This dataset consists of (i) 673 red-green-blue (RGB) images, (ii) precise coordinates of ground control points and harvest plot corners, (iii) the photogrammetrically reconstructed dense point cloud (comprising of 228,315,000 points with XYZ and RGB values), (iv) four normalised difference vegetation index (NDVI) maps, (v) aboveground biomass data from harvest plots, and (vi) observations of the ground surface obtained from a walkover survey with a GNSS instrument. These data were collected over the eastern part of Qikiqtaruk - Herschel Island, in the Canadian Yukon (69.5N, 138.8W). The images were collected in July and August 2016. Further details on the image processing are provided in the lineage section. This dataset was created by Andrew Cunliffe, with support from Isla Myers-Smith, Jakob Assmann, Jeffery Kerby and Gergana Daskalova (https://teamshrub.com/), in order to inform ongoing ecological monitoring studies in this area.
This research was supported by the Natural Environment Research Council (NE/M016323/1), and the NERC Geophysical Equipment Facility (GEF:1063).
Aboveground biomass, Arctic Ecosystems, Proximal Remote Sensing, Shrubs, Structure-from-motion photogrammetry, Vegetation change
|Use Constraints:||This data is covered by a UK Open Government Licence (http://www.nationalarchives.gov.uk/doc/open-government-licence/version/3/). Further by downloading this data the user acknowledges that they agree with the NERC data policy (http://www.nerc.ac.uk/research/sites/data/policy.asp).
Please cite this dataset as follow:
Cunliffe, A., Myers-Smith, I., Kerby, J., Assmann, J., & Daskalova, G. (2020). Canopy height, spectral reflectance (NDVI), and aboveground biomass of Salix richardsonii across a wet graminoid-shrubland ecotone on Qikiqtaruk - Herschel Island, Yukon, Canada (2016) (Version 1.0) [Data set]. UK Polar Data Centre, Natural Environment Research Council, UK Research & Innovation. https://doi.org/10.5285/61C5097B-6717-4692-A8A4-D32CCA0E61A9
|ISO Topic Categories:||
|Organisation||British Antarctic Survey|
|Name||Dr Andrew M Cunliffe|
|Organisation||School of GeoSciences, University of Edinburgh|
|Name||Dr Isla Myers-Smith|
|Organisation||University of Edinburgh|
|Name||Dr Jeffrey T Kerby|
|Organisation||Neukom Institute for Computational Science, Dartmouth College|
|Name||Dr Jakob Assmann|
|Organisation||University of Edinburgh|
|Organisation||University of Edinburgh|
Cunliffe, A., J. Assmann, G. Daskalova, J. Kerby, I. Myers-Smith (2020) Aboveground biomass corresponds strongly with drone-derived canopy height but weakly with greenness (NDVI) in a shrub tundra landscape, Environ. Res. Lett. DOI:10.1088/1748-9326/aba470.
Assmann, J.J., Kerby, J.T., Cunliffe, A.M., Myers-Smith, I.H., 2018. Vegetation monitoring using multispectral sensors - best practices and lessons learned from high latitudes. J. Unmanned Veh. Syst. 334730. https://doi.org/10.1101/334730
Cunliffe, A., Anderson, K., 2019. Measuring Above-ground Biomass with Drone Photogrammetry: Data Collection Protocol. Protoc. Exch. https://doi.org/10.1038/protex.2018.134
Cunliffe, A.M., Brazier, R.E., Anderson, K., 2016. Ultra-fine grain landscape-scale quantification of dryland vegetation structure with drone-acquired structure-from-motion photogrammetry.Remote Sens. Environ. 183, 129-143. https://doi.org/10.1016/j.rse.2016.05.019
Kachamba, D.J., Ørka, H.O., Gobakken, T., Eid, T., Mwase, W., 2016. Biomass estimation using 3D data from unmanned aerial vehicle imagery in a tropical woodland. Remote Sens. 8, 968. https://doi.org/10.3390/rs8110968
Wallace, L., Hillman, S., Reinke, K., Hally, B., 2017. Non-destructive estimation of above-ground surface and near-surface biomass using 3D terrestrial remote sensing techniques. Methods Ecol. Evol. 8, 1607-1616. https://doi.org/10.1111/2041-210X.12759
|Lineage:||We have conducted our study at one of the key sites for ecological monitoring in the circumpolar Arctic. Tundra communities on Qikiqtaruk - Herschel Island in the Canadian Arctic range from graminoid- to shrub-dominated, and are underlain by organic soils and ice rich permafrost. This site has undergone marked ecological changes in community composition, increases in canopy height and vegetation abundance, decreases in bare ground, and an advance in leaf emergence and flowering over the 18 years of ecological monitoring to date (Myers-Smith et al., 2019). We established a ca. 2 ha-1 site across a graminoid-shrub ecotone at the edge of a willow shrub dominated alluvial plain (69.34N, 138.53W).
To constrain the photogrammetric modelling and locate the point clouds in a coordinate reference system, 26 ground control markers (265 mm x 265 mm) were deployed across the site and geolocated to a relative 3D accuracy of <0.015 m with RTK-GNSS equipment (Leica GS10). Coordinates were relative to a local benchmark, geolocated in absolute terms to ±0.003 m in X and Y. and ±0.008 m in Z (95% confidence interval), using the AUSPOS web service. The markers were situated so as to be visible from the air, and a high density of markers facilitated image alignment in the texturally complex scenes. We selected 36 square plots of 50 cm × 50 cm for harvesting. The plots were arranged in a twelve blocks of three replicates across the range of canopy heights to allow a detailed assessment of the form of the allometric relationships. The corners of each harvest plot were geolocated using the GNSS. To minimise sinkage of the GNSS survey staff into the often soft ground, we used a ca. 25 cm2 'foot' on the bottom of the staff to dissipate pressure.
Point Cloud dataset
To obtain aerial images for modelling canopy heights, we used a 24 megapixel camera (Sony alpha6000), equipped with a prime lens (Sony SEL 20 mm F2.8), carried on a Tarot hexacopter controlled with a PixHawk running open source ArduPilot (http://ardupilot.org) software. Two sets of survey flights were undertaken, the first obtaining nadir imagery and the second obtaining oblique (ca. 20 degrees from nadir) images with a spatial grain of ca. 4-6 mm at the canopy top (Cunliffe and Anderson, 2019). The camera was triggered by the flight controller based on distance travelled, with both sets of flights together capturing > 22 photos for every part of the study area yielding (equivalent to forward overlap of 75% and sidelap of 65% for each flight). We collected 673 RGB photographs over our survey area. Image data were originally recorded in lossless RAW format (Sony ARW), and were converted to uncompressed TIFF using Sony's Image Data Converter (v4).
These aerial images were processed using structure-from-motion photogrammetry, using a high performance workstation and a workflow based on Cunliffe et al. (2016). Geotagged image data and marker coordinates were imported into Agisoft PhotoScan (v1.2.4) and converted into a common coordinate reference system (WGS84 UTM 7N; EPSG:32607). Image sharpness was assessed using PhotoScan's image quality tool, which assesses the sharpness of the sharpest part of each photograph; all images had a sharpness of > 0.77. Photos were matched and cameras aligned, using the highest quality setting, key point limit of 100,000; unlimited tie points, generic and reference pair preselection enabled, and adaptive camera model fitting disabled. Reference parameters were set to: camera location accuracy was set to 25 m; marker location accuracy was set to 0.01 m; marker projection accuracy was set to 2 pixels; tie point accuracy was set to 1.
The sparse cloud was filtered and tie points with reprojection error above 0.55 were excluded from further analysis. An operator reviewed the estimated camera positions to verify their plausibility and remove any obviously erroneous tie points from the sparse point cloud. Geolocated markers were manually placed on all projected images for each of the 26 ground control points (Cunliffe et al., 2016; Kachamba et al., 2016). Three markers used for independent accuracy assessment were deselected at this stage. The bundle adjustment was optimised using the filtered cloud of tie points and the following lens parameters: Focal length (f), principal point (cx, cy), radial distortion (k1, k2, k3), tangential distortion (p1, p2) aspect ratio and skew coefficient (b1, b2). Out of 673 images, 95% (636) were aligned and used for the multi-view stereopsis (dense cloud generation) using the ultrahigh quality setting, mild depth filtering and calculate point colours enabled. The dense point cloud was exported in the laz format, with point coordinate and RGB attributes.
The dense point cloud was analysed in PDAL (v1.9.1 PDAL Contributors, 2019). The corner coordinates were used to subset points for each harvest plot. Within each plot, the normalised height-above-ground (HAG) of each point was calculated relative to the horizontally closest corner coordinate. Any points with a negative height above ground were coerced to zero. In a few instances where corner marker posts were visible in the point cloud, these points were removed manually. We determined the maximum HAG for each cell across a fine grid with a spatial resolution of 0.01 m using the rasterstats package (v0.13.1). For cells containing no points, maximum heights were interpolated with inverse distance weighting considering an array of 11X11 cells using a power term of two, and cells with no neighbouring points in that area remained empty. We used the 1 cm spatial grain to preserve the fine-scale variability in the point cloud (Cunliffe et al., 2016; Wallace et al., 2017). Plot-level summary metrics were then extracted from this grid of local maxima elevations.
Spectral reflectance mosaic datasets
To obtain images for modelling spectral reflectance, we used a Parrot Sequoia multispectral sensors (firmware 1.0.0), mounted on a multi-rotor (as above) and fixed wing platform (Zeta Phantom FX-61 with a PixHawk flight controller). We undertook four multispectral surveys, at altitudes of 19 m, 50 m, 120 m and 120 m above ground level, to obtain a range of ground sampling distances. A MicaSense Reflection panel reflecting ca. 50% of light was photographed before and after each survey, and the most representative image was subsequently used to calibrate the spectral reflectance during processing (Assmann et al., 2018). The reflectance values of the panel were measured under laboratory conditions before and after the field campaign, and we used the mean to minimise errors arising from panel degradation. The Sequoia was triggered using a two-second intervalometer, to achieve an overlap of at least five images per part of the site.
The multispectral images were processed using Pix4Dmapper Pro (v4.0.25). Radiometric corrections were implemented using dowelling sun irradiance and pre- or post-flight images of reflectance panels following Assmann et al. (2018). GCPs were manually placed in <15 images, and then automatic placement was employed and manually verified. Normalised difference vegetation index (NDVI) maps were generated using the 'AG Multispectral Template' at the native resolution of the GSD. The mean NDVI value of all pixels in contact with each harvest plot were extracted using the rasterstats package.
Each of the 36 sub-plots was surveyed using point-intercept methodologies similar to ITEX protocols (Molau and Molgaard, 1996; Myers-Smith et al., 2019), on the 30th and 31st of July 2016. To do this, a grid with 36 points at 10 cm intervals was placed over each plot. A metal pin was placed vertically at each of the 36 points and the maximum height of the canopy above the moss/litter layer, as a pseudo-'ground surface' was recorded at every point.
Within each of the 36 sub-plots, all standing vascular plants were harvested down to the moss/litter layer, as a pseudo-'ground' surface (sensu Walker et al., 2003) on the 31st of July and 1st of August 2016. Biomass was dried at ca. 35°C for > 70 hours, until at a constant weight over a 24-hour period. To quantify biomass carbon contents, samples of each partition from 12 sub-plots were oven-dried at 70°C for 48 hours, homogenised with a ball mill, and flash combusted for measurement of evolved CO2 in an elemental analyser (CE Instruments, NC2500).
To enable canopy height to be modelled across the monitoring area, we undertook a walkover survey with the GNSS instrument to measure the ground elevation at regular intervals along transects, which yielded observations with an average spacing of 1 point per 10 m2. These terrain observations can be used to interpolate a continuous digital terrain model, while the first 36 observations could be reserved for validation of the interpolated canopy height model.
|Data Set Creator||Cunliffe, Andrew;Myers-Smith, Isla;Kerby, Jeffrey;Assmann, Jakob;Daskalova, Gergana|
|Data Set Title||Canopy height, spectral reflectance (NDVI), and aboveground biomass of Salix richardsonii across a wet graminoid-shrubland ecotone on Qikiqtaruk - Herschel Island, Yukon, Canada (2016)|
|Data Set Release Date||2019|
|Data Set Publisher||Polar Data Centre,Natural Environment Research Council,UK Research & Innovation|
|Other Citation Details||shortdoi:10/dzbq|
|Detailed Location||Qikiqtaruk, Herschel Island|
|Data Collection:||Red-green-blue images collected on 25th of July, 2016. Coordinates collected between June and August 2016. Images for normalised difference vegetation index (NDVI) maps collected in July 2018. Biomass harvests conducted in July & August 2016.|
|Distribution Media||Online Internet (HTTP)|
|Data Storage:||Data provided include:
(i) Precise coordinates of ground control points, in csv file: 'ground_control_marker_coordinates.csv', <1 mb in size.
(ii) Precise coordinates of harvest plot corners, in csv file: 'harvest_plot_corner_coordinates.csv', <1 mb in size.
(iii) Aboveground biomass data from harvest plots, in csv file: 'biomass.csv', <1 MB in size.
(iv) Terrain observations, in csv file: 'terrain_observations', <1 MB in size.
(v) 673 red-green-blue (RGB) images, in .tif format, each 68 mb in size.
(vi) Photogrammetrically reconstructed dense point cloud (derived product, comprising of 228,315,000 points with XYZ and RGB values), in .laz format, ca. 2.4 GB in size.
The free LAStools program can decrypt this compressed data: https://rapidlasso.com/lastools/.
Point cloud also provided in ascii .txt for data preservation purposes. Data conversion done using the LAStool las2txt using the -xyzRGB options, ca. 11GB in size.
(vii) Four normalised difference vegetation index (NDVI) maps, in .tif formats, ranging from 16 to 163 MB in size.