafrost zone, promoting thermokarst lake formation, whereas in discontinuous, sporadic and isolated permafrost zones, losses in lake area and number are attributed to subsurface drainage facilitated by permafrost loss [7].
A similar geographical shift of permafrost zones from continuous in northern territories to sporadic in the southern region is seen in Mongolia. In addition, climatic gradients occur along latitude as well; it is, colder and wetter in the northern territory, and warmer and drier in the southern territory. Although this environmental gradient would be interesting for comprehensive analysis of the factors controlling dynamics of thermokarst lakes, which have extensively developed on the depressions and valleys in the Altai, Hangai, Hovsgol and Hentii Mountain regions [13]- [16], the spatiotemporal changes of these lakes have not been investigated in Mongolia so far.
The primary objective of this study is to provide quantitative information on the temporal and spatial changes of thermokarst lakes in Mongolia using a time series of high-resolution satellite imagery. The other objective is to address the effects of the long-term trends of hydro-climatic regimes and permafrost degradation on the areal changes of lakes.

Study Sites
Study sites are located in the southern fringe of Siberian permafrost regions in Mongolia. We selected four study sites in the northern continuous permafrost zone including Darkhad depression, Mungut river valley, Chuluut river valley, and Khongor-Ulun, while selecting three other sites Nalaikh depression, Galuut depression, and Erdene in the southern isolated permafrost zone ( Figure 1). The sites are referred herein as sites 1-7 respectively. A set of basic environmental characteristics at all study sites are summarized in Table 1. In addition, these sites contain long-term permafrost monitoring boreholes located for our study sites.
Permafrost exists in almost two thirds of Mongolia, predominantly in the Altai, Hangai, Hovsgol, and Hentii Mountain regions and the surrounding areas [17]. The region is characterized by mountain and arid land permafrost and the permafrost distribution changes according to geomorphological and microclimatic conditions. Periglacial features include ice wedge polygons, thermokarst lakes, and pingos which are extensively developed in the permafrost zones [13]- [16] [18] [19]. Continuous and icerich permafrost occurs in the northern areas with volumetric ice contents higher than 30% [16], while discontinuous, sporadic and isolated permafrost has low ice content in the southern areas. Permafrost temperatures generally range from −3˚C to −0.1˚C [17].
The average thickness and mean annual temperature in the areas of permafrost are 50 -100 m and −1 to −2˚C in river valleys and depressions, and 100 -250 m and −1 to −3°C on mountains [17]. The thickness of the active layer in the continuous permafrost zone is 1 -3 m, while it ranges 4 -7 m in the discontinuous and isolated permafrost zone [16]. The continuous permafrost in northern Mongolia has warmed faster than southern isolated permafrost zone over recent decades [17] [20] [21]. The borehole data (10 -15 m depth) in the continuous permafrost zone (sites 1 -4) has shown a temperature increase between 0.38 and 0.95˚C [21]. In contrast, the borehole temperature (10 -15 m depth) in the isolated permafrost zone (sites 5 -7) has only increased between 0.1 and 0.64˚C in the last 30 years [21].
Mongolia has a typical continental climate. The lowest air temperatures often reach −34˚C in mid-January in the northern regions, while it reaches −20˚C in the southern regions [22]. In the warmest months, air temperatures range between 10 and 15˚C in the northern territories such as Hovsgol, Hangai, Altai, and Hentii Mountain regions and reach 20˚C in the southern territory (Gobi Steppe). Precipitation during the summer season contributes to more than 80% of the total annual precipitation in Mongolia [23]. The mean annual total precipitation ranges between 150 -300 mm in the north, and 50 -100 mm in south area [22]. The territory, thus, has a transition zone in terms of rainfall amount as well as vegetation, which changes from desert to grassland and boreal forest, within only several hundreds of kilometers over a south-to-north distance.

Remote Sensing Data and Processing
To estimate thermokarst lake changes at study sites, we used satellite imagery for three different time series : 1962-1968, 1999-2001 and 2006-2007 satellite data sets ( Table 2).
The oldest data (1960s) is especially useful in developing countries where aerial photographs in wide area coverage are rarely available. Beside the presented technical advantages, the high resolution Corona satellite images are also available at a reasonable price and provide an excellent opportunity for change detection studies. The Corona declassified images (1962)(1963)(1964)(1965)(1966)(1967)(1968) were acquired from the US Geological Survey Earth Table 1. General environmental characteristics of study sites. Permafrost temperature, active layer, and ice content datasets were obtained from the articles of Sharkhuu [16], and Jambaljav et al. [17]. N/A is not available data. ALOS images were a maximum RMSE of less than 1 m. Figure 2 summarizes the methodological framework of satellite images in this study.

Thermokarst Lake Area Delineation and Analysis
Firstly, we attempted the automated classification of lake areas based on the orthorectified images. However, we abandoned the automated spectral approaches commonly used in digital image processing due to the issues where cloud shadows creating dark patches that were spectrally similar to water, and sun glint near the edges of these images creating bright small lakes were confused bright target such as meadows [1] [9]. We visually delineated all lake areas from each satellite image that were above a minimum area of 0.1 ha (1000 m 2 ). The shoreline of each lake was manually traced as a polygon area using ArcGIS 9.1 software. The areas of the extracted lake polygons were computed ( Figure 2). Finally, we categorized the extracted lake areas into four distinct classes (i.e., 0.1 -1 ha, 1 -10 ha, 10 -100 ha, and 100 -1000 ha) in order to better understand the lake dynamics of individual lake size categories. Furthermore, we removed very large lakes (e.g., Dood Nuur (4810 ha) and Targan (1962 ha) in site 1 (Figure 1)) from the recent analysis, which are not likely thermokarst lakes [13].

Hydro-Climate Data and Analysis
Since Mongolia has a sparse distribution of meteorological stations with weather records spanning more than 50 years, we used the reanalysis data to evaluate the correlation between thermokarst lake changes and hydro-climatic parameters.

Thermokarst Lake Changes from 1962 to 2007
Contrasting changes in thermokarst lake dynamics were observed between the continuous and isolated permafrost zones in Mongolia. For the continuous permafrost zone     (Table 1; e.g., ice-rich permafrost and shallow active layers [17]). Decrease in sites 2 and 3 is a result of large lakes become fragmented following partial drainage. These results demonstrate the importance of analyzing both lake number and area to provide a meaningful information for understanding thermokarst lake dynamics.
The total changes of thermokarst lakes subdivided by lake area size are shown in Ta As discussed above, the increases in number and area of thermokarst lakes have been widely reported in the continuous permafrost zone [8] [28], while other studies observed increases in lake areas due to increasing precipitation [2] [11]. This study did not find any such correlation between precipitation and lake changes in the continuous permafrost zone. Our results differ from those of Jones et al. [3] in terms of the direction of lake changes, as that study found an increase in the number of small lakes but total lake area decreased between 1950 and 2007 for Alaska. However, similarly to Smith [8], we also found the increasing the number and area of lakes in the continuous permafrost zone in Mongolia, which strongly suggests that this phenomenon is a result of local thawing of ice-rich permafrost [21]. The temperature data from continuous permafrost boreholes exhibited an increase in permafrost temperature in the northern continuous permafrost zone. In fact, since the 1980s, permafrost has been degrading more significantly in the northern continuous zone [20] [21]. Warming and thawing of permafrost is consistent with thermokarst lake changes observed in this study for the continuous permafrost zone.
For the isolated permafrost zone, only two lakes were observed in 1962/68 in site 5 and one of them drained before 1999/01 (Figure 4(b), and Table 3). In terms of lake area change, there was a decrease of −86% (6 ha) for this site (Figure 3). The observed lakes in site 6 decreased both in number (−41%) and area (−10%) from 1962 to 2007, respectively. In site 7, lake number shows a decrease of −62% between 1962/68 and 1999/01, but remained stable afterwards for 2006/07. However, lake area decreased by −33% in 1999/01 and −50% in 2006/07. In the isolated permafrost zone, the investigated thermokarst lakes have greatly decreased in number and area between 1962 and 2007 as result of lake drainage. Overall, a total of 118 lakes (422 ha) were delineated in the isolated permafrost zone for the period of 1962/68. The number of these lakes decreased by −35% (77 lakes) in 1999/01, and −42% (68 lakes) in 2006/07. The total area of lakes in this permafrost zone decreased by −12% (from 422 ha to 371 ha) during the last 45 years.
In the isolated permafrost zone, there is a predominance of smaller lakes as shown in Table 5, which divides thermokarst lakes into four size classes. From 1962 to 2007, the number of lakes decreased in all size classes (except 100 -1000 ha), and the most reduction (from 117 lakes to 67 lakes) occurred in lake size classes between 0.1 -1 ha and 10 -100 ha. Large lake size class of 100-1000 ha was 181 ha (1962/68) and decreased by 34 ha (2006/07). Between 1962 and 2007, lake area decreased in all lake classes. In fact, fifty lakes completely disappeared in sites 5 -7 (Figure 4(b)). Most of the disappeared lakes had an area of less than 1 ha. This result is likely associated with the deep active layers (Table 1) and disappearing permafrost at those sites [21]. Shrinking and disappearing thermokarst lakes may become a common feature in the discontinuous, sporadic and isolated permafrost zones as a consequence of warming climate and disappearing permafrost [7]. Our results were consistent with lake reductions observed by other similar studies [7]- [9]. Smith et al. [8] used satellite imagery from 1970s to 2004 to document the decline of thousand lakes in Siberia. Their study showed an increase of total lake number and area in the continuous permafrost zone, and a contrasting decrease in discontinuous to isolated permafrost zone. We also found a substantial reduction in the total number and area of thermokarst lakes in the isolated permafrost zone in Mongolia.  Table 6). The correlations of the linear trend at those sites were insignificant since the 1962. However, the total annual potential evapotranspiration (PET) increased significantly at all sites mainly due to increasing MAAT, except for sites 5 and 7. The total annual PET had higher amounts at sites 5-7 in the isolated permafrost zone than sites 1 -4 in the continuous permafrost zone ( Figure 5 (c)). These differences were attributed to the large latitudinal climatic gradient in Mongolia (see Figure 1 and Table 1). According to these results, the decreasing total annual P and increasing PET contributed to the negative water balance (P-PET) at all sites, except site 7 ( Figure 5(d)).

Hydro-Climatic Change from 1962 to 2007
We attribute the increase in area and number of lakes observed in the continuous permafrost zone to ongoing warming ( Figure 5(a)), which has been found as the main driver for thermokarst lake expansion in this permafrost zone [8] [28]. During the last 45 years, MAAT and PET increased significantly in sites 1 -4. Such persistent warming trend has enhanced the permafrost degradation in northern Mongolia [21] including thawing of massive ground ice (e.g., Hinzman et al. [29]). As a result, new thermokarst lakes were formed in the continuous permafrost zone (Figure 4) due to ice-rich permafrost thaw and ground surface subsidence.  The observed reduction in number and area of lakes in the southern isolated permafrost zone may be due to a combination of several factors including the air temperature, the potential evapotranspiration, and negative water balance, as well as permafrost disappearance [21]. The MAAT trend increased slightly at sites 5 -6 in the isolated permafrost zone from 1962 to 2007 ( Figure 5(a)). Furthermore, a decreasing trend of MAAT with an increasing P was found at site 7, such results were not consisted with the lake reduction. From our analysis, we found no significant correlation between the total annual PET and reduction of lakes in these sites. However, the total annual PET was higher at sites 5 -7 in the isolated permafrost zone than sites 1 -4 in the continuous permafrost zone ( Figure 5(c)). Potentially, the reduction in area and number of thermokarst lakes, especially small lakes (Figure 3), it likely associated with negative water balance (P-PET), the disappearing permafrost and subsurface drainage observed in the isolated permafrost zone [7] [9].

Conclusion
This study provides the first baseline information of thermokarst lake changes across Mongolia, filling the gap in sub-Arctic lake inventories at regional scales such as the southern fringe of Siberian permafrost region. The time series data of high-resolution satellite imagery demonstrated useful in determining changes in the number and areal extent of thermokarst lakes greater than 0.1 ha in Mongolia from 1962 to 2007. We found contrasting changes of thermokarst lake dynamics in the continuous and isolated permafrost zones. Thermokarst lakes in the continuous permafrost zone have increased significantly in number and area while in the isolated permafrost zone we observed a decrease in both number and area over the 45 years of the study period. The dramatic increase in number of smaller lakes with sizes between 0.1 -1 ha and 1 -10 ha compared to larger lakes (10 -100 ha and 100-1000 ha) in the continuous permafrost zone is likely as a result of permafrost degradation due to increasing temperature and evaporations. However, small lakes (0.1 -1 ha) had a significant reduction in number in the isolated permafrost zone. The mechanism behind reduction of lake number and area may attribute to a combination of disappearing permafrost, deepening of the active layer and an increase water loss in this permafrost zone. Future research should focus on the temporal and spatial assessment of lake area changes across this region to better understand the detailed processes of lake area dynamics.