| 24 Mar 2026
Surface mass balance projections until 2100 for Folgefonna, a Norwegian ice cap
Abstract. Robust glacier projections are essential for mountain communities adapting to climate change, yet current projections are limited by climate data that either have coarse spatial resolution or span only a narrow range of future scenarios, thereby obscuring the true scale of predictive uncertainty. In this study, we quantify the cascading impact of uncertainty in climate projections on glacier surface mass balance (SMB) projections. We simulate the SMB of the Folgefonna glacier complex in western Norway for 1970–2100 using the energy-balance and snowpack model BESSI. To represent future climate forcing, we use the EURO-CORDEX ensemble, but show that the dataset is systematically too cold and too wet and exhibits unrealistic precipitation patterns for the western Norway region. We therefore develop a downscaling framework in which each EURO-CORDEX member is represented by analogs drawn from a high resolution convection-permitting model (NorCP). This enables SMB projections that account for a plausible spread in climate models and are based on high-resolution climate data with explicitly resolved physical processes. Using this method, we present the most detailed SMB projections of Folgefonna to date for three emission scenarios until 2100. Whether the glacier complex retains any accumulation zone by 2100 depends on the emission scenario. Ensemble medians indicate that Midtfonna loses its accumulation zone in all scenarios, Nordfonna does so in RCP4.5, and no accumulation zone remains on Folgefonna in RCP8.5. However, cumulative SMB (2026–2100) has an uncertainty of 65–75 m w.e. within each scenario due to climate model spread (25th to 75th quantiles). We furthermore find that the choice of global circulation model has a stronger influence on Folgefonna's SMB than the choice of regional climate model. These findings underscore the need for improving upon agreement between climate models. Detailed glacier mass change projections based on only a subset of the available and hence plausible climate projections underestimate uncertainty and should be considered with caution.
Review of “Surface mass balance projections until 2100 for Folgefonna, a Norwegian ice cap”
This study investigates the range of projected surface mass balance changes of the Folgefonna ice cap using a novel downscaling approach to drive an SMB model with a wide range of ESM projections. The study has two main aims: first, to develop a new framework in which RCM output is downscaled using a new “analog” method; and second, to apply this framework to force the SMB model BESSI and generate projections up to 2100 under three emission scenarios. The study is relevant and fits well within the scope of The Cryosphere. The authors demonstrate a substantial amount of work, incorporating a large number of climate simulations, multiple data products, a downscaling method, and SMB model calibration. Given this methodological complexity and the large number of simulations and data products used, it is particularly important that the methods are presented clearly. However, in the current stage this is lacking, and it constitutes my main concern. The presentation of the methods requires clarification before publication. I therefore provide general comments followed by a list of specific recommendations below. I recommend that the manuscript be considered for publication after major revisions.
General comments
Section 2.2: More details about the BESSI model are needed for the reader to fully understand and interpret the results. For example, it would be helpful to clarify how turbulent heat fluxes are calculated, how albedo is parameterized, and how melt and runoff are computed.
Section 2.3: The authors begin with a brief overview of the different data products and methods before describing each component in detail, which is a good structure. However, this section is currently difficult to follow because the overview remains too general. Improving this section would provide the reader with a clearer understanding of the overall methodology. More specific suggestions are provided below.
For the downscaling, analogs are created by selecting similar days from NorCP runs forced by two GCMs: EC-Earth and GFDL-CM3. Additional discussion is needed on how the climate of these models compares to the broader CMIP6 ensemble. Are these two models representative of the full range of plausible climates? Or might the spread of CMIP6 projections be underestimated by relying on only two ESMs?
Regarding using the analog downscaled fields as forcing, it is unclear whether all forcing fields are taken from NorCP to drive BESSI. The algorithm selects the most similar day based on precipitation and temperature, but are all variables (e.g., shortwave radiation, longwave radiation, humidity) taken from this selected day, or are some fields retained from the original RCM simulations?
How do BESSI simulations forced by historical EURO-CORDEX simulations compare to the SeNorge-forced BESSI run? Can the model output be trusted when forced by all these RCM–ESM combinations? Why was the decision made to include all EURO-CORDEX simulations rather than selecting only those models that perform well over the historical period?
A related point concerns the NorCP simulations: did you also run BESSI forced with the two NorCP runs over the historical period to assess their performance compared to observations or the SeNorge-forced reference run?
Specific comments
L35: What is meant by “glacier model uncertainty”? Does this refer to SMB, SEB, ice dynamics, or a combination?
L45: Please provide a reference for this statement.
L50: Please specify the regions referred to here.
L 52-54: Can you give the typical resolution of these models? Is the issue mostly in resolution or the fact the assume is hydrostatic or not?
L 62: “Improving SMB estimates”. If this is the goal it would be helpful to discuss the current state of SMB estimates and their limitations. Since this is addressed in Section 2.1, consider moving that discussion to the introduction.
L79-81: Please provide a reference. It would also help to include approximate values for average snowfall and melt rates.
L89–90: Please specify which glacier/SMB projections are currently available and why they are insufficient.
L92: What type of model was used to derive these estimates?
L109: Consider using “spread” instead of “width”.
L111-112: What resolution was NorCP run at? Also clarify here that it is forced by only two ESMs rather than the full CMIP6 ensemble.
L 113: A reference (figure or citation) is needed for the statement that EURO-CORDEX has unrealistic precipitation patterns over western Norway.
L119: Include the typical resolutions of the datasets here and clarify that SeNorge is a gridded observational product.
Figure 2: Given the complexity of the methods, this figure could be improved to better clarify the workflow. For example, indicate the number of NorCP simulations, the number of EURO-CORDEX RCM–ESM combinations, the role of ERA-Interim simulations, and which variables are subject to bias correction and analog downscaling.
L124: “Forced by dynamically downscaled GCMs”: Which model performed the dynamical downscaling? Or do the authors mean that NorCP is used to downscale EC-Earth and GFDL-CM3?
Table 1: first row; is this a mixed historical forcing of the two ESMs or one historical simulation for each? What version of EC-Earth is used?
L140: More prior explanation of the SeNorge dataset is needed here. Consider placing the section describing SeNorge before this section.
L141: ‘display similar patterns’: which patterns? Please specify and indicate where this is shown.
L 151: Can you comment on whether the bias is stable over the historical period? It would also be helpful to indicate the typical magnitude of the bias corrections.
Section 2.3.2: The manuscript would benefit from a table summarizing the ESMs, the RCMs used for downscaling, and their spatial resolutions.
L171-175: This section is unclear. If the bias correction is derived from the historical period and assumed constant, does it affect interannual variability in the SSP scenarios?
L 172: do you mean bias correction applied to the EURO-CORDEX instead of NorCP?
L 174: do you mean ‘climate variability’ in bias of the RCM?
L 178: ‘between models’: does this refer to RCMs? ESMs?
L 181: can you be more specific already what is meant with ‘most similar day’ in terms of what?
L 192: consider using ‘target day of the year’
L215: use ‘rectangular’ instead of square domain
L 219: ‘to not choose itself’: please clarify
L 234: It is unclear why an analog is constructed for SeNorge if it already represents the observational reference dataset.
L 247: How is this SMB for the basins used for calibration?
L253: ‘criteria’: do you mean evaluation criteria or performance metrics?
L 278: are these spreads based on the same number of RCM-ESM combinations within each scenario? How do the spreads change when considering only models available across all scenarios?
Fig 8a: can you explain why the inter-annual variability in the SeNorge-forced BESSI run is much larger than those forced by historcial RCM-ESM simulations? How does that affect your projections?
L 296. Please revise this statement, as historical simulations do not correspond to emission scenarios.
Table 3: What about summer precipitation? How is rainfall treated in the model?
L 329-330: this is unclear to me, please clarify statement
Section 3.4: did you assess how well BESSI performs in estimating runoff? Have SeNorge-forced simulations been compared to observed streamflow (if available)? It would also be useful to indicate what fraction of melt contributes to runoff.
L 355-257: this statement could be tested/demonstrated by plotting annual SMB against annual temperature
L 359-360: Is the precipitation gradient discussed here also present in the forcing datasets?
First paragraph of 4.2: Instead of the elevation explaining the stronger SMB response of the west, it seems from Fig 11b that for the same elevation, the west shows a stronger SMB decline compared to the same elevation on the west. Can you explain that?
L370: Is the relationship with winter temperature simply reflecting overall warming trends, or is there also a direct relationship with winter SMB?
L 383: What explains the large differences between model simulations? Could this be related to variables not included in the analog downscaling (e.g., radiation, humidity)?
L 400-402: How this SeNorge* was produced was not clear until now, this should be clarified earlier on in the methods.
Section 4.3: What is the advantage of the analog method compared to alternative approaches, such as training an emulator to reproduce NorCP simulations and applying it to the RCM–ESM ensemble?