This manuscript presents a historical reconstruction of calving front positions for 171 tidewater glaciers in Svalbard over the period 1936–1978, based on a combination of historical aerial photographs, early Landsat MSS imagery, and declassified KH-5 ARGON satellite photographs. The extension of calving-front observations into the pre-1980 satellite, together with the public release of georeferenced imagery and digitized fronts, gives this study clear long-term value as a data resource for Svalbard glacier research.

However, the scientific contribution of the manuscript is currently not articulated as clearly as it could be. The Introduction largely reads as a chronological listing of previous studies rather than a focused synthesis that motivates why this reconstruction is needed, what specific gap it fills relative to existing datasets, and how it advances current understanding. In addition, although 171 glaciers are mapped, the manuscript lacks a clear overview figure demonstrating the spatial coverage of the dataset, which makes it difficult to fully appreciate the scope and representativeness of the work.

I also have several major concerns related to (i) the treatment and interpretation of georeferencing uncertainties, (ii) the robustness of inferred average retreat and area-loss rates given the sparse temporal sampling, and (iii) the framing of some conclusions, which in places extend beyond what the data can securely support. Addressing these issues would substantially strengthen the scientific rigor, clarity, and long-term usability of the dataset. Overall, I recommend major revision.

Major comments

1. The study relies heavily on polynomial georeferencing of KH-5 ARGON images using many ground control points (typically 100–200 per image), while explicitly abandoning a physically based orthorectification approach (Sect. 2.2.2). Although the authors justify this choice by focusing on relative calving-front positions near sea level, the implications of this decision require a more critical and transparent discussion. High-order polynomial transformations can provide a good local fit at GCPs but may introduce non-physical distortions between control points, particularly over very large scenes such as ARGON images (>500 km wide). The manuscript also acknowledges that uncertainties in 1962–1963 are especially large, yet several key interpretations (e.g. identification and timing of advances, surge attribution) rely strongly on these data. Further clarification is needed:

• Provide quantitative diagnostics of georeferencing performance (e.g. RMSE at GCPs and, if possible, spatial patterns of residuals).
• Explicitly discuss how non-uniform geometric distortions could bias centerline-based distance estimates, especially for wide or curved glacier fronts.
• Clarify whether any spatial filtering or systematic exclusion (e.g. image edges or areas affected by severe distortions) was applied, beyond individual case decisions such as Emmabreen.

Without this information, it is difficult to assess whether some detected advances or retreats can be confidently distinguished from residual geometric artefacts.

2. The manuscript reports an average retreat rate of 26.3 m yr⁻¹ between 1936 and 1978 and discusses variations within this period (Sect. 3.1, Fig. 5a). However, the effective temporal resolution is extremely limited, with only three broad observation windows (1936–1938, 1962–1963, and 1976–1978). Given this sparsity: 1) Linear trends and apparent phases such as “plateaus” or accelerations are poorly constrained. 2)The reported R² values may give a misleading impression of robustness. I suggest that the authors:

• Reframe retreat and area-loss rates explicitly as period-averaged net changes, rather than temporally resolved trends.
• Avoid interpretative language implying acceleration or deceleration unless supported by sensitivity tests.
• Emphasize net changes between observation epochs as the primary result, with trends presented only as descriptive summaries.

3. Area change is calculated as the product of frontal displacement and average glacier width (Sect. 2.5, 3.1), following Li et al. (2025). While this approach is reasonable given the data limitations, it differs fundamentally from outline-based methods used in other studies cited for comparison (e.g. Kochtitzky and Copland, 2022). Although these differences are acknowledged, the manuscript still draws relatively strong comparative conclusions (Sect. 4.1, 5). I therefore recommend that the authors more clearly separate methodological differences from physical interpretation when comparing results across studies, explicitly discuss how width variability, fjord geometry, and non-parallel retreat may bias area estimates, and consider presenting cumulative frontal length change alongside area change as a more methodologically neutral metric.

4. The identification of surge events is one of the most interesting aspects of the manuscript. However, the criteria used to distinguish surges from non-surge advances are not always clearly defined. For example, Schweigaardbreen and Fonndalsbreen are treated cautiously as advancing glaciers, whereas Emmabreen is classified as having undergone a previously undocumented surge, largely based on frontal advance magnitude and geometry. Given the long time intervals between observations, this distinction remains ambiguous. I recommend that the authors explicitly state the diagnostic criteria used to classify an event as a surge versus a non-surge advance, clarify the relative roles of frontal advance magnitude, spatial coherence, and ancillary evidence (e.g. crevasse patterns or later velocity observations), and acknowledge that, in some cases, surge classification remains uncertain given the available data. This would improve the conceptual consistency of Sect. 4.2 and help avoid over-interpretation.

Minor comments:

1. Consider using the term “delineation” rather than “retrieval” consistently throughout the manuscript.
2. The Abstract and Introduction would benefit from clearer separation between background, methods, and the specific contribution of this study. The Introduction (particularly Sect. 1.2) could be substantially condensed; rather than listing previous studies, it should synthesize why DISP and early Landsat data are essential for the present work.
3. Use “calving-front position” rather than “locations” for consistency with common terminology.
4. An overview map showing all 171 calving fronts is necessary to demonstrate the spatial coverage of the dataset.
5. Clarify how many fronts are shown in Fig. 3 and how this relates to the total number of observations.
6. Several statements describing “faster” or “slower” retreat (e.g. Lines 184–186) are difficult to support given the sparse time sampling and should be revised.
7. Multi-panel figures (e.g. Fig. 7) are information-rich but visually dense; inset maps or simplified color schemes could improve readability.
8. 2.5 notes that centerlines were digitized manually; please clarify whether a single operator performed this task and comment briefly on consistency.
9. The uncertainty framework is thorough, but a concise summary table of typical positional uncertainties for each data source in the main text would aid readers.
10. Please ensure that metadata in the Zenodo archive clearly document coordinate reference systems and georeferencing procedures.