Article

SNOWCOVER DYNAMICS OVER A SMALL ARCTIC PROGLACIAL MORAINE: CONTRIBUTION OF PHOTOGRAMMETRY FOR SPATIO-TEMPORAL VARIABILITY OBSERVATION Jean Michel Friedt 1,§ , Eric Bernard 2,§ *, Florian Tolle 2 and Madeleine Griselin 2 1 2

*

CNRS / FEMTO-ST, University of Bourgogne Franche Comté, France CNRS / ThéMA, University of Bourgogne Franche Comté, France Correspondence: [email protected]

Academic Editor: name Version September 1, 2017 submitted to Remote Sens. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

20 21

Abstract: Current climate shift has significant impacts on both quality and quantity of snow precipitation. This directly influences spatial variability of snowcover as well as cumulative snow height. Contemporary glacier retreat reorganizes periglacial morphology: while the glacier surface decreases, the moraine area increases. It is becoming a new water storage potential almost as important as the glacier itself, but with a much more complex topography. We investigate combined Unmanned Aerial Vehicle (UAV) / Structure from Motion (SfM) image processing as a method to produce snow depth maps over the proglacial moraine area. Several UAV campaigns were carried out on a small glacial basin in Spitsbergen (Arctic): measurements were made at the maximum snow accumulation season (late April) while reference topography maps were acquired at the end of hydrological year (late September) when the moraine is mostly free of snow. Data accuracy was assessed by comparing Digital Elevation Models (DEM) generated by different SfM software. Snow depth is then determined from DEM subtraction. Using dedicated and natural ground control points for relative positioning of the DEMs, the relative DEM georeferencing with sub-meter accuracy removes the main source of uncertainty when assessing snow depth. The poor correlation between avalanche probe in-situ snow depth measurements and DEM differences is attributed to the different quantities measured: while the former only measures snow accumulation, the latter includes all ice accumulation during winter through which the probe cannot penetrate, in addition to the snow cover. Icing thicknesses are the source of the discrepency observed between avalanche probe snow cover depth measurements and difference of DEMs. Keywords: Snowcover ; Cryosphere ; Moraine ; Arctic ; SfM-UAV ; Spatial dynamics ; Photogrammetry

22

1. Introduction: snowcover dynamics as a key indicator in the cryosphere evolution

23

Crysophere dynamics are highly dependent on snowcover, which trigger further hydrological processes. Snowmelt runoff is part of fresh water fluxes reaching oceans and is thus strongly linked with snowcover spatio-temporal variability over a season ([1], [2]). Furthermore, in environments such as mountainous regions, snowcover dynamics often dominate water resource formation, storage and release ([3]). In high Arctic, year after year, while a significant glacier retreat trend is generally observed, the surface of the proglacial moraine significantly increasess in the same time ([4], [5]). Consequently, the corresponding snowcover surface also becomes wider. With respect to an glacial hydrosystem, it

24 25 26 27 28 29 30

Submitted to Remote Sens., pages 1 – 18

www.mdpi.com/journal/remotesensing

Version September 1, 2017 submitted to Remote Sens.

31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

2 of 18

now should be considered as a main contributor to outflows as well as the glacier snowcover itself ([6], [7]). However, due to the glacier forfield physical characteristics, snowcover on the moraine is much more challenging to monitor. Snow banks and massive accumulations contrast with convex area, particularly eroded by the wind or influenced by black-body effect ([8]). Indeed, the micro and local rough topography results in a high degree of seasonal and interannual variability in spatial extent ([9]). Ongoing dynamics induced by climate shift implies an increase of short events such as rain on snow ([10]), wind effects ([11]) or even heavy snowfalls ([12]). One trend of these many phenomena is that their occurence also increases since they are observed. This is very problematic in order to have an accurate monitoring of nivological processes. Especially in the case of a morainic structure, it is difficult to collect snow data that are representative of the spatial distribution of snow. Thus, remote sensing methods could be considered as an alternative or, better still, a complement to ground observations. In recent years, the use of UAV data acquisition has emerged as a well suited method for investigating geomorphologiocal changes due to climate shift ([13], [14]). This means that topographical changes as well as cryospheric processes can be observed and quantified ([15], [16], [17], [18]). Indeed, according to several works, the use of combined UAV with SfM (Structure from Motion) data processing is well suited for glacial/paraglacial environment ([19], [20], [21]), and especially in order to follow fast and short processes ([22], [23]). In addition, we showed in past works ([24]) that in Arctic, climatic conditions as well as harsh field campaigns need a flexible way for glacial/periglacial dynamics monitoring. The weather changes so fast that field campaigns has to be carried out as fast as possible to ensure both data acquisition and data homogeneity. Thus, there is all the more reason to use and apply this workflow for snowcover survey even if it was shown that photogrammetry on snow remains challanging ([25]). Here, we investigate snowcover spatial dynamics from one year to another, within a small proglacial moraine. Its ground surface covered by snow vs the same area free of snow exhibits significant differences with regard to hydrological processes as well as snowcover dynamics. That second topic is mostly discussed in this work, mainly through icings dynamics. The occurrence of icing fields has been described from several areas of Svalbard ([26], [27]). According to [28], water storage and release during the winter reflect the development of the subglacial drainage system and its capacity in the cold season. Icings fields phenomenon are well described in the litterature ([29], [30], [31], [32]). Our approach is here rather with respect to seasonnal evolutions and what triggers these dynamics. The two main purposes of this paper are as follows.

Figure 1. Two pictures, taken in April 2017 (right) and October 2016 (left), from the same spot, illustrating the snowcover implementation over the moraine topography.

64 65

1. to identify and analyse seasonnal spatial dynamics of icings over Austre Lovén proglacial moraine ;

Version September 1, 2017 submitted to Remote Sens.

66 67 68 69

3 of 18

2. to derive surface DEMs with and without snowcover in order to quantify theoric maximum snow accumulation. This point will be highlighted on areas such as river channels and icings (Fig. 1). The following step is to determinate the snow water equivalent (SWE) stock, which will fuel runoffs during melting season.

70

1.1. Study area and morphological characteritics

71

This work was carried out on a small glacial basin located on the West coast of Spitsbergen (high-Arctic), on the north side of the Brøgger peninsula (79◦ N, 12◦ E) (Fig. 2).

72

1

b

Kongsfjorden

a 2

Austre n Lové r cie gla

800m

c

d

Figure 2. (b) shows Austre Lovén glacier basin. The proglacial moraine is delimited in yellow line. The area of interest is in pale yellow, including the main water system, from the glacier outlet (red dot number 2) to the basin outlet (red dot number 1).(c) represents the moraine in Spring at its snow accumulation maximum while (d) exhibits the snow free moraine in Autumn. These 2 photos were taken from one of the highest point of the basin at app. 800 m.a.s.l. 73 74 75 76 77 78 79 80 81 82

In a 10.58 km2 basin, Austre Lovén is a small land-terminating valley and polythermal glacier. It covers an area of 4.5 km2 , with a maximum altitude of no more than 550 m.a.s.l. The proglacial moraine is today a 2.4 km2 large sedimantary complex which was formed since the Little Ice Age (LIA) period. Hence, it exhibits successive retreats of the glacier with a particular shape at the interface with the glacier snout, due to the fast retreat during the last decade. Under the glacial retreat dynamics, the proglacial moraine constantly reshapes from one year to another. This results in a complex and particularly brittle topography over a wide band at the front of the glacier. With such an heterogeneous topography coupled with a significant geomorphological and hydrological activity, the proglacial moraine is a key area for snowcover dynamics, especially in melting processes ([33], [34]) as well as in its role of water storage ([35]).

Version September 1, 2017 submitted to Remote Sens.

4 of 18

84

2. Method: use of combined Micro Unmanned Aerial Vehicule (UAV) and Structure from Motion (SfM)

85

2.1. Data acquisition

86

UAV survey was undertaken by using a low-cost, Commerical Off the Shelf (COTS) UAV DJI Phantom 3 Professional (Fig. 3). Some origine settings were kept, which means that the camera was not pre-calibrated. However, flight elevation was set at 110 m above ground, providing a spatial resolution of 5 × 5 cm pixels, considering the optical system lens properties. The camera parameters (ISO, shutter speed and focal aperture) have been set depending on the light conditions as well as the ground nature (bare stones, ice, snow). Several previous works ([36], [37], [38]) showed the interest of this kind of device’s features.

83

87 88 89 90 91 92

Figure 3. Experimental setup: natural Ground Control Points are selected for their good visibility even with the heavy Spring snow cover. (a) Their position is measured using a dual-frequency GPS receiver, or identified on a reference orthophoto. (b) A COTS UAV is used for nadir picture acquisition from an elevation of 110 m above take-off altitude. (c) Real time feedback of the camera view and GPS position of the UAV during acquisition improve safety and allow adaptating the flight path to features seen in the aerial views. Altizure (www.altizure.com) software is used for setting the raster scan flight path accounting for horizontal speed, image coverage and field of view. 93 94 95

This provides justification of its use in the context of this work. We used this quadcopter with its original camera and associated control hardware and software, using both the dedicated software (DJi GO) and dedicated mapping software (Altizure) allowing defining raster-patterned flight plannings

Version September 1, 2017 submitted to Remote Sens.

96 97 98 99 100 101 102 103 104 105 106 107 108 109

5 of 18

and storing such paths for later reproduction. This workflow is well suited as it allows to view and check flight parameters in real time. Moreover, pre-defined flights plans improve efficiency (especially in cold/wet conditions) and allow to be faithful to SfM rules (photos overlap, triggering interval). These plans could also be re-used aftwards to monitore an area according to a similar protocol of data acquisition. In the context of this work, data acquisition was made in Autumn (during the most likely snow-free moraine, beginning of October) and in Spring (late April) at the theoric maxiumum snow accumulation. These 2 periods present considerable technical difficulties. In Autumn, low lights imply to be very careful with camera settings (opening speed and ISO choice) and short measurement intervals to prevent cast shadow from inducing excessively variable observation of the same scenery during repeates passes of the UAV over the same region. In Spring, the high reflectance, the lack of structures on the smooth snow cover, and low contrasts also make photogrammetry challenging. Nevertheless, we observed protruding rocks or sastrugis to offer some usable tie points in most moraine areas.

Figure 4. Left: reference orthophoto fetched on the WMTS server of Norsk Polar Institutt (acquired in 2010), cropped to the region of interest. A water stream runs through the eastern valley East of the two hills visible in the upper central part of the region of interest. Right: associated DEM. 110 111 112 113 114 115 116 117 118 119 120

Past investigations and data analysis have demonstrated that high resolution geomorphological investigations cannot rely solely on single frequency GPS positioning of the UAV as each picture is acquired. In the current dataset, using a 2010 orthophoto provided by the Norsk Polar Institutt (available as a WMTS service at http://geodata.npolar.no/arcgis/rest/services/Basisdata/ NP_Ortofoto_Svalbard_WMTS_25833/MapServer/WMTS/1.0.0/WMTSCapabilities.xml) and the associated 5-m resolution DEM (available at https://publicdatasets.data.npolar.no/kartdata/ with 5 m resolution) as reference datasets (Fig. 4), an offset of up to 14 m was observed between the georeferenced orthophoto and the reference orthophoto (with both dataset analyzed in WGS84/UTM33N framework). Thus, fine positioning of the DEM using Ground Control Points is mandatory for assessing snow cover thickness if topographic variations are to be rejected. Two parallel processing flows were run for independent assessment of the error sources:

Version September 1, 2017 submitted to Remote Sens.

121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138

6 of 18

• In Autumn, dedicated GCPs were deployed in the moraine along the flight paths, and their position was recorded using a dual-frequency GPS receiver (Geo XH device with Zephir antenna) prior to the UAV flights. According to post processing, the accuracy obtained reached values of 15 cm for 98% of the markers, in the 3 directions (X, Y, Z). Since the GCPs were made of pink hard plastic gardening saucers (30 cm diameter) positioned on the ground, the first snow fall covered most of the surface used for positioning GCPs, even on convex topography, making them useless. • From a methodological point of view, it seems that GCPs on the ground are not the best or the most efficient solution in such an arctic context. Thus, according to previous works about this specific topic ([39]), some additional GCPs have been added: in complement to artificial GCPs, natural GCPs were selected in Spring, at maximum snow cover thickness, and measured using the dual-frequency GPS receiver. Large boulders located on upper parts of hills are assumed not to have moved and were selected, when clearly visible from ground, as GCPs. • Because GCPs positions were not collected during past flights used as reference (prior to September 2016), a third alternative solution is to locate the large boulders mentioned previously on the Norsk Polar Institutt orthophoto used as reference, and thanks to the associated high resolution DEM (as downloaded from the URL mentioned above), the three coordinates of these reference points are identified and used as GCPs.

139

2.2. Data processing

140

Photogrammetric image processing was completed using two software. On the one hand the opensource Micmac software provided by the French National Geographic Institute, available at https://github.com/micmacIGN/micmac.git with an extensive documentation at https://github. com/micmacIGN/Documentation is well suited for processing large datasets of several hundred pictures with a command line interface that does not limit the amount of processing data with a graphical user interface whose displayed is slowed down by large amount of data to display. Rather than use for embedded geographical coordinate transformation tool, all coordinate conversions from spherical to projected frameworks were processing with QGis (www.qgis.org). On the other hand, Agisoft Photoscan was used for independent assessment of the photogrammetric processing results, following the classical steps described by [40], [41], [42] in cold/glacial/paraglacial environments. The reconstruction of ground surface and objects by PhotoScan is a three-step process. For an accurate reconstruction, at least two photographs representing a single point must be available ([43]). Based on the collected geoereferenced nadir aerial pictures acquired on the field, the Micmac workflow is as follows:

141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169

1. tie points are identified on image pairs based on the GPS coordinates of the UAV at the time the pictures were acquired. While this coarse positioning is insufficient for accurate DEM positioning, it is sufficient to identify neighbouring image pairs in the raster scan flight pattern and reduce the computation time with respect to testing each image pair, even if taken too far apart for common tie points to be visible on both images. 2. the coarsely positioned orthophoto is used, over the reference orthophoto provided by the Norsk Polar Institutt, to identify and position GCPs over the area of interest. Additionnally, the associated DEM from Norsk Polar Institutt is used to identify the altitude of each selected GCP, so that all three coordinates are set for each GCP identified as relevant features on some of the UAV aerial pictures. A correlation map (Fig. 5, left) indicates areas where tie point identification and photogrammetric processing are either possible (light colors) or will induce poor results (darker shades) due to lack of usable features on the ground. 3. a second run combining camera GPS position and GCPs, the latter with lower uncertainty than the former, provides an accurately positioned orthophoto as assessed by comparing GCP position and some additional reference points that were not part of the GCPs with the reference orthophoto,

Version September 1, 2017 submitted to Remote Sens.

7 of 18

Figure 5. Left: correlation map generated by Micmac to assess the confidence in the tie point identification and the result of the photogrammetric processing. This chart results of the processing of the May 2017 data, when snow cover is maximum and usable features on the surface are sparse. Middle: resulting DEM for May 2017. Right: orthophoto resulting from processing the September 2016 dataset which is used for generating the reference DEM for assessing snow and ice volume accumulation during winter. Avalanche probe snow depth measurements are indicated as well to provide some context of the snow free underlying environment.

170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197

4. since the DEM is a byproduct of orthophoto generation, a properly positioned orthophoto guarantees a properly positioned DEM, which is much harder to assess due to the blured topographic features as opposed to the sharp structures observed on the orthophotos (Fig. 5, right). 5. DEMs acquired in Spring and Autumn are subtracted to compute snow and ice thickness, under the assumption that the underlying topography has not evolved after the freezing season has settled end of September. Micmac provides a detailed step by step workflow in which the quality of each intermediate step is assessed: tie points are identified (Tapioca tool), lens properties are identified on a subset of pictures with good contrast and significant topographic features (Tapas tool with a RadialStd lens model), the camera relative positions are computed based on photogrammetric information only (Tapas tool) and finally, after assessing visually the result of these initial calibration steps on a coarse point cloud (AperiCloud tool), the dense point cloud is computed (Malt). The associated orthophoto is assembled following topography compensation using the Tawny tool. For absolute georeferencing using GPS coordinates, CenterBascule is fed with the UAV GPS coordinates prior to the coarse and dense point cloud coordinate computations, and if GCPs have been identified, Campari is additionnaly used to combine photogrammetric, GPS and GCP results with relative uncertainties allowing for giving more weight to one dataset with respect to the others. The processing steps for identifying a few GCPs for coarse positioning of the pictures prior to a full GCP positioning has been described in detail at micmac.ensg.eu/index.php/Historical_Orthoimage which accurately summarizes the workflow we followed. The command line interface allows for handling huge datasets, with a dataset typically made of 500 to 600 pictures accounting for 3 GB worth of raw picture data. While all pictures, except oblique views collected for context of the environment at the date of the data collection, were processed with Micmac, Photoscan requires an additional step of selecting most appropriate pictures improving chances of photogrammetric processing. Indeed, for each field campaign, we recorded around 3000 pictures during multiple flight, which have been sorted following two steps. First, according to visual assesment (mainly obvious sharpness and UAV skid presence in the field of view), and secondly by using the automated processing procedure

Version September 1, 2017 submitted to Remote Sens.

25

0

25

50

8 of 18

75

100 m

14 m

Figure 6. Coarsely located orthophoto resulting from a positioning based on UAV GPS positioning as each picture was acquired (white transparent upper layer), and GCPs (numbered dots): the same features are observed on the orthophoto resulting from photogrammetric processing (GCP numbers above 10) and a reference background orthophoto, provided by Norsk Polar Institutt (GCP numbers under 10). Red dots and associated picture name tag are associated with the location at which the GPS onboard the UAV recorded its position as each photo was shot, helping select the dataset analyzed during coarse positioning of the pointcloud over the selected GCPs. The annotation “14 m” refers to the distance between the GCPs located on the orthophoto positioned solely using the GPS coordinates of the UAV (e.g. number 16), and the reference background picture (e.g. number 6).

198 199 200 201 202 203 204 205 206 207 208 209 210 211 212

provided by Photoscan. The latter step is based on quality and overlap estimation (photos with a score under 0.5 have been discarded), in order to remove blurred and under- or over-exposed images from further implemented processing and analyses. This last step represent approximately 5% of rejected photos. Hence, for each year, about 2000 photos were selected, with a full size definition. Micmac provides at the end of each processing step (lens calibration on a subset of acquired images, camera positioning for all images, conversion from arbitrary framework to georeferenced framework) information on the error resulting from photogrammetric processing. Flying at an elevation of 110 m with a DJI Phantom3 Professional UAV yields a 5-cm pixel size of ground features. We have processed orthophotos and DEMs at 1/8th of this resolution, with pixel sizes of 25 to 30 cm, yielding datasets only a few hundred megabytes large still usable under QGis for analysis. Full resolution does not add relevant information in this investigation and make the datasets much harder to handle. The camera lens property and camera positionning always yield to subpixel (typically 0.7 to 0.8, with a maximum value around 1.2 to 1.5) error after processing the several hundred picture datasets. Following the conversion from arbitrary framework to georeferenced framework, using both GPS and measured GCPs (Micmac’s GCPBascule followed by Campari tools), the resulting error

Version September 1, 2017 submitted to Remote Sens.

213 214 215 216 217

9 of 18

is again in the pixel range (one output example for the May 2017 dataset is a maximum error of 0.219468 for one picture and a mean error of 046345). Such results were achieved without freeing the camera parameters when running Campari: doing so (AllFree=1 option) did yield unacceptable errors of several pixels (7 to 10 maximum pixel error, mean error in the 2-pixel range) following the use of the GCPs. 25

100

0

100

200

300

0

25

50

75 100 m

400 m

Legend 2017050 DEM [m] 25.000000 30.480400 35.960800 41.441200 46.921600

Legend 170507dem-160925dem -3.0 -1.5 0.0 1.5 3.0

Figure 7. Left: DEM resulting from photogrammetric processing including GPS fine positioning using natural features visible on the aerial images acquired in May 2017. Right: difference of DEMs processed similarly using pictures acquired on September 25th 2016 and May 7th 2017. The snow and ice accumulation features are consistent with observations on the field and expected snowdrift accumulation next to hills and in valleys.

225

One surprising issue we met in the Spring 2017 acquisition is the erroneous tagging by the DJI Phantom3 acquisition system of negative altitudes in the picture EXIF headers. Indeed, this header includes two fields: Absolute Altitude: -17.78 and Relative Altitude: +111.40. We assume that the erroneous Absolute Altitude is associated with an erroneous initialization of the takeoff altitude. Micmac is unable to process datasets associated with negative altitudes, with the dense point cloud and orthophoto inconsistent despite proper camera calibration and positioning, as observed on the coarse pointcloud (output of AperiCloud). Replacing the Absolute Altitude field with the Relative Altitude field in the file configuring CenterBascule solved the problem.

226

2.3. Snow data extrapolation

227

Measuring snow water equivalent (SWE) requires substantially more effort than it does to only sample snow depth (HS). SWE and HS are known to be strongly correlated ([44]). This correlation could potentially be used to estimate SWE from HS even if sampling points are quiete few. Thus, studies have suggested enhancing sampling efficiency by substituting a significant part of the time-consuming SWE measurements by simple HS measurements ([45]). In our case, we carried out some snow sample measurements while flying the UAV, ensuring data acquisition at the same time. Snow samples were collected in snow pits at depths ranging from 20 to 100 cm by using 125 ml plastic bottles. Snow cover thickness was measured by using an avalanche probe. Despite varying snow conditions in various areas of the moraine, depending on the surrounding topography yielding more variable snow conditions than on the smooth glacier surface, the snow density was found to be homogeneous and constant at 0.43±0.3.

218 219 220 221 222 223 224

228 229 230 231 232 233 234 235 236 237

Version September 1, 2017 submitted to Remote Sens.

10 of 18

238

3. Snow and icings spatial dynamics

239

The analysis of ortho-images shows significant differences on icings size and distribution while the proglacial moraine is free of snow or not (Fig. 8). During the last years, firn areas get smaller or even completly desappear during the melting season. There is a dissymmetry on the icings localisation. In Autumn, remaining icings are mainly located in the deepest part of the proglacial moraine. Concerning the basin of Austre Lovén, this means that icings are essentially located on the right bank of the proglacial moraine. These old canyons concentrate most of the firn accumulation which persists over an hydrological year (i.e. October to September y+1). This situation contrasts with active periods, in Spring. Indeed, icing fields follow the stream bed of the main outlet, but including a large part of its floodplain. This results in a wider area, which evolves very quickly from one day to another. This is a point that we observed on the field and which is impossible to map, except if UAV flight sessions were made every day. If we compare images acquired in April 2017 with older data (satellite images from 2007-2009), icing fields are today less fragmented, but much wider.

240 241 242 243 244 245 246 247 248 249 250

Flight session, April 2016

NPI, 2010 400 m

a

400 m

b

Figure 8. (a) shows the localisation of residual icings (yellow surface) at the end of hydrological season. (b) exhibits active icings (orange areas) in Spring, connected with the glacier main outflow. While the same localization is observed during each Spring, icings extension has been observed to grow over the past 5 years. Yellow and violet dots are GCPs for orthophoto positioning. 251 252 253 254 255 256 257

On the active area (i.e.the main proglacial river), the shapes of icings are more complicated and elongated than in the inactive area. Moreover, while the inactive area exhibits residual icings, dynamics along the main rivers is more complex. During the melting season, the part of the icing spreading in the river channel usually melts completely. We observed that, into the proglacial moraine, flat proglacial zones favour the formation of larger icing fields. As already described, snowcover seems to play an significant role in the development of icing mounds. The water which flows out of a glacier moves in and on the snowcover until it runs out of energy in sub-zero air

Version September 1, 2017 submitted to Remote Sens.

11 of 18

266

temperatures. In the case of Autre Lovénbreen proglacial forefield, the compact structure of canyon as well as snow accumulation block the water which accumulate and and flow out under pressure. Processing DEM differences is quite challenging to apply over the whole moraine. Neverthless, to assess snowcover accumulation over time, a raster difference layer was created by subtracting the 2016 (October) and 2017 (April) DEMs. To compare snow and ice cover from one year to another, a second DEM difference was generated between October 2016 and May 2016: the same reference snow-free DEM is used in both cases. The entire area recorded was cropped to fit the area of interest. This area includes the outlet at the front of the glacier, following the main stream, up to the external moraine. This sequence represents the most rolling and changing topography.

267

4. Snow hight, SWE and further observations

268

As a reference measurement, avalanche probe snow depth measurements were carried out in the moraine during the same period the May UAV dataset was collected. Despite some fine visual correlation between the difference of DEM measurements and the avalanche probe snow depth measurements (Fig. 10), plotting the snow depth measurement with respect to the difference of DEM values at the locations where samples were collected does not show obvious correlations. This disappointing result at first is actually well explained by watching the locations at which the avalanche probe measurements were collected (Fig. 9).

258 259 260 261 262 263 264 265

269 270 271 272 273 274

5

West transect East transect

DEM difference (m)

4

3

2

1

0

0

0.2

0.4 0.6 0.8 1 1.2 avalanche probe measurement (m)

1.4

1.6

Figure 9. Comparison of the thickness measurement obtained by DEM difference with respect to avalanche probe measurement. The apparent lack of correlation is explained by the difference in the measured quantities: the avalanche probes only measures snow thickness, while the DEM difference measures both snow and ice accumulated during winter, which reaches thicknesses of several meters in some places where icings are thickest. 275 276 277 278 279 280 281 282 283 284 285

Indeed, on the northwestern part of the picture, a large depth measurement using the probe (blue dot) is well positioned on a snow drift next to a hill: this snow accumulation will show both on the DEM difference and the avalanche probe measurement. On the other hand, on the southeastern part of the picture, the snow depth measurement transect crosses the icing: while the DEM difference will exhibit a large elevation difference between Spring and Autumn measurement, the avalanche probe measured negligible snow cover thicknenss since the snow had been blown over the smooth ice surface, in addition to being transported by liquid water flow expelled by the high underground pressure even at sub-zero air temperatures, and the avalanche probe is unable to penetrate the dense ice accumulated during the winter. Hence, the visual correlation between larger snow depth measurements using the avalanche probe and visual snow drift accumulation makes us confident that the lack of correlation between the two techniques is associated with different

Version September 1, 2017 submitted to Remote Sens.

286 287 288

12 of 18

measured quantities – the snow probe penetrating solely in snow while DEM difference measuring snow and ice accumulation since Autumn – rather than the inappropriate use of DEM difference for our purpose. 25

0

25

50

75 100 m

Legend avalanche probe 0.00 - 0.15 0.15 - 0.29 0.29 - 0.44 0.44 - 0.58 0.58 - 0.73 0.73 - 0.88 0.88 - 1.02 1.02 - 1.17 1.17 - 1.31 1.31 - 1.46

Figure 10. Right: difference of DEMs between Septembre 25th 2016 and May 7th 2017, assumed to be representative of winter snow and ice accumulation. Dots represent snow depth survey measurements using an avalanche probe. The red dot at the center of the picture where rock is visible has been set as a 0-elevation differenc by adding 3 m to the DEM difference. Left: avalanche probe snow depth measurements in the context of the orthophoto acquired simultaneously. The western transect exhibits thicker snow cover measurements in areas consistent with snow drift and accumulation, while the eastern transect crosses the icings where snow cannot accumulate and difference of DEMs detects strong variations due to ice accumulation during winter yet negligible snow cover. 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304

One of the main topics of stuyding a small glacial basin is to better understand melting processes and their interaction with climate. As was demonstrated [29], icing fields constitute a very important element of the cryosphere in the High Arctic as a witness of thermal transformation of the glacier, and thus indirectly on the climate’s change impact. About icings dynamics, the seasonnal approach described in this work will need to be extend to several years to better understand how its mechansims are influenced by climate. Nevertheless, inter-seasonnal observation gave quite a few lessons. First, the presence or absence of icings indicates changes in the functioning of the proglacial moraine internal drainage system. Obviously, it appears that icings are not located in the same area in Spring and in Autumn. But the important point is that in Autumn, there is no significant dynamics recorded contrary to Spring where changes can be observed from an hour to another. By the way, the spatio-temporal scale at which processes are carrying out is too fast to be measurable, even by usind UAV survey. It was quite easy to observe the fast formation of massive icing mounds which raises questions about the icings dynamics. In Autumn the absence of any movement could be attributed to the fact that these icings are no longer in activity nor supplied by water outflows. According to [46], this means, in the case of Austre Lovén proglacial moraine, that almost all icings are associated with rivers or glacial water outflows but are not connected with groundwater outflows. This conclusion

Version September 1, 2017 submitted to Remote Sens.

305 306 307 308 309 310 311 312 313 314

13 of 18

is supported by Spring observations, which clearly indicate the strong relationship between outflows and icings. In addition to comparing the difference of DEM result with the avalanche probe measurement, we assess the evolution of the snow cover during two successive years. Taking the same reference DEM acquired in September 2016, we generate a second difference of DEMs with respect to the measurement obtained in April 2016. While no systematic avalanche probe measurements are available for this date, we assume that some correlation should be observed between icing distribution and snow cover from one year to another, considering the underlying topography was not observed to have significantly evolved in this part of the moraine and average aerological conditions to be reproducible over the years (Fig. 11).

Figure 11. Right: difference of DEMs between April 2016 and September 2016. Here again the red dot at the center of the picture, where rock is visible and no snow has accumulated, is used to set the 0-m offset altitude. Left: associated orthophoto. 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329

General snow accumulation on the western part of the area under investigation and icing distribution on the eastern part is consistent with the results obtained from DEM differences in September 2017. Similarly, snow accumulation on the areas north of the red reference point is observed in both datasets. Such results are also observed visually on pictures acquired on successive seasons from the same spot (Fig. 12). Moreover, difference of DEMs applied on snowcover shows the importance of ground topography. Indeed, Fig. 12 highlights the smooth effect (images below) when landforms exhibit very small differences. In that case, furthermore when wind effect is added, the estimlation of snow depth remains challenging while high variations are in the same scale as uncertainty. On the contrary, when the landforms are quite sharp (images above), even a strong wind effect can’t smooth the surface. Indeed, a rugged topography promotes cornices formation. This is a much more easy configuration for snow depth estimation since: 1. snowcover is quite deep and so easier to estimate by using photogrammetry ; 2. during data processing step, cornices create shadows and structures that are identifiable by processing algorithm.

Version September 1, 2017 submitted to Remote Sens.

14 of 18

Figure 12. Pictures taken from the same spot in April 2017 (right) and in September 2016 (left), towards the glacier (top) or the fjord (bottom), illustrating the smoothing effect of snow accumulation and hinting at the consistency of the observed ice and snow thicknesses quantitatively deduced from difference of DEMs.

341

The estimation of snowcover data is challenging by using SfM photogrammetry. However, the measurements performed on the proglacial moraine assessed that there are strong snow drift effects. Regardless of snow accumulation, it appears that morainic mounds evolve very little, contrary to canyons, that are constantly re-shaping and subject to strong melting processes that consequently dig under sediment transport action. Thus, the structure of the topography promotes massive snow accumulation as well as the orientation (orthogonal to the dominant winds). A lesson learnt while studying snowcover in the moraine, is that the comparison with the glacier snowcover remains random. Previous works showed that on the glacier a simple interpolation can be applied to estimate both SWE and height. In the case of the proglacial moraine, it is at the moment impossible to reach this goal. As often observed, the moraine constitutes a key area but still hard to monitore. Based on this paper and previous works, the coupling of LiDAR measurements as references, and several photogrammetric flight session appears as the most efficient method.

342

5. Conclusions

343

Two years of snow cover in an Arctic proglacial moraine area were investigated using difference of Digital Elevation Models, refering to the snow-free dataset acquired in Autumn. While spatial correlation is observed with respect to avalanche probe measurements in areas where snow accumulation over bare moraine rock is significant, the poor general corelation between in-situ measurement and remote-sensing techniques is attributed to the ice accumulation underlying the snowcover. This result is most striking in icings areas. Fine digital elevation model registration for snow cover thickness estimate requires ground-based control points. When lacking artificial reference points, natural ground control points were here used to register past and present acquisitions, refereing to large boulders clearly visible even at maximum snow cover and known not to have

330 331 332 333 334 335 336 337 338 339 340

344 345 346 347 348 349 350 351

Version September 1, 2017 submitted to Remote Sens.

352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377

15 of 18

moved in the last 7 years with respect to the reference orthophoto. Despite poor contrast under homogeneous snow cover conditions, Structure from Motion photogrammetric analysis appears suitable for mapping snow cover distribution even in the low-lying sun, cast shadow met in Arctic environments. Mapping a 2.4 km2 area proglacial moraine snow cover characteristics appears beyond the reach of a rotating wing quad-copter UAV: estimate SWE for the whole moraine is not possible with the current dataset acquired over multiple flight sessions due to the limited (20 minutes at most) autonomy. We conclude that a rotating wing UAV quadcopter is not suitable for such a large area. A fixed wing UAV seems to be a better suited solution as demonstrated by [20] in which a 5 km2 tongue of a glacier was mapped, an area similar to the one under investigation here, through flights spanning about 0.35 km2 each, an area about 1.5 to two times larger than those covered during our rotating wing UAV flights. Despite similar flight elevation and adjacent image coverage, their flight duration at 2500 m.a.s.l is about twice the one we met in Arctic conditions of close or sub-zero temperatures at sea level (15 minute flight durations for the DJI Phantom 3). In addition, combining SfM methods with satellite RADAR images analysis will open new opportunities for snowcover study in harsh condition as well as in rough topographic environement, thanks to the high resolution DEM generated by the former technique needed for interferometric analysis of the latter. Despite the poorer RADAR spatial resolution (5 m for Sentinel 1) and high operating frequency (C-SAR at 5.4 GHz or a 5.5 cm wavelength) inducing more complex interaction of the electromagnetic wave with the snow cover than an optical signal, such a technique [47,48] appears worth investigating in complement with DEM generated by UAV. Acknowledgments: This study was supported by a Franche Comté county grant. We akcknowledge IPEV for logistics support in Svalbard as well as J.-P. Culas at Photocoptère S.A.S (Besançon, France) for technical support on UAV maintenance and regulations. JMF is indebted to Luc Girod for identifying the altitude tag issue, as discussed in the excellent online forum of Micmac users (http://forum-micmac.forumprod.com/ inconsistent-dem-v-s-orhophoto-t1462.html).

381

Author Contributions: E. Bernard and J.M. Friedt conceived of and designed the experiments and analyzed the results. They both processed and analyzed the data and wrote the manuscript. F. Tolle assisted during field trips and provided initial funding for this investigation. M. Griselin shared her field experience and performed the snow cover thickness measurements.

382

Conflicts of Interest: The authors declare no conflict of interest.

383

Abbreviations

384

The following abbreviations are used in this manuscript:

378 379 380

385 386

COTS SfM UAV

387

Bibliography

388

1.

389 390

2.

391 392

3.

393 394 395 396 397

4.

Commercial Off the Shelf Structure from Motion Unmanned Aerial Vehicle

Nuth, C.; Moholdt, G.; Kohler, J.; Hagen, J.O.; Kääb, A. Svalbard glacier elevation changes and contribution to sea level rise. Journal of Geophysical Research 2010, 115, F01008. Hock, R. Glacier melt: a review of processes and their modelling. Progress in Physical Geography 2005, 29, 362–391. Lehning, M.; Ingo, V.; Gustafsson, D.; Nguyen, T.A.; St, M.; Zappa, M. ALPINE3D : a detailed model of mountain surface processes and its application to snow hydrology. Hydrological Processes 2006, 20, 2111–2128. Bernard, É.; Friedt, J.M.; Tolle, F.; Marlin, C.; Griselin, M. Using a small COTS UAV to quantify moraine dynamics induced by climate shift in Arctic environments. International Journal of Remote Sensing 2016, 0, 1–15.

Version September 1, 2017 submitted to Remote Sens.

398

5.

399 400 401

6.

402 403 404

7.

405 406 407

8.

408 409

9.

410 411

10.

412 413 414

11.

415 416 417 418

12.

419 420

13.

421 422 423

14.

424 425

15.

426 427

16.

428 429 430

17.

431 432 433

18.

434 435 436

19.

437 438

20.

439 440

21.

441 442 443

22.

444 445

23.

446 447 448 449

24.

16 of 18

Tonkin, T.N.; Midgley, N.G.; Cook, S.J.; Graham, D.J. Geomorphology Ice-cored moraine degradation mapped and quanti fi ed using an unmanned aerial vehicle : A case study from a polythermal glacier in Svalbard. Geomorphology 2016, 258, 1–10. Etzelmüller, B.; Ødegård, R.S.; Vatne, G.; Mysterud, R.S.; Tonning, T.; Sollid, J.L. Glacier characteristics and sediment transfer system of Longyearbreen and Larsbreen , western Spitsbergen. Norwegian Journal of Geography 2010, February 2013, 37–41. Martín-Español, a.; Navarro, F.; Otero, J.; Lapazaran, J.; Błaszczyk, M. Estimate of the total volume of Svalbard glaciers, and their potential contribution to sea-level rise, using new regionally based scaling relationships. Journal of Glaciology 2015, 61, 29–41. Barr, I.D.; Lovell, H. A review of topographic controls on moraine distribution. Geomorphology 2014, 226, 44–64. Derksen, C.; LeDrew, E. Variability and change in terrestrial snow cover: data acquisition and links to the atmosphere. Progress in Physical Geography 2000, 24, 469–498. Eckerstorfer, M.; Christiansen, H.H. Meteorology, Topography and Snowpack Conditions causing Two Extreme Mid-Winter Slush and Wet Slab Avalanche Periods in High Arctic Maritime Svalbard. Permafrost and Periglacial Processes 2012, 23, 15–25. Svendsen, H.; Beszczynska-Møller, A.; Hagen, J.O.; Lefauconnier, B.; Tverberg, V.; Gerland, S.; Ørbøk, J.B.; Bischof, K.; Papucci, C.; Zajaczkowski, M.; Azzolini, R.; Bruland, O.; Wiencke, C.; Winther, J.; Dallmann, W. The physical environment of Kongsfjorden–Krossfjorden, an Arctic fjord system in Svalbard. Polar Research 2002, 21, 133–166. Bednorz, E. Occurrence of winter air temperature extremes in Central Spitsbergen. Theoretical and Applied Climatology 2011, 106, 547–556. Stumpf, a.; Malet, J.P.; Allemand, P.; Pierrot-Deseilligny, M.; Skupinski, G. Ground-based multi-view photogrammetry for the monitoring of landslide deformation and erosion. Geomorphology 2015, 231, 130–145. Cook, K.L. An evaluation of the effectiveness of low-cost UAVs and structure from motion for geomorphic change detection. Geomorphology 2017, 278, 195–208. Bhardwaj, A.; Sam, L.; Akanksha.; Martín-Torres, F.J.; Kumar, R. UAVs as remote sensing platform in glaciology: Present applications and future prospects. Remote Sensing of Environment 2016, 175, 196–204. Fonstad, M.A.; Dietrich, J.T.; Courville, B.C.; Jensen, J.L.; Carbonneau, P.E. Topographic structure from motion: A new development in photogrammetric measurement. Earth Surface Processes and Landforms 2013, 38, 421–430. Lucieer, a.; Jong, S.M.D.; Turner, D. Mapping landslide displacements using Structure from Motion (SfM) and image correlation of multi-temporal UAV photography. Progress in Physical Geography 2013, 38, 97–116. Westoby, M.J.; Dunning, S.A.; Woodward, J.; Hein, A.S.; Marrero, S.M.; Winter, K.; Sugden, D.E. Instruments and methods: Sedimentological characterization of Antarctic moraines using uavs and Structure-from-Motion photogrammetry. Journal of Glaciology 2015, 61, 1088–1102. Nolan, M.; Larsen, C.; Sturm, M. Mapping snow depth from manned aircraft on landscape scales at centimeter resolution using structure-from-motion photogrammetry. Cryosphere 2015, 9, 1445–1463. Gindraux, S.; Boesch, R.; Farinotti, D. Accuracy assessment of digital surface models from Unmanned Aerial Vehicles’ imagery on glaciers. Remote Sensing 2017, 9, 1–15. Ryan, J.C.; Hubbard, A.L.; Box, J.E.; Todd, J.; Christoffersen, P.; Carr, J.R.; Holt, T.O.; Snooke, N. UAV photogrammetry and structure from motion to assess calving dynamics at Store Glacier, a large outlet draining the Greenland ice sheet. Cryosphere 2015, 9, 1–11. Smith, M.W.; Carrivick, J.L.; Quincey, D.J. Structure from motion photogrammetry in physical geography. Progress in Physical Geography 2015, 40, 247–275. Eltner, A.; Kaiser, A.; Castillo, C.; Rock, G.; Neugirg, F.; Abellán, A. Image-based surface reconstruction in geomorphometry-merits, limits and developments. Earth Surface Dynamics 2016, 4, 359–389. Bernard, E.; Friedt, J.M.; Prokop, A.; Tolle, F.; Marlin, C.; Griselin, M. Investigating snowpack volumes and icing dynamics in the moraine of an Arctic catchment using UAV/photogrammetry. The Photogrammetric Record accepted.

Version September 1, 2017 submitted to Remote Sens.

450

25.

451 452

26.

453 454 455

27.

456 457

28.

458 459

29.

460 461

30.

462 463

31.

464 465

32.

466 467

33.

468 469 470

34.

471 472 473

35.

474 475 476

36.

477 478

37.

479 480 481

38.

482 483

39.

484 485 486

40.

487 488

41.

489 490

42.

491 492

43.

493 494

44.

495 496

45.

497 498

46.

499 500 501

47.

17 of 18

Jagt, B.; Lucieer, A.; Wallace, L.; Turner, D.; Durand, M. Snow Depth Retrieval with UAS Using Photogrammetric Techniques. Geosciences 2015, 5, 264–285. Bennett, G.L.; Evans, D.J.A. Glacier retreat and landform production on an overdeepened glacier foreland: The debris-charged glacial landsystem at Kvíárjökull, Iceland. Earth Surface Processes and Landforms 2012, 37, 1584–1602. Sobota, I. Icings and their role as an important element of the cryosphere in High Arctic glacier forefields. Bulletin of Geography, Physical Geography series 2016, 10, 81–93. Bukowska-Jania, E.; Szafraniec, J. Distribution and morphometric characteristics of icing fields in Svalbard. Polar Research 2005, 24, 41–53. Bukowska-Jania, E. The role of glacier system in migration of calcium carbonate on Svalbard. Polish Polar Research 2007, 28, 137–155. Evans, D.J.A. Controlled moraines: origins, characteristics and palaeoglaciological implications. Quaternary Science Reviews 2009, 28, 183–208. Rutter, N.; Hodson, A.; Irvine-Fynn, T.; Solås, M.K. Hydrology and hydrochemistry of a deglaciating high-Arctic catchment, Svalbard. Journal of Hydrology 2011, 410, 39–50. Bælum, K.; Benn, D.I. Thermal structure and drainage system of a small valley glacier (Tellbreen, Svalbard), investigated by ground penetrating radar. The Cryosphere 2011, 5, 139–149. Bernard Éric.; Florian, T.; Michel, F.J.; Christelle, M.; Madeleine, G. How short warm events disrupt snowcover dynamics Example of a polar basin – Spitsberg, 79◦ N. ISSW Proceedings; IRSTEA: Grenoble, 2013. Ewertowski, M.W.; Tomczyk, A.M. Quantification of the ice-cored moraines’ short-term dynamics in the high-Arctic glaciers Ebbabreen and Ragnarbreen, Petuniabukta, Svalbard. Geomorphology 2015, 234, 211–227. Wittmeier, H.E.; Bakke, J.; Vasskog, K.; Trachsel, M. Reconstructing Holocene glacier activity at Langfjordjøkelen, Arctic Norway, using multi-proxy fingerprinting of distal glacier-fed lake sediments. Quaternary Science Reviews 2015, 114, 78–99. Westoby, M.; Brasington, J.; Glasser, N.; Hambrey, M.; Reynolds, J. ‘Structure-from-Motion’ photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology 2012, 179, 300–314. Lucieer, A.; Turner, D.; King, D.H.; Robinson, S.a. Using an Unmanned Aerial Vehicle (UAV) to capture micro-topography of Antarctic moss beds. International Journal of Applied Earth Observation and Geoinformation 2014, 27, 53–62. Colomina, I.; Molina, P. Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing 2014, 92, 79–97. Krsak, B.; Blistan, P.; Paulikov??, A.; Pu??k??rov??, P.; Kovani??, L.; Palkov??, J.; Zeliz??akov??, V. Use of low-cost UAV photogrammetry to analyze the accuracy of a digital elevation model in a case study. Measurement: Journal of the International Measurement Confederation 2016, 91, 276–287. Dietrich, J.T. Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry. Geomorphology 2016, 252, 144–157. Uysal, M.; Toprak, A.S.; Polat, N. DEM generation with UAV Photogrammetry and accuracy analysis in Sahitler hill. Measurement: Journal of the International Measurement Confederation 2015, 73, 539–543. Cook, K.L.; Turowski, J.M.; Hovius, N. A demonstration of the importance of bedload transport for fluvial bedrock erosion and knickpoint propagation. Earth Surface Processes and Landforms 2013, 38, 683–695. Harwin, S.; Lucieer, A.; Osborn, J. The impact of the calibration method on the accuracy of point clouds derived using unmanned aerial vehicle multi-view stereopsis. Remote Sensing 2015, 7, 11933–11953. Deems, J.S.; Painter, T.H.; Finnegan, D.C. Lidar measurement of snow depth: A review. Journal of Glaciology 2013, 59, 467–479. Debeer, C.M.; Pomeroy, J.W. Modelling snow melt and snowcover depletion in a small alpine cirque , Canadian Rocky Mountains. Hydrological Processes 2009, 2599, 2584–2599. Dietz, A.J.; Kuenzer, C.; Gessner, U. International Journal of Remote Remote sensing of snow – a review of available methods. International Journal of Remote Sensing 2013, 33, 37–41. Nagler, T.; Rott, H.; Glendinning, G. Snowmelt Runoff Modelling by Means of RADARSAT and ERS SAR. Canadian Journal of Remote Sensing 2000, 26.

Version September 1, 2017 submitted to Remote Sens.

502 503

504 505 506

48.

18 of 18

Luojus, K.; Pulliainen, J.; Metsamaki, S.; Hallikainen, M. Snow-Covered Area Estimation Using Satellite Radar Wide-Swath Images. IEEE Transactions on Geoscience and Remote Sensing 2007, 45.

c 2017 by the authors.

Submitted to Remote Sens. for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).