Multi-temporal image analysis of historical aerial photographs and recent satellite imagery reveals evolution of water body surface area and polygonal terrain morphology in Kobuk Valley National Park, Alaska

Multi-temporal image analysis of very-high-resolution historical aerial and recent satellite imagery of the Ahnewetut Wetlands in Kobuk Valley National Park, Alaska, revealed the nature of thaw lake and polygonal terrain evolution over a 54-year period of record comprising two 27-year intervals (1951–1978, 1978–2005). Using active-contouring-based change detection, high-precision orthorectification and co-registration and the normalized difference index, surface area expansion and contraction of 22 shallow water bodies, ranging in size from 0.09 to 179 ha, and the transition of ice-wedge polygons from a low- to a high-centered morphology were quantified. Total surface area decreased by only 0.4% during the first time interval, but decreased by 5.5% during the second time interval. Twelve water bodies (ten lakes and two ponds) were relatively stable with net surface area decreases of ≤10%, including four lakes that gained area during both time intervals, whereas ten water bodies (five lakes and five ponds) had surface area losses in excess of 10%, including two ponds that drained completely. Polygonal terrain remained relatively stable during the first time interval, but transformation of polygons from low- to high-centered was significant during the second time interval.


Introduction
Subarctic wetlands are transitional areas between hundreds of thousands of shallow ponds and lakes and the surrounding Figure 1. Context image for the Ahnewetut Wetlands study area. Inset provides geographic context. Yellow contour represents the Ahnewetut Wetlands boundary (Swanson 2001) within Kobuk Valley National Park (thick black contour). Study area is bounded by orange rectangle. Graphic created using the NIR channel (Band 4) of Landsat ETM+, acquired 30 August 2002 (displayed in red channel) and Landsat TM, acquired 6 July 1985 (displayed in green and blue channels). Data source: Global Land Cover Facility. terrestrial environment; they are particularly sensitive to minor shifts in the hydrologic cycle (e.g., Chapin et al 2006;Prowse et al 2006;Schindler and Smol 2006) and thermal regime. Wetlands are highly productive ecosystems that provide surface water and water filtration, flood protection, groundwater recharge through open taliks, carbon storage, and wildlife migration routes and habitat for moose, waterfowl, furbearers, and subsistence (Berkes and Jolly 2001). A warming global climate is expected to accelerate the exchange of mass and energy between the reservoirs that together comprise the regional hydrologic cycle. In recent decades, air and ground temperatures in the continuous and most of the discontinuous permafrost zones in Alaska have increased (e.g., Hinzman et al 2005;Osterkamp 2007). Warming in interior Alaska will affect snowfall and rainfall patterns, permafrost stability and distribution, surface water and groundwater hydrology, water quality, and vegetation dynamics. Pore-occluding permafrost, segregated ice lenses, and massive wedge ice inhibit infiltration and vadose zone drainage (Ford and Bedford 1987). Loss of permafrost and massive ice will increase both the storage capacity and hydraulic conductivity of shallow subsurface reservoirs, and will result in greater connectivity between surface water and groundwater, likely reducing the size and number of shallow subarctic ponds and lakes, altering adjacent drainage networks, and affecting ecosystem services provided by subarctic wetlands.
Extensive changes to shallow ponds and lakes, both a consequence and indicator of permafrost degradation (Osterkamp and Romanovsky 1999), vary along an evolutionary continuum from initial expansion of the number or size of shallow ponds and lakes (Smith et al 2005) to rapid drainage and significant permanent loss of surface water (Marsh et al 2009) and related changes in water body surface area (Yoshikawa and Hinzman 2003, Smith et al 2005, Riordan et al 2006, Roach et al 2011. Based on recent results of a broad survey of 15 lake pairs that exhibit opposing contraction/expansion behaviors, Roach et al (2011) posit that terrestrialization and thermokarsting are the most likely mechanisms causing simultaneous lake area contraction and expansion, with both mechanisms enhanced by recent climate warming.
Important information about the physical and hydrological conditions that drive wetland lake ecosystem dynamics in remote Kobuk Valley National Park (KOVA), northwestern interior Alaska, is lacking due to limited funding and study. Under the framework of the US National Park Service's vital signs monitoring program, park scientists recently observed the effects of local warming on permafrost instability within the Kobuk lowlands, including numerous catastrophically drained ponds and lakes (Larsen and Kristenson 2010). Larsen and Kristenson (2010) identified several examples of old lake coalescence, and both old and recent catastrophic channel breaches.
Various image analysis technologies have been used to quantify water body surface area changes occurring in thermokarst wetlands (e.g., Kaab et al 1997, Yoshikawa and Hinzman 2003, Grosse et al 2005, Jorgenson et al 2006, Riordan et al 2006, Mars and Houseknecht 2007, Hinkel et al 2007, Plug et al 2008, Marsh et al 2009, Sannel and Brown 2010, Sannel and Kuhry 2011. Analyses of high-resolution aerial imagery generally use fully manual approaches to water body surface area contouring, with the exceptions of Hinkel et al (2007) and Jones et al (2009Jones et al ( , 2011. Semi-automated boundary delineation is generally thought to result in an untenable level of misclassifications of land versus water, and yet accurate manual interpretation and mapping of land-water boundaries is a labor-intensive solution (Sannel and Brown 2010) that limits regional-scale, remote-sensing-based permafrost monitoring, partly due to the variable quality of historical aerial photographs. This paper demonstrates a semi-automated approach for using very-high-resolution imagery to quantify subtle landscape changes that are diagnostic of the health of near-surface permafrost in a wetland/tundra environment in the KOVA lowlands. We used high-precision co-registration techniques and active-contouring change-detection analysis to quantify water body surface area changes, and a differencenormalization approach to detect ice-wedge degradation during the periods between acquisition of early-1950s and late-1970s high-to very-high-resolution aerial photographs and post-2000 very-high-resolution optical satellite images. Change-detection analysis of more frequent but midresolution (Landsat) to high-resolution (SPOT) optical satellite imagery cannot identify ice-wedge degradation and thus such image sources were not used in the analysis. Despite potential limitations of semi-automatic interpretation and mapping, such methods can expedite quantitative analysis of changes in subarctic ponds, lakes and adjacent polygonal terrain.

Site description
Kobuk Valley National Park is located in interior northwestern Alaska, USA, at 67 • N (∼50 km north of the Arctic Circle). The low-lying Ahnewetut Wetlands study area (figure 1) is ∼170 km east of Kotzebue, Alaska. In this region, the Kobuk River lies within a hydrologic basin between the Baird and Waring Mountains to the north and south. Jorgenson et al (2008) suggested that the Kobuk River (figure 1) is an approximate dividing line between continuous permafrost to the north and discontinuous permafrost to the south where the Ahnewetut Wetlands are located. This is generally consistent with earlier mapping performed by Ferrians (1965), who indicated the region was just inside the demarcation boundary for continuous permafrost.
Ubiquitous ice-wedge polygons with scales ranging from <10 to 30 m, and subarctic ponds and lakes with scales ranging from 0.2 to 1.5 km, lie at elevations of 9 to 46 m above mean sea level in Kobuk Valley; these features are evidence of the longstanding presence of permafrost in the basin (since at least the Late Glacial Maximum-locally the Walker Lake glaciation) and of local permafrost degradation (Fernald 1964, Hamilton 1982. As permafrost thaws, land subsidence occurs, developing topographic lows in which shallow thaw ponds and lakes may form (Hopkins 1949). While further degradation at the margins may initially increase water body size, continued permafrost degradation may ultimately lead to increased net infiltration and lake drainage if warming is not reversed.
The Ahnewetut Wetlands (figure 1) are a 14 000 ha, roughly triangular-shaped fluvial terrace bounded by the Kobuk River to the north, Ahnewetut Creek to the northwest, and Niaktuvik Creek to the southeast (Kuhry-Helmens et al 1985, Swanson 2001. The relatively smooth fluvial terrace is characterized by small, abandoned meandering fluvial channels in a polygonal peat plateau, and many shallow ponds and lakes (Kuhry-Helmens et al 1985). Regarding surface water bodies in Kobuk Valley, Fernald (1964, p K29) stated, 'Actively caving banks are generally 5-20 feet (1.5-6 m) high and commonly serrated where thawing proceeds more rapidly along ice-wedge polygons. The lakes in general show various stages of development. Some are being enlarged completely around their margins, but others are being filled where thawing and caving is no longer active. Some of the lakes have been partly emptied where an outlet to a lower level has been established. Many of the lakes are migrating through enlarging on one shore and filling on opposite shores'.
Today, many of the surface water bodies in the Ahnewetut Wetlands are associated with polygonal patterned ground of the intermediate-to-high-centered polygon type, with raised central polygons surrounded by subsided, trough-like networks of ice wedges. In contrast, these polygons were low-centered depressions commonly filled with ponds when aerial photographic image data were first collected during the mid-last-century. Shallow wetland ponds and lakes occur within lowland organic-rich loams and peat, and the surrounding ecosystems are dominated by bogs and fens or shrubs and forests (Jorgenson et al 2009). Larsen and Kristenson (2010) describe 5 representative lakes in this system that range in area from 2.6 to 17.3 ha, with a mean lake depth of 1.68 m and maximum depth ranging from 1.4 to 5.8 m. The case study domain (figure 2) ranges in elevation from 24 to 30.5 m above mean sea level; it accounts for less than 20% of the Ahnewetut Wetlands, but contains 7 of its 17 lakes larger than 20 ha.
The climate of Kobuk Valley is moderately continental (Hartman and Johnson 1984), subarctic and semiarid (mean annual precipitation 360 mm; Gibson 2009a) with cold winters (January mean: −20.4 • C) and warm summers (July mean: 15.4 • C; Western Regional Climate Center 2011). Rainfall totals are significant from April to October, and precipitation peaks in August. Snowfall totals are not measured, but are included in 30-year gridded climate normals for Kobuk Valley precipitation (Gibson 2009a). Mean annual air temperature in Kobuk Valley was estimated to be (1) −6 • C based on Ambler West climate station data for a site located 45 km east (Leslie 1989;cf Fernald 1964); (2) −5 • C by the 30-year gridded climate normals for Kobuk Valley air temperature for the period from 1971 to 2000 (Gibson 2009b); and (3) has been measured locally over the last 18 years to be −4 • C (Western Regional Climate Center 2011).
Relatively long climate records are available for the first-order meteorological station located in Kotzebue, Alaska (e.g., figure 3). In contrast, local weather data for KOVA, which are available from the Kavet Creek remote automated weather station (Western Regional Climate Center 2011), are limited to the period from 1992 to present (e.g., figure 3); this station is located ∼18 km from the study area at an elevation 72 m above mean sea level. Air temperature readings are available year-round, but neither cold-season precipitation nor snow depth is recorded. The Kavet Creek mean annual air temperature tends to be ∼1 • C warmer than reported for Kotzebue (figure 3(a)). Mean annual air temperature (MAAT) records indicate a ∼1 • C increase at Kotzebue since 1975 This preliminary analysis estimated a 14% decrease in water surface area within the wetlands, mainly associated with smaller water bodies. The largest 25 lakes in the wetlands had a consistent number of water pixels in the analysis, which suggested little change in wetted lake area during this 23-year period, but the mid-resolution satellite images may have masked changes in shoreline geometry. Larger lakes are able to maintain their size via wave-cut erosion (long fetch translates to high potential wave energy; Sannel and Kuhry 2011) and thaw slumping (large water bodies have higher heat capacities than smaller water bodies).
Recognizing the possible limitations of fully automated classification methods (e.g. Sannel and Brown 2010), we have pursued new semi-automated methods to detect and quantify very small changes in these same surface water bodies and associated polygonal terrain using high-resolution imagery, high-precision orthorectification and co-registration, active-contouring techniques, and a normalized difference index. These analyses span 1951-2005, nearly the entire period of record for this wilderness area.

Data and methods
Lake shorelines have a fractal nature (e.g., Mandelbrot 1967) and their apparent length scale depends on image resolution. The image data used in this analysis were three 1951 black and white (B/W) aerial photographs, a 1978 color infrared (CIR) aerial photograph, a 2005 high-resolution Quickbird satellite image, and two 2008 ALOS PRISM (Advanced Land Observing Satellite Panchromatic Remote-sensing Instrument for Stereo Mapping) images. The B/W aerial photographs were acquired on 26 August 1951 by the US Navy KBV project. The CIR aerial photo was acquired by NASA's Johnson Space Center, project 386, on 5 July 1978. The satellite image is a panchromatic dataset acquired by DigitalGlobe on 18 August 2005 with a resolution of 0.68 m. Two ALOS PRISM images from an image triplet acquired on 29 September 2008 were used to derive a 2.5 m resolution digital surface model (DSM) essential to the orthorectification and co-registration process. Image numbers and camera details are provided in the online supplementary data and methods (available at stacks.iop.org/ERL/8/025007/mmedia).

Orthorectification and co-registration
High-precision orthorectification and co-registration of these images was performed using the COSI-Corr (Co-registration of Optically Sensed Images and Correlation) technique (Leprince et al 2007b) with the 2005 satellite image as the reference. The potential and limitations of this technique have been demonstrated by several studies using satellite-based remote sensing data; these studies included measurement of glacier-surface velocities in mountainous terrain (Scherler et al 2008, Herman et al 2011, co-seismic deformation For this study, the film-based aerial images were scanned with high spatial and radiometric resolution to accurately retrieve their information. Orthorectification and co-registration were performed as described by Leprince et al (2007a) with minimum available information (i.e., scanning resolution, fiducial points coordinates, principal point assumed to be at the crossing of the lines joining the opposite fiducial points) to set the camera geometry.
Preliminary work (Necsoiu et al 2010a(Necsoiu et al , 2010b revealed that even though the aerial photographs had correct quantification, co-registration was problematic because of scarcity of stable tie points. In addition, intrinsic factors such as (a) camera geometry and radial and tangential lens distortions, (b) spatial and radiometric resolution, (c) atmospheric diffraction, and (d) scratches and artifacts are major hindrances to obtaining satisfactory results in registering aerial datasets (Ayoub et al 2009, Necsoiu et al 2010b. These issues were addressed by carefully identifying, selecting and optimizing (i.e., through several iteration loops) stable tie points from the few available relatively stable landscape features. Eight tie points were selected from flat regions of the DSM to anchor the 1951 aerial photo to the 2005 satellite image, and most tie points coincided with relatively stable palsas (figure 5). The resulting average misregistrations were only a few centimeters (e.g., ∼0.02 ± 0.74 m for a B/W image and ∼0.05 ± 1.09 m for the CIR image).

Polygonal terrain analysis
Multi-temporal image analysis of recent changes in the morphology of polygonal terrain is in its infancy; this is one of the first papers to provide a broadly applicable methodology-the normalized difference index-to address the topic. Simple image comparison techniques such as blending/fading, swiping, blinking or flickering are effective for quick visualization of visible changes in ice-wedge polygons, but quantification of morphological change requires a more sophisticated approach. Before implementing a change-detection analysis of polygonal terrain (such as quantification of polygonal joint width), the following conditions should be satisfied (Lu et al 2004).
(2) Precise radiometric and atmospheric calibration or normalization between multi-temporal images.
(4) Selection of the same spatial and spectral resolution images, if possible.
Ideally each dataset will require a priori standardization by normalizing image pixel values for differences in sun illumination geometry, atmospheric effects and instrument calibration. For our case study, lack of camera calibration data made absolute standardization of the aerial data impossible. Instead, assuming that the relationship between the radiance obtained by sensors from regions with constant reflectivity could be approximated by a linear function, a relative radiometric normalization of the datasets was performed via a statistical adjustment approach (Mena and Malpica 2009). The recent 2005 satellite image was used as the reference because it had the best visual quality and the largest dynamic range, maximizing retention of available information (Lo and Yang 1998). The relatively high resolution historical 1951 image was used in the bi-temporal analysis, whereas the low resolution of the historical 1978 image was deemed insufficient for quantitative analysis (figure 6).
To qualitatively and quantitatively identify changes, a normalized difference index (NDI) was calculated as: NDI = NB norm (recent) − NB norm (historical) NB norm (recent) + NB norm (historical) Figure 6. Enhanced tri-temporal images of polygonal terrain (1951,1978,2005). See figure 2 for context. where NB is the normalized brightness of each pixel. NDI is a particular case of the normalized difference reflectance or NDR, an index introduced by Villa et al (2009) to assess bi-temporal land cover changes by exploiting the entire spectral range of panchromatic, multispectral and hyperspectral images. As with NDR, NDI follows a difference-normalization approach, introduced in remote sensing for vegetation studies with the normalized difference vegetation index (Kriegler et al 1969). NDI was calculated with the 1951 dataset normalized to the 2005 dataset for the image area shown in figure 7, which consists of 500 pixels × 500 pixels (∼310 m × 310 m). Using NDI, a preliminary assessment of changes in moisture, identified as an increase or reduction in brightness, was performed without information loss, in contrast to other change-detection techniques such as indices, compression or feature reduction. NDI ranges from −1 to +1, where −1 indicates maximum reflectance decrease, 0 is no change, and +1 indicates maximum reflectance increase. A threshold NDI of −0.6 was used to illustrate low-lying polygonal joints in which wedge ice had melted; likewise, a threshold of +0.6 was used to illustrate land that had become relatively high and dry. These selected thresholds were based upon image interpretation and professional judgment in consultation with David Swanson (US National Park Service).

Water body surface area analysis
Using the 1951, 1978, and 2005 image data, a series of algorithms were applied to define water body surface area changes in Ahnewetut Wetlands; 7 ponds (initially 0.09-0.5 ha) and 15 lakes (initially 2-180 ha) were assessed. All lakes in the study area larger than 2 ha were assessed, but fewer than half the ponds were considered (figure 2). The quantitative solution for assessing water body surface area change consisted of (1) a data-clustering algorithm with nonparametric density gradient estimation (Fukunaga and Hostetler 1975); (2) a Canny-Deriche edge-detection algorithm (Deriche 1987) to identify the shoreline (this commonly used algorithm is not discussed further); and (3) an active-contouring 'snake' algorithm (based on a deformation model originally proposed by Boudier 1997) to contour the shoreline. The nonparametric estimator of density gradient, essential to preprocessing, was used as a discontinuity-preserving filter. Using it, the set of pixels within a spatial radius and defined gray distances for each pixel were determined. Based on this set of neighboring pixels, a new spatial center and mean gray value was computed. The calculated mean values serve as the center for the next iteration; iterations continue until the spatial and grayscale mean values stop changing (Barthel 2007).
Active contours (Kass et al 1988) are widely used in image processing for applications that include edgedetection object recognition, shape modeling and segmentation (Abrantes and Marques 1996, Boudier 1997, Terzopoulos and Fleischer 1988, Xu and Prince 1998  2002), to cite a few of many such studies. Active contours, or snakes, are curves defined within an image domain that can move under the influence of internal forces coming from within the curve itself and external forces computed from the image data (Xu and Prince 1998).
Our method uses a geometric active-contouring technique developed by Boudier (1997) that circumvents elasticity constraints and the gradient-based driving forces that prevent mapping embayments (Andrey and Boudier 2006).
Compared with traditional active-contouring algorithms, this implementation can (1) start far from water body boundaries and converge to boundary concavities, or (2) start near some boundaries while being far from others, yet still converge to boundary concavities. The active-contouring algorithm is initiated by manually creating a seed oval or polygon snake within the water body image feature (figure 7). A polygonal seed may be defined for water bodies having complex geometry or affected by emergent vegetation with shallow shorelines. The algorithm concludes when the snake has expanded to positions adjacent pixels used as the edge threshold, resulting in a bounding contour that is shifted approximately one pixel inside the water body image feature (figure 7). This systematic shift yields water body area estimates that could be slightly smaller than if manually digitized, but these estimates are objective and consistent between water bodies and images, unlike manual digitization estimates. The image processing sequence is illustrated in figure 8. The results of the quantitative bi-temporal NDI analysis are illustrated in figure 9. Blue areas with NDI values less than −0.6 represent polygonal joints that had lower brightness values, consistent with relatively wet, melting wedge ice. Yellow areas with NDI values greater than 0.6 represent land that had became brighter, consistent with relatively dry, high-centered polygons. Overall, these observations are consistent with a transformation from low-centered to intermediate-to-high-centered polygons between 1951 and 2005. Although the cumulative distribution of NDI values approximates a normal distribution (dashed curve), ∼63% of the area had lower brightness values (figure 9). Using the ±0.6 NDI thresholds, troughs of melted wedge ice account for ∼3% of the 2005 image area, whereas polygonal centers that became drier account for ∼4% of the image area.

Polygonal terrain
Similar changes are shown in figure 10, which illustrates a classic example of ice-wedge degradation. Thermokarst pits that formed within a birch forest (D in figure 10

Water body surface areas
The shorelines of 22 water bodies were defined for each image, and per cent change in surface area for each time period were computed ( figure 11, table 1). For those who wish to compare these results against those of other studies that restricted analyses to lakes exceeding a certain size threshold, of the water bodies analyzed, seven were ponds smaller than 0.5 ha, seven were lakes ranging from 2 to 8 ha, four were lakes ranging from 13 to 28 ha, and four were large lakes in excess of 40 ha ( figure 12, table 1). Water body surface area trajectories are shown in figure 12 as cumulative per cent change since 1951 versus surface area.
Little change was detected during the 1951-78 timeframe with the exceptions of Lake 2, which increased in surface area by 15%; Lake 15, which decreased in surface area by 17%; and Pond 6, located at the edge of a meander bend where drainage by stream piracy may have played a role, catastrophically decreased in surface area (figures 11, 12 and table 1). The net water body surface area across the study domain was essentially stable during this early period (table 1), with large relative losses in small ponds balanced by small relative increases in large lakes (figure 12). Significant decreases in surface area of most shallow ponds occurred between 1978 and 2005; Ponds 4 and 6 went dry (table 1 and  figure 12).
Lake 18 expanded slightly during the first period and contracted equally during the second period (table 1 and figure 12). Four lakes occupying a local basin in the western part of the study domain decreased in surface area (Lakes 14,15,21,and 22); Lakes 14, 21 and 22 are linked by incised channels that drain to the north ( figure 12, inset). Two lakes greater than 25 ha (Lakes 20 and 21) experienced a significant decrease in surface area (table 1 and figure 12); Lake 20 occupies a local topographic high in contrast to basin-bound Lake 21. Water body surface area losses tended to be focused within (1) 'highland' ponds near Niaktuvik Creek and (2) basin-bound lowland lakes linked by incised channels. Three lakes greater than 40 ha (Lakes 1, 10 and 11) slightly increased in surface area throughout the 54-year period of record (table 1). Arctic lakes larger than this threshold are relatively stable in response to perturbations (Smith et al 2005, and their online supplementary material). Overall, 12 of 22 water bodies (10 lakes and 2 ponds) were relatively stable with net surface area decreases of ≤10% whereas 10 water bodies (5 lakes and 5 ponds) exhibited surface area losses in excess of 10% ( figure 12). The net change in water body surface area within the study domain was −5.9% (table 1).
Although orthorectification and co-registration are not required to measure changes in water body surface areas, they are required to measure lateral migration rates of lakes using the active-contouring method. Subtle changes were noted along the western shore of Lake 1 (informally named Child Lake, figure 13), including (1) shoreline advancement as emergent vegetation colonizes shallow water (vegetated area between yellow and cyan contours) and (2) very limited lake expansion. In stark contrast, differential eastward expansion of Lake 1 is apparent by inundation observed along its eastern shoreline (inundated area between yellow and cyan contours). Lake 1 likely expanded along ice-wedge exposures on the steeper/deeper shorelines (i.e., along much of the northern, eastern, and southern shores) by thaw slumping or wave-cut erosion (cf Pelletier 2005). Preferential eastward shoreline expansion may be in response to (1) differential insolation during the heat of the day when the sun is in the western sky or (2) frequent westerly winds that occur in open water season. The rate of erosion averaged over 54 years varies from less than ∼0.03 m yr −1 (i.e., the co-registration accuracy of the method) to ∼0.5 m yr −1 (<∼0.3 to ∼5 m/decade). The ∼5 m/decade erosion rate is similar to the 7.3 m/decade lateral expansion rate calculated by Sannel and Kuhry (2011) for a lake in the Hudson Bay lowlands at the southernmost boundary of the continuous permafrost zone; peat thickness is considerably greater at the Hudson Bay location (1.5 m compared to 0.5 m adjacent to Lake 1) and mean annual air temperatures are ∼2 • C cooler.
Lake 2 (figure 13) shows preferential westward and northwestward expansion (figure 12 and table 1) in the Figure 11. Panchromatic 2005 DigitalGlobe image draped with water body surface area contour lines estimated for three imagery dates. Yellow-1951Yellow- , magenta-1978Yellow- , cyan-2005. See figure 2 for context. direction of Lake 1 (figure 8(d)) suggesting that the peat plateau between the water bodies is susceptible to permafrost degradation. Based on field characterization, soil appears to erode along the lake margin, becomes entrained in the water, and is redeposited on the eastern edge of this small lake where newer margins are observed. In the absence of field data, these shoreline changes would not have been apparent using lower resolution images or without precise image co-registration. At the maximum rate of advancement, and given that the land bridge is 30-100 m wide, these lakes may coalesce in as few as 30 years if the ground temperature of the land bridge warms while the land area declines.
Challenges specific to surface-water-body-change analysis include the presence of emergent vegetation, shallow shorelines, and opacity of sediment discharge areas (cf Riordan et al 2006, Sannel andBrown 2010) because affected areas produce reflection values similar to those of bare soil. The active-contouring algorithm may underestimate lake area when faced with complicating factors like these, necessitating human-in-the-loop interpretation.

Discussion and conclusions
The wetlands in KOVA are remote and relatively untouched by humankind, which suggests that ongoing permafrost degradation in the region is a result of climatic factors that are disturbing the ground thermal regime, whether short-term .5 e −5.9 e a Per cent change between subsequent images is based upon the water body surface area of the preceding image; therefore, the overall per cent change is not necessarily the sum of the per cent changes of the 1978/1951 and 2005/1978 pairs. b Pond dried completely by 2005. c Lake geometry issues resulted in considerable uncertainty. d Difficult to define the shore in the W, NW, and N sides of the lake. e Per cent change in total water body surface area. seasonal variability, the multi-decadal PDO, anthropogenic climate change, or some combination thereof. While the PDO switched from a negative (cool) to positive (warm) phase ∼2 years before the 1978 image was acquired, some minimal lag time might be expected between the onset of enhanced thermal forcing associated with the new warm phase of the PDO and quantifiable morphological changes evident in imagery. Thus, it is fortuitous that the limited imagery available for this analysis is so well-matched to the 1976 PDO inflection point. Locally varying antecedent conditions, including lithology, ground ice distribution, hydrology, water body volume, vegetation, slope aspect and topography affect local active layer, snow, and ice thicknesses and the ground thermal regime-causing disparate evolutionary trajectories of water bodies within a single wetland; some bodies expand while others contract or drain completely when a local, critical threshold of instability is met (Marsh et al 2009). The observed decreases in small pond surface area while several large lakes exhibited slight growth is consistent with the understanding that ponds are more sensitive than large lakes to thermal perturbations; nevertheless, several large lakes had significant declines in surface area.
The total water body surface area across the study domain was essentially stable from 1951 to 1978 (figures 11 and 12), decreasing only 0.4% (table 1) during a relatively cool climate period (figure 3), with large relative losses in small ponds offset by small relative increases in large lakes. However, during the subsequent interval from 1979 to 2005-associated with a relatively warm climate and increased MAP (figure 3)-total water body surface area decreased by 5.5% (table 1), 16 of 22 water bodies exhibited decreased surface area, and polygonal terrain morphology evolved from low-centered polygons to intermediate-to-highcentered polygons. Overall, 12 of 22 water bodies (10 lakes and 2 ponds) were relatively stable with net surface area decreases of ≤10%, including 4 lakes that gained area during both time intervals; however, 10 of 22 water bodies (5 lakes and 5 ponds) had surface area losses in excess of 10%, including 3 lakes and 2 ponds with losses in excess of 20% and 2 ponds that drained completely.
Ice-wedge polygons observed in the Kobuk Valley lowlands near the present-day continuous permafrost zone boundary developed over hundreds to thousands of years. The low-centered polygons visible in the 1950s-70s aerial imagery may document well-preserved ice-wedge relics of a colder climate, but the intermediate-to-high-centered polygons observed in the 2005 satellite imagery illustrate that significant melt of wedge ice has occurred, which implies retreat of the permafrost table (Davis 2001), at least locally. Overall, the changes to polygonal terrain morphology that occurred between 1978 and 2005 represent a rapid orange-net surface area decrease of >20%; red-catastrophic drainage. Typical lake analysis size thresholds are presented as vertical lines at 2, 10, and 40 ha. Inset water body map based upon Creely and Brabets (2000). rate of permafrost degradation relative to generally accepted ice-wedge construction rates.
Co-registration of arctic images is particularly challenging due to a general lack of stable landmarks, but palsas may be relatively stable and therefore useful as tie points. Recognition of relatively stable landscape features during the timeframe of interest can yield a high-quality co-registration. The high-precision orthorectification and co-registration method and use of the normalized difference index for detecting changes in polygonal terrain is relatively fast to implement; thus, it has potential for use in production mode given an adequate supply of multi-temporal, very-highresolution imagery over large regional areas. Multi-temporal analysis also can be used to identify, characterize and quantify expansion/contraction/migration of shallow, subarctic ponds and lakes to reveal important multi-decadal changes in arctic wetland geomorphology. The ability to accurately rectify and co-register aerial photographs and satellite imagery expands the time horizon over which changes in arctic landscapes can be analyzed.
Active-contouring change-detection methods used to objectively estimate water body surface areas can best be described as semi-automatic because they also involve human-in-the-loop preprocessing and interpretation. When quantifying lake migration, human intervention is needed during the co-registration step to identify stable tie points and during the initial stage of lake shoreline mapping to construct one rough shoreline polygon (i.e., 'snake') per lake to be optimized during automated active contouring. Despite the potential limitations of semi-automatic methods, they can be used to expedite quantitative analysis of changes in subarctic ponds, lakes and adjacent polygonal terrain.
As water bodies drain and surface water is redistributed from polygon pools on the tundra to subsided polygonal troughs, wetland and aquatic habitat is rapidly converted to terrestrial habitat in the KOVA lowlands. Results from this remote sensing study can serve as baseline information about vital signs and decadal-scale vital-sign trends to inform strategic planning for monitoring the health and integrity of the KOVA wetlands. These analytical methods are also widely applicable to other conservation units in the region, and have widespread implications for management and wetland protection throughout northwestern Alaska. Semi-automated image analysis permits vital-sign quantification with respect to water body surface area and distribution, numbers of shallow ponds and lakes, and morphological indicators of permafrost characteristics. Figure 13. Lake 1 change detection (yellow-1951, cyan-2005); see figure 2 for context. East and west of Lake 1, old, abandoned meandering stream channels traverse terrain that generally has indistinct to no visible polygons. Indistinct polygons may represent an advanced stage of permafrost degradation where soil slumping has infilled previously deeply thawed polygonal troughs (Jorgenson and Osterkamp 2005), resulting in terrain leveling. Lake expansion was greatest in 2005 at locations where polygonal troughs were particularly deep and distinct. Where polygons are indistinct or lacking, little lake expansion is observed. The northern third of Lake 1 appears to be surrounded by a drained, thermokarst basin, suggesting it is currently smaller than its maximum extent and that it may have been directly connected at one time to Lake 10 (figure 12) located due north. Topographic downgradient in the image area is to the NNE.