Change and variability in Antarctic coastal exposure, 1979–2020

Increased exposure of Antarctica’s coastal environment to open ocean and waves due to loss of a protective sea-ice “buffer” has important ramifications for ice-shelf stability, coastal erosion, important ice-ocean-atmosphere interactions and shallow benthic ecosystems. Here, we introduce a climate and environmental metric based on the ongoing long-term satellite sea-ice concentration record, namely Coastal Exposure Length. This is a daily measure of change and variability in the length and incidence of Antarctic coastline lacking any protective sea-ice buffer offshore. For 1979–2020, ~50% of Antarctica’s ~17,850-km coastline had no sea ice offshore each summer, with minimal exposure in winter. Regional summer/maximum contributions vary from 45% (Amundsen-Bellingshausen seas) to 58% (Indian Ocean and Ross Sea), with circumpolar annual exposure ranging from 38% (2019) to 63% (1993). The annual maximum length of Antarctic coastal exposure decreased by ~30 km (~0.32%) per year for 1979–2020, composed of distinct regional and seasonal contributions.


Analysis methods
For this study, we developed and evaluated two new but different algorithms for quantifying and monitoring coastal exposure: the Coastal Exposure Index (CEI) and a more complex Coastal Exposure Length (CEL) method. The CEI technique is based on the detection of sea ice presence/absence radially out (northwards) from the coastline along each meridian (at one degree longitudinal spacing), following masking of the ice sheet (e.g., Supplementary Fig. 1a).
As such, the CEI is simply defined as the number of longitudes with no sea ice (threshold set to < 15% following convention) to the north of the continent, and hence runs from zero to 360.
While the relatively-simple CEI algorithm provides reasonable broad-scale assessment of coastal exposure (see the daily example in Supplementary Fig. 1a) and is computationally inexpensive, it fails to categorise certain coastal areas as exposed when they clearly are. As a result, it tends to underestimate total exposure in terms of both its physical length and duration.
Specifically, the application of CEI is limited where the (exposed) coastline being assessed: (1) is N-S trending (e.g., the Antarctic Peninsula); (2) has a zonal-trending promontory of land/ice sheet to its north; and/or (3) is semi-enclosed by a zonally-extending sea ice tongue offshore.
An example of the latter, across the Amundsen Sea at ~105-115°W, is given in Supplementary   Fig. 1a; such features tend to be common though sporadic and ephemeral, particularly during the late annual sea-ice retreat season (December-January).
For this reason, the more sophisticated CEL algorithm was developed, and the output from this is used in the main analysis as it does not have the disadvantages of the CEI. Here, CEL is defined as the length (in kms) of the Antarctic coastal perimeter with no adjacent sea ice anywhere offshore (i.e. total exposure of the coast to the open Southern Ocean with no intervening sea ice), but excluding coastal polynyas (recurrent areas of persistent open water/thin ice enclosed by sea ice 2 ) e.g. Supplementary Fig. 1b. By this method, we use the land mask to determine if each coastal grid point has an immediately-adjacent ocean grid point that is ice-free (i.e. has a sea-ice concentration of <15%). If this criterion is met, then a nearest (adjoining) neighbour-testing technique is used to determine whether that ocean grid point is exposed in some way to the wider open ocean or is bound by neighbouring sea ice offshore. If any of the neighbouring grid points are classified as "exposed", or if the total area of neighbouring ice-free grid points exceeds an arbitrary cut-off of 500,000 km 2 , then that coastal grid point is classified as "exposed". Otherwise, the grid point and all sea-ice-free neighbouring grid points are deemed to be bounded by sea ice and are classified as a coastal polynya. The 3 length of individual exposed coastal grid points is estimated by taking the square root of the respective pixel area. The length of coastal exposure, either regionally or net circum-Antarctic, is then simply the sum of the length of exposed coastal grid points.
Although the CEL algorithm is computationally more expensive than the CEI technique (especially where large coastal polynyas are present), it provides more complete coverage of coastal exposure at greater detail. Moreover, and as a by-product, the CEL method readily detects, maps and monitors regions of low (<15%) sea ice concentration that are bounded by higher-concentration sea ice, i.e. coastal polynyas, as shown in orange at ~80°W and ~90°E in Supplementary Fig. 1b. This additional information on polynya size and distribution (not provided here as it is outside the scope of this study) is also important from a coastal-exposure perspective in that Antarctic coastal polynyas make a regionally-important contribution to the seasonal melt-back of sea ice to the coast each austral summer 3 . The CEL method also has wider geographic applicability i.e. around the Arctic sea-ice regions, and a follow-up analysis of change and variability in Arctic coastal exposure is planned.
Within the analysis, delineation of regional sectors follows the widely-used protocol for seaice analysis 4 . Austral spring is taken to be September through November, summer is December Strictly speaking, the presence/absence of coastal sea ice in this study refers to both moving pack ice and stationary fast ice that is attached (in certain places 5 ) to the coast and icebergs grounded on near-coastal shoals <450 m deep 6 . However, the coarse spatial resolution of the passive microwave-based indices precludes specific detection of fast ice or its accurate distinction from pack ice -as Antarctic fast ice where present typically occurs in a relatively narrow coastal band 5 . Fine-scale analysis of coastal exposure involving lack/loss of fast ice 4 therefore requires finer-scale satellite data and analysis techniques, which are beyond the scope of this broad-scale study. Having said this, there is some correspondence between loss/lack of pack ice and loss/lack of adjacent sea ice, due to the susceptibility of coastal fast ice to breakup by ocean swells in the absence of a protective pack ice "buffer" 7,8 .

Brief comparison of CEL versus CEI
Here, we carry out a brief comparative assessment of the CEL and CEI methods for our Antarctic application, by replicating the CEL results shown in the main analysis in Figs. 1 to 6 using the CEI method i.e., Supplementary Figs. 2 to 7. This comparison confirms the wider applicability of the CEL method. Notably, the frequency of occurrence of CEL (Fig. 2a) is slightly higher than that of CEI ( Supplementary Fig. 3a), and CEL exposure has an earlier onset than CEI-derived exposure at the same longitude ( Fig. 2b versus Supplementary Fig. 3b