Land cover change across the major proglacial regions of the sub-Antarctic islands, Antarctic Peninsula, and McMurdo Dry Valleys, during the 21st century

Abstract

Land cover information is essential for understanding Earth surface processes and ecosystems. Here, we use K-means clustering to classify Landsat 8 Operational Land Imager (OLI) images covering six proglacial sites of sub-Antarctic islands, the Antarctic Peninsula, and the McMurdo Dry Valleys at 30-m resolution. We quantify spatial patterns of water, bedrock, vegetation, and sediments to an accuracy of 77 percent. Vegetation is most abundant on South Georgia (7 percent of the proglacial area) and the South Shetland Islands (1 to 2 percent). Furthermore, we use change vector analysis (CVA) to discriminate landcover change in the twenty-first century. A latitudinal pattern is evident in ice loss and proglacial landscape change; for example, loss of ice on South Georgia and proglacial landcover change is two orders of magnitude greater than in the McMurdo Dry Valleys. Four of the studied sites had similar landscape stability (64 to 68 percent unchanged), with Alexander Island an exception (50 percent change) due to recent enhanced glacier melt. Overall, we show how landcover of proglacial regions of the climaticallysensitive sub-Antarctic and Antarctica has changed since 2000, with a CVA accuracy of 80 percent. These findings inform understanding of geomorphological activity and sediment and nutrient fluxes and hence terrestrial and marine ecosystems.

Keywords: Antarctica; LULC; land cover; proglacial; sediment.

Figure 1.

Location of our study sites. The areas analyzed are highlighted in red and span a latitudinal gradient from 54° S to78° S. Proglacial regions not analyzed in this study are highlighted in black (Burton-Johnson et al. 2016) and are primarily mountains (e.g., Transantarctic Mountains) or are frequently covered by extensive cloud cover (e.g., King George Island). (A) Grytviken on South Georgia. Taken in 2009 by Simon Murgatroyd (CC BY-SA 2.0). (B) Camp Byers on South Beach (ESP) on Byers Peninsula. Taken in 2017 by “Inoceramid bivalves” (CC BY-SA 4.0). (C) Telefon Bay (background), as viewed from the rim of a crater on Deception Island. Taken in 2020 by Espen Mills (CC BY-SA 4.0). (D) Abernethy Flats on James Ross Island’s Ulu Peninsula, as viewed from Lachman Crags, above Triangular Glacier (looking West), taken in 2022. (E) The central station of Fossil Bluff on Alexander Island in 2003. Photo taken in 2003 by “Apacheeng lead” (Public Domain). (F) The Wright Valley of the McMurdo Dry Valleys (looking west toward Wright Upper Glacier) in 2013, taken by “Turkish D.” (CC BY-SA 4.0). Inset photos (A), (B), (D), (E), and (F) were sourced from Wikimedia Commons. Photo (C) by Christopher D. Stringer.

Figure 2.

Our approach to classifying land cover.

Figure 3.

The change detection (CVA) approach used in this study.

Figure 4.

A comparison between (a) the land classification produced in this study and (b) a geomorphology map, adapted from Jennings et al. (2021). Jennings et al. (2021) produced these data through a series of extensive field surveys on the Ulu Peninsula. Vegetation locations as collected in the field by Jan Kavan (of the Czech Antarctic Research Programme) in 2021 are also displayed. Note the similarities in the ice class, locations of river systems, and scree slopes. The colors in (b) have been adapted to allow a more direct comparison with the map produced in this study (a).

Figure 5.

Land cover maps of the six sites, including ten classes that describe eight distinct surfaces. Ice class may include limited areas of seasonal snow cover. Higher resolution maps can be found in the supplementary material (Section 2).

Figure 6.

Percentage land cover values (excluding ice, no data, and land [undifferentiated]) for each site, overlaying the coastline of Antarctica (coastline sourced from BAS). Error bars indicate the 95 percent confidence intervals.

Figure 7.

How the wet ice and turbid water classes compare to the images they are derived from, with a large area of saturated firn on Snow Hill Island (64°28′ S, 57°4′ W) and a sediment plume off the coast of Vega Island (63°52′ S, 57°16′ W).

Figure 8.

The proportion of the proglacial landscape that has changed at each site analyzed and the makeup of those changed regions.

Figure 9.

Maps of each site indicating the spatial variability in confidence. Very low confidence = <20 percent of points were accurate; low confidence = 21 to 40 percent; medium confidence = 41 to 60 percent; high confidence = 61 percent to 80 percent; very high confidence = >80 percent.

Figure 10.

Examples of the four most frequently observed change classes: 86 percent of the change identified in out data can be described by these four classes. The CTF example shows less active river channels in the modern image associated with drier sediments on Seymour Island. FTC shows the opposite, with more active river channels associated with wetter sediments on James Ross Island. The ITF example shows a reduction in the extent of glaciers and snowcover on Alexander Island, and the ITT example shows the development of proglacial lakes following glacier retreat on Snow Island in the South Shetland Islands. Though these four panel sets are designed to highlight the four main change classes, all change classes can be seen within these panels. Modern images are derived from Landsat 8 OLI, and the old images are derived from Landsat 7 ETM+.