Quantifying landscape change in an arctic coastal lowland using repeat airborne LiDAR

LiDAR was collected in the summers of 2006 and 2010 by Aero-metric Inc. The 2006 data were delivered as classified point clouds in NAD 83 Alaska State Plane Zone 3 in feet, with elevations referenced to Geoid99 in US Survey feet. The 2010 data were delivered as last return point clouds in NAD 83 UTM zone 6N in meters, with elevations referenced to the GRS 1980 Ellipsoid. The 2006 data were transformed to match the horizontal and vertical units and Ellipsoid reference frame of the 2010 data. Nominal point spacing for the data was on average one point per 1–1.5 m for both datasets. Both last return, point cloud datasets were clipped to their overlapping extents, and then converted to surfaces, first by using Terrascan (version 12)^® and a Triangulated Irregular Network (TIN) approach. The TIN connects LiDAR returns by a set of contiguous, non-overlapping, Delaunay triangles. The elevations between each triangle vertex are interpolated as definitions of planes and thus together construct a surface. These surfaces were then exported to lattice datasets, and finally converted to 1.5 m raster grids to be used in a GIS.

The two datasets were differenced to identify potential changes in elevation between 2006 and 2010. The reported vertical accuracies of the data were 0.12 m (2006 dataset) and 0.14 m (2010 dataset) RMSE. To determine if there were any errors introduced in the conversion/transformation process we compared elevations along transects that likely represented stable terrain features associated with oil infrastructure gravel pads between the two acquisition dates. The difference in elevation along the unvegetated land surfaces were less than ±0.10 m (figure 3). However to be conservative in our detection of change, and with the realization that comparison of the data acquired over gravel pads represent a best case scenario, we used the RMSE of the individual datasets to calculate (equation (1)) a threshold that describes the minimum difference between both datasets that meets a significant elevation change at the 99% confidence level or those pixels exceeding three standard deviations (Jaw 2001):

$$\begin{eqnarray} &&\fl 3\text{SQRT}\left ((2 0 0 6\text{ vertical accuracy})^{2}\right .\nonumber\\ &&\left .+~(2 0 1 0\text{ vertical accuracy})^{2}\right )=0.5 5~\mathrm{m}. \end{eqnarray} \tag{ 1 }$$

In the surface differenced raster, we considered any pixel with a height difference residual above or below 0.55 m as a statistically significant change in elevation. These areas were reclassified as 〈significant increase〉 or 〈significant decrease〉 per pixel. Objects demonstrating significant increases or decreases in elevation were extracted from the dataset by selecting areas that had at least five contiguous pixels of significant change. This resulted in a minimum object size greater than 10 m² and helped minimize potential errors associated with raster grid creation and horizontal positional accuracy (±0.60 m). While there were detectable changes of smaller amplitude in the difference image we did not consider these in our further assessment. In addition, all changes interpreted as indicating water level differences as well as a result of human-caused landscape changes, such as from infrastructure construction or material movement, were excluded from further analysis.

Objects that represented a statistically significant change in elevation were manually classified as one of 13 categories via visual inspection of the LiDAR imagery. Nine of these classes represented land subsidence or loss and four represented land uplift or deposition. Objects that were indicative of subsidence included thermokarst and thaw-related features associated with ice-bonded coastal, river, and lake bluffs, frost mounds, ice wedges, and thermo-erosional gullies. Non-thaw-related erosional features included beaches and spits, river bars and deltas, and sand dunes. Areas that represented landscape uplift or deposition included permafrost heave features and deposition associated with beaches and spits, river bars and deltas, and sand dunes (often in close proximity to an eroding feature).

The West Dock (WD) and Deadhorse (DH) Thermal State of Permafrost (TSP) permafrost observatories were established in the late 1970s (Osterkamp 2003). The WD TSP site is located 0.3–0.4 km from the coast and the DH TSP site is located 15 km from the coast (figure 1). Vegetation at each site consists of wet non-acidic tundra with a continuous cover of graminoid and moss species and wet and moist non-acidic tundra composed of graminoid–moss tundra and graminoid, prostrate-dwarf-shrub, moss tundra, respectively (Walker et al 2008). In 1986, a string of calibrated thermistors attached to a data logger were installed at each site (Osterkamp 2003, Romanovsky and Osterkamp 1995). In 1997, the measuring systems at each site were upgraded with new calibrated thermistor strings (MRC thermistor string) and Campbell Scientific data loggers (Romanovsky et al 2003). The old and new measuring systems were run concurrently for two years and differences in the temperature readings obtained from the two measuring systems at the same depth were typically within 0.2 ° C. Mean annual ground temperature at a depth of 70 cm for the time period of 1987–2010 was calculated using the daily averaged records from the WD and DH TSP sites. In 1996, a one hectare grid was established at each permafrost observatory to measure active layer thickness. Measurements were conducted every 10 m in accordance with the Circumpolar Active Layer Monitoring (CALM) program protocol (Brown et al 2000). To derive mean annual air temperature trends and TDD (based on 0 ° C) we used data from the Deadhorse airport climate station (figure 1). This station has been in operation since 1973, however due to data continuity issues we only present data for the period 1983–2010. These two permafrost monitoring sites and the climate station are considered representative of conditions within the repeat LiDAR study area.

Our results indicate that repeat airborne LiDAR measurements provide a straightforward, readily applied tool for quantifying landscape change in arctic coastal lowland regions. Comparison of the two LiDAR datasets revealed that 0.3% of the landscape area experienced a significant change in elevation between 2006 and 2010 that resulted from thaw and non-thaw-related processes. Thermokarst or thaw-related landscape features associated with ice-bonded coastal, river, and lake bluffs, thermokarst pits, thaw slumps, and thermo-erosional gullies accounted for 46% of the change in area over the four year study period and more than half of the significant change (56%) resulted from erosion and deposition associated with beach and spit deposits, riverine and deltaic flats, and sand dunes. These short-term landscape dynamics are likely a result of a combination of factors related to both natural processes and changes in the terrestrial, aquatic, and marine controls on the region.

Detailed examples of features detected in the study area include frost mound degradation (figure 4(a)); coastal bluff erosion, thermo-erosional gully expansion, and upland ice-wedge degradation (figure 4(b)); and lowland ice wedge degradation and thaw pit formation (figure 4(c)). Comparison between the 2006 and 2010 LiDAR data for the frost mound shows the excellent agreement (typically ±0.15 m) between the two datasets for a vegetated and sloping surface that experienced a minimal change in height. Degradation or slumping of the eastern margin of the feature accounted for quantifiable change. In the example given in figure 4(b), the ∼4 m high bluff-face migrated 13 m inland over the four year study period. Also, evident in this example is an overall lower tundra relief in 2010 compared to 2006. Tundra lowering in this example did not meet our minimum threshold for change detection however, a consistent decrease in the elevation is visible in the plotted data that could be a result of top down permafrost thaw and soil consolidation. This general pattern of tundra relief lowering is also evident in figure 4(c) for which we also detected a number of degrading ice wedges. The changes evident in this example may be a result of frequent inundation of the terrain during storm surges (Sultan et al 2011) which may lead to degradation of the ice-rich permafrost and degradation of ice wedges (figure 2(b)). With more local ground control statistically significant changes in elevation could be refined to capture these more subtle surface changes.

Arctic coastal lowland regions are particularly vulnerable to change given the interaction of atmospheric, terrestrial, aquatic, and marine influences on landscape configuration. Thermokarst and other thaw-related landscape features are driven by either local or regional disturbances. Jorgenson et al (2006) noted an abrupt increase in ice-wedge degradation and thermokarst pit formation on the Beaufort Sea coastal plain that was attributed to an increase in TDD in the late-1980s and throughout the 1990s. A compilation of TDD from 1983 to 2010 for the Deadhorse climate station shows that the TDD in 2006, 2008, and 2010 exceeded the ∼30 year average (figure 5). In addition, the 2006–2010 time period experienced some of the highest combined, continuous mean annual and mean summer temperatures over the course of the record for the Deadhorse climate station. No individual year rivaled the number of TDD that occurred in 1989, 1995, or 1998, which Jorgenson et al (2006) attributed to the widespread formation of thermokarst pits. It should be noted however that the warm summers in the late-1980s and mid-1990s could also be responsible for some of the observed increase in pit formation in our study region between 2006 and 2010, as the thermokarst pits that likely formed during the 1990s have continued to subside. A recent analysis of thermokarst pit formation over a 60 year period in the Prudhoe Bay oil fields indeed shows a substantial increase in thermokarst pit formation between 1990 and 2001 that has continued through 2010 (Raynolds et al 2013).

Zoom In Zoom Out Reset image size

Mean annual air temperature and the temperature of the ground as measured at a depth of 70 cm have warmed in the study region during the past ∼20 to ∼30 years, respectively (figures 5 and 6(a)). Interestingly, there is a mismatch between these warming trends and the trend in active layer thickness (figure 6(b)). This lack of coherence may be explained by the thaw of ice-rich permafrost and settlement of the ground surface as excess ice melts and the soil consolidates. Liu et al (2010) measured ground deformation using InSAR from 1992 to 2000 in this region and found that seasonal vertical displacements occurred as a result of freezing and thawing of the ground surface but that there was a secular subsidence of the land surface on the order of 1–4 cm/decade. Ground based studies on the Beaufort Coastal Plain from 2001 to 2006 indicate that >2.0 cm yr⁻¹ of subsidence occurred and that when combined with mechanical probing measurements of the active layer that there was an increase in the 'true' thaw depth (Streletskiy et al 2008). Thus, while general ground subsidence was not detected due to our conservative measure of change some of the lowering of the landscape observed in the data outside our defined level of confidence may indeed be a result of top down permafrost thaw.

Zoom In Zoom Out Reset image size

The Alaskan Beaufort Sea coast has experienced varying levels of increased rates of erosion since the 1950s (Jones et al 2009, Gibbs et al 2011, Ping et al 2011). Factors responsible for this likely include loss of sea ice, warming ocean temperatures, and an increase in total wave energy (Overeem et al 2011). While no comprehensive studies on river discharge or river bluff erosion exist in the study region, Arctic wide assessments indicate varying degrees of changes to river discharge (McClelland et al 2006) which may be impacting erosion of river bluffs. These changes are likely interacting to increase the effectiveness of thermal erosion and thermal abrasion of coastal and river bluffs in our study area. These processes likely accounted for the dominant permafrost-related degradational processes with respect to area (42%) and volume (51%) in the study region. In addition, areas lower than 1.4 m relative to mean sea level have been inundated during storm surges over the past two decades (Sultan et al 2011). Two of the ten highest storm surges between 1993 and 2010 occurred during our study period and all but one occurred between 2000 and 2010. Inundation of the low-lying coastal zone in our study area has likely contributed to the formation of thermokarst pits, with 60% of the observed change associated with these features below this approximate elevation (figure 4(c)). These inundation events may have also contributed to some of the general land lowering along the low-lying coastal zone that was observed in the data but outside our defined level of confidence.

Estimates of total organic carbon in the upper 1 m of the studied landscape range from 56 to 66 kg OC m⁻² at sample sites located in the study region (Ping et al 2008). Based on these estimates, the loss of land between 2006 and 2010 along coast and river bluffs may have mobilized 1.8–2.1 Gg yr⁻¹. However, given our conservative estimates of quantifiable change this number should be viewed as a minimum estimate since long-term estimates of carbon flux from this coastal stretch range from 3.0 to 7.0 Gg yr⁻¹ (Jorgenson and Brown 2005). This exercise simply illustrates a basic type of analysis for carbon-related studies in permafrost terrain where repeat airborne LiDAR surveys have been acquired.