Modern thermokarst lake dynamics in the continuous permafrost zone, northern Seward Peninsula, Alaska



[1] Quantifying changes in thermokarst lake extent is of importance for understanding the permafrost-related carbon budget, including the potential release of carbon via lake expansion or sequestration as peat in drained lake basins. We used high spatial resolution remotely sensed imagery from 1950/51, 1978, and 2006/07 to quantify changes in thermokarst lakes for a 700 km2 area on the northern Seward Peninsula, Alaska. The number of water bodies larger than 0.1 ha increased over the entire observation period (666 to 737 or +10.7%); however, total surface area decreased (5,066 ha to 4,312 ha or −14.9%). This pattern can largely be explained by the formation of remnant ponds following partial drainage of larger water bodies. Thus, analysis of large lakes (>40 ha) shows a decrease of 24% and 26% in number and area, respectively, differing from lake changes reported from other continuous permafrost regions. Thermokarst lake expansion rates did not change substantially between 1950/51 and 1978 (0.35 m/yr) and 1978 and 2006/07 (0.39 m/yr). However, most lakes that drained did expand as a result of surface permafrost degradation before lateral drainage. Drainage rates over the observation period were stable (2.2 to 2.3 per year). Thus, analysis of decadal-scale, high spatial resolution imagery has shown that lake drainage in this region is triggered by lateral breaching and not subterranean infiltration. Future research should be directed toward better understanding thermokarst lake dynamics at high spatial and temporal resolution as these systems have implications for landscape-scale hydrology and carbon budgets in thermokarst lake-rich regions in the circum-Arctic.

1. Introduction

[2] Thermokarst refers to the process by which characteristic landforms form following disturbance of the thermal equilibrium of the ground resulting in thaw of ice-rich permafrost or melting of massive ice [van Everdingen, 1998]. In areas with relatively low relief, a thick overburden of unconsolidated sediments, and high ground ice content, this often results in the formation of thermokarst lakes. Thermokarst lakes are one of the most dynamic features in Arctic and sub-Arctic lowland landscapes as the lakes, besides facilitating further thaw settlement, often expand laterally through thermal and mechanical erosional processes. Thus, the lateral and vertical dynamics of thermokarst lake change result in complex interactions with topography, streams, neighboring thermokarst lakes, and permafrost. Given current and projected Arctic climate change it is important to conduct studies on thermokarst lake dynamics as they are thought to be an indicator of the effects of climate change on the landscape at least over broad geographic scales [Smith et al., 2005].

[3] Thermokarst lakes have mainly formed during the course of the Holocene and are a sign of local permafrost degradation following post glacial climate warming [Rampton, 1988; Romanovskii et al., 2004; Walter et al., 2007]. Thermokarst lakes are abundant surface features across many high latitude regions, such as the Seward Peninsula [Hopkins, 1949; West and Plug, 2008; Plug and West, 2009], the Arctic Coastal Plain [Sellmann et al., 1975; Hinkel et al., 2005] as well as several areas in Interior Alaska [Jorgenson and Osterkamp, 2005]; in Canada on Banks Island [Harry and French, 1983], Tuktoyaktuk Peninsula [Mackay, 1988], the Yukon Coastal Plain [West and Plug, 2008; Plug and West, 2009], and Richards Island [Burn, 2002], and in large regions of Siberia [Romanovskii et al., 2004, Tomirdiaro and Ryabchun, 1973; Zimov et al., 1997; Smith et al., 2005; Grosse et al., 2006; Walter et al., 2007; Grosse et al., 2011]. While not all northern, high-latitude lakes are of thermokarst origin [Jorgenson and Shur, 2007; Smith et al., 2007], their importance to global climate change and northern high latitude soil and permafrost-stored carbon cycling has been noted [Zimov et al., 1997; Walter et al., 2006; Walter et al., 2007]. Therefore quantifying changes in thermokarst lakes is of importance for understanding potential positive and negative feedbacks to the atmospheric carbon budget.

[4] A suite of recent studies have utilized remotely sensed imagery to document changes in thermokarst lake extent in various locations across the Arctic and sub-Arctic (Figure 1). In general, thermokarst lakes are thought to be increasing in abundance and surface area in regions of continuous permafrost and decreasing in abundance and surface area in transitional permafrost zones (discontinuous, isolated, sporadic) [Smith et al., 2005]. For example, Smith et al. [2005] examined coarse resolution satellite imagery (150 m) to detect surface area changes in lakes (≥40 ha) from 1973 to 1997/98 over a broad expanse of West Siberia (∼515,000 km2). They found that lakes had decreased by 11% during this time period, though the changes were not uniform across the study area. Lake abundance and surface area decreased in the zones of discontinuous, sporadic, and isolated permafrost, while both increased in the zone of continuous permafrost. Lake change in Alaska also appears to follow this general pattern, though interpretation differs as to the relative roles of increasing aridity and permafrost dynamics [Riordan et al., 2006; Yoshikawa and Hinzman, 2003]. Recent studies have also shown the importance of precipitation on thermokarst lake surface area extent and its detection in moderate resolution (30 m) remotely sensed imagery [Plug et al., 2008; Jones et al., 2009a; Labrecque et al., 2009].

Figure 1.

Digital elevation model base map showing the location of Arctic and sub-Arctic lake change detection studies: (1) Yoshikawa and Hinzman [2003], (2) Smith et al. [2005], (3) Klein et al. [2005], (4) Walter et al. [2006], (5) Riordan et al. [2006], (6) Hinkel et al. [2007], (7) Plug et al. [2008], (8) Jones et al. [2009a], (9) Marsh et al. [2009], (10) Labrecque et al. [2009], (11) Arp et al. [2011] (12) Jones et al. (submitted manuscript, 2011). Exposed portions of the continental shelf during the last glacial maximum are shown in gray.

[5] It has long been shown that thermokarst lakes have a tendency to drain laterally [Hopkins, 1949; Mackay, 1988; Brewer et al., 1993; Hinkel et al., 2007; Marsh et al., 2009]. Typical thermokarst lake drainage mechanisms in the continuous permafrost zone consist of bank overflow, ice wedge degradation and development of a drainage network, headward stream erosion, lake tapping, coastal erosion, as well as expansion of a lake toward a drainage gradient. A few recent studies that have focused on drainage of thermokarst lakes in continuous permafrost environments found that rates of drainage have been fairly low [Hinkel et al., 2007] or even decreasing during the last ca. 50 years [Marsh et al., 2009]. However, the specific external or internal forcing mechanism remains poorly understood.

[6] In order to more fully understand landscapes dynamics in a thermokarst lake-rich landscape and their potential role in the northern high-latitude carbon cycle it is imperative to assess the balance between lake expansion and lake drainage. Since thermokarst lakes have been shown to act as an effective mechanism for the release of organic carbon previously frozen in permafrost and since drainage of thermokarst lakes allows for organic carbon sequestration through the accumulation of peat in drained basins, landscape-scale analyses of carbon cycling should account for thermokarst lake dynamics. The objectives of our study were (1) to employ high spatial resolution remotely sensed imagery (1 m) acquired between 1950 and 2007 to detect changes in thermokarst lake extent in continuous permafrost on the northern Seward Peninsula, Alaska; (2) to quantify rates of change and analyze whether these rates changed over time; (3) to identify potential mechanisms of lake expansion and lake area loss; and (4) discuss implications of thermokarst lake change for the carbon cycling in these landscapes.

2. Study Area

[7] Our study area is located on the northern Seward Peninsula in northwestern Alaska, USA (Figure 1, site 12) and bounded by the Chukchi Sea to the north, Kotzebue Sound to the east, and Devil Mountain volcano and a chain of large maar lakes to the south (Figure 2). The northern Seward Peninsula represents one of the major lake districts in Alaska, where 7% of the 6,418 km2 land area is covered in extant lakes larger than 1 ha [Arp and Jones, 2009]. This region is located in the central portion of Beringia, which during the Last Glacial Maximum (19–26.5 kyr ago) formed a largely nonglaciated landmass with exposed shelves between Siberia and Alaska (Figure 1) [Hopkins, 1967].

Figure 2.

Terrain unit map of the Cape Espenberg Lowland-Devil Mountain region (digitized from Charron [1995]). The 700 km2 study area is outlined in red and the location of Kotzebue relative to our study area is shown in the inset.

[8] Throughout the Wisconsinan, much of the lowlands in the Bering Land Bridge were characterized by the accumulation of eolian and alluvial silt originating from braided floodplains crossing the exposed shelves and the formation of syngenetic permafrost and ice wedges [Hopkins et al., 1955]. For the northern Seward Peninsula region, wind deposition is the favorable hypothesis for the aggradation of this land surface as silt/loam textures and grain size analyses are typical of loess, and also due to the fact that the silt drapes the paleolandscape [Höfle et al., 2000]. Roughly 21.5 kyr ago, a hydromagmatic eruption created the Devil Mountain Maar Lake, the largest known maar on Earth, and deposited tephra over a 2,500 km2 area in the Cape Espenberg lowlands of the northern Seward Peninsula [Beget et al., 1996]. Close to the maar lake the tephra thickness exceeds several 10 m, whereas in distal places thicknesses may still reach 1 m. The primary direction of the tephra outfall was to the north and west of the eruption [Beget et al., 1996], creating a matrix of ice-rich silt blanketed by volcanic tephra, which has subsequently been capped further by Late Pleistocene silt and a cover layer of Holocene soils and peat [Höfle et al., 2000].

[9] Throughout the Holocene, much of the land surface in this region has been subject to permafrost degradation and thermokarst processes [Hopkins, 1949; Charron, 1995]. With a mean annual ground temperature of ∼−3°C, the continuous permafrost in the study area is relatively warm today [Smith et al., 2010], but conditions remain favorable for the formation of permafrost and epigenetic ice wedges following drainage of thermokarst lakes [Hopkins et al., 1955]. The reworking of the landscape by thermokarst processes is clearly evident in the topography and landscape terrain. Within our 700 km2 study area, a subset of the Cape Espenberg Lowland-Devil Mountain region (Figure 2), 73% of the landscape has been influenced by the formation and/or drainage of thermokarst lakes [Charron, 1995]. The ratio of thermokarst lake-affected versus non lake-affected landscape units in the study region is very similar to other ice-rich permafrost regions in northern Alaska [Hinkel et al., 2005] and Siberia [Grosse et al., 2006].

[10] The vegetation of the northern Seward Peninsula is characterized as Bering Tundra. Ericaceous shrubs, including Empetrum nigrum, Vaccinium uliginosum, and V. oxycoccus, are common on the uplands, along with Betula nana, Salix spp., Eriophorum vaginatum, Sphagnum fuscum, Rubus chamaemorus, and Polytrichum strictum. Water-logged lowlands, including low-centered ice wedge polygons and drained lake basins, are dominated by Carex aquatilis, Eriophorum angustifolium, Eriophorum scheuchzeri, and Sphagnum riparium. Mean annual temperature of the region is −6.1°C based on the 1971 to 2000 climate normal period, recorded at Kotzebue, Alaska, which is located on the Baldwin Peninsula roughly 80 km to the northwest (Figure 2). Mean July temperature is 12.6°C (warmest month) while mean February temperature is −19.7°C (coldest month). Mean annual precipitation is 255 mm with slightly more than half (130 mm) falling between July and September, presumably in the form of rainfall.

3. Methods

3.1. Imagery and Classification

[11] We used high spatial resolution, contemporary satellite imagery and historical aerial photography, to quantify changes in thermokarst lakes and ponds larger than 0.1 ha from 1950 to 2007 (Table S1 of the auxiliary material). Pan-sharpened, multispectral IKONOS© satellite imagery from 2006 and 2007 with a resolution of 1 m was available for a large portion of the Cape Espenberg Lowland-Devil Mountain region and its cloud-free extent defines our 700 km2 study area (Figure 2). The imagery was georegistered to 1 m resolution, orthorectified, aerial photography available for a portion of the region [Manley et al., 2007a, 2007b, 2007c]. In areas without recent orthorectified aerial imagery, terrain-corrected Landsat ETM+ imagery (processing level L1T) was used with stable lake centroids as additional control points [Sheng et al., 2008].

Table 1. Change in Lake Number, Lake Area, Expansion Rate, and Drainage Events From 1950/51 to 1978 and 1978 to 2006/07 for Four Lake Size Classes
Lake Size Class (ha)Lake Number ChangeLake Area Change (ha)Lake Expansion Rates (m/yr)Lake Drainage Events
1950/51 to 19781978 to 2006/071950/51 to 19781978 to 2006/071950/51 to 19781978 to 2006/071950/51 to 19781978 to 2006/07
0.1 to 1+13+47+11+260.100.223138
1 to 10−2+14−47+370.220.281812
10 to 40+60+81+270.350.38710
40 to 400−3−4+3−8830.620.5958

[12] We compared lake and pond surface area from the registered IKONOS© imagery to orthorectified black and white photography from 1950/51 [Manley et al., 2007a] and orthorectified color infrared (CIR) photography from 1978 [Manley et al., 2007b]. For areas outside of the coverage provided by the Manley et al. [2007a, 2007b, 2007c] data sets, we acquired the appropriate aerial image frames from the U.S. Geological Survey, EROS Data Center in Sioux Falls, SD, USA and registered the frames to the IKONOS© imagery. In total, this required nine additional frames from ca. 1950 and three additional frames from 1978. Overall, mean RMS values associated with the Manley et al. [2007a, 2007b] orthodata sets were 1.7 m for the 1950/51 imagery and 1.1 m for the 1978 imagery. Mean RMS values for the additional frames required to cover the study area were 1.5 m and 1.6 m, respectively.

[13] We classified the imagery into water and nonwater binary raster files using object-oriented classification algorithms available in the image processing software eCognition® [Frohn et al., 2005]. Each frame was processed with the multiresolution segmentation parameter within eCognition® at a scale parameter of 10, shape factor of 0.1, and compactness and smoothness factor of 0.5 in the normal mode. This essentially converted the image from individual pixels, into pixels grouped as like objects, which nearly perfectly delineates lake and pond perimeters. The image was subjected to an additional segmentation in spectral differencing mode that grouped like objects further. Following the second segmentation, the objects were classified as water or nonwater based upon threshold values of the image objects. We employed this automated classification approach that provided initial perimeters of lakes and ponds, which was then supplemented with a semiautomated classification approach where we visually checked the accuracy of the water body perimeter and manually adjusted when needed, e.g., for shadows cast by steep lake banks and potentially flooded lake margins. This allowed for more rapid delineation of lakes and ponds over strict manual interpretation and allowed us to refine the classification and potentially account for differences in timing of imagery acquisition. The classified images were then converted from raster image files to vector files in order to assess changes in lake area and lake abundance for each of the three time slices. Owing to the high spatial resolution of the imagery (1 m) we were able to confidently assess area changes to lakes and ponds with a minimum mapping unit of 0.1 ha for each time slice. The 0.1 ha size was chosen for the cut off as it represented approximately three to five times the size of a typical low-centered ice wedge polygon pond.

[14] Recent remote sensing studies have shown that total precipitation during the year prior to image acquisition may impact measured surface water area of thermokarst lakes [Plug et al., 2008; Jones et al., 2009a]. For our study area and image set, precipitation during the year preceding imagery acquisition for each time slice was fairly similar with 211 mm (1950/51), 227 mm (1978), and 241 mm (2006/07). However, more precipitation in the most recent time slice may be reflected as a net increase in surface water area. Further, variations in the timing of imagery acquisition relative to breakup and freezeup within a particular year may impact lake surface area measurements. While all of the imagery was acquired during the summer (Table S1 of the auxiliary material), such seasonal differences may also impact measurements of lake surface area. However, we felt that the comparisons between the three time slices to be relevant due to the steep profile of lake banks along both lowland and upland lake bluffs [West and Plug, 2008; Plug and West, 2009] which would limit the impact of water level fluctuations on the surface area measurements, the imagery resolution and methodology used to delineate water bodies, and the removal of floodplain lakes from the analysis [Smith et al., 2005]. In addition, vegetation growing along lake margins and disproportionate growth rates over time may also impact the ability to delineate shorelines effectively. However, this potential source of error is likely negligible due to the 1 m resolution imagery used in this study and the overall low canopy height in the tundra-dominated study area.

3.2. Lake Shoreline Change

[15] Detailed lake expansion rates were determined with the Digital Shoreline Analysis System (DSAS) extension for ArcGIS© [Thieler et al., 2009]. DSAS generates orthogonal transects at user-defined intervals along a baseline, and calculates the rate of change between two vector files (lake perimeter from time 1 and time 2) based on the elapsed time and the linear distance. DSAS is generally used for coastal change studies; however, it is an appropriate tool for determining change rates with any time series vector file. We used a 5 m buffer of the most recent lake perimeter as the baseline and orthogonal transects were cast at 50 m intervals around each lake perimeter and change rates were determined from 1950/51 to 1978 and from 1978 to 2006/07.

[16] Error estimates of the linear expansion rate measurements were determined using a modified version of the equation used in determining errors associated with measurement of Arctic coastal erosion rates [Jones et al., 2009b]:

equation image

where Ep1 and Ep2 represent the pixel resolution of the imagery (all 1 m) from a particular year, RMS1 and RMS2 are the root mean square errors associated with georegistration of an image mosaic from a particular year, and Δt is the time interval associated with a given time period. Thus, error associated with the expansion rate measurements were 0.09 m/yr and 0.09 m/yr, during the first (1950/51 to 1978) and second (1978 to 2006/07) time period, respectively.

3.3. Climate Data Analysis

[17] Long-term (1950 to 2007) temperature and precipitation data were available from the climate station located in Kotzebue, Alaska (66.898°, −162.596°). Mean monthly air temperature (Ta) and precipitation (P) data were retrieved from the Alaska Climate Research Center ( to assess any variations in climate between the time periods (1950/51 to 1978 and 1978 to 2006/07) used to monitor lake change. These time periods also roughly coincide with a shift in the Pacific Decadal Oscillation from a negative phase to a positive phase in 1976 [Hartmann and Wendler, 2005].

[18] In order to estimate lake water balance, we calculated monthly evaporation (E) using the Blaney-Criddle method [McGuinness and Bordne, 1972]:

equation image

during the period of open water duration (June to September), where E is in cm and p is the monthly mean daily percentage of annual daylight hours for 67°N latitude. This empirical method has been compared to the Bowen ratio, energy budget method for a lake in temperate latitudes and performed reasonably for a model based on Ta and day length [Rosenberry et al., 2007]. Subtracting E from P was used to indicate lake water balance (P–E) from the month of snow accumulation (October) through the summer to September.

[19] Climate data were summarized according to hydroclimatic elements and periods hypothesized to influence interannual as well as long-term lake water balance and thermokarst processes. Winter precipitation (October – April) was analyzed as it determines snowmelt runoff, lake recharge, and the potential for overtopping of lake outlets, one aspect of catastrophic lake drainage. Annual water balance (October–September) summed to the late summer period of surface area observation was analyzed to assess interannual variability in water levels and cumulative moisture conditions for both time periods. Finally, mean annual air temperature (MAAT) was analyzed as it relates to both lake and ground temperatures, which play a role in driving and resisting thermal erosion [Burn, 2002]. For each set of hydroclimatic data, we compared linear regression models for the time periods 1950–1978 and 1978–2007 to evaluate trends and mean conditions that could explain observed lake change patterns.

4. Results and Discussion

4.1. Lake and Pond Abundance and Surface Area

[20] The number of thermokarst lakes and ponds larger than 0.1 ha in the 700 km2 Cape Espenberg Lowland study area has increased from 666 in 1950/51, to 680 in 1978, to 737 in 2006/07, or a 10.7% increase during the last 57 years. Analysis of the frequency of lakes and ponds based on four size classes (Table 1) shows an increase of 3.4% in the smallest size class (0.1 to 1 ha) between the 1950s and 1978, while an additional increase of 11.9% occurred in this size class between 1978 and 2006/7 (Figure 3). The next smallest size class (1 to 10 ha) also experienced its largest change in the second time period. The 1 to 10 ha size class actually decreased slightly between the first two time slices (−0.9%), however increased 6.8% during the latter time period. The size class ranging from 10 to 40 ha showed an increase of 13.0% between 1950/51 and 1978 and remained stable between 1978 and 2006/07. The largest size class was the only class to consistently show a decreasing trend of 10.0% and 15.3% in the first and second time period, respectively. Thus, it is apparent that there has been a loss of large lakes in the study area and an increase in the number of small lakes and ponds. This increase in the number of small water bodies may be a result of partial drainage of these larger lakes, leaving multiple remnant lakes and ponds, or may result from the formation of new lakes as a result of permafrost degradation.

Figure 3.

Comparison of the number of water bodies in four size classes in 1950/51, 1978, and 2006/07. The increase in smaller water bodies can largely be attributed to the partial drainage of larger water bodies.

[21] Of the 14 additional lakes mapped in 1978 relative to the 1950s, six resulted from the formation of new lakes (mean size of 0.12 ha), while the majority resulted from the partial drainage of larger thermokarst lakes and division into remnant water bodies. Similarly, the majority of the 57 additional lakes mapped in the 2006/07 imagery compared to the 1978 imagery were a result of the drainage of larger thermokarst lakes, with 96% representing lakes that resulted from drainage and division of a larger water body. During the latter time period, new thermokarst lake formation accounted for five lakes with a mean size of 0.27 ha. Further, all of these new lakes have formed in drained lake basins. There were a few instances (six) of new water bodies forming in remnant upland topography, however their size was below our minimum mapping unit of 0.1 ha. Thus, between 1950/51 and 2006/07, 85% of the increase in the number of lakes is actually a result of the partial drainage of larger thermokarst lakes.

[22] As shown above, sole analysis of lake abundance without addressing lake area changes over time may be misleading, since lake number may increase as a result of partial lake drainage. Only the combination of analysis of lake surface area changes and lake abundance provides meaningful information for understanding thermokarst lake dynamics. Total lake surface area during 1950/51 was 5,066 ha, during 1978 it was 5,115 ha, and during 2006/07 it was 4,312 ha. Therefore, a total surface area increase of 1.0% occurred between 1950/51 and 1978, which was followed by a decrease of 15.7% between 1978 and 2006/07, resulting in an overall lake surface area reduction of 14.9% between 1950 and 2007. Further, mean lake size for the study area over the period of record has decreased as a result of the increase in small water bodies as well as the loss of a number of larger lakes, declining from 7.6 ha (1950/51), to 7.5 ha (1978), to 5.9 ha (2006/07).

[23] A number of recent studies have focused on the pattern and rate of change in lakes located in the Arctic and sub-Arctic (Figure 1). Two broad-scale geographic studies indicate that thermokarst lakes in the zone of continuous permafrost are increasing in both number and area, while in the zone of discontinuous permafrost they are decreasing in both number and area [Smith et al., 2005; Riordan et al., 2006]. Other studies have shown that thermokarst lake surface area is tightly coupled with precipitation patterns [Plug et al., 2008; Jones et al., 2009a; Labrecque et al., 2009]. The results from our study area in a relatively warm region of the continuous permafrost zone document a different pattern for thermokarst lake change. We show an increase in the total number of water bodies, yet a decrease in the total area of thermokarst lakes. This pattern can be explained by the formation of several small lakes and ponds following partial drainage. Including water bodies as small as 0.1 ha, we found that total lake number in this study area has increased by 10.7% since the 1950s, yet total lake surface area has decreased by 14.9%. However, if we increase the minimum lake size to 40 ha in order to draw comparisons with changes documented for large lakes in Siberia [Smith et al., 2005], we find that between 1950 and 2007 there has been a reduction in lake number by 24.1% and a reduction in lake area of 26.5%. Therefore, large lakes on the northern Seward Peninsula are draining and are not being replenished by the growth and coalescence of smaller lakes at the same rate. In order to more fully understand these processes it is important to look further at the high resolution imagery to measure in detail thermokarst lake expansion rates and possible drainage mechanisms.

4.2. Thermokarst Lake Expansion Rates

[24] While expansion of thermokarst lakes in the continuous permafrost zone is a known phenomenon, there are very few data on linear expansion rates. Thermokarst lakes are thought to expand due to a number of different shoreline erosional processes, which include: (1) development of thermomechanic erosional niches [Tedrow, 1969]; (2) mass wasting through thaw slumps and block failures [Tomirdiaro and Ryabchun, 1973; Kokelj et al., 2009; Plug and West, 2009]; (3) mechanical erosion caused by ice shove during breakup; and (4) by incorporation of polygonal ponds into the lake [Billings and Peterson, 1980]. In order to determine linear expansion rates and whether they may have changed over time, we analyzed lakes and ponds that continually expanded over our three image time slices within our 700 km2 study region using the DSAS tool for two time periods, 1950/51 to 1978 and 1978 to 2006/07 (Figure 4). The number of lakes analyzed by this method was 370 and the water body size ranged from 0.1 ha to 378 ha.

Figure 4.

Example of expansion rate measurements at two lakes in the study area. Rates determined with the DSAS tool [Thieler et al., 2009] at 50 m increments around the perimeter of the lake. (a) Lake Rhonda expanded at a mean rate of 0.53 m/yr and (b) Lake Luna expanded at a mean rate of 0.44 m/yr over the observation period. The 1951 lake shoreline is shown as a yellow polygon, the 1978 lake shoreline as a green polygon, and the background image is from 2006. The 100 m grid in each frame shows the scale.

[25] Mean expansion rates for all lakes in the 1950/51 to 1978 time period was 0.34 m/yr, while in the 1978 to 2006/07 time period it was 0.39 m/yr. Owing to errors associated with image coregistration, lake perimeter delineation, and image pixel size (+/−0.09 m/yr) the small difference in expansion rate between the two time periods is within our measurement uncertainty. Mean expansion rate for an individual lake ranged from a low of 0.02 m/yr to a high of 1.81 m/yr in the 1950/51 to 1978 time period and from a low of 0.04 m/yr to a high of 1.55 m/yr in the 1978 to 2006/07 time period. Maximum expansion rate for an individual location was 4.25 m/yr and 6.01 m/yr in the first and second time period, respectively.

[26] Analysis of individual lake expansion rates by lake surface area shows a weak, but positive correlation for the early (r2 = 0.30) and late (r2 = 0.14) periods and only slight coherence in rates between periods (r2 = 0.17), suggesting somewhat variable expansion rates over time. However, categorizing expansion rates for lakes based on four distinct size classes (0.1 to 1 ha, 1 to 10 ha, 10 to 40 ha, and 40 to 400 ha) shows interesting results over time and between size classes (Table 1). Lakes in the smaller size class showed the largest discrepancy between the two time periods, with expansion rates increasing from 0.10 m/yr to 0.22 m/yr. Lakes in the middle size class expanded at slightly higher rates than the smallest size class, however the difference between the 1950/51 to 1978 and 1978 to 2006/07 was much smaller, increasing from 0.22 m/yr to 0.28 m/yr. For the two larger size classes, expansion rates were by far the highest. Both size classes showed fairly stable expansion rates (+/−0.03 m/yr) however, the largest size class was the only class to show a slight decrease, from 0.62 to 0.59 m/yr. Although smaller lakes have not expanded at the same rate as larger lakes it is interesting that the smaller water bodies exhibited an increase in rates between the time periods, whereas the larger lakes did not. Presumably, these smaller water bodies expand more as a result of thermal erosion since their small surface area and open water extent does not provide adequate fetch for effective wave action and mechanical erosion. Thus, the increased rate in smaller water bodies in the second time period may be a result of warmer water temperatures relative to the first time period.

[27] In addition to water body size, another potential source of variation in expansion rates relates to the height of a lake margin bluff. Lake margins in the study area can be divided into two distinct categories: lowland and upland bluff types (Figure 5). Of our 7,423 point measurements of thermokarst lake expansion, lowland bluffs accounted for 88% and upland bluffs accounted for 12%. Lowland margins are indicative of expansion into drained thermokarst lake basins and typically have bank heights from 0.5 m to 3.0 m. Upland margins refer to erosional remnants or yedoma-like terrain that have not been modified previously by thermokarst lake processes. Lake bluff heights along such upland margins typically range from 6 m to 17 m. The lowland margin types exhibited fairly consistent expansion rates between the two time periods, 0.37 m/yr and 0.42 m/yr, respectively, while erosion of the upland margin types also showed a similar pattern, increasing slightly from 0.15 m/yr to 0.18 m/yr, respectively, again the slight increase is within our measurement uncertainty. Thus, not surprisingly, it appears that expansion rate is largely driven by the height of the bluff and the composition and state of the material that the thermokarst lake is expanding into, with higher expansion rates along margins at which less sediment material has to be removed. Thus, total variation in the expansion rate of individual lakes may largely be explained by a combination of lake size and bluff height of surrounding lake margin. In addition, bathymetry likely also plays a role in the expansion rate of a bluff section due to warmer water temperatures associated with deeper lakes and greater disturbance to the ground thermal regime [Arp et al., 2011]; however, we lacked lake depth information to assess this for our data set.

Figure 5.

Field photos showing differences in lake bluff types. (a) Photo showing a typical lowland bluff and (b) a typical upland bluff. A Cessna 185 floatplane in each photo provides a scale.

[28] There is general agreement between thermokarst lake expansion rates from our study area and the limited data available from other thermokarst lake regions (Table 2). In various regions in Alaska, Canada, and Siberia, the long-term, mean expansion rate for thermokarst has been shown to vary between 0.10 and 0.70 m/yr for entire lake margins and upwards of 2.0 to 5.0 m/yr for individual locations along the lake perimeter. Thus, the mean thermokarst lake expansion rate of 0.35 to 0.39 m/yr that we have measured from two different time periods for the northern Seward Peninsula in Alaska falls within the range of expansion rates measured in other Arctic regions. However, all of the previous studies have included only a small number of lakes or provided hypothetical values based on limited data. Thus, our measurement of expansion rates at 7,423 points distributed across 370 lakes in our study region provides the first landscape-scale assessment of this typical process for thermokarst lakes located in the northern, high-latitude continuous permafrost zone.

Table 2. A Synthesis of Previously Published Measurements on Thermokarst Lake Expansion Rates in the Circum-Artic
RegionSubregionNumber of Lakes AnalyzedMean Expansion Rate (m/yr)Source
AlaskaEastern North SlopeUnknown0.30Schell and Ziemann [1983]
 Barrow Peninsula80.73Lewellen [1970]
 Fish CreekUnknown0.10Jorgenson and Shur [2007]
 Northern Teshekpuk Lake Special Area130.70Arp et al. [2011]
 Interior AlaskaUnknown0.10 to 2.0 (maximum rate)Jorgenson and Osterkamp [2005]
 Northern Seward Peninsula3700.35 to 0.39this study
CanadaMayo Territory30.70Burn and Smith [1990]
 Hudson Bay LowlandsUnknown2.0 (maximum rate)Dyke and Sladen [2010]
SiberiaCentral Yakutia20.12 to 0.52Are et al. [1979]
 Lower Anadyr LowlandsUnknown0.5 to 5.0 (maximum rate)Tomirdiaro and Ryabchun [1973]

[29] Further research should be directed toward conducting similar types of landscape analyses in other thermokarst lake-rich regions. The determination of expansion rates is important for developing long-term monitoring programs focused on the use of repeat remote sensing imagery to assess thermokarst lake expansion over time. Thus, for our study area, given a rate between 0.30 and 0.40 m/yr, it would be feasible to acquire high-resolution imagery (1 m) at three to five year increments to document change in rates overtime. Reporting detailed expansion rate estimates is also important for interpreting and understanding other lake change studies conducted with coarser resolution imagery. For example, widespread thermokarst lake expansion reported over a ∼25 year period for Siberia using imagery with a spatial resolution of 150 m [Smith et al., 2005], indicates that expansion rates in their study area would have to have averaged at least 6 m/yr for many lakes over a large area, in order for an increase due to shoreline erosion and permafrost degradation to be detected. Given the results from our analysis and expansion rates reported from other thermokarst lake-rich regions this seems unlikely and other factors controlling lake surface area fluctuation may have also been detected [Plug et al., 2008].

4.3. Thermokarst Lake Drainage

[30] As lakes expand, the chances for drainage increase due to the possibility of encountering a drainage gradient. As pointed out above, the majority of lakes in the study region are expanding, yet the increase in lake abundance can be explained by lake drainage and the division of a larger water body into several smaller water bodies. An analysis of the number of drainage events, defined as a >25% reduction in surface area [Hinkel et al., 2007], reveals that 130 lakes drained between 1950/51 and 2006/07, which has resulted in an average drainage rate of 2.3 lakes/yr. Analyzing the lake drainage events further, reveals that the thermokarst lake drainage rate has remained fairly stable over the last half century, with a drainage rate of 2.2 lakes/yr between 1950/51 and 1978, and 2.3 lakes/yr between 1978 to 2006/07. However, in the second period there was an increase in the drainage rate of larger lakes, accounting for the drastic reduction in thermokarst lake surface area on the landscape (Table 1).

[31] Hinkel et al. [2007] analyzed lakes larger than 10 ha for a portion of northern Alaska (34,000 km2) and found that 50 lakes drained (>25% reduction in surface area) between ca. 1975 and ca. 2000, for a drainage rate of ∼2 lakes/yr. Catastrophic lake drainage events have also been reported for a 5,000 km2 area on the Tuktoyaktuk Peninsula, Canada. Mackay [1988] found that between 1950 and 1986 roughly 65 lakes had drained completely or partially, yielding a drainage rate of ∼1.8 lakes/yr. More recently, Marsh et al. [2009] provide estimates of lake drainage events from the same region by looking at three time periods, 1950 to 1973, 1973 to 1985, and 1985 to 2000. Their results indicate a reduction in the drainage rate of thermokarst lakes in the region, from 1.13 to 0.93 to 0.33 lakes/yr in each time period, respectively. Thus, drainage rates for thermokarst lakes for our 700 km2 study area on the northern Seward Peninsula are slightly higher, yet fairly similar to those documented for other regions in northern Alaska and NW Canada. However, if these drainage rates held up across the entirety of the northern Seward Peninsula it is likely that this region would exhibit the highest thermokarst lake drainage rates thus far found in the Arctic.

[32] Drainage of thermokarst lakes can be divided into two distinct categories, lateral and internal, both of which relate to degradation of confining permafrost. Lateral thermokarst lake drainage has been reported from a number of regions in the circum-Arctic. Typical mechanisms thought to lead to the lateral drainage of thermokarst lakes in the zone of continuous permafrost are bank overflow, ice wedge degradation and development of a drainage network, headward stream erosion, lake tapping, coastal erosion, as well as expansion of a lake toward a drainage gradient [Hopkins, 1949; Mackay, 1988; Brewer et al., 1993; Hinkel et al., 2007; Marsh et al., 2009; Grosse et al., 2011]. In contrast, internal drainage of thermokarst lakes has been documented in discontinuous permafrost regions in instances where the talik or thawed zone beneath a lake penetrates the permafrost, allowing for drainage subterraneously [Hopkins, 1949; Yoshikawa and Hinzman, 2003]. Thus, it is important to try to determine the causal mechanism for lake drainage in a particular region to better understand the processes driving observed lake change. In the case of lateral lake drainage events, this can be accomplished with the use of high-resolution remotely sensed imagery because drainage channels can be visualized whereby in coarser resolution imagery they largely are not detectable because of their often small width.

[33] Through analysis of the high spatial resolution imagery we classified the causal mechanism of a particular lake drainage event. Based on the total number of drainage events between 1950/51 and 2006/07, the majority of lake drainage events (71%) appear to be a result of lake expansion into a low lying area, such as an adjacent lake, a stream corridor, the coast, or topographic gradient (Figure 6). This class was determined through visible evidence of lake expansion and development of a drainage channel (Figure 7). Analysis of those lakes draining during the 1978 to 2006/07 time period showed that nearly all lakes expanded at a rate (0.42 m/yr), from 1950/51 to 1978, roughly equal to that of the mean for the entire study area (0.35 m/yr). However, without elevation data over the entirety of our study region other mechanisms cannot definitively be ruled out. The second most important mechanism appeared to be related to bank overtopping or possibly ice wedge degradation (17%). This inference was based upon no noticeable bluff erosion and lake expansion, however development of a distinct drainage channel. Further, it is likely that lake expansion and bank overtopping or ice wedge degradation can occur in concert with one another at a given lake and also in areas where lake drainage clustering may have occurred due to expansion of one lake and subsequent drainage and flooding of a nearby lake, ultimately forcing it to overtop its bank. Thermokarst lake expansion into adjacent ice-rich permafrost also leads to incorporation of ground ice meltwater into the lake, which may also factor into bank overtopping. However, only excess ice that is situated above the lake water level can be counted for this additional water into the lake since melting excess ice below the lake water level may have an opposite effect due to the fact that produced water volume is smaller than original ice volume. Migration of river channels and subsequent lake tapping was likely responsible for the drainage of two lakes. For thirteen of the lakes that decreased in area over the study period no drainage outlet was visible in the high-resolution imagery, possibly indicating that these lakes shrunk as a result of drying rather than drainage. It is possible that these drained internally, however, they were all very small (mean area of 0.20 ha) and permafrost in this region may be up to 100 m thick. Thus, through the analysis of high-resolution imagery we have determined that the vast majority of lake drainage events in our study area result from lateral drainage and surface permafrost degradation.

Figure 6.

Graph showing a topographic profile adjacent to a lake that drained during our observation period. The topographic profile is from a LIDAR data set available for a small portion of the study region. The bluff line positions from 1950 and 1978 are marked with a vertical black line. Note expansion of the lake toward a drainage gradient. The drained of the lake created an incised channel ranging in depth from 0.7 to 1 m.

Figure 7.

Image time series showing expansion of a lake between (a) 1950 and (B) 1978 followed by its drainage between (b) 1978 and (c) 2006. The white polygon in the image from Figure 7b 1978 image shows the lake perimeter from 1950. The white polygon in the Figure 7c 2006 image shows the lake perimeter from 1978. The lake likely drained soon after 1978 as little additional expansion occurred prior to drainage.

[34] Hinkel et al. [2007] also tried to infer the causal mechanism for lake drainage events in northern Alaska and found that 38% resulted from lake expansion, 16% from stream meandering, 26% by headward erosion of a stream, and 2% through coastal erosion. For 18% it remained unclear as to how the lakes drained. Thus, with the exception of lakes draining via coastal erosion and positively identifying headward erosion (instead we consider bank overtopping or ice wedge degradation and development of a drainage channel), the relative pattern is similar. Further, the authors also reported a number of cases where human disturbance caused the drainage of lakes near Barrow, Alaska [Hinkel et al., 2007]. For our study, we are unaware of the impact of humans on the drainage of lakes in this region.

4.4. Climate Data Observations From 1950 to 2007

[35] Analysis of climate data from Kotzebue, Alaska, located 60 miles to the northwest of the study region, showed distinct differences in climatology between lake change observation periods. The earlier period (1950–1978) was characterized by low and stable winter precipitation of 6.8 cm (Figure 8a), decreasing P–E of 0.4 cm/yr (r2 = 0.18) (Figure 8b), and a MAAT of −6.3°C (Figure 8c). The latter period (1978–2007) was characterized by increasing winter precipitation 0.1 cm/yr (r2 = 0.10) (Figure 8a), a fairly stable yet slightly wetter annual water balance (Figure 8b), and a slightly warmer MAAT of −5.0°C (Figure 8c). The step change observed between these periods is consistent with a shift in the Pacific Decadal Oscillation from a negative to positive phase in 1976 [Hartmann and Wendler, 2005].

Figure 8.

Climate data showing (a) winter precipitation, (b) P-E index, and (c) MAAT variation from 1950 to 2007. The first (1950 to 1978) and second (1978 to 2007) time periods are separated by different regression lines.

[36] Thus, it is somewhat surprising that the expansion rate and drainage rate of thermokarst lakes in the study area has remained fairly constant over the last ∼60 years. However, the slight increase in expansion rate as well as drainage rate, although within measurement uncertainty, may reflect these shifts in climate. The possibility also exists, that the Kotzebue climate station data does not directly represent climatic conditions in our study region. Despite the close proximity, Kotzebue is located on a narrow isthmus of land that juts into the ocean. Future research and field studies should be directed toward gaining a better understanding of the factors controlling lake expansion and lake drainage as an accurate assessment of the causal mechanisms is critical for understanding how thermokarst lakes may respond to climate change.

4.5. Thermokarst Lake and Carbon Cycle Dynamics

[37] As demonstrated above, thermokarst lake expansion and drainage is an active landscape change mechanism operating on the northern Seward Peninsula. Thermokarst lakes have expanded at a mean rate of 0.35 to 0.39 m/yr since the 1950s. However, as lakes expand the possibility of drainage increases due to the encroachment toward a topographic gradient. For our study area, the lateral expansion of lakes has resulted in their lateral drainage through surface permafrost degradation at a rate of roughly 2.3 lakes/yr. In a simple analysis of the landscape that has been impacted by these two mechanisms we determined land lost through time as a result of thermokarst lake expansion and land gained through time as a result of thermokarst lake drainage (Figure 9). This indicates that during the first time period (1950/51 to 1978) the landscape was in near equilibrium, losing approximately 390 ha and gaining 340 ha of land area. However, due to the drainage of several large lakes in the second time period (1978 to 2006/07), land area gained (1200 ha) was nearly four times the area lost (410 ha) due to thermokarst lake expansion.

Figure 9.

Comparison of land lost (blue) through thermokarst expansion and land gained (green) as a result of thermokarst lake drainage between 1950 and 2007.

[38] Several studies have documented landscape scale controls on the emission of greenhouse gases from northern high-latitude regions [Bartlett et al., 1992; Christensen et al., 2007; Flessa et al., 2008; Schneider et al., 2009]. In general, the importance of Arctic and sub-Arctic freshwater systems as a net emitter has been noted for some time [Coyne and Kelley, 1974; Kling et al., 1992; Cole et al., 1994; Phelps et al., 1998]. More recently, Walter et al. [2006] highlighted the potential importance of northern high-latitude thermokarst lake methane fluxes on the global atmospheric carbon budget. However, high lake methane fluxes are linked to a specific type of thermokarst lake that has formed in thick ice-rich and organic-rich sediments (yedoma or yedoma-like permafrost), whereas thermokarst lakes in basin-rich lowlands largely occupy fully or partially the basins of previous lake generations filled with lacustrine sediments already depleted in labile carbon, resulting in lower CH4 emissions during subsequent lake generations (K. M. Walter Anthony et al., Methane emissions from 50 years of thermokarst in Alaskan lakes, submitted to Journal of Geophysical Research, 2011). Drainage of such low-emitting later generation thermokarst lakes and the formation of wetlands in the basin could, despite carbon accumulation in peat, result in a short-term increase in CH4 emissions.

[39] In the case of carbon dioxide fluxes from thermokarst lake and basin-rich lowland Arctic landscapes, Zona et al. [2010] noted that the formation and drainage of thermokarst lakes factor in prominently to net CO2 emissions at the landscape scale, with increased emissions in recently drained basins and progressively decreasing emissions as a drained basin ages and less productive plant species colonize. However, once vegetated, all basins served as a CO2 sink. Similarly, for a shrinking thermokarst lake in central Alaska, Wickland et al. [2009] found that within the first 15 years of drainage the freshly exposed lake sediments acted as a CO2 source. However, 30 years postdrainage, as a result of a decrease in labile compounds and establishment of terrestrial vegetation in the basin, CO2 emissions were reduced to the point where the basin acted as a net C sink [Wickland et al., 2009].

[40] Bastviken et al. [2011] recently found that globally, freshwater methane emissions act to offset the net continental or terrestrial carbon sink. Thus, if lakes on the landscape are draining as a result of surface permafrost degradation and the basins left behind begin to sequester carbon in the form of peat, lake drainage may serve as a negative feedback to global warming. However, in our study area, the net C budget for each lake/basin system is dependent on a complex set of thermokarst lake characteristics, lake history, substrate and organic carbon quality, environmental and climate conditions, and subsequent drainage and wetland characteristics (Walter Anthony et al., submitted manuscript, 2011) complicating extrapolation of the role of expanding and draining lakes on the landscape.

[41] Since thermokarst lake dynamics likely factor into landscape-scale carbon fluxes, we must gain a better understanding of the short-term and long-term C dynamics of these systems and regions [Frolking and Roulet, 2007] and incorporate these fluxes into terrestrial greenhouse gas emission scenarios. The balance between expanding lakes and draining lakes on the landscape is important for upscaling carbon emission and sequestration estimates over short as well as long time scales [Hinkel et al., 2003; Zona et al., 2010; M. C. Jones et al., Peat accumulation in drained thermokarst lake basins in continuous ice-rich permafrost, northern Seward Peninsula, Alaska, submitted to Journal of Geophysical Research, 2011]. Thus, further research is needed to more fully understand the role of thermokarst lake dynamics at the landscape-scale and how these prominent lowland Arctic landscapes factor in the northern, high-latitude carbon cycle.

5. Conclusion

[42] Thermokarst lakes are a dynamic component of lowland Arctic landscapes with ice-rich permafrost. Our assessment of lakes and ponds >0.1 ha in a 700 km2 area using high resolution remotely sensed imagery from 1950/51, 1978, and 2006/07 revealed that the majority of thermokarst lakes are actively expanding as a result of surface permafrost degradation. However, as lakes expand the opportunity for drainage increases due to the encroachment toward a drainage gradient. Thus, total surface area of lakes in the study region declined by 15% due to the lateral drainage of several large lakes. Long-term mean expansion rates of thermokarst lakes in the region ranged from 0.35 m/yr and 0.39 m/yr and long-term lake drainage rates from 2.2 lakes/yr to 2.3 lakes/yr in the first (1950/51 to 1978) and second (1978 to 2006/07) observation periods, respectively. Analysis of climate data over the 57 year study period did not reveal any definitive link in regards to the response of thermokarst lakes to climatic forcing. However, given future climate projections, it is likely that thermokarst lake-rich Arctic lowlands will change dynamically as a result of surface permafrost degradation. In turn, this will likely impact the northern high-latitude carbon budget.


[43] This study was supported by NASA grant NNX08AJ37G and NSF IPY grant 0732735. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily represent the views of the National Science Foundation or NASA. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. We thank the National Park Service Fairbanks Office for providing high-resolution satellite imagery of parklands and permits to do fieldwork in the Bering Land Bridge National Preserve. We also thank the GAM and LRS programs of the USGS for additional support. This manuscript benefited from the reviews of Karen Murphy, Amy Larsen, and one anonymous reviewer.