Analyzing Glacier Surface Motion Using Lidar Data

Home , Glacial motion, Glaciology

Portland State University PDXScholar

Geology Faculty Publications and Presentations Geology

3-17-2017 Analyzing Glacier Surface Motion Using LiDAR Data

Jennifer W. Tellig University of Houston

Craig Glennie University of Houston

Andrew G. Fountain Portland State University, [email protected]

David C. Finnegan Cold Regions Research and Engineering Laboratory Remote Sensing/GIS Center of Excellence

Follow this and additional works at: https://pdxscholar.library.pdx.edu/geology_fac

Part of the Geology Commons, and the Glaciology Commons Let us know how access to this document benefits ou.y

Citation Details Telling, J. W., Glennie, C., Fountain, A. G., & Finnegan, D. C. (2017). Analyzing Glacier Surface Motion Using LiDAR Data. Remote Sensing, 9(3), 1-12. doi:10.3390/rs9030283

This Article is brought to you for free and open access. It has been accepted for inclusion in Geology Faculty Publications and Presentations by an authorized administrator of PDXScholar. Please contact us if we can make this document more accessible: [email protected]. remote sensing

Article Analyzing Glacier Surface Motion Using LiDAR Data

Jennifer W. Telling 1,*, Craig Glennie 1, Andrew G. Fountain 2 and David C. Finnegan 3

1 National Center for Airborne Laser Mapping, University of Houston, 5000 Gulf Freeway, Building 4, Room 216, Houston, TX 77204-5059, USA; [email protected] 2 Department of Geology, Portland State University, P.O. Box 751, Portland, OR 97207-0751, USA; [email protected] 3 U.S. Army Engineer Research and Development Center Cold Regions Research and Engineering Laboratory Remote Sensing/GIS Center of Excellence, ATTN: CEERD-PA-H, 72 Lyme Road, Hanover, NH 03755-1290, USA; [email protected] * Correspondence: [email protected]; Tel.: +1-201-835-8478

Academic Editors: Guoqing Zhou, Xiaofeng Li and Prasad S. Thenkabail Received: 31 January 2017; Accepted: 14 March 2017; Published: 17 March 2017

Abstract: Understanding glacier motion is key to understanding how glaciers are growing, shrinking, and responding to changing environmental conditions. In situ observations are often difﬁcult to collect and offer an analysis of glacier surface motion only at a few discrete points. Using light detection and ranging (LiDAR) data collected from surveys over six glaciers in Greenland and Antarctica, particle image velocimetry (PIV) was applied to temporally-spaced point clouds to detect and measure surface motion. The type and distribution of surface features, surface roughness, and spatial and temporal resolution of the data were all found to be important factors, which limited the use of PIV to four of the original six glaciers. The PIV results were found to be in good agreement with other, widely accepted, measurement techniques, including manual tracking and GPS, and offered a comprehensive distribution of velocity data points across glacier surfaces. For three glaciers in Taylor Valley, Antarctica, average velocities ranged from 0.8–2.1 m/year. For one glacier in Greenland, the average velocity was 22.1 m/day (8067 m/year).

Keywords: terrestrial laser scanning; airborne laser scanning; LiDAR; morphology; glacier surface velocity

1. Introduction Glaciers are dynamic and in constant flux, and understanding glacial motion provides important information about their growth and retreat [1]. However, the size and extent of glaciers, along with challenging environmental conditions, can severely limit data collection in the field or, in particularly hazardous conditions, field work may be precluded entirely [2–4]. As remote sensing data becomes more widely available, it is becoming a primary data collection technique in the cryosphere, particularly in areas that are inaccessible to traditional field methods [2]. Traditional methods of collecting glacier velocity data rely on stakes drilled into the ice and/or GPS deployed on the glacier surface [5–8]. The optical and geodetic surveys require little post-processing but only provide data at discrete points, usually covering a fraction of the entire glacier. Remote sensing techniques, including synthetic aperture radar (SAR) and multispectral imagery, are increasingly being used to detect and monitor changes in the cryosphere [4,9–12]. In difficult to reach areas of the Himalayas, SAR has been used to measure flow velocities at the Kangshung and Khumbu Glaciers [11]. SAR was also used to measure flow velocities at the Shirase Glacier, Antarctica, and at Helheim Glacier, in Greenland [9,10,12]. Often, these methods rely on satellite based platforms, which offer regular, long duration coverage, but they can be inhibited by severe topography and viewing

Remote Sens. 2017, 9, 283; doi:10.3390/rs9030283 www.mdpi.com/journal/remotesensing Remote Sens. 2017, 9, 283 2 of 12 Remote Sens. 2017, 9, 283 2 of 13 offerangle regular, [11,12]. Inlong addition, duration SAR coverage, tends to but have they low ca backscatteringn be inhibited intensity by severe over topography dry snow, and reducing viewing the volumeangle [11,12]. of data In collected addition, in SAR some tends environments to have low [11 backscattering,12]. intensity over dry snow, reducing the volumeTerrestrial of data and collected airborne laserin some scanning environments (TLS and [11,12]. ALS, respectively) are two active remote sensing techniquesTerrestrial that useand lightairborne detection laser and scanning ranging (TLS (LiDAR), and ALS, typically respectively) in the near-infrared are two active wavelengths, remote sensingto provide techniques precise 3Dthat elevation use light models detection of surfaces. and ranging TLS and(LiDAR), ALS have typically both beenin the used near-infrared to survey wavelengths,glaciers to create to provide highly precise detailed 3D digital elevation elevation model modelss of surfaces. (DEMs) TLS for and the ALS purpose have ofboth determining been used tointerannual survey glaciers variability to create in surface highly elevation detailed in di ordergital toelevation estimate models mass balance (DEMs) or for long-term the purpose volume of determiningchange [2,3,13 interannual,14]. variability in surface elevation in order to estimate mass balance or long- term Involume this study, change we [2,3,13,14]. use LiDAR DEMs, collected over time, to calculate glacier surface velocity using two differentIn this study, methods—particle we use LiDAR DEMs, image collected velocimetry over (PIV) time, andto calculate manual glacier tracking. surface Applying velocity PIV, using an twoimage different processing methods—particle technique [15–18 image], to LiDAR velocimetry data provides (PIV) and an opportunity manual tracking. to measure Applying velocity PIV, across an imageentire glaciersprocessing rapidly technique and the [15–18], ability to measureLiDAR data high provides resolution an nuances opportunity in the to flow measure field thatvelocity may acrossbe missed entire using glaciers other rapidly techniques. and the When ability compared to measure to other high methodsresolution such nuances as feature in the extraction, flow field that and mayother be image missed correlation using other techniques, techniques. PIV When offers compar an approached to other that methods is sensitive such to as small feature scale, extraction, locally variableand other changes. image correlation This is especially techniques, important PIV offers on glacier an approach surfaces that where is sensitive common to features small scale, may deformlocally variableslightly betweenchanges. the This collection is especially of repeat important data. Successful on glacier application surfaces of where this method common will features make another may deformtechnique slightly available between for assessing the collection velocity of ofrepeat glaciers data. and Successful other slow application moving landforms. of this method The PIV will results make anotherare compared technique to manual available tracking for assessing results and, velocity where of available,glaciers and in situother data. slow The moving new method landforms. is tested The PIVon ALS results and are TLS compared point clouds to ofmanual six glaciers tracking in Antarcticaresults and, and where Greenland, available, with in data situ covering data. The a range new methodof spatial is andtested temporal on ALS resolutions. and TLS point clouds of six glaciers in Antarctica and Greenland, with data covering a range of spatial and temporal resolutions. 2. Study Sites 2. Study Sites The horizontal surface motion for six glaciers—five from Taylor Valley, Antarctica (Canada, Commonwealth,The horizontal Rhone, surface Suess, motion Taylor) for and six one glaciers—five from Greenland from (Helheim)—was Taylor Valley, analyzedAntarctica in the(Canada, study. TaylorCommonwealth, Valley (Figure Rhone,1) is oneSuess, of theTaylor) McMurdo and one Dry from Valleys Greenland (MDV) located(Helheim)—was in East Antarctica analyzed (77.5 in the◦S, 163study.◦E). Taylor The valley Valley landscape (Figure 1) is composedis one of the of McMurdo sandy, gravelly Dry Valleys valleybottoms (MDV) located with expanses in East ofAntarctica exposed (77.5°S,bedrock 163°E). and is populatedThe valley withlandscape perennially is composed ice-covered of sandy, lakes gravelly and ephemeral valley bo streamsttoms with originating expanses from of exposedglaciers thatbedrock flow intoand theis populated valleys from with the surroundingperennially ice-covered mountains [lakes19]. Air and temperatures ephemeral averagestreams originatingabout −17 ◦ Cfrom and glaciers summer that temperatures flow into typically the valleys hover from just belowthe surrounding freezing [20 ].mountains [19]. Air temperatures average about −17 °C and summer temperatures typically hover just below freezing [20].

Figure 1. TaylorTaylor Valley, Antarctica,Antarctica, withwith CommonwealthCommonwealth (Co),(Co), Canada (Ca), SuessSuess (S), Rhone (R), and Taylor (T) Glaciers indicated. The image was retrie retrievedved from the online Earth Ex Explorer,plorer, courtesy of the NASA EOSDISEOSDIS Land Land Processes Processes Distributed Distributed Active Acti Archiveve Archive Center (LPCenter DAAC), (LP USGS/EarthDAAC), USGS/Earth Resources ResourcesObservation andObservation Science (EROS) and Center,Science Sioux (EROS) Falls, South Center, Dakota, Sioux https://earthexplorer.usgs.gov/ Falls, South Dakota,. https://earthexplorer.usgs.gov/.

Remote Sens. 2017, 9, 283 3 of 12

Remote Sens. 2017, 9, 283 3 of 13 The MDV receive very little precipitation annually (3–50 mm water-equivalent) [21] and the glaciersThe move MDV only receive 0.3–20 very m/year, little precipitation though the fastestannually regions (3–50 weremm notwater-equivalent) in any of the areas[21] and surveyed the inglaciers this study move [7, 8only]. The 0.3–20 terminus m/year, of though Taylor Glacier,the fastes whicht regions is thewere area not examined in any of the in thisareas study, surveyed has beenin foundthis study to have [7–8]. surface The velocitiesterminus of ranging Taylor fromGlacier, 0.3–10 which m/year, is the area based examined on surveys in this from study, 1993–2001 has been and 2002–2004found to [have7,8]. Velocitiessurface velocities at the Canada ranging Glacier from 0. terminus3–10 m/year, were based found on to surveys be slower, from moving 1993–2001 at a and rate of up2002–2004 to 5 m/year [7–8]. and Velocities slowing at rapidly the Canada near theGlacier glacier terminus edges, were based found on theto be 1993-2001 slower, moving survey at [8 ].a rate of upHelheim to 5 m/year Glacier and (Figure slowing2 )rapidly is an outletnear the glacier glacier of edges, the Greenland based on the Ice 1993-2001 Sheet (66.38 survey◦N, [8]. 38.8 ◦W). AnnualHelheim variations Glacier in temperature (Figure 2) rangeis an betweenoutlet glacier±15 ◦ofC, the with Greenland summer temperaturesIce Sheet (66.38°N, averaging 38.8°W). around 5 ◦AnnualC[9]. In variations contrast toin glacierstemperature in Taylor range Valley, between Helheim ±15 °C, Glacier with movessummer quite temperatures rapidly with averaging velocities neararound the centerline 5 °C [9]. In of contrast the terminus to glaciers reaching in Tayl upor to Valley, 25 m/day Helheim and Glacier slowing moves up glacier quite rapidly(10–15 m/day)with andvelocities near the near glacier the centerline edges (0–5 of m/day) the terminus [9,10, 17reachi]. ng up to 25 m/day and slowing up glacier (10–15 m/day) and near the glacier edges (0–5 m/day) [9–10,17].

FigureFigure 2. 2.Helheim Helheim Glacier Glaciernear near thethe terminus.terminus. The blue blue triangle triangle indicates indicates the the TLS TLS scan scan position, position, the the orangeorange diamonds diamonds indicate indicate the the position position ofof thethe two GP GPSS units units on on the the glacier glacier surface surface during during the the TLS TLS data data collection,collection, and and black black boxes boxes indicate indicate the the bounds bounds of of study study regions regions 1 and1 and 2. The2. The image image was was retrieved retrieved from thefrom online the Earthonline Explorer, Earth Explorer, courtesy courtesy of the NASAof the EOSDISNASA EOSDIS Land Processes Land Processes Distributed Distributed Active Active Archive CenterArchive (LP Center DAAC), (LP USGS/EarthDAAC), USGS/Earth Resources Resources Observation Observation and Scienceand Science (EROS) (EROS) Center, Center, Sioux Sioux Falls, SouthFalls, Dakota, South Dakota, https://earthexplorer.usgs.gov/ https://earthexplorer.usgs.gov/. .

3.3. Methods Methods TheThe Taylor Taylor Valley Valley LiDARLiDAR point clouds clouds were were colle collectedcted during during two two aerial aerial surveys. surveys. NASA NASA flew the ﬂew theaerial aerial surveys surveys during during the the austral austral summer summer of of2000–2001 2000–2001 and and the the data data wa wass processed processed by bythe the US US GeologicalGeological Survey Survey to createto create a detailed a detailed DEM ofDEM the valleyof the bottomvalley [22bottom,23]. NASA’s[22–23]. AirborneNASA’s TopographicAirborne MapperTopographic (ATM), Mapper which has(ATM), a green which laser has with a green a scan laser angle with of a ±scan15◦ angleand aof scan ±15° frequency and a scan of frequency 20 Hz, was

Remote Sens. 2017, 9, 283 4 of 12 used for data collection. The NASA point cloud density was, on average, one point per 2.7 m2 [22,23]. The region was resurveyed in the summer of 2014–2015 by the National Center for Airborne Laser Mapping (NCALM), University of Houston, under contract from Portland State University, for the purpose of estimating landscape change over the intervening 14 years since the NASA survey [24]. The 2014 data were collected using a Teledyne Titan MW with three independent lasers operating at 532, 1064, and 1550 nm at an angle of ±30◦ and a scan frequency of 20 Hz [25]. An Optech Gemini ALTM, operating at 1064 nm with an angle of ±25◦ and a frequency of 35 Hz, was also used for some areas that required higher altitude collections due to extreme topography. Point cloud density ranged from 2 points per m2 to >10 points per m2 and averaged 4.7 points per m2 [26]. The Helheim Glacier surface was surveyed 23 times between 11 July and 12 July 2014 using a terrestrial laser scanner. The scanner was a RIEGL VZ-6000 operating at a wavelength of 1064 nm. For the purpose of this study, three TLS scans were selected for analysis. The first two scans were taken 51 min apart and the third scan was taken 12 h later, providing a test of fine time scale surface motion and overall surface change at Helheim. To examine the effect of varying point cloud density on the PIV analysis, two separate areas on Helheim glacier were studied in detail (Figure2). Region 1 was located further from the scanner and had an average point cloud density of 1 point per m2. Region 2, located near the scanner, had an average point cloud density of 5 points per m2. GPS data collected coincidentally with the TLS data at multiple points on the glacier surface is available to verify the velocity measurements derived from PIV. LiDAR point clouds cannot be used directly in PIV analysis, which is an image-based change detection technique [15–17]. The point clouds were first converted to greyscale images and colored by elevation using the Terrascan software package within MicroStation. Interpolation in regions with little or no data was limited to three pixels and interpolated regions accounted for less than 10% of each study area; though data gaps larger than three pixels exist in some places, they were not interpolated. The interpolation was done by taking an average of the elevation values from the surrounding pixels. The image generation process grids the point cloud data and the resolution for this process was chosen based on the manual tracking results. The Taylor Valley glacier images were produced at a resolution of 2 m and 0.5 m was used for Helheim Glacier. In both cases, the resolution was finer than the motion being measured, based on the results of the manual tracking measurements. Following the image conversion process, image pairs were analyzed using the PIVLab software, an open source MATLAB GUI [27–29]. Particle image velocimetry cross-correlates subsections of two images by searching for sets of features common to both images to solve for the component velocity of each feature [15,16]. When applied to glaciers, these trackable features are most commonly the peaks and valleys present on the glacier surfaces. PIVLab can complete multiple interpolation passes on a set of images at increasingly fine spatial resolution. The resolution of each pass was carefully chosen because a resolution smaller than the minimum displacement creates spurious results. A two pass interpolation was used; the resolutions of the first and second passes were scaled to be three times and two times, respectively, larger than the expected displacement, based on manual tracking results. For example, if the expected displacement was on the order of 1 m/year, the first pass would search the data in 3 m × 3 m grid cells and the second pass would use 2 m × 2 m grid cells. The grid resolution used to translate point clouds into images and the resolution of the PIV interpolation passes were two important factors in producing high resolution results and these values should always be chosen with respect to the temporal and spatial resolution of the data. Given the time interval between images and the vector field of displacement, the glacier surface velocities were calculated. Raw coordinates and u- and v-velocity vectors were exported from PIVLab in a text file format and plotted in MATLAB. Rather than plot thousands of individual vectors, the raw results were gridded to show the average motion in a given region. The Taylor Valley glaciers were gridded in 200 m × 200 m cells and the Helheim Glacier results were gridded in 100 m × 100 m cells. The grid sizes were chosen such that the data in each cell was normally distributed and the standard deviation for each was calculated. Remote Sens. 2017, 9, 283 5 of 12 Remote Sens. 2017, 9, 283 5 of 13

ManualManual tracking tracking was was also also completed completed to to validate validate th thee PIV PIV results. results. Thirty Thirty to to 40 40 points points were were chosen chosen toto measure measure displacement displacement on on each each glacier glacier by by hand hand.. These These measurements measurements were were done done directly directly on on the the pointpoint clouds clouds in in Microstation Microstation and and only only clearly clearly defin deﬁneded sets sets of of features, features, with with unique unique identifying identifying traits, traits, werewere chosen chosen for for the the manual manual tracking tracking analysis. analysis. Displacement Displacement was was measured measured either either from from peak peak to to peak peak oror trough trough to to trough, trough, depending depending on on which which had had a a high higherer density density of of LiDAR LiDAR points. points. Error Error for for the the manual manual trackingtracking results results was was calculated calculated using using the the standard standard deviation.

4.4. Results Results CommonwealthCommonwealth andand RhoneRhone Glaciers, Glaciers, in Taylorin Taylor Valley, Valley, both presentedboth presented unique unique challenges. challenges. Manual Manualtracking tracking could not could be completednot be completed at either at glacier.either glacier. At Commonwealth At Commonwealth Glacier, Glacier, which which has lower has lowerroughness roughness on average on average (0.3 m) (0.3 than m) than any ofany the of other the other glaciers glaciers examined, examined, identifying identifying features features that that can canbe usedbe used in manual in manual tracking tracking were were typically typically smaller smal thanler than the the resolution resolution ofthe of the NASA NASA data data (Figure (Figure3a). 3a).Rhone Rhone Glacier Glacier presented presented the the opposite opposite challenge challenge to manualto manual tracking, tracking, with with an an average average roughness roughness of of0.7 0.7 m m(Figure (Figure3b). 3b). Though Though Taylor Taylor Glacier Glacier has has a a higher higher surface surface roughness, roughness, withwith anan averageaverage of 1.7 m, valleysvalleys and and peaks peaks on on Rhone Rhone are are more more chaotic chaotic and and th thee lower lower resolution resolution of of the the NASA NASA data data makes makes it it difficultdifﬁcult to to identify identify the the unique unique features features that that permit permit manual manual tracking tracking with with confidence. conﬁdence. Commonwealth Commonwealth andand Rhone Rhone Glaciers Glaciers were were excluded excluded from from further further consideration consideration with with PIV PIV because because manual manual tracking tracking measurementsmeasurements for for reference reference were were not available at either site.

(a) (b)

FigureFigure 3. (a) CommonwealthCommonwealth GlacierGlacier roughness roughness from from the th 2014e 2014 NCALM NCALM data data is shown is shown in the in left the ﬁgure. left figure.The rougher The rougher (lightblue) (light area blue) indicates area indicates the extent the ofextent theglacier. of the glacier. White areasWhite contain areas contain no data no and data are andmostly are mostly located located on the valleyon the valley ﬂoor, rather floor, rather than the than glacier the glacier surface. surface. (b) Rhone (b) Rhone Glacier Glacier roughness roughness from fromthe 2014 the NCALM2014 NCALM data is data shown is onshown the right. on the The right. extent Th ofe theextent glacier of the is still glacier fairly is clearly still fairly described clearly by describedthe higher by values the higher of roughness values of (light roughness blue). (light blue).

The flow field at Canada Glacier was highly coherent closer to the glacier terminus, where The flow field at Canada Glacier was highly coherent closer to the glacier terminus, where surface roughness increased (Figure 4). However, the flow field away from the rougher terminus surface roughness increased (Figure4). However, the flow field away from the rougher terminus region, towards the smoother ice up glacier showed a higher degree of scatter. The average velocity, region, towards the smoother ice up glacier showed a higher degree of scatter. The average velocity, as determined by PIV, on Canada Glacier was 1.41 ± 0.3 m/year. Manual tracking results yielded an as determined by PIV, on Canada Glacier was 1.41 ± 0.3 m/year. Manual tracking results yielded average velocity of 1.62 ± 0.5 m/year, well within the uncertainty of the PIV results. When the outer an average velocity of 1.62 ± 0.5 m/year, well within the uncertainty of the PIV results. When 200 m of the glacier were considered separately from the center region, manual tracking produced the outer 200 m of the glacier were considered separately from the center region, manual tracking faster velocities than PIV in both regions. The manual tracking center velocity estimate, 1.75 ± 0.6 produced faster velocities than PIV in both regions. The manual tracking center velocity estimate, m/year, fell within the error of the PIV results, 1.60 ± 0.4 m/year. The results for the edge region, 0.97 1.75 ± 0.6 m/year, fell within the error of the PIV results, 1.60 ± 0.4 m/year. The results for the ± 0.1 m/year from PIV and 1.50 ± 0.3 m/year from manual tracking, did not agree within error though

Remote Sens. 2017, 9, 283 6 of 12 edgeRemote region,Sens. 20170.97, 9, 283± 0.1 m/year from PIV and 1.50 ± 0.3 m/year from manual tracking, did not agree6 of 13 within error though there were far fewer points to analyze in this region when compared to the center. Thethere results were far for fewer the Canada points Glacier to analyze and in the this other regi Tayloron when Valley compared glaciers to are the reported center. The in Table results1. for the Canada Glacier and the other Taylor Valley glaciers are reported in Table 1.

Figure 4. Canada Glacier PIV Results. Surface roughness (average(average 0.4 m) of Canada Glacier is overlaid by PIV results, averaged over 200 mm × 200200 m m cells. cells.

Table 1.1. Results for Canada, Suess, and Taylor Glaciers in Taylor Valley. AllAll refersrefers toto the entire glacier surface that was used in the PIV analysis. Edge refers to the 200 m of glacie glacierr nearest to the glacier edge and center refers toto everythingeverything otherother thanthan thethe edge.edge.

GlacierGlacier Region Region PIV PIV (m/year) (m/year) Error Error Manual Manual Tracking Tracking (m/year) (m/year) Error Error Canada All 1.41 0.3 1.62 0.5 Canada All 1.41 0.3 1.62 0.5 CanadaCanada Center Center 1.60 1.60 0.4 0.4 1.75 1.75 0.6 0.6 CanadaCanada Edge Edge 0.97 0.97 0.1 0.1 1.50 1.50 0.3 0.3 SuessSuess All All 0.86 0.86 0.3 0.3 1.25 1.25 0.7 0.7 SuessSuess Center Center 1.00 1.00 0.4 0.4 2.45 2.45 0.3 0.3 Suess Edge 0.75 0.2 1.03 0.5 TaylorSuess Edge All 0.75 1.98 0.2 0.6 1.03 3.60 0.5 1.3 TaylorTaylor CenterAll 1.98 2.03 0.6 0.9 3.60 3.68 1.3 1.0 TaylorTaylor Center Edge 2.03 1.71 0.9 0.6 3.68 3.42 1.0 1.3 Taylor Edge 1.71 0.6 3.42 1.3 The results for Suess Glacier were more coherent in areas of higher roughness and showed more scatterThe in results the smoother for Suess region Glacier in were the center more coherent of the glacier in areas (Figure of higher5). The roughness average and velocity showed at Suess more Glacierscatter in was the determined smoother region to be 0.86 in the± 0.3 center m/year, of th whiche glacier is lower (Figure than 5). the The manual average tracking velocity average at Suess of 1.25Glacier± 0.7 was m/year, determined though to the be two 0.86 values ± 0.3 m/year, agree within which the is statedlower error.than the Similar manual to the tracking results foraverage Canada of Glacier,1.25 ± 0.7 the m/year, manual though tracking the velocities two values for the agree glacier with centerin the and stated edge error. regions Similar werefaster to the than results the PIVfor velocitiesCanada Glacier, in the same the manual areas. The tracking PIV center velocities velocity, for 1.00the glacier± 0.4 m/year, center and was edge signiﬁcantly regions slowerwere faster than thethan manual the PIV tracking velocities velocity, in the 2.45same± areas.0.3 m/year. The PIV However, center velocity, there was 1.00 agreement, ± 0.4 m/year, within was error, significantly between theslower edge than velocities the manual (0.75 ±tracking0.2 m/year velocity, from 2.45 PIV, ± and0.3 m/year. 1.03 ± 0.5 However, m/year fromthere manualwas agreement, tracking). within error,Most between of Taylor the edge Glacier velocities is too smooth(0.75 ± to0.2 produce m/year from velocity PIV, results and 1.03 either ± 0.5 through m/year PIV from or manual trackingtracking). of surface features. However, the region near the terminus is sufﬁciently rough to use both methods with success (Figure6). The PIV results produced an average velocity of 1.98 ± 0.6 m/year. Manual tracking results produced an average velocity of 3.6 ± 1.3 m/year. The center and edge results

Remote Sens. 2017, 9, 283 7 of 13

Remote Sens. 2017, 9, 283 7 of 12 from manual tracking (3.68 ± 1.0 m/year and 3.42 ± 1.3 m/year, respectively) were both faster than those for PIV (2.03 ± 0.9 m/year and 1.71 ± 0.6 m/year, respectively) but, in both cases, the results agreedRemote within Sens. 2017 the, 9, error.283 7 of 13

Figure 5. Suess Glacier PIV results for 200 m × 200 m gridded data. The color scale shows the glacier surface roughness (average 0.4 m). Glacier surface velocities are higher in the center of the glacier and decrease towards the edges.

Most of Taylor Glacier is too smooth to produce velocity results either through PIV or manual tracking of surface features. However, the region near the terminus is sufficiently rough to use both methods with success (Figure 6). The PIV results produced an average velocity of 1.98 ± 0.6 m/year. Figure 5. Suess Glacier PIV results for 200 m × 200 m gridded data. The color scale shows the glacier ManualFigure tracking 5. Suess results Glacier produced PIV results an for average 200 m × velocity200 m gridded of 3.6 ± data.1.3 m/year. The color The scale center shows and theedge glacier results surface roughness (average 0.4 m). Glacier surface velocities are higher in the center of the glacier and fromsurface manual roughness tracking (average (3.68 0.4± 1.0 m). m/year Glacier and surface 3.42 velocities ± 1.3 m/year, are higher respectively) in the center were of theboth glacier faster and than decrease towards the edges. thosedecrease for PIV towards (2.03 the± 0.9 edges. m/year and 1.71 ± 0.6 m/year, respectively) but, in both cases, the results agreedMost within of Taylor the error. Glacier is too smooth to produce velocity results either through PIV or manual tracking of surface features. However, the region near the terminus is sufficiently rough to use both methods with success (Figure 6). The PIV results produced an average velocity of 1.98 ± 0.6 m/year. Manual tracking results produced an average velocity of 3.6 ± 1.3 m/year. The center and edge results from manual tracking (3.68 ± 1.0 m/year and 3.42 ± 1.3 m/year, respectively) were both faster than those for PIV (2.03 ± 0.9 m/year and 1.71 ± 0.6 m/year, respectively) but, in both cases, the results agreed within the error.

FigureFigure 6. PIV6. PIV derived derived surface surface ﬂow flowﬁeld field atat thethe TaylorTaylor Glacier terminus terminus overlaid overlaid on on a amap map of ofsurface surface roughnessroughness (average (average 1.7 1.7 m). m). The The raw raw results results werewere averagedaveraged over 200 m m ×× 200200 m m regions. regions.

The PIV results for Helheim Glacier show predominantly west to east motion on the glacier surface with a slight SE motion near the eastern boundary, which is less pronounced in the 51 min results (Figure7) than the 12 h results (Figure8). On average, PIV velocities ranged from 17.4 ± 0.5 Figure 6. PIV derived surface flow field at the Taylor Glacier terminus overlaid on a map of surface roughness (average 1.7 m). The raw results were averaged over 200 m × 200 m regions.

Remote Sens. 2017, 9, 283 8 of 13 Remote Sens. 2017, 9, 283 8 of 13 Remote Sens.The2017 PIV, 9 ,results 283 for Helheim Glacier show predominantly west to east motion on the glacier8 of 12 The PIV results for Helheim Glacier show predominantly west to east motion on the glacier surface with a slight SE motion near the eastern boundary, which is less pronounced in the 51 min surface with a slight SE motion near the eastern boundary, which is less pronounced in the 51 min results (Figure 7) than the 12 h results (Figure 8). On average, PIV velocities ranged from 17.4 ± 0.5 – results (Figure 7) than the 12 h results (Figure 8). On average, PIV velocities ranged from 17.4 ± 0.5 – –28.328.3± ±2.5 2.5 m/day m/day across the two regions and and the the two two time time frames frames examined examined (Table (Table 2).2). These These results results 28.3 ± 2.5 m/day across the two regions and the two time frames examined (Table 2). These results werewere in goodin good agreement agreement with with those those from from GPS, GPS, 20.7 20.7± ± 0.03–22.20.03–22.2 ± 0.160.16 m/day m/day (Table (Table 3),3), and and manual manual were in good agreement with those from GPS, 20.7 ± 0.03–22.2 ± 0.16 m/day (Table 3), and manual tracking,tracking, 18.6 18.6± ±3.1–33.5 3.1–33.5± ± 4.5 m/day (Table (Table 2).2). tracking, 18.6 ± 3.1–33.5 ± 4.5 m/day (Table 2).

Figure 7. Region 1 results on the left and Region 2 results on the right, both for a time interval of 51 Figure 7. Region 1 results on the left and Region 2 results on the right, both for a time interval of 51 Figureminutes. 7. Region The Helheim 1 results Glacier on the PIV left andresults Region have 2been results plotted on the on right,top of botha contour for a timeplot of interval point density,of 51 min. minutes. The Helheim Glacier PIV results have been plotted on top of a contour plot of point density, Therather Helheim than Glacierroughness. PIV It results is important have been to note plotted that point on top density of a contour is scaled plot differently of point density,for Regions rather 1 and than rather than roughness. It is important to note that point density is scaled differently for Regions 1 and roughness.2, in order It to is highlight important the to attributes note that of point each. density Raw PIV is scaledresults differentlywere averaged for Regionsat a resolution 1 and of 2, in100 order m 2, in order to highlight the attributes of each. Raw PIV results were averaged at a resolution of 100 m to highlight× 100 m. Though the attributes not shown, of each. average Raw surface PIV results roughnesses were averagedfor Regions at 1 a and resolution 2 were 0.8 of 100m and m ×1.0100 m, m. × 100 m. Though not shown, average surface roughnesses for Regions 1 and 2 were 0.8 m and 1.0 m, Thoughrespectively. not shown, average surface roughnesses for Regions 1 and 2 were 0.8 m and 1.0 m, respectively. respectively.

FigureFigure 8. Region8. Region 1 results1 results on on the the left left and and Region Region 2 2 results results on on the theright, right, bothboth forfor aa timetime interval of 12 h. Figure 8. Region 1 results on the left and Region 2 results on the right, both for a time interval of 12 hours. The Helheim Glacier PIV results have been plotted on top of a contour plot of point density, Thehours. Helheim The GlacierHelheim PIV Glacier results PIV have results been have plotted been on pl topotted of on a contourtop of a plotcontour of point plot density,of point ratherdensity, than rather than roughness. It is important to note that point density is scaled differently for Regions 1 and roughness.rather than It roughness. is important It is to important note that to point note density that point is scaleddensity differently is scaled differently for Regions for 1Regions and 2, in1 and order 2, in order to highlight the attributes of each. Raw PIV results were averaged at a resolution of 100 m to2, highlight in order theto highlight attributes the of attributes each. Raw of each. PIV resultsRaw PIV were results averaged were averaged at a resolution at a resolution of 100 mof ×100100 m m. × 100 m. Though not shown, average surface roughnesses for Regions 1 and 2 were 0.8 m and 1.0 m Though× 100 m. not Though shown, not average shown, surface average roughnesses surface roug forhnesses Regions for 1 andRegions 2 were 1 and 0.8 2 m were and 0.8 1.0 m and respectively. 1.0 m respectively. respectively. Table 2. PIV and manual tracking results for Helheim Glacier. Table 2. PIV and manual tracking results for Helheim Glacier. Table 2. PIV and manual tracking results for Helheim Glacier. GlacierGlacier Region Region ∆tΔt PIV PIV (m/day)(m/day) Error Error Manual Manual Tracking Tracking (m/day) (m/day) Error Error Glacier Region Δt PIV (m/day) Error Manual Tracking (m/day) Error Helheim 1 51 m 28.3 2.5 33.5 4.5 HelheimHelheim 11 5151 m m 28.3 28.3 2.5 2.5 33.5 33.5 4.5 4.5 HelheimHelheim 22 5151 m m 23.0 23.0 1.1 1.1 24.0 24.0 4.5 4.5 Helheim 2 51 m 23.0 1.1 24.0 4.5 HelheimHelheim 11 12 12 h h 19.8 19.8 1.5 1.5 20.3 20.3 3.4 3.4 Helheim 1 12 h 19.8 1.5 20.3 3.4 HelheimHelheim 22 12 12 h h 17.4 17.4 0.5 0.5 18.6 18.6 3.1 3.1 Helheim 2 12 h 17.4 0.5 18.6 3.1

Table 3. GPS results for Helheim Glacier.

Glacier Unit ∆t GPS (m/day) Error Helheim hg02 51 m 21.6 0.17 Helheim hg03 51 m 22.2 0.16

Helheim hg02 12 h 20.7 0.08 Helheim hg03 12 h 20.7 0.03 Remote Sens. 2017, 9, 283 9 of 12

5. Discussion The results from PIV on image pairs from four glaciers show good agreement with the results of other techniques, including manual tracking and, in the case of Helheim Glacier, GPS measurements. Overall, Canada Glacier showed the closest agreement between the manual tracking and PIV results, with an average difference of just 0.21 m/year between the two measurements. This is most likely due to the well distributed across-glacier peaks and valleys present on Canada Glacier. Importantly, these features are more present even in areas of low roughness than on either Suess or Taylor Glaciers. The manual tracking and PIV results agreed, within error, in every case except for the edge region of Canada Glacier and the center region of Suess Glacier. These two discrepancies may be the result of bias inherent in manual tracking since the method requires features to be clear and well defined in order to be accurately tracked. Very smooth or very rough regions become difficult to track manually with confidence but the PIV algorithm, which tracks brightness as a function of elevation, is able to track more subtle patterns of motion. Despite the good agreement between PIV and manual tracking overall, PIV derived velocities were consistently lower than those determined through manual tracking. Particle image velocimetry is more sensitive to smaller glacier motions than manual tracking. Glaciers do not move at one speed across their entire surface and PIV can more easily track smaller, slower features that appear when an area of motion is examined as a whole but are not significant enough to detect when looking for individual features [7,8,17]. As an additional method of verification, GPS data collected at ablation stakes on glacier surfaces can be used where they are available. Along with the GPS data collected at Helheim Glacier (Table3), GPS data are also available for Canada and Taylor Glaciers [30]. The time period represented in this data is from 1995 to 2001 so it is not contemporaneous with the LiDAR data collection but it is still a valuable point of comparison. Canada Glacier was found to be moving at an average of 0.77 ± 0.45 m/year over the whole glacier surface and at a slower rate of 0.45 ± 0.12 m/year near the terminus, which agrees well with the average and edge velocities determined using PIV. Taylor Glacier showed an average velocity of 1.0 ± 0.35 m/year with a velocity near the terminus of 1.11 ± 0.21 m/year, which also shows good agreement with the PIV results. Published averages for Taylor Glacier as a whole tend to be higher than these estimates, ranging from 0.3–25 m/year [6–8] but maps of Taylor Glacier published by Fountain et al. (2006) and Kavanaugh et al. (2009) show the highly variable nature of velocity across the glacier, with higher velocities located outside of the area included in this study. In situ data was not available for Suess Glacier. The results from PIV and manual tracking agreed for the average and edge glacier velocities but not for the center region. Suess Glacier’s surface characteristics, seen more clearly at a resolution of 100 m × 100 m (Figure9), offer some clues to this discrepancy. The SW center region has a higher roughness, and correspondingly clearer surface features, than the NE center region. There were relatively few features available for manual tracking in the NE region, which the PIV results show to be moving significantly slower than the SW center. The manual tracking results for Suess Glacier are, consequently, biased towards higher velocities. At a resolution of 100 m × 100 m, the distribution of data for Suess Glacier is still predominantly Gaussian. Though the 200 m × 200 m resolution results were used to compute the statistics shown in Table1, the finer resolution results, coupled with the glacier roughness, more fully describe the reason for the discrepancy between PIV and manual tracking velocities in the glacier center. Surface roughness and the shape of surface features both play a key role in the utility of PIV and manual tracking to analyze glacier LiDAR data. Taylor Glacier has a higher surface roughness than any of the other glaciers considered in the study, with an average surface roughness of 1.7 m. However, PIV and manual tracking both worked well on Taylor Glacier but not on Rhone Glacier, with an average surface roughness of 0.7 m. While surface roughness provides a useful first pass test for interpreting PIV results and understanding possible sources of error, the nature of the surface features themselves are also important. Features on Taylor Glacier are more easily defined in a visual analysis and they have distinct shapes in both the NASA and NCALM LiDAR point clouds. Despite having Remote Sens. 2017, 9, 283 10 of 12 lower geometric roughness, features on the surface of Rhone Glacier are more closely packed and lessRemote distinct. Sens. 2017, 9, 283 10 of 13

Figure 9. Suess Glacier PIV results forfor 100100 mm ×× 100100 m m gr griddedidded data. The color scale shows the glacier surface roughness (average(average 0.40.4 m).m). GlacierGlacier surface velocities are higher in the center of the glacier and decrease towards thethe edges.edges.

Surface roughness and the shape of surface features both play a key role in the utility of PIV and A comparison of Helheim Glacier to the Taylor Valley glaciers shows that spatial and temporal manual tracking to analyze glacier LiDAR data. Taylor Glacier has a higher surface roughness than resolution is also vital to the use of PIV and manual tracking. Spatially, on average, all three of the any of the other glaciers considered in the study, with an average surface roughness of 1.7 m. Helheim Glacier TLS scans used here have higher resolution than the Antarctica ALS scans. While However, PIV and manual tracking both worked well on Taylor Glacier but not on Rhone Glacier, the NCALM data collected in 2014 has a similar average resolution, the earlier NASA data collected with an average surface roughness of 0.7 m. While surface roughness provides a useful first pass test in 2001 has a substantially lower resolution on average (<1 point/m2 compared to ~5 points/m2). for interpreting PIV results and understanding possible sources of error, the nature of the surface This significantly limits the size of detectable features. Additionally, while large features on the Taylor features themselves are also important. Features on Taylor Glacier are more easily defined in a visual Valley glaciers remain easily recognizable, smaller features could be obscured by changes on the glacier analysis and they have distinct shapes in both the NASA and NCALM LiDAR point clouds. Despite surface. The high frequency and resolution of the Helheim Glacier TLS scans make even very small having lower geometric roughness, features on the surface of Rhone Glacier are more closely packed features trackable manually and digitally. and less distinct. 6. ConclusionsA comparison of Helheim Glacier to the Taylor Valley glaciers shows that spatial and temporal resolution is also vital to the use of PIV and manual tracking. Spatially, on average, all three of the HelheimParticle Glacier image TLS velocimetry scans used here has beenhave shownhigher resolution to be a useful than techniquethe Antarctica for analyzingALS scans. glacierWhile surfacethe NCALM velocity data using collected remotely in 2014 collected has a similar LiDAR average data. resolution, Four glaciers—Helheim, the earlier NASA Canada, data collected Suess, andin 2001 Taylor—were has a substantiallyanalyzed lower in this resolution study using on average PIV, manual (<1 point/m tracking,2 compared and, where to ~5 points/m available,2). GPS.This Strongsignificantly agreement limits was the found size of between detectable average features PIV.and Additionally, manual tracking while velocitieslarge features at all on of thethe glaciers Taylor andValley these glaciers findings remain were easily verified recognizable, using GPS smaller data at features Canada, could Taylor, be andobscured Helheim by changes Glaciers. on Most the ofglacier the glacierssurface. alsoThe high showed frequency strong and agreement resolution between of the Helheim PIV and Glacier manual TLS tracking scans velocitiesmake even when very examiningsmall features the trackable glacier center manually and edges and digitally. separately, a strong indicator that PIV can help to determine full, across-glacier velocity fields, rather than providing a few discrete data points. 6. ConclusionsSurface roughness, surface feature shapes and distribution, LiDAR point cloud density, and temporal resolution were all found to be important when applying PIV to glacier surface motion.Particle While image PIV velocimetry can be applied has been to glaciers shown that to be are a useful difficult technique to examine for analyzing using manual glacier tracking surface (Commonwealth,velocity using remotely Rhone), collected there is LiDAR no way data. to check Four that glaciers—Helheim, the results are reasonable. Canada, Suess, In addition, and Taylor— the PIV resultswere analyzed for both in Commonwealth this study using and PIV, Rhone manual Glaciers tracking, showed and, awhere great dealavailable, more GPS. scatter Strong in the agreement data than atwas any found of the between other four average glaciers PIV used and in themanual study, tracking suggesting velocities that the at same all of surface the glaciers feature and geometry these findings were verified using GPS data at Canada, Taylor, and Helheim Glaciers. Most of the glaciers also showed strong agreement between PIV and manual tracking velocities when examining the glacier center and edges separately, a strong indicator that PIV can help to determine full, across- glacier velocity fields, rather than providing a few discrete data points.

Remote Sens. 2017, 9, 283 11 of 12 that inhibits manual tracking also inhibits PIV. However, on glaciers with distinct features, PIV offers a method to rapidly and systematically collect thousands of velocity vectors, characterizing glacier motion. Ideally, PIV should be complementary to other types of glacier velocity data. Regardless though, it has the potential to significantly expand our understanding of the complex dynamics that characterize glacier flows by offering a wider, more comprehensive data distribution than other currently available techniques. Whether these techniques are used to measure glacier flow or other moving landforms, special attention needs to be paid to the time scale of data collection and the spatial scales used to rasterize and analyze LiDAR data. At the data collection stage, LiDAR data acquisitions must be temporally spaced to allow movement beyond the uncertainty of the data while still maintaining common features. For example, a rapid warming event or large volume of snowfall could significantly change the appropriate timescale for repeat measurements, since both events have the potential to occlude glacier surface features. Once data is collected, the key to rasterizing for use with PIV is to choose a scale that is larger than the uncertainty but smaller than the expected motion. Finally, the interpolation windows selected for PIV analysis should be larger than the anticipated motion so that faster moving regions will not be falsely truncated. These guidelines should be applicable to LiDAR-based PIV studies across a number of disciplines.

Acknowledgments: This work was supported in part by grants from the National Science Foundation (#1246342 and #1339015) and by the U.S. Army Engineer Research and Development Center Cold Regions Research and Engineering Laboratory Remote Sensing/GIS Center of Excellence. The data from Antarctica and Fountain’s participation in the paper was funded by NSF grant, PLR-1246342 to Portland State University. Special thanks to Preston Hartzell, who provided valuable editorial help, and to Pete Gadomski, who provided the Helheim Glacier data. Author Contributions: J.W.T. and C.G. conceived the project idea; J.W.T. carried out the work with data provided by A.G.F. and D.C.F.; A.G.F., C.G., and D.C.F. helped to develop the methods and analysis; J.W.T. wrote the paper. Conﬂicts of Interest: The authors declare no conﬂict of interest.

References

1. Cuffey, K.M.; Paterson, W.S.B. The Physics of Glaciers, 3rd ed.; Academic Press: Cambridge, MA, USA, 2010. 2. Bhardwaj, A.; Sam, L.; Bhardwaj, A.; Martín-Torres, F.J. LiDAR remote sensing of the cryosphere: Present applications and future prospects. Remote Sens. Environ. 2016, 177, 125–143. [CrossRef] 3. Favey, E.; Pateraki, M.; Baltsavias, E.P.; Bauder, A.; Bösch, H. Surface modelling for Alping glacier monitoring by Airborne Laser Scanning and Digital Photogrammetry. Int. Arch. Photogramm. Remote Sens. 2000, XXXIII, 269–277. 4. Sam, L.; Bhardwaj, A.; Singh, S.; Kumar, R. Remote sensing ﬂow velocity of debris-covered glaciers using Landsat 8 data. Prog. Phys. Geogr. 2015, 408.[CrossRef] 5. Flowers, G.E.; Jarosch, A.H.; Belliveau, P.T.A.P.; Fuhrman, L.A. Short-term velocity variations and sliding sensitivity of a slowly surging glacier. Ann. Glaciol. 2016, 57, 1–13. [CrossRef] 6. Johnston, R.R.; Fountain, A.G.; Nylen, T.H. The origin of channels on lower Taylor Glacier, McMurdo Dry Valleys, Antarctica, and their implication for water runoff. Ann. Glaciol. 2005, 40, 1–7. [CrossRef] 7. Kavanaugh, J.L.; Cuffey, K.M.; Morse, D.L.; Conway, H.; Rignot, E. Dynamics and mass balance of Taylor Glaicer, Antarctica: 1. geometry and surface velocities. J. Geophys. Res. Earth Surf. 2009, 114, 1–15. [CrossRef] 8. Fountain, A.G.; Nylen, T.H.; MacClune, K.; Dana, G.L. Glacier mass balances (1993–2001) Taylor Valley, McMurdo Dry Valleys, Antarctica. J. Glaciol. 2006, 52, 451. [CrossRef] 9. Bevan, S.L.; Luckman, A.; Khan, S.A.; Murray, T. Seasonal dynamic thinning at Helheim Glacier. Earth Planet. Sci. Lett. 2015, 415, 47–53. [CrossRef] 10. Voytenko, D.; Stern, A.; Holland, D.M.; Dixon, T.H.; Christianson, K.; Walker, R.T. Tidally driven ice speed variation at Helheim Glacier, Greenland, observed with terrestrial radar interferometry. J. Glaciol. 2015, 61, 301–308. [CrossRef] 11. Luckman, A.; Quincey, D.; Bevan, S. The potential of satellite radar interferometry and feature tracking for monitoring ﬂow rates of Himalayan glaciers. Remote Sens. Environ. 2007, 111, 172–181. [CrossRef] Remote Sens. 2017, 9, 283 12 of 12

12. Wakabayashi, H.; Nishio, F. Glacier flow estimation by SAR image correlation. In Proceedings of the 2004 IEEE International Geoscience and Remote Sensing Symposium, Anchorage, Alaska, USA, 20–24 September 2004. 13. Joerg, P.C.; Weyermann, J.; Morsdorf, F.; Zemp, M.; Schaepman, M.E. Computing of a distributed glacier surface albedo proxy using airborne laser scanning intensity data and in-situ spectro-radiometric measurements. Remote Sens. Environ. 2015, 160, 31–42. [CrossRef] 14. Joerg, P.C.; Morsdorf, F.; Zemp, M. Uncertainty assessment of multi-temporal airborne laser scanning data: A case study on an Alpine glacier. Remote Sens. Environ. 2012, 127, 118–129. [CrossRef] 15. Aryal, A.; Brooks, B.A.; Reid, M.E.; Bawden, G.W.; Pawlak, G.R. Displacement fields from point cloud data: Application of particle imaging velocimetry to landslide geodesy. J. Geophys. Res. Earth Surf. 2012, 117, 1–15. [CrossRef] 16. Prasad, A.K. Partcile Image Velocimetry. Curr. Sci. 2000, 79, 51–60. [CrossRef] 17. Gadomski, P.J. Measuring Glacier Surface Velocities with LiDAR: A Comparison of Three-Dimensional Change Detection Methods; University of Houston: Houston, TX, USA, 2016. 18. Telling, J.; Dufek, J.; Shaikh, A. Ash aggregation in explosive volcanic eruptions. Geophys. Res. Lett. 2013, 40, 2355–2360. [CrossRef] 19. Hoffman, M.J.; Fountain, A.G.; Liston, G.E. Surface energy balance and melt thresholds over 11 years at Taylor Glacier, Antarctica. J. Geophys. Res. Earth Surf. 2008, 113, 1–12. [CrossRef] 20. Doran, P.T.; Mckay, C.P.; Clow, G.D.; Dana, G.L.; Fountain, A.G.; Nylen, T.; Lyons, W.B. Valley floor climate observations from the McMurdo dry valleys, Antarctica, 1986–2000. J. Geophys. Res. 2002, 107.[CrossRef] 21. Fountain, A.G.; Levy, J.S.; Gooseff, M.N.; Van Horn, D. The McMurdo Dry Valleys: A landscape on the threshold of change. Geomorphology 2014.[CrossRef] 22. Csatho, B.; Schenk, T.; Krabill, W.; Wilson, T.; Lyons, W.; McKenzie, G.; Hallam, C.; Manizade, S.; Paulsen, T. Airborne laser scanning for high-resolution mapping of Antarctica. Eos Trans. Am. Geophys. Union 2005, 86, 237–238. [CrossRef] 23. Schenk, T.; Csatho, B.; Ahn, Y.; Yoon, T.; Shin, S.W.; Huh, K.I. DEM Generation from the Antarctic LIDAR Data; Site Report (unpublished); Ohio State University: Colombus, OH, USA, 2004. 24. 2014–2015 LIdar Survey of the McMurdo Dry Valleys, Antarctica. Available online: http://opentopo.sdsc. edu/datasetMetadata?otCollectionID=OT.112016.3294.1 (accessed on 10 June 2016). 25. Fernandez-Diaz, J.; Carter, W.; Glennie, C.; Shrestha, R.; Pan, Z.; Ekhtari, N.; Singhania, A.; Hauser, D.; Sartori, M. Capability Assessment and Performance Metrics for the Titan Multispectral Mapping Lidar. Remote Sens. 2016, 8, 936. [CrossRef] 26. Fountain, A.G.; Fernandez-Diaz, J.; Levy, J.; Obryk, M.; Gooseff, M.; Van Horn, D.; Shrestha, R. High-resolution elevation mapping of the McMurdo Dry Valleys, Antarctica and surrounding regions. Earth Syst. Sci. Data Discuss. 2017. in review. [CrossRef] 27. Thielicke, W.; Stamhuis, E.J. PIVlab—Towards user-friendly, Affordable and Accurate Digital Particle Velocimetry in MATLAB. J. Open Res. Softw. 2014, 2, 1. [CrossRef] 28. Thielicke, W.; Stamhuis, E.J. PIVlab—Time-Resolved Digital Particle Image Velocimetry Tool for MATLAB. 2014. [CrossRef] 29. Thielicke, W. The Flapping Flight of Birds—Analysis and Application; University of Groningen: Groningen, The Netherlands, 2014. 30. McMurdo Dry Valleys LTER—Glacier Stake Locations. Available online: http://www.mcmlter.org/content/ glacier-stake-locations (accessed on 5 December 2016).

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).