A decade of remotely sensed observations highlight complex processes linked to coastal permafrost bluff erosion in the Arctic

The ABSC is composed of a low-lying (maximum elevation of ∼10 m) tundra plain that extends ∼1950 km from the Canadian Border to Utqiaġvik (formerly Barrow), Alaska, USA. Spatial and temporal rates of coastal change along the ABSC are known to be highly variable (Jorgenson and Brown 2005, Lantuit et al 2012, Gibbs and Richmond 2015, 2017), due to variability in ground-ice content (and wedge-ice content in particular) as well as variation in erosional processes, geomorphology, lithology, coastal orientation, near shore bathymetry, and the presence of barrier islands (Jorgenson and Brown 2005). Jorgenson and Brown (2005) and Gibbs and Richmond (2015) reported that the long-term average erosion rate along the ABSC between the late-1940s and early-2000s was ∼2 m yr⁻¹. However, some particular sites eroded as much as 16–20 m yr⁻¹. Ping et al (2011) assessed 48, 1 km segments distributed across the ABSC and found that mean annual erosion between 1950–1980 was 0.6 m yr⁻¹, but increased to 1.2 m yr⁻¹ between 1980–2000. Mars and Houseknecht (2007) compared land loss due to erosion by differencing Landsat satellite imagery with legacy topographic map sheets and also found a doubling in the rate of erosion between 1985–2005 relative to 1955 and 1985. Jones et al (2009b) used more precise techniques based on aerial photography for the exposed and north-facing, 60 km segment of the ABSC between Cape Halkett and Drew Point and found that the erosion rate increased from 6.7 m yr⁻¹ (1955–979), to 9.7 m yr⁻¹ (1979–2002), to 13.6 m yr⁻¹ (2002–2007). Barnhart et al (2014a) reported that the mean erosion rate over a 7-km stretch of coast at Drew Point was 15 m yr⁻¹ (2008–2011) and 19 m yr⁻¹ (2011–2012).

We focus on a 9 km stretch of the Drew Point coastline located in the western region of the ABSC about 100 km east of Utqiaġvik and 200 km west of Prudhoe Bay (figure 1). The dominant erosional process at Drew Point consists of thermo-abrasion (Jones et al 2009b), although thermo-denudation also occurs here (Wobus et al 2011) (figure 2). Bluff height ranges from 1.6–7.1 m, with a mean of 4.4 m above the mean water level during LiDAR data acquisition on 6 August 2011. The near surface sediments consist mainly of ice-rich Holocene-aged lacustrine silts with local peat accumulations and contain large ice wedges. Sediments underlying lacustrine silts consist of transgressive marine Pleistocene silts and clays with sandy horizons near the base of the eroding bluffs (Farquharson et al 2018a). Estimates of total volumetric ground-ice content for permafrost along these bluffs approaches 80%–90%, (Kanevskiy et al 2013), with segregated and pore ice volumes accounting for 50%–80%, and wedge ice contributing nearly 30% in some locations (Wobus et al 2011). The fine grained composition of the bluffs, means that eroded sediment is easily transported away and does not accumulate and protect the base of the bluffs as is common elsewhere. Estimates of ice-wedge polygon dimensions, range from 6 to 25 m across with a mean size of ∼15 m (Wobus et al 2011, Kanevskiy et al 2013). Ice wedges are approximately 1–4 m wide near the surface and typically penetrate 3–5 m down from the surface. The Drew Point area is underlain by continuous permafrost with mean annual ground surface temperatures of about −9 °C (Smith et al 2010). Permafrost at a depth of 20 m at coastal sites along the ABSC has warmed by 0.6 °C–2.2 °C between 1989 and 2008 (Smith et al 2010).

Zoom In Zoom Out Reset image size

Offshore, water depths are shallow, the open water season is short, and the tidal range is on average only 15 cm. Nearshore water depth is less than 2 m within a distance of 0.5 km from the shoreline and increases to 3 m at a distance of 2.0 km from the coast. The nearshore open water duration at Drew Point has more than doubled between 1979–2009, increasing from ∼45 days to ∼90 days, with a higher proportion of the increase in open water duration occurring in the fall (∼0.9 days yr⁻¹) relative to the early summer (∼0.7 days yr⁻¹) (Overeem et al 2011). However, this area is prone to highly variable open water seasons and is influenced by sea-ice transport and break-up patterns from both the east and the west (Barnhart et al 2016). Between 2007–2012, the Beaufort Sea experienced the lowest September sea ice extents yet observed since the late 1970s (Ballinger and Rogers 2013) and has continued to exhibit similar patterns through 2017 (Perovich et al 2017). This increase in open water days has been accompanied by a warming trend in SST in the Beaufort Sea (Steele and Dickinson 2016). Air temperature has continued to increase in this region since 2000 as measured near Utqiaġvik, AK (Wendler et al 2012).

Rapid shoreline retreat rates observed along the ABSC may partially be explained by erosional processes uniquely associated with ice-rich permafrost coastal bluffs (Are 1988, Dallimore et al 1996). Lantuit et al (2008a) demonstrated a weak but statistically significant relation between ground-ice content and mean retreat rate, with higher mean annual retreat rates typically corresponding to coastlines with higher ground-ice content. Block failure following undercutting caused by thermo-abrasion and thaw slump activity (thermo-denudation) are common modifiers of Arctic coastal morphology and tend to be dominant erosional processes along ice-rich permafrost bluffs (Are 1988, Walker 1988, Günther et al 2012). Melting of ground ice is an important consideration as it can substantially reduce the volume of sediment input and cause thaw settlement in the nearshore, deepening the nearshore profile. Interestingly, observations made along this coast in 1901 (Schrader 1904) indicate that collapsed blocks could persist for 4–5 years (Leffingwell 1919). Such observations highlight that both the formation of erosional-niches followed by block collapse have been modifying this coast for at least the last century and that the combined impacts of climatic-oceanographic-geomorphologic conditional states have changed dramatically since the early 1900s.

The primary objective of this study is to map coastal permafrost bluff changes and compare annual retreat rates with annual open water season duration and other factors to better understand the potential mechanisms responsible for the reported increase in erosion observed at Drew Point since the early 2000s (Jones et al 2009b, Overeem et al 2011, Barnhart et al 2016). We acquired ten suitable high spatial resolution satellite images from five different satellites: Quickbird, IKONOS, GEOEYE-1, and Worldview-1 and -2 (figure 3) for a 9 km segment of eroding permafrost bluffs located at Drew Point, Alaska, USA between 2008–2017. We only used the high-resolution panchromatic band provided by each of these satellites, with spatial resolutions between 0.5–1.0 m. The number of shoreline observations acquired in this study is 10, a significant increase from the previously available high spatial resolution observations, which was 4, for this site since the 1950s.

Zoom In Zoom Out Reset image size

Airborne LiDAR data was acquired on 6 August 2011 for our study area, which provided a common base layer for georectifying all of the imagery. Initially, optical images were automatically orthorectified using the RPC information embedded in the image file and the LiDAR DTM (1 m postings), but the results showed variability in the position of ice-wedge intersections on the order of 2–5 m. To improve image rectification, we selected 20 ground control points per image using the LiDAR DTM as the base map. A second order polynomial transformation was applied resulting in the images being georectified to UTM NAD83 Zone 5N, with spatial resolutions ranging from 0.5–1.0 m. The mean rms associated with the georegistration process ranged from 0.44–0.85 m (SOM table 1), with a maximum individual registration point rms error always less than 1.5 m. Visual comparison of each optical image strip for our study area showed excellent spatial agreement and suitability for further analysis in spite of differing image acquisition conditions. Difficultlies in the use of automated approaches for delineating blufflines in high-spatial resolution optical imagery (as recently noted by Lantuit et al (2011) and Günther et al (2013, 2015)) required manual delineation of the coastal permafrost bluff line. The bluff line was manually digitized in each image independent of one another at a scale of 1:1000. We also included the bluff line position from 2007 aerial photography as reported in Jones et al (2009b) to expand annual coverage and have a complete decade of annual observations.

Bluff position measurements were made at 10 m increments along the study coast using the Digital Shoreline Analysis System (DSAS v. 4.4) (Thieler et al 2017). This tool measures the change in distance between two vector lines relative to a baseline and is widely used to measure coastal changes in the Arctic (Jones et al 2008, 2009a, 2009b, Gibbs and Richmond 2015, 2017, Farquharson et al 2018). The baseline in our study was created by taking a buffer of the 2007 shoreline and isolating the offshore line vector. Transects were cast every 10 m along this baseline using a 200 m smoothing algorithm to account for subtle undulations in the coastline and to ensure perpendicular transects. This resulted in 888 transects along the ∼9 km baseline. Since two small segments of this coast represent areas with small streams flowing into the ocean without exposed coastal bluffs, these were removed from further analysis. The end result provided a measure of bluff line erosion along the study coast at 876 measurement points annually for the past decade.

While it is difficult to accurately assess errors in erosion rate measurements associated with this type of analysis (Lantuit et al 2011), we adopted techniques used in previous coastal change detection studies (Hapke 2005, Lantuit and Pollard 2008b, Jones et al 2009b, Gorkhovich and Leiserowiz 2011, Gibbs and Richmond 2017). These are based on the identification of factors that contribute to the error associated with feature delineation in the images under comparison (SOM table 1). Potential sources of error include the spatial resolution of the imagery, the rms error associated with image registration, and the ability to accurately map the bluffline in the same optical image, as a proxy for producers uncertainty as averaged from the digitization of the same image three times (SOM table 1).

We extracted daily and bi-daily sea-ice concentrations at Drew Point between 1979–2016 using Nimbus-7 SMMR and DMSP SSM/I-SSMIS Passive Microwave Data from the National Snow and Ice Data Center (NSIDC) to define annual open water periods (Overeem et al 2011). Using three, 25 km² nearshore pixels, sea-ice concentrations <15% were flagged as open water. The open water duration was defined as the average of these three pixels exhibiting less than 15% sea ice concentration in a given year. The first, last, and total number of open-water days per year for each sampled pixel were compiled for the study period (figure 4). SST data were derived from the NOAA Optimum Interpolation (OI) SST V2 dataset (Reynolds et al 2002) for the three grid cells located between 71°N–72°N and 154°W to 152°W. Weekly SST data were averaged for the various open water periods determined with the NSIDC open water duration dataset. Locally, a time lapse camera was also installed on a pipe anchored into the subsea permafrost in August 2016 and provided hourly images for determining the wind speed and direction necessary for conducting geomorphic work which was used to determine storm conditions of interest (figure 5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Onshore, we collected hourly data for wind speed and direction and air and ground temperatures using the US Geological Survey meteorological station which has operated at Drew Point since 1998 (Urban and Clow 2016). We compiled hourly air temperature data from June–October to characterize the summer season, wind speed/direction data for the open water period for each respective year, and near-surface summer/fall (June–October) permafrost temperature data from 2007–2016. The hourly air temperature data have been summed to daily means and used to calculate the number of thawing degree days (based on 0 °C) for each period. The wind data and the time lapse camera (figure 5) were used to identify wind events or storms capable of forming erosional niches at the bluff base and/or collapsed block degradation (figure 5). The time lapse images showed that the geomorphologically significant winds were generally those with wind speeds greater than 5 m s⁻¹ from directions of 240°–360° and 0°–90°. Thus, we modified the methods of Atkinson (2005) to represent winds exceeding 5 m/s from the directions mentioned above for a period of at least 12 h with no lulls >6 consecutive hours. Each wind or storm event was further summarized according to a storm-power metric (Atkinson 2005) taken as the square of a storm's average wind velocity relative to its duration. The various open water duration assessments were used to summarize storms or winds indicative of conducting geomorphic work in a given open water period. Permafrost temperature data were aggregated to summer/fall (June–November) seasonal means.

Early 21st century, mean annual erosion has increased at Drew Point, ABSC when compared to the latter half of the 20th century (figure 6(a)). The increase in erosion reported in Jones et al (2009b) for the period 2002–2007 (16.3 m yr⁻¹) relative to the 1955–1979 (7.0 m yr⁻¹) and 1979–2002 (9.4 m yr⁻¹) time periods has been sustained between 2007–2016 (17.2 m yr⁻¹). This indicates that changes observed at this particular site are likely linked to ongoing shifts in the atmospheric, terrestrial, and/or marine conditions increasingly typical of the warming 21st century Arctic and not the result of enhanced erosion associated with a few catastrophic events where 25–40 m of erosion in a single year can have a big impact on the decadal-scale average (Are 1988, Lantuit et al 2012). In spite of a sustained increase in erosion of 17.2 m yr⁻¹ at Drew Point, year to year variability in open water season erosion was as high as 15.9 m. The range in mean annual erosion of 6.7 m in 2010 to more than 22.0 m in 2007, 2012, and 2016 (figure 6(b)) provided the basis for standardizing nearly annual observations of coastal bluff change using the number of open water days between image acquisitions to explore various environmental drivers.

Zoom In Zoom Out Reset image size

Erosion rates are typically reported on annual to decadal time-scales in the Arctic but focusing on the open water period when erosion is occurring may better resolve the processes driving coastal permafrost bluff retreat (Overeem et al 2011). Our nearly annual time series of high resolution satellite images allowed us to constrain open water season erosion between 2007–2016. In table 1, we report an erosion year which refers to the roughly annual period of image observations available for our study coast. Between 2007–2016, the average open water duration (OWD) was 91 days, but it ranged from 71 days (2014) to 107 days (2008 and 2016). In 2010, open water duration erosion was 0.08 m d⁻¹ and more than 0.20 m d⁻¹ in 2007, 2012, 2014, 2015, and 2016 (table 1). However, the difference in open water duration season did not correspond to periods of the lowest and highest observed coastal bluff losses. In 2008, 2009, and 2011–2014 the ability to bracket the open water period in a given year was not possible. However, OWD as derived from satellite remote sensing data constitutes our erosion year and thus we have considered the timing of image acquisition relative to measured erosion and accounted for this when summarizing erosional losses and open water days. Thus, when assessing erosion on a near-annual basis, the hypothesis that OWD is a good first order predictor of coastal erosion at Drew Point does not hold up.

Factors contributing to patterns of coastal bluff retreat include open water season, SST, summer air temperature, and permafrost temperature, yet few studies have explored their correlation with rates of erosion (figure 7). Barnhart et al (2014b) indicated that the combination of OWD and the number of storms during this period were important factors controlling erosion at Drew Point. On average, there were ∼11 storms per year between 2007–2016. In the 2010 erosion year, the year with the lowest measured bluff retreat of 6.7 m, the fewest storms occurred (n = 8) and in the 2012 erosion year, the year with the highest measured bluff retreat 22.6 m, the most storms occurred (n = 17). While the assertion that the combination of the number of storms during an open water period holds true at Drew Point on the extreme end of observations, we find that the correlation between the two variables over the study period yields a low R² (0.21) (figure 7) and an attempt to correlate variability in cumulative storm strength in a given erosion year yielded even lower relations (R² = 0.09). We also correlated mean erosion year variables indicative of SST, summer air temperature, and permafrost temperature, and all were weak and not statistically significant (figure 7). Multiple linear regression, forward stepwise regression, and best subsets regression of our erosion year open water season time series at Drew Point did not reveal any statistically significant relations either.

Zoom In Zoom Out Reset image size

Do the dynamics of permafrost coastlines serve as critical indicators of changes in the Arctic System?

What factors appear to be responsible for an increase in permafrost coastal erosion?

How do various environmental forcing factors interact with one another to drive coastal permafrost bluff erosion?

Our study underscores the challenge in using remotely-sensed snapshots of landscape change to confidently identify the processes driving the observed increase in coastal permafrost bluff erosion rates along the ABSC. While our datasets facilitated a continuous suite of observed erosion over a decade for Drew Point, complex oceanographic and geomorphic feedbacks limit the ability of our approach to discern the impact of various environmental forcing factors. For example, empirically-based modeling approaches that have been employed in the Drew Point area have experienced a similar kind of limitation regarding process-based understanding. Our work, taken within the context of contributions from the rapidly-emerging Arctic coastal research community, encourages the pairing of carefully-designed field monitoring and multi-physics (i.e. oceanographic, thermal, and mechanical) model development. Taken together, this kind of 'measure and model' approach may further elucidate the sensitivities of Drew Point (and other indicator sites in the Arctic) to uncertain environmental futures.