| 30 Apr 2026
Characterizing melt ponds on sea ice in the Norwegian Earth System Model (NorESM2)
Abstract. Melt Ponds (MPs) are pools of open water that formed on Arctic sea ice during the warm months. They significantly affect the surface radiation budget of the Arctic Ocean and play an important role in the Arctic sea ice mass balance, climate and ecological systems. The second generation of the coupled Earth System Model developed by the Norwegian Climate Center, NorESM2, includes an explicit representation of MPs. To characterize the spatial distribution and the seasonal evolution of MPs, we introduced four key variables: melt pond fraction (MPF), relative melt pond fraction (RMPF), MP area (MPA), and pond area fraction (PAF). Using these variables, particularly PAF, we conducted a comprehensive evaluation of MPs in NorESM2 across the Pan-Arctic region and its four sub-regions: the Pacific, Atlantic, Laptev and Canadian sector. Additionally, we classified MPs base on ice type, distinguishing between first-year ice (FYI) and multi-year ice (MYI), over the period from May to September during 2002–2011. This study utilized observational data, including MPF from MODIS and MERIS, sea ice age from NSIDC, sea ice concentration from MODIS and AMSR-E, and surface downward solar radiation from SYN1deg, and reanalysis atmospheric data from ERA5. Our study revealed that while differences existed between MODIS and MERIS, these discrepancies were relatively small during May, June and September, which corresponds to the early melt season and the refreezing season. In contrast, larger discrepancies were observed during July and August, the peak melt season. These significant deviations are likely attributed to the challenges associated with retrieving MPF using coarse-resolution optical sensors. Despite these differences, the overall spatial patterns and temporal evolutions of MPs derived from MODIS and MERIS were largely consistent. MPs were found to form earlier and more extensively on FYI compared to MYI, leading to more abundance of MPs on FYI. NorESM2 successfully reproduced these general characteristics. However, the model consistently simulated the maximum PAF was in August (except for FYI in the Pacific region), which was typically one month later than in the observations. NorESM2 exhibited a systematic underestimation of MPs in May and June, not only in the pan-Arctic but also across the sub-regions, and on both FYI and MYI. Conversely, MPs on MYI in the Pacfic sector were overestimated, likely due to the higher prevalence of MYI in NorESM2 during these months in this region. The systematic underestimation of MPs in NorESM2 during the early melt season can be attributed to the too low surface downward solar radiation in NorESM2 and the too large retrieved MPF from MODIS and MERIS. The former attribution was further confirmed by the experiments conducted with the 1D sea ice model ICEPACK. Giving the critical role of MPs play in Arctic albedo and energy budgets, advancing the representation of MPs in the model has a profound implications for the Arctic system.
Review of “Characterizing melt ponds on sea ice in the Norwegian Earth System Model (NorESM2)” by Wang et al.
Summary and general comment
This study evaluates how well the Norwegian Earth System Model, NorESM2, simulates the evolution of melt ponds over Arctic sea ice by comparing NorESM2 model output with MODIS and MERIS satellite observations from 2002 to 2011. The observations show that melt ponds form earlier and are generally more abundant on first-year ice than on multi-year ice. NorESM2 captures the very broad cycle, but underestimates melt pond coverage during the early melt season and tends to simulate the seasonal peak in melt pond area about one month later than observed. The authors attribute much of this bias to insufficient downward solar radiation in the model, which delays melt onset and pond formation, highlighting the need for improved melt pond representation to better simulate Arctic sea ice and climate processes.
Overall, this is a useful topic and the paper is potentially a good contribution to this journal, but the analysis and discussion need to be tightened substantially. It’s evaluating only one climate model which obviously limits its scope, so it’s important to instead try and provide sufficient detail and care in the analysis to make it interesting to a wider audience. I think the May/June bias is probably the most robust result, while the more detailed intercomparison of raw melt pond fraction values is much harder to interpret and I think should be condensed (and ideally improved). The manuscript would be stronger if it focused more on the broad seasonal cycle and general timing biases, rather than presenting detailed numbers as though this were a robust quantitative intercomparison.
The treatment of the observational products was also quite basic, with e.g. limited consideration of significant sampling biases. The sensitivity experiment is also too limited. It mainly demonstrates that increased downward solar radiation leads to earlier melt pond development in general, rather than fully resolving the May bias or isolating the cause of the model-observation differences. The role of clouds/DSR in driving the bias is plausible, but not demonstrated robustly and is thus quite speculative. Other causes could also contribute, including melt physics and broader atmospheric differences. The conclusion statements need to be softened accordingly or the analysis improved considerably.
Specific comments
The introduction jumps around from introductory physics to available observations and then back again. This needs to be reorganized to improve flow. I suggest focusing first on the science problems, then the models and data/observations needed to investigate and constrain them.
We appreciate this was only recently submitted and is only a preprint, but there is a very relevant paper that should be considered and cited (Smith et al., 2026), including an analysis of melt ponds across multiple CMIP6 models including two versions of NorESM (MM and LM). Melt pond area in the MM is pretty similar to the LM and only slightly larger in July/August, despite being colder. This could be cited in the models section as well, where NorESM-MM is introduced.
The observational products are quite different, which makes it hard to ascertain the bias. More importantly, the results need to discuss the observational sampling gaps much more carefully, including their potential impact on the conclusions. As is, the treatment of the obs data is quite weak.
I would avoid quoting so many raw numbers in the model vs obs comparisons. Instead, focus more qualitatively on the broad seasonal cycle and how the simulations compare with the observations. Given the crude treatment of the observational sampling and data gaps/averaging, the current level of detailed numerical intercomparison is not justified.
Similarly, phrases like “clearly overestimated” should be removed or softened. The analysis is not robust enough to support that kind of language in our view. There is too much detailed comparison of results that are not especially reliable given the observational limitations.
The sensitivity experiment mainly demonstrates earlier melt pond development during June rather than resolving the May bias. As it stands, the sensitivity study does not add much beyond the basic point that increased DSR leads to earlier melt pond formation. It needs at least one additional test to be more convincing. The DSR sensitivity study is also extremely basic. Other causes could also contribute to the differences. For example, melt physics is not explored, and atmospheric differences beyond DSR could also be important.
r1i1p1f1, r2i1p1f1, and r3i1p1f1 are used here but we noticed that r1i1p4f1 also provides melt pond output. Is there a reason to not use perturbed physics members?
The justification for using MODIS-Low is quite weak. This needs to be explained more clearly.
There should also be more justification for using the period 2002-2011. There are other observational datasets, e.g. later MPD1 (2017 to 2023) and MPD2 (2017 to 2025) that could have easily been used to extend this forward in time.
We recommend computing the NorESM MPA over the full domain (pole included) and adding to the current Figure 4 or stating how adding the pole hole in the calculations impacts MPA and PAF
The authors do not provide an exhaustive list of causes of the MPF bias. DSR may be the most plausible cause, but more context is needed.
The conclusions are too strongly stated. I would stick to explaining what the sensitivity study does and does not demonstrate.
The northern sea ice age distribution is quite different, so I am not sure an ice-age comparison is especially logical here.
Technical comments
This is not the best written manuscript and needs careful editing. There are a lot of spelling and grammar issues throughout. I have not listed all of these individually.
All the melt pond acronyms are a little confusing. Please simplify where possible and make sure all acronyms are defined clearly. Define acronyms in the figure captions please.
I would simplify the discussion of Figure 5. The current discussion does not seem to add very much.
L51: The statement “AI techniques” is too vague, as this likely just applies to the processing of remote sensing data.
L60: The discussion of AI is not clear and I think it should be dropped. As noted later, the MPF dataset from MODIS uses ANNs.
L125: It is not really a “column version” of CICE; it is the column physics model of CICE, ICEPACK. CICE essentially adds a dynamic core around ICEPACK. This only needs a slight rewording for clarity.
L137: This statement feels too strong given the limited validation. I would suggest softening it.
L143: How are these variables actually used in the analysis?
L160: Why the specific emphasis on ERA-Interim here? The rationale is not clear.
MPF calculation: the calculator of MPF and RMPF is done on monthly timescales here, but it would be better to do this first at daily timescales before averaging monthly (we assume you have access to the full daily output).
L205: The notation here seems unnecessary for simply saying that a mean is calculated over time and across grid cells. More importantly, how are the sampling differences handled, especially given the significant quantity of missing data in the observations? We recommend just stating that the multi-year seasonal mean is computed over all months from May to September and all years from 2002 to 2011. Also, maybe a good idea to remove September due to a lack of observational coverage, especially since the MODIS dataset only includes half of September. You should also consider removing it for any seasonal mean plot for a more relevant model-observation comparison.
L461: This is not an exhaustive list of causes. Presenting it as such creates problems for the discussion that follows and is a major weakness of the current framing. Figure 7 of Smith et al., (2026) also suggests an issue with the parameterization promoting melt pond concentrations that are a strong function of sea ice concentration, that does not seem to be supported by the observations.
L530: Does it really show that? I would stick to explaining what the sensitivity study does and does not demonstrate.
References:
Smith, M., C. Cardinale, A. Petty, H Niehaus, (2026) Regional Biases in Arctic Melt Ponds Between CMIP6 Models and Satellite Observations, ESS Open Archive [in review at JGR Oceans], https://doi.org/10.22541/essoar.15001935