An Accurate and Automated Method for Identifying and Mapping Exposed Rock Outcrop in Antarctica Using Landsat 8 Images

Antarctica plays a significant role in global change studies. As a typical land cover in Antarctica, exposed rock is considered indispensable in many studies, and the mapping of exposed rocks is seen as a key basis of work to meet the demand for more accurate and updated datasets with the consecutive development of satellite technology. Although the normalized difference snow index has been commonly used for differentiating exposed rocks and snow, it often misidentifies clouds as rocks. The British Antarctic Survey has used Landsat 8 data to create a new rock outcrop map for Antarctica, overcoming the limitations of previous techniques and generating Antarctic Digital Database (ADD) New Rock Outcrop with higher accuracy than previously achieved. However, there are still some omission and commission errors apparent in the shaded areas, which affect the accuracy. Widespread shaded areas in Antarctica due to low solar elevation angles and extreme topography cause difficulty in accurately mapping exposed rock. In addition, major differences are present between existing products. Addressing the existing issues about extraction of exposed rock, this study used the near infrared band and shortwave infrared 2 band of Landsat 8 reflectance data to build a specific exposed rock index for the extraction of exposed rocks. A shadow detection method combined with a blue reflectance threshold is used for the shadowed rock identification. Accuracy assessment of these extraction results showed that the accuracy of the new product is higher than all existing exposed rock products. The conversion from DN values to top of atmosphere reflectance and the solar elevation correction for each pixel individually eliminate a variety of errors associated with the different acquisition times of each image. From the statistics of the reflectance related to the training samples, this paper established the threshold of exposed rock extraction so as to ensure the applicability of the same threshold for exposed rock extraction in all images. This method is applied to a total of 1100 high-quality images that were collected for covering the Antarctic continent from November 2013 to February 2014. The results show that 253 of the images contain exposed rocks, and these images were used for mapping work. The map showed that the main exposed rock areas are mainly distributed in four coastal regions: The Antarctic Peninsula, Queen Maud Land, Lambert Glacier basin, and Victoria Land regions. We also compared our results with ADD New Rock Outcrop and Bedrock Mapping Project 2 (Bedmap2) data in the four main regions. Our results were close to the ADD rock outcrops and exhibited remarkable differences with Bedmap2. We explored the possibility of analyzing and explaining these differences. Especially, as using the same data source Landsat 8 but with a different method, the comparison between our results and ADD New Rock Outcrop is discussed and concludes in shadowed rocks extraction, mixed-pixels, and omission disagreement. The results also show that shadowed rocks accounts for nearly 12% of the total exposed rocks and cannot be neglected. The method we developed can be quickly applied for extraction and mapping of large areas of exposed rocks using Landsat 8.