Please note our website will be undergoing maintenance on Tuesday, May 28, 2024. e-Commerce transactions and new registrations will be temporarily unavailable during this time. We apologize for any inconvenience this may cause.
Open access

Thaw slump activity measured using stationary cameras in time-lapse and Structure-from-Motion photogrammetry

Publication: Arctic Science
6 September 2018


Thaw slumps are one of the most dynamic features in permafrost terrain. Improved temporal and spatial resolution monitoring of slump activity is required to better characterize their dynamics over the thaw season. We assess how a ground-based stationary camera array in a time-lapse configuration can be integrated with unmanned aerial vehicle (UAV)-based surveys and Structure-from-Motion processing to monitor the activity of thaw slumps at high temporal and spatial resolutions. We successfully constructed point-clouds and digital surface models of the headwall area of a thaw slump at 6- to 13-day intervals over the summer, significantly improving the decadal to annual temporal resolution of previous studies. The successfully modeled headwall portion of the slump revealed that headwall retreat rates were significantly correlated with mean daily air temperature, thawing degree-days, and average net short-wave radiation and suggest a two-phased slump activity. The main challenges were related to strong JPEG image compression, drifting camera clocks, and highly dynamic nature of the feature. Combined with annual UAV-based surveys, the proposed methodology can address temporal gaps in our understanding of factors driving thaw slump activity. Such insight could help predict how slumps could modify their behavior under changing climate.


Les glissements dûs au dégel sont l’une des caractéristiques les plus dynamiques au niveau du terrain pergélisol. La surveillance d’activité des glissements au moyen de la résolution temporelle et spatiale améliorée est requise pour mieux caractériser leur dynamique au cours de la saison de dégel. Nous évaluons comment une série d’appareils photo stationnaires au sol en configuration pour prises de vues à intervalle peut être intégrée aux levés par véhicules aériens sans pilote (UAV) suivi du traitement de structure à partir du mouvement afin de surveiller l’activité des glissements dû au dégel à de hautes résolutions temporelles et spatiales. Nous avons réussi à construire des modèles de nuages de points et surfaciques numériques de la zone du mur de rimaye d’un glissement de dégel à intervalles de 6–13 jours au cours de l’été, améliorant significativement la résolution temporelle de décennale à annuelle des études précédentes. La partie du mur de rimaye du glissement, modelée avec succès, a révélé que les taux de retrait du mur de rimaye étaient significativement corrélés avec la moyenne quotidienne de température de l’air, les degrés-jours de dégel et la moyenne de rayonnement de courtes longueurs d’onde et suggèrent une activité de glissement à deux phases. Les principaux défis étaient liés à la forte compression d’images JPEG, aux horloges d’appareil photo à la dérive et à la nature très dynamique de la caractéristique. De pair avec des levés annuels au moyen d’UAV, la méthodologie proposée peut adresser les écarts temporels dans notre compréhension des facteurs déterminant l’activité de glissements dûs au dégel. Un tel aperçu pourrait aider à prévoir comment les glissements pourraient modifier leur comportement sous le changement du climat. [Traduit par la Rédaction]


The warmer and wetter northern climate during recent decades has led to increased thermokarst activity in permafrost terrain (Kokelj and Jorgenson 2013; Liljedahl et al. 2016; Segal et al. 2016; Fraser et al. 2018). Retrogressive thaw slumps are one of the most dynamic thermokarst features as each can thaw hectares of permafrost annually (Burn and Lewkowicz 1990). In northwestern Canada, thousands of thaw slumps have been inventoried (Lantuit et al. 2012; Lacelle et al. 2015; Obu et al. 2016; Segal et al. 2016; Kokelj et al. 2017) and these have been shown to modify the discharge of streams and rivers (Kokelj et al. 2013) and the geochemical composition and sediment loads of streams and lakes (Kokelj et al. 2009; Malone et al. 2013; Rudy et al. 2017; Tanski et al. 2017). This has negatively affected aquatic ecosystems, including benthic macroinvertebrates communities (Chin et al. 2016). Therefore, increased knowledge on the dynamics of thaw slump activity is required to better characterize their impact on the environment.
To date, thaw slump activity has been investigated at regional scales using a variety of remote sensing approaches, including (i) historical air photographs acquired at decadal time-scales (Lantuit and Pollard 2005, 2008; Lacelle et al. 2010; Swanson 2012); (ii) medium-resolution satellite imagery acquired at an annual time-scale, such as visual observations using SPOT or Landsat images (Kokelj et al. 2017; Swanson and Nolan 2018) or tasseled cap trend analysis of Landsat image stacks (Brooker et al. 2014; Fraser et al. 2014); and (iii) radar interferometry performed at sub-seasonal time-scales (Short et al. 2011; Zwieback et al. 2018). Thaw slump dynamics have also been characterized at more local scales using (i) digital surface model (DSM) differencing of LiDAR datasets acquired at annual time-scales (Obu et al. 2017); (ii) optical imagery acquired with an unmanned aerial vehicles (UAVs) at annual to sub-seasonal time-scales (van der Sluijs et al. 2018); (iii) high-frequency repeat terrestrial laser scanning (Barnhart and Crosby 2013); (iv) dense high-resolution TerraSAR-X time-series (Stettner et al. 2018); or (v) time-lapse photography from a single ground-based stationary camera from which a qualitative index of change over the thaw season was developed (Kokelj et al. 2015). A combination of remote sensing tools, such as optical satellite imagery combined with historical air or radar imagery was also used to study the dynamics of these features (Lantuit and Pollard 2008; Balser et al. 2014; Segal et al. 2016). Despite the variety of remote sensing approaches, a gap exists in the monitoring of thaw slumps (and other thermokarst landforms) at high spatio-temporal resolution, which could improve the understanding of short-term drivers modifying the features (Kokelj and Jorgenson 2013).
The recent emergence of Structure-from-Motion (SfM) photogrammetry combined with automated time-lapse ground-based cameras now allow for the characterization of landforms at high temporal and spatial resolutions (Westoby et al. 2012). The SfM processing methods perform the photogrammetric processing of 2D images and generate 3D spatial data in the form of point-cloud models from which change detection and deformation analysis of landforms can be performed (Westoby et al. 2012; Hugenholtz et al. 2013; Turner et al. 2015). To date, very few studies have produced continuous datasets with time-lapse SfM photogrammetry (James and Robson 2014; Eltner et al. 2016; Mallalieu et al. 2017). In this study, we assessed how a ground-based stationary camera array in time-lapse configuration, combined with annual UAV-based surveys, can be integrated with SfM processing to monitor the activity of thaw slumps (i.e., headwall retreat rate, volume of sediments transported along the slump floor) over the thaw season (i.e., daily to weekly interval), one of the most dynamic feature in permafrost terrain. The proof-of-concept study was conducted at a typical thaw slump on the ice-rich Peel Plateau (NWT, Canada) where the time-lapse dataset of slump activity was compared with local weather conditions.

Study site

The study site is situated in the Vittrekwa River watershed on the Peel Plateau of northwestern Canada (Fig. 1). The plateau lies in the southeastern foothills of the Richardson Mountains and gently slopes eastward from 650 m above sea level (a.s.l.) in the foothills of the Richardson Mountains to 100 m a.s.l. at the Peel River. Permafrost in the region is continuous and warm (e.g., in the −3 to −1 °C temperature range; O’Neill et al. 2015) with ground-ice content ranging from 50%vol to tabular bodies of pure ground ice (Lacelle et al. 2004, 2010, 2013). The ice-rich landscape of the Peel Plateau is dotted by hundreds of active thaw slumps of various sizes that developed in tills and glacio-lacustrine sediments deposited between 18 and 15 k cal year B.P. (Lacelle et al. 2013, 2015). In a 15 700 km2 area of the Peel Plateau and Richardson Mountains, 189 active slumps ≥0.4 ha were identified using trend analysis of Landsat imagery; the average surface area was 5.4 ha, and included 10 slumps >20 ha (Lacelle et al. 2015). A random sample showed headwall retreat rates from 5 to 60 m year−1, with an average rate of 12.4 ± 7.4 m year−1 (Lacelle et al. 2015). Over the last few decades, the region has seen a pronounced increase in the number of large and active thaw slumps (Kokelj et al. 2015; Lacelle et al. 2015).
Fig. 1.
Fig. 1. Map showing location of thaw slumps, including D1 slump and GNWT automated meteorological station, on the Peel Plateau, NWT, Canada. Distribution of thaw slumps in the area is from Lacelle et al. (2015).
The selection of a thaw slump for this study, referred to as D1, was based on its accessibility, size, and feasibility to install stationary cameras and operate a UAV. A second site was instrumented; however, grizzlies damaged the camera set-up (hence the need of a low-cost system). The D1 slump is situated on the south side of the Dempster highway in the foothills of the Richardson Mountains (67°10.4′N, 135°43.1′W; 580 m). The slump has a northerly exposure and a surface area of 0.5 ha. The slump floor has a small creek running through it at the topographic low near the toe of the slump and the Dempster highway. Vegetation around the slump consists of low shrubs, sedges, and lichens.
The climate in the region is characterized by long cold winters and short cool summers. The mean annual air temperature in Fort McPherson (1986–2010), the closest Environment Canada meteorological station, is −6.8 ± 1.6 °C and total annual precipitation averages 295 mm, with rainfall accounting for approximately half (Environment Canada 2010). In summer 2010, an automated meteorological station was installed by the Government of Northwest Territories (GNWT) along the Dempster Highway, in proximity to D1 (67°14.77N, 135°13.22W; 410 m). In summer 2016 (June to August), air temperature averaged 11.3 ± 6.6 °C with rainfall totaling 399 mm (data provided by GNWT Water Resources Division of the Department of Environment and Natural Resources).


Image acquisition for SfM-based mapping is typically performed using a single camera mounted on a piloted airborne platform (Nolan et al. 2015) or UAV (Hugenholtz 2012; Hugenholtz et al. 2013; Turner et al. 2015), which generally limits temporal coverage due to required presence of a human operator in the field. Continuous time-lapse datasets with SfM photogrammetry have only been produced for less dynamic features such as lava flows (photographed by a human operator over a 37 min period; James and Robson 2014) and glacier margins (18 SfM models over a 426 days period using autonomous cameras; Mallalieu et al. 2017). Here, we test if an automated array of ground-based cameras in time-lapse mode can be used to track changes in thaw slump activity over the thaw season (a highly dynamic feature) at high spatio-temporal resolution with a targeted 1-day interval, but dependent on weather conditions. Twelve stationary trail cameras were used to collect images between 27th June and 2nd September 2016 to produce georeferenced point-cloud models of the D1 slump using SfM photogrammetry. Digital cameras mounted to UAVs were also used to systematically collect overlapping photographs on 27th July 2015 and 3rd August 2016. The ground- and UAV-based point-cloud models were then used to quantify changes in thaw slump activity. Given that headwall retreat rate in the study area averages 5–10 cm day−1 (Lacelle et al. 2015), the accuracy of the SfM models should be at least half of that (2.5–5 cm) for expected daily reconstruction.

Acquisition of datasets

UAV imagery and ground control points (GCP)

In July 2015, digital photos of the D1 slump were acquired using a 24 megapixel Sony a6000 mirrorless camera with a Sony f/2.8 20mm compact lens mounted to a Spyder PX8 Plus 1000 octocopter. The UAV used the 3DR Pixhawk flight 4 controller running ArduCopter version 3.2.1. Mission Planner version 1.3 software was used to plan a gridded survey with 85% forward and 75% side photo overlap at 70 m height above the ground (∼1 cm ground resolution). The photos were captured at 400 ISO, shutter speed of 1/1000 s, and with focus fixed at infinity. Global positioning system (GPS) tags were added to the JPEG Exif (Exchangeable image file) information using the log file from the onboard 3DR UBlox GPS.
The accuracy of the georeferencing was optimized with four GCP (small disks with a cross marked in their center) surveyed using a Trimble R9s dGPS and a local base station that logged for 6 h 47 min. The base station position was corrected against the NRCan Online Precision Point Positioning System (, yielding an estimated 95% sigma position error of ±1.1 cm (X), ±1.8 cm (Y), and ±3.3 cm (Z).
In August 2016, the D1 slump was surveyed using a 16 megapixel DJI Zenmuse X5 camera mounted to a DJI Inspire Pro quadcopter. The Litchi flight app for DJI was used to program a survey with similar overlap, height, and ground resolution to the 2015 gridded survey. The photos were captured at 200 ISO and shutter speed of 1/1000 s. Four GCPs were surveyed in a similar manner as in 2015 using a Leica GS14 Performance Smart Antenna dGPS and yielded the same accuracy as the 2015 survey.

Time-lapse imagery using stationary trail camera array

The slump was photographed between 27th June and 2nd September 2016 using an array of 12 Solar SpyPoint trail cameras (12 megapixel, f/3 aperture, and 7.3 mm focal length) in time-lapse mode (Fig. 2A). These cameras were selected for their weather-protective casing, solar power source with back-up batteries, and affordability given the high number of cameras required and likelihood of wildlife damaging the set-up. The cameras were programmed to acquire images at hourly intervals, which increased the likelihood of obtaining a set of daily images under optimal conditions since SfM algorithms can be sensitive to certain environmental conditions (i.e., textural differences between images and variability in shadows due to sun angle; Micheletti et al. 2015). We attempted to calibrate the cameras individually before field deployment using the Agisoft lens camera calibration software. Unfortunately, without a view finder, manual focus, and shutter on the SpyPoint cameras, proper capture of the calibration target screen proved very difficult and precalibration was ultimately unsuccessful. However, given the wide variation in temperature at the field site during the 72 days acquisition period, the lens internal geometry and optimal predetermined camera calibration parameters would likely have drifted over the study period.
Fig. 2.
Fig. 2. (A) Field photograph showing the installation of cameras in the array along D1 slump on the Peel Plateau, NWT, Canada. Arrows point to targets. (B) Geometry of camera array (numbered from 1 to 12) and associated fields of view.
The camera array was configured to provide a range of orthogonal and oblique views to maximize coverage (>60% image overlap) of the slump. Due to the configuration of the slump and feasible installation points, the camera array was separated into two groups (Fig. 2B). Cameras 1 through 5 were positioned ∼50–70 m distance from the slump headwall and provided ∼100% image overlap. A small stream cutting through the slump floor forced cameras 6–12 to be installed away from the first array. The second array was located on a topographic high above the stream closer to the target (∼25–50 m distance) and provided ∼70% image overlap. All cameras were mounted to iron posts that were drilled >1 m into the permafrost to prevent camera movement and minimize data loss due to shift in orientation. The latter proved to be minimal from visual inspection of differences in the field of view of images; only camera 6 began to experience a slight change in orientation (but not a translational movement) in late summer causing a slight deviation in image overlap. The set-up of the camera array gave variable baseline offsets (1:5–1:9) and vertical offsets (1–2.2 m above the ground surface), which are important to reduce the image distortion in ground-based photogrammetry and SfM (Chandler et al. 2002; Fonstad et al. 2013; Micheletti et al. 2015).
The locations of the trail cameras to their tops (the offset was then subtracted to the center of lens for each camera) were surveyed using a Leica GS14 Performance Smart Antenna dGPS because it has been demonstrated that accurate SfM models can be generated from only the known camera positions and orientations, called direct georeferencing (Hugenholtz et al. 2016; Nolan and DesLauriers 2016). Targets were also positioned above the headwall of the slumps (visible in the field of view of the cameras because the terrain slopes upward) to provide additional anchoring points. These were not directly surveyed, but the coordinates of the targets were extracted from the 2016 UAV-survey and point-cloud model and represent “pseudo GCPs” (e.g., Stöcker et al. 2015); no targets was positioned on the slump floor because it is often saturated by ground-ice meltwater and becomes fluid (Kokelj et al. 2015).

Point-clouds processing

UAV imagery and rectification to real-world coordinates

Agisoft Photoscan Pro version 1.4 SfM software was used to generate point-cloud models of the D1 slump. The UAV models were initially aligned with the onboard camera Exif GPS data, which provides positional coordinates for each image. After initial alignment and camera calibration, georeferencing and camera calibration were optimized using the four GCPs before a dense point-cloud model was generated. The use of 3–4 GCPs has shown to generate accurate point-cloud models and increasing their number does not greatly improve model accuracy (Tonkin and Midgley 2016). The SfM processing report for the model indicated a residual mean square error of <1 cm. However, comparison of stable objects within small barren sites just upslope of the headwall in the 2015 and 2016 UAV point-cloud models suggested that there was a 2–4 cm x-y offset and a 5–10 cm vertical offset. This assessment of real-world accuracy of the point-clouds model likely better represents their true registration accuracy. The dense point-cloud models in LAS data format (NAD83 UTM Zone 8N projection) were used to generate orthomosaics and DSMs.

Stationary camera imagery and rectification to real-world coordinates

Following the recovery of the trail cameras, it was found that 11 of the 12 cameras had collected hourly images over the 72 days acquisition period, resulting in ∼20 000 images, all saved on the built-in storage of the cameras (camera 2 failed 2 days after installation due to firmware issues; out of the ∼20 000 images, only ∼800 of them would have been used for targeted daily reconstruction). Prior to processing in Agisoft Photoscan, the images were manually examined to remove those captured (1) at night, which triggered the infrared setting; (2) during or after precipitation events when the camera lens was wet; (3) under poor lighting conditions; (4) reflective glare on the ablating headwall and water-saturated slump floor; and (5) during periods of rapid terrain deformation. The cameras were programmed to capture images every 60 min; however, the time-stamp of each image in their Exif tags varied between cameras by 56–63 min and caused a random offset in the timing of image capture that compounded over the monitoring period. These limitations, some of which due to climate conditions and others due to the dynamic behavior of the target, resulted in high-quality sets of images for only 8 dates (27 June, 7 July, 17 July, 25 July, 1 August, 14 August, 20 August, and 2 September). Point-cloud models for each of these dates were generated using the SfM software with prior masking of the images to remove the banner and background and the camera calibration parameters were allowed to vary at all steps during bundle optimizations (Fig. 3). Upon first attempt, the software was unable to align the images from the 11 cameras, even when the coordinates of the cameras were provided; the software was only able to build separate point-cloud models for camera array #1 and #2 (array #1: cameras 1–5; array #2: 8–12; cameras 6 and 7 were removed due to presence of growing vegetation that partially obstructed the view). Since the camera locations did not improve the SfM model accuracy, we therefore provided the coordinates of the “pseudo GCPs” to the SfM software prior to photo alignment and reconstruction of the ground-based point-cloud models from camera array #1 and #2 (Fig. 4; e.g., Stöcker et al. 2015). Although only as accurate as the UAV point-cloud models (∼2–4 cm), these “pseudo GCPs” provided the anchoring points required to align, merge, and georeference the two separate models for each date. The resulting point-cloud models for each date covered most of the headwall and the slump floor in proximity to the headwall; the slump floor at increasing distance from the headwall was poorly reconstructed, likely due to the highly reflective, wet surface that confused the SfM algorithms. Given that our approach aims to bridge the gap between annual UAV-based surveys, we ensured integration of datasets by conforming to a horizontal orthoplane. We therefore generated from the ground-based point-cloud models in LAS data format (NAD83 UTM Zone 8N projection), a series of orthomosaics and DSMs that allowed direct comparison with those of the UAV-based surveys. This standard choice in Agisoft required no translation or reprojection as orthomosaics are one of the standard SfM outputs. The accuracy of the ground-based point-cloud models to the 2016 georeferenced UAV model was assessed using the average distances differencing algorithm in the Cloud Compare software (; EDF R&D 2012).
Fig. 3.
Fig. 3. Photographs from stationary cameras of D1 slump on the Peel Plateau (NWT, Canada) showing example of (A) full image and (B) its masking in Agisoft Photoscan. Photographs from stationary cameras showing variable lighting between adjacent cameras over short time frames: (C) 2145, 31 July 2016 and (D) 2158, 31 July 2016. Arrows point to targets positioned above headwall and used as anchoring points.
Fig. 4.
Fig. 4. (A and B) Example of resultant two point-cloud chunks that required subsequent merging using the targets (shown as blue flags) installed above the headwall and from which their coordinates were extracted from the unmanned aerial vehicle (UAV) point-cloud models and provided to the Structure-from-Motion (SfM) software prior to photo alignment and the reconstruction of the ground-based point-cloud models. (C) Example of the full reconstruction of merged ground-based point-cloud model. Camera positions used in reconstruction are indicated as black lines (camera number indicated) and unaligned cameras are indicted with blue circles. Note that cameras are out of frame in panel (A).

Quantifying thaw slump activity

Among the different methods available to quantify changes in landforms and landscapes represented by point-cloud models, the most common is to derive raster DSMs from the point cloud and calculate their difference [i.e., DSM of difference (DoD); Lague et al. 2013; Turner et al. 2015]. For the UAV surveys, the 3rd August 2016 DSM generated in Agisoft from the point-cloud model was differenced from the 27th July 2015 DSM. For the ground-based point-cloud models, the DSMs between successive dates were differenced against each other and the difference values interpolated with kriging in ArcGIS (parameters defined to variable search radius, 12 points and maximum distance, 50 cm). The DoD results were displayed in ArcMap on a common scale and overlaid on the georeferenced 2016 UAV orthophoto.
Headwall retreat rate was calculated by delineating a linear topographic profile that started on the slump floor and intersected the apex of the arcuate-shape headwall in the most active section. Based on the change in elevation between the point-clouds models, we determined headwall retreat. A linear correlation analysis was then performed (Aable statistical software) between the calculated average slump headwall retreat rates for the different time intervals (6–13 days) between the eight ground-based DSM models (n = 7) and the mean daily air temperatures, thaw degree-days (index of mean daily air temperature above 0 °C for a specified period), total precipitation, and mean net short-wave radiation measured at the nearby GNWT meteorological station.


UAV-derived point-cloud models of D1 slump

The UAV point-cloud models consisted of ∼12 million points with a density of ∼2500 pts m−2 (Table 1). Based on the orthophotos, changes in the D1 slump between 27th July 2015 and 3rd August 2016 included upslope expansion, mainly in the south–west section of the headwall, and areas of stabilization on the slump floor (Fig. 5). Similar to the retreat pattern at other nearby slumps, D1 also exhibited differential retreat rates along the headwall, causing the headwall to develop lobate forms. Near the center of the headwall, a small region was more resistant to backwasting, whereas adjacent surfaces retreated upslope. The differential retreat along the headwall is likely caused by differences in ground-ice content in permafrost, or it is possible that the local slope was not sufficient to allow parallel headwall retreat. Based on differences in vegetation cover on the slump floor, the lower section of the floor appears to have stabilized given the more abundant vegetation in 2016.
Table 1.
Table 1. General characteristics of parameters in point-clouds models.

Note: Differences in values for the unmanned aerial vehicle (UAV) surveys relate to the full extent of the survey and not to the cropped analysis area.

Fig. 5.
Fig. 5. Orthophotos (true color) of D1 slump on the Peel Plateau (NWT, Canada) derived from Structure-from-Motion (SfM) models of unmanned aerial vehicle (UAV) imagery in (A) 27 July 2015 and (B) 3 August 2016. A pale rendering of the UAV-based orthomosaic is used as a backdrop.
The visual changes to D1 were quantified using the 2015 and 2016 DSM models (Fig. 6A). Spatially, the DoD method agrees with the visual description of changes to the slump: material was removed along the headwall and the topographic high just north of the creek cutting through the slump floor (evidenced from loss of elevation); material accumulated near the toe of the slump floor (evidence from gain in elevation), although part of it may be due to low shrub growth. The average DoD over the spatial extent of the thaw slump was 0.2 m, indicating that the icy sediments that ablated along the headwall accumulated on the slump floor; however, elevation loss along the retreating headwall was in the order of 2 m (consistent with the height of the headwall) and a maximum of 2 m of sediments accumulated in certain locations near the toe of the slump floor. Based on the width of the zone showing elevation loss along the headwall, it retreated by ∼5–10 m between 27th July 2015 and 3rd August 2016 (Fig. 6A).
Fig. 6.
Fig. 6. Elevation change results [DSM of difference (DoD)] along D1 slump of (A) digital surface model (DSM) derived from Structure-from-Motion (SfM) models of unmanned aerial vehicle (UAV) imagery in 27 July 2015 and 3 August 2016. (B and C) DSM derived from SfM models of stationary cameras: 7 July and 27 June 2016, 17 July and 7 July 2016. (D) Orthophoto of D1 slump showing position of headwall for various days in summer 2015 and 2016, as well as location of cross-section from which topographic profiles were derived (E).

Ground-based point-cloud models of D1 slump

The 8 point-cloud models reconstructed from ground-based stationary cameras had much lower point counts (1.5–4 million) than the UAV ones due to their smaller spatial extent; however, their point densities were comparable (Table 1). Although only the slump headwall and the slump floor in proximity to the headwall were successfully modeled, the orthophotos showed changes between the eight dates, which allowed visualizing slump activity during the 2016 thaw season (Fig. 7). Between 27th June and 2nd September 2016, the orthophotos revealed backwasting of the headwall, thawed sediments accumulating at the base of the headwall, and transport of material along the slump floor.
Fig. 7.
Fig. 7. Orthophotos (true color) of D1 slump on the Peel Plateau (NWT, Canada) derived from stationary camera array and projected over the 3 August 2016 unmanned aerial vehicle (UAV) orthophoto. (A) 27 June 2016, (B) 20 July 2016, (C) 17 July 2016, (D) 25 July 2016, (E) 1 August 2016, (F) 14 August 2016, (G) 20 August 2016, and (H) 2 September 2016. A pale rendering of the UAV-based orthomosaic is used as a backdrop.
These visual changes to D1 slump were quantified by sequentially differencing of the eight DSMs derived from the ground-based point clouds (Figs. 6B, 6C). The changes in elevation loss or gain were in the order of 1–2 m between the dates and the spatial agreement of change suggest that ground-based SfM method is detecting changes in locations where one would expect for thaw slumps.

Thaw slump activity and weather conditions

Topographic profiles were constructed for the UAV and ground-based SfM models along a cross-section that traversed the headwall of the slump (Figs. 6D, 6E). Despite the topographic profiles of the ground-based SfM models being noisier relative to the UAV-derived ones (because of errors in Z in the order of 5–10 cm and that the UAV–DSM rasters were derived from kriging interpolation), the ground-based SfM models were able to track headwall retreat during summer 2016: the position of the headwall was moving upslope as summer progressed (Figs. 6D, 6E). Based on these changes, we calculated the cumulative headwall retreat and average retreat rate between each date (6- to 13-day intervals; Fig. 8). Along the cross-section, the cumulative headwall retreat between 27 July 2015 and 3 August 2016 was 7 m; whereas between 27 June and 2 September 2016 it was 4.2 m. The calculated headwall retreat rates of the D1 slump for the different time intervals are in the range of 1.4–14.0 cm day−1, with an average of 6.1 ± 5.2 cm day−1. In 2016, the active thaw season was from early May to early October. Statistical analysis revealed significant correlations for the linear regression analysis between average headwall retreat rates and mean daily air temperatures, thawing degree-days and mean net short-wave radiation measured at the nearby GNWT meteorological station (Fig. 9).
Fig. 8.
Fig. 8. Comparison of (D) cumulative headwall retreat and (E) average headwall retreat rate for the time periods between each date (6- to 13-day intervals) of D1 slump on the Peel Plateau (NWT, Canada) in summer 2016 (derived from stationary cameras) with (A) mean daily air temperature (MDAT), (B) total rainfall, (C) net short-wave radiation measured by the GNWT meteorological station near D1 along the Dempster Highway.
Fig. 9.
Fig. 9. Scatter plots and linear regression analysis of average headwall retreat rates for the time periods between each date (6- to 13-day intervals; summer 2016) of D1 slump on the Peel Plateau (NWT, Canada) against (A) mean daily air temperature (MDAT), (B) thawing degree-days (TDD), (C) total rainfall, (D) mean net short-wave radiation measured by the meteorological station near D1 along the Dempster Highway (climate variables were calculated for the same time intervals).


The study is a proof-of-concept on the use of low-cost ground-based stationary cameras in time-lapse combined with SfM photogrammetric processing to quantify changes in thaw slumps at high spatio-temporal resolution. This approach can (1) bridge the temporal gap between UAV surveys because it doesn’t require human presence in remote locations; and (2) provide an alternative to monitor landforms at high temporal resolution in places where the geomagnetic field lines interfere with UAV navigation, especially in some mid to high Canadian Arctic regions where the compass becomes erratic. The approach proved that it can be used to derive accurate DSMs from the SfM point-cloud models that can be combined with UAV surveys; however, there were some challenges in generating the point-cloud models at our site and targeted 1-day interval, most of which were related to the cameras, poor weather during summer 2016, and dynamic behavior of the target. Despite such limitations, we were able to produce point-cloud models at a 6- to 13-day interval, which represents a considerable improvement compared with previous studies (i.e., decadal to annual temporal resolution). The DSMs derived from the point clouds allowed tracking thaw slump activity during the thaw season and exploring relations with local weather variables. Below we first explore local factors driving slump growth on the Peel Plateau, NWT and then discuss some of the challenges encountered with the study.

Time-lapse slump activity and weather conditions

The average headwall retreat rates (6- to 13-day interval) of the D1 slump were significantly correlated with mean daily air temperatures, thawing degree-days, and mean net short-wave radiation measured at the nearby GNWT meteorological station (Fig. 9). This finding is in agreement with previous studies that examined climate factors driving slump growth at sub-seasonal to decadal time-scales (Lewkowicz 1985, 1987; Grom 2008; Lacelle et al. 2015; Zwieback et al. 2018). Variations in headwall retreat rates between sites are attributed to internal factors (i.e., ground-ice content and its vertical distribution) and environmental factors (i.e., slope, aspect, and late lying snow-banks) where very large slumps can become self-perpetuating (Kokelj et al. 2015). Therefore, the higher temporal resolution of observations provided by a camera time-lapse approach did not yield new insights into general factors governing headwall retreat. However, our high spatio-temporal observations present evidence of a two-phased slump activity. The first phase involves higher headwall retreat rates during the period of maximum solar insolation in early summer with the thawed material accumulating at the base of the headwall (decreasing apparent headwall height as summer progresses). The second phase is the downslope transport and removal of accumulated materials on slump floor which progresses through the summer (a small mound on 27 June 2017 that is progressively eroded; Fig. 6E). These changes were not observed in the UAV-based models and other remote sensing approach due to their coarser temporal intervals (i.e., 1 year or greater repeat surveys).
The methodology has potential to track daily to weekly changes along the slump floor, which represents the conduit that transport meltwater and sediments to the fluvial environments (Malone et al. 2013; Kokelj et al. 2015). The slump floor is a highly dynamic feature and mainly qualitative changes have been described thus far (Kokelj et al. 2015). van der Sluijs et al. (2018) demonstrated that daily UAV surveys from which high spatial resolution DSMs are derived can detect rapid movement on the slump floor and allow establishing vectors of movement. With improvement to some of the issues we encountered, this will become possible in the near future to detect movement without a human operator.

Challenges and improvement strategies

Four main challenges were encountered with using the ground-based stationary trail cameras in time-lapse for SfM photogrammetry: (1) image compression, (2) drifting internal camera clocks, (3) illumination and variable weather, and (4) growth dynamic of the target. Unlike most cameras used in SfM for mapping applications, RAW format images or minimally compressed JPEG images were not available and the JPEG files were encoded using a strong compression, even at the highest quality camera image setting. The 8-by-8 pixel JPEG compression block structure is clearly visible when images are viewed in close-up. This means that despite an appropriate sensor size and resolution, the loss of fine detail and blocky compression artefacts presented challenges for the first step of SfM processing, which relies on matching features between cameras using a SIFT-like method (Lowe 2004). For comparison, JPEG images from the DJI Inspire Pro UAV, which had a slightly higher 16 MP resolution, were ∼7.5 MB, whereas those from the 12 MP stationary cameras were compressed to ∼1 MB. The strong JPEG image compression is likely the major reason for the higher degree of noise in the trail camera point-cloud models by comparison to the UAV-based models.
The second issue was the drifting of the internal camera clocks. The cameras were set to a 1-h time-lapse capture mode; however, it was discovered after the camera recovery that all camera clocks had drifted from their original synchronization. In most cases, photos were captured every 56–63 min, with the random interval within individual cameras. Over the study period of 2.5 months, the issue of clock drifting compounded, resulting in time-stamp errors of 1–30 min at the end of the observation period. This at first glance may not prove to be a major issue to study the dynamics of thaw slumps at daily time-scale, but becomes a significant issue when near-synchronous acquisition of photos is necessary to avoid variability in solar illumination and shadowing between photos that degrade the quality of SfM feature matching.
The third issue related to poor alignment of point-cloud pairs due to variable weather conditions causing variability in illumination. At the northerly exposed study site, the best lighting conditions occurred in early morning when both sun angle and ablation of headwall were low (the latter reduced the glare on the headwall). The window of ideal lighting was however quite short, usually lasting less than 1 h (8–9 h Mountain Standard Time, UTC – 7). Unfortunately, due to the internal clock drifting, full sets of photos from the 11 cameras were not often available within that ideal illumination window. In addition, periods with extended rainfall could not be used due to condensation and rain drops on the camera lenses in addition to greater sun-glint from wet slump surfaces.
Finally, the rapid movement of materials on the slump floor, of specific interest for our study, presented a major issue for the SfM matching algorithms. This issue was compounded by the drifting camera clocks that prevented synchronous acquisition of photos from the 11 cameras. Even if the time-stamps of the photos were offset by <5 min, this was sufficient to confuse the SfM algorithm (identify keypoint matching) during periods of rapid terrain deformation. This resulted in the slump floor not being reconstructed in the point-cloud models as they cannot model areas which moved or changed in shape between image acquisitions.
Overall, the main challenges were related to image compression, drifting clocks in the cameras, and highly dynamic nature of the target, acknowledging that trail cameras are generally intended for a much different purpose than SfM photogrammetry. The use of digital single-lens reflex cameras (dSLR) should resolve the image compression issue, but with separate concerns related to increasing cost, durability, and risk of damage by wildlife in Arctic regions (i.e., a separate instrumented slump had its camera set-up damaged by grizzlies). Clock drifting is common with digital cameras, including dSLR, and was also an issue in other studies (e.g., Mallalieu et al. 2017). Welty et al. 2013 proposed to integrate GPS receivers to the cameras if very precise time-stamp is required, but at the expense of greatly increasing cost and complexity of the system. Despite our ambition to generate point-clouds models at a daily temporal resolution, we were limited to useable photos on eight dates during the summer season with 6- to 13-day intervals, still greatly improving the temporal resolution to study these features. This allowed us to demonstrate the potential of the approach to monitor at high spatio-temporal resolution highly dynamic features and thaw slump activity. Time-lapse stationary camera combined with UAV-based surveys and SfM can become a powerful tool to monitor landforms changes in Arctic regions where the operation of UAVs can be challenged by weather and close proximity to magnetic north.


The purpose of the study was to explore the potential of a cost-effective ground-based stationary camera array in a time-lapse integrated with SfM processing to monitor the activity of thaw slumps in remote locations. Based on the findings, the following conclusions can be made:
We were able to successfully construct point-clouds and DSM models of a thaw slump at 6- to 13-day intervals over the thaw season, a significant improvement over previous studies.
The successfully modeled headwall portion of the slump revealed that headwall retreat rates over the 2016 thaw season were significantly correlated with mean daily air temperature, cumulative thaw degree-days, and mean net short-wave radiation measured at a nearby meteorological station.
The main challenges were related to strong JPEG image compression and drifting camera clocks. The latter prevented construction of point-cloud models of highly dynamic sections of the slump. The varying sun angles also caused some issues of shadowing, and it was found that in the case of D1 slump, the optimal sun angle for photogrammetry occurred between 8 and 9 h MST. This would limit the best temporal resolution to be at daily time-scales. The rapid technological advances in digital cameras will likely reduce most of the challenges encountered in this study.
The time-lapse datasets for the study of thaw slumps have potential to address temporal data gaps in our understanding of factors governing their activity. Such insight could help advance the understanding of thaw slumps and how we may expect them to modify their behavior under changing climate.


This work was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant with logistical support provided by NSERC Northern Supplement, and the Northern Scientific Training Program. We thank H. Crites, B. Wilson, and J. van der Sluijs (NWT Centre for Geomatics) for valuable field assistance and Shawne Kokelj (Water Resources Division, GNWT) for supplying Peel Plateau meteorological station data.


Balser A.W., Jones J.B., and Gens R. 2014. Timing of retrogressive thaw slump initiation in the Noatak Basin, northwest Alaska, USA. J. Geophys. Res. Earth Surf. 119: 1106–1120.
Barnhart T.B. and Crosby B.T. 2013. Comparing two methods of surface change detection on an evolving thermokarst using high-temporal-frequency terrestrial laser scanning, Selawik River, Alaska. Remote Sens. 5: 2813–2837.
Brooker A., Fraser R.H., Olthof I., Kokelj S.V., and Lacelle D. 2014. Mapping the activity and evolution of retrogressive thaw slumps by tasselled cap trend analysis of a Landsat satellite image stack. Permafr. Periglac. Process. 25: 243–256.
Burn C.R. and Lewkowicz A.G. 1990. Canadian landform examples — 17 retrogressive thaw slumps. Can. Geogr. 34: 273–276.
Chandler J., Ashmore P., Paola C., Gooch M., and Varkaris F. 2002. Monitoring river-channel change using terrestrial oblique digital imagery and automated digital photogrammetry. Ann. Assoc. Am. Geogr. 92: 631–644.
Chin K.S., Lento J., Culp J.M., Lacelle D., and Kokelj S.V. 2016. Permafrost thaw and intense thermokarst activity decreases abundance of stream benthic macroinvertebrates. Glob. Chang. Biol. 22: 2715–2728.
EDF R&D. 2012. Cloud compare (version 2.4) [gpl software]. Technical report, Telecom ParisTech, Paris.
Eltner A., Kaiser A., Castillo C., Rock G., Neugirg F., and Abellán A. 2016. Image-based surface reconstruction in geomorphometry — merits, limits and developments. Earth Surf. Dyn. 4: 359–389.
Environment Canada. 2010. Canadian Climate Normals 1981–2010 Station Data. [Online]. Available from [23 Feb. 2016].
Fonstad M.A., Dietrich J.T., Courville B.C., Jensen J.L., and Carbonneau P.E. 2013. Topographic structure from motion: a new development in photogrammetric measurement. Earth Surf. Processes Landforms, 38: 421–430.
Fraser R., Kokelj S., Lantz T., McFarlane-Winchester M., Olthof I., and Lacelle D. 2018. Climate sensitivity of high Arctic permafrost terrain demonstrated by widespread ice-wedge thermokarst on Banks Island. Remote Sens. 10: 954.
Fraser R.H., Olthof I., Kokelj S.V, Lantz T.C., Lacelle D., Brooker A., Wolfe S., and Schwarz S. 2014. Detecting landscape changes in high latitude environments using Landsat trend analysis: 1. Visualization. Remote Sens. 6: 11533–11557.
Grom, J.D. 2008. Retrogressive thaw slump process and morphology, Eureka Sound Lowlands, Ellesmere Island, Nunavut, Canada. M.Sc. thesis, McGill University, Montreal, Que. 106 pp.
Hugenholtz C. 2012. Small unmanned aircraft systems for remote sensing and earth science research. Eos Trans. AGU, 93: 236–237.
Hugenholtz C.H., Whitehead K., Brown O.W., Barchyn T.E., Moorman B.J., LeClair A., Riddell K., and Hamilton T. 2013. Geomorphological mapping with a small unmanned aircraft system (sUAS): feature detection and accuracy assessment of a photogrammetrically-derived digital terrain model. Geomorphology, 194: 16–24.
Hugenholtz C., Brown O., Walker J., Barchyn T., Nesbit P., Kucharczyk M., and Myshak S. 2016. Spatial accuracy of UAV-derived orthoimagery and topography: comparing photogrammetric models processed with direct geo-referencing and ground control points. Geomatica, 70: 21–30.
James M.R. and Robson S. 2014. Sequential digital elevation models of active lava flows from ground-based stereo time-lapse imagery. ISPRS J. Photogramm. Remote Sens. 97: 160–170.
Kokelj S.V. and Jorgenson M.T. 2013. Advances in thermokarst research. Permafr. Periglac. Process. 24: 108–119.
Kokelj S.V., Lantz T.C., Kanigan J., Smith S.L., and Coutts R. 2009. Origin and polycyclic behaviour of tundra thaw slumps, Mackenzie Delta region, Northwest Territories, Canada. Permafr. Periglac. Process. 20: 173–184.
Kokelj S.V., Lacelle D., Lantz T.C., Tunnicliffe J., Malone L., Clark I.D., and Chin K.S. 2013. Thawing of massive ground ice in mega slumps drives increases in stream sediment and solute flux across a range of watershed scales. J. Geophys. Res. Earth Surf. 118: 681–692.
Kokelj S.V., Tunnicliffe J., Lacelle D., Lantz T.C., Chin K.S., and Fraser R. 2015. Increased precipitation drives mega slump development and destabilization of ice-rich permafrost terrain, northwestern Canada. Glob. Planet. Change, 129: 56–68.
Kokelj S.V., Lantz T.C., Tunnicliffe J., Segal R., and Lacelle D. 2017. Climate-driven thaw of permafrost preserved glacial landscapes, northwestern Canada. Geology, 45: 371–374.
Lacelle D., Bjornson J., Lauriol B., Clark I.D., and Troutet Y. 2004. Segregated-intrusive ice of subglacial meltwater origin in retrogressive thaw flow headwalls, Richardson Mountains, NWT, Canada. Quat. Sci. Rev. 23: 681–696.
Lacelle D., Bjornson J., and Lauriol B. 2010. Climatic and geomorphic factors affecting contemporary (1950–2004) activity of retrogressive thaw slumps on the Aklavik Plateau, Richardson Mountains, NWT, Canada. Permafr. Periglac. Process. 21: 1–15.
Lacelle D., Lauriol B., Zazula G.D., Ghaleb B., Utting N., and Clark I.D. 2013. Timing of advance and basal condition of the Laurentide Ice Sheet during the last glacial maximum in the Richardson Mountains, NWT. Quat. Res. 80: 274–283.
Lacelle D., Brooker A., Fraser R.H., and Kokelj S.V 2015. Distribution and growth of thaw slumps in the Richardson Mountains-Peel Plateau region, northwestern Canada. Geomorphology, 235: 40–51.
Lague D., Brodu N., and Leroux J. 2013. Accurate 3D comparison of complex topography with terrestrial laser scanner: application to the Rangitikei canyon (N-Z). ISPRS J. Photogramm. Remote Sens. 82: 10–26.
Lantuit H. and Pollard W.H. 2005. Temporal stereophotogrammetric analysis of retrogressive thaw slumps on Herschel Island, Yukon Territory. Nat. Hazards Earth Syst. Sci. 5: 413–423.
Lantuit H. and Pollard W.H. 2008. Fifty years of coastal erosion and retrogressive thaw slump activity on Herschel Island, southern Beaufort Sea, Yukon Territory, Canada. Geomorphology, 95: 84–102.
Lantuit H., Pollard W.H., Couture N., Fritz M., Schirrmeister L., Meyer H., and Hubberten H.W. 2012. Modern and late Holocene retrogressive thaw slump activity on the Yukon Coastal Plain and Herschel Island, Yukon Territory, Canada. Permafr. Periglac. Process. 23: 39–51.
Lewkowicz A.G. 1985. Use of an ablatometer to measure short-term ablation of exposed ground ice. Can. J. Earth Sci. 22: 1767–1773.
Lewkowicz A.G. 1987. Headwall retreat of ground-ice slumps, Banks Island, Northwest Territories. Can. J. Earth Sci. 24: 1077–1085.
Liljedahl A.K., Boike J., Daanen R.P., Fedorov A.N., Frost G.V., Grosse G., Hinzman L.D., Iijma Y., Jorgenson J.C., Matveyeva N., Necsoiu M., Raynolds M.K., Romanovsky V.E., Schulla J., Tape K.D., Walker D.A., Wilson C.J., Yabuki H., and Zona D. 2016. Pan-Arctic ice-wedge degradation in warming permafrost and its influence on tundra hydrology. Nat. Geosci. 9: 312–318.
Lowe D.G. 2004. Distinctive image features from scale-invariant keypoints. Int. J. Comput. Vis. 60: 91–110.
Mallalieu J., Carrivick J.L., Quincey D.J., Smith M.W., and James W.H.M. 2017. An integrated Structure-from-Motion and time-lapse technique for quantifying ice-margin dynamics. J. Glaciol. 63: 937–949.
Malone L., Lacelle D., Kokelj S.V, and Clark I.D. 2013. Impacts of hillslope thaw slumps on the geochemistry of permafrost catchments (Stony Creek watershed, NWT, Canada). Chem. Geol. 356: 38–49.
Micheletti N., Chandler J.H., and Lane S.N. 2015. Investigating the geomorphological potential of freely available and accessible structure-from-motion photogrammetry using a smartphone. Earth Surf. Process. Landforms, 40: 473–486.
Nolan M. and DesLauriers K. 2016. Which are the highest peaks in the US Arctic? Fodar settles the debate. The Cryosphere, 10: 1245–1257.
Nolan M., Larsen C., and Sturm M. 2015. Mapping snow depth from manned aircraft on landscape scales at centimeter resolution using structure-from-motion photogrammetry. Cryosphere, 9: 1445–1463.
Obu J., Lantuit H., Fritz M., Pollard W.H., Sachs T., and Günther F. 2016. Relation between planimetric and volumetric measurements of permafrost coast erosion: a case study from Herschel Island, western Canadian Arctic. Polar Res. 35: 30313.
Obu J., Lantuit H., Grosse G., Günther F., Sachs T., Helm V., and Fritz M. 2017. Coastal erosion and mass wasting along the Canadian Beaufort Sea based on annual airborne LiDAR elevation data. Geomorphology, 293: 331–346.
O’Neill H.B., Burn C.R., Kokelj S.V., and Lantz T.C. 2015. “Warm” tundra: atmospheric and near-surface ground temperature inversions across an alpine treeline in continuous permafrost, Western Arctic, Canada. Permafr. Periglac. Process. 26: 103–118.
Rudy A.C.A., Lamoureux S.F., Kokelj S.V., Smith I.R., and England J.H. 2017. Accelerating thermokarst transforms ice-cored terrain triggering a downstream cascade to the ocean. Geophys. Res. Lett. 44: 11,080–11,087.
Segal R.A., Lantz T.C., and Kokelj S.V 2016. Acceleration of thaw slump activity in glaciated landscapes of the Western Canadian Arctic. Environ. Res. Lett. 11: 34025.
Short N., Brisco B., Couture N., Pollard W., Murnaghan K., and Budkewitsch P. 2011. A comparison of TerraSAR-X, RADARSAT-2 and ALOS-PALSAR interferometry for monitoring permafrost environments, case study from Herschel Island, Canada. Remote Sens. Environ. 115: 3491–3506.
Stettner S., Beamish A.L., Bartsch A., Heim B., Grosse G., Roth A., and Lantuit H. 2018. Monitoring inter- and intra-seasonal dynamics of rapidly degrading ice-rich permafrost riverbanks in the Lena Delta with TerraSAR-X time series. Remote Sens. 10: 51.
Stöcker C., Eltner A., and Karrasch P. 2015. Measuring gullies by synergetic application of UAV and close range photogrammetry — a case study from Andalusia, Spain. Catena, 132: 1–11.
Swanson, D. 2012. Mapping of erosion features related to thaw of permafrost in the Noatak National Preserve. Alaska Natural Resource Data Series NPS/ARCN/NRDS–2012/248. National Park Service, Fort Collins, CO, USA.
Swanson D. and Nolan M. 2018. Growth of retrogressive thaw slumps in the Noatak Valley, Alaska, 2010–2016, measured by airborne photogrammetry. Remote Sens. 10: 983
Tanski G., Lantuit H., Ruttor S., Knoblauch C., Radosavljevic B., Strauss J., Wolter J., Irrgang A.M., Ramage J., and Fritz M. 2017. Transformation of terrestrial organic matter along thermokarst-affected permafrost coasts in the Arctic. Sci. Total Environ. 581–582: 434–447.
Tonkin T.N. and Midgley N.G. 2016. Ground-control networks for image based surface reconstruction: an investigation of optimum survey designs using UAV derived imagery and structure-from-motion photogrammetry. Remote Sens. 8: 786.
Turner D., Lucieer A., and de Jong S.M. 2015. Time series analysis of landslide dynamics using an Unmanned Aerial Vehicle (UAV). Remote Sens. 7: 1736–1757.
van der Sluijs, J., Kokelj, S.V., Fraser, R.H., Tunnicliffe, J.F., and Lacelle, D. 2018. Permafrost dynamics and infrastructure impacts revealed by UAV photogrammetry and thermal imaging. Remote Sens. In press.
Welty E.Z., Bartholomaus T.C., O’Neel S., and Pfeffer W.T. 2013. Cameras as clocks. J. Glaciol. 59: 275–286.
Westoby M.J., Brasington J., Glasser N.F., Hambrey M.J., and Reynolds J.M. 2012. ‘Structure-from-Motion’ photogrammetry: a low-cost, effective tool for geoscience applications. Geomorphology, 179: 300–314.
Zwieback S., Kokelj S., Günther F., Boike J., Grosse G., and Hajnsek I. 2018. Sub-seasonal thaw slump mass wasting is not consistently energy limited at the landscape scale. Cryosphere, 12: 549–564.

Information & Authors


Published In

cover image Arctic Science
Arctic Science
Volume 4Number 4December 2018
Pages: 827 - 845


Received: 7 June 2018
Accepted: 28 August 2018
Accepted manuscript online: 6 September 2018
Version of record online: 6 September 2018

Key Words

  1. thermokarst
  2. permafrost
  3. remote sensing
  4. Arctic


  1. thermokarst
  2. pergélisol
  3. télédétection
  4. Arctique



Lindsay Armstrong
Department of Geography, Environment and Geomatics, University of Ottawa, Ottawa, ON K1N 6N5, Canada.
Denis Lacelle [email protected]
Department of Geography, Environment and Geomatics, University of Ottawa, Ottawa, ON K1N 6N5, Canada.
Robert H. Fraser
Canada Centre for Mapping and Earth Observation, Natural Resources Canada, Ottawa, ON K1S 5H4, Canada.
Steve Kokelj
Northwest Territories Geological Survey, Government of Northwest Territories, Yellowknife, NWT X1A 1K3, Canada.
Anders Knudby
Department of Geography, Environment and Geomatics, University of Ottawa, Ottawa, ON K1N 6N5, Canada.


This article is open access. This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Competing Interests

The authors have no conflicts of interest to report.

Metrics & Citations


Other Metrics


Cite As

Export Citations

If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.

Cited by

1. Three-dimensional subsurface architecture and its influence on the spatiotemporal development of a retrogressive thaw slump in the Richardson Mountains, Northwest Territories, Canada
2. Evaluation of low-cost Raspberry Pi sensors for structure-from-motion reconstructions of glacier calving fronts
3. The Potential of UAV Imagery for the Detection of Rapid Permafrost Degradation: Assessing the Impacts on Critical Arctic Infrastructure
4. Rapid seafloor changes associated with the degradation of Arctic submarine permafrost
5. Landform mapping, elevation modelling, and thaw subsidence estimation for permafrost terrain using a consumer-grade remotely-piloted aircraft1
6. CubeSat Technology and Periglacial Landscape Analysis
7. Thermokarst
8. Sub-seasonal variability of supraglacial ice cliff melt rates and associated processes from time-lapse photogrammetry
9. Automatically quantifying evolution of retrogressive thaw slumps in Beiluhe (Tibetan Plateau) from multi-temporal CubeSat images
10. Automated Registration of SfM‐MVS Multitemporal Datasets Using Terrestrial and Oblique Aerial Images
11. Detailed Characterization and Monitoring of a Retrogressive Thaw Slump from Remotely Piloted Aircraft Systems and Identifying Associated Influence on Carbon and Nitrogen Export
12. Debris cover on thaw slumps and its insulative role in a warming climate
13. Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images
14. Assessment of the Ice Wedge Polygon Current State by Means of UAV Imagery Analysis (Samoylov Island, the Lena Delta)
15. Permafrost Terrain Dynamics and Infrastructure Impacts Revealed by UAV Photogrammetry and Thermal Imaging

View Options

View options


View PDF

Get Access





Share Options


Share the article link

Share on social media

Cookies Notification

We use cookies to improve your website experience. To learn about our use of cookies and how you can manage your cookie settings, please see our Cookie Policy.