| 29 Sep 2025
Evidence and interpretation of non-linear recession behaviour in a periglacial cliff at Port Foster, Deception Island (South Shetlands, Antarctica)
Abstract. Cliff recession in periglacial coastal environments is highly sensitive to climate-driven changes in temperature, sea ice, and permafrost dynamics. While previous studies have predominantly relied on linear models to describe shoreline retreat, these methods often fail to capture the non-linear, episodic, and threshold-driven nature of coastal erosion in cold regions. Moreover, the scarcity of high-resolution, long-term datasets in polar regions, particularly in Antarctica, has limited the development of predictive models tailored to these dynamic systems. This study aims to improve the understanding of long-term cliff recession patterns in periglacial environments by applying advanced non-linear statistical modelling to a multitemporal dataset. Focusing on the coastal bluffs of Port Foster, Deception Island (South Shetlands, Antarctica), we examined geomorphological changes over a 66-year period (1956–2022), using a unique combination of historical aerial photographs and high-resolution satellite imagery. Photogrammetric pre-processing, orthorectification, and manual digitisation of reference lines were integrated into a transect-based statistical analysis framework. The study applied both linear and non-linear least squares regression models – quadratic and sigmoidal – to reconstruct spatial-temporal erosion trends, with uncertainty-weighted parameters incorporated into the estimation. Results reveal a distinct shift from quasi-stable to accelerated recession after 2000, particularly in areas exposed to dominant marine and thermal forcing. Linear models underestimated these trends, while sigmoidal logistic models more accurately identified inflection points in erosion rates. Maximum recession rates reached up to 5 m/year in the central bluff segment. The findings underscore the importance of integrating non-linear modelling into coastal monitoring and management frameworks, especially in vulnerable and data-scarce polar environments. This approach provides a more realistic understanding of periglacial coastal dynamics and highlights the critical need for adaptive strategies to address climate-induced instability near strategic infrastructure such as research stations.
Major Comments
This manuscript is a technically sound photogrammetric reconstruction of cliff recession at Port Foster over 66 years. The manuscript is well executed and methodologically coherent. However, in my opinion, the work does not fit The Cryosphere standards for several fundamental reasons that extend beyond revision. The manuscript lacks novelty and significance to meet the journal’s standards.
The 1.5 km study segment is too geographically isolated for a journal emphasising cryospheric processes at regional to global scales. While the location adjacent to BAD and BAEGC research stations provides practical relevance for infrastructure risk assessment, this should be better emphasised in the introduction and study area chapters.
The analysis could reasonably span a larger coastal area, such as the whole island, to enable broader insights into the South Shetland Islands’ coastal dynamics, making it more Antarctic coastal science. In this way, it remains rather isolated.
Coastal retreat accelerated post-2000 is a major finding, but the manuscript does not explain why – this is, for me, the main scientific flaw. Why does the inflexion point occur specifically around 2000? What climate threshold or permafrost property change drives this transition?
“Sigmoidal curves fit better than linear models” (L23-24) is pattern recognition, not mechanism understanding. Regional warming is cited (0.54°C/decade, L145; recent average -1.1°C, L148), yet climate forcing is never mechanistically linked to observed retreat trends. Hypotheses missing that could improve the manuscript are: (1) time-series correlations between temperature anomalies and retreat rates; (2) sea ice extent variability affecting wave fetch and storm frequency; (3) differentiation of thermal versus mechanical driver contributions.
L641-644: “This study advances the understanding of Antarctic periglacial coastal recession by: … (ii) validating sigmoidal models as superior alternatives for characterising episodic erosion” – this claim is not justified by any evidence. Missing: correlation values for linear versus quadratic versus sigmoidal models, for example, among other possibilities like confidence intervals or statistical significance tests. For a paper claiming model superiority, the absence of comparative statistics is worrying, and the central claim is not properly justified without this evidence.
Temporal bias: Dense post-2000 sampling could create an artificial acceleration trend.
Absent but could improve: sensitivity testing, refitting the sigmoidal model using only 1956-2000 data to predict 2001-2022, and assessing how inflexion point timing shifts with gap interpolation variations.
The introduction reads like an Arctic coastal manuscript, with several references to Arctic-specific studies, while there is minimal Antarctic-specific content. Missing are Antarctic Peninsula climate studies, wave-forcing dynamics, storminess patterns, and sea-ice loss relevant to the Antarctic Peninsula and South Shetland Islands region.
The discussion, while generally well performed, lacks a marine forcing analysis, such as wave energy, storm surge, and tidal dynamics, which should be substantively addressed, even if data are limited, through reasonable assumptions grounded in regional oceanographic literature.
The 5 m transect spacing (implied in methods) for a 1.5 km study area is coarse. For bluff proxy analysis with low morphodynamic complexity and sinuosity compared to waterline proxies, denser spacing (1-2 m) is necessary to identify adequate spatial variability. Coarse spacing may artificially favour non-linear model performance due to spatial smoothing effects.
Uncertainty assessments address georeferencing errors (Eq. 1, L334-337) but omit shoreline position uncertainty and shoreline change rate uncertainty independent of imagery quality. It is standard and necessary for coastal analysis studies to include shoreline position or shoreline change trends uncertainty. The uncertainty equations also appear overly complex for this study's scope.
Discussion could incorporate and project the risks for the scientific bases, which would give more significance to the manuscript
Section 5.1 (L383-420) is best fit as a Methods subchapter; it is not a result.
Minor Comments
L60, L63: Acronyms not defined at first use.
L68: The term bluff top or bluff edge is more commonly used than bluff crest. Consider using Bluff top or bluff edge
L82: Roland et al. 2024 cited but absent from the reference list.
L156: “wave heights up to 1.165 m“, while other values have two decimal places, keep consistency.
L170: Vieira et al. (2008) cited for permafrost thickness, but broader Antarctic permafrost dynamics literature is underrepresented, and there is a need for better context.
L178: "Transient dejection cones" terminology is questionable. Dejection cones typically refer to alluvial fans. Consider "erosion cones," "washout cones," or "fan-shaped gully complexes."
L184: In general, I do not agree with the term coastal recession applied in the manuscript. Coastal retreat, or more specifically, coastal erosion, should be used.
3.1. Archive aerial imagery subchapter: several details not relevant to the manuscript can be cut from the text (captured altitudes, exposure ranges, DPIs). Table 1 details of scale, camera model, focal length, and number of photos are enough
L273-276: 37 GCPs identified as "immobile rocky outcrops." Given volcanic activity (eruptions in 1967, 1969, and 1970; L134-135), GCP stability over 66 years is merely an assumption.
L279: Satellite optical images downloaded from Google Earth Pro??? Google Earth Pro or Google Earth Engine?
Some figures in the results section could be combined to join related temporal and spatial patterns, improving readability and focusing attention on key findings.
L484-493: Shoreline change rates should have uncertainty values for the periods (example: 0.50 +- 0.16 m/y)