Distribution of landfast, drift and glacier ice in Hornsund, Svalbard
Abstract. Co-occurring landfast, drift and glacier ice in fjords respond to climate differently and have diverse impacts on the environment. Here we describe a new method to separate ice types in fjord environments on 2639 binary satellite-derived ice/open water maps at 50 m resolution. We used a set of thresholds to create near-daily maps of landfast, drift and glacier ice of Hornsund, Svalbard over 11.5 years (2012-01-02 to 2023-06-29). The ice was first divided into stationary and moving classes based on ice-pixel persistence through time. The ice was then polygonised and the polygons were ascribed a set of parameters describing their class, time, size and location. Temporal and spatial constraints were imposed on landfast ice. Drift and glacier ice were split based on timing, location and size. Finally, the data were re-rasterized and refined at the pixel level. Over the 11.5 years, the fjord ice was classified as 53 % drift, 35 % as landfast, 8.5 % as glacier, 1.4 % as uncertain ice type while 2.1 % was masked due to radar shadows. There was a great interannual variability in the length of sea ice and landfast ice seasons, and in ice type extent, with no clear long-term trend. Negative correlation existed between the water temperature in the winter months (January–March) and the length of the sea ice and landfast ice season, as well as between the air temperature in winter months and sea ice and landfast ice coverage. Glacier ice coverage depended on air temperature in summer months (July–September) and water temperature autumn months (October–December) where lower temperatures enhanced ice persistence. The method can be adapted to other areas and used in a wide range of analyses including fjord hydrography, nearshore wave transformation or ecological studies.
This study provides valuable data and an interesting perspective on fjord ice and its environmental drivers. I do, however, have a few suggestions that might strengthen the manuscript.
First, the classification of pixels into "stationary" and "moving" is central for distinguishing fast ice, but the description is not sufficiently clear. The manuscript states that "The ice pixels were first divided into 'stationary' or 'moving' classes depending on their persistence through time", but does not explain how this was actually implemented in practice. It seems possible that a feature-tracking approach was applied, but this is not stated. Clarification would improve transparency and reproducibility.
Second, the manuscript presents correlations between fjord ice parameters, air temperature, and water temperature, but does not provide p-values. Since the time series are relatively short, especially for water temperature where gaps further reduce the effective sample size, it is difficult to know whether the reported Pearson coefficients are statistically significant or could arise from chance. This is particularly important because the conclusions rely heavily on these correlations to link environmental drivers with ice variability. Including p-values would make the statistical basis of these conclusions much stronger.
Finally, the statement that no interannual trend was observed is currently descriptive. I suggest that the authors quantify the trends in the key time series using a Mann–Kendall test combined with Sen's slope, and report both the trend values and their p-values, regardless of whether they are significant. This would provide a clearer and more transparent basis for the conclusion on long-term changes.