Boike, Julia; Veh, Georg; Piel, Konstanze; Stoof, Günter; Muster, Sina (2015): Orthomosaic of Samoylov Island, Siberia, summer 2006 (382 MB) [dataset]. Alfred Wegener Institute - Research Unit Potsdam, PANGAEA, https://doi.org/10.1594/PANGAEA.845722

High-resolution land cover mapping is needed in the heterogeneous arctic landscapes that change land surface parameters over a range of a few meters. Polygonal tundra on Samoylov Island features a network of dry polygonal rims interspersed with patches of wet tundra and polygon ponds.

We obtained sub-meter resolution aerial images of Samoylov Island by mounting a Canon PowerShot S3 IS on a helium filled balloon. Images were acquired in the visibile (RGB) range on August 25, 2006.

In total, 60 images (Appendix A) were used for stereo photogrammetric processing using Agisoft Photoscan (V 1.1.6). The decisive criteria for image selection were the absence of clouds and the image sharpness. Only pictures from the central part of the island met these conditions.

Camera positions for raw image alignment were estimated by the software due to the lack of an internal GPS log. After image alignment, all points with a reprojection error greater than 0.5 pixels were deleted from the raw point cloud. Since there were no Ground Control Points (GCPs) measured during field work, an artificial network of GCPs had to be developed. We evenly distributed 91 "Virtual Ground Control Points (Appendix B) across the whole island, using the Orthomosaic Samoylov from the year 2007 (Boike et al., 2012) as base reference. The same virtual GCPs were already used for georeferencing the VIS and NIR orthomosaics from the year 2008 (Boike et al., 2015). 54 of these points (excluding points 32-34, 38-43, 45, 46, 49-51, 53, 54, 56, 58, 59, 63, 67-75, 77, 82, 84-86, 88 & 89) could be detected in the VIS images of 2006. The coordinates of the GCPs (WGS 1984, UTM Zone 52N) were imported into Photoscan and placed in each image. The georeferenced and optimized point cloud was filtered once again, using the same reprojection error threshold of 0.5 pixels to delete misaligned points. The overall spatial error of the placed markers is 1.12 m or 2.352 pixels, respectively.

A mesh was built from the sparse point cloud and exported as Geotiff with a planimetric resolution of 0.18 m.

See the developer's tutorial (Appendix C) to retrace the orthophoto processing chain in Agisoft Photoscan.

See Appendix D for the reconstruction parameters of each GCP and the estimated position of each image.

Slight differences compared to the Orthomosaic Samoylov 2007 (Boike et al., 2012) occur especially at the eastern and southern margins of the orthomosaic. The other parts of the mosaic show a very good agreement with the reference mosaic. Geomorphological units, i.e. the polygonal tundra, as well as major waterbodies are clearly detectable.

Image meta data:

Image format: 3 bands (RGB)

Coordinate system: WGS 1984 UTM Zone 52N

WKID: 32652 Authority: EPSG

Resolution: 0.18 m

Image extent:

Top: 8033890.28175

Left: 414632.431443

Right: 416121.931443

Bottom: 8031437.24175

Pixel type &depth: 8bit unsigned integer

File size: 430 MB (uncompressed)

Appendices:

App. A: Images used for stereophotogrammetric processing in Agisoft Photoscan

App. B: Virtual Ground Control Points used for image alignment in Agisoft Photoscan

App. C: Tutorial (Beginner level): Orthophoto and DEM Generation with Agisoft PhotoScan Pro 1.1 (with Ground Control Points)

App. D: Marker and image placement properties in Agisoft Photoscan