Methodology:

The rock outcrop dataset was derived though automated image analysis of Landsat 8 data and provides an updated map of exposed rock in Antarctica at greatly improved accuracies compared to the previous dataset. Classification accuracy has been calculated to be 74 ±9% at 1 standard deviation error compared to 39 ±19% for the previous manually derived dataset.

The image analysis algorithm used to produce the updated dataset used an NDSI threshold, thermal infrared to blue ratio, thermal infrared threshold, an NDWI threshold and a coastal mask to identify sunlit rock whilst excluding sunlit snow, clouds and liquid water. In addition, a blue reflectance threshold, an NDWI threshold and a coastal mask were used to identify shaded rock whilst excluding shaded snow and liquid water. The extents of shaded and sunlit rock exposure were then merged. To produce the final dataset the exposed rock extents were merged with the previous manually derived dataset in areas lacking suitable Landsat 8 images at the time of analysis (most extensively south of 82°40 S). These polygons are identified within the attribute table of the shapefile under the field Source (the origin of each polygon identified as Manually derived or Landsat 8 automated rock outcrop identification). To aid this differentiation the field GRIDCODE; identifies automatically derived polygons as 1 and manually derived polygons as 0.

Data quality:

The rock outcrop dataset was derived though automated image analysis of Landsat 8 data and provides an updated map of exposed rock in Antarctica at greatly improved accuracies compared to the previous dataset. Classification accuracy has been calculated to be 74 ±9% at 1 standard deviation error compared to 39 ±19% for the previous manually derived dataset.