combination of uav and terrestrial photogrammetry to assess … · combination of uav and...

Nat Hazards Earth Syst Sci 18 1055ndash1071 2018httpsdoiorg105194nhess-18-1055-2018copy Author(s) 2018 This work is distributed underthe Creative Commons Attribution 30 License

Combination of UAV and terrestrial photogrammetry toassess rapid glacier evolution and map glacier hazardsDavide Fugazza1 Marco Scaioni2 Manuel Corti2 Carlo DrsquoAgata3 Roberto Sergio Azzoni3 Massimo Cernuschi4Claudio Smiraglia1 and Guglielmina Adele Diolaiuti31Department of Earth Sciences ldquoA Desiordquo Universitagrave degli Studi di Milano 20133 Milan Italy2Department of Architecture Built Environment and Construction Engineering Politecnico di Milano 20133 Milan Italy3Department of Environmental Science And Policy Universitagrave degli Studi di Milano 20133 Milan Italy4Agricola 2000 SCPA 20067 Tribiano (MI) Italy

Correspondence Davide Fugazza (davidefugazzaunimiit) and Marco Scaioni (marcoscaionipolimiit)

Received 1 June 2017 ndash Discussion started 6 June 2017Revised 6 March 2018 ndash Accepted 10 March 2018 ndash Published 5 April 2018

Abstract Tourists and hikers visiting glaciers all year roundface hazards such as sudden terminus collapses typical ofsuch a dynamically evolving environment In this study weanalyzed the potential of different survey techniques to an-alyze hazards of the Forni Glacier an important geosite lo-cated in Stelvio Park (Italian Alps) We carried out surveysin the 2016 ablation season and compared point clouds gen-erated from an unmanned aerial vehicle (UAV) survey close-range photogrammetry and terrestrial laser scanning (TLS)To investigate the evolution of glacier hazards and evalu-ate the glacier thinning rate we also used UAV data col-lected in 2014 and a digital elevation model (DEM) cre-ated from an aerial photogrammetric survey of 2007 Wefound that the integration between terrestrial and UAV pho-togrammetry is ideal for mapping hazards related to theglacier collapse while TLS is affected by occlusions andis logistically complex in glacial terrain Photogrammetrictechniques can therefore replace TLS for glacier studiesand UAV-based DEMs hold potential for becoming a stan-dard tool in the investigation of glacier thickness changesBased on our data sets an increase in the size of collapseswas found over the study period and the glacier thinningrates went from 455plusmn 024 m aminus1 between 2007 and 2014to 520plusmn 111 m aminus1 between 2014 and 2016

1 Introduction

Glacier- and permafrost-related hazards can be a seriousthreat to humans and infrastructure in high mountain regions(Carey et al 2014) The most catastrophic cryospheric haz-ards are generally related to water outbursts either throughbreaching of moraine- or ice-dammed lakes or from theenglacial or subglacial system causing floods and debrisflows Ice avalanches from hanging glaciers (Vincent et al2015) and debris flows caused by the mobilization of accu-mulated loose sediment on steep slopes (Kaab et al 2005a)can also have serious consequences for downstream popu-lations Less severe hazards but still particularly threaten-ing for mountaineers are the detachment of seracs (Riccardiet al 2010) or the collapse of ice cavities (Gagliardini etal 2011 Azzoni et al 2017) While these processes are inpart typical of glacial and periglacial environments there isevidence that climate change is increasing the likelihood ofspecific hazards (Kaab et al 2005a) In the European Alpsaccelerated formation and growth of proglacial moraine-dammed lakes has been reported in Switzerland amongstconcern of possible overtopping of moraine dams provokedby ice avalanches (Gobiet et al 2014) Ice avalanches them-selves can be more frequent as basal sliding is enhanced bythe abundance of meltwater in warmer summers (Clague2013) Glacier and permafrost retreat which has been re-ported in all sectors of the Alps (Smiraglia et al 2015 Fis-cher et al 2014 Gardent et al 2014 Harris et al 2009) is amajor cause of slope instabilities which can result in debrisflows by debuttressing rock and debris flanks and promot-

Published by Copernicus Publications on behalf of the European Geosciences Union

1056 D Fugazza et al Combination of UAV and terrestrial photogrammetry to assess rapid glacier evolution

ing the exposure of unconsolidated and ice-cored sediments(Keiler et al 2010 Chiarle et al 2007) Glacier downwast-ing causes changes in water resources with an initial increasein discharge due to enhanced melt followed by a long-termreduction affecting drinking water supply irrigation and hy-dropower production (Kaab et al 2005b) along with a risingoccurrence of structural collapses (Azzoni et al 2017) Fi-nally glacier retreat and the increase in glacier hazards bothnegatively influence the tourism sector and the economicprosperity of high mountain regions (Palomo 2017)

The growing threat from cryospheric hazards under cli-mate change calls for the adoption of mitigation strategiesRemote sensing has long been recognized as an importanttool for producing supporting data for this purpose such asdigital elevation models (DEMs) and multispectral imagesDEMs are particularly useful for detecting glacier thicknessand volume variations (Fischer et al 2015 Berthier et al2016) and for identifying steep areas that are most prone togeomorphodynamic changes such as mass movements (Bla-sone et al 2015) Multispectral images at a sufficient spatialresolution make it possible to recognize most cryospherichazards (Quincey et al 2005 Kaab et al 2005b) Whilesatellite images from Landsat and ASTER sensors (15ndash30 mground sample distance ndash GSD) are practical for regional-scale mapping (Rounce et al 2017) the assessment of haz-ards at the scale of individual glaciers or basins requires ahigher spatial resolution which in the past could only beachieved via aerial laser scanner and photogrammetric sur-veys (Vincent et al 2010 Janke 2013) or dedicated fieldcampaigns with terrestrial laser scanners (TLS) (Kellerer-Pirklbauer et al 2005 Riccardi et al 2010) Recent yearshave seen a resurgence of terrestrial photogrammetric sur-veys for the generation of DEMs (Piermattei et al 20152016 Kaufmann and Seier 2016) due to important tech-nological advances including the development of structure-from-motion (SfM) photogrammetry and its implementationin fully automatic processing software as well as improve-ments in the quality of camera sensors (Eltner et al 2016Westoby et al 2012) In parallel unmanned aerial vehicles(UAVs ndash Colomina and Molina 2014 OrsquoConnor et al 2017)have started to emerge as a viable alternative to TLS formultitemporal monitoring of small areas UAVs promise tobridge the gap between field observations notoriously diffi-cult on glaciers and coarser-resolution satellite data (Bhard-waj et al 2016) Although the number of studies employingthese platforms in high mountain environments is on the rise(see eg Fugazza et al 2015 Gindraux et al 2017 Seier etal 2017) their full potential for monitoring glaciers and par-ticularly glacier hazards has yet to be explored In particularthe advantages of UAV and terrestrial SfM photogrammetryand the possibility of data fusion and volume change esti-mation to support hazard management strategies in glacialenvironments need to be investigated and assessed

In this study we investigated a rapidly downwastingglacier (almost 5 m aminus1 water equivalent Senese et al 2012)in a protected area and highly touristic sector of the Ital-ian Alps Stelvio National Park We focused on the glacierterminus and the hazards identified there ie the formationof normal faults and ring faults The former occur mainlyon the medial moraines and glacier terminus and are due togravitational collapse of debris-laden slopes The latter de-velop as a series of circular or semicircular fractures withstepwise subsidence caused by englacial or subglacial melt-water creating voids at the icendashbedrock interface eventuallyleading to the collapse of the cavity roofs While often over-looked these collapse structures are particularly hazardousfor mountaineers and they are likely to increase under a cli-mate change scenario (Azzoni et al 2017) They are moredangerous than crevasses because of their larger size

We conducted our first UAV survey of the glacier in 2014in the summer of 2016 the glacier was surveyed using threedifferent techniques for the generation of point clouds DEMsand orthophotos The aims were (1) to compare the differentmethods and select the most appropriate ones for monitor-ing glacier hazards (2) to identify glacier-related hazards andtheir evolution between 2014 and 2016 and (3) to investigatechanges in ice thickness between 2014 and 2016 and between2007 and 2016 by comparing the two UAV DEMs and a thirdDEM obtained from stereo-processing of aerial photos cap-tured in 2007

2 Study area

The Forni Glacier (see Fig 1) has an area of 1134 km2 basedon the 2007 data from the Italian Glacier Inventory (Smi-raglia et al 2015) an altitudinal range between 2501 and3673 m asl and a north-northwesterly aspect The glacierhas retreated markedly since the Little Ice Age when itsarea was 1780 km2 (Diolaiuti and Smiraglia 2010) withan acceleration of the shrinkage rate over the past threedecades typical of valley glaciers in the Alps (Diolaiuti et al2012 DrsquoAgata et al 2014) It has also undergone profoundchanges in dynamics in recent years such as the loss of iceflow from the eastern accumulation basin towards its tongueand the evidence of collapsing areas on the eastern tongue(see Fig 2d Azzoni et al 2017) Continuous monitoring ofthese hazards is important as the site is highly touristic (Gar-avaglia et al 2012) The glacier is in fact frequently visitedduring both summer and winter months During the summerhikers heading to Mount San Matteo take the trail along thecentral tongue accessing the glacier through the left flank ofthe collapsing glacier terminus (see Fig 2b and c) Duringwintertime ski mountaineers instead access the glacier fromthe eastern side crossing the medial moraine and potentiallycollapsed areas there (see Figs 1 and 2a)

Nat Hazards Earth Syst Sci 18 1055ndash1071 2018 wwwnat-hazards-earth-syst-scinet1810552018

D Fugazza et al Combination of UAV and terrestrial photogrammetry to assess rapid glacier evolution 1057

Figure 1 The tongue of Forni Glacier The map shows the locationof take-off and landing sites for the 2014 and 2016 UAV surveysstandpoint of TLS survey GCPs used in the UAV photogramme-try surveys and trails crossing the glaciers Letters andashe identify thelocation of features described in Fig 2 Base map from 2015 cour-tesy of IIT Regione Lombardia WMS Service Trails from Kom-pass online cartography at httpswwwkompass-1039italiaitinfomappa-online

3 Data sources acquisition and processing

In this study we took advantage of a UAV survey performedin 2014 (Fugazza et al 2015) Then through a field cam-paign in 2016 we conducted different surveys using a UAVterrestrial photogrammetry and TLS In the 2014 UAV sur-vey no ground control points (GCPs) were collected whilein 2016 we specifically set up a control network for georef-erencing purposes Processing of the UAV and terrestrial im-ages was carried out using Agisoft Photoscan version 124(httpwwwagisoftcom) implementing a SfM algorithm forimage orientation followed by a multi-view dense-matchingapproach for surface 3-D reconstruction (Westoby et al2012) In addition we employed a DEM from an aerial sur-vey of 2007 to calculate glacier thickness changes over a pe-riod of 7 to 9 years

31 UAV photogrammetry

311 2014 dataset

The first UAV survey took place on 28 August 2014 us-ing a SwingletCam fixed wing aircraft (see Fig 3a) Thiscommercial platform developed by SenseFly carries a CanonIxus 127 HS compact digital camera The UAV was flown inautopilot mode with a relative flying height of approximately380 m above the glacier surface which resulted in an averageGSD of 12 cm The flight plan was organized by using theproprietary software eMotion by which the aircraft followspredefined waypoints with a nominal along-strip overlap of70 In our study side lap was not regular because of thevarying surface topography but it averaged approximately60 Flight operations started around 0730 LT and endedaround 0830 LT Early-morning operations were preferredto avoid saturating camera pictures as during this time ofday the glacier is not yet directly illuminated by the sun andto minimize blurring effects due to the UAV motion sincewind speed is at its lowest on glaciers during morning hours(Fugazza et al 2015) Pictures were automatically capturedby the UAV platform selecting the best combination of sen-sor aperture (F27) sensitivity (between 100 and 400 ISO)and shutter speed (between 1125 and 1640 s) The surveycovered an area of 221 km2 in two flight campaigns with alow-altitude take-off (see Fig 1) Both the terminal parts ofthe central and eastern ablation tongue were surveyed

Since no GCPs were measured during the 2014 campaignthe registration of this data set into the mapping referencesystem was based on GNSS (Global Navigation SatelliteSystem) navigation data only Consequently a global biason the order of 15ndash2 m resulted after georeferencing andno control on the intrinsic geometric block stability was pos-sible After the generation of the point cloud a DEM andorthophoto were produced with spatial resolutions of 60 and15 cm respectively

312 2016 dataset

Two UAV surveys were carried out on 30 August and1 September 2016 both around midday with 88 of thesky covered by stratocumulus clouds The UAV employedin these surveys was a customized quadcopter (see Fig 3b)carrying a Canon Powershot 16 Megapixel digital cameraTwo different take-off and landing sites were chosen to gainaltitude before take-off and maintain line-of-sight operationwith a flying altitude of 50 m above ground which ensured anaverage GSD of 6 cm To reduce motion blur camera shutterspeed was set to the lowest possible setting 12000 s withaperture at F27 and sensitivity at 200 ISO

Several flights were conducted to cover a small sectionof the proglacial plain and different surface types on theglacier surface including the terminus a collapsed area onthe central tongue the eastern medial moraine and some

wwwnat-hazards-earth-syst-scinet1810552018 Nat Hazards Earth Syst Sci 18 1055ndash1071 2018


Figure 2 Collapsing areas on the tongue of Forni Glacier (a) faults cutting across the eastern medial moraine (b) glacier terminus (c) near-circular collapsed area on the central tongue (d) large ring fault on the eastern tongue at the base of the icefall (e) Close-up of a vertical icecliff at the glacier terminus The location of features is reported in Fig 1 Photos courtesy of G Cola

Figure 3 The UAVs used in surveys of the Forni Glacier and their characteristics (a) The SwingletCam fixed-wing aircraft employedin 2014 at its take-off site by Lake Rosole (b) the customized quadcopter used in 2016 in the lab



debris-covered parts of the eastern tongue A ldquozig-zagrdquo fly-ing scheme was followed to reduce the flight time The UAVwas flown in autopilot mode using the open-source softwareMission Planner (Oborne 2013) to ensure 70 along-stripoverlap and side lap In total two flights were performed dur-ing the first survey and three during the second lasting about20 min each The surveyed area spanned over 059 km2

Eight GCPs (see Fig 1) were measured for the registra-tion of the photogrammetric blocks and its byproducts intothe mapping system The root mean square error (RMSE) ofthe GCP location was 40 cm which can be used as an in-dicator of the internal consistency of the photogrammetricblock The point cloud obtained from the 2016 UAV flightwas interpolated to produce a DEM and orthophoto withthe same cell resolution as the 2014 dataset ie 60 and15 cm respectively Both products were exported in theITRS2000UTM 32N mapping reference system

32 Terrestrial photogrammetry

The terrestrial photogrammetric survey was carried out on29 August 2016 to reconstruct the topographic surface of theglacier terminus which presented several vertical and sub-vertical surfaces (see Fig 2e) whose measurement was notpossible from the UAV platform carrying a camera in nadirconfiguration

Images were captured from 134 ground-based stationsmost of them located in front of the glacier and some onboth flanks of the valley in the downstream area A single-lens-reflex Nikon D700 camera was used equipped with a50 mm lens and a full-frame CMOS sensor (36times 24 mm)with 4256times 2823 pixels In this case since no preliminaryinformation about approximate camera position was col-lected the SfM procedure was run without any initial infor-mation

Seven natural features visible on the glacier front wereused as GCPs to be included in the bundle adjustment com-putation Measurement of GCPs in the field was carried outby means of a high-precision theodolite The measurementof points previously recorded with a GNSS geodetic receivermade it possible to register the coordinates of GCPs in themapping reference system The RMSE of 3-D residual vec-tors on GCPs was 34 cm

33 Terrestrial laser scanning

On the same days as the first UAV survey of 2016 a long-range terrestrial laser scanner Riegl LMS-Z420i was used toscan the glacier terminus One instrumental standpoint lo-cated on the hydrographic left flank of the glacier terminus(see Fig 1) was established The horizontal and vertical scan-ning resolution were set up to provide a spatial point densityof approx 5 cm on the ice surface at the terminus Georefer-encing was accomplished by placing five GCPs consisting incylinders covered by retroreflective paper The coordinates of

GCPs were measured by using a precision theodolite follow-ing the same procedure adopted for terrestrial photogramme-try Considering the accuracy of registration and the expectedprecision of laser point measurement the global uncertaintyof 3-D points was estimated on the order of plusmn75 cm

34 GNSS ground control points

Prior to the 2016 surveys eight control targets were placedboth on the periglacial area and on the glacier tongue (seeFig 1) Differential GNSS data were acquired at the target lo-cation for the georeferencing of UAV terrestrial photogram-metry and TLS data GCPs were used (1) to georeferenceUAV data directly by identifying the targets on the imagesin Photoscan and (2) to register theodolite measurements forgeoreferencing terrestrial photogrammetry and TLS The tar-gets consisted in a square piece of white fabric (80times 80 cm)with a circular marker in red paint chosen to provide con-trast against the background Except for the one GCP locatedat the highest site such GCPs were positioned on large flatboulders to provide a stable support and reduce the impact ofice ablation between flights

GNSS data were acquired by means of a pair of Le-ica Geosystems 1200 geodetic receivers working in RTK(real-time kinematics) mode (see Hoffman-Wellenhof et al2008) One of them was set up as master on a precise pointbeside Branca hut with known coordinates in the mappingreference system ITRS2000UTM 32N The second receiverwas used as a rover communicating via radio link with themaster station The maximum distance between master androver was less than 15 km but some points were measured instatic mode with measurement time of approximately 12 mindue to the local topography preventing the radio link and thelack of mobile phone services (for RTK) The theoretical un-certainty of GCPs provided by the processing code was onthe order of 2ndash3 cm

35 2007 DEM

The 2007 TerraItaly DEM was produced by the BLOMCGR company for the Lombardy region It is the final prod-uct of an aerial survey over the entire region conductedwith a multispectral push broom Leica ADS40 sensor ac-quiring images from a flying height of 6300 m with an av-erage GSD of 65 cm The images were processed to generatea DEM with a cell resolution of 2 mtimes 2 m and a plusmn3 m un-certainty We converted the DEM from the ldquoMonte Mariordquoto the ldquoITRS2000rdquo datum and the height from ellipsoidal togeodetic using the official software for datum transformationin Italy (Verto ver 3)



Figure 4 Location of different glacier features or hazard-prone areas on the tongue of Forni Glacier were the point cloud comparison wasperformed The background image is the merged point cloud generated from the 2016 UAV and terrestrial photogrammetry survey

4 Methods

41 Analysis of point clouds from the 2016 campaignUAVterrestrial photogrammetry and TLS

The comparison between point clouds generated during the2016 campaign had the aim of assessing their geometricquality before their application for the analysis of hazardsThese evaluations were also expected to provide some guide-lines for the organization of future investigations in the fieldat the Forni Glacier and in other Alpine sites Specificallywe analyzed point density (points mminus2) and completenessie of area in the ray view angle Point density partlydepends upon the surveying technique used since it is con-trolled by the distance between sensor and surface and de-termines spatial resolution In SfM photogrammetry pointdensity is affected by image texture sharpness and resolu-tion which influence the performance of dense matching al-gorithms (DallrsquoAsta et al 2015) while in TLS it can be setup as a data acquisition input parameter In this study thenumber of neighbors N (inside a sphere of radius R= 1 m)divided by the neighborhood surface was used to evaluatethe local point density D in CloudCompare (httpwwwcloudcompareorg) To understand the effect of point densitydispersion (Teunissen 2009) the inferior 125 percentile ofthe standard deviation σ of point density was also calculatedThe use of these local metrics allowed us to distinguish be-tween point densities in different areas since this may largelychange from one portion of surface to another A further met-ric in this sense was point cloud completeness referring to

the presence of enough points to completely describe a por-tion of surface In this study the visual inspection of selectedsample locations was used to identify occlusions and areaswith lower point density

To analyze these properties five regions were selected (seeFig 4) located on the glacier topographic surface and char-acterized by different glacier features and the presence ofhazards (1) a glacial cavity composed of subvertical andfractured surfaces over 20 m high and forming a typical semi-circular shape (2) a glacial cavity over 10 m high with thesame typical semicircular shape as location 1 covered byfine- and medium-sized rock debris (3) a normal fault over10 m high (4) a highly collapsed area covered by fine- andmedium-sized rock debris and rock boulders and (5) a planarsurface with a normal fault covered by fine- and medium-sized rock debris and rock boulders The analysis of lo-cal regions was preferred to the overall analysis of all thepoint clouds due to (1) the incomplete overlap between pointclouds obtained from different methods and (2) the oppor-tunity to investigate the performances of the techniques indiverse geomorphological situations

Within the same sample locations we compared the pointclouds in a pairwise manner Since no available benchmark-ing data set (eg accurate static GNSS data) was concur-rently collected during the 2016 campaign the TLS pointcloud was used as a reference When comparing both pho-togrammetric data sets the one obtained from the UAV wasused as reference because of the even distribution of pointdensity within the sample locations The presence of resid-ual non-homogenous georeferencing errors in the data sets



required a specific fine registration of each individual sam-ple location which was conducted in CloudCompare usingthe ICP (iterative closest point) algorithm (Pomerleau et al2013) ICP iteratively matches a source point cloud to a ref-erence point cloud in Euclidean space and calculates the nec-essary rotation and translation to align the source point cloudwith the reference based on minimization of a distance met-ric in a point-to-point fashion After fine registration pointclouds in corresponding sample areas were compared usingthe M3C2 algorithm implemented in CloudCompare (Lagueet al 2013) As discussed in Fey and Wichmann (2016) thedistance between a pair of point clouds is often evaluated bycomparing elevations at corresponding nodes of DEMs afterresampling of the original data This approach works prop-erly when both point clouds are approximately aligned alongthe same planar direction but not when there are structureswith different alignments as in the case of the glacier surfacesunder investigation In fact the M3C2 algorithm does not al-ways evaluate the distance between two point clouds alongthe same directional axis but computes a set of local normalsusing points within a radiusD depending on the local rough-ness which is directly estimated from the point cloud dataand also considering the accuracy of preliminary local regis-tration refinement using ICP In this case a radiusD of 20 cmand a pre-registration accuracy of 5 cm were considered thelatter obtained from ICP residuals This solution allowed usto remove registration errors from the analysis and focus onthe capability of the adopted techniques to reconstruct thelocal geometric surface of the glacier in an accurate way

42 Point cloud merging

To improve coverage of different glacier surfaces includingplanar areas and normal faults photogrammetric point cloudsfrom the 2016 campaign (UAV and terrestrial surveys) weremerged We chose to avoid TLS and employed the two lower-cost techniques (Chandler and Buckley 2016) to assess theirpotential for combined future use Prior to point cloud merg-ing a preliminary co-registration was performed on the basisof the ICP algorithm in CloudCompare Regions common toboth point clouds were used to minimize the distances be-tween them and find the best co-registration The point cloudfrom UAV photogrammetry which featured the largest ex-tension was used as reference during co-registration whilethe other was rigidly transformed to fit with it After manyiterations both point clouds were aligned according to thebest solution found by the ICP In order to remove redundantpoints and to obtain a homogenous point density the mergedpoint cloud (see Fig 5) was subsampled keeping a minimumdistance between adjacent points of 20 cm The final size ofthis merged data set is approximately 44 million points TheRGB color information associated with each point in the fi-nal point cloud was derived by averaging the RGB informa-tion of original points in the subsampling volumes Whilethis operation resulted in losing part of the original RGB in-

Figure 5 Maps of point density in sample location 2

formation it helped to provide a realistic visualization of thetopographic model and therefore to interpret glacier hazards

43 Glacier hazard mapping

The investigation of glacier hazards was conducted using thepoint cloud and orthophoto from the 2014 UAV data set aswell as the merged (UAV and terrestrial) point cloud andorthophoto from 2016 In this study we focused on ringfaults and normal faults which were identified by visuallyinspecting their geometric properties in the point clouds andmanually delineated while color information from orthopho-tos was used as a cross check On orthophotos both typesof structures generally appear as dark linear features ow-ing to shadows projected by fault scarps As these structuresmay look similar to crevasses further information concern-ing their orientation and location needs to be assessed fordiscrimination The orientation of fault structures is not co-herent with glacier flow with ring faults also appearing incircular patterns Their location is limited to the glacier mar-gins medial moraines and terminus whereas crevasses canappear anywhere on the glacier surface (Azzoni et al 2017)After delineation we also analyzed the height of vertical fa-cies using information from the point clouds

44 DEM co-registration for glacier thickness changeestimation

Several studies have found that errors in individual DEMsboth in the horizontal and vertical domain propagate whencalculating their difference leading to inaccurate estima-



tions of thickness and volume change (Berthier et al 2007Nuth and Kaab 2011) In the present study different ap-proaches were adopted for georeferencing all the DEMs usedin the analysis of the volume change of the Forni Glaciertongue (2007 2014 2016) To compute the relative dif-ferences between the DEMs a preliminary co-registrationwas therefore required The method proposed by Berthier etal (2007) for the co-registration of two DEMs was separatelyapplied to each DEM pair (2007ndash2014 2007ndash2016 2014ndash2016) Following this method in each pair one DEM playsas reference (ldquomasterrdquo) while the other is used as ldquoslaverdquoDEM to be iteratively shifted along x and y axes by frac-tions of a pixel to minimize the standard deviation of eleva-tion differences with respect to the master DEM Only ar-eas assumed to be stable are considered in the calculationof the co-registration shift The ice-covered areas were ex-cluded by overlaying the glacier outlines from DrsquoAgata etal (2014) for 2007 and Fugazza et al (2015) for 2014 Theoldest DEM which is also the widest in each comparisonwas always set as the master To co-register the 2014 and2016 DEMs with the 2007 DEM both were resampled to2 m spatial resolution whereas the comparison between 2014and 2016 was carried out at the original resolution of thesedata sets (60 cm)

All points resulting in elevation differences greater than15 m were labeled as unreliable and consequently discardedfrom the subsequent analysis Such greater discrepanciesmay denote errors in one of the DEMs or unstable areas out-side the glacier Values exceeding this threshold howeverwere only found in a marginal area with low image over-lap in the comparison between the 2014 and 2016 DEMswith a maximum elevation difference of 36 m Once the finalco-registration shifts were computed (see Table 1) the coef-ficients were subtracted from the top left coordinates of theslave DEM the residual mean elevation difference was alsosubtracted from the slave DEM to bring the mean to zeroAfter DEM co-registration the resulting shifts reported inTable 1 were applied to each slave DEM including the en-tire glacier area Then the elevations of the slave DEM weresubtracted from the corresponding elevations of the masterDEM to obtain the so-called DEM of differences (DoD)Over a common glacier area (Fig 1) we estimated the vol-ume change and its uncertainty which can be expressed asthe combination of (1) uncertainty due to errors in elevationand (2) the truncation error caused by the use of a discretesum (sum of DoD at each pixel multiplied by pixel area)in place of the integral in volume calculation (Jokinen andGeist 2010) We calculated the former following the ap-proach of Rolstad et al (2009) taking into account spatialautocorrelation of elevation change over stable areas con-sidering a correlation length of 50 m for the latter we usedthe method described by Jokinen and Geist (2010)

Table 1 Statistics of the elevation differences between DEM pairsbefore and after the application of co-registration shifts DEM 2007is from aerial multispectral survey DEM 2014 and DEM 2016 arefrom UAV photogrammetry

DEM pair Elevation Co-registration shifts Elevation

differences X (m) Y (m) differences withwithout co-registration

co-registration shiftsshifts (micro1H plusmn σ1H )

(micro1H plusmn σ1H ) (m)(m)

2007ndash2014 196plusmn 260 111 minus111 000plusmn 1702007ndash2016 minus043plusmn 348 244 minus111 000plusmn 2602014ndash2016 minus292plusmn 321 minus020 minus130 000plusmn 222

5 Results

51 Point cloud analysis

The analysis of point density shows significant differencesbetween the three techniques for point cloud generation (seeTable 2) Values range from 103 to 2297 points mminus2 depend-ing on the surveying method but the density was gener-ally sufficient for the reconstruction of the different surfacesshown in Fig 4 except for location 5 Terrestrial photogram-metry featured the highest point density while UAV pho-togrammetry had the lowest In relation to UAV photogram-metry similar point densities were found in all sample lo-cations especially for the standard deviations that were al-ways in the 22ndash29 point mminus2 range Mean values were 103ndash109 points mminus2 in locations 2ndash4 while they were higher inlocation 5 (141 points mminus2) Due to the nadir acquisitionpoints the 3-D modeling of verticalsubvertical cliffs in lo-cation 1 was not possible In relation to TLS a mean value ofpoint density ranging from 141 to 391 points m2 was foundwith the only exception of location 5 where no sufficientdata were recorded due to the position of this region withrespect to the instrumental standpoint Standard deviationsranged between 69 and 217 points m2 moderately correlatedwith respective mean values The analysis of the complete-ness of surface reconstruction also revealed some issues re-lated to the adopted techniques (see Fig 5) Specifically TLSsuffered from severe occlusions which prevented acquisitionof data in the central part of the sample area while UAV pho-togrammetry was able to reconstruct the upper portion of thesample area but not the vertical cliff Only terrestrial pho-togrammetry acquired a large number of points in all areas

In terms of point cloud distance (see Table 3) the compar-ison between TLS and terrestrial photogrammetry resultedin a high similarity between point clouds with no greatdifferences between different sample areas Conversely thecomparison between TLS and UAV photogrammetry andterrestrial and UAV photogrammetry provided significantlyworse results which may be summarized by the RMSEs in



Table 2 Area and number of points in each sample window on the Forni Glacier terminus mean and standard deviation of local pointdensity and number of points above the lower 125 percentile in each window k stands for thousands of points UAV refers to UAVphotogrammetry TP to terrestrial photogrammetry and TLS to terrestrial laser scanning

Sample Area Number of points abovewindow (m2) Number of points in sample Mean and standard deviation of point the lower 125

windows density (points mminus2) percentile

UAV TP TLS UAV TP TLS UAV TP TLS

1 2793 ndash 1984k 141k ndash 1654plusmn 637 226plusmn 100 ndash 880 262 1806 76k 2175k 130k 109plusmn 29 2297plusmn 708 391plusmn 217 61 881 03 495 43k 712k 25k 103plusmn 27 1978plusmn 606 151plusmn 60 49 766 314 672 62k 557k 33k 108plusmn 22 1384plusmn 530 141plusmn 69 62 324 25 3960 406k 810k ndash 141plusmn 22 485plusmn 227 ndash 97 31 ndash

Table 3 Statistics on distances between point clouds computed on the basis of the M3C2 algorithm showing mean standard deviation androot mean square error (RMSE) of each point cloud pair UAV refers to UAV photogrammetry TP to terrestrial photogrammetry and TLS toterrestrial laser scanning Ref stands for reference and ldquondashrdquo means no comparison was performed

Sample Means and SDs of M3C2 distances RMSE of M3C2window (cm) distances (cm)

Ref TLS TLS UAV TLS TLS UAVSlave TP UAV TP TP UAV TP

1 45plusmn 74 ndash ndash 87 ndash ndash2 minus11plusmn 105 148plusmn 347 minus145plusmn 267 106 377 3043 84plusmn 41 147plusmn 151 minus85plusmn 189 94 211 2074 28plusmn 53 94plusmn 222 minus23plusmn 249 60 240 2505 ndash ndash minus85plusmn 253 ndash ndash 267

the ranges of 211ndash377 and 207ndash304 cm respectively Thegreater deviations were in both cases obtained in the analy-sis of location 2 which mostly represents a vertical surfacewhile the best agreement was found within location 3 whichis less inclined As the UAV flight was georeferenced on a setof GCPs with an RMSE of 405 cm the ICP co-registrationmay have not totally compensated the existing bias

In terms of spatial coverage considering the entire surfaceexamined using each technique outside the sample locationsthe UAV survey extended over the widest area (059 km2)including part of the proglacial plain the entire terminus andthe glacier tongue up to the collapsed area on the centralpart but with data gaps on the vertical and subvertical walls(see Fig 6a) The point cloud obtained from terrestrial pho-togrammetry covered approximately a third of the area sur-veyed with the UAV (see Fig 6b) including the full glacierterminus at very high spatial resolution with the exception ofa few obstructed parts while the TLS point cloud covered theterminus although with some holes due to the obstructions

52 Glacier-related hazards and risks

The tongue of Forni Glacier hosts several hazardous struc-tures including crevasses normal faults and ring faults Inthis study we focused on the latter two due to their relation-

ship with glacier downwasting While most collapsed areason Forni Glacier are normal faults two large ring fault sys-tems can be identified the first located in the eastern sec-tion (see Figs 2d and 7) covered an area of 256times 103 m2

and showed surface dips of up to 5 m in 2014 This area wasnot surveyed in 2016 since field observation did not showevidence of further significant subsidence Conversely thering fault that only emerged as a few semicircular fracturesin 2014 grew until cavity collapse with a vertical displace-ment up to 20 m Additional fractures extending southeast-ward (see Figs 2c and 7) formed between 2014 and 2016suggesting that the extent of the collapse might widen in thenear future In 2014 further ring faults were also identifiedat the eastern glacier margin and the one that was surveyedin 2016 showed evidence of increasing subsidence (+2 m)and the formation of additional subparallel fractures

Normal faults are mostly found on the eastern medialmoraine and at the terminus Between 2014 and 2016 thefirst (see Fig 2a) developed rapidly in the vertical domainreaching a height of 12 m in 2016 The latter increased innumber as size as the terminus underwent collapse leadingto the formation of three major ice cavities up to 24 m high(see Fig 2b and e) In 2016 ice thickness above the cavityvault was between 5 and 10 m (see Fig 4 location 1) Several



Figure 6 Spatial coverage of UAV and terrestrial photogramme-try point clouds and merged point cloud from the two techniques(a) UAV photogrammetry point cloud (b) terrestrial photogramme-try point cloud (c) merged point cloud

fractures identified as normal faults also appear in conjunc-tion with the large ring fault located in the central sectionof the glacier extending the fracture system to the westernglacier margin

The presence of cavities at the terminus which is easilyreached in a 45 min walk from Branca hut is particularlyhazardous for tourists because of (1) the danger of cavitycollapse and (2) the potential fall of large boulders or blocksof ice from the ice cliffs Other fracture systems locatedhigher up glacier are presumably reached by more experi-enced hikers Most ring faults however show evidence ofvertical and horizontal expansion and further cavity collapsewould imply a severe risk of injury and death if hikers wereinvolved

The location of structures in 2016 suggests that the glacierterminus will recede through ice cliff backwasting and cavitycollapse along the fault system on the eastern medial moraineand along the ring faults at the eastern and western mar-

Table 4 Average ice thickness change thinning rates and vol-ume loss from DEM differencing over a common reference areaof 032 km2 for all DEM pairs Uncertainty of thickness changeexpressed as 1 standard deviation of residual elevation differencesover stable areas after DEM co-registration

DEM pair Mean thickness Mean thinning Volume changechange (m) rates (m aminus1) (106 m3)

2007ndash2014 minus3191plusmn 170 minus455plusmn 024 minus1000plusmn 017 (174 )2007ndash2016 minus4286plusmn 260 minus476plusmn 029 minus1346plusmn 020 (147 )2014ndash2016 minus1041plusmn 222 minus520plusmn 111 minus329plusmn 008 (260 )

gins potentially compromising access to the glacier Sincethis system of fractures has been developing very rapidlyand new collapses have already been documented in Septem-ber 2017 the risk for people walking on the glacier tongueduring summer should be carefully evaluated Although sur-face features may be visually detected the availability of de-tailed 3-D models that depict the entire outer surface of theglacier is a great advantage because it allows quickly cap-turing the glacier topography remotely helping predict thepossible development of new collapses and understand theirmechanisms of formation

53 Glacier thickness change

The Forni Glacier tongue was affected by substantial thin-ning throughout the observation period Between 2007and 2014 the greatest thinning occurred in the eastern sec-tion of the glacier tongue with changes persistently lowerthan minus30 m (more than 4 m aminus1 thinning) whereas the up-per part of the central tongue only thinned by 10ndash18 m (be-tween approximately 1 and 25 m aminus1) The greatest ice lossoccurred in correspondence with the normal faults localizedin small areas at the eastern glacier margin (see Fig 8a)with local changes generally lower than minus50 m (more than7 m aminus1 thinning) and a minimum of minus6680 m owing tothe formation of a lake Conversely between 2014 and 2016the central and eastern parts of the tongue had similar thin-ning patterns with average changes ofminus10 m (5 m aminus1) Thegreatest losses are mainly found in correspondence with nor-mal faults with a maximum change of minus3871 m at the ter-minus and local thinning greater than 25 m on the lower me-dial moraine The ring fault at the left margin of the centralsection of the tongue also shows thinning of 20 to 26 m (10ndash13 m aminus1) In the absence of faults little thinning occurredinstead on the upper part of the medial moraine where athick debris cover shielded ice from ablation with changesof minus2 to minus5 m (1 to 25 m aminus1 see Fig 8c) Considering acommon reference area (see Fig 1 Table 4) an accelerationof glacier thinning seems to have occurred over recent yearsover the lower glacier tongue from minus455plusmn 024 m aminus1

in 2007ndash2014 to minus520plusmn 111 m aminus1 in 2014ndash2016 alsoconfirmed by the value ofminus476plusmn 029 m aminus1 obtained fromthe comparison between 2007 and 2016 Looking at the first



Figure 7 Location of collapse structures ie normal faults and ring faults and trails crossing the Forni Glacier (a) Collapse structuresin 2014 with 2014 UAV orthophoto as base map The red box marks the area surveyed in 2016 (b) Collapse structures in 2016 with2016 UAV orthophoto as base map Trails from Kompass online cartography at httpswwwkompass-1039italiaitinfomappa-online

Figure 8 Ice thickness change rates from DEM differencing over (a) 2007ndash2014 (b) 2007ndash2016 and (c) 2014ndash2016 Glacier outlinesfrom 2014 and 2016 are limited to the area surveyed during the UAV campaigns Base map from hill shading of 2007 DEM



two DoDs the trend seems to be caused by the increase incollapsing areas (Fig 8a and b) In all DoDs the uncertaintyin ice thickness change affects less than 3 of the respectivevolume change (see Table 4)

6 Discussion comparison of techniques for point cloudgeneration

The choice of a technique to monitor glacier hazards and theglacier thickness changes depends on several factors includ-ing the size of the area the desired spatial resolution andaccuracy logistics and cost In this study we focused on spa-tial metrics ie point density completeness and distancebetween point clouds to evaluate the performance of UAVclose-range photogrammetry and TLS in a variety of condi-tions

61 Point density and completeness

Considering point density terrestrial photogrammetry re-sulted in a denser data set than the other techniques Thisis mostly motivated by the possibility of acquiring data fromseveral stations using this methodology depending only onthe terrain accessibility reducing the effect of occlusionswith a consequently more complete 3-D modeling Howeverthe mean point density achieved when using terrestrial pho-togrammetry is highly variable both between different sam-ple locations and within each location as shown by the stan-dard deviations of D Point densities related to UAV pho-togrammetry and TLS are more regular and constant In thecase of UAV photogrammetry the homogeneity of point den-sity might be due to the regular structure of the airbornephotogrammetric block In the case of TLS the regularity ismotivated by the constant angular resolution adopted duringscanning Since any technique may perform better when thesurface to survey is approximately orthogonal to the sensorrsquospoint of view terrestrial photogrammetry is more efficientfor reconstructing vertical and subvertical cliffs (sample ar-eas 1 and 2) and high-sloped surfaces (sample areas 3 and 4)In contrast airborne UAV photogrammetry provided the bestresults in location 5 which is less inclined and consequentlycould be well depicted in vertical photos In general pointclouds from terrestrial photogrammetry provide a better de-scription of the vertical and subvertical parts (see eg Win-kler et al 2012) while point clouds obtained from UAV pho-togrammetry are more suitable to describe the horizontal orsubhorizontal surfaces on the glacier tongue and periglacialarea (Seier et al 2017) unless the camera is tilted to an off-nadir viewpoint (Dewez et al 2016 Aicardi et al 2016)Results obtained from photogrammetry based on terrestrialand UAV platforms can thus be considered quite complemen-tary and they support the concept of merging the point cloudsfrom these two techniques as seen in Fig 6c In agreementwith other studies of vertical rock slopes (eg Abellaacuten et al

2014) we found that the TLS point cloud was affected byocclusions (see eg location 2 in Figs 4 and 5) which canonly be compensated by increasing the number of stationsData acquisition with this platform was in general difficult inregions subparallel to the laser beams and in the presence ofwet surfaces

62 Point cloud and DEM uncertainty

In this study the distance between the UAV and TLS pointclouds (211ndash377 cm RMSE) assumed as a measure of theuncertainty of the 2016 UAV data set was slightly higherthan previously reported in high mountain glacial environ-ments (eg Immerzeel et al 2014 Gindraux et al 2017Seier et al 2017) although in these studies the comparisonwas between DEMs and GNSS control points Contributingfactors might include the suboptimal distribution and den-sity of GCPs (Gindraux et al 2017) the delay between theUAV surveys as well as between the UAV and TLS and thelack of coincidence between GCP placement and the UAVflights This means the UAV photogrammetric reconstruc-tion was affected by ice ablation and glacier flow which onForni Glacier range between 3 and 5 cm dayminus1 (Senese et al2012) and between 1 and 4 cm dayminus1 respectively (Urbiniet al 2017) We thus expect a combined 3-day uncertaintyon the 2016 UAV data set between 10 and 20 cm and loweron GCPs considering reduced ablation due to their place-ment on boulders A further contribution to the GCP errorbudget might stem from the intrinsic precision of GNSS andtheodolite measurements and image resolution The compar-ison between close-range photogrammetry and TLS was lessaffected by glacier change because data were collected 1 dayapart and the RMSE of 6ndash106 cm is in line with previousfindings by Kaufmann and Landstaedter (2008) To reducethe uncertainty of UAV photogrammetric blocks a better dis-tribution of GCPs or switching to an RTK system should beconsidered while close-range photogrammetry could benefitfrom measuring a part of the photo-stations as proposed inForlani et al (2014) instead of placing GCPs on the glaciersurface

The uncertainty in UAV photogrammetric reconstructionalso factored in the standard deviation still present after theco-registration between DEMs in areas outside the glacier(222 m between 2014 and 2016) Another important factorhere is the morphology of the co-registration area ie theoutwash plain still subject to changes due to the inflowof glacier meltwater and sediment reworking UAV pho-togrammetric products permitted us to investigate ice volumechanges over 2 years with an uncertainty of 260 while theintegration with close-range photogrammetry was required toinvestigate hazards related to the collapse of the glacier ter-minus



63 Logistics and costs

In our surveys it became evident that the main disadvantageof TLS compared to photogrammetry is the complexity of in-strument transport and setup In terms of logistics and work-load up to five people were involved in the transportation ofthe TLS instruments (laser scanner theodolite at least twotopographic tripods and poles electric generator and ancil-lary accessories) while two people were required for UAVand close-range photogrammetric surveys which were alsoconsiderably faster Meteorological conditions and the lim-ited access to unstable areas close to the glacier terminus alsoprevented the acquisition of TLS data from other viewpointsas done with photogrammetry Concerning UAV surveys weconducted them under different meteorological scenarios andobtained adequate results in early-morning operations with08 cloud cover and midday flights with 88 cloud coverBoth scenarios can provide diffuse light conditions allow-ing collection of pictures suitable for photogrammetric pro-cessing but camera settings need to be carefully adjusted be-forehand (OrsquoConnor et al 2017) If early-morning flights arenot feasible in the study area for logistical reasons or whensurveying glaciers with eastern exposures the latter scenarioshould be considered

In terms of costs UAV and terrestrial photogrammet-ric surveys are also advantageous since TLS instrumentsare much more expensive at EUR 70 000ndash100 000 comparedto UAVs (EUR 3500 for our platform) and DSLR (digitalsingle-lens reflex) cameras used in photogrammetry in theEUR 500ndash3500 range

64 Additional remarks

In summary although TLS point clouds are regarded as themost accurate (Naumann et al 2013) they suffer from in-homogeneous point density and cumbersome logistics andtheir potential in glacial environments is limited unless amaximum uncertainty of 5ndash10 cm can be tolerated Laserscanners are also employed on aerial platforms includingUAVs where they can reconstruct terrain morphology withonly slightly higher uncertainty than the terrestrial counter-parts with a much greater coverage (Rayburg et al 2009)but the high operational cost has limited the diffusion ofthis technique Lastly photogrammetry from higher-altitudeaerial platforms (mostly planes but also helicopters andsatellites) can similarly achieve low uncertainty (3 m An-dreassen et al 2002) and extensive coverage at the price ofa lower spatial resolution compared to UAVs (eg 2 m inour case) and due to its popularity in the past it is often theonly means to acquire good quality archive data to investi-gate glacier changes over broad timescales (Andreassen etal 2002 Moelg and Bloch 2017)

In our pilot study we covered part of the Forni Glaciertongue and investigated different techniques to mapmonitorhazards related to the glacier collapse Our maps can help

identify safer paths where mountaineers and skiers can visitthe glacier and reach the most important summits Howeverthe increase in collapse structures owing to climate changerequires multitemporal monitoring A comprehensive risk as-sessment should also cover the entire glacier to investigatethe probability of serac detachment and provide an estimateof the glacier mass balance with the geodetic method Whileour integrated approach using a multicopter and terrestrialphotogrammetry should be preferred to TLS for the investi-gation of small individual ice bodies fixed-wing UAVs ide-ally equipped with an RTK system and the ability to tilt thecamera off-nadir might be the platform of choice to coverlarge distances (see eg Ryan et al 2017) potentially re-ducing the number of flights and solving issues with GCPplacement Such platforms could help collect sufficient datafor hazard management strategies up to the basin scale inStelvio National Park and other sectors of the Italian Alpseventually replacing higher-altitude aerial surveys Cost anal-yses (Matese et al 2015) should also be performed to eval-uate the benefits of improved spatial resolution and lowerDEM uncertainty of UAVs compared to aerial and satellitesurveys and choose the best approach for individual cases

7 Conclusions

In our study we compared point clouds generated from UAVphotogrammetry close-range photogrammetry and TLS toassess their quality and evaluate their potential in mappingand describing glacier hazards such as ring faults and normalfaults in a specific campaign carried out in summer 2016 Inaddition we employed orthophotos and point clouds from aUAV survey conducted in 2014 to analyze the evolution ofglacier hazards as well as a DEM from an aerial photogram-metric survey conducted in 2007 to investigate glacier thick-ness changes between 2007 and 2016 The main findings ofour study are as follows

ndash UAVs and terrestrial photogrammetric surveys providereliable performances in glacial environments and out-perform TLS in terms of logistics and costs

ndash UAV and terrestrial photogrammetric blocks can be eas-ily integrated providing more information than individ-ual techniques to help identify glacier hazards

ndash UAV-based DEMs can be employed to estimate thick-ness and volume changes but improvements are neces-sary in terms of area covered and to reduce uncertainty

ndash The Forni Glacier is rapidly collapsing with an increasein ring fault sizes providing evidence of climate changein the region

ndash The glacier thinning rate increased due to collapses to520plusmn 111 m aminus1 between 2014 and 2016



The maps produced from the combined analysis of UAV andterrestrial photogrammetric point clouds and orthophotos canbe made available through GIS web portals of the Stelvio Na-tional Park or the Lombardy region (httpwwwgeoportaleregionelombardiait) A permanent monitoring programmeshould be set up to help manage risk in the area issuingwarnings and assisting mountain guides in changing hikingand ski routes as needed The analysis of glacier thicknesschanges suggests a feedback mechanism which should befurther analyzed with higher thinning rates leading to in-creased occurrence of collapses Glacier downwasting is alsoof relevance for risk management in the protected area pro-viding valuable data to assess the increased chance of rock-falls and to improve forecasts of glacier meltwater produc-tion

While our test was conducted on one of the largest glaciersin the Italian Alps the integrated photogrammetric approachis easily transferable to similarly sized and much smallerglaciers where it would be able to provide a comprehen-sive assessment of hazards and thickness changes and be-come useful in decision support systems for natural hazardmanagement In larger regions UAVs hold the potential tobecome the platform of choice but their performances andcost-effectiveness compared to aerial and satellite surveysmust be further evaluated

Data availability All data except for the 2007 photogrammet-ric DEM were acquired by the authors Data can be re-quested by email from the authors davidefugazzaunimiit ormarcoscaionipolimiit

Competing interests The authors declare that they have no conflictof interest

Special issue statement This article is part of the special issueldquoThe use of remotely piloted aircraft systems (RPAS) in monitoringapplications and management of natural hazardsrdquo It is a result of theEGU General Assembly 2016 Vienna Austria 17ndash22 April 2016

Acknowledgements This study was funded by DARA the De-partment for Autonomies and Regional Affairs of the Italian gov-ernmentrsquos Presidency of the Council of Ministers The authorsare grateful to the central scientific committee of CAI (ClubAlpino Italiano ndash Italian Alpine Club) and Levissima SanpellegrinoSPA for funding the UAV quadcopter The authors also thankStelvio Park Authority for the logistic support and for permitting theUAV surveys and IIT Regione Lombardia for the provision of the2007 DEM Our gratitude also goes to the GICARUS lab of Politec-nico Milano at Lecco Campus for providing the survey equipmentFinally the authors would also like to thank Tullio Feifer Livio Pi-atta and Andrea Grossoni for their help during field operations andthe reviewers for their helpful comments

Edited by Paolo TarolliReviewed by three anonymous referees

References

Abellaacuten A Oppikofer T Jaboyedoff M Rosser N JLim M and Lato M J Terrestrial laser scanning ofrock slope instabilities Earth Surf Proc Land 39 80ndash97httpsdoiorg101002esp3493 2014

Aicardi I Chiabrando FGrasso N Lingua A M Noardo Fand Spanograve A UAV photogrammetry with oblique images firstanalysis on data acquisition and processing in InternationalArchives of the Photogrammetry Remote Sensing and Spatial In-formation Sciences 12ndash19 July 2016 Prague Czech Republic41-B1 835ndash842 httpsdoiorg105194isprs-archives-XLI-B1-835-2016 2016

Andreassen L M Hallgeir E and Kjollmoen B Using aerialphotography to study glacier changes in Norway Ann Glaciol34 343ndash348 httpsdoiorg1031891727564027818176262010

Azzoni R S Fugazza D Zennaro M Zucali M DrsquoAgata CMaragno D Cernuschi M Smiraglia C and Diolaiuti GA Recent structural evolution of Forni Glacier tongue (Ortles-Cevedale Group Central Italian Alps) J Maps 13 870ndash878httpsdoiorg1010801744564720171394227 2017

Berthier E Arnaud Y Kumar R Ahmad S WagnonP and Chevallier P Remote sensing estimates of glaciermass balances in the Himachal Pradesh (Western Hi-malaya India) Remote Sens Environ 108 327ndash338httpsdoiorg101016jrse200611017 2007

Berthier E Cabot V Vincent C and Six D DecadalRegion-Wide and Glacier-Wide Mass Balances Derived fromMulti-Temporal ASTER Satellite Digital Elevation Mod-elsValidation over the Mont-Blanc Area Front Earth Sci 4 63httpsdoiorg103389feart201600063 2016

Bhardwaj A Sam L Akanksha Martin-Torres F J and Ku-mar R UAVs as remote sensing platform in glaciology Presentapplications and future prospects Remote Sens Environ 175196ndash204 httpsdoiorg101016jrse201512029 2016

Blasone G Cavalli M and Cazorzi F Debris-Flow Monitor-ing and Geomorphic Change Detection Combining Laser Scan-ning and Fast Photogrammetric Surveys in the Moscardo Catch-ment (Eastern Italian Alps) in Engineering Geology for So-ciety and Territory Vol 3 edited by Lollino G ArattanoM Rinaldi M Giustolisi O Marechal J C and GrantG Springer Cham 51ndash54 httpsdoiorg101007978-3-319-09054-2_10 2015

Carey M McDowell G Huggel C Jackson M Portocar-rero C Reynolds J M and Vicuntildea L Integrated ap-proaches to adaptation and disaster risk reduction in dy-namic sociocryospheric systems in Snow and Ice-relatedHazards Risks and Disasters edited by Haeberli W andWhiteman C Elsevier Amsterdam the Netherlands 219ndash261httpsdoiorg101016B978-0-12-394849-600008-1 2014

Chandler J H and Buckley S Structure from motion (SFM) pho-togrammetry vs terrestrial laser scanning in Geoscience Hand-book 2016 AGI Data Sheets 5th Edn Section 201 edited by



Carpenter M B and Keane C M American Geosciences In-stitute Alexandria USA 2016

Chiarle M Iannotti S Mortara G and Deline P Re-cent debris flow occurrences associated with glaciersin the Alps Global Planet Change 56 123ndash136httpsdoiorg101016jgloplacha200607003 2007

Clague J Glacier Hazards in Encyclopedia of NaturalHazards edited by Bobrowski P Springer Dordrechtthe Netherlands 400ndash405 httpsdoiorg101007978-1-4020-4399-4_156 2013

Colomina I and Molina P Unmanned aerial sys-tems for photogrammetry and remote sensing A re-view ISPRS J Photogram Remote Sens 92 79ndash97httpsdoiorg101016jisprsjprs201402013 2014

DrsquoAgata C Bocchiola D Maragno D Smiraglia C and Dio-laiuti G A Glacier shrinkage driven by climate change duringhalf a century (1954ndash2007) in the Ortles-Cevedale group (StelvioNational Park Lombardy Italian Alps) Theor Appl Cima-tol 116 169ndash190 httpsdoiorg101007s00704-013-0938-52014

DallrsquoAsta E Thoeni K Santise M Forlani G Giacomini Aand Roncella R Network design and quality checks in auto-matic orientation of close-range photogrammetric blocks Sen-sors 15 7985ndash8008 httpsdoiorg103390s150407985 2015

Dewez T J B Leroux J and Morelli S Cliff collapse haz-ard from repeated multicopter UAV acquisitions return on ex-perience in The International Archives of the PhotogrammetryRemote Sensing and Spatial Information Sciences XXIII IS-PRS Congress 12ndash19 July 2016 Prague Czech Republic41-B5 805ndash811 httpsdoiorg105194isprs-archives-XLI-B5-805-2016 2016

Diolaiuti G A and Smiraglia C Changing glaciersin a changing climate how vanishing geomorphositeshave been driving deep changes in mountain land-scapes and environments Geacuteomorphologie 2 131ndash152httpsdoiorg104000geomorphologie7882 2010

Diolaiuti G A Bocchiola D DrsquoAgata C and Smiraglia CEvidence of climate change impact upon glaciersrsquo recessionwithin the Italian Alps Theor Appl Climatol 109 429ndash445httpsdoiorg101007s00704-012-0589-y 2012

Eltner A Kaiser A Castillo C Rock G Neugirg F and Abel-laacuten A Image-based surface reconstruction in geomorphometryndash merits limits and developments Earth Surf Dynam 4 359ndash389 httpsdoiorg105194esurf-4-359-2016 2016

Fey C and Wichmann V Long-range Terrestrial laser scan-ning for geomorphological change detection in alpine terrain ndashhandling uncertainties Earth Surf Proc Land 42 789ndash802httpsdoiorg101002esp4022 2016

Fischer M Huss M Barboux C and Hoelzle M Thenew Swiss Glacier Inventory SGI2010 relevance of us-ing high-resolution source data in areas dominated byvery small glaciers Arct Antarct Alp Res 46 933ndash945httpsdoiorg1016571938-4246-464933 2014

Fischer M Huss M and Hoelzle M Surface elevation and masschanges of all Swiss glaciers 1980ndash2010 The Cryosphere 9525ndash540 httpsdoiorg105194tc-9-525-2015 2015

Forlani G Pinto L Roncella R and Pagliari D Terrestrial pho-togrammetry without ground control points Earth Sci Inform7 71ndash81 httpsdoiorg101007s12145-013-0127-1 2014

Fugazza D Senese A Azzoni R S Smiraglia C CernuschiM Severi D and Diolaiuti G A High-resolution mappingof glacier surface features The UAV survey of the Forni glacier(Stelvio national park Italy) Geografia Fisica e Dinamica Qua-ternaria 38 25ndash33 httpsdoiorg104461GFDQ201538032015

Gagliardini O Gillet-Chaulet F Durand G Vincent C and Du-val P Estimating the risk of glacier cavity collapse during arti-ficial drainage The case of Tecircte Rousse Glacier Geophys ResLett 38 L10505 httpsdoiorg1010292011GL047536 2011

Garavaglia V Diolaiuti G A Smiraglia C Pasquale V andPelfini M Evaluating Tourist Perception of EnvironmentalChanges as a Contribution to Managing Natural Resources inGlacierized areas A Case Study of the Forni Glacier (StelvioNational Park Italian Alps) Environ Manage 50 1125ndash1138httpsdoiorg101007s00267-012-9948-9 2012

Gardent M Rabatel A Dedieu J-P and Deline P Mul-titemporal glacier inventory of the French Alps from thelate 1960s to the late 2000s Global Planet Change 120 24ndash37httpsdoiorg101016jgloplacha201405004 2014

Gindraux S Boesch R and Farinotti D Accuracy As-sessment of Digital Surface Models from Unmanned AerialVehiclesrsquo Imagery on Glaciers Remote Sensing 9 2ndash15httpsdoiorg103390rs9020186 2017

Gobiet A Kotlarski S Beniston M Heinrich G RajczakJ Stoffel M 21st century climate change in the Euro-pean Alps ndash A review Sci Total Environ 493 1138ndash1151httpsdoiorg101016jscitotenv201307050 2014

Harris C Arenson L U Christiansen H H EtzelmuellerB Frauenfelder R Gruber S Haeberli W Hauck CHoelzle M Humlum O Isaksen K Kaab A Kern-Luetschg M Lehning M Matsuoka N Murton J BNoetzli J Phillips M Ross N Seppaelae M Spring-man S M and Vonder Muehll D Permafrost and climatein Europe Monitoring and modelling thermal geomorpholog-ical and geotechnical responses Earth-Sci Rev 92 117ndash171httpsdoiorg101016jearscirev200812002 2009

Hoffmann-Wellenhof B Lichtenegger H and Wasle E GNSSndash GPS GLONASS Galileo amp more Springer Vienna Austriahttpsdoiorg101007978-3-211-73017-1 2008

Immerzeel W W Kraaijenbrink P D A Shea J M ShresthaA B Pellicciotti F Bierkens M F P and de Jong S MHigh-resolution monitoring of Himalayan glacier dynamics us-ing unmanned aerial vehicles Remote Sens Environ 150 93ndash103 httpsdoiorg101016jrse201404025 2014

Janke J R Using airborne LiDAR and USGS DEM data for as-sessing rock glaciers and glaciers Geomorphology 195 118ndash130 httpsdoiorg101016jgeomorph201304036 2013

Jokinen O and Geist T Accuracy aspects in topographicalchange detection of glacier surface in Remote sensing ofglaciers CRC PressBalkema Leiden the Netherlands 269ndash283 httpsdoiorg101201b10155-15 2010

Kaab A Huggel C Fischer L Guex S Paul F Roer I Salz-mann N Schlaefli S Schmutz K Schneider D Strozzi Tand Weidmann Y Remote sensing of glacier- and permafrost-related hazards in high mountains an overview Nat HazardsEarth Syst Sci 5 527ndash554 httpsdoiorg105194nhess-5-527-2005 2005a



Kaab A Reynolds J M and Haeberli W Glacier and Per-mafrost hazards in high mountains in Global Change andMountain Regions Advances in Global Change Research editedby Huber U M Bugmann H K M and Reasoner M ASpringer Dordrecht 225ndash234 httpsdoiorg1010071-4020-3508-X_23 2005b

Kaufmann V and Ladstaumldter R Application of terrestrial pho-togrammetry for glacier monitoring in Alpine environments inInternational Archives of the Photogrammetry Remote Sensingand Spatial Information Sciences Beijing China 37-B8 813ndash818 2008

Kaufmann V and Seier G Long-term monitoring of glacierchange at Goumlssnitzkees (Austria) using terrestrial photogram-metry in The International Archives of the PhotogrammetryRemote Sensing and Spatial Information Sciences XXIII IS-PRS Congress 12ndash19 July 2016 Prague Czech Republic41-B8 495ndash502 httpsdoiorg105194isprs-archives-XLI-B8-495-2016 2016

Keiler M Knight J and Harrison S Climate changeand geomorphological hazards in the eastern Euro-pean Alps Philos T Roy Soc A 368 2461ndash2479httpsdoiorg101098rsta20100047 2010

Kellerer-Pirklbauer A Bauer A and Proske H Terrestriallaser scanning for glacier monitoring Glaciation changes ofthe Goumlszlignitzkees glacier (Schober group Austria) between 2000and 2004 in 3rd Symposion of the Hohe Tauern National Parkfor research in protected areas 15ndash17 September 2005 castle ofKaprun Austria 97ndash106 2005

Lague D Brodu N and Leroux J Accurate 3D comparison ofcomplex topography with terrestrial laser scanner application tothe Rangitikei canyon (N-Z) J Photogram Remote Sens 8210ndash26 httpsdoiorg101016jisprsjprs201304009 2013

Matese A Toscano P Di Gennaro S F Genesio L Vaccari FP Primicerio J Belli C Zaldei A Bianconi R and Gioli BIntercomparison of UAV Aircraft and Satellite Remote SensingPlatforms for Precision Viticulture Remote Sensing 7 2971ndash2990 httpsdoiorg103390rs70302971 2015

Moelg N and Bolch T Structure-from-Motion Using HistoricalAerial Images to Analyse Changes in Glacier Surface ElevationRemote Sensing 9 1021 httpsdoiorg103390rs91010212017

Naumann M Geist M Bill R Niemeyer F and Grenzdo-erffer G Accuracy comparison of digital surface models cre-ated by Unmanned Aerial Systems imagery and Terrestrial LaserScanner in International Archives of the Photogrammetry Re-mote Sensing and Spatial Information Sciences UAV-g20134ndash6 September 2013 Rostock Germany 61-W2 281ndash286httpsdoiorg105194isprsarchives-XL-1-W2-281-2013 2013

Nuth C and Kaab A Co-registration and bias corrections of satel-lite elevation data sets for quantifying glacier thickness changeThe Cryosphere 5 271ndash290 httpsdoiorg105194tc-5-271-2011 2011

Oborne M Mission planner software available at httpardupilotorgplanner (last access 18 May 2017) 2013

OrsquoConnor J Smith M J and James M R Camerasand settings for aerial surveys in the geosciences op-timising image data Prog Phys Geogr 41 1ndash20httpsdoiorg1011770309133317703092 2017

Palomo I Climate Change Impacts on Ecosystem Servicesin High Mountain Areas A Literature Review Mount ResDev 37 179ndash187 httpsdoiorg101659MRD-JOURNAL-D-16-001101 2017

Piermattei L Carturan L and Guarnieri A Use of terrestrialphotogrammetry based on structure from motion for mass bal-ance estimation of a small glacier in the Italian Alps Earth SurfProc Land 40 1791ndash1802 httpsdoiorg101002esp37562015

Piermattei L Carturan L de Blasi F Tarolli P Dalla FontanaG Vettore A and Pfeifer N Suitability of ground-based SfMndashMVS for monitoring glacial and periglacial processes EarthSurf Dynam 4 325ndash443 httpsdoiorg105194esurf-4-425-2016 2016

Pomerleau F Colas F Siegwart R and Magnenat S Compar-ing ICP variants on real world data sets Autonomous Robots 34133ndash148 httpsdoiorg101007s10514-013-9327-2 2013

Quincey D J Lucas R M Richardson S D Glasser NF Hambrey N J and Reynolds J M Optical remotesensing techniques in high-mountain environments applica-tion to glacial hazards Prog Phys Geogr 29 475ndash505httpsdoiorg1011910309133305pp456ra 2005

Rayburg S Thoms M and Neave M A compari-son of digital elevation models generated from dif-ferent data sources Geomorphology 106 261ndash270httpsdoiorg101016jgeomorph200811007 2009

Riccardi A Vassena G Scotti R and Sgrenzaroli M Recentevolution of the punta S Matteo serac (Ortles-Cevedale GroupItalian Alps) Geografia Fisica e Dinamica Quaternaria 33 215ndash219 2010

Rolstad C Haug T and Denby B Spatially integratedgeodetic glacier mass balance and its uncertainty basedon geostatistical analysis application to the westernSvartisen ice cap Norway J Glaciol 55 666ndash680httpsdoiorg103189172756409787769528 2009

Rounce D R Watson C S and McKinney D C Identificationof Hazard and Risk for Glacial Lakes in the Nepal Himalaya Us-ing Satellite Imagery from 2000ndash2015 Remote Sensing 9 654httpsdoiorg103390rs9070654 2017

Ryan J C Hubbard A Box J E Brough S Cameron KCook J M Cooper M Doyle S H Edwards A Holt TIrvine-Fynn T Jones C Pitcher L H Rennermalm A KSmith L C Stibal M and Snooke N Derivation of HighSpatial Resolution Albedo from UAV Digital Imagery Appli-cation over the Greenland Ice Sheet Front Earth Sci 5 1ndash18httpsdoiorg103389feart201700040 2017

Seier G Kellerer-Pirklbauer A Wecht M Hirschmann SKaufmann V Lieb G K and Sulzer W UAS-BasedChange Detection of the Glacial and Proglacial TransitionZone at Pasterze Glacier Austria Remote Sensing 9 549httpsdoiorg103390rs9060549 2017

Senese A Diolaiuti G A Mihalcea C and Smiraglia CEnergy and Mass Balance of Forni Glacier (Stelvio Na-tional Park Italian Alps) from a Four-Year Meteorologi-cal Data Record Arct Antarct Alp Res 44 122ndash134httpsdoiorg1016571938-4246-441122 2012

Smiraglia C Azzoni R S DrsquoAgata C Maragno D FugazzaD and Diolaiuti G A The evolution of the Italian glaciers fromthe previous data base to the new Italian inventory Preliminary



considerations and results Geogr Fis Dinam Quat 38 79ndash87httpsdoiorg104461GFDQ20153808 2015

Teunissen P J G Testing theory An introduction in Series onMathematical Geodesy and Positioning VSSD Delft Delft theNetherlands 2009

Urbini S Zirizzotti A Baskaradas J A Tabacco I E CafarellaL Senese A Smiraglia C and Diolaiuti G Airborne radioecho sounding (RES) measures on alpine glaciers to evaluate icethickness and bedrock geometry Preliminary results from pilottests performed in the ortles-cevedale group (Italian alps) AnnGeophys 60 G0226 httpsdoiorg104401ag-7122 2017

Vincent C Auclair S and Le Meur E Outburst flood hazardfor glacier-dammed Lac de Rochemelon France J Glaciol 5691ndash100 httpsdoiorg103189002214310791190857 2010

Vincent C Thibert E Harter M Soruco A and GilbertA Volume and frequency of ice avalanches from Tacon-naz hanging glacier French Alps Ann Glaciol 56 17ndash25httpsdoiorg1031892015AoG70A017 2015

Westoby M J Brasington J Glasser N F HambreyM J and Reynolds J M Structure-from-Motionrsquophotogrammetry A low-cost effective tool for geo-science applications Geomorphology 179 300ndash314httpsdoiorg101016jgeomorph201208021 2012

Winkler M Pfeffer W T and Hanke K Kilimanjaro ice cliffmonitoring with close range photogrammetry in InternationalArchives of the Photogrammetry Remote Sensing and Spa-tial Information Sciences XXII ISPRS Congress 25 Augustndash1 September 2012 Melbourne Australia 39-B5 441ndash446 2012


Abstract

Introduction

Study area

Data sources acquisition and processing

UAV photogrammetry

2014 dataset

2016 dataset

Terrestrial photogrammetry

Terrestrial laser scanning

GNSS ground control points

2007 DEM

Methods

Analysis of point clouds from the 2016 campaign UAVterrestrial photogrammetry and TLS

Point cloud merging

Glacier hazard mapping

DEM co-registration for glacier thickness change estimation

Results

Point cloud analysis

Glacier-related hazards and risks

Glacier thickness change

Discussion comparison of techniques for point cloud generation

Point density and completeness

Point cloud and DEM uncertainty

Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests

Special issue statement

Acknowledgements

References


ing the exposure of unconsolidated and ice-cored sediments(Keiler et al 2010 Chiarle et al 2007) Glacier downwast-ing causes changes in water resources with an initial increasein discharge due to enhanced melt followed by a long-termreduction affecting drinking water supply irrigation and hy-dropower production (Kaab et al 2005b) along with a risingoccurrence of structural collapses (Azzoni et al 2017) Fi-nally glacier retreat and the increase in glacier hazards bothnegatively influence the tourism sector and the economicprosperity of high mountain regions (Palomo 2017)

The growing threat from cryospheric hazards under cli-mate change calls for the adoption of mitigation strategiesRemote sensing has long been recognized as an importanttool for producing supporting data for this purpose such asdigital elevation models (DEMs) and multispectral imagesDEMs are particularly useful for detecting glacier thicknessand volume variations (Fischer et al 2015 Berthier et al2016) and for identifying steep areas that are most prone togeomorphodynamic changes such as mass movements (Bla-sone et al 2015) Multispectral images at a sufficient spatialresolution make it possible to recognize most cryospherichazards (Quincey et al 2005 Kaab et al 2005b) Whilesatellite images from Landsat and ASTER sensors (15ndash30 mground sample distance ndash GSD) are practical for regional-scale mapping (Rounce et al 2017) the assessment of haz-ards at the scale of individual glaciers or basins requires ahigher spatial resolution which in the past could only beachieved via aerial laser scanner and photogrammetric sur-veys (Vincent et al 2010 Janke 2013) or dedicated fieldcampaigns with terrestrial laser scanners (TLS) (Kellerer-Pirklbauer et al 2005 Riccardi et al 2010) Recent yearshave seen a resurgence of terrestrial photogrammetric sur-veys for the generation of DEMs (Piermattei et al 20152016 Kaufmann and Seier 2016) due to important tech-nological advances including the development of structure-from-motion (SfM) photogrammetry and its implementationin fully automatic processing software as well as improve-ments in the quality of camera sensors (Eltner et al 2016Westoby et al 2012) In parallel unmanned aerial vehicles(UAVs ndash Colomina and Molina 2014 OrsquoConnor et al 2017)have started to emerge as a viable alternative to TLS formultitemporal monitoring of small areas UAVs promise tobridge the gap between field observations notoriously diffi-cult on glaciers and coarser-resolution satellite data (Bhard-waj et al 2016) Although the number of studies employingthese platforms in high mountain environments is on the rise(see eg Fugazza et al 2015 Gindraux et al 2017 Seier etal 2017) their full potential for monitoring glaciers and par-ticularly glacier hazards has yet to be explored In particularthe advantages of UAV and terrestrial SfM photogrammetryand the possibility of data fusion and volume change esti-mation to support hazard management strategies in glacialenvironments need to be investigated and assessed

In this study we investigated a rapidly downwastingglacier (almost 5 m aminus1 water equivalent Senese et al 2012)in a protected area and highly touristic sector of the Ital-ian Alps Stelvio National Park We focused on the glacierterminus and the hazards identified there ie the formationof normal faults and ring faults The former occur mainlyon the medial moraines and glacier terminus and are due togravitational collapse of debris-laden slopes The latter de-velop as a series of circular or semicircular fractures withstepwise subsidence caused by englacial or subglacial melt-water creating voids at the icendashbedrock interface eventuallyleading to the collapse of the cavity roofs While often over-looked these collapse structures are particularly hazardousfor mountaineers and they are likely to increase under a cli-mate change scenario (Azzoni et al 2017) They are moredangerous than crevasses because of their larger size

We conducted our first UAV survey of the glacier in 2014in the summer of 2016 the glacier was surveyed using threedifferent techniques for the generation of point clouds DEMsand orthophotos The aims were (1) to compare the differentmethods and select the most appropriate ones for monitor-ing glacier hazards (2) to identify glacier-related hazards andtheir evolution between 2014 and 2016 and (3) to investigatechanges in ice thickness between 2014 and 2016 and between2007 and 2016 by comparing the two UAV DEMs and a thirdDEM obtained from stereo-processing of aerial photos cap-tured in 2007

2 Study area

The Forni Glacier (see Fig 1) has an area of 1134 km2 basedon the 2007 data from the Italian Glacier Inventory (Smi-raglia et al 2015) an altitudinal range between 2501 and3673 m asl and a north-northwesterly aspect The glacierhas retreated markedly since the Little Ice Age when itsarea was 1780 km2 (Diolaiuti and Smiraglia 2010) withan acceleration of the shrinkage rate over the past threedecades typical of valley glaciers in the Alps (Diolaiuti et al2012 DrsquoAgata et al 2014) It has also undergone profoundchanges in dynamics in recent years such as the loss of iceflow from the eastern accumulation basin towards its tongueand the evidence of collapsing areas on the eastern tongue(see Fig 2d Azzoni et al 2017) Continuous monitoring ofthese hazards is important as the site is highly touristic (Gar-avaglia et al 2012) The glacier is in fact frequently visitedduring both summer and winter months During the summerhikers heading to Mount San Matteo take the trail along thecentral tongue accessing the glacier through the left flank ofthe collapsing glacier terminus (see Fig 2b and c) Duringwintertime ski mountaineers instead access the glacier fromthe eastern side crossing the medial moraine and potentiallycollapsed areas there (see Figs 1 and 2a)







311 2014 dataset



312 2016 dataset





















35 2007 DEM





4 Methods



























5 Results



























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References






311 2014 dataset



312 2016 dataset





















35 2007 DEM





4 Methods



























5 Results



























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References


















35 2007 DEM





4 Methods



























5 Results



























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References



4 Methods



























5 Results



























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References





















5 Results



























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References























































7 Conclusions
















References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References









References












































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References

































































Abstract

Introduction

Study area


UAV photogrammetry

2014 dataset

2016 dataset




2007 DEM

Methods


Point cloud merging



Results







Logistics and costs

Additional remarks

Conclusions

Data availability

Competing interests


Acknowledgements

References

combination of uav and terrestrial photogrammetry to assess … · combination of uav and...

Documents