| 20 Mar 2026
Transient snow line altitudes of glaciers in the European Alps from multi-mission remote sensing data (2000–2025)
Abstract. Observations of glacier snow line altitude (SLA) provide important information for estimating glacier surface mass balance and glaciological modelling. Optical satellite remote sensing enables the repeated measurement of SLA at glacier-specific to regional scales, yet producing continuous multi-temporal SLA time series requires the processing and analysis of large volumes of medium-resolution imagery. In addition, traditional image brightness thresholding may suffer from temporal changes in illumination, variations in reflectivity from evolving snow and ice surface properties, and cross-sensor radiometric differences. Here, we present a fully automated cloud-processing workflow for SLA retrieval based on optical image segmentation and adaptive, scene-specific thresholding, to map the snow-ice boundary on glaciers under various acquisition conditions and based on multi-mission data. Our method uses atmospherically corrected reflectance data from the Landsat-5 to -9 and Sentinel-2 satellite missions and accounts for cloud cover, non-glacier pixels, terrain shadows and temporal glacier surface elevation change. Our SLA dataset comprises a total of ∼200,000 observations across 408 glaciers in the European Alps between 2000 and 2025. In addition, we provide estimates of end-of-summer SLAs and multi-annual change trends for most glaciers and years. Validation against visually delineated SLAs at selected reference glaciers demonstrates high agreement, with a root mean square error of ∼100 m vertical deviation. Applying the workflow to glaciers in the Alps reveals a mean regional SLA rise of 145 m since 2000 (+6 m year−1), characterized by pronounced interannual variability. Particularly during the last decade, seasonal SLA change can be tracked due to the short revisit times of the current Landsat-8, -9 and Sentinel-2 earth observation missions. All data are publicly available at Zenodo (https://doi.org/10.5281/zenodo.18223929; Sommer et al. (2026)).
I would like to congratulate the authors on this interesting work, which is already in very good shape. The study presents a methodology based on Landsat and Sentinel-2 optical data to derive ice/snow classification maps using a segmentation approach, and subsequently to analyse trends in snowline altitude. A major strength is that the method is designed for a large area (the Alps) and a long time frame (2000–2025), enabling insightful analyses over a period of more than 20 years.
The methodology is described very clearly. I also particularly appreciate the authors for providing open-source code that is well documented. So I believe this work can meet the ESSD standards for publication.
I think the manuscript does not need extensive revision since it is already in a very good shape. I am suggesting just some minor revisions.
The only major point I would raise is that a comparison with other state-of-the-art algorithms is missing. I understand that previous studies may not have focused on large-scale applications; however, a key limitation here is that the approach remains threshold-based. Even if the threshold is determined automatically, it may fail in many cases, as the authors also note. Why not consider more advanced techniques, such as deep learning methods?
Another issue pertains to code execution. I attempted to reproduce the results but encountered an error related to the lack of a billing account.
Forbidden: 403 POST
https://storage.googleapis.com/storage/v1/b?project=snowmapping-492110&prettyPrint=false: The billing account for the owning project is disabled in state absent.
It appears that having an active billing account is necessary for running your workflow, which may be considered a limitation. Could you please provide further clarification regarding this requirement?
Some minor issues:
L205: What do you consider to be the minimum and maximum solar azimuths and elevations? Are you not using the actual solar azimuth and elevation for the specific scene acquisition? Please explain this more clearly. Also, I understand that you are only considering topography for shadow characterization, correct? You are not using spectral features—if so, is that sufficient?
L262: How do you determine the snow-class area a priori (i.e., prior to your segmentation)? Is it derived from the NDSI threshold?
L335: I understand that you apply outlier detection, but could this procedure remove abrupt changes (e.g., due to climate change)? Do outliers always need to be excluded, or can they sometimes be informative?
L358: Do the numbers 120,000 (Landsat) and 103,000 (Sentinel-2) refer to scenes or pixels? Later you mention 13,000 satellite images, so it is unclear what these values represent.
L372: The values 127.4 m and 105.5 m—could you clarify where these numbers come from?
L389: How do you explain the behaviour in Figure A2, with deviations from the 1:1 regression line, especially at high altitudes?
L398: Do you mean an SLA drop in 2023 in Figure 4c?
Figure 6: Please indicate the definition of North (I assume 0/360°) to make the figure easier to interpret.