Baseline data for monitoring geomorphological effects of glacier lake outburst flood: A very high-resolution image and GIS datasets of the distal part of the Zackenberg River, northeast Greenland

The Arctic regions experience intense transformations, such that efficient methods are needed to monitor and understand Arctic landscape changes in response to climate warming and low-frequency high-magnitude events. One example of such events, capable of causing serious landscape changes, is glacier lake outburst floods. On 6 August 2017, a flood event related to glacial 10 lake outburst affected the Zackenberg River (NE Greenland). Here, we provided a very high-resolution dataset representing unique time-series of data captured immediately before (5 August 2017), during (6 August 2017), and after (8 August 2017) the flood. Our dataset covers a 2.1-km-long distal section of the Zackenberg River. The available files comprise: (1) unprocessed images captured using an unmanned aerial vehicle (UAV): https://doi.org/10.5281/zenodo.4495282 (Tomczyk and Ewertowski, 2021a); and (2) results of structure-from-motion (SfM) processing (orthomosaics, digital elevation models, 15 and hillshade models in a raster format), uncertainty assessments (precision maps) and effects of geomorphological mapping in vector formats: https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b). Potential applications of the presented dataset include: (1) assessment and quantification of landscape changes as an immediate result of glacier lake outburst flood; (2) long-term monitoring of high-Arctic river valley development (in conjunction with other datasets); (3) establishing a baseline for quantification of geomorphological impacts of future glacier lake outburst floods; (4) assessment of 20 geohazards related to bank erosion and debris flow development (hazards for research station infrastructure – station buildings and bridge); (5) monitoring of permafrost degradation; and (6) modelling flood impacts on river ecosystem, transport capacity, and channel stability.

On 6 August 2017, a flood event related to a glacier lake outburst affected the Zackenberg River (NE Greenland), leaving behind serious substantial geomorphological impacts on the riverbanks and channel morphology (see . 65 Here, we provided a very high-resolution dataset representing time-series of data captured immediately before (5 August 2017), during (6 August 2017), and after (8 August 2017) the flood. This unique set of data makes it possible to study the immediate landscape response to the GLOF event and can be used as a baseline for any long-term monitoring exercise. Our dataset covers approximately a 2.1-km-long distal section of the Zackenberg River. Available files comprise: (1) unprocessed images captured using an unmanned aerial vehicle (UAV): https://doi.org/10.5281/zenodo.4495282 (Tomczyk and Ewertowski, 70 2021a); and (2) results of structure-from-motion (SfM) processing (orthomosaics, digital elevation models, and hillshade models in a raster format), uncertainty assessments (precision maps) and effects of geomorphological mapping in vector format: https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b). The availability of unprocessed images means that the potential user can derive their own photogrammetric products using more advanced technologies (potentially available in the future) to ensure coherence with future-collected monitoring data. 75 Potential applications of the presented dataset include: (1) assessment and quantification of landscape changes as an immediate result of glacier lake outburst flood ; (2) long-term monitoring of high-Arctic river valley development (in conjunction with other datasets); (3) establishing a baseline for quantification of geomorphological impacts of future glacier lake outburst floods; (4) assessment of geohazards related to bank erosion and 80 debris flow development (hazards for research station infrastructurestation buildings and bridge); (5) monitoring of permafrost degradation; (6) modelling flood impacts on river ecosystem, transport capacity, and channel stability.
Typical discharges during summer month were from 20 m 3 s −1 to 50 m 3 s −1 , and usually lower at the end of melting season (September-October) (Søndergaard et al., 2015;Ladegaard-Pedersen et al., 2017). One of the Zackenberg River's characteristics is regular floods during summer related to sudden lake drainageprobably due to rupture of the glacier dam 90 (see Jensen et al., 2013;Behm et al., 2017;Behm et al., 2020). Between 1996 and 2018, 14 extreme flood events with discharges of over 100 m 3 s -1 were recorded , while two additional ones were observed in the winter period . Such events had an enormous impact on the riverscape geomorphology , discharge and sediment transport (Hasholt et al., 2008;Søndergaard et al., 2015;Ladegaard-Pedersen et al., 2017), and delivery of nutrients and sediments into the fiords and delta development 4 Kroon et al., 2017). In this context, the given dataset aims to establish a baseline for monitoring the consequences of future extreme floods by documenting the state of the riverscape before, during, and after the 2017 glacier lake outburst flood.

UAV surveys
According to the guidelines for using structure-from-motion (SfM) photogrammetry in geomorphological research (see James 100 et al., 2019), details about UAV surveys are presented in Sect. 2.2, and the parameters used for SfM processing are detailed in Sect. 3. In that way, other researchers can use the data to replicate our results; alternatively, as new approaches become available, novel processing methods can be utilised. Figure 1: Location of the study area (Reprinted from Journal of Hydrology, Vol 591, Tomczyk et al., Geomorphological impacts of a glacier lake outburst flood in the high arctic Zackenberg River, NE Greenland, 125300, Copyright (2020), with permission from Elsevier).

Rationale 110
There were three primary goals for conducting the UAV surveys: (1) collect data that would enable quantifying medium-term (i.e., temporal scale of several years) changes in the river landscapecompared to the available high-resolution 2014 data (COWI, 2015); (2) to document river state and immediate landscape response during the 2017 flood; and (3) to establish a baseline for the monitoring of geomorphological changes in response to future glacier lake outburst floods, including potential geohazards to research infrastructure (i.e. bridge and buildings of the research station). To achieve these aims, it was necessary 115 to collect data with high spatial resolution, preferably better than 0.05 m ground sampling distance (GSD) (Fig. 2), covering a relatively 2.1-km-long section (2.1 km) of the river from the bridge to the delta. We decided to use a small portable UAV as it was more economical in terms of financial and time requirements compared to terrestrial laser scanning (TSL).

Equipment
We used a lightweight, consumer-grade UAVmultirotor DJI Phantom 4 Pro. The little weight (1.4 kg) combined with a 120 small size (0.35 m diagonal) ensure that the UAV could be easily transported in the field using a backpackthis was essential, as mechanised transport is not allowed due to fragility of the vegetation. The UAV was equipped with a DJI 20MP, 1-inch size CMOS RGB sensor and a global shuttercamera model FC6310 (Table 1). There was a prime lens with 8.8 mm focal length (24 mm equivalent for 35 mm), aperture range from f/2.8 to f/11 and autofocus. A 3-axis (pitch, roll, yaw) gimbal stabilised the camera, enabling it to take sharp pictures while the craft was in motion. The UAV was equipped with a global navigation 125 satellite system (GNSS) receiver, capable of receiving signals from GPS and GLONASS satellite positioning systems.

Survey design and execution
To collect the necessary data, we designed an initial survey plan comprised of five lines approximately parallel to the main river channel's course routed over the centre of the main channel and both banks. During the surveys, this design was modified, as the river sections containing meandering segments were too wide to be captured with five lines of images with necessary 130 overlap. Therefore, we turned to surveying N-S lines of the images, covering both the river channel and its neighbourhood. Individual flights were operated manually, using DJI GO 4 app for Android, for in such high latitudes onboard GNSS and magnetometer were potentially prone to erroneous reading. Related unexpected behaviours (e.g., errors in compass reading or lost GNSS signal) were easier to tackle in the manual than automated mode. We captured mostly nadir images with a high 140 overlap (> 80%). Additional oblique images were collected to cover the steep, near-vertical riverbank sections so as to ensure their proper representation in the model. Due to the length of the studied river section, and to comply with the visual line of sight (VLOS) flight operations, three take-off/landing sites were used each day. The weather condition for each day was good (i.e., no precipitation nor strong winds), and illumination conditions were sunny. The UAV surveys were performed at average nominal altitudes (from 70 to 110 m above ground level) to achieve the desired GSD (Table 1). In total, 1972 images were 145 taken on 5 August 2017 (before-flood dataset), 887 images on 6 August 2017 (during-flood dataset), and 1929 images on 8 August 2017 (after-flood dataset). As the river level was fluctuating during the flood (6 August survey), we used a higher flight altitude, which translated into a lower number of images captured on 6 August, but enabled us to cover the area more quickly with approximately the same water level during the survey. SoTherefore, it was a compromise between photogrammetric quality (i.e., the image network geometry), desired GSD, and rapidly changing flood conditions. 150 The unprocessed images captured during the surveys are available at: https://doi.org/10.5281/zenodo.4495282 (Tomczyk and Ewertowski, 2021a). They can be used by interested parties to generate their own photogrammetric products using different methods and/or software than those described in Sect. 3.

Structure-from-Motion processing
The UAV-captured images were processed using Agisoft Metashape Professional Edition 1.5.2. The values used for processing settings in each step were the following: 1) Camera settingscamera type: Frame; enable rolling shutter compensation: unchecked (as the UAV was equipped with global shutter) 160 2) Images alignment and sparse point cloud generationaccuracy: High; generic preselection: Yes; reference preselection: Yes; key point limit: 100,000; tie point limit: 0 (i.e., unlimited) 3) Gradual selection and removal of the outliers and erroneous pointsthree-stage selection based on reconstruction uncertainty: 10; reprojection error: 0.5; projection accuracy: 6 4) Optimisation of the sparse point cloudparameters: f, b1, b2, cx, cy, k1, k2, p1, p2 165 5) Dense point cloud generationquality: High, Depth filtering: Aggressive 6) DEM generationsource data: Dense Cloud, Interpolation: Enabled The external orientation of the reconstructed scene was established using coordinates of each camera position obtained from the onboard GNSS system. To further constrain the geometry of the scene, additional control points were used (CPs). As we were not able to collect high-quality ground control points (we did not have access to cm-accuracy survey equipment, and it 170 was not possible to cross the river during the flood, because of the high water level), control points (CPs) were then generated post-survey using previous UAV dataset from 2014(COWI, 2015. In total, 100 points were selected, located mostly on stable, flat boulders, which were easy to identify in the images. CPs were distributed on level terrain to minimize the impact of potential permafrost creep. Distribution of CPs was along both sides of the river to ensure that the distance between individual points is less than 100 m, which was suggested as optimal by Tonkin and Midgley (2016). The projection used was UTM 27N. 175 The number of points used as "control" to optimise the exterior orientation was: 61 (5 August), 57 (6 August), 61 (8 August).
The remaining points were used as independent checkpoints: 39 (5 August), 21 (6 August), 22 (8 August). A smaller number of points used for data collected on 6 August and 8 August were related to differences in coverage.

SfM processing results
The produced tie points clouds consisted of between 1.2 million (6 and 8 August) and 1.4 million (5 August) filtered points, 180 with low tie point reprojection errors from 0.28 to 0.44 pixels, which was indicative of the high quality of the image geometry network (Table 1). Dense cloud point density varied from 322 points m -2 (6 August) to 778 points m -2 . These translated to orthomosaics with GSDs from 0.018 m (5 August) to 0.028 m (6 August), and DEMs with GSDs from 0.036 m to 0.056 m (Fig. 3). RMS discrepancies for control points and checkpoints were between 0.12 m and 0.15 m, which was expected, as the control and checkpoints were transferred from previously existing data. The coherence between models was also estimated 185 based on test areas selected in stable fragments of moraine and palaeo-delta to ensure significant systematic differences in elevation between datasets do not exist. The final products of SfM processing (Orthomosaic and DEMs) and their derivative (hillshade models) for each data are available at: https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b).

Mapping
The mapping process was based on the approach proposed by Chandler et al. (2018), i.e. identification and interpretation of the geomorphological features were based on a combined analysis of remote sensing products and their derivatives 195 (orthomosaics, DEMs, slope maps, hillshade models) as well as ground-based truthing. Final vector shapefile datasets were vectorised on-screen in ArcMap 10.6 software. The main geomorphological units (e.g., relict fluvial terraces, modern floodplain, slopes) and areas affected by mass movements of various types (e.g., debris flows, debris slumps) were mapped as polygons. Additional layers of polylines included features such as scarps or thermal contraction cracks. River extent (i.e., area covered by water) is provided for each day as a separate polygon layer. Geomorphological features are provided as a separate 200 file for before-the-flood (5 August 2017) and after-the-flood (8 August 2017) dataset. The mapping results in the form of vector files in the shp format (compatible with most GIS software) are available to download from at: https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b). Vector data combined with the hillshade models were presented as a series of geomorphological maps (see Tomczyk and Ewertowski, 2020 for details) The quality of the presented datasets was assessed in relation to the outside world (i.e., external or absolute accuracy) and in relation to each survey (internal precision). Quality assessment based on data presented in Table 1 indicates a high quality of internal image network geometry, illustrated by low sub-pixel values of tie points reprojection errors. The external accuracy was estimated based on root-mean-square errors (RMSE) and standard deviations (SD) of errors on checkpoints, which were between 0.12 and 0.15 m ( Table 1) Moreover, if better relative accuracy (i.e. survey-to-survey accuracy) is necessary in the future monitoring applications, coalignment of UAV time-series can provide better relative accuracy than the classic approach of individual SfM processing of each survey using GCP -as demonstrated in several studies (e.g., Feurer and Vinatier, 2018;Cook and Dietze, 2019;de Haas et al., 2021). Therefore, we provided also unprocessed images so the potential user can perform their own SfM processing. 225 The internal quality of the reconstructed scenes was based on tie point precision. To estimate the spatial variability of the models' photogrammetric and georeferencing uncertainties, the precision estimates for sparse point clouds were generated in Agisoft Metashap and exported using the Python script provided by James et al. (2020). The precision analysis indicated that the vertical component was less spatially consistent than the horizontal ones for all three surveys (Fig. 4). For the models' 230 ground parts, the overall precision was limited by the precision of control points, which is not surprising as they were derived from the older, less detailed remote sensing dataset. The internal accuracy of each survey was assessed based on The the mean point precision estimates, which varied from 4 to 6 mm for the horizontal component and from 11 to 15 mm for the vertical one (Table 1)the weakest values were for the 6 August 2017 dataset, which was expected as the average flying altitude was highest then. Precision maps are available to download from at: https://doi.org/10.5281/zenodo.4498296 (Tomczyk and 235 Ewertowski, 2021b). Z discrepancies on control points were calculated using Doming Analysis software (v.1.0) (James et al., 2020). The analysis indicated no doming distortion (Fig. 5), which is probably related to the generally very high overlap of images and the inclusion of oblique images of the steep riverbanks.

250
Individual orthomosaics and DEMs were also inspected, resulting in the discovery of the following several problems, which ought to be taken into account in any future analysis: 1) In general, the interpretation of riverbank conditions can be tampered by vegetation cover and/or bank undercutting (Niedzielski et al., 2016;Hemmelder et al., 2018). The proposed solutions included taking the mean elevation value of the bank in between the vegetated areas and then using it as the reference height (Hemmelder et al., 2018) or 255 interpolating a line between the last exposed sections of the riverbank, not covered by trees and bushes (Niedzielski et al., 2016). While vegetation cover is usually not a problem in the case of Arctic rivers, such an approach might be useful when other obstacles (e.g., shadows, infrastructure) might prevent the direct measurements of the bank's heights. In the case of the presented dataset, some sections of riverbanks were steep, near-vertical, before the flood (Fig. 6a, 7). However, during the flood, some of the sections were significantly undercut, forming deeply incised 260 nichesthese overhanging banks obstructed the view of the bottom part of some studied sections from the air. During the UAV campaigns, we took oblique images to at least produce a proper representation of steep slopes; however, it was not possible to take horizontal images due to the presence of water. As a result, it was impossible to calculate the volume of sediments eroded from the niches under these overhanging sections.
2) Structure-from-motion is based on reconstructing the image network geometry based on characteristic points that 265 appear in several images (Westoby et al., 2012). It therefore fails where there are rapidly moving objects, which changed their position in time between the images captured. The structure-from-motion photogrammetry can reconstruct the location of points in dry areas, and, in the case of transparent water, also points located underwater (Carrivick and Smith, 2019). However, in our study, the high turbidity of water and sediment suspension prevented viewing of the riverbed. As an Arctic river, the Zackenberg River has suspended sediment concentrations within a 270 range of 50 to 500 mg L −1 (Søndergaard et al., 2015), which can increase even up to 4000 mg L −1 during glacial lake outburst floods, indicated by the lack of transparency and the yellowish yellow or brownish brown colours of water in the orthomosaics. The turbidity of water is also very high (Ladegaard-Pedersen et al., 2017), as was also found in our surveys. The fact that the water surface was full of ripples gave rise to bi-directional reflectance problems. Therefore, it was not possible to adequately resolve the surface of flowing water (Fig. 6b). To partly address this 275 issue, the water bodies were masked from DEMs and hillshade models. They are, however, visible in orthomosaics which enables the user to assess the character of water flow (Fig. 8).
3) Some fragments of the models revealed artefacts associated with mismatches in points generation. These areas can generate erroneous elevation values, which can be identified in the DEM and hillshade model as unexpectedly rough surfaces in places where the ground level should be uniform (Fig. 6c). These areas were indicated with polygon files 280 for easy identification in case of future analysis.

Data and code availability
All described data are available in the Zenodo repository. The structure of the dataset is as follows: 1) Unprocessed UAV-captured images (~46 GB) are available at: https://doi.org/10.5281/zenodo.4495282 (Tomczyk and Ewertowski, 2021a). The images are zipped into three folders following naming convention: 295 2017_08_05_before_flood_unprocessed_UAV_images, 2017_08_06_during_flood_unprocessed_UAV_images, 2017_08_08_after_flood_unprocessed_UAV_images. The images are in jpg format and contain embedded positions in geographic coordinate system WGS84 obtained from the on-board GNSS receiver.
2) The results of photogrammetric processing (~18 GB) are available at https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b) in the file Sfm_products.zip, and are grouped into subfolders with the following 300 names: dem (containing digital elevation models), orthomosaic (containing orthomosaics), hs (containing hillshade models); all data are in GeoTIFF format in the UTM 27N projected coordinate system. Individual files are named as 3) The mapping results (25 MB) are in the same repository entry as SfM processing results, i.e. at https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021b) in the folder "mapping.zip". Inside, there are four sub-folders: a. Generalcontains general vectors that did not change over the course of three days (e.g., station buildings, 310 4x4 trail) b. River_extentcontains polygons for river extent for 2014 (generated from older UAV data (COWI, 2015)), and for 05, 06, and 08 August 2017. The 2017 data are named as yyyy_mm_dd_river c. Before_flood_geomorphologycontains polygon and lines illustrating geomorphological features before the flood with separate files providing extent of mass movements which can be potentially hazardous e.g., 315 debris flows, debris falls, rockfalls, slumps (names of individual files are provided in Table 2) d. After_flood_geomorphologycontains polygon and lines illustrating geomorphological features after the flood with separate files providing extent of mass movements which can be potentially hazardous e.g., debris flows, debris falls, rockfalls, slumps (names of individual files are provided in Table 2) All data are in shp vector format in the UTM 27N projected coordinate system. 320 4) Precision estimates for tie points and precision maps for X, Y, Z coordinates are in the same repository entry as SfM processing results, i.e. at https://doi.org/10.5281/zenodo.4498296 (Tomczyk and Ewertowski, 2021a) in the folder " uncertainty_assessment.zip". Individual files are named as follows:  (James et al., 2020) are available to download from https://www.lancaster.ac.uk/staff/jamesm/software/sfm_georef.htm.  The ability to detect changes in the geomorphology of the riverbed and riparian areas remains a crucial issue in monitoring and modelling the geomorphic effects of flood events. Using a UAV survey for rapid assessment, as in the case of the studied 2017 flood, can be more beneficial than other methods (like high-resolution satellite imagery, terrestrial laser scanning) (cf. Smith et al., 2016), as it allows for covering the substantial length of the river with high-resolution data.
Such data are intended to be a baseline for future monitoring projects. Potential applications of the presented dataset include: 340 1) Establishing a long-term monitoring of high-Arctic river valley development in a permafrost terrainclimate warming in the Arctic is more intense than in other regions (see Moritz et al., 2002;Walsh et al., 2011;Duarte et al., 2012), with the thawing of permafrost in Greenland being one of the effects (Elberling et al., 2013;Anderson et al., 2017). In such a dynamic environment, riverscapes are also likely to transform rapidly (Chassiot et al., 2020). As our data covers river section located close to the Zackenberg Research Station, it facilitates logistics and can potentially 345 enable developing long-term remote sensing data series illustrating the dynamic response of the riverscape to ongoing climate change, which is essential from the standpoint of long-term landscape evolution.
2) Quantification, monitoring and modelling of geomorphological impacts of glacier lake outburst floodthe presented dataset was meant to quantify changes related to the 2017 GLOF (see ; however, these studies only described the immediate impacts of a single flood event. An example of 350 geomorphic change detection is presented in Fig. 9, demonstrating the acceleration of debris flows resulting from sediment entrainment at the base of the river bans by floodwater. Overall, the observed changes were spatially variable erosion dominated along steep banks as expected; however, understanding of differences in erosion rates between sites requires further studies, which will consider differences in lithology as well as modelling of water flow to investigate potential erosion forces in relation to channel characteristics. The first GLOF at Zackenberg was observed 355 in 1996 and since then floods occurred every year or at the two-year interval . The lake, which is the source of GLOF, is located more than 3 km from the current ice margin, so we expect a similar or higher frequency as more water will be melting from glaciers and stored in the lake. Thus, future monitoring is needed to investigate whether the GLOFs will be observed more frequently but with lower discharge magnitude or less often but with higher discharge. 360 2)3) Processed-based modelling studies -As the high-magnitude low-frequency events are typically rare and difficult to predict, our understanding of the quantitative aspect of geomorphological changes related to them remains limited compared to the "normal" processes (Tamminga et al., 2015b). These arise particularly from difficulties in collecting high-resolution data before and after these innately unpredictable and rare flood events. However, investigation into the geomorphological response of river morphology to "extreme" events is key to understanding the evolution of 365 river morphology and being crucial from the standpoint of river modelling and monitoring (Tamminga et al., 2015a;Tamminga et al., 2015b). Moreover, the relationship between the magnitude of the flood and geomorphological effects is not fully understood. For example, in the case of Zackenberg River, immediate (2-days) lateral erosion compared to three-year erosion was spatially very diversified. In some sections, immediate lateral erosion after the 2017 flood reached up to 10 m, whereas the same section was stable between 2014 and 2017, even though higher 370 peak discharges characterised 2015 and 2016 GLOFs than 2017 GLOF . Further process-based studies are necessary to observe and model links between the magnitude of a flood and the severity of erosion. It is especially important in periglacial landscapes where lateral bank erosion can be responsible for delivering a large quantity of organic matter and widespread changes in ecosystems. especially combined with other weather extreme events (see Christensen et al., 2021). Using the provided dataset as a baseline for the monitoring of future changes, it 375 should be possible to quantify the difference between geomorphological effects of "normal" (i.e., high-frequency, low-magnitude) processes on the one hand, and extreme (i.e., low-frequency, high-magnitude) events on the other. Also, by linking the intensity of a geomorphological response to hydrological data about flood characteristics, it should be possible to improve modelling routines (cf. Carrivick, 2007a, b;Carrivick et al., 2011;Guan et al., 2015;Staines and Carrivick, 2015). 380 4) Geo-hazards assessmentthe Zackenberg Research Station premises are located close to the riverbank which is regularly affected by floods. The development of debris flows, which has started to threaten the Station's infrastructure, is one outcome of the removal of sediments from the channel by flood. Another example of geohazards is washing out the foundation of the bridge located up the valley. These hazards require regular monitoring to prevent damage to the infrastructure, and the presented database can be used to assess current hazards and establish a baseline 385 for future monitoring.