A comparison of TerraSAR-X, RADARSAT-2 and ALOS-PALSAR interferometry for monitoring permafrost environments, case study from Herschel Island, Canada

Remote Sensing of Environment 115 (2011) 3491–3506

Contents lists available at SciVerse ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

A comparison of TerraSAR-X, RADARSAT-2 and ALOS-PALSAR interferometry formonitoring permafrost environments, case study from Herschel Island, Canada

Naomi Short a,⁎, Brian Brisco a, Nicole Couture b, Wayne Pollard c, Kevin Murnaghan a, Paul Budkewitsch a

a Canada Centre for Remote Sensing, 588 Booth St., Ottawa, ON, Canada K1A 0Y7b Geological Survey of Canada, 601 Booth St., Ottawa, ON, Canada K1A 0E8c Department of Geography, McGill University, 805 Sherbrooke St. W., Montreal, QC, Canada H3A 2K6

⁎ Corresponding author. Tel.: +1 613 947 1264; fax:E-mail address: [email protected] (N. Short).

0034-4257/$ – see front matter. Crown Copyright © 20doi:10.1016/j.rse.2011.08.012

a b s t r a c t

Article history:Received 9 March 2011Received in revised form 18 August 2011Accepted 21 August 2011Available online 15 September 2011

Keywords:SARInterferometryPermafrostHerschel IslandTerrain stabilityArctic

Interferometric synthetic aperture radar (InSAR) data sets from TerraSAR-X, RADARSAT-2 and ALOS-PALSARare compared for their ability to detect ground movement over the continuous permafrost site of HerschelIsland, Yukon Territory, Canada. All three sensors maintain good coherence within a summer season andcan be used to create summer displacement products. Stacking is advantageous for the TerraSAR-X andRADARSAT-2 data sets, although mottling, possibly an interaction of the SAR with vegetation, or residual tro-pospheric noise, is visible, reducing the reliability of the results. RADARSAT-2 and ALOS-PALSAR provide themost promising results with the ability to form one year interval interferograms. PALSAR can also form twoand three year interval interferograms. Long interval data sets spanning 2007 to 2010 identify a band ofmovement of 20 to 30 cm/year along the north-east coast, and a region of movement of up to 5 cm/yearnear the northern tip of the island. The ability to form long interval displacement products holds the mostpromise for permafrost monitoring, since long-term trends are of greater interest for permafrost stabilitythan short-term seasonal changes. TerraSAR-X data have the disadvantage that year to year interferogramscannot be formed. InSAR is not the ideal monitoring technique for the large thaw slumps of Herschel Island.Although general areas of instability can be identified, specific slump detection is limited by radar look direc-tion, and the large and abrupt slump movement, often accompanied by disintegration and collapse of slumpsections, causes loss of coherence in the InSAR data. Thaw slumps may require a different interferometric ap-proach, such as slump extent mapping from coherence loss, or the installation of corner reflectors and pointtarget techniques. The frequent revisit and high spatial resolution of TerraSAR-X provide the best chance ofmaintaining coherence over thaw slumps. In general, InSAR is more successful at identifying broad areas ofsubtle subsidence in gentle relief, areas of terrain instability, possibly due to permafrost thaw or ground icemelt and the removal of water volume, and prior to significant slumping.

Crown Copyright © 2011 Published by Elsevier Inc. All rights reserved.

1. Introduction

Permafrost is ground that remains frozen for two or more consec-utive years. Permafrost is overlain by a layer that thaws and refreezesseasonally, called the active layer. The presence of permafrost and anactive layer presents particular challenges for infrastructure and engi-neering, and for monitoring environmental and ecological stability.While permafrost is a subsurface thermal phenomenon, changes inthe state of permafrost are eventually manifested at the ground sur-face, to various degrees depending on the ice content of the perma-frost and the surficial geology. Permafrost aggradation (growth)causes ground heave and surface uplift, while permafrost thaw causesground subsidence. Persistent thaw can lead to the formation of

+1 613 947 1385.

11 Published by Elsevier Inc. All righ

hummocky terrain called thermokarst, and landslides and slumpson significant slopes.

Synthetic aperture radar interferometry (InSAR) is now a wellestablished technique for measuring ground movement (Gabrielet al., 1989; Massonnet & Feigl, 1995, 1998). Since changes in perma-frost often result in movement at the ground surface, it is a logicaltechnique to exploit. However, to date monitoring success has beenmodest. Until recently, the C-band satellites in orbit, namely Envisatand RADARSAT-1 had trouble maintaining coherence over the north-ern environments with vegetative cover, albeit sparse. ERS-1 and -2had more success, particularly in the tandem mode with short repeatperiods (Li et al., 2003; Wang & Li, 1999a,b). Experiments over ToolikLake in Alaska using ERS data demonstrated that distinct periods ofthaw could be identified within a summer, although snow accumula-tion and melt, and freeze and thaw of the ground pose particularproblems for maintaining coherence at C-band year round (Wang &Li, 1999a). Additional work demonstrated that spatial variability inthe amount of heave could be detected with InSAR, and this was

ts reserved.

3492 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

related to the depth of the active layer and soil moisture content(Wang & Li, 1999b). Further work at Toolik Lake showed that settle-ment and heave were generally greatest on exposed hilltops andalong river channels, and less on slopes and in low areas, where athick organic layer insulates the ground. While these spatial patternsof variation could be detected, they were not successful in extractingquantitative estimates of heave and settlement (Li et al., 2003). Morerecently, Liu et al. (2010) used 1992–1999 ERS data to measure sea-sonal and long-term rates of deformation over the North Slope ofAlaska. Using an area of relatively stable ground as calibration (flood-plain deposits with coarse gravel thought to be insensitive to frostheave or thaw settlement), they measured seasonally varying dis-placements of 1–4 cm during the thawing season, and 1–4 cm of cu-mulative settlement over the decade.

L-band InSAR, with its longer wavelength and greater penetrationof vegetative cover, has proven to be much better at maintaining co-herence over permafrost environments (Rykhus & Lu, 2008; Short etal., 2009). Using the JERS-1 satellite, patterns presumed to be thawsettlement were detected over the Alaskan Arctic Coastal Plain frommid-June to the end of September (Rykhus & Lu, 2008). Exploratorywork with ALOS-PALSAR showed that coherence was very good inthe northern Yukon, Canada, but that an accurate DEM was criticalfor the extraction of reliable measurements from PALSAR pairs, par-ticularly because PALSAR baselines can be very large (Short et al.,2009).

Work comparing the performance of different wavelength SARsfor interferometry has generally been carried out in the mid andlow latitude regions and has utilised both airborne and spaceborneSARs. Some general conclusions have emerged. Early work over Ha-waii using airborne X-, C- and L-band SARs demonstrated that varia-tions in the wet troposphere (atmospheric water vapour) causedartefacts in all interferograms, but that these are proportionately larg-er for the X-band short wavelength data, therefore in non-desert re-gions, the use of the longest wavelength available is recommended(Zebker et al., 1997). The superior coherence of L-band interfero-grams over vegetated areas was also identified in the early Hawaiiwork (Rosen et al., 1996). Other work has demonstrated that thewavelengths perform equally well over desert and arid regions butthat only L-band data have the potential for maintaining coherenceover forested areas (Werner et al., 1997) and rural areas (Takeuchi& Yamada, 2002). Over temperate regions L-band offers the possibil-ity of long time interval interferograms (Werner et al., 1997), poten-tially five years or more (Takeuchi & Yamada, 2002). C-band can alsodeliver long interval interferograms but these are limited to a maxi-mum of four years (Yonezawa & Takeuchi, 2000).

The work here investigates the performance of three different sen-sors for InSAR specifically over a northern permafrost environment.We evaluate three sensors launched relatively recently: TerraSAR-X(X-band SAR with 3.1 cm wavelength), RADARSAT-2 (C-band SARwith 5.6 cm wavelength) and ALOS-PALSAR (L-band SAR with23.6 cm wavelength). Collectively, these new sensors offer finer spa-tial resolution, better baseline control, shorter revisit periods, and avariety of SAR wavelengths not previously readily available. Usingthe study site of Herschel Island, which lies in the continuous perma-frost zone of northern Canada, the performance, strengths and poten-tial contribution of InSAR with each sensor for permafrost monitoringis evaluated.

2. Study site

Herschel Island (69°35′N, 139°06′W) is a small island off the north-ern Yukon coast in the southern Beaufort Sea. It is approximately108 km2, (15 by 8 km) and is characterised by rolling topography anda maximum elevation of 183 m a.s.l. (Mackay, 1959). The island ispart of an ice-pushed structure formed by a lobe of the Laurentide IceSheet during the Wisconsin Glaciation (Mackay, 1959). The island is

composed mainly of fine-grained marine sediments dredged fromHer-schel Basin in the Beaufort Sea, with coarser grained coastal deposits(Lantuit & Pollard, 2008). The seasonally thawed active layer penetrates45 to 90 cm into the silty diamicton that blanketsmost of the island, andup to 110 cm in the coarser coastal sediments of the sand spits and bea-ches (Smith et al., 1989). The island is characterised by continuous per-mafrost with widespread ground ice. This ground ice occurs as icewedges, ice lenses, pore ice, buried snowbank/glacier ice and somemassive tabular ice bodies (Pollard, 1990; Pollard & Dallimore, 1990).The fine grained sediments and the widespread presence of groundice make the island particularly susceptible to permafrost degradationand therefore an ideal study site. The island surface is characterised bylow growing, Arctic tundra vegetation. This is also helpful for theInSAR analysis as the complications of dense vegetation are reduced.The relatively gentle topography of the island precludes the complica-tions of layover, foreshortening and shadow in the InSAR analysis.

Mean monthly air temperatures on Herschel Island range from−26 °C in February to +9 °C in July, with a 1979–2007 annual meanof −9.5 °C (Burn & Zhang, 2009). The mean annual air temperaturehas warmed 2.5 °C since 1905, and this warming has resulted in a risein the temperature of the permafrost near the ground surface (Burn &Zhang, 2009). Annual measurements of the active layer made between2003 and 2007 showed that the active layer had increased in thicknessby 15 to 25 cm compared to similar measurements made in 1985 bySmith et al. (1989) (Burn & Zhang, 2009).

Patterns of coastal erosion and thaw slumping on the island havebeen studied by Lantuit and Pollard (2008). Their study shows thehighest rates of coastal retreat are on the north-west facing sections,since these are primarily exposed to high energy wave action. Com-parable high rates of erosion are also documented along the north-east coast, where slopes and coastal erosion cause instability and ac-tive slope processes. The south and south-east facing shorelines haveseen increasing numbers and areal extents of retrogressive thawslumps.

Herschel Island has been identified as at risk from rising sea levels,eroding coastline and melting permafrost by both UNESCO WorldHeritage and the World Monuments Fund. The island has no perma-nent residents and is managed as a territorial park. Some historic in-frastructure from whaling activities at the end of the 19th centuryremains on the spit at Pauline Cove. Fig. 1 shows the location andthe main features of the island. Fig. 2 is a photograph of CollinsonHead, the eastern point of the island, showing the eroding coastlineand the thin vegetation.

3. Data

3.1. SAR data

Synthetic aperture radar (SAR) data sets are from TerraSAR-X,RADARSAT-2 and ALOS-PALSAR. Table 1 lists the relevant characteris-tics of the three sensors. All SAR data sets were acquired from the ap-propriate distribution agencies in Single Look Complex (SLC) format.Since permafrost terrain is frozen and relatively inactive in the wintermonths (Alasset et al., 2008), deep snow cover can affect phase valuesin the late winter and early spring (Short et al., 2008), and snowmeltcauses loss of coherence during the spring (Wang & Li, 1999a), inter-ferometric data sequences were planned and acquired only for Maythrough to November. For TerraSAR-X and RADARSAT-2 acquisitionswere planned and ordered. For ALOS-PALSAR, the pre-planned Sys-tematic Observation Strategy meant that data acquisitions could notbe specifically planned. Observation requests were submitted butrarely acquired. July, August, September and October data sets wereobtained from the PALSAR archive.

Tables 2 through 5 list the characteristics of the data sets, including ameasure of InSAR quality. Successful interferometry depends on theground surface characteristics maintaining similarity between SAR

Fig. 1. Map and location of Herschel Island, from National Topographic Mapsheet 117D.

3493N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

acquisitions, the statistical measure of which is called coherence. If thereis significant change in the surface characteristics between data acquisi-tions (as opposed to just displacement), such as vegetation change,snowfall or melt, or ground freeze or thaw, surface features will notmatch and it can be difficult or impossible to co-register images success-fully. And, even if registration is successful, coherence may be low andthe phase will be too noisy for successful phase unwrapping. Therefore,each data set is given an InSAR quality measure: ✗ means unworkable,either failed registration or unstable phase due to change betweenimage acquisitions; ~ means registration and formation of the interfero-gram was successful, but that noise rendered unwrapping unreliable sothe results are questionable;✓means the data setwas used successfully.

3.2. Elevation data

A high resolution digital elevation model (DEM) was acquired forHerschel Island. The DEM is a PhotoSat product, created using a pro-prietary stereo image matching process with IKONOS satellite imagesfrom September 18, 2004. Although vertical accuracy is not specifical-ly given for the Herschel product, the image matching process hasbeen proven to generate DEMs from IKONOS imagery with verticalaccuracy on the order of 50 cm (Mitchell & MacNabb, 2010). The Her-schel DEM has a pixel spacing of 2 m. Fig. 3 shows the distribution ofelevation across the island.

Fig. 2. Photograph of Collinson Head, Herschel Island, showing the rolling topography,low growing vegetation and eroding coastline.

3.3. Soil moisture data

Soil moisture is a factor that has been shown to affect the penetra-tion depth of a radar wave into the ground and, therefore, to affectInSAR measurements (Nolan et al., 2003; Nolan & Fatland, 2003a,b).While it was not logistically possible to acquire soil moisture mea-surements for all the periods covered by image acquisitions, somesoil moisture measurements were made by a McGill University fieldparty on August 2, 2010. Using a Delta-T Devices Moisture Meterwith a 6 cm ThetaProbe, % volumetric soil moisture content was mea-sured at six sites across the island. Since soil moisture is known tovary significantly over a small area, four measurements were madeat 6 cm depth, over a 1×1 m representative area at each site and av-eraged. The six sites are marked on Fig. 3 and the measured valuesgiven in Table 6.

3.4. Slump headwall data

The headwalls of thaw slumps in the Thetis Bay area are surveyedannually by McGill University using a Trimble 4700 diffferential Glob-al Positioning System (GPS) unit. The outlines of the informallynamed Hawk, Ice Wedge and ABC slumps as mapped in 2009 areshown in Fig. 3.

4. Methodology

The InSAR processing was carried out using the GAMMA InSARprocessing software (Werner et al., 2000). The processing steps areoutlined in Fig. 4. Most processing steps followconventional InSARpro-cessing practise, apart from the two step co-registration of the SLCs. Theresampling of slave SLCs to the geometry of themaster using the DEM is

Table 1SAR sensors and characteristics.

Mission Launch–terminationdates

SARwavelength

SARfrequency

Repeatperiod

(cm) (GHz) (days)

TerraSAR-X June 2007- 3.1 9.65 11RADARSAT-2 December 2007- 5.6 5.405 24ALOS-PALSAR January 2006–April 2011 23.6 1.270 46

Table 2TerraSAR-X data sequence - descending path (right looking, Stripmap mode 007, orbit 24, satellite heading 195.88°, acquisition time 16:07 UTC or 09:07 local, HH polarisation, mid-incidence angle 31.14°, approximate resolution 3 m). September 2009 scenes missing due to satellite anomaly.

Data pairs Perpendicular baseline (m) Days separation InSAR quality

20090509–20090520 −19 11 ✗

20090520–20090531 −32 11 ✗

20090531–20090611 86 11 ✗

20090611–20090622 −22 11 ✓

20090622–20090703 −62 11 ✓

20090703–20090725 −123 22 ✓

20090725–20090805 234 11 ✓

20090805–20090816 −194 11 ✓

20090816–20090827 −53 11 ✓

20091010–20091021 81 11 ✗

20091021–20091101 118 11 ✗

20091101–20091123 −54 22 ✗

20090703–20100712 69 374 ✗

20090725–20100701 124 341 ✗

20090725–20100723 43 363 ✗

20090725–20100803 −60 374 ✗

20090805–20100803 −293 363 ✗

20090816–20100825 164 374 ✗

20090827–20100825 217 363 ✗

20090827–20100905 31 374 ✗

Table 3TerraSAR-X data sequence — ascending path (right looking, Stripmap mode 004, orbit 152, satellite heading 342.26°, acquisition time 02:17 UTC or 19:17 local, HH polarisation,mid-incidence angle 24.04°, approximate resolution 3 m).

Data pairs Perpendicular baseline (m) Days separation InSAR quality

20100618–20100629 74 11 ✓

20100629–20100710 −62 11 ✓

20100710–20100721 −117 11 ✓

20100721–20100801 −23 11 ✓

20100801–20100812 226 11 ✓

20100812–20100823 −175 11 ✓

20100823–20100903 −52 11 ✓

20100903–20100914 −134 11 ✓

20100914–20101006 214 22 ✗

Table 4RADARSAT-2 data sequence — ascending path (right looking, Beam Mode Ultra-Fine 10, relative orbit 43, satellite heading 346.61°, acquisition time 02:39 UTC or 19:39 local. HHpolarisation, mid-incidence angle 37.83°, approximate resolution 3 m).

Data pairs Perpendicular baseline (m) Days separation InSAR quality

20090614–20090708 9 24 ✓

20090708–20090801 −397 24 ✓

20090801–20090825 −61 24 ✓

20090825–20090918 52 24 ✓

20090918–20091012 −157 24 ✗

20090708–20100703 −10 360 ~20090801–20100727 251 360 ✓

20090801–20100820 55 384 ✗

20090825–20100820 117 360 ~20090825–20100913 103 384 ✗

3494 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

relatively newbut has been shown to improve coherence and results forscenes with large baselines and/or significant relief (pers. comm.Charles Werner, 2010). Although the relief of Herschel Island is only183 m, some of the PALSAR baselines are 1000 s of metres. Initial

Table 5ALOS-PALSAR data pairs— ascending path (right looking, Fine Beam Dual Mode, HH polarisalocal, mid-incidence angle 38.77°, approximate resolution 20 m).

Data pairs Path and frame Perpendicular baseline (m) Days separ

20070819–20071004 235/1390 529 4620070831–20080902 233/1390 −5230 36720070831–20100908 233/1390 1769 110420080902–20091021 233/1390 6961 41420080902–20100724 233/1390 6653 69020080902–20100908 233/1390 7000 73620081001–20091004 232/1390 3483 36820100724–20100908 233/1390 346 46

experiments proved that this step did improve the results and thisfirst stepwas therefore routinely implemented. The second step is to re-fine the co-registration of the initially resampled SLCs using a grid of in-tensity samples. These samples measure the range and azimuth offsets

tion used for processing, satellite heading 347.20°, acquisition time 07:10 UTC or 00:10

ation InSAR quality Ionospheric effects

✓ No✓ Some over mainland, no significant variation over island✓ Yes, small variation across the island✗ Not done~ Noisy — inconclusive~ Noisy — inconclusive✗ Not done✓ No

Fig. 3. High resolution digital elevation model of Herschel Island. Soil moisture sampling sites are marked SM1-SM6. The 2009 outlines of the Hawk, Ice Wedge and ABC slump fea-tures monitored annually by McGill University are shown in white. The short red line at the eastern tip of the island marks the profile of permafrost sites studied by Burn and Zhang(2009). The black oval marks the control area used for assessment of displacement errors.

3495N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

between the scenes and allow a polynomial model to be derived. Sincinterpolation is then used to resample the slave to themaster accordingto the derived polynomial model.

The interferogramswere generated using range spectrum filtering, toaccount for the spectral shift induced by the difference in incidence an-gles between the master and slave scenes due to the baseline (Gatelli etal., 1994). Only the range spectrum interval common to the two sceneswas retained. Common band filtering was also applied in azimuth. Thehigh resolution DEM was used to remove the phase due to topographyand thus generate the differential interferograms. Adaptive filteringwas applied to smooth the differential phase (Goldstein & Werner,1998). Phase unwrapping of the differential interferograms was doneusing the Minimum Cost Flow phase unwrapping algorithm (Costantini,1998; Werner et al., 2002).

Residual phase trends resulting from inaccurate orbit data and in-correct baseline estimates were corrected for using the baseline re-finement method of Rosen et al. (1996) and ground control points(GCPs). This method compares measured phase values at GCP loca-tions with phase values simulated from the DEM, and identifieslong-scale statistical trends. Since areas of significant deformationwould produce large phase variations in addition to baseline trends,a mask was used to prevent GCPs from being located in the knowndynamic area along the north coast. Identified long-scale phasetrends were used to refine the baseline estimate and the differential

Table 6Percentage volumetric water content measured at six locations across Herschel Island on A

% Volumetric water content

Site Coordinates 1 2 3

SM1 69°33′56.6″N 38.3 30.5 30.1139°04′25.9″W

SM2 69°33′56.8″N 32.9 35.7 36.9139°03′10.1″W

SM3 69°37′33.9″N 39.4 59.8 51.4139°01′48.2″W

SM4 69°35′23.7″N 47.0 43.7 40.3139°07′51.4″W

SM5 69°35′03.5″N 35.9 34.3 29.4139°11′28.7″W

SM6 69°34′41.9″N 39.1 42.2 71.3138°51′51.6″W

interferograms were recreated using the improved baseline estimateswith residual phase trends removed.

Final unwrapped phase values were converted to line-of-sight dis-placement in slant range metres. Displacement is presented as move-ment towards (positive) or away from (negative) the satellite. Ingenerally flat areas, the ground movement is probably entirely verti-cal, such that positive movement would indicate uplift and negativewould be subsidence. On slopes movement is likely a combinationof horizontal and vertical. Line-of-sight displacement is used for pre-sentation to avoid introducing errors by assuming a direction for themovement. To facilitate quantitative comparisons between data sets(since slant range distance is a function of scene incidence angle),the TerraSAR-X displacements were corrected to an incidence angleof 38°, to better match those of the RADARSAT-2 and PALSAR datasets, using Eq. (1).

dnorm ¼ d�cos θ1–θ2ð Þ ð1Þ

where:

dnorm is the corrected line-of-sight displacement (m)

d is the original line-of-sight displacement (m)

θ1 is the incidence angle of the primary data set

θ2 is the incidence angle of the data set to be corrected

ugust 2, 2010.

4 Average Description

30.4 32.3 Hummocky tundrawith mud boils

49.7 38.8 Hummocky tundra with mudboils and small tussocks

55.3 51.5 Hill top, wet tussocks

78.4 52.3 Highest point on island,wet tussocks

49.7 37.3 Valley side, several ponds

36.5 47.3 Tussock tundra with mud boils, deepice wedge troughs and small ponds

Fig. 4. InSAR data processing sequence.

3496 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

While this is a very simple correction, and assumes that the move-ment is in the RADARSAT-2 line-of-sight direction (as opposed topurely vertical or horizontal), it appeared to improve the quantitativecomparison between the different frequencies and geometries.

Final displacement and image products are geocoded using the regis-tration LookUp Table created during the initial registration of the DEM tothe master SAR scene. This Look Up Table is used to reproject the SARimage and any derived products from the slant range domain into the co-ordinate system of the DEM. In our case, the Universal TransverseMerca-tor projection of the DEM is used for the final products.

Since the influence of the ionosphere is known to be more signifi-cant at L-band, testing for ionospheric effects was done using the speck-le tracking technique and examining the patterns of azimuth pixelshifts, looking specifically for linear features or waves across the scene

(Gray et al., 2000). These data set observations are included in Table 5.Fortunately, ionospheric streaks or waves did not affect the one yearand shorter PALSAR interferometric pairs (some streaks were observedover themainland in the 20070831–20080902 data set, but Herschel Is-landwas not affected). The two year PALSAR pairwas noisy and the ion-ospheric test inconclusive, and the three year pair did show some subtlevariation across the island. These factors add to the uncertainties in theinterpretation of these particular PALSAR data sets.

4.1. Stacking

Stacking is a technique that has evolved to extract small deforma-tion signals out of multiple repeat InSAR data sets. By averaging manyinterferograms over the same area, random noise such as

3497N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

atmospheric signals can be subdued and small, cumulative trends canbe extracted. Because ground movement in permafrost environmentsis cyclical on an annual basis, we have applied stacking only to thesummers of 2009 and 2010, and for data availability reasons, only tothe TerraSAR-X and RADARSAT-2 data sets. All possible interfero-grams during the months of June, July, August and September wereformed and stacked. A small amount of interpolation (4×4) wasused to fill holes in the unwrapped interferograms in order to createa more complete picture. The stacking process takes the unwrappedphase results, still in the slant range coordinate system of the masterscene, and determines a rate of deformation for each pixel over thesummer months. This approach assumes that the displacement isconstant and linear through one summer, which unfortunately, maynot be true, since thaw could well occur in distinct periods (Wang &Li, 1999a), nevertheless, the technique is a simple method that pro-vides a good first estimate of cumulative summer displacement.

5. Results

5.1. Seasonal effects

For TerraSAR-X and RADARSAT-2 snowmelt and freeze up pro-cesses produced large changes in the May, October and Novemberscenes, making it impossible to co-register the scenes accuratelyenough for interferometry. These observations are captured in theInSAR quality measure given in Tables 2 through 4. PALSAR imageregistration was successful for July, August, September and someearly October scenes, but our late October scene produced poorimage registration statistics and poor coherence so was droppedfrom the analysis. In general therefore, June, July, August and Septem-ber data sets are considered useful for interferometry for permafrost

Fig. 5. Line-of-sight displacement results from TerraSAR-X descending sequence through sBlack areas are no displacement data, due to loss of coherence.

applications in this Arctic region. L-band data may also be usefulinto early October but this is less reliable.

5.2. TerraSAR-X

The TerraSAR-X results are very variable. Fig. 5 is an example of asummer sequence of displacement results over this area, and the var-iation is quite obvious. While the coherence is generally high, groundmovement trends are not immediately obvious or consistent. The dif-ferential phase patterns that are present are not related to topogra-phy or geomorphology, this suggests that the majority of the signalis either tropospheric variation, or an interaction of the radar withvegetation, rather than ground movement. The influence of the atmo-sphere on repeat track interferometry data sets is well established(Goldstein, 1995; Zebker et al., 1997), and the phase delay due tothe wet component of the troposphere is more obvious in the shortwavelength X-band results when the display scale is only +/−1.5 cm.

Since individual TerraSAR-X interferograms were not obviouslyuseful for this area, stacking was pursued. The upper panel of Fig. 6shows the results of stacking the June, July and August 2009 datasets. The descending path and consequent look direction identifiessome movement over the upper portion of Bell Bluff (on the north-east coast), on the order of−2 to−4 cm over the summer, and peak-ing at −8 cm in one small pocket. There is a very small area of move-ment towards the satellite at Collinson Head (the narrow strip of darkblue on the island's eastern tip), up to 4 cm, but no major patterns ofmovement along the north-east coast are identified. Movement alongthe north-west coast is more clearly identified, up to a maximum of−11 cm. And some movement (b −5 cm) is noted in the vicinity ofthe lake in the middle of the island, although, this is likely a spatial in-terpolation artefact around this small landscape feature.

ummer 2009. Displacement is in colour, with brightness a function of radar intensity.

3498 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

In an effort to better identify the area of known movement along thenorth-east coast, an ascending data set was acquired in 2010. Due to thesame variability in the individual interferograms, stacking was also pur-sued for this data set. The results are shown in the lower panel of Fig. 6.The opposing look direction brings out the major patterns of groundmovement along the north-east coast more clearly, (−5 to −15 cm).But, there is mottling over the central portion of the island suggestingmovement of up to −5 cm which is hard to explain. The patterns onlyvaguely align with local topography thus we cannot be certain that theyare ground displacement; they could be an interaction of the X-bandSAR with the vegetation, or residual tropospheric noise.

Wewere unfortunately not able to form any one year interferogramswith TerraSAR-X data. Image feature changes were so significant be-tween years that image co-registration could not be reliably performed.

5.3. RADARSAT-2

The RADARSAT-2 C-band results appear more stable, as shown inthe Fig. 7 sequence of summer 2009 results. While there is some subtlevariability in the displacements, regions of known movement are con-sistently identified. These RADARSAT-2 results also lend themselves

Fig. 6. Stacking results for TerraSAR-X, summer 2009, descending sequence (upper panel) aused in each stack are listed in the lower right corner of each panel.

well to stacking and the upper panel of Fig. 8 shows the stacked resultsfrom summer 2009. The displacement along the north-east coast, andthe spatial variability within that region is well detected with an as-cending pass and at C-band. Movement over the summer of 2009 isgenerally −5 to −8 cm along Bell Bluff, with a maximum of −11 cmdetected at Collinson Head. No significant movement on the north-west coast is captured in this data set. Generally the island shows littlemovement, on the order of+/−2 cm, but there is some subtlemottling,particularly near Orca Cove and the southern tip of the island, showingpatches of movement on the order of −2 to −4 cm over the summer.Again, the patterns do not clearly align with local topography, nor dothey necessarily concur with the TerraSAR-X observations, thereforethey could be an interaction of the C-band SAR with the vegetation, orresidual tropospheric noise, rather than ground movement.

Because of our repeat acquisitions in 2009 and 2010, we were ableto make one year repeat pairs with the RADARSAT-2 data. Of our fivepossible pairs, only one was successfully processed to a good dis-placement product. The lower panel of Fig. 8 presents these results.There are some patches where data are missing, due to loss of coher-ence, mostly along stream channels and gullies, but in general, goodpatterns can be seen. The north-east coast shows movement on the

nd TerraSAR-X summer 2010, ascending sequence (lower panel). The acquisition dates

Fig. 7. Line-of-sight displacement results from RADARSAT-2 data sequence through summer 2009. Displacement is in colour, with brightness a function of radar intensity. Blackareas are no displacement data, due to loss of coherence.

3499N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

order of −10 to −20 cm over the year. Some terrain instability alongthe coast and slumps of Thetis Bay can be seen, and along streamchannels. The north-west coast manifests some movement towardsthe satellite, on the order of 2 to 5 cm over the year. A patch of subtlemovement is visible near the northern tip of the island, on the orderof−2 to−4 cm between 2009 and 2010, and outlined with a red cir-cle. For RADARSAT-2 with an incidence angle of 37.8° and a perpen-dicular baseline of 251 m, it would require an error of 48 m in theDEM to produce an error of 2 cm in the displacement results, thuswe are confident that even though this area was not previouslyknown to be unstable, this movement is real and not an artefact of aDEM error.

5.4. ALOS-PALSAR

We have two successful summer data sets for ALOS-PALSAR, 2007and 2010, shown in the upper and lower panels of Fig. 9 respectively.Summer movement is relatively modest and varies slightly betweenthe years. In 2007 there was movement on the order of −2 to−3 cm noted for the highest point on the island, 2 to 3 cm of move-ment towards the satellite along the north-west coast and −2 to−6 cm of movement along the north-east coast. In 2010, the highestpoint on the island shows more modest movement of −1 to −2 cm.Over the steep slopes in the interior (just east and south of the high-est point)−3 to−6 cm of movement is seen, and movement reaches−13 cm at Collinson Head, which is not covered in the 2007 pair.Given that the periods of coverage correspond to different parts ofthe summer (2007 is late summer/fall (August 19–October 04) and2010 is mid to late summer (July 24–September 08)), we shouldprobably not conclude too much from the differences, suffice to saythat the 2010 mid to late summer data set appears to be more dynam-ic, with more areas experiencing a loss of coherence, and greateramounts of movement, particularly over slopes.

The one year separation PALSAR results are more dramatic. In the2007–2008 pair shown in the top panel of Fig. 10, the displacementalong the north-east coast is very well developed, showing move-ment of −10 to −30 cm over the full year. The area at CollinsonHead has lost coherence, indicating that either the surface structurehas changed significantly or that the local spatial deformation gradi-ent is too high for the InSAR method. Other pockets of movementare visible along the Thetis Bay coast, and associated with topograph-ical features in the interior of the island. Some displacement towardsthe sensor is visible along the north-west coast (3 to 5 cm), and again,the patch of movement near the northern tip of the island is visible,on the order of −5 cm between 2007 and 2008.

The middle panel of Fig. 10 shows the results from the two yearPALSAR data pair (2008–2010). Movement is again visible over theactive Bell Bluff cliffs, but there is significant loss of coherence in themost dynamic sections and phase unwrap errors are beginning tocreep in due to disconnects between coherent regions. The northernpatch is again apparent (−5 to −7 cm between 2008 and 2010), asis some movement and coherence loss along the Thetis Bay coast.Some movement is apparent at Pauline Cove, and in the vicinity ofOrca Cove, but the results are noisy with many patches of coherenceloss, so we hesitate to draw conclusions from these. The perpendicu-lar baseline for this pair was very large (6653 m) and baseline decorr-elation may be responsible for some of the coherence loss.

The lower panel of Fig. 10 shows the results from the PALSARthree year pair (2007–2010). The patterns of movement are surpris-ingly complete apart from the very active north-east coastline andalong the coast of Thetis Bay, in both these areas there is significantslumping and coherence has been lost. The patch of movement atthe north of the island is again visible, showing movement of up to−12 cm over the 3 years. This concurs well with the RADARSAT-2and PALSAR one year results which demonstrated −2 to −5 cm ofmovement per year over this feature. Movement along the north-

Fig. 8. Line-of-sight displacement measured from RADARSAT-2 with stacking of summer 2009 ascending sequence (upper panel) and acquisitions of August 1, 2009 and July 27,2010, almost one year separation (lower panel). The acquisition dates used are listed in the lower right corner of each panel. Red circle outlines the northern patch of subtle move-ment. The underlying background images are SAR intensity, these are visible where there are no displacement results.

3500 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

west coast is also quite noticeable over the 3 years, up to 12 cm. Cen-tral portions of the island also show patterns of ground movement.Many are clearly slope related processes, others suggest a more gen-eral lowering of the ground surface. But we know this data set hassome ionospheric variation across the island, therefore the valuesare questionable. We include these results more to demonstrate thecoherence potential of the L-band PALSAR data over long time periodswhen the baseline is relatively well maintained, 1769 m in this case.

5.5. Slump features

Retrogressive thaw slumps are important features in ice-rich per-mafrost environments. They are of great interest to permafrost geo-scientists, since their presence identifies areas experiencing changeand unstable slopes, and they are a key mechanism for ground ice ex-posure, subsequent thaw and the release of stored carbon. Thawslumps are common on Herschel Island and can be large, measuringtens to hundreds of metres in width. Ground displacement on theorder of tens of metres may be observed in any given year. Obviously,such large scale movement is a challenge for the InSAR method, sinceabrupt changes of this magnitude will cause jumps in phase valuesleading to unwrap errors and erroneous results. While frequent

observations may mitigate this problem in some measure, the disin-tegration of slump sections and the presence of mud-flows is alsocommon in thaw slump areas, these cause complete loss of coherenceand hence no InSAR results at all.

Fig. 11 shows the InSAR results for three slump areas along ThetisBay. The outlines of the Hawk, Ice Wedge and ABC slumps as mappedin 2009 are shown in red. The TerraSAR-X descending results maintaincoherence and showmodest groundmovementwithin the slump areasduring the summer of 2009. The RADARSAT-2 ascending results overthe same period do not show that as clearly, possibly due to the lowertemporal resolution and the less favourable line-of-sight direction.The 2010 ALOS-PALSAR results show little movement at Hawk, subtlemovement for ABC and Ice Wedge, and a region of movement directlyabove the Ice Wedge slump. The RADARSAT-2 2009–2010 results pro-vide the least ambiguous results; coherence has been lost in mostslump areas presumably due to surface change, and the area above IceWedge concurs with the PALSAR summer 2010 results by providingsubtle evidence of a bowl feature. The PALSAR 2007–2008 resultsshow significant movement in the Hawk slump general area, andsome movement and loss of coherence in the Ice Wedge and ABCslumps. Movement above Ice Wedge is barely evident in the 2007–2008 results, so presumably the activity here accelerated in 2010.

Fig. 9. Seasonal line-of-sight displacement measured from ALOS-PALSAR acquisitions of August 19 and October 04, 2007 (upper panel) and July 24 and September 8, 2010 (lowerpanel). Underlying background images are SAR intensity, these are visible where there are no displacement results.

3501N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

One of the problems in this area is that the direction of groundmovement is perpendicular to the SAR line-of-sight and thereforenot well detected (especially at Ice Wedge and ABC). Acquiring satel-lite data on a descending path and hence looking from the oppositeside might improve this. The results from the TerraSAR-X descendingpath show that this does bring some improvement. However, sincesatellites always travel on an approximately north–south trajectory,there is limited flexibility. Thus any monitoring method could poten-tially miss some areas of considerable slope instability due to satellitelook direction.

6. Discussion

6.1. Accuracy and errors

In an attempt to quantify the level of noise in the results we calcu-lated the statistics for an area observed to be repeatedly stable in theInSAR results, the control area outlined in Fig. 3. Table 7 records themean movement and the standard deviation for each data set. Fromthese we can see that the standard deviation of the measurementsis b1 cm for TerraSAR-X and RADARSAT-2, and b1.5 cm for PALSAR.We suggest these values as approximate margins of error, resulting

from general noise, registration errors and inaccuracies in satellitebaseline estimates.

These values represent the accuracy with which the signal isreturned from the same effective scattering centre on the ground.This is not necessarily the fidelity with which the signal capturesmovement of the ground. The scattering centre returning the signalmay be the ground surface, a few millimetres below the surface, orwith the X- and C-band, possibly vegetative elements on the surface.We are confident that we are measuring distance to the same scatter-ing centre to this accuracy, but it remains to be seen whether this isthe accuracy of the ground movement measurement. We also haveto be alert to tropospheric or ionospheric artefacts in specific datasets, these errors are over and above the general margins of error.

A potential source of error that deserves special consideration isthe effect of soil moisture. Because a radar wave will penetrate dryground, the signal may in fact be returned by the wet/dry interfacewithin the soil, thus any displacement signal could actually be themovement of the soil moisture front, rather than ground surface dis-placement (Nolan et al., 2003; Nolan & Fatland, 2003a). Nolan andFatland (2003b) developed a model for this effect and displayedmod-elled penetration depths for X-, C-, and L-band SARs in their Fig. 2.Their model, which includes transmission loss as well as attenuation

Fig. 10. Line-of-sight displacement measured from ALOS-PALSAR acquisitions of August 31, 2007 and September 2, 2008 — one year separation (upper panel), September 02, 2008and July 24, 2010, an interval of almost two years (middle panel) and August 31, 2007 and September 8, 2010, an interval of three years (lower panel). Underlying backgroundimages are SAR intensity. Red circle outlines the northern patch of subtle movement.

3502 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

Fig. 11. InSAR displacement results for thaw slumps of Thetis Bay. Red lines are slump headwall locations as mapped in 2009 by field surveys. Slump outlines are Hawk (left), Ice Wedge (centre) and ABC (right).

3503N.Short

etal./

Remote

SensingofEnvironm

ent115

(2011)3491

–3506

Table 7Mean ground movement and standard deviation for control area outlined in Fig. 3.

Data set Mean movement(cm)

Standarddeviation (cm)

TerraSAR-X descending stack 2009 −0.72 0.69TerraSAR-X ascending stack 2010 −0.14 0.58RADARSAT-2 ascending stack 2010 0.86 0.94RADARSAT-2 one year 2009–2010 0.27 0.45ALOS-PALSAR seasonal 2007 −0.34 1.44ALOS-PALSAR seasonal 2010 0.57 0.82ALOS-PALSAR one year 2007–2008 0.12 0.63ALOS-PALSAR two year 2008–2010 0.98 0.99ALOS-PALSAR three year 2007–2010 −0.32 0.84

3504 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

loss, which is presumed to bemore realistic than attenuation loss alone,shows that for volumetricwater content above 35% the radarwave doesnot penetrate into the ground at X-, C- or L-band. Soil moisture levels of11% or less would be required to achieve penetration depths of greaterthan 1 cm for all bands. Our field data collected on August 2nd, 2010,show that the volumetric water content ranges from 29 to 78%(Table 6). According to the Nolan and Fatland model (2003b), 29% vol-umetric water content would result in ground penetration of 0.7 mmatX- and C-band, and 2.1 mm at L-band. These values are well within thestandard deviations observed for the measurements and our generallyaccepted margins of error. While we only have one set of soil moisturemeasurements, which does not inform us of seasonal drying trends forthis area, the August measurements are late-summer, which would beafter the primary drying transition of spring to early summer. Soil mois-ture values might therefore not decrease much further. We thereforesuggest that this cool maritime environment sustains generally highsoil moisture levels that preclude significant SAR penetration, andeven if there is minor penetration, it is well containedwithin the gener-al margins of error of 1 cm for TerraSAR-X and RADARSAT-2 and 1.5 cmfor PALSAR.

6.2. Validation

Unfortunately, it is difficult to quantitatively evaluate the InSARmeasurements of movement. The remote location, the long repeat pe-riods between data acquisitions and, as was the case for ALOS-PALSAR,the lack of user knowledge about satellite acquisition dates,made a con-current field campaign difficult to realise. There are however, two re-cent studies of Herschel Island that permit some evaluation of theInSAR results.

The first study is by Burn and Zhang (2009). From steel pipes withbottom plates anchored below the permafrost table, they observedincreasing active layer thaw depths between 2005 and 2007, on theorder of 11 cm along a profile at Collinson Head (red line in Fig. 3).Related subsidence caused the increasing protrusion of the steelpipes. The mean protrusion increase was 6.5 cm over the two years.

The summer of 2009 TerraSAR-X results show movement of 0 to−2 cm for most of the Collinson Head profile. The summer 2010 Ter-raSAR-X results show similar values. The RADARSAT-2 summer 2009results vary from +1 to −2 cm along the profile. These values are allvery close to the margin of error, and they only cover the summer pe-riod, in which one would expect small amounts of subsidence relatedto seasonal settlement of the active layer. Qualitatively they seem rea-sonable, if considerably less than the Burn and Zhang measurements.The PALSAR summer 2010 results show−2 cm at the top of CollinsonHead, but +2 cm at the western end of the profile, which is move-ment towards the sensor; this is also the local downslope directionand PALSAR could therefore be capturing normal slope processessuch as creep or solifluction.

The RADARSAT-2 one year data set from 2009 to 2010 showsvalues of +1 to −1 cm over the majority of the profile, hence mostlywithin the margin of error and suggesting no significant change

during this period. The PALSAR 2007–2008 results manifest a rangeof values, from −5.0 cm at the top of Collinson Head, to −1.0 cmfor the majority of the profile, but with 3.0 cm of movement towardsthe sensor in the lower portion of the profile, close to Pauline Cove.Again, this movement towards the sensor could be natural slope pro-cesses rather than change in permafrost.

The PALSAR two year data set suffers from a loss of coherence inthis region, but the few available points suggest movement of −1.0to −3.5 cm over the profile between 2008 and 2010. The three yearPALSAR data set is more complete but quantitatively suspect due tothe ionospheric influence across the island, observed values are 0 to−2.5 cm over most of the profile between 2007 and 2010.

Many of the InSAR values are very close to the margin of error of~1 cm, and therefore do not indicate significant change. However,all the data sets report negative values over the middle portion ofthe profile, indicating downward movement rather than uplift, andvalues are on the order of magnitude that one might expect in oneyear due to the seasonal settlement of the active layer. The InSARvalues are all lower than the Burn and Zhang (2009) observation of6.5 cm between 2005 and 2007, but it is possible that rates of subsi-dence have declined since the 2005 and 2007 field measurements.In addition, the InSAR results are presented as line-of-sight displace-ment, whereas field measurements would be purely vertical mea-surements. If the InSAR line-of-sight displacements were convertedto vertical displacement the InSAR measurements would be evensmaller. We therefore tentatively suggest that values provide qualita-tively correct information, but not yet quantitatively conclusive.

Unfortunately, ground truth measurements are always point mea-surements and InSAR results are always representative of a patch onthe ground, from 12 m2 (RADARSAT-2 and TerraSAR-X) to 350 m2

(ALOS-PALSAR), therefore exact agreement between field observa-tions and InSAR observations might never be obtained. The installa-tion of point targets such as corner reflectors might be the only wayto obtain an exact match between field and InSAR observations. Forthe cumulative seasonal results there is also the possible impact ofthe assumption of linear deformation, as specific thaw events maybe smoothed out in the statistical extraction of the summer trend.An alternative stacking approach might be preferable. And, thawmay begin in late April or mid-May (Alasset et al., 2008), and ourInSAR acquisitions begin in June, therefore we might have missedsome of the earliest movement.

The PALSAR data are unique in demonstrating movement towardsthe sensor near Pauline Cove. Potentially the L-band detects slopeprocesses that the other sensors do not, such as creep or solifluction.It is not clear why this is, it is possible that surface characteristics suchas vegetation obscure this subtle signal in the shorter wavelengthTerraSAR-X and RADARSAT-2 data sets.

A second study providing field data is from Lantuit and Pollard(2008). They measured rates of erosion along sections of the Herschelcoastline using aerial photography, optical satellite imagery and fieldsurveys. In their map of coastal retreat rates for the 1970 to 2000 pe-riod, they showed retreat rates ranging between 0.3 and 3.1 m peryear along the north–east facing section. The InSAR measurementsonly detect a maximum of 30 cm of movement here, greater thanthis and coherence seems to be lost, presumably due to significantchange in the surface structure, or possibly due to large spatial defor-mation gradients. Since the InSAR method cannot succeed at measur-ing rates of movement that are very large and abrupt, such as cliffcollapse and retreat, its value lies in identifying areas of terrain insta-bility that may be a precursor to slumping or cliff collapse.

The potential subsidence of the ground due to thawing of nearsurface permafrost and a deepening of the active layer, is related toits excess ice content (i.e. water (ice) content over and above thesoil porosity). Burn and Zhang (2009) measured excess ice contentat 10 sites on Collinson Head by drilling; excess ice content variedfrom 22 to 63% with the average being 47%. They calculated that a

3505N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

10 cm deepening of the active layer and subsequent melt of excess icecontent, could result in between 0.5 and 17 cm of subsidence(depending on the excess ice content at a particular site). A 20 cmdeepening of the active layer could produce between 3 and 34 cmof subsidence. Since this deepening of the active layer would takeplace gradually over several years or even decades, the settlementrates per year would be on the order of centimetres, and this is theorder of magnitude that the InSAR method detects. The net changefrom year to year, detectable with the long interval C- and L-band in-terferograms could make a special contribution to detecting this cu-mulative settlement.

Like Li et al. (2003) we conclude that spatial patterns of variationcan be detected, and we believe the results are qualitatively correct,and the correct order of magnitude, but precise and simultaneousground measurements would be needed to thoroughly validate thedisplacements. Even then, the agreement between point field mea-surements and pixel scale InSAR measurements might not be exact.

6.3. Thaw slump monitoring

Thaw slump activity on Herschel Island was not consistentlyidentified or quantified in these InSAR results, although generallyunstable areas were identified if the line-of-sight direction wasfavourable. TerraSAR-X seemed tomaintain the best summer coherenceover these features, presumably due to the high temporal and spatialresolution of the sensor. Since the loss of coherence is the least ambig-uous result in slumps, coherent change detection may be a more suit-able technique for these features. Such an approach has been used todelineate the area of active landslides in more southerly locations, forexample Calabro et al. (2010). An alternative InSAR approach for thawslumpmonitoring is the use of coherent targets. This involves the instal-lation of radar corner reflectors in and around a slump, these maintaincoherence even as the slump moves, for example Singhroy et al.(2008). This is not an ideal approach since the logistics associatedwith corner reflector installation are large in remote northern environ-ments and the results would be very sparse. And, if the movement islarge and abrupt and targets move dramatically, more than half a wave-length between acquisitions, the results could be incorrect due to phaseunwrap errors. For the remote northern permafrost environment thestrength of the InSAR method may lie in identifying flatter or gentleslope areas experiencing subsidence, rather than the movement withinslumps themselves, for example, the unstable areas above slumps or thepreviouslymentioned patch ofmovement detected in the northern partof the island. Such areas of modest subsidence indicate change, possiblydue to ground ice melt and the subsequent loss of water volume, andmay identify instability prior to the development of active layer detach-ments and slumps.

7. Conclusions

The InSAR results for this permafrost environment are very promis-ing. Broad areas of known instability are well mapped by all three sen-sors with a favourable look direction, andwith quantitatively consistentresults. Having several independent InSAR data sets is advantageous foridentifying and confirming real trends. Our results spanning 2007 to2010 suggest a band of displacement along the north-east coast ofHerschel Island of 20 to 30 cm/year, with displacements exceeding30 cm/year on the north-eastern edge of Collinson Head. This activityalong the north-east coast is known to be driven by coastal erosionand unstable slopes. A region of movement of b5 cm/year was alsodetected near the northern tip of the island; this was not previouslyknown and will be the subject of an upcoming field study.

The TerraSAR-X data demonstrated good coherence in the sum-mer data sets, but stacking was essential to overcome noise and ex-tract results. Even after stacking some vegetation effects ortropospheric noise seemed to remain. The high temporal resolution

and fine spatial resolution of TerraSAR-X make the data potentiallybest for detecting thaw slump dynamics. The TerraSAR-X data havethe disadvantage that year to year interferograms are not possiblein this environment, therefore detecting long-term ground move-ment trends is more difficult.

The RADARSAT-2 C-band results were more reliable. Areas of sig-nificant deformation could be identified in single interferograms andstacking further improved the summer results. Areas of knownmove-ment were well identified, although some noise related to vegetationor tropospheric effects remained in the stacked products. The possi-bility of forming one year interferograms exists for C-band data, andthese net annual change interferograms are particularly valuable foridentifying long-term change in a permafrost environment.

The ALOS-PALSAR L-band data provide the best coherence and themost complete results. Summer interferograms produced patterns ofmovement that could be intelligently related to local topography,suggesting that active layer dynamics are captured in the PALSAR sea-sonal pairs. Areas of significant movement on the island were consis-tently identified, as well as some more subtle slope processes. Thepossibility of one year interferograms is most hopeful for identifyingareas of permafrost instability and long-term trends. Multiple yearPALSAR data sets can also be successful, and provide interesting andconsistent patterns of movement; but there are problems with base-line decorrelation, detection of ionospheric effects, and phaseunwrapping in these data sets. An L-band data source with tightlycontrolled baselines and a shorter revisit period would overcomemost of these limitations.

Specific slump activity and movement within slumps on HerschelIsland was not reliably identified or quantified by the InSAR, althoughunstable areas were identified if the line-of-sight direction wasfavourable. Depending on the location, operational monitoring ofslump features may need to combine two look directions to mapthe deformation activity, and even then, some movement may goundetected. The abrupt and large scale nature of slumping, and thedynamics within slumps, are not ideally suited to the InSAR method.The frequent revisit and high spatial resolution of TerraSAR-X providethe best chance of maintaining coherence over these features. Theloss of coherence due to slump activity is the least ambiguous result,therefore coherent change detection may be a more suitable tech-nique for these features. The installation of corner reflectors andpoint target InSAR methods could yield better results, but theywould be sparse and the installation logistics would be hard to justifyin remote northern locations. In general, repeat pass InSAR is bettersuited for detecting broad areas of terrain instability in gentle relief,potentially caused by permafrost thaw or ground ice melt and the re-moval of water volume, and prior to significant slumping.

In terms of monitoring permafrost stability, it is really the long-term changes that are of most interest, since we know that seasonalchanges in the active layer are normal and cyclical. Therefore, theability of RADARSAT-2 and particularly ALOS-PALSAR to form oneyear and possibly longer interval interferograms is the greatest po-tential contribution for identifying permafrost and landscape change.

Acknowledgements

This work was funded by the Remote Sensing Science Program ofNatural Resources Canada. This publication forms ESS contribution20110051. Additional resources were provided by the CanadianSpace Agency Government Related Initiatives Program. Field datawere processed and provided by Heather Cray of McGill University.The ALOS-PALSAR data are copyright JAXA, METI and were processedand provided by the Americas ALOS Data Node. The TerraSAR-X dataare copyright DLR and were generously provided under projectLan0440. RADARSAT-2 data are copyright McDonald, Dettwiler andAssociates and were provided through the Government of CanadaData Allocation. We thank Francois Charbonneau and Laurence Gray

3506 N. Short et al. / Remote Sensing of Environment 115 (2011) 3491–3506

for their review of the paper and helpful comments. We thank threeanonymous reviewers who provided feedback and significantly im-proved the quality of the manuscript.

References

Alasset, P. -J., Poncos, V., Singhroy, V., & Couture, R. (2008). SAR monitoring of perma-frost melt in the Lower Mackenzie Valley, Canada. Proceedings of IGARSS 2008, 6–11July, Boston, Massachusetts. New York, NY: IEEE.

Burn, C. R., & Zhang, Y. (2009). Permafrost and climate change at Herschel Island(Qikiqtaruq), Yukon Territory, Canada. Journal of Geophysical Research, 114,F02001, doi:10.1029/2008JF001087.

Calabro, M. D., Schmidt, D. A., & Roering, J. J. (2010). An examination of seasonal defor-mation at the Portuguese Bend landslide, southern California, using radar interfer-ometry. Journal of Geophysical Research, 115, F02020, doi:10.1029/2009JF001314.

Costantini, M. (1998). A novel phase unwrapping method based on network program-ming. IEEE Transactions on Geoscience and Remote Sensing, 36(3), 813–821, doi:10.1109/36.673674.

Gabriel, A. K., Goldstein, R. M., & Zebker, H. A. (1989). Mapping small elevation changesover large areas: Differential radar interferometry. Journal of Geophysical Research,94(B7), 9183–9191.

Gatelli, F., Monti Guarnieri, A., Parizzi, F., Pasquali, P., Pratti, C., & Rocca, F. (1994). Thewavenumber shift in SAR interferometry. IEEE Transactions on Geoscience and Re-mote Sensing, 32(4), 855–865.

Goldstein, R. M. (1995). Atmospheric limitations to repeat-track radar interferometry.Geophysical Research Letters, 22, 2517–2520.

Goldstein, R. M., & Werner, C. L. (1998). Radar interferogram filtering for geophysicalapplications. Geophysical Research Letters, 25(21), 4035–4038.

Gray, A., Mattar, K., & Sofko, G. (2000). Influence of ionospheric electron density fluctuationson satellite radar interferometry. Geophysical Research Letters, 27(10), 1451–1454.

Lantuit, H., & Pollard, W. (2008). Fifty years of coastal erosion and retrogressive thawslump activity on Herschel Island, southern Beaufort Sea, Yukon Territory, Canada.Geomorphology, 95, 84–102.

Li, S., Romanovsky, V., Lovick, J., Wang, Z., & Peterson, R. (2003). Application of satelliteSAR imagery in mapping the active layer of Arctic permafrost. Final Report NAG5-8614. : NASA Centre for Aerospace Information.

Liu, L., Zhang, T., & Wahr, J. (2010). InSAR measurements of surface deformation overpermafrost on the North Slope of Alaska. Journal of Geophysical Research, 115,F03023, doi:10.1029/2009JF001547.

Mackay, J. R. (1959). Glacier ice-thrust features of the Yukon coast. Geographical Bulle-tin, 13, 5–21.

Massonnet, D., & Feigl, K. (1995). Discrimination of geophysical phenomena in satelliteradar interferograms. Geophysical Research Letters, 22, 1537–1540.

Massonnet, D., & Feigl, K. (1998). Radar interferometry and its application to changes inthe earth's surface. Reviews of Geophysics, 36, 441–500.

Mitchell, G., & MacNabb, K. (2010). High resolution stereo satellite elevation mappingaccuracy assessment. Proceedings of the ASPRS Annual Conference, April 26–30, 2010,San Diego, California.

Nolan, M., & Fatland, D. (2003). New DEMs may stimulate significant advancements inremote sensing of soil moisture. EOS Transactions of the American GeophysicalUnion, 84(25), 233–237.

Nolan, M., & Fatland, D. (2003). Penetration depth as a DInSAR observable and proxyfor soil moisture. IEEE Transactions on Geoscience and Remote Sensing, 41(3),532–537.

Nolan, M., Fatland, D., & Hinzman, L. (2003). DInSAR measurement of soil moisture.IEEE Transactions on Geoscience and Remote Sensing, 41(12), 2802–2813.

Pollard, W. H. (1990). The nature and origin of ground ice in the Herschel Island area,Yukon Territory. Proceedings, Fifth Canadian Permafrost Conference, Québec(pp. 23–30).

Pollard, W. H., & Dallimore, S. R. (1990). Petrographic characteristics of massive groundice, Yukon Coastal Plain, Canada. Paper Presented at the 5th International Conferenceon Permafrost, Trondheim, Norway, August 1988.

Rosen, P. A., Hensley, S., Zebker, H. A., Webb, F. H., & Fielding, E. (1996). Surface defor-mation and coherence measurements of Kilauea Volcano, Hawaii from SIR-C radarinterferometry. Journal of Geophysical Research, 101(E10) 23,109 - 23,125.

Rykhus, R. P., & Lu, Z. (2008). InSAR detects possible thaw settlement in the AlaskanArctic Coastal Plain. Canadian Journal of Remote Sensing, 34(2), 100–112.

Short, N., Brisco, B., Gray, A. L., Budkewitsch, P., & Murnaghan, K. (2008). ALOSInSAR for permafrost monitoring applications. Proceedings of the 2nd ALOS PISymposium, Nov 3–7, 2008, Rhodes, Greece. : ESA Communication ProductionOffice SP-664.

Short, N., Brisco, B., Budkewitsch, P., & Murnaghan, K. (2009). ALOS-PALSAR interfer-ometry for permafrost monitoring in Canada. Proceedings of the 3rd ALOS PI Sympo-sium, November 9–13, 2009, Big Island, Hawaii. Fairbanks, Alaska: Alaska SARFacility.

Singhroy, V., Alasset, P. -J., Couture, R., & Froese, C. (2008). InSAR monitoring of land-slides in Canada. Proceedings of IGARSS 2008, 6–11 July, Boston, Massachusetts.New York, NY: IEEE.

Smith, C. A. S., Kennedy, C. E., Hargrave, A. E., & McKenna, K. M. (1989). Soil and vege-tation of Herschel Island, Yukon Territory. Yukon Soil Survey Report, Vol. 1, Ottawa:Land Resource Research Centre, Agriculture Canada.

Takeuchi, S., & Yamada, S. (2002). Comparison of InSAR capability for land subsidencedetection between C-band and L-band SAR. Proceedings of IGARSS 2002, 24–28 June,Toronto, Canada (pp. 2379–2381). New York, NY: IEEE.

Wang, Z., & Li, S. (1999). Thaw deformation of permafrost active layer near Toolik Lake,Alaska, imaged by DInSAR technique during summer time. EOS Transactions of theAmerican Geophysical Union, November 16, 1999, Fall Meeting Supplement (H41A-37).

Wang, Z., & Li, S. (1999). Detection of winter frost heaving of the active layer of Arcticpermafrost using SAR differential interferograms. Proceedings of IGARSS 1999.June28–July2, Hamburg, Germany (pp. 1946–1948). NJ: IEEE Piscataway.

Werner, C. (2010). Canada Centre for Remote Sensing, Ottawa March 24, 2010.Werner, C. L., Hensley, S., Rosen, P., & Wegmueller, U. (1997). Comparison of repeat

track interferometric correlation signatures from ERS-1, ERS tandem, SIR-C andJERS-1. ERS SAR Interferometry; Volume 2, ESA SP 1997, 406(1) (pp. 117). Paris: Eu-ropean Space Agency.

Werner, C., Wegmüller, U., Strozzi, T., & Wiesmann, A. (2000). GAMMA SAR and inter-ferometric processing software. Proceedings of ERS-Envisat Symposium, Gothenburg,16–20 October, 2000.

Werner, C., Wegmüller, U., Strozzi, T., & Wiesmann, A. (2002). Processing strategies forphase unwrapping for InSAR applications. Proceedings of EUSAR, Cologne 4–6 June,2002.

Yonezawa, C., & Takeuchi, S. (2000). Land subsidence detection using long intervalERS/SAR data pairs. Proceedings of IGARSS 2000, 24–28 July, Honolulu, Hawaii.New York, NY: IEEE.

Zebker, H. A., Rosen, P. A., & Hensley, S. (1997). Atmospheric effects in interferometricsynthetic aperture radar surface deformation and topographic maps. Journal ofGeophysical Research, 102(B4), 7547–7563.