Methodology:

Images and annotations were produced from UAV imagery taken between December 2019 and February 2022. Flights over the South Sandwich Islands (SSI) were conducted using a DJI Mavic Pro 2 between elevations of 40-100m and sequential images had at least 60 percent overlap when possible. The images were then used to create 3D models of all the islands using Structure-from-Motion (SfM), a computer 3D rendering technique where points are aligned based on overlapping images, which produces a three-dimensional landscape that is orthorectified and georeferenced. The resulting orthotiff created a composite birds-eye view and was then used to export a Digital Elevation Model (DEM). Where orthotiffs were particularly large, or had trouble being exported as one tiff image, they were broken up into smaller tiles. These tiles are then individually georeferenced so, when loading into QGIS or a similar software, they will automatically recreate the entire site.

The subsequent orthotiff and DEM were imported into QGIS, where nests were located using the vector layer editing tool. Nests were categorised using existing outlines of colonies made using collected GPS coordinate data. Nest coordinates were then exported into standalone csv files. Categories are as follows:

chinstrap penguin nests

gentoo penguin nests

macaroni penguin nests

mixed species sub-colony

unknown species

Using the total number of nests on each site, the total count for each grouping and the surveyed site as a whole was calculated into another table. For Zavodoski Island, which suffered distortion due to the difficult flying conditions, an estimate of nests was calculated for regions where counting was not possible. In order to estimate the number of nests in areas that could not be counted, we compared other parts of the site (that had been counted) to sites where counts were not possible.

To do this, we outlined areas that could not be counted, and extracted the ellipsoidal area, which is the default area calculation for QGIS. Then, we outlined a random sample of areas that had nest coordinates and repeated the area calculation to get a density estimate (how many nests were present in a certain space). We then multiplied the density estimate of the random sample against the individual area calculations for areas that were distorted to get an estimate of the number of nests present. These nests were finally categorised using the existing GPS coordinate outlines of colonies.

In addition, an additional flight for one region of Zavodovski Island was carried out in 2022. The resulting orthotiff and DEM had better resolution and were overlaid over the original image to help distinguish if nests were present. This method was used as an estimate of the chinstrap nests in 2020, as yearly nest locations are unlikely to significantly shift outside of annual variation. This step should be taken into account for any future analyses.

Lastly, only Adelies were visible in the Candlemas images and were the only species counted.

Data collection:

Structure-from-Motion (SfM)

Quantum GIS - QGIS

Data quality:

Some areas stitched had poor resolution and distorted imaging. For regions difficult to annotate, estimates of the number of nests were calculated as described in the lineage/methods.

When orthotiffs were too large or difficult to export as one file, they were broken up into tiles that together made up the site. This is also described in the lineage/methods.

For Zavodovski Island, a 2022 resurvey of one region of the island was included to better distinguish sub-colonies and individual nests where possible. This orthotiff is included in the data.

For Candlemas Island, only Adelie nests were countable due to problems with distortion and poor resolution.