Polygonal tundra geomorphological change in response to warming alters future CO2 and CH4 flux on the Barrow Peninsula
Abstract
The landscape of the Barrow Peninsula in northern Alaska is thought to have formed over centuries to millennia, and is now dominated by ice-wedge polygonal tundra that spans drained thaw-lake basins and interstitial tundra. In nearby tundra regions, studies have identified a rapid increase in thermokarst formation (i.e., pits) over recent decades in response to climate warming, facilitating changes in polygonal tundra geomorphology. We assessed the future impact of 100 years of tundra geomorphic change on peak growing season carbon exchange in response to: (i) landscape succession associated with the thaw-lake cycle; and (ii) low, moderate, and extreme scenarios of thermokarst pit formation (10%, 30%, and 50%) reported for Alaskan arctic tundra sites. We developed a 30 × 30 m resolution tundra geomorphology map (overall accuracy:75%; Kappa:0.69) for our ~1800 km² study area composed of ten classes; drained slope, high center polygon, flat-center polygon, low center polygon, coalescent low center polygon, polygon trough, meadow, ponds, rivers, and lakes, to determine their spatial distribution across the Barrow Peninsula. Land-atmosphere CO2 and CH4 flux data were collected for the summers of 2006–2010 at eighty-two sites near Barrow, across the mapped classes. The developed geomorphic map was used for the regional assessment of carbon flux. Results indicate (i) at present during peak growing season on the Barrow Peninsula, CO2 uptake occurs at -902.3 106gC-CO2 day−1 (uncertainty using 95% CI is between −438.3 and −1366 106gC-CO2 day−1) and CH4 flux at 28.9 106gC-CH4 day−1(uncertainty using 95% CI is between 12.9 and 44.9 106gC-CH4 day−1), (ii) one century of future landscape change associated with the thaw-lake cycle only slightly alter CO2 and CH4 exchange, while (iii) moderate increases in thermokarst pits would strengthen both CO2 uptake (−166.9 106gC-CO2 day−1) and CH4 flux (2.8 106gC-CH4 day−1) with geomorphic change from low to high center polygons, cumulatively resulting in an estimated negative feedback to warming during peak growing season.
Introduction
High latitude permafrost landscapes have accumulated vast amounts of carbon over thousands of years (Smith et al., 2004; Schuur et al., 2008; Strauss et al., 2013), and now contain ~50% of the world's soil organic carbon pool (Tarnocai et al., 2009). Projections indicate these regions will experience greater increases in temperature, precipitation, and growing season length than elsewhere on the globe (Kattsov et al., 2005; Stocker et al., 2013), which suggests that the arctic soil carbon pools may be impacted. There are considerable uncertainties about the present and future state of the atmospheric CO2 sink strength of arctic tundra at the pan-arctic scale (Mcguire et al., 2012), and the associative role climate change may have on land cover change (Pearson et al., 2013). Therefore, it is important to conduct more tangible and comprehensive landscape-level studies in data-rich subregions of the Arctic tundra biome to evaluate how warming is likely to impact carbon exchange trends and dynamics in response to future climate and land cover change scenarios, and how such responses may enhance or mitigate greenhouse warming.
This study is focused on the Alaskan Arctic coastal plain tundra spanning the Barrow Peninsula, which is a mosaic of ice-wedge polygons, meadows, ponds, lakes, and rivers that are irregularly distributed across Drained Thaw Lake Basins (DTLBs) and interstitial tundra. Sub-surface ice-wedge growing and thawing and related heaving and subsidence of the ground surface in arctic coastal plain tundra influence geomorphology and fine-scale surface microtopography, to affect local hydrology (Liljedahl et al., 2011, 2012), snow pack depth and density (Kershaw, 2008), plant community composition (Villarreal et al., 2012), soil organic carbon storage (Bockheim et al., 2001), and associated land-atmosphere CO2 and CH4 fluxes (Rhew et al., 2007; Olivas et al., 2011). Although the relative distribution of lakes, DTLBs, and interstitial tundra on the Barrow Peninsula is well-established (Hinkel et al., 2003; Frohn et al., 2005), the spatial distribution of smaller geomorphic types (i.e., high, flat, low center polygons) which control fine-scale processes, is generally unknown. Although recent progress has been made (Muster et al., 2012, 2013; Skurikhin et al., 2013), difficulty in discriminating between relatively homogeneous moisture and vegetation cover in tundra geomorphic types is challenging. Identifying the spatial distribution of geomorphic types in arctic coastal plain tundra where landscape heterogeneity is high, is essential to reduce uncertainty in model applications and estimates of annual/seasonal carbon exchange.
Arctic coastal plain tundra can evolve geomorphologically over long- and short-time scales. The thaw-lake cycle acts over long-time scales, typically forming over centuries to millennia (i.e., over one to several thaw-lake cycles; Billings & Peterson, 1980). Although many revisions have been made after its initial formulation (Jorgenson & Shur, 2007), this process successively transforms one land form to another, beginning with lake formation and ending with lake drainage (Billings & Peterson, 1980). Specific to the Barrow Peninsula, approximately 28% of this region (i.e., interstitial tundra) has been unaffected by thaw-lake processes during the last 15–20 ka (Hinkel et al., 2003; Eisner et al., 2005). However, the vast majority of surficial features are younger than 5 500–8 300 years and have likely not completed a single thaw-lake cycle (Brown 1965, Hinkel et al., 2003), suggesting geomorphic change is ongoing. Alternate hypotheses have been presented to explain potentially noncyclic geomorphic changes in other arctic coastal tundra sites where ice content and subsurface properties differ (Jorgenson & Shur, 2007; Jones et al., 2012). Abrupt or relatively rapid (i.e., short-time scale) geomorphic changes can occur in these landscapes with altered drainage patterns as a result of (i) thermal erosion along coasts, river and stream margins, and adjacent to slumps and gullies (Berhe et al., 2007; Fortier et al., 2007), (ii) landscape drying (Smith et al., 2005; Lin et al., 2012), and (iii) thermokarst (i.e., ground subsidence) formation which produces pits, troughs, and small lakes (Mackay & Burn, 2002; Cory et al., 2013; Raynolds et al., 2014).
Recent observations suggest ice-wedge degradation may be increasing as mean annual ground temperatures have increased 2–5 °C over past decades (Romanovsky et al., 2003; Shiklomanov et al., 2010), causing a substantial redistribution of surface water, that transitions geomorphology from low to high center polygons. For example, low to high center geomorphic transitions have been observed in response to (i) abrupt increases in thermokarst pits and trough depths, from 1982–2001 near the Colville River, AK (Jorgenson et al., 2006), from 1990–2010 near Prudhoe Bay, AK (Raynolds et al., 2014), and from 1978–2005 near Kobuk Valley National Park, AK (Necsoiu et al., 2013), associated with recent warming, and (ii) accelerated thermal erosion by gullies or streams in DTLBs near Barrow, AK (Britton, 1957), along the Meade river in AK (Billings et al., 1978), and in Bylot Island, CA (Fortier et al., 2007). Thermokarst formation is likely to occur rapidly (e.g., over 10–20 years) and quickly stabilize with the accumulation of vegetation and organic matter (Jorgenson et al., 2006; Gamon et al., 2012), while altering landscape geomorphic structure. However, the timing, direction, and magnitude of climate feedback to warming associated with long and short-term landscape change processes in arctic tundra regions is highly uncertain (Luo et al., 2011; Anthony et al., 2012; Mcguire et al., 2012).
To date, a variety of plot and landscape-level CO2 and CH4 exchange studies near Barrow have focused on characterizing peak growing season (i.e., mid July to early August) carbon exchange using various land cover or geomorphic units across the Barrow peninsula, such as soil moisture (Rhew et al., 2007; Von Fischer et al., 2010), plant functional type (Brown et al., 1980), plant community composition (Lara et al., 2012), microtopographic position (Olivas et al., 2011; Zona et al., 2011), and DTLB age and/or interstitial tundra (Zona et al., 2010; Zulueta et al., 2011; Sturtevant & Oechel, 2013). These studies find differences in CO2 and/or CH4 exchange over small or large spatial scales associated with varying soil moisture and microtopographic variation. However, each study specifically and intensively focused on a particular land cover unit or spatial scale and no study has integrated spatial patterns in carbon flux from plot to landscape scales in a manner that is appropriate to validate previous estimates and reduce uncertainty of landscape-level estimates (Luo et al., 2011).
This study evaluates the future impact of 100 years (e.g., 2002–2102) of geomorphic change scenarios on peak growing season carbon exchange including: (i) successional geomorphological changes associated with the thaw-lake cycle and (ii) thermokarst pit formation spanning incremental levels (10%, 30%, and 50%) of those reported near the Colville River (Jorgenson et al., 2006) and Prudhoe Bay (Raynolds et al., 2014). We hypothesized that: (i) successional landscape-level changes caused by the thaw-lake cycle will offset carbon uptake/loss as new polygons develop, while others are degraded, (ii) increased thermokarst pit formation will decrease carbon uptake with geomorphic transitions from wetter low center polygons to drier high-centered polygons, and increase CH4 flux with greater inundated soils caused by pit formation, resulting in a significant reduction of the overall sink strength of the tundra due to the extensive rise in CH4 flux. We develop a 30 × 30 m tundra geomorphology map of the Barrow Peninsula and spatially upscale plot-level fluxes by polygon type in combination with estimates of geomorphic thaw-lake succession and scenarios of thermokarst degradation, to estimate 100-year change in peak growing season fluxes. We assess change in peak growing season CO2 and CH4 fluxes solely associated with geomorphic landscape change associated with the thaw-lake cycle and scenarios of thermokarst formation on the Barrow Peninsula. This study is a contribution to the Next-Generation Ecosystem Experiments (NGEE-Arctic) and draws from data collected by the International Polar Year-Back to The Future (IPY-BTF) project.
Materials and methods
Study site
The ~1800 km² Barrow Peninsula lies at the northernmost tip of the Arctic Coastal Plain on the North Slope of Alaska. The region, is underlain by continuous permafrost >400 m thick (Sellmann & Brown, 1973), with a maximum thaw depth ranging from 30 to 90 cm (Nelson et al., 1998; Hinkel & Nelson, 2003). Mean annual air temperature, precipitation, and snowfall are −11.2 °C, 115 mm, and 958 mm, respectively (climate normal 1981–2010; Alaska Climate Research Center:www.akclimate.org). The region has warmed by approximately ~3 °C since 1950, with the majority of warming occurring since the mid-1980s (Lachenbruch & Marshall, 1986; Stafford et al., 2000; Romanovsky et al., 2002). Barrow-area soils are ice-rich, with silty, deltaic, and loess deposits, which are regionally identified as, ‘true thaw lakes’ (Jorgenson & Shur, 2007) and the thaw-lake cycle is the primary catalyst of landscape change within lakes and DTLBs that cover ~72% of land area (Hinkel et al., 2003; Frohn et al., 2005). The landscape generally has a low relief with elevations ranging from 0–22 m.a.s.l. (Brown et al., 1980). Higher elevated regions are accented by erosional remnants that emerged following marine regressions (Brighamgrette & Hopkins, 1995; Eisner et al., 2005). Seasonal freeze-thaw cycles have influenced ice-wedge geomorphological structure, creating a distinct array of geomorphic types and associated plant communities (Webber, 1978; Villarreal et al., 2012).
Polygonal tundra geomorphic types
- Drained Slopes (DS) are characterized by very high relief found on the margins of DTLBs, lakes, and rivers. These geomorphic types have dry soils, low productivity, and high albedo. Two plant communities dominate DS dry lichen heath and dry Arctagrostis, Luzula, Poa, Carex graminoid tundra.
- High-Center (HC) polygons have a relatively high relief and are generally found in well drained interstitial tundra regions or elevated sections of old-ancient DTLBs. Relief results from the thawing of ice-wedge troughs surrounding low- or flat-centered polygons, with the polygon center remaining as a topographic high once thermo-erosion has lowered the troughs. Dry Arctagrostis, Luzula, Poa, Carex graminoid tundra dominates this geomorphic type.
- Flat-Center (FC) polygons have an intermediate relief and are most common in dry-moderate soil moisture regimes. Plant communities of FC polygons include dry Arctagrostis, Luzula, Poa, Carex graminoid tundra and moist Carex, Poa, Luzula graminoid tundra.
- Low Center (LC) polygons are similar geomorphologically to HC polygons with the exception of having a submerged moist-aquatic center. Surrounding this moist-aquatic center are dry-moist rims. Four plant communities typically occur within this geomorphic type and are strongly influence by microtopography. On polygon rims: dry Arctagrostis, Luzula, Poa, Carex graminoid tundra is common whereas in polygon centers: moist Carex, Poa, Luzula graminoid tundra, wet Carex, Sphagnum graminoid tundra, and seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra dominate.
- Coalescent Low Center (CLC) polygons occur through erosional and fragmentation of LC polygon rims, which creates a series of interconnected ponds or troughs.. This geomorphic type is often found within old-ancient DTLBs. The wet-aquatic soil moisture status of this geomorphic type results in seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra and aquatic Arctophila, Carex, Dupontia graminoid tundra being common.
- Meadows (Mdw) are most common in recently drained areas such as Young-Medium age DTLBs, where ice-wedge networks have a minimal development. This geomorphic type is characterized by moist-wet soils, low relief, and high productivity (Hinkel et al., 2003). Three plant communities dominate Mdw including moist Carex, Poa, Luzula graminoid tundra, wet Carex, Sphagnum graminoid tundra, and seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra.
- Troughs (Tr) are drainage channels that are found on the perimeter of HC, FC, LC, and in some cases CLC polygons. These channels can have substantially different soil moisture ranging from moist to aquatic dependent on local drainage patterns. Plant communities can include Arctagrostis, Luzula, Poa, Carex graminoid tundra, moist Carex, Poa, Luzula graminoid tundra, wet Carex, Sphagnum graminoid tundra, and seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra.
- Ponds or thermokarst ponds are generally small and shallow (<0.5 m), and result from a range of geomorphic processes. Ponds typically expand from LC polygon centers as a result of thermal erosion of the upper layer of permafrost. Subsidence occurs as ground ice melts leading to an increase in standing water area and depth(Hobbie, 1980). The dominant plant community found in this geomorphic type is aquatic Arctophila, Carex, Dupontia graminoid tundra.
- Lakes or thermokarst lakes form when ponds coalesce due to areal expansion or ground subsidence. Lakes form an elliptical shape with an elongated north–south axis due to wind produced circulation cells that erode banks (Livingstone, 1954).
- Rivers and streams transport freshwater and dissolved solids from the terrestrial environment into surrounding Oceans.
| Geomorphic Unit | Moisture Regime | Relief | Vegetation Community |
|---|---|---|---|
| Drained Slope | Very Dry | Very High | (i) Dry Lichen Heath, (ii) Dry Arctagrostis, Luzula, Poa, Carex graminoid tundra |
| High-Center | Dry | High | (i) Dry Arctagrostis, Luzula, Poa, Carex graminoid tundra |
| Flat Center | Dry-Moist | Intermediate | (i) Dry Arctagrostis, Luzula, Poa, Carex graminoid tundra, (ii) Moist Carex, Poa, Luzula graminoid tundra, |
| Low Center | Moist-Wet | Intermediate | (i) Moist Carex, Poa, Luzula graminoid tundra, (ii) Wet Carex, Sphagnum graminoid tundra, (iii) Seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra |
| Coalescent Low Center | Wet | Intermediate | (i) Seasonally flooded Carex, Dupontia, Eriophorum graminoid tundra, (ii) Aquatic Arctophila, Carex, Dupontia graminoid tundra |
| Meadow | Moist-Wet | Low | (i) Moist Carex, Poa, Luzula graminoid tundra, (ii) Wet Carex, Sphagnum graminoid tundra, (iii) Seasonally flooded Carex, Dupontia, Eriophorum graminoid |
| Trough | Moist-Aquatic | Low | (i) Dry Arctagrostis, Luzula, Poa, Carex graminoid tundra, (ii) Moist Carex, Poa, Luzula graminoid tundra, (iii) Wet Carex, Sphagnum graminoid tundra, (iv) Seasonally flooded Carex, Dupontia, Eriophorum graminoid |
| Pond | Aquatic | Low | (i) Aquatic Arctophila, Carex, Dupontia graminoid tundra |
| Lakes & River | Aquatic | Low | N/A |
Polygonal tundra geomorphic classification
We used geomorphology, surface water, vegetation composition, and spectral geomorphic characteristics from literature and field-based studies to inform our geomorphic classification. Due to similarity in surface moisture and vegetation across geomorphic types, traditional image classifications schemes such as supervised and unsupervised approaches (geospatial analysis of grouping pixels with similar spectral properties into classes with and without prior knowledge of existing land cover) had difficulty differentiating between geomorphic classes with respect to ground truth data. Therefore, to overcome this classification challenge and differentiate between geomorphic types within Landsat-7 (30 m resolution) and pan-sharpened Quickbird (0.6 m resolution) imagery, we use three classification approaches that allowed for the extraction of specific attributes based on geomorphic characteristics: (i) supervised classification identified Open water, HC, LC, FC, and CLC, (ii) Object Based Image Analysis further differentiated the open water class into Rivers, Lakes, Ponds, and Tr, and (iii) threshold feature extraction differentiated DS and Mdw using vegetation indices. Classification approaches (Figure S1) differed slightly with satellite imagery varying in spatial resolution and number of multispectral bands. Decision rules and thresholds were informed by plot and transect measurements of albedo, hyperspectral reflectance, and soil moisture from Lara et al. (2012), plant community assessments (Villarreal et al., 2012; C.E. Tweedie, C.G. Andresen, R.D. Hollister, J.L. May, D.R. Bronson, A. Gaylord and P.J. Webber, in prep), digital elevation models (1 m and 20 m ground resolution; Hubbard et al., 2013; Manley et al., 2006), DTLB relative age maps (Hinkel et al., 2003), and previous observations and geomorphic assessments (Brown et al., 1980).
Image processing
Two cloud and snow-free Landsat-7 ETM+ satellite images were acquired for the Barrow Peninsula (1792 km²) on 18 July 2002 and 29 July 2002. Images include seven multispectral bands (30 m) and one panchromatic band (15 m). Landsat-7 imagery was rectified using ground control points from two previously orthorectified images (NASA Landsat Program, 2002) and resampled at 30 × 30 m pixels using a nearest neighbor algorithm. To normalize reflectance across satellite scenes, top-of-atmosphere reflectance was calculated according to (Chavez, 1996) which minimized the radiometric differences between images due to atmospheric conditions, acquisition dates, and satellite solar zenith angle. Finally, the two Landsat-7 images were mosaiced.
Vegetation indices used for image classification and threshold feature extraction were calculated for both Landsat-7 and Quickbird satellite imagery. With the exception of albedo calculated only for Landsat-7 imagery using bands 1, 3, 4, 5, and 7 (B, R, NIR, MIR, and SWIR Liang, 2001), spectral indices, normalized difference vegetation index NDVI (Rouse et al., 1974), modified soil adjusted vegetation index MSAVI (Qi et al., 1994), normalized difference standing water index NDSWI (Goswami et al., 2011), and normalized difference water index NDWI (Gao, 1996) were calculated for all imagery. ArcGIS 10.1 software was used to carry out all image processing tasks.
In the Landsat-7 image classification (Figure S1a) spectral bands 1–5 and 7 (VIS, NIR, SWIR) were used, similar to other Landsat-7 ETM+ classifications derived for arctic coastal tundra ecosystems (Grosse et al., 2006; Schneider et al., 2009). We applied a maximum likelihood supervised classification using (n = 21) training sites synthesized from (i) field observations, (ii) Quickbird high resolution imagery, and (iii) additional ancillary data (e.g., GPS geo-tagged field photos) provided by the Barrow Area Information Database (BAID: http://www.barrowmapped.com). The maximum likelihood algorithm groups image pixels with similar reflectance values across the multispectral space. In our study, we specifically targeted polygon tundra geomorphic types, HC, FC, LC, CLC, and open water with this method. Open water surface area was further differentiated by area into lakes (>1 ha) and ponds (<1 ha; Figure S1a), similar to Hinkel et al. (2003). Thresholds for DS (e.g., albedo >0.241 and NDSWI of < −0.428, or slope >0.95) and Mdw (NDVI > 0.6, NDWI <−0.26 > −0.46, and DTLB age ≤300 years) were developed as the Supervised Classification was unable to differentiate DS and Mdw from HC and LC due to similar spectral reflectance properties. Specific thresholds were defined using spectral indices and slope derived from high resolution satellite imagery and DEMs relative to geomorphology ground control points.
A cloud- and snow-free Quickbird satellite image mosaic was acquired for 1 August 2002 from Manley et al. (2006), to: (i) identify geomorphic types not resolved in Landsat data (Tr; <2 m), (ii) train classification for the Landsat-7 imagery, and (iii) determine sites suitable for validating the Landsat classification. The image mosaic included four multispectral (2.4 m) bands and one panchromatic (0.6 m) band. The Quickbird extent that we used (p-002) represented approximately 17% of the Barrow Peninsula. Similar to Landsat-7 pre-processing, top-of-atmosphere reflectance was calculated for Quickbird imagery following Schowengerdt (1997).
In preparation for Quickbird image classification (Figure S1b), multispectral bands were pan-sharpened using the Gram-Schmidt panchromatic scene sharpening method, which maintains spatial and spectral pixel quality. A K-means clustering algorithm was used on a raster image stack (e.g., R, G, B, NIR, MSAVI), to separate ~0.6 m pixels into open water, wet, moist, and dry classes. Similar to Landsat-7 classification, open water was separated into respective classes by areal coverage (Figure S1b). Estimated area, perimeter, and thickness (i.e., the diameter of the largest disc that touches the boundary of a polygon in two or more locations) were used to determine the spatial extent of lakes, ponds, and troughs, resulting in a similar product to that produced by Skurikhin et al. (2013). Trough location was determined by merging standing water <625 m² with wet and moist classes, while restricting perimeter to ≥140 m, and thickness to ≥1.65 m. All remaining open water, wet, and moist pixels are classed into LC, while dry pixels were classed into HC polygons. Thresholds were produced for DS and Mdw similar to Landsat-7 classification (Figure S1a). We combined high and flat-center polygons as these classes were not able to be extracted with high confidence outside the extent of a high resolution (~1 m) airborne light detection and ranging (LIDAR) digital elevation model (DEM), data product used in Hubbard et al. (2013). To estimate the percentage of trough by geomorphic type, we overlaid the tundra geomorphology map on the trough layer (i.e., Quickbird image extent) and calculated the ratio of trough area by geomorphic type. Lastly, we extrapolated Tr by geomorphic type across the Barrow Peninsula and adjusted total area by geomorphic type.
Map accuracy assessment
An accuracy assessment was computed for both low and high resolution geomorphic classifications based on 0.6 m Quickbird satellite imagery and 0.5 m LIDAR DEM ground control points (low and high resolution classification n = 360 and n = 142, respectively). A subset of 40 validation points derived from satellite imagery was compared with ground truth data, which were distributed across the International Biological Program research site and the Barrow Environmental Observatory during the 2010 field campaign. We determined satellite-derived validation points to be in 92.5% agreement with ground truth data indicating satellite-derived validation data are an acceptable data type for validation of the geomorphic classification for this tundra ecosystem. The LIDAR DEM was primarily used to create reference data for high resolution trough validation as satellite imagery may not adequately represent moist or dry trough drainage networks with seasonal/inter-annual variation in surface runoff. Training sites were selected >100 m from validation sites to minimize an artificially inflated classification accuracy (Hammond & Verbyla, 1996). Site validations were further guided by aerial/ground based photography, high resolution plant community distribution maps (C.E. Tweedie, C.G. Andresen, R.D. Hollister, J.L. May, D.R. Bronson, A. Gaylord and P.J. Webber, in prep), and local knowledge. We used an error matrix (Stehman, 1997) to derive an overall map accuracy and user/producer accuracies by geomorphic type for low and high resolution classifications, in addition to the Kappa and Weighted Kappa Coefficient (Fleiss et al., 1969; Congalton, 1988).
Spatial data and analysis
To assess the spatial distribution of polygonal tundra across dominant landforms such as DTLBs, interstitial tundra, and elevation ranges, we use a variety of spatial datasets developed on the Barrow Peninsula. Datasets include relative age estimates for DTLBs (i.e. Young <50 years, Medium 50–300 years, Old 300–2000 years, Ancient 2000–5500 years), calculated using radiocarbon-dating techniques and associated degree of plant succession (Hinkel et al., 2003); estimates of interstitial tundra areas, calculated by considering the tundra region where DTLBs, lakes, ponds, and rivers were absent; and elevation ranges computed using a DEM developed by Manley et al. (2006), which identify elevated erosional remnants of older surfaces (i.e. >12 m.a.s.l; Bockheim et al., 2004; Eisner et al., 2005).
We estimated geomorphic successional change rates in response to the thaw-lake cycle by (i) determining the distribution of geomorphic types within the 558 DTLBs mapped by Hinkel et al. (2003), (ii) assessing the difference in geomorphic type percentage between DTLB age categories (i.e., Young to Medium, Medium to Old, Old to Ancient), and (iii) associating median age estimates for DTLBs as carbon isotope dates, to determine an estimate of geomorphic change with respect to DTLB age on a century-millennial time scale. Due to uncertainty of successional change in response to recent climate warming, we assume change rates over the next 100 years will be similar to those over the past few millennia. We did not consider landscape change on the interstitial tundra in response to the thaw-lake cycle, as these areas are thought to be unaffected by the thaw-lake cycle (Hinkel et al., 2003; Eisner et al., 2005), but known to exhibit thermokarst disturbance (Jorgenson et al., 2006).
Within the interstitial tundra, thermokarst pit formation was estimated to occur over the next 100 years, by assuming a low, modest, or extreme (10%, 30%, or 50%) ice-wedge degradation, based on estimates reported for nearby Alaskan coastal plain tundra sites near the Colville River (Jorgenson et al., 2006) and Prudhoe Bay (Raynolds et al., 2014). Jorgenson et al. (2006) reports a 2.5 fold increase of new thermokarst pits in upland or interstitial tundra from 1982–2001, while Raynolds et al. (2014) documents a 1.8 fold increase in thermokarst pits from 1990–2010. Thermokarst pits manifest on the landscape by expanding/merging with trough networks, resulting in a substantial redistribution of surface water from LC to Tr, facilitating the transition from previously LC to HC polygons, which is also documented at other Alaska tundra sites (Necsoiu et al., 2013). We compensate for increased thermokarst pit area, by increasing Tr area across all adjacent tundra geomorphic types using landscape-level spatial statistics calculated from ‘Tr area by geomorphic type’ (Fig. 2). These estimates enable the growth of pits at the expense of adjacent polygon edges, altering all geomorphic type areas, respectively.
Upscaling peak growing season CO2 and CH4 fluxes
Carbon flux measurements including Net Ecosystem Exchange (NEE), Gross Ecosystem Exchange (GEE), Ecosystem Respiration (ER), and Methane (CH4) flux, used in this study are represented by 82 sites, (n = 304 CO2 and 76 CH4 measurements) distributed across geomorphic classes (DS = 11, HC = 13, FC = 6, LC = 18, CLC = 6, Tr = 10, Mdw = 6, and Pond = 12) using a closed chamber technique. Twenty-seven sites are presented in Lara et al. (2012), six sites are presented in Olivas et al., 2010; and forty-nine sites are presented in this study (DOI:10.5440/1156852). At each site (exception of Olivas et al., 2011 sites, which are well represented diurnally due to repeat sampling), four different thicknesses of shade cloth were used to generate CO2 light response curves (sensu Shaver et al., 2007). Light response curves were used to normalize photosynthetically active radiation that is diurnally variable to a peak growing season average ~400 umolm−2 s−1. Although, CH4 fluxes used in this study are measured prior to max thaw depth and potentially max CH4 flux, seasonal CH4 trends in the Barrow area, suggest peak growing season (mid July-early August) CH4 fluxes may be similar to (Zona et al., 2009) or higher (Sturtevant et al., 2012) than late season flux likely due to differences in soil temperature and moisture. Sites were generally clustered in groups of three and dispersed across a relatively large 25 km2 area (centroid, lat: 71.291 long: −156.645) to maximize representativeness of this tundra region near Barrow. CO2 and CH4 flux was measured by a LI-COR 6200 Photosynthesis System (LI-COR Inc., Lincoln, NB, USA) and a photo-acoustic multi-gas analyzer (INNOVA 1312 AirTech Instruments A/S, Denmark) in this study (sensu Lara et al., 2012). All carbon flux data used in this assessment to represent tundra geomorphic types were collected between the summers of 2006–2010 during peak growing season (doi:10.5440/1156852). Flux sites were categorized into tundra geomorphic classes as previously defined, using plant community composition, site photos, and high resolution satellite imagery. Due to the morphological similarity and the inability to spatially separate thermokarst pits from trough networks, we conservatively assume these carbon fluxes to be equivalent. Carbon fluxes were weighted by geomorphic type across the mapped study area. C-CO2 equivalents was calculated by assuming a 100 year atmospheric residence time (i.e., CH4 = 28 times the greenhouse warming potential as CO2, Stocker et al., 2013) of both NEE and CH4 fluxes. In addition, we summarized plot-level measurements for thaw depth (TD), water table depth (WTD), leaf area index (LAI), and normalized difference vegetation index (NDVI) by geomorphic type (Table 4). We estimate uncertainty in both plot and landscape-level flux estimates using 95% confidence intervals based on two sigma standard deviation.
Results
Spatial distribution of geomorphic types on the Barrow Peninsula
The tundra geomorphology map classified the Barrow Peninsula into ten dominant geomorphic types by area (Fig. 2). The polygon map determined the dominant geomorphic types (>10%), were high-centered polygon (HC), flat-centered polygon (FC), low-centered polygon (LC), and Lakes, representing 11%, 17%, 24%, and 24%, respectively, while the other six geomorphic types [i.e., drained slope (DS), coalescent low center (CLC), meadow (Mdw), trough (Tr), Pond, River] cumulatively represented ~25% of the total land cover on the Barrow Peninsula (Fig. 2, Table 2). We found Tr to be disproportionately represented among tundra geomorphic types. The dominant geomorphic types where Tr networks may be found are HC, FC, and LC representing ~93% of the total trough area on the Barrow Peninsula (Fig. 2).
| Geomorphic Type | Reference data | Total | User accuracy (%) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Urban | DS | HC | FC | LC | CLC | Mdw | Pond | Lake | River | |||
| Classification | ||||||||||||
| Urban | 3 | 3 | 1.00 | |||||||||
| Drained Slope | 15 | 1 | 1 | 17 | 0.88 | |||||||
| High-Center | 2 | 43 | 11 | 2 | 58 | 0.74 | ||||||
| Flat Center | 4 | 15 | 28 | 5 | 52 | 0.54 | ||||||
| Low Center | 3 | 7 | 17 | 92 | 4 | 1 | 1 | 125 | 0.74 | |||
| Coalescent Low Center | 17 | 2 | 1 | 20 | 0.85 | |||||||
| Meadow | 1 | 1 | 5 | 1 | 20 | 1 | 1 | 30 | 0.67 | |||
| Pond | 8 | 1 | 9 | 0.89 | ||||||||
| Lake | 1 | 1 | 41 | 43 | 0.95 | |||||||
| River | 3 | 3 | 1.00 | |||||||||
| Total | 3 | 24 | 67 | 59 | 104 | 23 | 23 | 9 | 45 | 3 | 360 | |
| Producer accuracy (%) | 1.0 | 0.63 | 0.64 | 0.47 | 0.88 | 0.74 | 0.87 | 0.89 | 0.91 | 1.0 | ||
| Overall accuracy (%) | 0.75 | |||||||||||
| Kappa | 0.694 | |||||||||||
| Weighted Kappa | 0.836 | |||||||||||
- Trough (High res. classification) producer and user accuracy was 96 and 73%, n = 142; Kappa: 95% CI = 0.64–0.75, strength of agreement considered ‘good’, weighted Kappa: strength of agreement considered ‘very good’.
The distribution of geomorphic types was also determined within DTLBs, interstitial tundra, and within increasing elevation ranges. We found DTLBs, interstitial, and Lakes to represent 44%, 34%, and 24% (total not equal to 100 because lakes present in DTLBs), respectively, of the total land area on the Barrow Peninsula (Fig. 3). Relative DTLB ages of Young, Medium, Old, and Ancient, represented 3%, 8%, 27%, and 6%, respectively, of the total peninsula land area (Table 3). The primary difference in geomorphic types in Young through Ancient DTLBs was found in LC, CLC, and Mdw, where LC and CLC increased 15% and 11%, from Young to Ancient, respectively, and Mdw decreased 24%. Additionally, FC increased 4% from Young to Ancient DTLBs, with no other changes ≥2% identified. Elevation categories, 0–2, 2–8, 8–13, and >13 m, represented 6%, 73%, 19%, and 2%, of the total elevational range of the Barrow Peninsula, respectively. Generally, the extent of wetter geomorphic types decreased with increasing elevation, as geomorphic types Lakes, LC, CLC, and Mdw decreased by 32%, 8%, 3%, and 3%, respectively, between elevation categories, 0–2 and >13 m; in contrast, dry geomorphic types DS, HC, FC, and Tr increased 3%, 14%, 28%, and 4% (Fig. 4).
| Terrain Feature | Area (km²) | Area (%) | DS | HC | FC | CLC | LC | Tr | Mdw | Pond | Lake | River |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Elevation | ||||||||||||
| 0–2 m | 115 | 6% | 11% | 8% | 9% | 7% | 22% | 2% | 3% | 1.3% | 36% | 0.8% |
| 2–8 m | 1299 | 73% | 7% | 10% | 14% | 8% | 26% | 3% | 3% | 1.0% | 28% | 0.6% |
| 8–13 m | 344 | 19% | 10% | 17% | 26% | 7% | 28% | 5% | 0.8% | 0.2% | 7% | 0.1% |
| >13 m | 34 | 2% | 14% | 23% | 36% | 4% | 14% | 6% | 0.2% | 0.1% | 4% | 0.1% |
| Drained Thaw Lake Basin | ||||||||||||
| Young | 59 | 3% | 3% | 12% | 15% | 4% | 28% | 4% | 24% | 1.0% | 5% | 0.0% |
| Medium | 146 | 8% | 2% | 8% | 13% | 13% | 37% | 5% | 12% | 0.8% | 5% | 0.0% |
| Old | 476 | 27% | 4% | 7% | 15% | 16% | 44% | 5% | 1.7%a | 0.2% | 1% | 0.0% |
| Ancient | 101 | 6% | 4% | 10% | 19% | 15% | 43% | 2% | 0.1%a | 0.2% | 0% | 0.0% |
| All DTLB | 782 | 44% | 4% | 8% | 16% | 15% | 43% | 5% | 4% | 0.4% | 2% | 0.0% |
| Interstitial | 609 | 34% | 17% | 23% | 30% | 3% | 19% | 6% | 1.4% | 0.5% | 1% | 0.0% |
| All tundra | 1792 | 100% | 8% | 11% | 17% | 8% | 24% | 5% | 2% | 0.9% | 24% | 0.6% |
- a Estimated from manual delineation of quickbird imagery.
Map accuracy assessment
The accuracy assessment (Table 2) determined an overall map accuracy of 75%, with a Kappa and Weighted Kappa of 0.694 and 0.836 respectively, suggesting the strength of agreement between an independent validation dataset and classification to be ‘good’ and ‘very good’ (Fleiss et al., 1969; Congalton, 1988). Given the resolution (i.e., 30 × 30 m) and spectral characteristics of the Landsat-7 imagery, mixed pixels and misclassifications of spectrally similar classes were expected. However, most geomorphic classes were well represented by this geomorphic classification. The lowest accuracy was found in FC polygons, as this geomorphic type was difficult to determine even with the use of both high resolution imagery and DEMs. Additionally, the high resolution Quickbird classification, determined Tr user and producer accuracy to be 73% and 96% (Table 2).
Carbon fluxes by geomorphic type
We identified considerable differences in plot-level CO2 and CH4 fluxes among geomorphic types (Table 4). The lowest NEE (e.g., greatest uptake, where negative values indicate carbon uptake from the atmosphere and positive values indicate carbon loss to the atmosphere) was found in aquatic-wet geomorphic types, Pond (−3.4 gC-CO2 m−2 day−1), Tr (−2.2 gC-CO2 m−2 day−1), and Mdw (−1.7 gC-CO2 m−2 day−1), which also had the largest uptake in GEE. Drier geomorphic types, DS, HC, FC, varied less in NEE (−0.2 to −1.2 gC-CO2 m−2 day−1), GEE (−1.9 to −3.8 gC-CO2 m−2 day−1), and ER (1.6 to 2.6 gC-CO2 m−2 day−1). Further, both CO2 and CH4 fluxes in DS were among the lowest of all geomorphic types (Table 4). Low center NEE was 0.2 gC-CO2 m−2 day−1, as ER was greater than GEE, making LC the only geomorphic type to have a positive NEE (i.e., carbon loss) during the peak growing season. Not surprisingly, CH4 fluxes were highest in aquatic-wet anaerobic geomorphic types, Ponds (114.5 mgC-CH4 m−2 day−1), CLC (70.0 mgC-CH4 m−2 day−1), Mdw (46.1 mgC-CH4 m−2 day−1), and Tr (43.2 mgC-CH4 m−2 day−1). The remaining CH4 fluxes for moist-dry geomorphic types were <16.0 mgC-CH4 m−2 day−1 (Table 4). Uncertainty ranges (i.e. 95% confidence intervals) for all carbon flux estimates by geomorphic type and change scenarios are shown in Table 4 and 5.
| Measurement | DS | HC | FC | LC | CLC | Tr | Mdw | Pond | Total |
|---|---|---|---|---|---|---|---|---|---|
| Area (km², landscape %) | 136, 8 | 201, 11 | 296, 17 | 434, 24 | 137, 8 | 90, 5 | 42, 2 | 16, 1 | 1342, 75 |
| NEE (gC-CO2 m−2 day−¹) | −0.2 ± 0.2 | −1.2 ± 0.4 | −1.0 ± 0.9 | 0.2 ± 0.4 | −0.6 ± 0.4 | −2.2 ± 1.1 | −1.7 ± 0.4 | −3.4 ± 1.1 | – |
| GEE (gC-CO2 m−2 day−¹) | −1.9 ± 0.5 | −3.8 ± 1.0 | −3.4 ± 1.2 | −2.3 ± 0.8 | −1.5 ± 0.6 | −5.6 ± 1.4 | −5.0 ± 1.0 | −5.1 ± 1.7 | – |
| ER (gC-CO2 m−2 day−¹) | 1.6 ± 0.4 | 2.6 ± 0.8 | 2.4 ± 0.6 | 2.5 ± 0.8 | 1.0 ± 0.4 | 3.4 ± 0.7 | 3.3 ± 1.2 | 1.7 ± 0.8 | – |
| CH4 flux (mgC-CO2 m−2 day−¹) | 6.2 ± 4.4 | 16.3 ± 5.1 | 7.9 ± 6.7 | 12.9 ± 3.3 | 70.0 ± 22.0 | 43.2 ± 23.5 | 46.1 ± 12.9 | 114.5 ± 48.1 | – |
| TD (cm) | 37 ± 7 | 35 ± 8 | 27 ± 5 | 27 ± 1 | 36 ± 2 | 34 ± 3 | 36 ± 3 | 40 ± 4 | – |
| WTD (cm) | −37 ± 7 | −31 ± 10 | −6 ± 3 | −3 ± 1 | 0 ± 1 | −2 ± 2 | −4 ± 3 | 4 ± 2 | – |
| LAI (Index) | 0.3 ± 0.1 | 0.3 ± 0.1 | 0.4 ± 0.2 | 0.4 ± 0.2 | 0.2 ± 0.1 | 0.6 ± 0.1 | 0.4 ± 0.1 | 0.3 ± 0.1 | – |
| NDVI (Index) | 0.5 ± 0.04 | 0.5 ± 0.05 | 0.5 ± 0.08 | 0.5 ± 0.05 | 0.6 ± 0.02 | 0.7 ± 0.07 | 0.7 ± 0.03 | 0.4 ± 0.08 | – |
| NEE (106gC-CO2 day−¹) | −29.8 ± 30 | −245.1 ± 82 | −298.9 ± 283 | 69.6 ± 159 | −80.1 ± 49 | −193.9 ± 101 | −68.9 ± 17 | −54.4 ± 18 | −902.3 ± 464 |
| GEE (106gC-CO2 day−¹) | −252.4 ± 76 | −770.7 ± 212 | −1012.7 ± 348 | −1015.9 ± 334 | −209.9 ± 81 | −498.1 ± 124 | −207.1 ± 41 | −81.9 ± 28 | −4048.7 ± 667 |
| ER (106gC-CO2 day−¹) | 222.6 ± 60 | 525.5 ± 170 | 713.8 ± 192 | 1085.5 ± 349 | 129.9 ± 49 | 304.2 ± 60 | 138.2 ± 49 | 27.4 ± 13 | 3147.2 ± 427 |
| CH4 flux (106gC-CH4 day−¹) | 0.8 ± 0.6 | 2.9 ± 1.1 | 2.3 ± 2.0 | 5.6 ± 1.4 | 9.6 ± 3.0 | 3.9 ± 2.1 | 1.9 ± 0.5 | 1.8 ± 0.8 | 28.9 ± 16 |
| C-CO2-eq (106gC-CO2 day−¹) | −6.5 | −164.1 | −233.4 | 226.7 | 187.1 | −85.2 | −14.9 | −2.8 | −93.2 ± 428 |
| Unit | Geomorphic change | NEE | GEE | ER | CH4 | C-CO2-eq |
|---|---|---|---|---|---|---|
| gC m−2 day−¹ | 2002_Present | −0.67 ± 0.34 | −3.0 ± 0.49 | 2.3 ± 0.32 | 0.021 ± 0.01 | −0.07 ± 0.32 |
| 2102_TLC | −0.67 ± 0.34 | −3.0 ± 0.49 | 2.3 ± 0.32 | 0.021 ± 0.01 | −0.07 ± 0.32 | |
| 2102_TLC+TK10% | −0.71 ± 0.34 | −3.1 ± 0.51 | 2.4 ± 0.33 | 0.022 ± 0.01 | −0.09 ± 0.32 | |
| 2102_TLC+TK30% | −0.79 ± 0.35 | −3.3 ± 0.54 | 2.5 ± 0.37 | 0.023 ± 0.01 | −0.14 ± 0.33 | |
| 2102_TLC+TK50% | −0.88 ± 0.36 | −3.5 ± 0.57 | 2.6 ± 0.40 | 0.025 ± 0.01 | −0.18 ± 0.34 | |
| 106gC day−¹ | 2002_Present | −902.3 ± 464 | −4048.7 ± 667 | 3147.2 ± 427 | 28.9 ± 16 | −93.2 ± 428 |
| 2102_TLC | −900.1 ± 464 | −4045.8 ± 667 | 3145.7 ± 427 | 28.9 ± 16 | −90.9 ± 428 | |
| 2102_TLC+TK10% | −955.9 ± 469 | −4137.3 ± 674 | 3181.4 ± 440 | 29.7 ± 17 | −125.5 ± 433 | |
| 2102_TLC+TK30% | −1069.2 ± 476 | −4411.0 ± 689 | 3341.8 ± 453 | 31.7 ± 18 | −182.7 ± 441 | |
| 2102_TLC+TK50% | −1182.5 ± 476 | −4684.7 ± 688 | 3502.3 ± 451 | 33.7 ± 18 | −239.9 ± 441 |
Upscaling carbon fluxes for present day
Extrapolated peak growing season carbon fluxes varied for Young, Medium, Old, and Ancient DTLBs. As DTLBs age, we found NEE to decrease with Young, Medium, Old, and Ancient basins represented −0.9, −0.6, −0.4, and −0.4 gC-CO2 m−2 day−1, respectively. Additionally, CO2 uptake and respiratory losses decreased with DTLB age, as GEE and ER were estimated at −3.5, −3.0, −2.7, −2.7 gC-CO2 m−2 day−1, and 2.7, 2.4, 2.3, and 2.2 gC-CO2 m−2 day−1, respectively. The highest NEE within DTLBs by geomorphic type was recorded for FC and HC, while LC was the lowest (Fig. 4). Generally, NEE was highest for Mdw in Young and Medium DTLBs but dramatically reduced in Old and Ancient, while CLC had the inverse response. Differences in CH4 fluxes across DTLBs were similar, as Young, Medium, Old, and Ancient are estimated as 26.1, 27.0, 24.2, and 21.7 mgC-CH4 m−2 day−1. Patterns for CH4 fluxes were similar to that of NEE, GEE, and ER for Mdw, as fluxes were high in Young and Medium DTLBs and reduced in Old and Ancient DTLBs (Fig. 4). In addition, across elevation ranges on the Barrow Peninsula (Fig. 4), we found NEE to vary by geomorphic type, while CH4 fluxes generally decreased with increasing elevation which was associated with drier geomorphic types (Fig. 4). NEE increased with elevation with the categories 0–2, 2–8, 8–13, and >13 m, representing −0.6, −0.6, −0.7, and −0.8 gC-CO2 m−2 day−1, respectively. While, GEE and ER were estimated at −2.8, −2.9, −3.0, −3.2 gC-CO2 m−2 day−1, and 2.2, 2.3, 2.3, and 2.4 gC-CO2 m−2 day−1 respectively, CO2 uptake increased and respiratory losses remained fairly consistent with increasing elevation. Conversely, CH4 fluxes decreased with increasing elevation, as 0–2, 2–8, 8–13, and >13 m, representing 22.5, 21.6, 17.4, and 14.7 mgC-CH4 m−2 day−1, respectively.
At the scale of the Barrow Peninsula, we estimate total NEE and CH4 flux to presently account for −902.3 106gC-CO2 day−1 (uncertainty between −438.3 and −1366 106gC-CO2 day−1) and 28.9 106gC-CH4 day−1 (uncertainty between 12.9 and 44.9 106gC-CH4 day−1; Table 5). Across this landscape, the geomorphic types highest in GEE and ER were LC, FC, HC, and Tr, which accounted for 25%, 25%, 19%, and 12%, respectively, of the total daily GEE and 35%, 23%, 17%, and 10%, respectively, of the total daily ER. While GEE and ER were largely neutral in DS, CLC, Mdw, and Pond, representing 6%, 5%, 5%, and 2%, respectively in GEE and 7, 4, 4, and <1%, respectively in ER. Similarly geomorphic types highest in CH4 fluxes were CLC, LC, Tr, HC, FC, Mdw, Pond, and DS, which accounted for 33%, 19%, 14%, 10%, 8%, 7%, 6%, and 3%, respectively, of the total CH4 flux on the Barrow Peninsula. Our analysis indicates that while CLC and Tr only represent 12% of the total area on the Barrow Peninsula, they account for ~47% of the total CH4 flux.
One-hundred years of geomorphic change
When assessing the difference in tundra geomorphic type relative area, between DTLBs classified as Young-Medium (Y-M), Medium-Old (M-O), and Old-Ancient (O-A), to provide an estimate of geomorphic change over time (Fig. 5), we found change rates to vary by DTLB age. Generally, differences by geomorphic type appeared gradual and linear over a 3500 year period (Fig. 5), with the percentage of change reducing from Younger to Older DTLBs. Furthermore, after normalizing geomorphic change rates over a 100 year time period, Young DTLBs emerged as the predominant DTLB where geomorphological change can be expected relative to older DTLBs (Fig. 5). Specific to DTLB age, we estimate rates of change over a 100 year time period by geomorphic type, HC, FC, LC, CLC, and Mdw for Young (−2.26, −1.27, +5.60, +4.83, −6.80%), Medium (−0.10,+0.15, +0.54, +0.25, −0.92%), and Old (+0.09, +0.12, 0.01, −0.03, −0.05%). We assumed no change over a 100 year period in Ancient DTLBs (i.e., likely lower then Old DTLBs). Across the Barrow Peninsula, due to 100 years of successional landscape change via the thaw-lake cycle, we estimate HC, FC, LC, CLC, and Mdw to change by −0.17, −0.03, +1.38, +0.20, −1.31 km2 (Fig. 6), respectively. In addition, we estimated the change in area by geomorphic type in response to the thaw-lake cycle and incremental levels, low, modest, or extreme (10%, 30%, or 50%), of thermokarst pit formation on the interstitial tundra (Fig. 6). We found that ice-wedge degradation associated with changes in thermokarst pits can substantially influence landscape geomorphology (e.g., increase in Tr and HC, decrease in LC and FC: Fig. 6) and carbon fluxes.
Lastly, we upscale carbon fluxes across this landscape according to 100 years of change via the thaw-lake cycle and incremental thermokarst pit formation (Fig. 6, Table 5). Little change in fluxes are found in response to the thaw-lake cycle relative to present-day fluxes, as landscape-level NEE, GEE, ER, changed −2.2, −2.9, −1.5 106 gC-CO2 day−¹, while CH4 and C-CO2-eq changed -0.037 106 gC-CH4 day−¹, and −2.3 106 gC-CO2-eq day−¹. Generally we find increased carbon uptake with increasing thermokarst formation, as seen with change in NEE and GEE (Table 5). Although ER increased with thermokarst formation, CO2 uptake was greater than that lost, therefore the tundra CO2 sink strength increased. Further, with the increase in thermokarst pit scenarios, we found an increase in the area of inundated anaerobic soil, thus, CH4 increased with higher thermokarst formation (Table 5). However, after compensating for the warming potential of both CO2 and CH4 trace gases (i.e. C-CO2-eq), CO2 uptake was greater than that lost by ER and CH4 during peak growing season, suggesting the Barrow Peninsula is currently a carbon sink during peak growing season and if thermokarst pits increase over the next 100 years, the peak growing season sink strength of this tundra region may increase.
Discussion
Prior to this mapping initiative, geomorphologic distribution of polygonal tundra was highly uncertain, with few studies subjectively characterizing patterns and distribution across the landscape (Sellmann et al., 1972; Brown et al., 1980; Hinkel et al., 2003; Frohn et al., 2005). Our 30 × 30 m resolution mapping approach yielded results that could be utilized by a multitude of ecological applications and discriminates well among geomorphic types. Using this map, we (i) identify differences in tundra geomorphic distribution within varied age DTLBs and interstitial tundra, (ii) use previously determined DTLB age estimates (Hinkel et al., 2003) to approximate geomorphic change associated with the thaw-lake cycle, and (iii) determine the impact of thermokarst pit formation on geomorphology over the next 100 years. We find the Barrow Peninsula to presently be a carbon sink during peak growing season, although LC and CLC were found to be net sources of carbon to the atmosphere (Table 4). Additionally, for the first time reported in the literature, we present estimates of CH4 flux across the Barrow Peninsula, although flux magnitudes were likely underestimated as ebullition rates from biologic and geologic seeps (Anthony et al., 2012) were not quantified for the study region. Our initial hypotheses failed to be rejected as we find (i) that the thaw-lake cycle has little effect on carbon flux in response to 100 years of geomorphic change, highlighting the extremely slow transition between geomorphic types in the Barrow Peninsula, and (ii) that increasing thermokarst pit formation increases CH4 fluxes and CO2 uptake. However contrary expectations, our analysis indicates that the magnitude of CO2 uptake vastly outweighed the combined losses of ER and CH4, resulting in an estimated negative feedback to warming during the peak growing season.
Geomorphic classification
The use of multiple image classification methods in conjunction with prerequisite understanding of geomorphic characteristics can be a powerful tool for landscape characterization studies in the heterogeneous arctic coastal plain. Three classification approaches were used to identify specific attributes characteristic of geomorphic types: (i) Supervised Classification, (ii) Object Based Image Analysis, and (iii) Threshold feature extraction. Map accuracy was good with an overall relative accuracy of 75%, and are comparable with land cover classification accuracies in similar Arctic tundra environments in a NE Siberian coastal lowland (79%: Grosse et al., 2006) and in the Lena river delta (77.8%: Schneider et al., 2009). As the Landsat-7 ETM+ mosaic provides a snapshot of the tundra landscape during the peak growing season within the summer of 2002, seasonal variation in vegetation cover, soil moisture, or surface water runoff can typically lead to classification errors. However, by characterizing geomorphic classes based on geomorphological and hydrological properties, we substantially decrease possible temporal heterogeneities. Seasonal variation in soil moisture or surface water runoff may influence the optical detection of dry-moist Tr delineation within high resolution imagery, as spectral reflectance in dry troughs using satellite imagery can be indiscernible from dry geomorphic types. This likely resulted in slightly lower Tr accuracies relative to other Tr delineation assessments (Skurikhin et al., 2013), which used satellite imagery vs DEMs for validation. Additionally, the lowest accuracy for mapped geomorphic types was found for FC, as observationally, this class was difficult to discriminate from HC as the primary geomorphological differences are with lower topographic relief and +25 cm WTD based on field data (Table 4). To increase Mdw class accuracy, we assumed that Mdw was not present in Old and Ancient DTLBs, as the high density of ponded LC and CLC polygons in Old and Ancient DTLBs were often misclassifed as Mdw. This assumption is in line with Hinkel et al. (2003) as Mdw are typically only found in Young or Medium DTLBs, with an exception of Mdw presence in Old or Ancient DTLBs after smaller thermokarst lakes within DTLBs are drained. However, we manually delineated Mdw within the Quickbird extent to represent approximately 1.7 and 0.06% of the total surface area within Old and Ancient DTLBs. Although studies suggest ~<4 m image resolution is needed to adequately represent arctic polygonal tundra geomorphology (Muster et al., 2012, 2013), we demonstrate that coarser resolution (i.e. 30 m) imagery spanning larger regions can be used to upscale tundra geomorphological types in this region. However, a primary limitation of our map is its inability to detect subtle ice-wedge geomorphological differences due to its coarse resolution. Fine-scale (<4 m) geomorphological mapping remains essential for understanding long and short-term geomorphological changes in response to thermokarst pit development in arctic tundra regions where significant warming has occurred.
Upscaling carbon fluxes on the Barrow Peninsula
The plot-level land-atmosphere CO2 and CH4 fluxes measurements that we used for upscaling carbon fluxes are representative of all geomorphic types in our classification and were similar to those reported for dry-aquatic geomorphic types for CO2 (Oberbauer et al., 2007; Olivas et al., 2011) and CH4 (Zona et al., 2009; Von Fischer et al., 2010; Sturtevant & Oechel, 2013). Similar to Oechel et al. (1995), we find LC polygons to be net sources of CO2 to the atmosphere (uncertainty between −0.2 and 0.6 gC-CO2 m−2 day−1); however, the importance of this result is amplified as we identify LCs to be the most common geomorphic type representing 24% of the Barrow Peninsula. As expected, all other geomorphic types were weak to strong sinks of CO2 during the peak growing season. Understanding the difficulties in upscaling fluxes from plot to landscape scales (Bubier & Moore, 1994), we compare landscape-level eddy covariance CO2 measurements derived from tower (Zona et al., 2010; Sturtevant & Oechel, 2013) and aircraft based eddy covariance platforms (Zulueta et al., 2011). We find striking similarity between this plot-level upscaling assessment (−902.3 106 gC-CO2 day−1; Table 4) and upscaled eddy covariance (-1431.0 106 gC-CO2 day−1; Zulueta et al., 2011) estimates spanning the Barrow Peninsula, supporting the spatial representativeness of both landscape-level carbon flux estimates and demonstrating the applicability and spatial representativeness of the tundra geomorphology map. Our uncertainty analysis suggests that the Barrow Peninsula was a sink of CO2 during the growing season between −438 and −1366 106gC-CO2 day−1. Although we did not consider fluxes from Lakes and Rivers when upscaling, which account for ~25% of the total land area (Table 3), Zulueta et al. (2011) reports these regions to account for ~95.0 106gC-CO2 day−1 of the Barrow Peninsula. In addition, we estimate peak growing season CH4 fluxes for the Barrow Peninsula to represent 28.9 106 gC-CH4 day−1(uncertainty between 13 and 45 106gC-CH4 day−1; Table 5). Although CH4 fluxes were considerably smaller than CO2 exchange, by considering the combined warming potential of CO2 and CH4 trace gases represented as C-CO2-eq, the landscape sink status was estimated at -93.2 106 gC-CO2-eq day−1, suggesting CH4 fluxes, although relatively small, must be considered to quantify landscape-level net carbon exchange. The importance of C-CO2-eq is highlighted as we identify CLC to have a net sink of CO2 −80.1 106 gC-CO2 day−1. Yet, when the warming potential of CH4 is considered at the landscape scale, CLC was dramatically altered from a sink to a source of carbon to the atmosphere at 187.1 106 gC-CO2 day−1 (Table 4).
Specific to tundra within DTLBs, our plot-based upscaling assessment validated previous spatial patterns in carbon fluxes, similar to that reported in Oechel et al. (1998). We found Young DTLBs to be the most productive, relative to Medium, Old, and Ancient DTLBs, similar to that reported by Zona et al. (2010) and Sturtevant & Oechel (2013). Differences were likely a function of differing geomorphology with increasing age, as highly productive Mdw were most abundant in Young DTLBs (i.e., 24%; Table 3) and were replaced by less productive LC and CLC in Medium, Old, and Ancient DTLBs, which may also be associated with decreased nutrient availability (Bliss & Peterson, 1992). Our NEE estimates were markedly similar in magnitude and pattern for all DTLBs, to that reported by Zona et al. (2010), however our carbon flux estimates were not derived from seasonal averages as reported for Zona et al. (2010). Additionally, we find the average CH4 flux for Young, Medium, Old, and Ancient DTLBs to represent 26.1, 27.0, 24.2, and 21.7 mgC-CH4 m−2 day−1. With little difference among DTLBs of different ages, patterns were analogous to those reported by Sturtevant & Oechel (2013). Methane fluxes were likely similar among DTLBs of different ages as dry-moist geomorphic types compose ~60% of the land surface area. Soil moisture and age-related influences on net CO2 and CH4 exchange appear to offset each other, with decreasing Mdw with increasing LC and CLC as DTLBs age.
Geomorphic change assessment
In response to the present (Stafford et al., 2000; Romanovsky et al., 2002) and predicted (Sazonova & Romanovsky, 2003; Stocker et al., 2013) warming near Barrow, it is likely that the interaction of warming on permafrost and ice-wedge thaw will affect future tundra geomorphic change. Although due to our limited knowledge of geomorphological change over time according to the thaw-lake cycle, we assumed change rates over the next 100 years will be similar to those over the past few millennia, despite predicted non-linear warming trends for this region. Over a 100-year time period we determine Young DTLBs to transition between geomorphic types several orders of magnitudes faster than older DTLBs (Fig. 5), as initial ice-wedge growth after lake drainage can be as high as 3 cm yr−1 (Mackay & Burn, 2002), and growth may slow considerably as vegetation and increased snow capture reduces winter temperature fluctuations (Jorgenson et al., 2006). For example, in 1950 four shallow thaw-lakes were drained to minimize the risk of flooding on Barrow infrastructure, and observations over the first few years following this drainage find significant revegetation of once barren sediments with low relief FC and LC polygons emerging (Britton, 1957). Additionally, Britton (1957) reported the formation of HC polygons ~1 m high around the margins of DTLBs, where thermokarst subsidence in troughs commonly delineates HC polygons. The successional geomorphic transitions illustrated in this paper as DTLBs age generally also correspond to polygonal tundra development found for the floodplain of the Colville River, AK (Jorgenson et al., 1998) and the northern Seward Peninsula (Jones et al., 2012; Regmi et al., 2012), indicating that LC and CLC polygons increase in density and cover as basins age.
Although thermokarst encompasses a variety of geomorphic effects on landforms (Higgins et al., 1990), we focus our efforts on pit formation. A growing body of evidence suggests that thermokarst pit formation has rapidly increased over recent decades in some areas (Jorgenson et al., 2006; Necsoiu et al., 2013; Raynolds et al., 2014), likely in response to increased mean annual ground temperatures (MAGT), which are predicted to gradually increase by ~3.5 °C over the next century near Barrow (Sazonova & Romanovsky, 2003). Numerical permafrost models estimate differences in MAGT of Barrow relative to Colville and Prudhoe Bay, where significant thermokarst pit formation has been observed, to be ~+0.7 °C (Marchenko & Tipenko, 2008). Further, these models suggest MAGT near Barrow will surpass that of Colville and Prudhoe Bay by 2030–2040, and by 2100 will approach temperatures similar to current soils on the foothills of the Alaska Brooks Range (i.e. −4.4 °C; Romanovsky & Marchenko, 2009). However, differences in thermokarst related processes, associated with ‘Lake regions’ (Jorgenson & Shur, 2007), increase future uncertainties related to thermokarst formation in the Barrow region. Although, we conservatively assess pit formation assuming only 10-50% of change recently reported over 19 (Jorgenson et al., 2006) and 20 years (Raynolds et al., 2014), future thermokarst formation under a warmer climate may substantially surpass our estimates. Analysis presented here suggests 100 year scenarios of thermokarst influenced geomorphic change may have a much greater effect on geomorphic change than successional change due to the thaw-lake cycle alone.
We use space-for-time substitution, to assess the impact of geomorphic change in the Barrow Peninsula, as this approach has been effective in predicting ecological responses to climate change (Blois et al., 2013). Although space-for-time limitations exist (Rastetter, 1996), we satisfy these by identifying nearby regions where a perturbation to the ecosystem has occurred (i.e. warming mediated permafrost degredation; Jorgenson et al., 2006), and conservatively estimate the future impact of land cover change and associated carbon exchange. We assess change in peak growing season CO2 and CH4 fluxes solely associated with geomorphic change and do not consider other environmental factors that may impact future carbon cycling such as elevated atmospheric CO2 concentrations (Stocker et al., 2013), lengthening of the growing season (Euskirchen et al., 2006), shrub expansion (Myers-Smith et al., 2011; Elmendorf et al., 2012), and associated surface energy balance (Chapin et al., 2005). Therefore, further investigation of the dynamically coupled ecosystem responses to climate in continuous permafrost tundra landscapes is necessary. In particular, coordinated efforts are needed to understand, model, and project future thermokarst dynamics in arctic and boreal high latitude ecosystems for purposes of determining the potential effects of thermokarst disturbance on ecosystem services related to changes in carbon dynamics and fish and wildlife habitat. Although our approach focused on the potential change in peak growing season carbon flux in response to change in tundra geomorphology, further investigation using the developed tundra geomorphology map, may reveal new insights previously unexplored due to lack of spatial information such as the characterization of winter/annual carbon balance, energy balance, herbivore impacts, shorebird habitat, soil structure, or improved ground ice quantification, which all present current data limitations. This study justifies the further development of these efforts to better understand and model the transient dynamics of polygonal tundra in a changing climate.
Acknowledgements
The long history of research in the Barrow region and the accessibility of data products from the National Snow and Ice Data Center (NSIDC), Digital Globe, and United States Geological Survey (USGS) made this analysis possible. This study was supported through (i) the Next-Generation Ecosystem Experiments (NGEE-Arctic) project supported by the Office of Biological and Environmental Research in the U.S. Department of Energy Office of Science, (ii) the Integrated Ecosystem Model for Alaska and Northwest Canada project supported by the USGS Alaska Climate Science Center and by the Arctic, Western, and Northwest Boreal Landscape Conservation Cooperatives, and (iii) the National Science Foundation, Office of Polar Programs (grant no. ANS-0732885). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
