Journal of Geophysical Research: Biogeosciences

Twentieth century erosion in Arctic Alaska foothills: The influence of shrubs, runoff, and permafrost

Authors


Abstract

[1] Recent changes in the climate of Arctic Alaska, including warmer summers and a lengthened growing season, have increased vegetation productivity and permafrost temperatures. In this study, we use (1) time series imagery to examine the landscape pattern of tall shrub distribution and expansion in Arctic Alaska and (2) lake sediments from watersheds where shrub expansion is occurring to compare twentieth century temporal trends between shrub cover, erosion, and runoff. Landsat thematic mapper data from 1986 and 2009 were used to evaluate the expansion of tall shrubs across three regional subscenes in the Arctic foothills in northeast Alaska. We found that tall shrubs occupied floodplains and streams in 1986 and have been expanding their coverage along these corridors. The interaction between shrub expansion and erosion was examined by reconstructing the last 60–100 years of erosion from sediment cores in four lakes with shrub expansion in the surrounding watersheds. Three of the four lake cores show a steadily increasing or fluctuating erosion rate until 1980, after which these cores show a synchronous decline. We postulate that the increase in shrubs since 1980 in landscape positions prone to erosion has contributed to the decline in erosion. A decrease in the magnitude and frequency of runoff events has likely also contributed to the decline in erosion. Our results indicate a general decline in erosion since 1980 that is contemporaneous with shrub expansion and peak runoff decline, punctuated by episodic erosion events in one of four catchments.

1. Introduction

[2] Recent changes in the climate of Arctic Alaska, including warmer temperatures and a lengthened growing season [Chapin et al., 2005; Serreze and Francis, 2006; Shulski and Wendler, 2007], are linked with increases in vegetation productivity and permafrost temperatures. Increased vegetation productivity, as measured using time series of satellite vegetation indices such as Normalized Difference Vegetation Index (NDVI) [Bhatt et al., 2010; Goetz et al., 2005; Jia et al., 2003; Myneni et al., 1997], has been partly attributed to the expansion of shrubs, which has been documented using time series of aerial photography [Sturm et al., 2001; Tape et al., 2006], plot studies [Joly et al., 2007], and shrub growth ring chronologies [Forbes et al., 2010; Hallinger et al., 2010].

[3] The evidence of shrub increase from repeat photography is most reliable for tall, dark Alnus viridis ssp. fruticosa (Siberian alder) shrubs, and evidence from other repeat photography (Figure 1), NDVI trends (herein), and vegetation plot studies [Joly et al., 2007] indicate an expansion of Salix spp. (willow), and possibly Betula spp. (birch) shrubs. Shrub expansion may include clonal reproduction, sexual reproduction [Douhovnikoff et al., 2010], or new shrubs germinating from seeds that have been either frozen or otherwise dormant until recently [Ebersole, 1989]. These various modes of shrub propagation are manifest in a heterogeneous spatial pattern of expansion and a shifting dominance within tundra ecosystems. Here, we focus on patterns of shrub propagation (shrubs >1 m tall; hereafter tall shrubs) and whether it is associated with permafrost aggradation or degradation.

Figure 1.

Repeat photograph pair from the Jago River valley. (top) The old photograph is from 1957 (G. Kunkle and E. Sable). (bottom) The new photograph is from 29 July 2007 (K. Tape). The camera positions are nearly identical in the two photographs, but in the new photograph, willow shrubs have replaced alpine flora and are obscuring the ponds seen in the old photograph. On the left and right are close-up photographs of two foreground rocks seen in the old photograph. Both rocks have migrated downhill. The left rock has retained its initial orientation. The rock on the right has been forced upright by the large willow shrub growing underneath it. Also, the steep slope to the right of the person in the old photograph has collapsed as the underlying permafrost has thawed. Three large boulders in the old photograph are marked in the new photograph. The assumption in this study is that if this gully drained into a lake, then the collapse evident in this photograph would be recorded as a sediment pulse in the lake.

[4] Observations of associated shrub expansion and thermokarst formation in boreal Alaska indicate a coupling between shrub expansion and permafrost dynamics [Jorgenson and Osterkamp, 2005; Jorgenson et al., 2010; Schuur et al., 2007]. Repeat photography from the Arctic Alaska foothills suggests that similar associations of shrubs and permafrost degradation are present in the Arctic (Figure 1), and growing evidence from the Mackenzie River delta region substantiates this connection [Lantz et al., 2009]. It is sometimes unclear whether the shrub expansion or thermokarst disturbance occurs first, but in either case, snow depth is increased by drifting around shrubs, which increases ground temperatures [Schimel et al., 2004; Sturm et al., 2005b] and facilitates permafrost degradation [Nowinski et al., 2010]. Other feedbacks, such as drainage impoundment increasing thermal conductivity, can also facilitate thermokarst development [Jorgenson and Osterkamp, 2005; Jorgenson et al., 2006; Osterkamp et al., 2009].

[5] Borehole evidence shows that permafrost temperatures in Alaska are warming in response to a combination of elevated air temperatures and greater or earlier snowfall during the second and fourth quarters of the twentieth century [Clow and Urban, 2002; Lachenbruch, 1959; Osterkamp, 2007a, 2007b; Shulski and Wendler, 2007]. Thawing the top of the ice-rich permafrost [Jorgenson and Shur, 2007] can wet the soil, reduce shear strength, and cause subsidence, accelerating processes like soil creep [Matsuoka, 2001], ice wedge degradation [Jorgenson et al., 2006], and thermal erosion along gullies [Osterkamp et al., 2009] (Figure 1). Permafrost degradation and subsequent erosion could potentially expose and release large amounts organic carbon currently stored in frozen and unfrozen soils, thereby accelerating feedbacks to climate warming [Tarnocai et al., 2009; Zimov et al., 2006].

[6] The increase in erosion is obvious from catastrophic permafrost degradation, such as thaw slumps that muddy the waters of rivers and lakes [Lamoureux and Lafreniere, 2009; Lantz and Kokelj, 2008]. Reports from northern Alaska and NW Canada using time series of imagery document a deterioration of massive ice wedges and subsequent thermokarst formation on the North Slope coastal plain since 1945 [Jorgenson et al., 2006], and an increase in permafrost related mass movements (hereafter “thaw slumps”) in the Arctic Alaskan foothills tundra [Bowden et al., 2008; Gooseff et al., 2009] and Arctic Canada [Lantuit and Pollard, 2008; Lantz and Kokelj, 2008; Wolfe et al., 2004]. Warm periods from 13,600 to 12,800 calendar years before present (cal yr B.P.) and 11,200 to 10,000 cal yr B.P. caused permafrost degradation and rapid erosion from slopes into stream channels and floodplains of the North Slope [Mann et al., 2010] and Brooks Range (13,500 cal yr B.P. to <12,600 cal yr B.P. (B. Gaglioti et al., unpublished data, 2011)). Alluvial deposition from hillslopes overwhelmed the transport capacity of streams and floodplains, causing them to aggrade [Mann et al., 2010].

[7] The increase in vegetation productivity and shrub expansion, in some cases linked to permafrost degradation, could instead stabilize soil particles [Lewkowicz and Kokelj, 2002] (Figure 2), reduce summer heat flux, and promote permafrost development [Blok et al., 2010]. In northern Alaska during the early Holocene, rapid alder expansion and associated paludification – the accumulation of organic material resulting from aboveground production outpacing decomposition – led to the development of a widespread peat layer and the modern tundra [Mann et al., 2002; Oswald et al., 1999; Walker et al., 2001]. This association between shrub expansion and soil stabilization is in contrast to the association of shrub expansion and permafrost degradation (Figure 1).

Figure 2.

Photographs showing shrub expansion emanating from a drainage along the Nimiuktuk River, in the Noatak basin, from (a) 1948 and (b) 2001. An interesting exercise is to imagine a photograph taken a half-century prior to 1948: about the time that the Little Ice Age waned and the current shrub expansion presumably initiated. Such an extrapolation yields a landscape devoid of large shrubs on the facing slope or along the creek, much like in the old photograph from Figure 1. This supports the notion of drainages as corridors for rapid expansion and dispersion of shrubs over the landscape.

[8] Is the current warming and concurrent shrub expansion on older Arctic landscapes associated with increased or decreased erosion? In this study, we explored this question by using (1) time series imagery to examine the landscape pattern of tall shrub distribution and expansion in Arctic Alaska and (2) lake core sediments from watersheds where shrub expansion is occurring to compare twentieth century temporal trends between shrub expansion and erosion.

2. Methods

2.1. Study Area

[9] The vegetation of the Arctic foothills, located on the north side of the Brooks Range, Alaska, is mostly tussock-sedge, dwarf-shrub, moss tundra, with some erect, dwarf-shrub tundra [Walker et al., 2005], dissected by riparian corridors ranging in size from floodplains to small streams that often host tall shrubs [Beck et al., 2011; Walker et al., 1994]. Most of this foothills region is moist acidic tundra (MAT), while moist nonacidic tundra (MNT) is confined to areas of ongoing loess deposition, limestone bedrock, and naturally disturbed systems, such as floodplains, snow beds, windblown ridges, and recently deglaciated areas. Vegetation succession and possibly changes in climate spurred peat development during the Holocene, and most of the landscape was converted from dry vegetation on mineral-rich loess and till deposits to MAT [Walker et al., 1998]. Older MAT surfaces typically have higher NDVI values, resulting from greater biomass [Munger et al., 2008].

[10] The climate of the Alaskan North Slope is dominated by long, cold winters and short, cool summers. Snow covers the landscape for 7 to 9 months [Benson and Sturm, 1993], is highly variable in distribution, and melts rapidly, constituting the hydrologic event of the season and the onset of the growing season [McNamara et al., 1998]. Snow drifts can persist into August, though the duration of these residual drifts is likely diminishing in response to warmer and longer summers [Euskirchen et al., 2006].

[11] Soils of the Brooks Range and North Slope of Alaska are dominated by continuous permafrost, often within one meter of the surface, except for talik underneath some large lakes and river floodplains [West and Plug, 2008]. Spatial variability of the depth of the active layer and temperature of the permafrost is a function of the air temperature, the thermal properties (e.g., thermal conductivity, heat capacity) of the overlying soil and vegetation, and, when present, snow. Permafrost at the upland/lowland interface (toe slopes and stream channels) is vulnerable to thawing owing to the close proximity of the talik, and ecological processes that inhibit the development of peat. Toe slopes and stream channels also receive considerable runoff from upslope, which transfers heat via flowing water [Jorgenson et al., 2010; Kane et al., 2001].

[12] We selected the Chandler River corridor (Figure 3) for lake coring because it contains abundant shrubs, and because repeat photography shows a large increase in shrub cover since 1950 (28% to 38% = 36% relative increase [Tape et al., 2006]). The section of the Chandler River basin with abundant shrubs has never been glaciated [Kaufman and Manley, 2004], but sediment inputs from weathering Ca-rich rock outcrops provide continual disturbance that prevent peat development near outcrops and along stream channels and floodplains. The primary deciduous shrubs are alder (height = 1–4 m), willow (0.25–4 m), and birch (0.25–1.5 m). The Chandler River and its tributaries had an earlier breakup than other rivers on the North Slope in 2003 (K. D. Tape, unpublished aerial photography, May 2003), and their floodplains host gallery forests dominated by Populus balsamifera along approximately 100 river km, centered around the Ninngolik Valley (K. D. Tape, personal observations, 2001, 2009, 2010). Large shrubs, a longer growing season, and gallery poplar forests collectively suggest this region to be a good analog for a future with improved growing conditions across on the North Slope.

Figure 3.

Locations of SPOT and Landsat imagery, lakes cored, and repeat photography images referred to in the paper.

[13] Approximately 100 km east of the Chandler River are four subscenes from a single Landsat image, used here. Two subscenes (1 and 2) were only glaciated during maximum glacial extent (regionally the Sagavanirktok advance: >125,000 yr), while the other two subscenes (3 and 4) contain some areas glaciated during the late Wisconsinan advance (regionally the Itkillik advance: 24,000 to 11,500 cal yr B.P.) [Kaufman and Manley, 2004].

2.2. Detecting Tall Shrubs and Their Patterns of Distribution and Expansion

[14] Time series of photography overlapping with the Système Pour l'Observation de la Terre (SPOT) imagery and the area of the lake cores (Figure 3), and time series of Landsat imagery from a nonoverlapping area within the general study region (Figure 3), were used to establish patterns of shrub distribution and expansion. Repeat photography from 1948 and 2001 was used to manually delineate 10 polygons (approximately 0.04 ha each) of rapidly expanding shrub patches and 10 polygons of slowly expanding or stable shrub patches on SPOT satellite imagery (bands 2, 3, 4; 2.5 m pixel; 12 July 2008) along a 20 km stretch of the Chandler River. The difference in growth rates deduced from widespread repeat photography is confirmed by shrub growth rings from 22 shrub patches across Arctic Alaska. Shrub rings show greater growth in expanding patches responding to warmer spring and summers, while stable patches show slower growth and only weak relationships to climatic influences (K. D. Tape et al., Landscape heterogeneity of shrub expansion in Arctic Alaska, submitted to Ecosystems, 2011). Using the pixel characteristics of the expanding shrub polygons and stable shrub polygons, ENVI software applied the supervised classification to the SPOT image to interpolate areas of shrub expansion (and stability) throughout the SPOT image.

[15] To validate this method, if only qualitatively, additional remote sensing techniques were applied to imagery of the broader study region. Landsat thematic mapper (TM) data from 1986 and 2009 were used to evaluate the expansion of tall shrubs across four regional subscenes in the Arctic foothills (Figure 3). The subscenes were selected as clear-sky areas from a 6 July 1986 and 5 July 2009 Landsat scene (Path73Row11) and were 25 km by 20 km in size. Images were calibrated and NDVI was computed using red and near-infrared spectral reflectance [Chander et al., 2009] from each subscene. Tall shrub areas, because of relatively high leaf area, typically have the highest NDVI among tundra vegetation types [Jia et al., 2003]. NDVI values were thresholded >0.6 and considered pixels dominated by tall shrubs [Olthof et al., 2008].

2.3. Chandler River Lake Cores

[16] We cored four lakes near the Chandler River on the central North Slope of Alaska (Figure 3). Lakes were chosen with watersheds where shrub expansion was either observed using repeat photography or inferred using the SPOT imagery. Additionally, lakes with small watersheds were selected to limit intrawatershed spatial variation of explanatory factors. In August 2009, using two small boats, a piston gravity corer with clear polycarbonate tubing was used to extract sediment cores 30 cm long and 6.7 cm in diameter from near the center of each lake. Because of logistical constraints of transporting intact cores, cores from all four lakes were sectioned in the field at 0.5 cm intervals to 10 cm depth, 1 cm intervals from 10 to 20 cm, and then 2 cm intervals from 20 to 30 cm. Samples were stored in whirl pak bags and refrigerated for 10 days before being frozen.

[17] Lakes 1 (watershed = 2.0 km2) and 2 (watershed = 1.5 km2) are thaw ponds with evidence of shrub expansion along stream inlets identified using the SPOT imagery. These two lakes showed eroding polygonal ground along one margin. The inlet to lake 1 is incised <1 m and enters the lake at a gentle gradient. The inlet to lake 2 is a steep-sided gully >3 m deep incised into polygonal ground (Figure 4). Alder shrubs 1 to 3 m tall, frost boils, and tussocks are prominent in both watersheds, and smaller willow and birch shrubs are also common. Lakes 3 (watershed = 0.3 km2) and 4 (watershed = 2.7 km2) are located on elevated river terraces at the edge of the valley fill. Lake 3 has a 300 m shrub-covered 12° slope adjacent to it, with gullies and evidence of shrub expansion (Figure 5). Lake 4 has no identifiable inlet, and is also located at the edge of the valley fill. There were no signs of erosion along lake 3 and 4 shores. All lakes were 2.5 m at the deepest point, where the cores were taken, and had nearly flat bathymetry (determined by ≥20 depth measurements), suggestive of a uniform underlying permafrost table.

Figure 4.

(a) A false-color SPOT image showing most of the watersheds for lakes 1 and 2 (dashed line), including the shrub-filled inlets. (b) Supervised classification of Figure 4a into expanding shrub patches (green), stable shrub patches (blue), and nonshrub (white) showing the gully inlet to lake 2 densely vegetated with expanding shrubs. Numbers also denote locations of cores. The image is 2 km wide.

Figure 5.

(a) A small watershed for lake 3 photographed in 1948–1950 (U.S. Geological Survey/U.S. Navy) and (b) the same watershed photographed on 17 August 2004 (Google Earth imagery Google Inc., used with permission), showing a gentle bluff at the edge of the Chandler River valley with shrubby gullies leading down to a lake. Shrub expansion is prominent along the perimeter of the snow drift (dashed lines), one of many such drifts persisting along the valley edge in the larger image (not shown). (c and d) Shrub expansion along the perimeter of residual snow drifts filling slope breaks (these examples from an underlying geologic stratum) is also evident in repeat photography from other locations, along the nearby Oolamnagavik River on 11 August 1948 and 27 July 2002, respectively. Lake 4 is in a setting similar to that of lake 3, 1 km to the north and on the same river terrace.

[18] The cores were dated assuming constant deposition and decay of 210Pb, and mass depositional rates in the cores were used to infer the last 60 to 100 years of erosion in the watersheds. Core analysis and age modeling was conducted by Flett Research Ltd., Manitoba, Canada. Core sections were dried, weighed, and dated using sediment 210Pb activity determined by α-spectroscopy. Typically, spectroscopy was applied to all core sections from 0.0 to 3.0 cm, alternating sections from 3.0 to 6.0 cm, and with decreasing frequency along the remainder of the core. Profiles of sediment 210Pb activity (half-life ∼22 years), decreased logarithmically with depth to supported (background) levels, which was determined statistically [Binford, 1990]. The age at the bottom of a given section was calculated using the constant rate of supply (CRS) model [Appleby and Oldfield, 1978], by taking the natural log of the fraction of unsupported 210Pb (measured in disintegrations per minute, (DPM = Becquerels) in the section to the total unsupported 210Pb below that section. Accumulation rates are then calculated by dividing the mass in a given section by the time to deposit that section.

[19] 137Cs was used as an independent tracer to validate the 210Pb chronology. The maximum 137Cs activity in a core profile typically corresponds to the date of 1963, the year of maximum 137Cs direct deposition of bomb testing radionuclides from the atmosphere in the northern hemisphere. Elevated 137Cs activity above the maximum 137Cs core section would indicate that the majority of the 137Cs is from terrestrial erosion sources. In such cases, the 137Cs maximum can be delayed to the year 1966, the date when maximum 137Cs soil inventory occurred. To have high confidence in the 210Pb model, it should predict a core date between 1963 and 1966 for the 137Cs peak.

[20] Because the lakes have nearly flat bathymetry, we assume that deposition is uniform over the area of the lake. Using this assumption, we scaled the mass of each 0.5 cm section to the size of the lake (as determined in Google Earth) and calculated the mass deposited in each section, which was then averaged over the period of the section to obtain an annual mass influx. That term was then divided by the area of the watershed to obtain an annual average erosion rate for each watershed (g/m2 watershed/yr). These rates averaged over the watershed area are only relevant if the source of the erosion was approximately uniform. If the erosion instead came from point sources, then the spatially explicit values are misleading. The calculations are also misleading if the deposition in the lake is nonuniform (see initial assumption), though temporal changes in deposition at the core location still reflect changes in watershed erosion. In calculating erosion, we assume aquatic production (in a vertical column above the core area) to be approximately equal between lakes and constant over the period of study. We also assume the eolian deposition onto the lake surface to be negligible.

2.4. Core Chronology Statistics

[21] Standard error in the 210Pb activity in each section is converted to standard error in the 210Pb-derived dates ranging from 0.1 years on the recent dates to a maximum of 6.8 years on the oldest date, with a mean of 2.0 years. These are relatively small (all <10%) and not represented graphically, because they are likely outweighed by unknown potential errors resulting from nonconstant 210Pb inputs, and from sediment mixing and 210Pb diffusion in upper sections.

3. Results

3.1. Detecting Tall Shrubs and Their Pattern of Distribution and Expansion

[22] The spatial pattern that emerges from extrapolating the repeat photography using a supervised classification of a SPOT image is one of rapidly expanding shrub patches emanating from floodplains, stream channels, and sedimentary rock outcrops. In contrast, slowly expanding or stable shrub patches are located on gentler tussock tundra slopes and upland benches (Figure 6). Shrub growth rings confirm that rapidly expanding patches have added more biomass than stable shrub patches during the twentieth century (Tape et al., submitted manuscript, 2011).

Figure 6.

Similar pattern of expanding shrubs in two different areas, produced using two methods. (a) Magenta indicates tall shrubs (NDVI >0.6) in 1986, and light green indicates additional tall shrubs in 2009, as evaluated using Landsat imagery (pixels are 30 m by 30 m). (b) SPOT image extrapolation of repeat photography indicates a similar pattern of expanding tall shrubs (green) along stream networks. Images are approximately 4 km wide, and locations of Figures 6a and 6b are shown in Figure 3.

[23] This landscape pattern of expansion is confirmed and refined by the analysis of paired Landsat imagery. Time series of four Landsat subscenes (1986 and 2009) shows an 18% increase in high NDVI (>0.6) pixels. This increase is due to tall shrubs expanding on floodplains, along stream channels, and adjacent to rock outcrops (Figure 6). Tall shrubs in the Landsat imagery are alder and willow. The Landsat imagery better distinguishes between previously existing tall shrub patches and new tall shrub patches, while the extrapolated repeat photography essentially classifies entire patches of tall clumped shrubs as expanding (Figure 6).

[24] One alternative explanation is that the Landsat results might simply be due to interannual variation in canopy production related to climate, with 2009 possibly a more favorable growing season relative to 1986. Under this scenario we would expect the change in NDVI from high NDVI patches that existed in 1986 to be significantly greater in 2009. However, no increase was observed between aforementioned pixels in 1986 and 2009 (Table 1), indicating that the observed increase in the time series of adjacent (high NDVI) pixels was not an artifact of a single enhanced growing season, but rather was an increase in the number of high NDVI pixels between 1986 to 2009.

Table 1. Change in Mean NDVI (±Standard Error) From Tall Shrub Pixels Within Each Framea
 Mean NDVI Within 1986 High NDVI PatchesMean NDVI Within New 2009 High NDVI Patches
6 July 19865 July 20096 July 19865 July 2009
  • a

    Within the extent of the 1986 tall shrub pixels, the 2009 NDVI values are equal or slightly lower than those in 1986. The new tall shrub pixels evident in the 2009 image have, in contrast, changed significantly since 1986, suggesting that those pixels represent new tall shrubs.

  • b

    p < 0.01.

Frame 10.627 ± 0.00020.621 ± 0.00040.567 ± 0.0004b0.616 ± 0.0002b
Frame 20.615 ± 0.00010.602 ± 0.00030.571 ± 0.0003b0.611 ± 0.0001b
Frame 30.627 ± 0.00010.614 ± 0.00020.569 ± 0.0002b0.616 ± 0.0001b
Frame 40.628 ± 0.00010.621 ± 0.00030.566 ± 0.038b0.615 ± 0.014b

3.2. Chandler River Lake Cores

[25] Lake sediment cores 1, 3, and 4 show fluctuating or increasing erosion rates until 1980, after which these cores collectively show a decline (cumulative for 3 lakes: r = −0.47, p < 0.03; Figure 7c), though when lakes are considered individually only the decline in lake 1 is significant. The two lakes located on a river terrace (3 and 4) have similar temporal trends in deposition, showing that between 1904 and 1980 there was a period of increasing erosion (r2 = 0.998, p < 0.001 and r2 = 0.82, p < 0.05, respectively). The lake 1 record has a higher temporal resolution revealing more fluctuations prior to 1980.

Figure 7.

(a) Shrub growth ring widths for the largest stem from expanding (thick line) and stable (thin line) shrub patches, only the former strongly correlated with spring/summer temperatures (Tape et al., submitted manuscript, 2011). (b) A decline in the number of extreme snowmelt discharge events on the Kuparuk River, and a 1 week earlier (1.8 days/decade) peak snowmelt discharge indicative of earlier onset of the growing season (dotted line, r2 = 0.143, p < 0.01, discharge data: U.S. Geological Survey Open File Reports, 1971–2010). (c) Depositional rate in four lakes over the last century. A decline around 1980 in lakes 1, 3, and 4 reflects a decline in erosion in those watersheds. We speculate that trends in shrub growth (Figure 7a) and a decline in peak discharge events (Figure 7b) are contributing to trends in erosion (Figure 7c), except in lake 2, where the deposition is episodic and seemingly independent of trends in Figures 7a or 7b.

[26] The record of deposition in lake 2 is unlike that in lakes 1, 3, and 4. Lake 2 deposition is highly episodic, showing an order of magnitude difference between relative maximum and minimum sediment deposition rates (∼0.3 to ∼0.03 g cm−2 yr−1). Even the lower rates in lake 2 are substantially higher than the deposition rates in the other three lakes (∼.008 g cm−2 yr−1).

[27] Assuming that the lake deposition record can be used to infer average basin erosion rates, we compute that since 1951 (the shortest record in four lakes), watersheds from lakes 1 and 3 were eroded 11.1 and 11.8 g m−2 yr−1, respectively. The lake 4 watershed, a large area with a small lake and no obvious inputs, eroded 0.25 g m−2 yr−1. Using the same computational method, and over the same interval, the watershed of lake 2 eroded 51.6 g cm−2 yr−1, more than twice as much erosion per area as the other three lakes combined.

4. Discussion

[28] Tall shrubs occupy floodplains, gullies, and stream banks, which act as propagation corridors for shrubs across the broader landscape (gullies and streams; Figures 1, 2, and 6). Within the Landsat imagery the tall shrub areal expansion rate was +0.78% yr−1 between 1986 and 2009. This is a change in tall shrub area of +18%, and it is similar to the 0.68% yr−1 areal increase between 1948 and 2002 based on repeat photography reported by Tape et al. [2006] for the Arctic foothills.

[29] One consequence of the warmer spring and summer temperatures in the Arctic [Shulski and Wendler, 2007] is that the growing season in the study area has lengthened, asymmetrically toward spring [Chapin et al., 2005]. A longer growing season means that snow drifts disappear earlier in the spring [Euskirchen et al., 2007; Verbyla, 2008], a trend possibly accelerated by taller shrubs decreasing albedo and further increasing snowmelt temperatures [Chapin et al., 2005]. The same trend toward earlier snowmelt is evident in the Kuparuk River (which flows across Landsat subscene 1), where gauging since 1971 shows a 1 week earlier peak discharge, from 10 May to 3 May (r2 = 0.143, p < 0.01).

[30] Shrub expansion in northern Alaska often occurs along the perimeter of shrinking residual snow drifts (Figure 5), where the growing season has lengthened. Indeed, alder growth rings in expanding patches from Arctic Alaska show positive correlations with current and previous year spring and early summer temperatures, implying a connection between the onset of the growing season and shrub growth (Tape et al., submitted manuscript, 2011). Correlations between temperature in the first half of the growing season and shrub growth have also been observed in Scandinavia [Hallinger et al., 2010] and Siberia [Blok et al., 2011; Forbes et al., 2010]. High Arctic Salix arctica shrubs in Greenland grow more during years of reduced snow extent early in the growing season [Schmidt et al., 2006]. Other studies have shown that the relationship in Arctic Alaska between 1 and 15 June peak NDVI and annual NDVI was not significant, supporting a decoupling of early growing season photosynthesis from peak season photosynthesis [Verbyla, 2008], though this may be explained by graminoids being more sensitive to interannual temperature fluctuations than shrubs [Jia et al., 2006].

[31] Erosion in pristine nonperiglacial settings with precipitation similar to the Arctic is primarily controlled by topography, lithology, vegetation, and precipitation. Precipitation dislodges soil particles and creates runoff, but typically increases plant growth, which in turn stabilizes soil. The net effect is an inverse relationship between precipitation and erosion, though this relationship is not valid at extremely low precipitation, where plants have difficulty growing and stabilizing (nonperiglacial) soils [Dendy and Bolton, 1976; Douglas, 1967; Langbein and Schumm, 1958]. The tundra environment in this study is underlain by continuous permafrost that impedes drainage and causes most soils to be moist or wet [Walker et al., 2002], and plants unlikely to be water limited [Chapin et al., 1989; Tape et al., submitted manuscript, 2011], despite the semiarid climate. The role of precipitation in controlling erosion in continuous permafrost landscapes is therefore confined to eroding and transporting sediment, and transferring heat. Thermal erosion from permafrost degradation is an additional variable of importance in the Arctic environment [Mann et al., 2010]. When examining temporal trends in erosion in an Arctic watershed, we assume topography and lithology to be constant, and the variables controlling erosion become vegetation, runoff, permafrost, and climate.

[32] An increase in shrub cover and vegetation productivity, specifically in landscape positions where the ground was previously exposed to erosion, coincided with and may have contributed to the decline in erosion observed in lake cores 1, 3, and 4 (Figure 7). Beginning in 1980, alder shrub rings from rapidly expanding shrub patches along streams, gullies, and outcrops in Arctic Alaska became much wider than those from alders in tussock tundra (Figure 7a) (Tape et al., submitted manuscript, 2011). Because only the largest stems were cut and analyzed, the decline in the tussock tundra alder rings is predominantly an age trend of decreasing ring width over the record, including few young stems. The lack of decline in the expanding patches since 1980 is due to wide rings from new stems and to maintained growth of aging stems during warmer and longer summers. Based on the timing evident in the ring record, and the strong link between temperature and productivity [Walker et al., 2003a], and between temperature and shrub growth [Chapin et al., 1995; Forbes et al., 2010; Walker et al., 2003b], the pattern of expanding shrubs along drainages was likely concurrent with the erosion decline around 1980. Increased vegetation cover is correlated with reduced erosion in the high Arctic Canadian archipelago [Lewkowicz and Kokelj, 2002], and across North America [Langbein and Schumm, 1958]. Surfaces prone to erosion are those landscape positions without peat or other vegetation blanketing the soil [Mann et al., 2010]. Though our study focused on tall shrub expansion along streams and proximal to shrinking residual snow drifts, a general increase in vegetation productivity [Goetz et al., 2005; Munger et al., 2008; Verbyla, 2008] is likely stabilizing soil in many landscape positions.

[33] A decrease in the magnitude and frequency of peak runoff events may also be contributing to the decline in erosion in lake cores 1, 3, and 4. The largest runoff events of the year occur at snowmelt [Cockburn and Lamoureux, 2008; McNamara et al., 1998], and four of the top five peak annual discharge events for the Kuparuk River occurred between 1971 and 1982, with only one similar event occurring since 1982 (Figure 7b). We assume the peak discharge events, more so than the cumulative discharge, to be erosive [Dugan et al., 2009]. The decline in the magnitude of peak discharge events is contemporaneous with both the erosion decline and shrub expansion. The decrease in stream and river silt depositional events, as inferred from decreased lake core deposition, is consistent with and could be partly responsible for the increase in vegetation and stabilization of bars along floodplains of rivers and streams [Tape et al., 2006]. One explanation for the decline in peak discharge events is that there is less snow (technically, snow water equivalent, or SWE), though it is possible that the mere occurrence of snowmelt earlier in the spring, when solar insolation is smaller, is sufficient to explain a slower, and thus less flashy runoff. Taller shrubs reducing snow surface albedo could also advance the snowmelt date [Chapin et al., 2005; Sturm et al., 2005a].

[34] We consider the erosion trends from lakes 1, 3, and 4 to be slow and steady erosion shaping the landscape, and these trends are probably controlled by changes in vegetation and runoff. In contrast, the erosion trend from lake 2 is episodic, fluctuating tenfold within a decade. Lakes 1 and 2 watersheds have similar topography and are about as similar in vegetation and morphology as two watersheds could be, except that lake 2 has a more incised stream inlet leading up 600 m from the mouth (Figure 4). There are no obvious slope failures in the watershed, so the episodic sediment pulses probably emanate from the gully inlet that is cut through polygonal ground and hosts expanding shrubs (Figure 4). Erosion at the lake margin could also be a source of sediment. So, while trends from lakes 1, 3, and 4 indicate recent soil stabilization resulting from increased vegetation, the episodic record of deposition in lake 2 hints at the potential instabilities and heterogeneity in tundra landscapes. More specifically, if the episodic events are from point sources, then the background erosion rates are, relatively speaking, small and unimportant, because the landscape is being shaped by catastrophic events, rather than gradual processes [Hooke, 2000].

[35] In three of four watersheds studied here, the addition and enhanced growth of shrubs along streams and gullies since 1980 did not coincide with a destabilization of permafrost in those watersheds. If the permafrost degradation represented by Figure 1 were occurring in watersheds of lakes 1, 3, and 4, then an increase, rather than a decrease, in erosion should have been detected, despite concurrent increases in shrubs. The general stability in those three watersheds therefore stands in contrast to Figure 1 and recent reports of increasing permafrost degradation. That contrast may be explained by the recent thermal erosion being concentrated in younger, more recently glaciated terrains [Bowden et al., 2008; Gooseff et al., 2009; Lantz and Kokelj, 2008], such as in Figure 1. Thinner or nonexistent peat layers in the younger terrains may be permitting greater heat flux and permafrost degradation, whereas older terrains with thicker peat layers are most susceptible in limited areas where the peat is not present.

[36] Erosion includes any process that redistributes soil particles, and the coring methods in this study cannot separate erosion due to permafrost degradation from erosion due to water dislodging soil particles. We speculate that the episodic events in lake 2 are the signature of catastrophic thermal erosion, shown to be recently increasing [Bowden et al., 2008; Gooseff et al., 2009; Lantz and Kokelj, 2008], though no trend in the episodic signal was observed in this study. This study highlights the need for a more integrated understanding of the factors that control permafrost stability and erosion. Widespread coring of varying terrains is necessary to assess the post Little Ice Age soil and permafrost stability over large areas, and to correlate erosion trends with landscape and climate variables such as vegetation productivity and runoff events.

5. Conclusion

[37] Our results indicate a background decline in erosion (collectively, in 3 cores) since 1980, superimposed by episodic erosional events (in 1 core). The background decline in erosion is associated with trends of increasing shrubs and declining peak runoff events. In contrast to the positive feedbacks associated with shrub expansion and climate change [Chapin et al., 2005; Mack et al., 2004; Sturm et al., 2005b; Swann et al., 2010], our results suggest a negative feedback from shrubs stabilizing soil (including carbon), though this response could be reversed by an increase in permafrost-related erosional events, or an increase in decomposition [Mack et al., 2004]. Results from lakes 1, 3, and 4 appear counter to the recent reports of increased thaw slump activity, but results from lake 2 are consistent with these recent reports and probably reflect the episodic contribution of thermal erosion to the erosion regime within that watershed. Determining the relative contribution of enhanced vegetation productivity, changes in the hydrologic regime, and permafrost degradation to erosion trends is a challenging problem that warrants attention in future studies.

Acknowledgments

[38] This work was supported in part by a Director's Fellowship from the Institute of Arctic Biology at the University of Alaska Fairbanks, the Environmental and Natural Resource Institute at the University of Alaska Anchorage, and the Water and Environmental Research Center at the University of Alaska Fairbanks. Thanks to the Alaska Stable Isotope Facility for supplying the coring equipment. Partial support was provided by NSF OPP grant 0612534 awarded to Jeffrey M. Welker, and the Bonanza Creek Long-Term Ecological Research program (funded jointly by NSF grant DEB-0620579 and USDA Forest Service, Pacific Northwest Research Station grant PNW01-JV11261952-231). Thanks to Roger W. Ruess for improving this manuscript with his critical review.